4th Hull Performance & Insight Conference

HullPIC 19
Gubbio, 6-8 May 2019

Edited by Volker Bertram

 Sponsored by

[Link]

[Link]

[Link] [Link] [Link] [Link]

[Link] [Link] [Link]

[Link] [Link] [Link] [Link]

[Link] [Link] [Link]

[Link] [Link] [Link]

[Link] [Link] [Link]

1
 Index
Johnny Eliasson 4
Discussion on Ecological Risks of Bio-Fouling: Harmonizing and Vetting Related Approaches
Volker Bertram 9
Some Fairy Tales in Performance Monitoring Revisited
13
Fully Automated CFD Analysis for Full-Scale Ship Hydrodynamics
Naoto Sogihara, Takashi Yonezawa, Tsuyoshi Ishiguro, Yoshihiko Sugimoto, Shinta 21
Masuyama, Hirohisa Mieno, Mamoru Shimada, Yoshiro Matsubara
Introduction of OCTARVIA Project: Project for Ship Performance in Actual Seas
Matthew Patey 31
Performance Monitoring Information Feedback to Design
Tetsuro Ashida, Naoki Mizutani, Pekka Pakkanen, Arttu Väisänen, Teemu Manderbacka 38
Performance Monitoring Insight at MOL
Erik van Ballegooijen, Torben Helsloot 50
An Approach to Monitor the Propeller Separately from the Hull
Nikos Themelis, Christos Spandonidis, Christos Giordamlis 56
Evaluating the Efficiency of an Energy-Saving Device by Performance Monitoring
Florian Kluwe, Lars-Uve Schrader, Herbert Bretschneider 65
From Bow to Stern: Hydrodynamic Measures for Increased Hull Performance
Paolo Becchi 76
ISO 19030: Onboard Monitoring System for Real-time Performance Feedback, Prediction and
Optimization
Miyeon Jeon, Yoojeong Noh, Kyung Hwan Jeon, Sang Bong Lee, Inwon Lee 86
Real Ship Maritime Big Data Analysis for Prediction of Fuel Consumption
Simon Doran 97
A Short History of Hull Cleaning and Where Do We Go Now
Serena Lim, Roger Teo, Teck Chong Sia 103
A Digital Business Model for Vessel Performance Monitoring
Gunnar Prytz, Rune Gangeskar, Vemund Svanes Bertelsen 114
Distributing Real-Time Measurements of Speed Through Water from Ship to Shore
Jean-Marc Bonello, Tristan Smith 128
Exploring the Effect of Vessel Performance Information Barriers on Decision-Making Practice:
A Time Charter Application
Carlos Gonzalez, David Lara Arango 143
Techniques for the Automated Detection of Anomalies and Assessment of Quality in High-
Frequency Data Collection Systems
Anders Östman, Kourosh Koushan, Luca Savio 153
Study on Additional Ship Resistance due to Roughness using CFD
Daniel Schmode, Ole Hympendahl, Francesco Bellusci 162
Influence of Data Sources on Hull Performance Prediction
Tomoki Wakabayashi, Yasuhiko Inukai, Takeshi Yonezawa, Norihide Igarashi, Masahiko 172
Mushiake, Satoshi Kawanami
Full-Scale Measurement of a Flow Field at Stern using Multi-Layered Doppler Sonar (MLDS)

2
Iason Stefanatos, André Kauffeldt, Eirinaios Petoumenos 179
Case Study on Performance Improvement using Advanced Hull Performance Analysis Metho-
dologies
Beom Jin Park, Myung-Soo Shin, Gyeong Joong Lee, Min Suk Ki 193

Luke De Freitas, Noah Silberschmidt, Takis Pappas, David Connolly 201

Full-Scale Performance Measurement and Analysis of the Silverstream Air Lubrication System
Najmeh Montazeri, Jakob Buus Petersen, Søren Vinther Hansen 211
Autolog Data Processing for Vessel Performance Application
Ditte Gundermann, Danny McLaughlin 220
When ISO19030 Fails: Utilizing Basic Machine Learning/Data Fitting Methods for Perfor-
mance Analysis without Reference Data
Angelos Ikonomakis, Jesper Dietz, Klaus Kähler Holst, Roberto Galeazzi, Ulrik Dam Nielsen 229
Application of Sensor Fusion to Drive Vessel Performance
Abel Vargas, Hua Shan, Eric Holm 242
Using CFD to Predict Ship Resistance due to Biofouling and Plan Hull Maintenance
Bastian Marquardt, Alessandro Castagna, Amirhossein Rahimi, Aaqib Gulzar Khan, Johanna 254
Marie Daniel, Christian Voosen, Nikhil Mathew
Reducing the Resistance of Container Ships with a FLUME® Roll Damping Tank
Sergiu Paereli,Manolis Levantis 261
ISO 19030 2.0: How to Improve
Kevin Murrant, Allison Kennedy, Rob Pallard, Marcel Montrose 272
Effects of Hull and Propeller Cleaning on Propulsion Efficiency of an Offshore Patrol Vessel
Wojciech Gorski 292
Information Sharing Concept in Ship Performance Management Systems
Philip Chaabane, Catherine Austin, Dan Isaksson, Oliver Weigenand, Cecilia Ohlauson 301
Application of Biotechnology in Antifouling Solutions for Hard Fouling Prevention

Index of authors 312

Call for Papers for next HullPIC

3
 Discussion on Ecological Risks of Bio-Fouling:
Harmonizing and Vetting Related Approaches
Johnny Eliasson, Chevron Shipping LCC, San Ramon/USA, [Link]@[Link]

Abstract

This paper highlights some of the issues of decreasing the environmental footprint of shipping we are
faced with and the complex environment under which new guidelines, and possibly future regulations
are developed. To serve the marine industry NACE International has started working group to discuss
the environmental concerns of biofouling on ships. This working group is not tasked to write guide-
lines or standards. It is simply tasked with discussing the subject. The group has, however been asked
to act as a review group for other organizations that are developing guidelines. This can help with
harmonization and vetting of such developments. The NACE TEG532X is open to all, and participa-
tion is encouraged particularly by ship operators.

1. Introduction

The part of the United Nations tasked with governing worldwide shipping is the International
Maritime Organization (IMO). Shipping is in effect governed by international law. IMO develop
mandatory regulations that the Nation States enact into law and enforce. The regulations are adopted
by consensus by all member states. Each member state may impose stricter local regulations above
and beyond the minimum required by IMO regulations. This is however generally discouraged. To
have uniform and global rules facilitate smooth transport an essential part of global trade.

With the increased concern about human influenced global warming trends IMO must take action to
reduce the footprint from shipping. There are also other environmental related issues IMO has been
addressing in recent decades, such as the emission of acid exhaust gases. This paper will focus on the
effect of ship biofouling and highlight some of the activities under way at IMO and outside of IMO.

Biofouling present primarily two risks to the marine environment;

1.
2. It presents a risk of spreading invasive aquatic species

The increase in hull resistance directly influence fuel usage and as such greenhouse gas emissions.
Introduction of invasive aquatic organisms into a new marine environment not only affects
biodiversity and the health of eco systems but can also have detrimental impacts on a number of
economic sectors such as fisheries, aquaculture and ocean energy.

2. IMO and the environment

IMO adopted the Ship Energy Efficiency Management Plan (MEPC.282(70) and the Guidelines for
ouling to minimize the transfer of invasive aquatic species,
Resolution MEPC.207(62), in July of 2011. The latter sets out the format and recommendations for a
vessel specific Biofouling Management Plan and Biofouling Record Book. United States Coast Guard
(USCG) required vessels to have such a plan effective June 21, 2012 (CFR 151 2000). California
imposed the same as of 1/1/2018. Add to this New Zeeland, Australia and other. These documents are
subject to Port State Control vetting.

The IMO has launched a new international effort to combat the negative environmental impacts of
biofouling - The GloFouling Partnership. The official launch was in London on October 25th, 2018, in
conjunction with the IMO Marine Environmental Protection Committee (MEPC) 73, and the first
GloFouling event was organized in London, in February 2019.

4
GloFouling is a collaborative project involving the Global Environment Facility (GEF) and the United
Nations Development Programme (UNDP) and the IMO tasked it with pushing for the implementa-
of best
practices and standards.

Although the core group behind this effort involves 12 nations, a mix of developing nations and small
island states, it has received endorsement from over 40 major stakeholders, representing academia,
industry associations, technology developers and private sector companies in the marine field.

IMO will focus on shipping while the Intergovernmental Oceanographic Commission of UNESCO
(IOC) will join to lead the approach to other marine sectors.

Hiroyuki Yamada, director of the Maritime Environmental Division at IMO, stated

to implement the IMO Biofouling Guidelines and best practices for other marine industries will help
He
also highlighted the additional contribution of biofouling management to the reduction of greenhouse
gas (GHG) emissions from shipping through energy-efficiency gains resulting from clean hulls.

3. Other active organizations

International Paint & Printing Ink Council (IPPIC) lead a GloFouling session at the Sustainable Ocean
Summit (SOS) in Hong Kong, 14 November 2018, with members from NACE TEG532X, IPPIC
presenting among others. Some highlights included improvements in performance monitoring,
optimum antifouling selection, data to evaluate fouling risk via diver inspections, the balance of risk
AIS vs biocide usa

offshore structures, and other key infrastructures involves not only technical research hurdles, but also
a host of regulatory challenges producers must face bringing these products to markets. Coatings
researchers must balance the requirement for products that not only control biofouling effectively, but
which do so in a way that is cognizant of the need to minimize ancilliary environmental damage,
including harm to other (non-
airish@[Link])

Shipping conference / workshop Shiptec in China is very focused on the environment with high goals
to exercise control. The efforts are presently apparently mainly focused on greenhouse gas (GHG),
Sulfur and Nitrous oxide gas reductions.

Active Shipbuilding Experts Federation (ASEF) is actively involved in the GloFouling project and
keep a close relationship to NACE. Members of ASEF are also contributing to the NACE TEG532X
working group.
of 2017 discussions was held on what the shipbuilder industry can contribute to Biofouling
Management.

Baltic and International Maritime Council (BIMCO) is developing a hull cleaning standard or best
practice guideline. This work started in 2018. This attempt to standardize terms and tools for hull
grooming and cleaning will facilitate use of these important tools. NACE TEG532X is acting as a
review group in support of that important effort. Hull grooming or cleaning is an important part of

hull performance plan. BIMCO also launched a membership survey to gain better insight into how
shipowners are dealing with the biofouling issues.

Institute of Marine Engineering (ImarEst) in cooperation with IPPIC launched a template Biofouling
Management Plan on August 2017 (Allen Irish, airish@[Link]). Such plans are mandatory to have

5
onboard ships delivered on or after 1 January 2018 and to other ships on completion of their first
regularly scheduled out-of-water maintenance (i.e. drydocking) on or after that date by mandate in
California. An ImarEst TV recording of the 22 February 2018 presented a technical lecture to the
Operational and Environmental impacts and
IMarEST Biofouling Expert Group was formed following the
inaugural Australia /New Zealand /Pacific (ANZPAC) Workshop on Biofouling Management for
Sustainable Shipping, held in Melbourne, Australia, in May 2013. The aim of the group was to assist
and promote further discussions and international consultation on the development and
implementation of practical, effective and globally consistent biofouling management measures for
shipping. The key issues identified were:

1. Effective & practical biofouling management measures

2. Biofouling management guidelines, requirements & regulations: present & future
3. In-water cleaning of ship hulls: costs, benefits, impacts & regulation
4. Regulation & scrutiny of new and existing fouling control coatings and antifouling biocides
5. Costs & impacts of biofouling: ship energy efficiency & harmful aquatic species transfer
6. Ship biofouling management: best practice guidance

levels
of allowable biofouling depending on the vessels itinerary. The State of California tightened up its
regulations on biofouling as off 1 January 2018, to include the mandatory management of biofouling
hulls. Australia issued a Marine Notice in September 2017 advising about
revisions of the 2015 Antifouling and In-water Cleaning Guidelines for Australia.

International Association of Classification Societies (IACS) holds the view that biofouling regulations
are likely to grow in scope and geographical reach.

Niche areas (sea chests, bow thrusters etc.) generally have restricted access meaning the application of
antifouling control products can be challenging. When in service, changes in water flow in these areas
can also lead to increased fouling challenge compared to the main hull areas. Biofouling control in sea
chests are often assisted by the use of Marine Growth Prevention Systems (MGPS). The IMO guide-
lines give advice on design and construction to minimize small niche areas. However, effective
fouling control can only be achieved via correct use of antifouling paints and operation of the MGPS
in conjunction with regular in-service inspections.

4. NACE TEG532X

The TEG532X working group at NACE International is tasked with Discussion on Ecological Risks
of BioFouling. The group is not tasked with producing standards or other specific documents. It is
tasked simply with discussing the subject. It is a very large and international group with members that
are also members of many of the other groups, some mentioned herein, working on various related
solutions, best practices and standards. This does not mean that other NACE working groups cannot
evolve as a spin-off of these discussions. One example is a new NACE working group that was
formed around developing a best practice standard on hull treatment of ships in a dry dock.

5. Chevron experience

There are already known best practices that might not be utilized at present to the fullest. A project
started at Chevron Shipping in 2014 with the aggressive target of reducing the fuel used per miles
travelled by 20%. The result after 4 years is a reduction of 29% well above the original target. It
should also be mentioned that the performance data used, noon reports in combination with publicly
available data, is not granular enough, and the result might best be described as a range between 25%
and 35%. The analysis used ISO19030 with some necessary modifications.

This is the result of many contributing factors:

6
 1. Strict adherence to our propeller polishing policy
2. Structured monitoring of the fouling degree / evolution by divers
3. Early hull grooming rather than late hull cleaning
4. Optimizing hull treatment in dry dock; squaring preferred over spot blasting
5. actual operative profile
6. Monitoring of hull performance
7. Power trail improvements
8. Raised awareness onboard

In other words, it is important to employ all cost-effective solutions we have at our disposal to yield
the most promising result.

6. Mapping the various biofouling activities

There are many organizations worldwide developing best practices and other solutions with the aim of
reducing biofouling on ships. Fig.1 shows many of these organizations and how they interact. Harmo-
nization of efforts is imperative that the outcome is practical and effective.

References

IMO MEPC.282(70) - The Ship Energy Efficiency Management Plan

IMO MEPC.207(62) -
mize the transfer of invasive aquatic species

7
Fig.1: Organizations developing best practices and other solutions to reduce biofouling on ships, see
Fig.2 for abbreviations

Fig.2: Abbreviations used in Fig.1

8
 Some Fairy Tales in Performance Monitoring Revisited
Volker Bertram, DNV GL, Hamburg/Germany, [Link]@[Link]

Abstract

This paper looks critically at some assumptions and allegations floating around in performance moni-
toring: data frequency as a cure-all; capability to separate hull degradation and propeller degrada-
tion; capability to correct for sea state 4 and above. Lack of error analyses is pointed out as a fre-
quent root cause for questionable confidence and assertions.

1. Introduction

The HullPIC conference has promoted our collective insight into hull and propeller performance
monitoring. And, by and large, the performance monitoring solutions employed now are much better
than what was on the market when the Working Group for ISO 19030 started its work.

We have moved from the dark middle ages to a period of (early) enlightenment. However, some old
and half-truths are hard to eradicate, especially if they make for convenient short-cuts in
performance monitoring models. Most
therefore sound so convincing. And covered in a cloak of a nice- -age
mumbo-jumbo, the middle-age beliefs keep coming back. Maybe this paper can contribute to more
transparency and enlightenment on some of the most popular half-
tales, urban myths, or marketing).

2. Urban myths concerning data quantity

2.1. More data sets = better results, always

ever-higher sampling frequency will improve insight. ISO 19030 uses 0.07 Hz (1 data set every 15 s)
as minimum requirement for the default method, but some in the industry boast sampling rates of
1 Hz and above. But how useful is higher data sampling frequency really?

We could copy and paste the same data set 1000000 times. Obviously, that would not give any more
insight. More data sets are no good if they are exactly the same, and little good if they are almost the
same. We need independent data sets with sufficient variation in (steady) variables to derive useful
insight. If the sampling frequency is higher than the frequency of ship motions (encounter frequency),
there will be spurious changes in key variables. While added resistance in waves is negligible for most
ships up to sea state 3, the periodic surge motion in longitudinal direction induces non-negligible
changes in speed, propeller rpm and torque. These make averaging a necessity. The recommendations
of ISO 19030 make sense here. With 10-minutes averaging, the fluctuations due to ship motions will
average out for most cases. Only in following seas, very low encounter frequencies may occur.

There is no harm (and little gain) in higher data sampling frequency. The average values over 10
minutes are probably quite constant if data frequency is increased. Perhaps someone with access to
high-frequency data could verify this assumption. The advantage of high-frequency data is that both
average value and standard deviation (or another measure of variation) can be derived. This can help
in filtering data sets where above-average variation indicates atypical conditions during the sampling
interval, e.g. maneuvering.

In any case, averaging intervals should be short enough to ensure that the statistical characteristics of
the seaway (significant height and period, direction) and ship speed are constant. Again, the ISO
19030 default recommendations of 10 minutes fulfills this requirement.

9
2.2. Everything will average out for the best

Everything will average out for the best This is another hopeful adage in performance monitoring,
particularly if correction methods are weak. But this assumes that there are no long-term operational
changes (e.g. different routes due to different demand), no large-scale ambient changes such as
fluctuations in the Gulf stream or sea state climates.

For ferries on a fixed route, with little vari

-purpose vessel, most likely by the time averaging has
taken care of your errors, any useful insight comes too late.

3. Hull and propeller lived separately ever after

We can separate propeller and hull p And we can make it look

scientific or we can make it simple and appealing to the common sense. We have three (virtually)
independent variables: speed, propeller rpm and torque. If, compared to the clean reference condition
at same propeller rpm, we have

a) Lower speed, then the resistance has increased and we blame hull fouling
b) Higher torque, then we blame propeller fouling/degradation

In reality, for same rpm, we will typically see both speed loss and torque increase. The hull fouling
-
action effects are small; then we could use as a rule of thumb the percentage loss in speed and the
percentage torque increase and split the total degradation accordingly. But this is very coarse and the
insight is mostly that you should clean both propeller and hull.

One can try to estimate the contributions from propeller and from hull by using more elaborate theo-
ries. Maybe I should be pleased that many such attempts give reference to my book, Bertram (2012).
But should I be pleased when the formulas are used without the hydrodynamic understanding that I

The formulas use assorted efficiencies and/or wake fraction w and thrust deduction t. If you only
look at the design condition (design draft, zero trim, design speed, no wind, no sea state), these
variables are constants as the single symbol suggests. But for performance monitoring, they are
functions of many parameters. For example, we should write instead of w; and any
approach should document how each function is modelled. For example: w will change with draft,
and therefore always take the value at design draft. w will change with trim,
ue for zero trim. w
get the idea.

The quantities depend also on scale, i.e. they differ between model tests and full-scale ship. This
-
ba (10% have been reported in oral communication
by Maersk) between different model basins. If such approximate full-scale extrapolation for a specific
ship are not available from model-basin reports, some people use an approximation of the approxima-
tion: design formulas (for design conditions and typically based on ships tested in the 1960s) as found
in Bertram (2012).

models are
approximations. Some may be very good approximations, some may just get the order of magnitude
right. ISO 19030 quantified the uncertainty for the performance indicator for the default method
described in Part II of the standard, using random variations within the range of uncertainties of the
input variables to see how this would affect the final results. A corresponding approach would be
needed in the assorted hydrodynamic models trying to separate propeller and hull performance. We
need to estimate uncertainty (or accuracy) of variables and functions and see how the errors propagate

10
to the end result. Maybe some very rough estimates are OK, maybe some simplifications lead to 50%
variation in the final numbers. Assuming uncertainties and seeing how they propagate is a task any
developer can do and we should ask for this at least in scientific publications.

Quantifying the effect of speed, draft, trim, possibly induced motions by ambient waves on e.g. the
wake number at full scale would be a nice research project for the CFD (computational fluid
dynamics) community.

4. My formula / machine learning can correct for sea state 4 and above

We can correct for sea state 4 and above This may come in disguise with filters set at higher sea
states. We can apply formulas and software, but the errors will be high, easily 50%, possibly 100%.
Among experts, we may argue whether correcting for sea state 4 is possible with acceptable errors.
For sea states up to 3, you can use any correction or none at all (as in the default method of ISO
19030). If your wave heights are derived from sea state estimates from the crew, you may as well
omit any correction. See Bertram (2016) for a detailed discussion.

If we have good measurements of the actual near-field of waves around the ship and good three-
dimensional methods to compute the added power in oblique waves, we may correct for higher sea
states. But probably the best approach would be to compute the speed loss or added power and filter
based on percentage of the calm-water power for that case, as promoted in Schmode et al. (2018).
Vendors of wave measurement equipment and services should publish comparisons with wave buoy
measurements to get realistic estimates for errors in wave measurements. Then error propagation
analyses should give insight into the effect on performance indicators.

Machine learning is just another (and more obscure) way of approximating a relation between waves
and added power or speed loss. It does not change the fundamental dilemma.

5. Conclusion

Data frequency is not a cure-all. Averaging over sampling intervals is necessary to remove fluctua-
tions from ship motions in waves. When averages and standard deviations no longer significantly
change with higher frequency, further increase in frequency becomes pointless.

Nobody can correct reasonably for higher sea states, because the initial information on waves
becomes too uncertain and the correction methods have large errors in real seaways. Filtering at sea
state 3 is fine, filtering at sea state 4 often an uncomfortable necessity; beyond that, the data sets cause
more harm than good.

We should collectively work more on error estimates, particularly on error propagation in the
performance monitoring models.

Acknowledgement

I am grateful to Prof. Heinrich Söding, who helped sharpening my views on the finer points of
(practical) ship hydrodynamics as required for this paper.

References

BERTRAM, V. (2012), Practical Ship Hydrodynamics, Butterworth & Heinemann, Oxford

BERTRAM, V. (2016), Added power in waves Time to stop lying (to ourselves), 1st HullPIC Conf.,
Pavone, pp.5-13, [Link]

BERTRAM, V. (2017), Some heretic thoughts on ISO 19030, 2nd HullPIC Conf., Ulrichshusen, pp.4-

11
11, [Link]

SCHMODE, P.; HYMPENDAHL, O.; GUNDERMANN, D. (2018), Hull performance prediction

beyond ISO 19030, 3rd HullPIC Conf., Redworth, pp.135-140, [Link]
hullpic2018_redworth.pdf

12
 Fully Automated CFD Analysis for Full-Scale Ship Hydrodynamics
Inno Gatin, Wikki Ltd, London/UK, [Link]@[Link]
, Wikki Ltd, London/UK, [Link]@[Link]
Hrvoje Jasak, Wikki Ltd, London/UK, [Link]@[Link]

Abstract

A Computational Fluid Dynamics (CFD) software environment for establishing full-scale ship
hydrodynamic characteristics with low cost and time to delivery is presented in this paper. The
framework features fully automated process of CFD analysis from geometry input to report
generation. Two applications are supported: prediction of resistance curve and power curve. The
automated power curve prediction enables trim optimisation knowledge database calculation for low
cost. The software requires minimal user interaction by using pre-set numerical settings that are
tailored for steady resistance in calm water and self-propulsion, providing accurate CFD simulations.
The framework is based on an open-source software eliminating licence-related fees, enabling low
overall cost.

1. CFD in the marine industry

Many speak of Computational Fluid Dynamics (CFD) as the great new technology that can solve all
problems in designing ships of all kinds. Yet, we do not seem to see this in everyday life of ship
design offices. CFD is getting more accurate, fast and reliable but still it did not spread in the marine
industry like many have anticipated. Sure, there is a plethora of research groups and R&D
departments in big companies doing CFD but this is still not the market-wide application that can
really transform the industry. Small design offices seldom rely on CFD, and only a fraction of the
fleet uses CFD-generated trim optimisation data-bases to save fuel in daily operation
(Reichel, Minchev, & Larsen, 2014).

s the problem? It is no secret that CFD is computationally expensive, but is this really the
number one reason why it is not spreading through small design offices and ships in service like wild
fire? Not really. The cost of High Performance Computing (HPC) has reduced so much that almost
any CFD software can calculate ship steady resistance at five speeds for less than 200 EUR of CPU-
related costs. Cloud-based HPC services are widely available and easy to use. The problem seems to
lie on the other side of the computer screen the human effort and expertise that are necessary to
prepare and interpret the simulations. A man-hour of a highly skilled expert is far more expensive
than computer resources, and this is a two-fold obstacle for everyday application of CFD. First, the
highly skilled expert can usually do only the one thing CFD calculations, and second, he is very
expensive! This prevents many small to medium ship design offices and ship owners to use CFD:
their intermittent needs for CFD do not justify employing an expensive expert to sit around most of
the year. Even if they do, it does not guarantee a successful application of the technology.

In order for CFD to really contribute to the shipping industry on a larger scale, these problems need to
be addressed. The amount of human effort needs to be minimised to reduce the man-hour cost, while
at the same time the number of input parameters needs to be brought down to ship particulars and load
condition details, leaving CFD-specific parameters away from the user. This can be done by
automating the entire process, which also solves the second problem: the need for a highly skilled

complete CFD calculation process, from pre-processing to post-processing. Generating the

computational grid is the most time-consuming part of the process, and requires heavy user-
interaction in an iterative process. Running the simulation, i.e. processing, requires specific know-
ledge of CFD in order to properly set-up numerical parameters, often entailing considerable
uncertainty of the user with respect to different parameters leading to human error and poor results.
The post-processing step is straight forward and does not differ from other similar engineering

13
activities, where a comprehensive document containing relevant results and graphs should be
produced. Still, it consumes a significant amount of precious man-hours.

Despite the difficulties, fully automating the CFD process is possible, but only for narrowly
specialised applications. Luckily, calm water ship resistance and self-propulsion are specific enough
to allow automation. This paper presents a fully automated numerical framework for these two
specific applications that is a result of years of experience in performing such calculations, which
yielded a set of numerical settings that are broadly applicable for different vessel types. A
parametrised grid generation methodology is developed that is fully autonomous, requiring only basic
ship information. The automatic self-propulsion calculation capability is applicable to generating low-
cost knowledge data-bases needed for trim optimisation, where the delivered propeller power is
calculated depending on the draught, trim and vessel speed. The numerical framework is based on in-
house CFD software specialised for naval hydrodynamics called the Naval Hydro Pack.

2. About the software

The Naval Hydro Pack is a CFD software based on collocated Finite Volume method. In this work,
we rely on Level Set for interface capturing. Special discretisation techniques are employed based on
the Ghost Fluid Method to guarantee high accuracy of the two-phase flow model ( , Jasak &
Gatin 2017). The ship propellers are modelled using the actuator disc model where a pressure jump is
prescribed on a circular surface representing the propeller. The key feature of the algorithm is the
ability to assess the undisturbed propeller inflow velocity without the need to perform a separate open
water calculation ( , 2019).

3. Calm water resistance

The developed automated framework enables various vessel types to be considered, i.e. no
assumptions are made on the ratios of L, B and T, number of hulls, or on the Froude numbers. Fig.1
shows a few examples of different vessels during a calm water resistance simulation produced with
the fully-automated procedure. Calculation of ship resistance in calm water conditions is very
important and therefore deserves to be verified and validated in detailed. The Naval Hydro Pack has
shown to be sufficiently accurate and precise in the past (see e.g. evi , 2015);
however, some recent results from , 2018 will be shown here.

Fig.1: Various hull forms sailing in calm water

14
To test the automated procedure, three different, publicly available hull forms are selected and
simulated. Namely the KCS, JBC and DTMB 5512 hull forms are used, [Link]
[Link] both due to the availability of the
geometries and experimental results. Accuracy is reported in this section, while man-hour and
computational costs are presented in Section 5.

Table 1 shows the main particulars of the three vessels. Here, denotes the scale of the model. Note
that the simulations are performed in model scale to allow direct comparison against experimental
results; however the numerical framework makes no assumptions on the scale of the ship. Table 2
shows the characteristics of computational grids generated by the automatic framework. Fig.2 shows
the side-view of the generated grids for the three hull forms.

Table 1: Ship characteristics

Item KCS JBC DTMB
LPP , m 230 280 142
,m 3 52 030.0 178 369.9 8 702.7
V, kt 24 14.5 14.6
Fr 0.26 0.142 0.201
31.6 40.0 46.6

Table 2: Characteristics of computational grids automatically generated for the three test hull forms
KCS JBC DTMB
No. Cells 1 609 281 1 733 267 2 219 751
No. Hexahedra 1 526 009 1 699 753 2 183 431

Fig.2: Side-view of the generated computational grids for KCS, JBC and DTMB hulls, in that order
from top to bottom

15
Table 3 shows the results of the steady resistance simulations for KCS, JBC and DTMB hull together

dynamic trim angle, while E rr stands for relative error calculated as Err (EFD CFD) , where
EFD denotes the experimental result and CFD the simulation result. The differences are within 4% for
resistance and within small range for sinkage and trim as well. Note that the results are obtained from
the first try using the automated framework, and no simulations were repeated in order to get better
agreement.

Table 3: Results of ship resistance in calm water for KCS, JBC and DTMB compared to EFD
Item KCS JBC DTMB
3.64 4.13 4.54
3.71 4.29 4.50
1.9 3.7 -0.9
,m -0.441 -0.31 -0.08
,m -0.445 -0.24 N/A
,m 0.004 0.07 N/A
,o 0.182 0.104 0.21
,o 0.169 0.103 N/A
o
, -0.013 -0.001 N/A

4. Self-propulsion

-propulsion
simulation needs to be performed where the resistance of the hull is balanced by the thrust force of the
propeller. An actuator disk model is applied in the present numerical model, which is fed with the
open-water curves of the actual propeller. A proportional-integral controller is applied in order to
adjust the rotation rate of the propeller until the resistance of the hull is balanced by the thrust. The
result is the power needed to be delivered to the propeller and rotation rate.

To gain confidence in the current approach the code was tested on three different cases:

1. Self-propelled model-scale JBC test case with experimental comparison published in Bakica,
, 2019. It was shown that the resistance force acting on the
hull in the self-propelled condition agreed with experimental measurements within 0.3%,
while the thrust and torque coefficients were assessed with accuracy of around 8%.
2. Full-scale case of a car-carrier produced by the Uljanik shipyard in Croatia, where the
measured mile results were compared to CFD simulations ( ,
2019). Here, instead of fixing the ship speed, the power delivered to the propeller was fixed,
while the result was the ship speed on the measured mile and propeller rotation rate. The
calculated ship speed was within 0.1% of the measured speed on the fine grid, while the
calculated 126.27 RPM gave a difference of 0.24%. Fig.3 shows the forward speed calculated
on three different computational grids compared to the value measured at the measured mile
on sea-trials. Fig.4 shows the perspective view of the ship in the simulation, where the
actuator disc model can be observed.

16
Fig.3: Forward speed of the full-scale, self-propelled car-carrier: CFD results and measured mile data
( , 2019).

Fig.4: Car-carrier in the self-propulsion simulation

Fig.5: Forward speed of the general cargo carrier Regal, as measured on the sea trial and calculated
using CFD ( , 2019)

17
 3. Full- Computer
Simulations, Ponkratov (2017), [Link]
[Link], where sea trials were conducted to serve as benchmark
data. The ship in question is a general cargo carrier Regal. In this case, a fixed propeller
rotation rate is applied and the final ship speed is reported. Fig.5 shows the forward speed
convergence compared to measured values during the sea trial. It can be observed that CFD
converges towards the mean of the two measured values (measured mile in one direction and
the opposite, denoted with blue lines). The calculated ship speed falls within 0.2% of the ISO
15016 speed.

Fig.6: General cargo ship regal during the self-propulsion simulation

5. Automatic framework

The automated procedure starts with user input, where data such as main ship particulars, loading
condition particulars (draft and displacement), speed range and similar are supplied to the software. A
Python based environment takes over from there, first generating the computational grid and then
setting up all the necessary simulations. The simulations are then ready to be executed using the Naval
Hydro Pack software. When the simulations are finished, a document is automatically generated
comprising results for all load conditions and speeds. Graphs are included where relevant items are
plotted against speed, such as resistance, power or RPM. Additionally, for the calm-water resistance
simulations polar graphs of inflow velocity distribution in the propeller plane are automatically
produced, as shown in Fig.7.

Fig.7: Automatically generated wake field plots for KCS, JBC and DTMB, from left to right

6. Cost analysis

For the three calm water resistance simulations performed in Section 3, an analysis of invested man-
hours and computational time is conducted. Note that running the self-propulsion simulations using
the automated framework exerts essentially the same amount of human effort; hence the data
presented below can be applied to self-propulsion analysis as well. All simulations are conducted on
64 cores of Intel Xeon Processors, E5-2637 v3, 15M Cache, 3.50 GHz.

Table 4 shows the amount of man-hours, computational time and cost required for individual hull

18
forms. A cost of 0.1 EUR per core-hour is assumed in the computational cost assessment. In average,
it takes around 40 minutes of human effort for a single hull form. Note that this number would not
change even if more simulations for individual ships were performed, e.g. at five different ship speeds
or different static trim angles. Thus, 40 minutes represents an average amount of human effort per
ship, irrespective of the total number of simulations conducted for that ship. The computational time
scales linearly with the number of simulations, and on average it costs 8.7 EUR to perform a single
simulation. Note that self-propulsion simulations can exert more computational resources; hence the
price of computations can go upward.

Given the computational cost of individual simulations, it is obvious that the cost of man-hour is
indeed dominant. Reducing the man-hour cost to less than one hour per ship is an essential key to
bring CFD analyses of this type to widespread industrial application.

Table 4: Man-hour cost and CPU time needed for individual calm water resistance simulations
KCS JBC DTMB Average
Man-hours per ship, h 0:25 0:46 0:48 0:39
CPU wall-clock time per simulation, h 1:06 1:01 1:58 1:22
Total core-hours per simulation, h 70.4 64.0 126.9 87.1
CPU cost per simulation, EUR 7.0 6.4 12.7 8.7

6. Conclusion

The main reason why CFD is not being applied widely in the marine industry is the cost of conducting
accurate and reliable calculations, which is a consequence of high complexity of the method. In an
effort to address this issue, and to make a step forward towards CFD application in every-day marine
industry activities, a fully automated computational framework is developed and presented in this
paper. The framework is based on a CFD software called the Naval Hydro Pack, specialised for
problems encountered in marine and offshore hydrodynamics.

The automated framework reduces the required man-hours to 40 minutes per ship, which is an order
of magnitude reduction with respect to current practice. This applies to calm water resistance and self-
propulsion simulations, which are essential in ship design and for generating knowledge data-bases
for trim optimisation onboard ships in service, respectively. As an example, the prediction of calm
water resistance for ten different vessel speeds would take 40 minutes of human effort, and around 90
EUR of computational resources.

References

BAKICA, A.; GATIN, I.; ; JASAK, H.; VLADIMIR, N. (2019), Accurate assess-
ment of ship-propulsion characteristics using CFD, Ocean Engineering, pp.149-162

GATIN, I.; JASAK, H.; VUK EVI , V. (2015), Validation and Verification of Steady Resistace KCS
Simulations with Sinkage and Trim using Embedded Free Surface Method, Tokyo 2015: A workshop
on CFD in Ship Hydrodynamics

GATIN, I.; .; JASAK, H.; RUSCHE, H. (2017), Enhanced coupling of solid body
motion and fluid flow in finite volume, Ocean Engineering, pp.295-304

GATIN, I.; .; .; JASAK, H. (2018), Fully Automated Ship Resistance

Prediction using the Naval Hydro Pack, Technology and Science for the Ships of the Future, Trieste,
pp.256-263

JASAK, H.; .; GATIN, I.; . (2019), CFD validation and grid sensitivity
studies of full scale ship self propulsion, Int. J. Naval Architecture and Ocean Engineering, pp.33-43

19
PONKRATOV, D. (2017), 2016 Workshop on Ship Scale Hydrodynamic Computer Simulations,
, Southampton

PONKRATOV, D.; ZEGOS, C. (2015), Validation of Ship Scale CFD Self Propulsion Simulation by
the Direct Comparison with Sea Trial Results, 4th Int. Symp. Marine Propulsors

REICHEL, M.; MINCHEV, A.; LARSEN, N.L. (2014), Trim Optimisation - Theory and Practice,
Int. J. Marine Navigation and Safety of Sea Transport, pp.387-392

.; JASAK, H. (2015), Seakeeping Validation and Verification using Decomposition

Model based on Embedded Free Surface Method. Tokyo 2015: A workshop on CFD in Ship
Hydrodynamics

.; JASAK, H. (2015), Validation and Verification of Decomposition Model based on

Embedded Free Surface Method for Oblique Wave Seakeeping Simultions, Tokyo 2015: A workshop
on CFD in Ship Hydrodynamics

; JASAK, H.; GATIN, I. (2017), Implementation of the Ghost Fluid Method for the
Free Surface Flows in Polyhedral Finite Volume Framework, Computers and Fluids 153, pp.1-19

20
 Introduction of OCTARVIA Project:
Project for Ship Performance in Actual Seas
Naoto Sogihara, National Maritime Research Institute, Tokyo/Japan, sogihara@[Link]
Takashi Yonezawa, Monohakobi Technology Institute, Tokyo/Japan,
takashi_yonezawa@[Link]
Tsuyoshi Ishiguro, Japan Marine United Corporation, Tokyo/Japan, ishiguro-tsuyoshi@[Link]
Yoshihiko Sugimoto, Mitsui O.S.K. Lines, Tokyo/Japan, [Link]@[Link]
Shinta Masuyama, Kawasaki Kisen Kaisha, Tokyo/Japan, [Link]@[Link]
Hirohisa Mieno, Chugoku Marine Paints, Hiroshima/Japan, hirohisa_mieno@[Link]
Mamoru Shimada, Nippon Paint Marine Coatings, Tokyo/Japan, [Link]@nippe-
[Link]
Yoshiro Matsubara, Kansai Paint Marine, Hyogo/Japan, [Link]@[Link]

Abstract

This paper describes the OCTARVIA project, a Japan Maritime Cluster Collaborative Research by 25
stakeholders (including ship operators, shipyards, paint makers, machinery and equipment makers,

objectively evaluate and compare the ship performance. Three main topics are addressed: evaluating
ship performance in operation; predicting ship performance in design; presenting ship performance in
operation to owners and operators. The first proposal is validated through the monitoring data of
various ship types.

1. Introduction

The Greenhouse Gas (GHG) reduction strategies are decided at International Maritime Organization
(IMO), which results in the start of Data Collection System on fuel consumption from January 2019.
The evaluation on the performance of ships in service is becoming a trend. The fair implementation of
the evaluation strongly requires the accurate measurement of the ship performance in actual seas for
each ship. Making the ship performance in actual seas visible under certain indices contributes to
realization of more efficient maritime transport and GHG reduction .

A project for the evaluation of ship performance in actual seas , in which 25

companies of a Japanese maritime cluster collaborate, has been launched as a three-years project aiming
to establish a Participants
of OCTARVIA project are composed of eight sectors:

Ship Owners
Classification society
PainT makers
ShipyArds
PRopeller & Rudder makers
GoVernor maker
Research Institute
WeAther consulting company

The project organizes three working groups for the sub-themes as following;

S1-WG : Working Group for establishment of ship performance monitoring method in actual seas
S2-WG : Working Group for establishment of estimation method of ship performance in actual seas
S3-WG : Working Group for establishment of evaluation of ship performance in actual seas

21
This paper describes the outlines of OCTARVIA project and introduces the activities of S1-WG which
is closely related to the topics of HullPIC.

2. The outlines of OCTARVIA project

OCTARVIA project has discussed the technical issues on the evaluation of ship performance in actual
seas as shown in Table 1. To address these issues, OCTARVIA has set up three sub-themes shown in
Fig.1. S1-WG treats the performance evaluation at the operation phase, and S2-WG treats the
performance evaluation at the design stage. S3-WG treats how to present the ship performance in actual
seas to ship owners and operators based on the evaluation method developed by S1-WG and S2-WG.

Table 1: Technical issues on ship performance discussed in OCTARVIA

Phase Ship performance in calm seas Ship performance in winds and waves
Design The validation is enough carried out by The procedure for model tests and pre-
tank tests and sea trials. diction is not established.
The correlation between prediction and
full-scale is not validated.
The performance cannot be compared
objectively among ships.

Operation The evaluation method of the perfor- The procedure for measurement, analy-
mance in service is not established be- sis, and evaluation based on monitoring
cause the draft condition in speed tri- data is not established.
als is limited. The accuracy of monitoring data is not
The current monitoring cannot evalu- sufficiently ensured.
ate the performance enough to assess The performance resulted from monitor-
fouling and aging effect. ing cannot be compared among ships.

CFD
in winds
in waves

Added Towing tank

resistance in
waves

Self-propulsion Actual sea

factors in waves model basin
Onboard monitoring data and relations
between SHP-Vs-NE in calm sea condition

Case study on evaluation fuel

consumption in an operation model.
Presentation method of ship
performance in actual seas to ship
owners etc.

Fig.1: Outlines of OCTARVIA project

22
OCTARVIA focuses on the development of a
performance in actual seas. The participants of OCTARVIA are able to apply the outcomes of the
project for their own research and development.

2.1 Working Group for establishment of ship performance monitoring method in actual seas
(S1-WG)

S1-WG is aiming to establish a performance monitoring method of ships in actual seas and the activity
is outlined in section 3. The performance monitoring method has been globally spread and ISO19030
(2016) has been developed as an international standard for measurement of changes in hull and
propeller performance. However, ISO19030 has noted that the wave correction would be considered in
the future revisions of the standard. There is no international standard for evaluation of ship
performance in actual seas by analyzing monitored data.

The ship performance in actual seas is derived from the combination of the ship performance in calm
seas and the effect of external forces acting of a ship such as winds and waves. In other words, the ship
performance in calm seas should be evaluated first. In this regard, shipyards have a lot of experiences
and knowledge through sea trials because conventionally they conduct weather corrections of sea trial
data. ISO15016 (2015) is recognized as an international standard for the assessment of ship performance
by an analysis of sea trial data.

One of the objectives of S1-WG is the establishment of the method of a displacement correction. It is
well-known that in each voyage a ship varies the displacement resulted from the change in cargo. The
Admiralty coefficient can be one solution for the displacement correction, however, its application is
limited within small displacement range. The method for a displacement correction applicable to wide
displacement range is expected to be developed.

Furthermore, the performance monitoring system involves various kinds of instruments. The accurate
evaluation of the ship performance in actual seas requires incorporation of sufficiently accurate
instruments. In order to accomplish this, S1-WG discusses the required accuracy of each instrument
based on the estimation on the effect of instruments on the ship performance in actual seas.

2.2 Working Group for establishment of estimation method of ship performance in actual seas
(S2-WG)

Although a ship performance in actual seas has been evaluated at the design stage based on model tests
or theoretical estimation, the method should be further validated through model-scale data and full-
scale data. Making an objective and visible evaluation on the ship performance in actual seas requires
to improve accuracy of the estimation method. S2-WG addresses the issue above and aims to establish
the estimation method of ship performance in actual seas and discusses the following topics.

1) Evaluation on the ship performance in waves

2) Model test procedures for measurement of ship performance in waves
3) Prediction of wind forces and moment
4) Model test procedures in wind tunnel

2.2.1 Evaluation on the ship performance in waves

S2-WG incorporates VESTA, Tsujimoto et al. (2015), developed by National Maritime Research
Institute (NMRI) as the base program for the evaluation on the ship performance in actual seas. VESTA
can predict added resistance in waves by involving the test data obtained in short waves, Tsujimoto et
al. (2008), and estimate added resistance in winds by the empirical formula, Fujiwara et al. (2005).
VESTA also can calculate self-propulsion factors in waves based on the algorithm of load variation
test.

23
S2-WG addresses the improvement on the estimation on added resistance in beam to following waves
and self-propulsion factors in waves. Accomplishment of the improvement requires model test data
with high accuracy for the validation of the estimation method. The model tests are carried out at the
towing tank and the actual sea model basin, Fig.2.

Fig.2: Test in waves of a bulk carrier model at a towing tank

2.2.2 Model test procedure for measurement of ship performance in waves

S2-WG is aiming to establish a model test procedure for the measurement of ship performance in waves.
The procedure targets measurement of added resistance and self-propulsion factors in waves with
practicality and high accuracy.

The model tests can provide the benchmark data for the validation of the estimation method of added
resistance and self-propulsion factors in waves. The model tests in waves also provide the data for the
validation of Computational Fluid Dynamics (CFD) simulation. For example, both of the tank test and
the numerical simulation are carried out with the models which have different shapes of bow.

2.2.3 Prediction of wind forces and moment

Many studies have been conducted on the prediction of wind forces and moment. They proposed
empirical formulae for the prediction of wind forces and moment based on wind tunnel tests. In a few
decades CFD technology has been evolutionally advanced and is applied to various fields of industry.

S2-WG aims to develop a comprehensive guideline for the prediction of wind forces and moment by
means of CFD. NMRI in-house code "NAGISA", Ohashi et al. (2018), has been utilized in the present
study. The following items are mainly discussed with comparing the computed results to the measured
data in a wind tunnel.

Effect of ship type

Effect of wind velocity and direction
Effect of wind velocity profile

2.2.4 Model test procedure in wind tunnel

The International Towing Tank Conference (ITTC) introduces wind tunnel tests in its Recommended
Procedures and Guidelines as one of means to estimate wind force on speed and power trials, ITTC
(2017). In wind tunnel tests, a model fixed to a load cell is set in winds for a measurement of wind
forces and moment. Measured forces and moment are transformed into non-dimensional parameters by
a representative wind velocity, an air density, a model length, and a projected area of a model.

24
 Fig.3: Wind tunnel test with a bulk carrier model

The key parameter of the wind tunnel tests is the representative wind velocity. While the representative
wind velocity can be identified in uniform flow, generally the wind velocity profile shows boundary
layer generated on the bottom of wind tunnel. The boundary layer is caused by specific phenomena for
wind tunnels and some wind tunnels apply exponential wind velocity profile. Therefore, S2-WG
discusses carefully the definition of the representative wind velocity and the wind velocity profile to be
used for the wind resistance test.

Throughout the discussion including the treatment on the representative wind velocity, S2-WG aims to
establish a reliable test procedure in wind tunnels. For the confirmation of the effectiveness of the test
procedure, many wind tunnel tests are conducted which provide benchmark data for the CFD guidelines
for the prediction of wind forces and moment.

2.3 Working Group for establishment of evaluation of ship performance in actual seas (S3-WG)

S1-WG and S2-WG address the establishment of the estimation method of ship performance in actual
seas by means of the ship performance monitoring and the ship performance simulation, respectively.
In order to make the best use of the outputs from S1-WG and S2-WG, an index based on the outputs
which is easily understood to ship owners and operators is necessary to be developed. S3-WG aims to
develop the index of the ship performance in actual seas in order to establish the evaluation method of
the ship performance.

Fig.4 Concept of Life Cycle Fuel Consumption

S3-WG discusses a concept -

the long-term prediction of fuel consumptions in actual seas. The concept of LFC is shown in Fig.4.

25
LFC is calculated based on the estimation method of ship performance in actual seas developed by S2-
WG. LFC is provided by a main engine output derived from the long-term prediction in a standard
operation model and a separately prepared specific fuel consumption (SFC). A standard operation
model prescribes the conditions for the calculation of the index, such as weather conditions and draft
conditions. The calculation of LFC illustrated in Fig.5 includes the effect of fouling and aging
deterioration on ship performance in actual seas as well as the occurrence probability of the weather
condition.

S3-WG also discusses the target accuracy of OCTARVIA project, which results in 5% accuracy of
added resistance and self-propulsion factors in waves required in the evaluation of the ship performance
in actual seas. Taking into account that LFC is calculated on the basis of simulated fuel consumption,
S3-WG set the target accuracy 2% of fuel consumption in actual seas.

Fig.5: Illustration of calculation of Life-cycle Fuel Consumption

3. The activities of SI-WG: Analysis of ship monitoring data

3.1 Approaches for the analysis of ship monitoring data

S1-WG discusses two approaches for the analysis of ship monitoring data. One approach
derived from simplified formulae and
does not use the detailed hull form data and tank test data which are usually confidential data for ship-
yards. Simplified Analysis is a helpful approach for the companies such as ship owners and operators
who cannot easily access the detailed hull form data and performance data based on model tests or CFD.

Table 2: Data arrangement for analysis of ship monitoring data

the detailed data mentioned above and

can carry out more realistic analysis than Simplified Analysis. Shipyards can perform Detailed Analysis
because they can access the detailed data. Table 2 shows data arrangement for the analysis of ship

26
monitoring data by Simplified Analysis and Detailed Analysis.
Table 2 requires only limited information such as a ship type and ship principal particulars.
Therefore, it is not hard for the companies other than shipyards to analyze ship monitoring data. The
effectiveness of Simplified Analysis is verified based on the discussion in the teams for Detailed
Analysis in which both Simplified Analysis and Detailed Analysis are performed.

S1-WG has selected 10 ships shown in Table 3 for discussion on the analysis of ship monitoring data,
bearing in mind that typical ship types are chosen, and organizes teams for performing analysis of ship
monitoring data for every objected ship. While the team for Simplified Analysis performs only
Simplified Analysis, the team for Detailed Analysis performs both Simplified Analysis and Detailed
Analysis which enables a verification of Simplified Analysis.

Table 3: Ships and approaches for analysis

3.2 Expression of a mathematical model on performance in calm seas

As shown in Table 2, Simplified Analysis estimates a ship performance in calm seas based on empirical
formulae and ship monitoring data. In Simplified Analysis, the data deemed in a calm condition are
extracted and expressed as Eqs.(1) and (2) where VS is the ship speed in knot, NE is the engine revolution
in rpm, and P is the engine output in kW.

bn
P an N E cn (1)
N E d nv Vs (2)

Eqs.(1) and (2) give discretized performance data: the relationship among ship speed, engine revolution,
and engine output. The inverse analysis of a power estimation based on the discretized performance
data can provide a resistance curve and an effective wake coefficient in full-scale, by involving propel-
ler open characteristics by optimized-design, a thrust deduction factor, and a propeller rotative efficient
estimated by empirical formulae. The obtained resistance curve and the effective wake coefficient in
full-scale are treated as input data to estimate the ship performance speed drop and fuel oil consumption
in actual seas by VESTA.

3.3 Analysis interval of monitoring data

The use of automatic monitoring systems for ship performance evaluation in service is rising. Time
histories of ship performance or weather are recorded with high frequency (e.g. 1 s). Some monitoring
systems can calculate statistical values such as a mean value and a standard variation onboard and
transmit them to land. S1-WG discusses the analysis interval with which statistical values are calculated
including the sampling interval. While it is preferable to be provided with statistical values to the
analysis, some monitoring systems can transmit only instantaneous values. Therefore, S1-WG also
discusses a proper interval which is effective for an instantaneous value instead of statistical value.

27
 transmission

statistical value transmission

sampling instantaneous value

interval
time
Analysis interval
Fig.6: Discussion on analysis interval

3.4 Data filtering conditions

In estimating a ship performance in calm seas by ship performance monitoring, the data in calm seas
condition should be appropriately extracted. On this regard, ISO19030 recommends that the data for
evaluation on performance in calm seas should be extracted under the condition in which the wind
speed is less than 7.9m/s. However, S1-WG considers that it is not preferable to apply such condition
to all ships uniformly since the effect of weather on ship performance depends on the size of a ship.

S1-WG focuses on the rate of resistance increase due to winds and waves as a candidate parameter for
filtering. The rate of resistance increase is expected to be applied irrespective of ship size since it is
relative parameter representing the extent of weather effect. The rate of resistance increase is estimated
in accordance with ISO15016, which requires not only wind data but also wave data for the better
estimation. Although ISO19030 postpones wave correction, S1-WG incorporates the accurate wave
data provided by Japan Weather Association and investigates the effect of waves.

3.5 Example of Simplified Analysis

The example of Simplified Analysis for c

Table 3 is demonstrated in this chapter. The principal particulars of the container ship and the tanker
are shown in Table 4.

Table 4: Principal particulars of the container ship and the tanker

C

Simplified Analysis was performed twice. Simplified Analysis starts with data fitting in calm condition
by the mathematical model expressed as Eqs.(1) and (2), which requires the filtering condition for
extracting the data in calm seas. Though the discussion on the filtering condition is continued in parallel,
the tentative filtering condition is set as shown in Table 5.

Table 5: Tentative filtering condition for extracting the data in calm seas in Simplified Analysis
Wind speed Significant wave height

1st Simplified Analysis none

less than 7.9 m/s
LPP
2nd Simplified Analysis less than 1.35 [m]
100

28
20 voyages for the container ship and 18 voyages for the tanker are analyzed. For each the voyage, data
fitting by the mathematical model expressed as Eqs.(1) and (2) is conducted, which provides with
performance data (i.e. resistance curve and effective wake coefficient in full-scale). Such performance
data are input data of the simulation by VESTA of the time history for predicting the ship speed, the
engine output, and the fuel oil consumption in actual seas.

The comparison on the total fuel oil consumption in each the voyage between monitoring data and the
simulation by VESTA can confirm the accuracy of ship performance in calm seas obtained from the
monitoring data. The ship performance in calm seas is essential for the evaluation on that in actual seas.
Difference of total fuel oil consumption between monitoring data and the simulation by VESTA is
defined as expressed in Eq.(3) where FOCVESTA is the total fuel oil consumption by VESTA simulation
and FOCMEAS is that based on monitoring data.

FOCVESTA FOC MEAS

FOC (3)
FOC MEAS

The histogram of FOC for the container ship and that for the tanker are shown in Figs.7 and 8,
respectively. The figures confirm that the inclusion of wave height to the filtering condition improves
mean and standard deviation of FOC. On the total fuel oil consumption, VESTA overestimates abt.
3% for the container ship and abt. 5% for the tanker.

The ship performance in calm seas input to VESTA includes the effect of winds and waves, which
results in an overestimation of total fuel oil consumption. In other words, the application of the
mathematical model to the weather corrected data can give more accurate simulation.

Fig.7 The histogram of FOC for the container ship

(left: 1st Simplified Analysis, right: 2nd Simplified Analysis)

Fig.8 The histogram of FOC for the tanker

(left: 1st Simplified Analysis, right: 2nd Simplified Analysis)

29
4. Concluding remarks

This paper outlines the activities of OCTARVIA project and introduces the working groups for its sub-
theme. S1-WG addresses establishment of ship performance monitoring method in actual seas, S2-WG
addresses establishment of estimation method of ship performance in actual seas at the design stage,
and S3-WG addresses establishment of evaluation of ship performance in actual seas. These working
groups work together in close cooperation to produce the maximum output in the project.

OCTARVIA project addresses not only technical issues discussed in S1-WG and S2-WG but also the
application of the indices on ship performance in actual seas in S3-WG, which contributes to the
establishment of a the evaluation of the ship performance. The global
the promotion of building eco-friendly ships and the low-
emission operation of GHG. These can contribute to realization of more efficient maritime transport
and GHG reduction, which is the goal that OCTARVIA is aiming to achieve.

References

FUJIWARA, T.; UENO, M.; IKEDA, Y. (2006), Cruising performance of a large passenger ship in
heavy sea, ISOPE2006, San Francisco

ISO 19030 (2016), Ship and marine technology -Measurement of changes in hull and propeller perfor-
mance, ISO, London

ISO 15016 (2016), Ship and marine technology -Guidelines for the assessment of speed and power
performance by analysis of speed trial data, ISO, London

ITTC (2017), Preparation,Conduct, and Analysis of Speed/Power Trials (7.5-04-01-01.1), Interna-

tional Towing Tank Conference
[Link]

OHASHI, K.; HINO, T.; KOBAYASHI, H.; ONODERA, N.; SAKAMOTO, N. (2018), Development
of a structured overset Navier Stokes solver with a moving grid and full multigrid method, JMST,
[Link]

TSUJIMOTO, M.; SHIBATA, K.; KURODA, M.; TAKAGI, K. (2008), A Practical Correction
Method for Added Resistance in Waves, J. Japan Society of Naval Architects and Ocean Eng. 8

TSUJIMOTO, M.; SOGIHARA, N.; KURODA, M.; SAKURADA, A. (2015), Development of a ship
performance simulator in actual seas, OMAE2015, Newfoundland

30
 Performance Monitoring Information Feedback to Design
Matthew Patey, Foreship Ltd., Helsinki/Finland, [Link]@[Link]

Abstract

Performance monitoring solutions and services are primarily intended for ship owners to manage and
improve the performance of vessels in operation. The data that is collected can also be useful in the
design of new vessels as well as in conversion and retrofit projects. This paper presents some ideas
how this can be done as well as an example of the application of collected performance monitoring
data in design studies.

1. Eco-efficiency Issues in Ship Design and Conversion Projects

Improving the energy efficiency and environmental impact of shipping operations has continued to be
a significant factor in new ship design, as well as a motivating factor for many conversion projects.
The decision to make these improvements may be due to legislation, as is the case with some
environmental measures like ballast water management systems or exhaust gas scrubbers, or they may
be due to other business decisions by the shipowner e.g. to reduce fuel expenses or improve brand
appeal. The drive for improvements in performance is always limited by the available resources for
the development and implementation of new technologies or procedures. The assessment of new
technologies or procedures should therefore be done as realistically as possible, not just to judge the
technical effectiveness of a new technology, but also to answer the question does it make practical
business sense to implement it.

It is with this practical issue in mind that this paper attempts to present the benefit of using
performance monitoring data in the design stage. The issue is not just whether a new technology will
whether or not normal design decisions are actually going to work in practice,
given what the owner knows about the performance of the ships in operation compared to the
traditional assumptions employed in standardized design procedures.

2. Uncertainty and Complexity in Ship Performance Design

Whenever possible, ship design projects normally work on the basis of experience, for example,
through the use of standard procedures to assess the design or by implementation of proven
technologies. In a lot of cases the designer has no choice but to follow procedure e.g. because the
administration and class society rules dictate how the design assessment is to be carried out. It is true

may be used if it can be shown that they provide an equivalent level of

in proving something - to the authorities is something the designer avoids unless they
are forced to use an alternative procedure.

But in a lot of other cases the designer has the freedom to do what they like. This can obviously be a
problem if experience does not exist in-house
not yet been established. Even in cases when standard procedures do exist and there is time to carry
them out, there can be uncertainty in the validity and usefulness of the results.

One familiar example of this is the estimation of required propulsion power. It is standard procedure
to evaluate the performance of the hull using RANS CFD methods at an early stage of the design
together with some typical propulsion efficiencies and then later by assessing a final version of the
hull with self-propulsion tests in a model basin. Two uncertain issues related to this include whether
or not the model test procedures sufficiently represent the real full-scale ship and whether or not the
CFD procedures model the flow with sufficient detail. While these methods may be sufficient to say
The same sort of uncer-

31
tainty exists with the wind resistance. When it comes to additional power required due to the ship
operating in waves, there is even more uncertainty in the measurements and calculations.

In the end we get some estimate of required propulsion power based on standard model test proce-
dures (which vary in the details from one test basin to another) at constant speed in calm water plus
ess as to the wind resistance. At the design
stage this is considered to be the most trustworthy estimate of the required propulsion power. The
hydrodynamicist then has to balance this propulsion power estimate with the commercial pressure to
try to use as little engine power as possible. At some point, however, there must be a decision about
the propulsion solution to implement, the engine arrangement and all their related systems. Uncer-
tainty in the propulsion power means uncertainty in the fuel consumption, uncertainty in the optimal
engine arrangement and uncertainty in e.g. the size of the fuel tanks.

e problem of the energy

consumption on the vessel and the required installed power. For example:

How is marine growth accounted for in the propulsion power estimate?

The hotel load on a large cruise ship can be 50% of the total required power for some cruising
speeds. How is this estimated in the total power requirement?
We are considering exhaust gas steam turbine generators, fuel cells, air lubrication system,
Flettner rotors, solar panels, etc. but they only work well in certain conditions. So how much
less power do we need if we implement this technology and will it cover the cost of imple-
mentation?
We know from experience the engine test bed SFOC curve is never how much fuel the engine
burns in practice. How do you account for that when sizing the fuel tanks?
We are going to use LNG. What is the experience with the fuel consumption for this type of
engine?

All of these issues make the challenge of assessing what power plant is the best highly uncertain and
difficult to evaluate. Ultimately the energy efficiency and commercial effectiveness of the vessel
could be compromised due to conservative decisions made in order to alleviate the risk due to these
uncertainties.

Looking at the bullet list above, it can be seen that these examples all have a lot to do with the
operation of the vessel. This implies that in order to assess the contribution of these items to the
energy balance on the vessel we need to know how the vessel is operated and where it is operating. So
we come to th

The operational profile of the vessel includes the distribution of vessel speed, and the distribution of
time spent in certain operations (manoeuvring, in transit, at anchor, in port, etc.) but this is only part
of the information needed. When speaking of the operational profile we need to know not just the
speed profile, but the actual route and time of year the vessel is operating on these routes. This would
help us extract the expected environmental conditions, if we actually can get the environmental data
from some source. Putting this information together in a useful form to assess e.g. electrical power
demand for a diesel electric powered cruise ship is not a simple task, but it is something that is always
desired.

3. Use of Performance Data and Voyage Simulations in Design Decisions

Performance data collected onboard an existing ship can help provide some of the data which com-
prises a new vessel operational profile, e.g. speed profile, propulsion power profile, engine mode
profile, service load, etc. The post-processing of this data can provide insight into the real expected
behaviour of the ship. The quality of the collected data is of course a concern, as is the way in which
the raw measured data is post-processed before being made available to the shipowner and the

32
designer. Even if the data can be trusted, there still is the problem of applying this data to a new
design, which may be intended to operate on routes for which very little or even no operational data
exists.

There is also the issue of putting all of the measured data together with other data used in the design.
This is usually a bit of a mess, with the design team dealing with many sources of data in many
different forms from several different design disciplines. Consistency between data sets can be
difficult to maintain, especially as the design progresses in different departments in the design organi-
zation. It should be more convenient if all of the questions related to the energy balance on the ship
could be handled in a more consistent, realistic manner with one approach while reducing the amount
of uncertainty in the evaluation of each contributor to the energy consumption onboard.

One way to address this issue is through the use of voyage simulations, Fig.1. In a voyage simulation
a mathematical performance model of the vessel is used to calculate the performance of the ship on a
given route. can also include the seakeeping
behaviour, as well as the effectiveness of any equipment contributing to the energy consumption and
production. Departure and arrival dates and times at given ports are set and the voyage proceeds
according to a series of discrete steps in time or distance. Environmental conditions, typically from a
hindcast weather database, are varied during the voyage according to the vessel position, date and
time. By varying the date and time of departure, many years of simulations can be conducted, giving
many of data about fuel consumption, propulsion power, etc. The complexity of the
model and the simulation is limited by the software tool employed to conduct the simulations, but it
should be apparent that the tool must be capable of modelling the systems which are involved in the
energy balance to the level of detail which provides confidence in the related design decision.

Fig.1: The voyage simulation process, illustrated using a sample voyage in an older version of the NAPA
Voyage Optimization software used onboard some ships for voyage planning.

The benefits of using voyage simulations include:

it is easier to manage many aspects of the energy efficiency analysis of the vessel than with a
dozen different spreadsheets and separate analysis programs used by different design depart-
ments and disciplines

33
 simulations can realistically represent the operation and performance of the vessel along the
actual planned routes using hindcast weather databases, easing concerns about the realism of
the information used in the assessment
it is easy to modify the vessel design and the performance model to conduct design studies re-
lated to energy efficiency
real operational data can be taken into account, as explained below
need the operational profile because the operational profile is actually a result of the
simulations

Realistic simulations depend very much on the nature and quality of the performance model used in
the simulations. Ultimately performance model predictions are based on analytical or statistical
formulations, which in turn have been verified and validated by experience. So we get back to the
same problem that old experience should be corrected by new observations.

This is where the measured performance data can be used to calibrate the performance model. This is
what is normally done in voyage planning software systems used onboard ships the initial
installation of the system contains a baseline performance model used to make recommendations
about planned speed and weather routing. The base model is adjusted over time using the measured
data from the performance monitoring system. The same approach can be used in simulating voyages
using a hindcast weather database and performance data from an existing ship. The measured perfor-
mance data, e.g. fuel consumption, can be compared to that predicted by simulating the exact same
voyage for the ship to derive correction factors and correction methods to apply to the analytical
models used.

To apply this data to a new design is the next problem. The same analytical models would be used for
the new design, but the predictions should incorporate the calibration factors from an existing
reference vessel for which performance data is available. In order to reduce the risk in applying this
approach the reference vessel should be as similar as possible to the new design.

Using a reference vessel may seem like a risk if the ships are somewhat dissimilar, or call into
question the need to conduct such comprehensive analysis, i.e. why not just use the reference vessel
the new ship? B from all these details and ask ourselves the
really fundamental question:

What information do you trust?

The engineer
experience, it is easier to convince people of the trustworthiness of measured operational data than it
is to convince them of model test results, CFD or some empirical formula, as long as common sense
and experience agree with the observations. If real operational data is available, it should be used. In a
new design this is not available, but neither are the model test or CFD results early in the design
process and the empirical formulas are not necessarily reliable. We always have to rely on reference
ships and reference data at some point. It would seem from a common-sense point of view to be better
to rely on measured operational data from a reference vessel than model tests from the reference
vessel. Later in the new vessel design the empirical models get replaced by CFD or model test results
for the new vessel, but these too should be corrected taking into account the real operational data from
the reference vessel. How to make these corrections is a completely new issue and will not be
But the
basic principle we try to follow always is use the best data available.

4. An Example from Recent Experience

At Foreship we have been conducting voyage simulations as part of some newbuilding consultancy
projects for about 3 years now, primarily focussing on comfort analysis onboard cruise ships and

34
prediction of fuel consumption on sample voyages, taking into account the weather conditions
encountered. The results have been used in making decisions about operational feasibility of
particular routes and aspects of the general arrangement design, amongst other things.

We had the opportunity to take advantage of full-scale operational data recently in one of these
studies. We approached one owner saying we had a great new way to calculate fuel consumption for
their newbuilding project. They were sceptical about the approach in general and they were also
questioning the accuracy of the analytical models used in the simulations. There were also several
aspects from operational experience that they needed to take into account in any energy efficiency
analysis. These included:

Marine growth effect on powering had to be taken into account, considering also the normal
drydocking schedule and cleaning of the hull.
Engine availability due to planned and unplanned maintenance had to be considered.
Energy saving technologies were to be employed, namely exhaust gas steam turbine genera-
tors, absorption chillers and hull air lubrication systems (ALS).
The service load should be taken into account as accurately as possible.

To address these concerns, we suggested that we try to use the full-scale performance data they had
for their ships, first by calibrating the fuel consumption and propulsion power models with the
measurements on specific voyages for their reference vessel and then applying the correction factors
to the new design.

4.1 Use of Performance Monitoring Data in the Voyage Simulations

The first thing was to develop the correction factors for the performance model for the reference
vessel. The performance model in the simulations consists of, among other things:

total propulsion power as a function of speed

propulsion efficiencies
engine powers, modes and SFOC curves
service load profile i.e. service load as a function of time of day

We were primarily interested in the required propulsion power and fuel consumption. For this we
asked for whatever performance data the owner had which could be used for this, preferably in the
raw data format, for as many voyages as possible. We got a flood of data from a 6 month period in a
series of text files dumped out of the performance monitoring system using 5 minute or 15 minute
averages. From this data we could figure out the actual measured fuel consumption, develop a daily
average power profile for the service load and figure out the actual average propulsion power used for
any specific section of a voyage.

The calibration process consisted of simulating a section of the measured voyages over which the
speed was more or less constant using hindcast weather data for the same dates and times of these
voyages. Sections of 36 individual port-to-port voyages were simulated from a single itinerary which
was repeated several times over the six month period. The SFOC curves were corrected by a factor
provided by the owner based on their experience. The service load profile was developed from the
measured service load data. Overall propulsion efficiencies were provided by the owner as standard
reference values they wished to use. Performance of the exhaust gas steam turbine generator was
taken as an average power from the measured data for each voyage. Based on this a simple estimate of
the reduction in the average service load demand as a function of speed could be developed. Finally
the performance of the air lubrication system (ALS) was taken into account as a further reduction in
propulsion power demand. This was based on an earlier analysis of measured data of propulsion
power with the ALS on and off, carried out by the supplier of the performance monitoring system.

35
Other factors which were tuned in the model were the propulsion power curve and the auxiliary power
demand from the activation of an engine. The fit of the performance model was judged by comparing
the total fuel consumption and average propulsion power on each voyage according to the simulation
to that of the actual measured data. In the beginning, we tried to be clever and employ a genetic
algorithm to minimize the discrepancy between the predicted and measured fuel consumption and the
predicted and measured propulsion power. This turned out not to be so useful, and in the end the
propulsion power curve was easier to adjust manually in 2 or 3 iterations to get already a quite good
correlation between predicted and measured average propulsion power. With this manual tweaking of
the power curve, the achieved correlation between predicted and measured total fuel consumption was

The tweak -power curve was then compared to the

eed-power curve for the reference vessel to develop a set of correction factors
to apply to the new vessel.

Statistics of the marine growth effect on propulsion power were provided by the owner and used to
adjust the corrected propulsion curve as the simulation progressed. At the beginning of the simulation
this effect was set to zero and increased until the next scheduled drydocking, when it was assumed the
hull cleaning was so good that the marine growth effect was again zero.

to remove engines from the set of available engines. Planned maintenance was handled by removing
one engine on a regular basis for a single port-to-port voyage. Unplanned maintenance i.e. engine
failure, was simulated using a random number generator and given probability of failure to remove an
engine for a single port-to-port voyage.

In summary, the performance models of both the reference vessel and the new design used data from
performance monitoring systems for the following items:

propulsion power and efficiencies

fuel consumption
marine growth
engine availability
service load
ALS net power saving

4.2 Results of the Study

The two vessels (reference vessel currently in operation and the new design) were then run through a
series of voyage simulations. Four round-trip itineraries in four different sea areas were simulated,
with each itinerary consisting of several port-to-port voyages. The departure and arrival times at each
port were fixed. Simulations were run for each itinerary for a 25 year period from 1993 to 2017, with
the ship leaving the departure port the next day after it arrived from the previous round trip. During a
particular voyage the speed was optimized for minimum fuel consumption over several legs of the
voyage and, where applicable, weather routing was used. A lot of information was generated as a
result of the simulations, including:

speed, engine mode and power profiles

voyage buffer times and late arrival times
fuel consumption, energy consumption, CO2 emissions and EEOI
experienced levels of passenger comfort and seasickness
experienced weather conditions
wind, wave and current contributions to the ship powering problem

One of the main conclusions from the simulation process was that the current selection of engines for

36
the new vessel appeared to be insufficient when it came to the total installed power. Several late
arrivals on one of the itineraries were observed in the simulations for the new vessel (but not for the
reference vessel). These occurred when there was a high level of marine growth or when one or more
engines were out of service in combination with moderate or high marine growth. The installed power
of the engines was, however, well in excess of the required power according to the design speed-
power curve, the expected service load and the traditional 15% sea margin. In addition the selection of
engines appeared to offer less possibility for optimizing fuel consumption than in the reference vessel.
This was because all the engines were the same in the new ship while the reference vessel had engines
of two different sizes, allowing many more combinations of engines and more flexibility to adapt to
the variations in power demand encountered. These conclusions helped support the decision to
increase the installed power and change the size and configuration of the engines used.

The most interesting part of the simulations from our point of view, however, was that these
conclusions were only reached because the measured propulsion power, marine growth effect and
engine availability statistics were included in the analysis.

5. Conclusions

The use of voyage simulations coupled with data from performance monitoring systems has for us
proven to be a valuable tool in assessing the expected performance of a new vessel design. Voyage
simulations allow the study of the energy efficiency of the vessels to be treated in a more
comprehensive and realistic manner than what is traditionally done in the ship design process. They
also help to remove doubt about the assumptions used in the analysis because they essentially mimic
the way the vessel will be operated. Finally, they can bring together all of the essential data needed in
studying the performance of the vessel and eliminate a lot of work and guessing in managing and
connecting different sources of information.

The key to an accurate assessment of is a realistic performance model.

Operational data collected by performance monitoring systems is the best information which can be
used to develop these models. This information should be used as much as possible in the design
stage, preferably in conjunction with voyage simulations, to really understand how new ship designs
will perform in operation.

37
 Performance Monitoring Insight at MOL
Tetsuro Ashida, Mitsui O.S.K. Line, Tokyo/Japan, [Link]@[Link]
Naoki Mizutani, NAPA, Kobe/Japan, [Link]@[Link]
Pekka Pakkanen, NAPA, Helsinki/Finland, [Link]@[Link]
Arttu Väisänen, NAPA, Helsinki/Finland, [Link]@[Link]
Teemu Manderbacka, NAPA, Helsinki/Finland, [Link]@[Link]

Abstract

This paper describes a process for analyzing deviations and trends of ship performance by combining
open data- Mitsui O.S.K. Line (MOL), as all commercial
operators, has collected departure, noon and arrival reports from all the ships in their fleet. There is a
massive volume of data. But it also requires lots of effort to make use of these data, for further ship,
fleet performance analysis. NAPA has introduced a system where algorithms automatically, and
constantly, estimates the fuel consumption of each day for all vessels in the world, and compares the
noon-
performance and operational performance by identifying which vessels are underperforming compared
to benchmark and which vessels could use maintenance, and how effectively past voyages were
operated.

1. Introduction

In recent years due to increase in regulations and high prices of crude oil, have many companies started
developing fuel and performance monitoring systems for cost saving. Most of these developed systems
are onboard systems that require a costly hardware installation that can be a problem for some charters
due to the charter party contracts.

Freighting business consists usually of three different types of charter party contract models. Models
are voyage-, time- and bareboat charters. These models differ from each other in how the costs are
distributed. Costs of a ship consist of capital investment made when ship is ordered and built, operating
costs which include supplies, maintenance, crew salaries, insurances etc. and voyage costs that consists
of fuel, port and canal fees. In voyage chartering ship owner pays all the above costs. In time chartering
ship owner pays capital cost and operating costs and charterer pays voyage costs and hire rate to the
ship owner. In bareboat chartering, the ship owner pays the capital cost of the ship and rents the ship to
charterer who pays all other costs, Stopford (2009).

On voyage and relatively short time charter cases charterers might hesitate to install costly system to
their vessels. Also, contract might prohibit it. This creates a need for flexible systems that can provide
estimates of performance without costly installations and hardware.

MOL operates around two hundred bulk carriers of which more than half are chartered in. Monitoring
and predicting the changes in the technical performance is relevant from charter party point of view and
as it has a large effect on whether the operated spot voyages are profitable or not. However, as these
vessels are typically chartered for 1-2 years and consume approximately 15-40 tons of fuel per day, it
does not make commercially sense to install torque or fuel flow sensors with high frequency data
collection.

So far MOL has been using their internal engineering department for analyzing
noon reports, but there have been some downsides in the approach. For example, the only data source
has been the noon reports and therefore it has taken a long time to form a baseline consumption for a
vessel. Sometimes the baseline can also be false or inaccurate due to false input data, either by human
accident or on purpose, Aldous et al. (2013), Safaei et al. (2018). In addition, the data frequency in noon
report is only one data point per day, which makes any wind, wave, sea current corrections difficult or

38
in many cases even meaningless, Aldous et al. (2015), Mannheim (2017). Attempts to overcome the
challenges of relying purely on noon-report data are made by improving the filtering techniques or
introducing alternative methods for the analysis, (2016, Davies and Bevan (2017).

In this article we will present how a reference value for the daily fuel consumption is calculated in a
service called NAPA Fleet Intelligence. Calculation is based on high frequency location data from
Automatic Identification System (AIS), weather nowcasts and vessel performance models, and the
article describes what type of benefits can be achieved with comparing the reported values to the results
of the simulations.

2. Challenges of noon-report data

Noon-report is once a day crew made report. It contains a snapshot of the ships daily performance and
it contains information on: consumption, location, course, speed, shaft rotational speed (RPM), wind
speed, loading conditions (laden or ballast) and timestamp.

The challenge of applying noon-report data lies in the sparse resolution (once per day) and on possible
inaccuracies or even human errors since the data is collected using various methods and typed in
manually by the crew members. The available sensor readings are dependent on the level of
instrumentation on board, which generally varies between the ship types. More expensive ships, such
as cruise vessels, ropax or even container and LNG ships can be equipped with fuel flow meters and
fixed tank level sensors connected with onboard systems providing better accuracy of consumption
monitoring. Whereas tankers and dry bulk vessels might not have such sophisticated installations. In
these cases, the crew must take the reading of fuel remaining on board applying a manual level sensor.
The manual readings are affected by sea state, causing motions and thus disturbing the accurate reading.
Moreover, the trim and possible heel might not be taken properly into account when calculating the
amount of the fluid from the level reading. More on the challenges of measuring quantities of fuel on
board can be found e.g. in Pecci (2007) related to liquid bulk cargo measurements.

Consumption is a key variable and if we look closer to it we can see that, the available data on the
consumption depends on the ship type and the quality and the extent of the systems the ship is equipped
with. Some ships have level sensors installed in all fuel tanks and the readings are connected to systems
like NAPA Loading Computer, which gives the exact quantity of fluid in the fuel tank based on the
level sensor reading and geometry of the fuel tank.

Fig.1: Wind speed distribution of case vessel

39
Also other reported variables suffer from varying quality. Sometimes this is caused by unclear
instructions on what to report, sometimes human mistakes, sometime the unconscious habit of reporting
towards what is expected rather than what is happening. Below example of a Fig.1 shows a case where
wind speed close to 5 Beaufort are commonly reported as 5 Beaufort.

In addition to being affected by human error, the sparseness of the reported average values causes
significant error if utilized solely. For example, only one value for the wind speed is reported as e.g.
14kn / 335° or sometimes even just 4 BF. However, there are commonly significant variations the
environmental conditions (force, direction) during a day and wave height is either not reported at all or
is based on visual observations and thus prone to bias based on observer. Approximately on 30% of
operational days the wind speed changes over 5 m/s or wind wave height more than 1 m.

Fig.2 Variation in weather during a day

The impacts of weather and wave conditions are not linear, thus using the average value, even if it could
be 100% correct, would be problematic. As seen on Fig.3, variations in only wind direction on 4BF
conditions can decrease vessel speed by 15%, or if maintaining a fixed speed, to increase consumption
by 50%. Thus, using only one average speed and average weather conditions value for the 24h period
is clearly inadequate.

Fig.3 Variations in performance in 4 BF conditions

3. Methodology in this study

The estimation of fouling is based on comparison between the reported consumption and calculated

40
reference consumption. The reference consumption is calculated for all the days where the noon-reports
are obtained. The reference consumption calculation is based on the full hydrodynamic model of the
ship including propulsion and engine arrangement. The calculation of reference consumption accounts
for the actual weather the ship encountered from noon to noon, including the weather variations. Since
the noon-reports obtained from the ships contain manual input, they are filtered, and data is prepared
for analysis of level of fouling. The calculation of reference consumption, the preparation of noon-
report data, and the estimation of fouling is described in detail in the following subchapters.

3.1. Calculation of reference consumption in NAPA Fleet Intelligence

High frequency ship location data is used to calculate reference consumption. Ship location, speed and
heading are obtained from the AIS messages, which are collected in few minute intervals. Besides the
location and other information, the AIS message contains information on the draft of the ship. AIS
location is based on the Global Navigation Satellite System (GNSS), which usually is Global
Positioning System (GPS), these signals can be considered reliable. The draft is included in the AIS
message as voyage-related information and is manually entered by the crew and may be inaccurate,
Adland et al. (2017).

Global ocean weather now-cast data from independent weather forecast provider is used to obtain the
weather at the ship location, including; wave, wind and sea currents. The weather now-casts are
provided at spatial resolution of 1.25°, which corresponds to a grid size of approximately 100 km. The
weather is updated every 180 minutes. To accurately match the vessel timestamp and position, the now-
casts are interpolated to the ship position at each time instant the position is obtained. This is done by
trilinear spatio-temporal interpolation, which is described in Haranen et al. (2017). Wave conditions
are given as two main wave directions, namely swell and wind waves. For each of these the significant
wave height, zero crossing period and direction is given. Wave conditions are based on the data from
WAVEWATCH III (WW3) model, which is developed by National Centers for Environmental
Prediction (NCEP). Also, the wind the speed with direction, as well as speed and direction of sea current
including the effect of tide are given.

The performance of the ship is estimated with first principle hydrodynamics calculation methods based
on the ship location, weather data and generic ship specific performance model. The performance model
constitutes of; a generic model of ship hull (i.e. lines drawing), which is formed based on the available
information of the main characteristics of the ship, a model of ship propulsion and maneuvering
arrangement, and a model of the ship powering i.e. installed engine model. Essentially, a digital twin
of the ship is produced, as accurately as possible, with the available ship data. The level of the data
availability depends on the use of the model, for public use of calculation results, publicly available
data is applied. Whereas more accurate data of the ship provided e.g. by the ship-owner, can be applied
for the calculations reserved for the ship-

Calculation of consumption based on the publicly available data is done by calculating the resistance
of the ship in the given operational conditions: speed, loading condition, water depth, wave and wind
conditions. All the forces acting on the ship are calculated with hydrodynamics models. Quartering
waves and wind cause drifting angle to ship s propagation, which need to be balanced by the rudder.
The rudder forces are also included into the model together with the hydrodynamic coefficients
accounting for the additional resistance due to the drift angle. Factoring in all the above-mentioned
forces, the required thrust to propel the ship at given speed is calculated by solving the force balance.
The thrust is calculated considering the propulsion arrangement; propeller diameter, pitch ratio, thrust
deduction, and wake factor. Then the required propeller revolutions per minute (rpm) and corresponding
required power from the main engine are calculated, and finally, based on these, the consumption is
calculated. The calculated consumption will be used as reference consumption to compare with the
reported consumption.

Similar approaches have been taken by Jalkanen et al. (2012) for estimating the Green House Gas
emissions from ships. These studies have taken the effect of weather conditions as average sea margin,

41
which is a common way in the ship design, however that type of approach can only approximate long-
time average consumption and is not suitable for our purpose. The approach taken here calculates the
actual consumption in the actual weather conditions and can be deemed as the most advanced one.

Fig.4 Calculation of required propulsion power for each timestamp

3.2. Noon-report consumption

In this research we used noon-report data provided by MOL. Ships were bulk carriers of size range 30
000-230 000 dead weight tonnage (DWT). Data collection period was approximately 3 years for all
ships and altogether there were several thousand points of operational data.

Data filtering and preparation were carried out for both datasets, NAPA Fleet Intelligence data and
Noon-report data. NAPA Fleet intelligence data sampling frequency depends on the availability of AIS
data and is typically once per hour while at sea. NAPA Fleet Intelligence data was averaged to once per
day sample. For Noon-report data a time correction was applied for days that was 23 or 25 hours long.
After this were the obvious outliers filtered out of the data.

From the relation of noon-reports load and rpm a variation in rpm can be observed with constant load
values. This variation can be assumed as pseudo weather impact to the ship. This can be justified
because power is rpm times torque and load is power divided by the maximum continues rating. By
combining these two we get equation represented below.

(1)

From this we can see that if load is constant and rpm changes the torque needs to change also. Relation
of noon-report s load and rpm is presented in Fig.5 and from this we can see that rpm varies so we can
assume that this is due to weather.

Because of this variation a new continuous variable that represents weather can be created. A fit was
created between noon-report rpm and engine load. With this fit we calculated the residual of measured
and predicted rpm. This residual was used to make the continuous variable representing weather
condition. Weather variable was named pseudo weather. With new pseudo weather, rpm, consumption
and loading condition a model was made to predict consumption. Model is presented in Fig.6.

This model is used for filtering out abnormal data points from noon-report data. This is done so that a
consumption prediction is made of noon-report data and the residual is calculated from this prediction
and the measured consumption. After all residuals have been calculated 5% of all smallest and largest

42
points are filtered out.

Fig.5: Noon-reports Rpm vs Load

Fig.6: Model

3.3. Fouling estimation

The reported vessel speed is normally an average speed, and similarly for the wave height, if reported,
and for wind speed. However, the consumption, or in this case the speed reduction has non-linear
dependency on the wave height and wind speed. For this reason, the average values cannot be used to
represent the daily speed-power-or-consumption figure to estimate the level of fouling.

Usually, noon reports with harsh weather conditions are filtered out and not used, or these effects are

43
intended to be factored out, by calculating separately the added resistance of waves and wind etc.
However, the challenge is that these effects do not superpose, instead they affect altogether to the
resistance and consumption. For instance, side wind causes drift, which introduces additional resistance
through counter maneuvering of the rudder and changes the propeller efficiency and so on. Thus, we
will use the performance model described earlier and calculate the performance in each available
timestamp.

Fig.7: A day of a bulker on voyage from Australia to Japan

In above example we can see how during one day the significant wave height decreased from 1.3 m and
at the same time the vessel speed increased from 11.4 kn to 13.6 kn, while the average speed obtained
from noon-report shows only one value, 12.3kn. In the noon report also the daily consumption, 24.7
metric tons, has been reported. NAPA Fleet Intelligence calculates the reference fuel consumption for
each moment of the day and we can then calculate the average consumption, 21.5mt. From this
difference we can calculate, as an example value, the effect of hull fouling as 24.7mt 21.5mt = 3.2mt,
shown as a red dot in Fig.8.

Similar calculations are performed on all available data points, and as a result we can get a time series

speed FOC power curves of the vessel.

Fouling estimate is done by subtracting the NAPA Fleet Intelligence consumption from noon-report
consumption and then fitting a trend to this data. In Fig.9 is presented a simple normal least square fit.
Obvious outliers are marked with ovals.

We can deal with these outliers by making a robust fit using iteratively reweighted least squares. This

44
fit is presented in Fig.10. The smaller the weight the less it affects the fit, Huber (2009).

Fig.8: All noon-report and reference calculation based datapoints of increased consumption due to
fouling

Fig.9: Normal fit

4. Results

The described calculations are implemented in a manner which is easily reproducible for any number
or types of vessels. In this study, the main focus is on the methodology and proof of concept using a set
of small to medium size bulk carriers operated by MOL.

4.1. Typical changes in the performance

The calculation process allows to easily quantify the decreasing performance of the vessels in the fleet

45
of interest, which can be then used for decision making regarding for example timing of dry docking or
revising the charter party consumption.

Fig.10: Robust fit

This dataset contained slow steaming bulk carriers. A typical annual increase in consumption on a fixed
speed was 7-9%, which is in line with general understanding. However, the variation in performance
changes was huge; on some vessels the performance remained stable for the whole monitoring period,
on some the performance deviated substantially and the effect of maintenances was very large. This
shows that in addition to general understanding of typical scenarios, a constant vessels specific analysis
is beneficial.

The maintenance periods of the case fleet varied and there was from zero to two maintenances during
this three-year period. The typical benefit of the maintenance was slightly over 10% reduction in fuel
consumption.

4.2. Uncertainly of the model prediction

In addition to the performance model inaccuracies, the difference between the reported and predicted
values are caused by for example weather forecast inaccuracy, AIS data availability and erroneous input
by the crew, either accidental or even purposeful. Thus, as the true value of the consumption is not fully
known, we do not talk about error of the model, but rather uncertainty of the prediction. In these
comparisons we are not filtering out any harsh conditions, days where half of the time is spent for
drifting etc.

In Fig.11, we estimated the uncertainty of the daily hull fouling rate value with an iterative process with
certain 95% confidence limits. In other words, with this, we can say that the fouling is at the calculated
level and with 95% confidence it is within the confidence limits. For example, after just few weeks of
operation after the maintenance on autumn 2016, we could state with 95% confidence that that the
vessel daily consumption had improved by 5+1mt / day. With more data, the limits tend to narrow down
and the accuracy improve.

To demonstrate the improvement of uncertainty level, we have calculated the confidence limits after n-
days of data collection after the maintenance of the hull for all vessels utilized in this study. Average

46
value of confidence limits as percentages of average consumption, is shown in Fig.12 as a function of
number of days the data is collected.

Fig.11: Uncertainty of the hull fouling rate

Fig.12: Daily error vs data points curve

The above means that after using all the available data, with 95% confidence the predicted daily fuel
consumption values have less than 3% difference to the reported ones. This difference is coming from
the typical inaccuracies in the noon report readings, from small inaccuracies in the weather now-cast
data and from periodic longer gaps in AIS data.

In this comparison the model uncertainty started quite poor, around 15%, due to large differences in
technical performance of bulkers, but rapidly improves. With this method, we have around 5-6%
uncertainty after 3-6 months, which is enough for large majority of use cases of ship performance
monitoring and analysis. For reference, commonly the model training on purely machine learning based
services working with high frequency automation data take from three to six months for bulk carriers.

When comparing longer periods, for example voyages, the uncertainty level would be lower as some
of the inaccuracies in for example weather forecasts cancel each other during different days. This is
more meaningful comparison for voyage planning purposes and will be estimated on separate studies.

47
4.3. Estimate stability of the hull fouling trend prediction

To estimate the near future, 3-12 months, performance for a vessel, it is important to be able to estimate
the current hull fouling rate of the vessel. The stability of the fouling rate is done by taking two points
of data from the start of estimation period and then fitting a line on them. After this the slope is stored
in memory and a data point is added and process repeated until whole period is processed. Previous
slope data point is subtracted from the following slope data point to get the change size of fits slope.
This is done to all periods of all ships to get the slope development vs time curve. In Fig.13 is presented
all the data of the periods slope stability relative to the amount of data.

Fig.13: Slope stability vs data points

The above chart tells us that during the first days of operation the variation in the fouling rate prediction
is huge. The estimation of the hull fouling trend however stabilizes fast and takes only a few voyages
since the data collection start to be able to have some indication of the performance trend. After 50 days
there is hardly any variation in the performance trend prediction.

5. Conclusions

In the paper methodology for using noon-reports as a source of information to assess the technical
performance was presented. A digital twin of the ship is created based on the naval architecture
principles, hydrodynamic models, and experience in ship design. With the proposed methodology;
public data, hydrodynamic models and manually reported data are combined into NAPA Performance
Model i.e. the digital twin. The methods are implemented in the NAPA Fleet Intelligence system. The
challenge of sparse data, one daily data point, obtained form the noon-reports, is overcome with the
combination of hydrodynamics and big-data based model. The reference consumption is calculated with
these models and compared with the reported one. This enables improvement of the absolute accuracy
of the NAPA Fleet Intelligence performance models without onboard installations.

The accuracy of the presented method was studied based on noon-reports over a several years from fleet
of dry bulk vessels. The accuracy was shown to improve in good level rapidly with the increasing data.

These models can be used for evaluating or predicting the need for dry docking or other maintenance,
for planning of forthcoming schedules and negotiation of contracts, for weather and speed profile

48
optimization or for several other use cases of which the shipping companies can gain benefit from.
Discussed methodology is applicable for small or large fleets without large investments. The service
can be started with significantly less interference to ship operations and significantly lower investment
than high frequency data collection systems and can enable access to fleet wide analytics in a matter of
some weeks or months.

References

ADLAND, R.; JIA, H.; STRANDENES, S.P. (2017), Are AIS-based trade volume estimates reliable?
The case of crude oil exports, Maritime Policy and Management 44 (5), pp.657-665

ALDOUS, L.; SMITH, T.; BUCKNALL, R. (2013), Noon report data uncertainty, Low Carbon ship-
ping Conf., London

ALDOUS, L.; SMITH, T.; BUCKNALL, R.; THOMPSON, P. (2015), Uncertainty analysis in ship
performance monitoring, Ocean Eng. 110, pp.29-38,

.; ARSLANA, O.; TURANB, O.; ÖLÇERC, A.I. (2016),

, Computers & Operations Research
66, pp.393-401

DAVIES, H.; BEVAN, S. (2017), A Consultative Approach to Charter Party Agreements Based on
Virtual on Time Arrival, TRANSNAV the Int. J. Marine Navigation and Safety of Sea Transp. 11(2)

HARANEN, M.; MYÖHÄNEN, S.; CRISTEA, D. S. (2017), The Role of Accurate Now-Cast Data in
Ship Efficiency Analysis, 2nd Hull Performance & Insight Conference (HullPIC), Ulrichshusen, pp.25-
38, [Link]

HUBER, P.J. (2009), Robust Statistics, John Wiley & Sons

JALKANEN, J.P.; JOHANSSON, L.; KUKKONEN, J.; BRINK, A.; KALLI, J.; STIPA, T. (2012),
emissions for particulate matter and carbon
monoxide, Atmos. Chem. Phys. 12, pp.2641-2659

MANNHEIM, D. (2017), Application of AIS Data in Vessel Performance Analysis, Master Thesis, DTU
and Aalto University, [Link]

PECCI, J. (2007), Liquid Bulk / Quantity Measurements and other challenges, SafeWaters UM,
[Link]

SAFAEI, A.A.; GHASSEMI, H.; GHIASI, M.; (2018),

Data Using Statistical and Data Mining Methods, European Transport \ Trasporti Europei 67/7

STOPFORD, M. (2009), Maritime Economics, Taylor & Francis

49
 An Approach to Monitor the Propeller Separately from the Hull
Erik van Ballegooijen, VAF Instruments, Dordrecht/Netherlands, evballegooijen@[Link]
Torben Helsloot, VAF Instruments, Dordrecht/Netherlands, thelsloot@[Link]

Abstract

This paper describes an approach to monitor hull performance separately from propeller performance.
Use is made of thrust measurements in addition to power, speed and environmental variables. With this
data the extra propulsive power required due to fouling and roughness can be ascribed to either the
hull, or to the propeller. Without thrust measurements, making such a division would be impossible.
Instead of estimating the exact interaction effects, which are difficult to determine in full scale,
are defined that are easy to interpret and make a pragmatic division between efficiency losses related
to the hull and losses related to the propeller. By
to changes in efficiency and make informed decisions regarding the maintenance of their ships. A case
study is presented to demonstrate the usefulness of this approach.

1. Introduction

Ships in service regularly undergo cleanings of both their hulls and their propeller(s). This is done to
keep every surface of the ship that is in contact with the water smooth, so that friction decreases and
the overall hydrodynamic efficiency increases. Ultimately this saves money, because less fuel needs to
be bought to propel a more efficient ship. However, hull cleaning and propeller polishing costs money
as well, so there is a tradeoff to be made between maintenance costs and costs incurred by hull and
propeller fouling.

Finding the optimal times to schedule a maintenance is impossible without an accurate quantification
of the costs of hull fouling and propeller roughness. Poor knowledge of the hydrodynamic performance
of a ship is therefore costly, which is why more and more ship owners are investing in tools that will
give them the insight that is needed. Sensor data can help provide this insight. With accurate
measurements of the power consumption over time the gradual increase in resistance caused by fouling
can be detected. Data alone however, does not paint a clear picture. Sensor data of ships in service is
extremely complex and noisy, which means a careful analysis is needed to arrive at a performance
indicator that can be readily understood and acted upon.

This paper discusses a set of such can be used to quantify the costs of propeller roughness
and hull fouling.
ship, one to indicate hull fouling, and one to indicate propeller roughness. To do this, the performance
of the hull and the propeller must be separated.

2. Separating hull and propeller

Much has been written about separating the efficiencies of propeller and hull, but the consensus is that
thrust measurements are a necessity, see ISO19030-1. A promising approach to separate hull and
propeller efficiency has been presented by Paereli et al. (2016), Paereli and Levantis (2017). They
identify some practical obstacles in obtaining useful performance indicators, such as the
immeasurability of the effective power PE, but are nevertheless able to get some sensible results. In the
following paragraphs a method not dissimilar to theirs is presented. Because the effective towing power
cannot be determined in full scale

With thrust T and ship speed VS. In conventional naval architecture terms, PH relates to the effective
towing power and thrust deduction factor t.

50
The other quantity that is needed for the analysis in the present paper is the shaft power PS.

The main reason for using these quantities is that they are measurable. PH can be measured with a thrust
meter, PS can be measured with a shaft power meter. With these two powers, we can already make a
separation. The intuition behind the separation is the following: some of the power that is available at
PH propeller. In the
form of an equation this reads:

To understand what the separation precisely means, the characteristics of PS and PH must be understood.
In equal sailing conditions, PS changes over time when the overall propulsive efficiency of a ship
changes, it is influenced by changes in
Hull efficiency
Open water efficiency
Relative rotative efficiency
Shaft efficiency

In equal sailing conditions, PH changes over time either when the resistance of a ship changes or when
the thrust deduction factor changes.

When only the resistance increases, both PH and PS will increase with the same percentual amount. A
five percent increase in resistance leads to five percent increase in PH and when the rest of the efficiency
stays the same a five percent increase in PS. From this it follows that when the relative increase in PS is
greater than the relative increase in PH, it must be due to something that is not hull resistance.

The approach that is proposed in this paper to separate hull and propeller effects works with a set of
three KPIs. The first KPI is defined in such a way that it tracks the percentual change in total shaft
power PS, compared to a certain baseline.

This KPI lsion power los tracks the aforementioned changes in:
Hull efficiency
Open water efficiency
Relative rotative efficiency
Shaft efficiency

The second KPI is defined in such a way that it tracks the percentual change in hull power PH, compared
to a certain baseline.

This KPI ower los tracks the changes in:

Hull resistance
Thrust deduction factor

51
The third KPI is equal simply to the difference between the other two KPIs. This results in the fact that
the third KPI tracks those effects that are included in the first KPI but are not present in the second KPI.

This KPI ower los tracks the changes in:

Wake fraction
Open water efficiency
Relative rotative efficiency
Shaft efficiency (not expected to change)

Loosely speaking, tracks the total hydrodynamic performance, tracks performance

related to the flow around the hull, and tracks performance related to the losses in the wake.
The main contributor to will be the resistance increase of the hull. The main contributor to
will be the propeller roughness, but also anything that changes the inflow to the propeller will show up
in the propeller power loss. (Changes in propeller inflow will go together with changes in ship
resistance. Draught and trim for example, will both change the inflow to the propeller, and resistance.
Also energy saving devices such as a wake equalizing duct will have an effect on both resistance and
wake.)

Granted, this approach does not perfectly separate the efficiency of the propeller from the efficiency of
the hull, which is possible only in theory. However,
understandable and ties into conventional naval architecture theory in a clear way.

Another advantage of using the proposed KPIs is that they are all the same thing, so that they can be
directly compared to each other. All three of them are expressed as a percentage of baseline power. This
percentage may for example be applied to the expected fuel costs to get an estimation of the savings
potential in dollars. Or, more accurately, the savings potential can be calculated by multiplying lost
power with specific fuel oil consumption and fuel price. The KPIs are thus more closely related to costs
than alternatives such as propeller efficiency or speed loss.

3. Case study

The following section describes a case study in which the use of the proposed method is exemplified.
The data used in this case study was also used in a previous publication, Ballegooijen et al. (2018), in
which a different set of KPIs was used. The case study uses thrust and torque measurements made with
the TT-Sense®, shown in Fig.1.

Fig.1: The TT-Sense® thrust and torque meter

52
3.1 Description of data

The ship under consideration is a large (>50,000GT) passenger vessel that underwent maintenance
where both the hull and propeller were cleaned. The data used encompasses 11 months, roughly 5.5
months before and 5.5 months after the maintenance. The data points are chosen to represent quasi-
static conditions. Filters are applied for low windspeeds, similar draft and deep water. The goal of the
case study is to estimate how effective the maintenance has been.

To account for a noticeable bias in the original speed log a correction method has been applied. The
details of this correction method can be found in Ballegooijen et al. (2018).

3.2 Creation of the baseline

In this paper simple baselines are used for the expected thrust and expected power. In practice it would
be better to use a more accurate prediction model as a baseline, or use baselines determined from model
tests. To demonstrate the proposed KPIs however it suffices to use baselines that are simply a power
law fit of the data after the cleanings.

Where aP and bP were determined using least squares estimation. Similarly:

Where aT and bT were again determined using least squares estimation.

3.3 Visualisation of results

Fig.2 shows the set of KPIs computed using the simple power law relations as a baseline. The propulsion
KPI shows the percentage of power lost within the total propulsion chain. The Hull KPI shows the
percentage of power lost that can primarily be ascribed to effects related to the hull.

Fig.2

53
The Propeller KPI shows the percentage of power lost that can be ascribed to losses in the propeller
wake. (Losses in the propeller wake are a result of both the innate (in)efficiency of the propeller and
extra losses relating to the quality of the inflow in behind condition.) Please note the y-axis of the bottom
subplot is scaled differently.

The data shown can be summarized in two different ways. The first summary is shown in Fig.2 and
Table 1, and is based on the average power loss before and after the cleanings. The individual data
points show variance due to several reasons. Most causes of variation are expected to average out in the
long run. Comparing the averages before and after a maintenance will therefore give a reliable estimate
of the maintenance effects. Table 1 displays the average power loss before and after the cleaning event.
The estimate says that six percent more power was needed in the period before the cleaning.

Table 1: The effects of the cleaning event on lost power using averages
Mean Power loss Before Cleaning After Cleaning Average Effect
Propulsion losses 6.1% 0.1% 6.0%
Hull losses 4.5% 0.1% 4.4%
Propeller losses 1.6% 0.0% 1.6 %

Fig.3

Table 2: The effects of the cleaning event on power loss using linear trends
Linear Power loss Before Cleaning After Cleaning Instant Effect
Propulsion losses 4.7% 7.2% -3.4% 5.5% 10.6 %
Hull losses 3.2% 5.6% -2.9% 4.8% 8.5 %
Propeller losses 1.5% 1.6% -0.4% 0.7% 2.0 %

Fig.3 shows the same data as Fig.2, but this time the data summaries are linear trends rather than
averages. The linear trendlines reflect the expectation that hull fouling and propeller roughness
gradually increase over time. Using the endpoints of the trendlines to estimate the cleaning effect yields
larger estimates of the cleaning effect, as shown in Table 2. Whereas the estimate in Table 1 says
something about the average effect of the cleaning, Table 2 says something about the instantaneous

54
decrease in power demand. The difference between them arises because the fouling rates of the two
time periods are not the same.

Assuming that the average specific fuel oil consumption (SFOC) of both time periods is
approximately equal it follows from the presented analysis that this maintenance has yielded around ten
percent instantaneous reduction of fuel consumption, of which roughly eight percent was due to a
decrease in resistance and roughly two percent due to improvements of the propeller efficiency and
wake. Because the fouling rate was higher in the second period the average reduction of fuel
consumption was roughly six percent, of which four-and-a-half percent came from the decrease in hull
resistance, and one-and-a-half percent from improvements in propeller efficiency and wake. These
numbers can be used to estimate the return on investment of similar cleanings.

Figure 3 also shows that the amount of power loss related to the hull is approaching the value it had
before the cleanings, which indicates that it might be a good idea to schedule a new cleaning in the
coming months. The power loss related to the propeller has increased somewhat, but is not on the same
level as it used to be, which could lead to the decision to not invest in a propeller cleaning yet.

4. Conclusions

Through the use of thrust measurements a pragmatic separation between performance degradation of
the hull and the propeller can be made. One way to do this is to use the three KPIs that are introduced
in this paper, each of which reflects the percentage of extra power needed to propel a ship with respect
to a predefined baseline.

The first KPI, power loss , is affected by changes in the total propulsive efficiency of a
ship. Secondly, hull power loss is affected by changes in ship resistance and thrust deduction factor.
Lastly, a change in propeller power loss reflects changes in propeller efficiency and wake fraction.

When all other effects are properly accounted for, increasing power loss will reflect increases in
propeller roughness and hull fouling. As such, the power loss KPIs can be used to aid scheduling of
hull cleaning and propeller polishing. In the presented case study an average and instantaneous effect
on the power consumption after a cleaning was determined for both the hull and the propeller. This
makes it possible to determine the return on investment of the cleaning.

References

BALLEGOOIJEN, W.; HELSLOOT, T.; TIMMER, M.; (2018), The propeller as a speedlog: using full
scale thrust measurements to quantify propeller efficiency and hull fouling, RINA Full Scale Ship Per-
formance, London

PAERELI, S.; KRAPP, A.; BERTRAM, V.; (2016), Splitting propeller performance from hull perfor-
mance A challenge, 1st HullPIC Conf., Pavone

PAERELI S.; LEVANTIS M., Quantification of Changes in Propeller Performance Method Valida-
tion, 2nd HullPIC 2017 Conf., Ulrichshusen

ISO 19030-1 (2016), Ships and marine technology - Measurement of changes in hull and propeller
Performance - Part 1: General principles, ISO, Geneva

55
 Evaluating the Efficiency of an Energy-Saving Device by Performance
Monitoring
Nikos Themelis, Prisma Electronics, Greece, [Link]@[Link]
Christos Spandonidis, Prisma Electronics, Greece, [Link]@[Link]
Christos Giordamlis, Prisma Electronics, Greece, christos@[Link]

Abstract

In the current paper, we utilize a performance monitoring system in order to examine the fuel savings
for an oil tanker in which an energy saving device (mewis type duct) had installed during dry-dock.
Three years of operational data utilizing the LAROS platform for the signal processing, data collection
and analysis have been used, covering a period before and after the installation of the duct. The
propulsion performance is monitored through several KPIs, whose sensitivity is tested for previous
events such as propeller polishing and cleaning. Furthermore, we carried out a comparative study on
a sister vessel without such energy saving device, but having similar hull and propeller cleaning history
in order to access the long-term performance assessment.

1. Introduction and aim of the study

Nowadays the shipping industry is seeking ways to improve the energy efficiency of ships either for
reasons related with cost savings, regulatory compliance or environmental protection. Several design
and operational solutions have been proposed. Bouman et al. (2017) provides a detailed presentation of
studies quantifying the energy reduction potential provided by several technological advances targeting
for example optimizations in hull, power and propulsion as well as route planning. For existing ships,
an option to increase their energy efficiency is by implementing devices aiming at their propulsion
improvement. Such a device can be a mewis type duct mounted in front of a propeller targeting the
enhancement of its working conditions and specifically by accelerating and straighten the hull wake
into the propeller as well as by producing a net forward thrust. The installation includes also a fin system
which provides a pre-swirl to the ship wake which reduces losses in the propeller slipstream and thus
resulting in an increase in propeller thrust at a given propulsive power. Model and CFD tests are usually
carried out in order to properly design and quantify the benefit from such an installment (see
Schneekluth and Bertram (1998). Nevertheless, for several reasons, operators still need to measure and
verify the claimed gains during the real operation of a ship when conditions can differ extensively from
the design ones.

On the other hand, performance monitoring can be utilized for a variety of reasons, spanning from
situational awareness and optimization, triggering maintenance events as well as evaluating the effect
of technological interventions such as dry-dock (see e.g. Themelis et al. (2018), while for several case
studies see Bertram (2018)). Performance assessment of ships is based on the wealth of data produced
, while the shipping industry has started
to adopt frameworks such as ISO19030, ISO (2016), aiming at standardizing the assessment procedures
in a comparable manner.

In the current paper, we utilize a performance monitoring system in order to quantify the fuel savings
for an oil tanker fitted with a mewis type duct during dry-dock. Our aim is to examine whether a set of
Key Performance Indicators (KPIs) are capable of capturing the effect of the duct in ship performance.
Three years of operational data employing the LAROS platform for the data collection and analysis will
be utilized, covering a period before and after the installation of the duct. The same set of KPIs will be
tested for their sensitivity for changes in performance due to dry-dock and propeller polishing and
cleaning. The scope of the analysis is to study whether a practical index could be used before one shall
proceed with a more detailed analysis involving methods for the decomposition of hull and propeller
performance (see for example Logan (2011), Grigoropoulos and Theodosiou (2012)).

56
2. The Performance monitoring framework

A practical, yet comprehensive approach is to monitor the performance through using specific metrics,
the so-called Key Performance Indicators (KPIs). Performance indicators may be characterized by the
following aspects:

The principal components they describe; hull, propeller and /or engine condition
The ability of the indicator to account for changes in speed, environmental conditions, displace-
ment, etc.
The physical representation of the indicator; certain parameters are more intuitive and easy to
interpret than others
Ability to identify slow time-varying performance changes, e.g. due to fouling.

A general framework for the calculation of a KPI is shown in Fig.1, where it is pointed out that the
required step for data pre-processing.

Fig.1: A general framework for KPI calculation.

The set of KPIs to be used in this study is defined as follows:

KPIa = % power increase, following the concept of ISO 19030, ISO (2016), and targeting hull
and propeller performance
KPIb = ME FOC / d, in 1 hr. ME FOC and d are the fuel oil consumed in main engine and the
distance travelled respectively. This KPI includes also engine efficiency
KPIc = P /n3, where P and n are the shaft power and revolutions respectively. This KPI corre-
sponds to propeller curve coefficient or else power ratio, e.g. Logan (2011)

The % power increase KPI is defined as:

(%) (Pm-Pe)/Pe

It expresses
Vm. The expected power Pe is read from a speed-power reference curve (e.g. obtained by sea trials) at
the measured speed through water (Vm) and at the measured displacement and trim as shown in Fig.2.
This KPI follows the filtering requirements as described in Part 2 of ISO 19030 (ISO 2016).
Specifically, the next parameters shall be considered for filtering:

A minimum water depth is defined as h=min [3(B*Tm)1/2, 2.75 Vs2/g], where B, Tm, and Vs are

rudder angle larger than 50

true wind speed values larger than 7.9 m/s

57
Additionally, we exclude data points that correspond to current speed greater than 1 kn. Furthermore,
no extrapolation of speed-power curves is allowed, thus we utilize data only for the speed range of the
reference curves. Furthermore, we are not considering measurements that corresponds to values of
displacement a BP% from the respective values of the reference
conditions. A study on the levels of confidence achieved due to uncertainty issues spanning this
performance indicator, and mainly due to sampling frequency, has been presented in Themelis et al.
(2018).

Fig.2: Expected and measured power for a fouled ship.

3. Set up of the case study

The examined ships are two sister Suezmax crude oil tankers with summer deadweight of 158000 mt.
High-frequency data with sampling frequency of 1 minute are collected from the ships utilizing the
LAROS platform (see Spandonidis et al. 2018). As we will rely on the data collected for the evaluation
of the energy saving device, our aim is to consider a time period covering a period before and after the
installation of the duct which is placed during dry-dock. Specifically, the available operational data
cover a period 3 years. Table 1 shows the key dates for the 2 ships. Due to the fact that the duct was
installed during dry-dock, where hull and propeller cleaning were also taken place, the introduction to
the analysis of the 2nd sister ship will provide the benchmark for comparison spanning the period after
dry-docking. We also have to notice that Ship 1 undergone propeller polishing and cleaning a few
months before her scheduled dry dock, while such event did not occur for Ship 2, at least during the
monitoring period. Among the whole set of signals monitored using the LAROS platform,

Table 2 presents the required parameters for the study.

Table 1: Key dates (month/year) for performance intervention.

Ship Mewis Duct Dry-Dock Propeller Start date of End date of
installed (DD) polishing and monitoring monitoring
cleaning (PP) period period
Ship 1 1/15 1/15 9/14 6/14 6/17

Ship 2 No 4/15 No 6/14 6/17

Fig.3 to Fig.5 present indicative distributions of operational and weather parameters. Current speed is
calculated using speed through water and speed over ground. A pre-processing stage was carried out
targeting the filtering of data based on the next criteria as well as the normalization of power values for
the loading conditions that correspond to ballast and design drafts (see Table 3) using the Admiralty
formula. Correction of power values due to wind resistance was carried out using the method proposed
by ISO 19030-Part 2 (ISO 2016).

58
 Table 2: Set of monitoring parameters.
Parameter Units Sensor type
Speed over ground knots GPS
Speed through water knots Speed log
Shaft revolutions Revolutions Shaft torque meter
per minute
Shaft power kW Shaft torque meter
ME fuel oil consumption Tons/day Mass flow meter
Drafts m Draft indicator
Ship direction degrees GPS
Rudder angle degrees autopilot
Relative wind speed and direction m/s, degrees anemometer
Water depth m Echo sounder

Table 3: Characteristics of the reference loading conditions.

Loading condition Mean draft (m) Trim (m)
Ballast 7.3 -2.1
Laden @ design draft 15.5 0

Fig.3: Probability density functions of speed through water, true wind and current speed.

59
 Fig.4: Mean draft versus trim. Positive trim values correspond to trim by the bow.

Fig.5: Contours of the joint probability density function of

correspond to the lower values of the pdf.

4. Analysis based on KPIs

Distance travelled
the power - speed curves were obtained by utilising the data of 3 months of operation after the dry-dock
and using the form P = a Vb. The parameters a and b are calculated by the linear least square method.
Fig.6 presents the obtained reference curves for the two loading conditions for Ship 1.

Fig.6: New baselines obtained from operational data for Ship 1. R2 values are 0.9951 and 0.9949
respectively.

For the calculation of the KPIs, we use the filtered and normalized dataset as obtained by applying the

60
procedure described in sections 2 and 3. In the first instance, KPIs are derived for the reference loading
condition (ballast and laden). Such results are shown in Fig.7, while KPI values, including both loading
conditions, are shown in Fig.8 where trend lines have been included for each period.

Both 3 KPIs present significant large values for Ship 1 before her propeller polished and cleaned,
revealing inefficient ship operations. After this event there was a significant decrease of the KPIs.
Additionally, after dry-docking, the KPIs of Ship 1 were slightly decreased compared with the
respective values before dry-docking nd
line reveal a slight increase for the next 2.5 years of operation following dry-dock, which will be
examined later.

On the other hand, the condition of Ship 2 before dry-docking was better than those of Ship 1 and no
propeller polishing occurred before dry-docking. However, there was an increasing trend in power
deviation and fuel oil consumed per nautical mile of the KPIs. Following dry-dock, KPI values were
reduced with an increasing long-term trend.

Fig.7: Time history of % power increase (KPIa). Red and blue dots correspond to ballast and laden
condition, respectively. PP and DD stands for propeller polishing/cleaning and dry-docking,
respectively.

KPIa

KPIb

KPIc
Fig.8: Time histories of KPIs for both ships including trends.

Therefore, as shown in Fig.8, the effects of propeller polishing and cleaning and dry-dock in
performance were adequately captured. Nevertheless, the main target is to examine the effect of the

61
mewis duct provided that it was installed at the same time hull and propeller cleaned in dry-dock.
Therefore, as a first step, we need to examine more detailed the long-term behaviour following dry-
dock. Fig.9 presents information of the distribution of each KPI, and specifically the mean value, the
25th and 75th percentiles as well as the maximum and minimum values excluding the outliers.
Furthermore, these plots have been derived for specific time periods either defined as periods before
and among specific events (e.g. propeller polishing or dry-dock) or as yearly intervals following dry-
-dock. Such graphs
provide additional information as present the behaviour of the statistical significance zone of the KPI
values, except from its average value.

Fig.9: Synopsis of the distribution of KPI values in box and whiskers type of diagrams for specific
time periods. Horizontal thick black lines correspond to mean values.

For Ship 1, we have included two periods before dry-dock; one before propeller polishing (named with
the number 1 in Fig.9) and the other stands for the period among propeller polishing and dry-dock
(named with number 2). Fig.9 includes also a case that include all the period following dry-dock (named
as 3 and 2 for Ship 1 and 2, respectively). It can be derived that propeller polishing/cleaning resulted in

62
a significant decrease of KPI values. For power increase (KPIa), Ship 1 presented less deviated KPI
values, while for year 2 achieved less increase than the respective one of Ship 2. Similar results present
the analysis of KPIb, whereas KPIc does not indicate significant difference between the ships.

Therefore, in order to focus on KPIs trend, we produce through LAROS Analysis Engine the data sets
used for the graphs of Fig.10 which show the statistical trend zone of the KPIs for three key periods:
before dry-docking, for the first and second year after dry-docking. For Ship 1 we have selected the
period before dry-docking to correspond to the one between propeller polishing/cleaning and dry-dock.
For KPIa (% power increase), we can observe that both ships present a decrease with the one of Ship 1
to be more pronounced, while after year 1 the power increase of Ship 2 is sharper by an amount of 3.5%
in average. A similar trend increase between year 1 and 2 occurs also for KPIb, where Ship 2 presents
a severer increase. If we compare at kn and equal percentage of time for loading
conditions at ballast and design draft, then the difference in fuel oil consumption would be in the range
of 0.8-1.5 tons/day or else 3.5-5%. This difference is increasing to a range of 5-7% as time is passing
further from dry-dock. Such effect could be attributed to the mewis duct as it comparatively seems to
be more effective in the later stages of the monitoring period, where probably hull fouling were in-
creased. On the other hand, this trend cannot be verified when using the third KPI of the study, where
for year 1 both ships present similar behaviour, while for the year 2, Ship 2 had a steadier KPI trend.

Fig.10: KPIs statistical trend zones in time. Dashed lines correspond to mean values, while the upper
and lower lines to the 75% and 25% percentile, respectively.

63
5. Concluding remarks

We examine the capability of a practical performance monitoring framework to quantify the effect of
an energy saving device such as a mewis duct. The key output of the framework is a set of KPIs, whose
calculation is relied on the LAROS platform for the collection, processing and analysis of high-
frequency data. The case study examined refers to two sister vessels, where one of them had been fitted
with a mewis type duct The effect of other interventions aiming at increasing the energy efficiency of
the ships and specifically hull and propeller cleaning were also quantified within this assessment
framework.

Our aim is to analyse whether such a quite practical approach could identify any fuel consumption gains
provoked by the mewis type duct by comparing the performance of the ships for the period following
dry-dock. Fuel savings of the order of 3.5-5% were identified with an increasing trend as hull and
propeller fouling is increasing. Such a difference could be owed to the presence of the mewis duct, but
we suggest that this result deserves a more detailed examination. Dynamic speed profiling of the vessel
may produce a better result for this energy saving device. Therefore, as a next step, an analysis targeting
in the separation of the hull and propeller effect in performance in real operating conditions is planned
to be carried out, where the option of directly measuring thrust shall not be ignored as an option to the
problem. Furthermore, issues such as the performance of such devices in conditions deviating
significantly from calm sea is also a subject that needs attention.

References

BOUMAN, E.; LINDSTAD, E.; RIALLAND, A.; STRØMMAN, A. (2017), State-of-the-art technol-
ogies, measures, and potential for reducing GHG emissions from shipping A review, Transportation
Research Part D, 52, pp.408-421

BERTRAM, V. (ed.), (2018), Proceedings of 3rd Hull Performance & Insight Conference, Redworth

GRIGOROPOULOS, G.J.; THEODOSIOU, D.K.N. (2012), Harmonising SEEMP with effective vessel
operation, SNAME 4th Int. Symp.

ISO, (2016), ISO19030: Ships and marine technology - Measurement of changes in hull and propeller
performance. Part 2 Default method, ISO, Geneva

LOGAN, K.P. (2011), , ASNE Intelligent Ships

Symp. IX, Philadelphia

SCHNEEKLUTH, H.; BERTRAM, V. (1998), Ship design for efficiency and economy, Butterworth-
Heinemann

SPANDONIDIS, C.C.; THEMELIS, N.; CHRISTOPOULOS, G.; GIORDAMLIS, C. (2018), Evalua-

tion of Ship Energy Efficiency predictive and optimization models based on Noon report and Condition
Monitoring datasets, Proceedings, 7th Int. Conf. on Data Analytics, IARIA

THEMELIS, N.; SPANDONIDIS, C.C.; CHRISTOPOULOS, G.; GIORDAMLIS, C. (2018), A

Comparative Study on Ship Performance Assessment based on Noon Report and Continuous
Monitoring System datasets, Proceedings, 12th Conf. Hellenic Institute of Marine Technology, pp.55-
64

64
 From Bow to Stern:
Hydrodynamic Measures for Increased Hull Performance
Florian Kluwe, Lars-Uve Schrader, Herbert Bretschneider, HSVA Hamburg Ship Model Basin,
Hamburg/Germany, kluwe@[Link]

Abstract

This paper reviews a comprehensive set of innovative technologies for improved ship-hull
performance which have been studied at HSVA in the past or are currently under investigation. The
focus is on the reduction of frictional drag, the dominant component of the total hull resistance for
large slow-speed vessels. Fundamentally, all concepts build on a modification of the boundary layer
via different physical mechanisms such as stabilisation for transition delay through compliant
coatings, manipulation of coherent flow structures by grooved surfaces, injection of a low-friction
fluid via air lubrication and momentum transport through deflector vanes. They often follow bionic
principles inspired by nature, e.g. by dolphins, sharks and penguins, and are associated with a
preferred region of application along the hull from bow to stern. The individual drag-reduction
potentials range from about 2% to well above 5% such that the technologies may in combination yield
savings in excess of 10%. Whereas some technologies are on the market, others are still a matter of
fundamental research. The results have been obtained through in-house studies and within the
publicly funded research projects FLIPPER, HAI-TECH, eSHaRk and TARGETS. Finally, an outlook
on a new concept based on super-hydrophobic surfaces is given which is currently being studied in
the EU project AIRCOAT. New ideas of boundary-layer separation control are discussed as well.

1. Introduction

IMO emission-reduction targets will definitely require a fundamental change in seaborne trade pat-
terns. As alternative fuels will come at a substantially higher price than the present marine fuels, in-
vestment into efficiency improvements will become more attractive and solutions currently failing to
prove their return on investment will likely stimulate the market in the near future. The efficiency of
shipping is among other aspects determined by the hydrodynamic performance of the merchant fleet,
with ship-hull resistance and propulsive efficiency being the key factors. The hull resistance of large,
slow-speed vessels such as bulk carriers, tankers and general-cargo ships is dominated by the viscous
resistance components; these are composed of the friction between the hull surface and the water
including three-dimensional effects and the friction-induced pressure resistance caused by head loss
along the streamlines past the hull. Since the aforementioned ship types make up a large portion of
the worldwide fleet, Fig.1, a reduction of viscous hull resistance will significantly contribute to a
more ecological and economical seaborne transportation.

Fig.1: Fraction of different vessel types am

January 2017 (data source: [Link])

65
The frictional forces on the ship hull are confined to a limited region, the boundary layer, which
changes its state along the hull: at the bow, the initially laminar flow quickly transitions to the
turbulent regime associated with higher skin friction. Towards the stern, the boundary-layer flow
decelerates, loses energy and may even separate from the hull surface. All these phenomena cause
increased drag. This paper considers four different technologies, of which three directly act on the hull
surface and inside the boundary layer to provide reduced skin friction and thus improved hydro-
dynamic performance. The fourth technology also modifies the boundary layer but focusses on an
improved interaction between the hull-wake flow and the propeller. These four hull-performance
enhancement technologies are tailor-made for different hull zones, Fig.2, and are illuminated in the
subsequent chapters.

Fig.2: From bow to stern: different concepts of drag reduction for improved hull performance

2. Compliant coatings

In the research project FLIPPER (2014-2017), HSVA joined forces with the Fraunhofer Institute for
Manufacturing Technology and Advanced Materials (Fraunhofer IFAM), the Hamburg University of
Technology (TUHH) and the chemicals manufacturer ARKEMA in order to investigate the potential
of compliant ship-hull coatings to increase the hydrodynamic hull performance. Thanks to their
softness and responsiveness, compliant coatings are able to interact favourably with the boundary
layer along the ship hull (fluid-structure interaction). Two principal physical mechanisms are at play:
(i) a delay of laminar-turbulent boundary-layer transition and (ii) an attenuation of coherent flow
structures in fully turbulent flow. FLIPPER focussed on the former mechanism, drawing inspiration
from dolphins and their soft, pliable skin, Fig.3a.

Fig.3: (a) Schematic and (b) mechanical model of dolphin skin; reproduced after Carpenter and
Garrad (1985)

The project goal of FLIPPER

hulls with the ability to postpone the boundary-layer transition to turbulence at the bow of the vessel.

66
To this end, numerical calculations of the fluid-structure interaction between the laminar boundary
layer at the bow and the compliant coating were conducted using a mechanical model of dolphin skin,
Fig.3b, see also Carpenter and Garrad (1985). This model was coupled to a numerical solver of the
Orr-Sommerfeld equation governing the development of the so-called Tollmien-Schlichting (TS)
waves, Schmid and Henningson (2001). TS waves are minute flow disturbances inside the laminar
boundary layer which grow in amplitude when travelling downstream and eventually break down to
the fully turbulent state. The role of the compliant coating is to attenuate the growth rate of these TS
waves such that the laminar state can be maintained along a longer stretch of the boundary layer,
leading to lower skin friction and reduced frictional hull resistance. The compliant-coating model and
the Orr-Sommerfeld solver were used to identify suitable coating parameters layer thickness,
ulus) and damping for effective transition delay. The procedure was applied
to the laminar boundary layer along the bow of a small search-and-rescue (SAR) vessel, Fig.4, in
model scale 1:3.2 (5.72 m length). The details of the computations are described in Schrader (2019).

Fig.4: SAR vessel of the German Maritime SAR Association DGzRS (source: [Link]). The
FLIPPER coatings were designed for a 1:3.2 scale model of this ship (5.72 m length). HSVA
thanks the Fassmer group and DGzRS for their permission to use the hull geometry in
FLIPPER

Fig.5: TS-wave growth curves (blue) in laminar flow past the bow of a 1:3.2 SAR ship model and
envelope (black) of total disturbance growth. Transition prediction at a ship-model speed of 10
m/s according to the eN method for (a) a rigid hull surface and (b) a 9.16 mm-thick compliant
coating

TS-wave growth curves for a rigid and a coated bow surface of the SAR ship model were computed,
Fig.5. The calculations assumed a coating made of a 9 mm-thick soft silicone layer covered by a 0.16
mm-thick polyethylene foil mimicking the thick blubber and the thin epidermis of dolphin skin,
Fig.3a. The TS waves amplified at a substantially slower rate on the compliant surface, Fig.5b, than
on the rigid counterpart, Fig.5a, such that the transition threshold as per the eN prediction criterion was
reached farther downstream. In this particular example, transition to turbulence occurred 1.37 m
downstream of the stem for the compliant coating versus 0.48 m for the rigid surface (red arrows in

67
Fig.5). This transition shift allowed for a calculated friction-drag reduction by almost 56 N
corresponding to 5.6% of the frictional drag of the hull model (see Schrader (2019) for details).

The FLIPPER Hydrodynamics and

Cavitation Tunnel (HYKAT). A wooden 1:3.2-scale model of the SAR-ship hull was used, featuring a
removable bow segment with an integrated load cell for drag-force measurement, Fig.6a. Several
compliant coatings of different thickness and stiffness were formulated and manufactur
partner Fraunhofer IFAM. It could be demonstrated in the experiments that these
skins were indeed able to reduce the drag force on the bow with respect to the uncoated reference. At
a tunnel speed of 10 m/s (design speed used in the calculations), the 9.16 mm-thick coating led to a
drag reduction by almost 21 N, Fig.6b which is less than the predicted value of 56 N. This dis-
crepancy is explained by inherent simplifications of the transition-prediction method, the mechanical
coating model and the numerical setup as well as some uncertainty in the experimental determination
of the viscoelastic coating parameters, Schrader (2019).

Fig.6: (a) SAR ship-hull model in HYKAT test section. Coatings applied on the bow segment only.
Drag force on the bow is measured by an integrated load cell. (b) Frictional drag reduction on
the bow w.r.t. the rigid surface thanks to compliant coatings of 3.16 mm and 9.16 mm thickness

In summary, the FLIPPER project successfully demonstrated the functionality of compliant coatings
as a means of frictional drag reduction in marine applications. The physical mechanism behind these
coatings is a passive boundary-layer control via delayed laminar-turbulent transition. The passive

energy is needed to operate the system in contrast to active technologies such as air lubrication (see
Sec. 4). The obtained frictional drag reduction is moderate but it was also shown in FLIPPER that
there is potential for further optimisation of the coating parameters such that a drag reduction by 2-3%
based on the total hull resistance is deemed realistic. The compliant-coating technology is applied at
the bow of ships where the transition to turbulence occurs. It is best-suited to small vessels as the
relative savings through the technology decrease with increasing ship size, Gad-el-Hak (1996). It is
also pointed out that the compliant-coating technology is not yet market-ready because practical
aspects such as robustness, anti-fouling properties and application technologies on shipyard scale still
need to be investigated.

3. Riblet surfaces

Riblets are micro-textured surface protrusions aligned with the flow which impose an anisotropic
roughness distribution on a surface. The original idea of using riblets was inspired by the skin of fast-
was first applied in the aerospace, automotive and energy
industries. The application to Olympic rowing shells (Los Angeles 1984) and racing yachts
(America s Cup 1987) resulted in banning this technology from these disciplines. Extensive
experimental investigations on various geometries have been made in a limited range of Reynolds
numbers, Bechert et al. (1997), indicating a drag reduction of up to 10% for plane micro-textured

68
surfaces compared to hydraulically smooth flat surfaces. Although the mechanism is not yet fully
understood, a reduction of turbulent fluctuations and a decrease in turbulent shear stress is supposed.

In the German research project HAI-TECH (2009-2011), Wilke et al. (2010), a consortium of Fahrion
Engineering, Blohm + Voss Naval, Beluga Shipping, LimnoMar and HSVA led by Fraunhofer IFAM
developed a coating technique combining paint application with micro-texturing in a single
continuous production process ( Dual-Cure-Paint ) for ship hull applications, Fig.7a. The trials were
carried out on an 8 m-long torpedo-shaped body in the HYKAT water tunnel at HSVA, Fig.7b. The
comparison between riblet-structured and smooth surfaces revealed a reduction of frictional resistance
by more than 5% at near-operational test conditions at ab. 20 knots flow speed. This demonstrated the
enormous potential of adapted surface structuring on ship hulls in terms of fuel savings and fuel cost
reductions for the shipping industry.

Fig.7: (a) HAI-TECH riblet structure (© Fraunhofer IFAM). (b) Development and testing of the
ribleted foils in the HYKAT water tunnel using a cylindrical test body

In the European research and innovation project eSHaRk (2015-2019), PPG (hull paint supplier,
coordinator), Mactac (adhesive-film manufacturer), ND Coating/Meyer Werft (hull coating and anti-
corrosion services), VertiDrive (robotic solutions for ship hull treatment) and HSVA formed a
consortium to develop and manufacture a self-adhesive, non-toxic fouling release foil, produced by
applying a state-of-the-art fouling release coating on top of a self-adhesive plastic film. The micro-
texture of the film surface for friction reduction was accomplished by appropriate embossing during
the manufacturing process. Three test campaigns in the HYKAT water tunnel demonstrated a friction
reduction of ab. 4% compared to standard fouling-release paint; moreover, a sufficient strength of the
film system and appropriate adhesion forces of the self-adhesive layer could be verified.

4. Active air lubrication

For the last eight years HSVA has been investigating air lubrication systems in the HYKAT water
tunnel (partial model in full scale) for application to the flat-bottom area of ships in order to reduce
the surface friction force, the propulsion power and the GHG emissions. These studies have been con-
ducted on behalf of a company which has since then equipped several new buildings with air lubrica-
tion systems and is nowadays the leading provider of that technology on the market.

The working principle is as follows: the air lubrication system generates small air bubbles and intro-
duces them into the turbulent water boundary layer on the ship hull, producing a fluid-gas mixture of
lower viscosity than pure water. The bubbles reduce the wetted hull surface and may also favourably
interact with the turbulent flow structures in the boundary layer, thereby significantly diminishing the
frictional hull resistance. Although the effect of skin friction reduction of air-lubricated plane surfaces
often referred to as micro-bubble drag reduction (MBDR) has been tested on a laboratory scale by
various researchers for about 40 years, see e.g. Madavan et al. (1984), HSVA was among the first to
investigate the system in full scale, using partial models integrated into a flat plate of 8 m by 2 m,
Fig.8a, connected to a force balance to measure the friction force at the top of the HYKAT test section

69
(dimensions: 11 m × 2.8 m × 1.6 m). Apart from different geometries of the air-release openings, ship
draughts of 2-13 m, air flow rates up to 120 m³/h and water flow speeds (ship speeds) up to 18 knots
were tested with focus on the bubble creation and carpet behaviour, Fig.8b. For the partial model of
the air lubrication system a friction reduction of 40% was achieved compared to a flat plate of the
same dimensions. Since the air lubrication system is an active system the total power balance needs to
include the necessary power for providing the required compressed air.

Fig.8: (a) Air lubrication model in the HYKAT. (b) Bubble carpet during testing

In 2014 HSVA participated in the sea trial of a medium-range chemical tanker of 40,000 DWT which
had been refitted with the HSVA-tested air lubrication system during the dry-docking period. The in-
stallation of the system took two weeks and could be accomplished within the standard docking
schedule. Net average energy savings of 4.3% for the vessel in ballast and 3.8% in laden condition
could be confirmed based on a conservative interpretation of the speed-power measurements, Shell
and Silverstream (2015).

5. Boundary-layer alignment

The wake flow behind a ship plays a crucial role for the propulsive efficiency. Especially the bulky
hull forms of full-block vessels suffer from massive axial-momentum losses above the propeller shaft;
in addition, rotational losses occur in the propeller slipstream. In order to address these two sources of
loss, HSVA developed a novel type of hull appendage in the research project TARGETS (2010-2014)
the boundary-layer alignment device (BLAD) consisting of a pair of flow deflectors, Fig.9.

Fig.9: BLAD deflector and position on the aft ship of a Capesize bulk carrier. The two deflectors on
the port and starboard sides feature different profiles and are asymmetrically arranged

The purpose of the BLAD is to deflect the streamlines towards the hull surface in order to accelerate
the wake flow locally and reduce the boundary-layer thickness for a more homogeneous flow through
the propeller. Moreover, the BLAD deflectors are intended to create a swirling flow against the pro-

70
peller rotation for diminished rotational losses in the slipstream along with extra thrust at a given en-
gine power. This swirl is accomplished by different profile shapes and an asymmetric arrangement of
the two deflectors.

Fig.10 ots speed: improvement of

wake homogeneity and swirl generation. (b) Nominal wake in the propeller plane

The BLAD was developed and tested for a Capesize bulk carrier at 14 knots speed with an 8.5 m-
diameter four-bladed propeller, -house codes FreSCo+ and
QCM. Thanks to the asymmetric deflection of the port and starboard sided streamlines towards the
propeller, Fig.10a, a nominal wake with increased homogeneity and a swirling flow component could
be achieved, Fig.10b. The flow acceleration into the propeller plane led to an increased advance num-
ber at which the propeller could be operated more efficiently, Fig.11a. This effect in combination with
diminished rotational slipstream losses allowed for a lower required power to obtain the same ship
speed, Fig.11b. Three different retrofit scenarios were considered: (i) a refit of BLAD deflectors only,
(ii) a BLAD retrofit plus cutting and grinding of the propeller-blade trailing edges and (iii) a BLAD
retrofit along with a new tailor-made propeller. As expected, the most expensive retrofit option (sce-
nario iii) allowed for the largest savings of propulsive power of almost 7%, Fig.11b. The details of the
BLAD development and achievements are compiled in Schrader and Marzi (2017).

Fig.11: (a) Open-water efficiency and (b) delivered propulsive power at a speed of 14 knots for three
different propellers of a Capesize bulk carrier: (i) original propeller, (ii) original propeller with
cut and ground blade trailing edges, (iii) new propeller each combined with the BLAD
deflectors in comparison to the baseline (original propeller, no BLAD)

Apart from a power reduction, the BLAD deflectors also led to a more symmetric distribution of the
mean propeller-thrust force with a reduced peak value, Fig.12. This is expected to yield lower

71
propeller-blade loads along with reduced cavitation risks as well as a better course stability of the
vessel. The latter aspect will allow for fewer course corrections by rudder manoeuvres such that
additional fuel savings are anticipated.

Fig.12: Mean thrust distribution across the propeller disc for a Capesize bulk carrier at 14 knots speed
using the original propeller: (a) without and (b) with BLAD deflectors

In summary, the BLAD concept provides an attractive technology for improved hull-propeller
performance along with reduced fuel costs and emission of pollutants. The design of the BLAD
deflectors is considerably less complex than that of established energy-saving devices like stator fins
or various duct products. Since the deflectors are mounted quite far upstream of the propeller they are
hardly exposed to the unsteady propeller-induced flow field and the associated dynamic loads,
reducing the risk of fatigue and failure.

6. Conclusion and discussion

For many years ship designers have focussed their efforts on the wave-making resistance when
optimising the efficiency of their hull designs. Viscous flow was largely out of the focus for various
reasons: in pre-CFD times the wave-making resistance was the only component that could be studied
in detail by visual observation while measurements delivered integral values for the total resistance
only. Technologies like CFD simulations allow for a much more straightforward decomposition of
hull resistance and therefore open the opportunity to include these aspects into the optimisation
process.

Even more important, the growing ship sizes, the d the focus on hull lines
excessively optimised for minimum wave-making resistance have increased the relative contribution
of the viscous drag components to the total hull resistance. Consequently the last years have seen a lot
of new and improved paint products that are aiming to reduce the frictional resistance by keeping the
surface roughness at a minimum. Looking beyond these approaches, this paper presents several
technologies that also target the reduction of viscous resistance by different approaches. Their state of
maturity is quite diverse: while technologies like the boundary-layer alignment device (cf. Sec. 5) are
close to production readiness, other ideas like compliant coatings (cf. Sec. 2) are still mainly
fundamental research.

Experimental model testing plays an important role in these research studies. Model basins provide
the large-scale experimental facilities needed to conduct hydrodynamic investigations at Reynolds
numbers with practical relevance for the shipping industry. Numerical studies are certainly an
attractive alternative at low to medium Reynolds numbers (see Sec. 2); however, fully turbulent high-
Reynolds number boundary layers are mostly beyond the scope of industrial CFD and therefore still
need to rely on physical testing.

72
An important aspect is to identify which technology fits to which type of ship. Taking for example the
boundary-layer alignment device the effect will be most prominent in the case of blunt full-block
vessels like tankers and bulkers while the impact will be limited in the case of slender hulls such as
RoRo or passenger vessels at least with respect to resistance reduction. Air bubble injection (cf. Sec.
4) can be placed only in regions with fairly horizontal, large surfaces and thus becomes more difficult
to be applied on vessels with significant deadrise. The previous example also demonstrates another
crucial aspect: to make best use of the measures they need to be placed at the right position on the
hull. The transition-delaying compliant coatings, for instance, need to be located in the bow
region and would not make much sense in the aftbody region. The same applies to air bubble systems
as already discussed above. Hence, the task and their know-how do not only become
manifest in a mature, reliable and resilient technical solution but also in developing hull designs
which allow for a beneficial placement and combination of these technologies as illustrated in Fig.2.

The experience shows that the main market barrier for these technologies is often the durability under
service conditions. This includes their resilience against mechanical failure as well as the robustness
of the working principle against dirt and fouling, which is an obvious challenge in the case of grooved
surfaces (cf. Sec. 3). In order to generate trust into these new approaches, continued research and
even more important prototype installations are needed to demonstrate their practical feasibility.
The overall savings that are assumed realistic based on the currently available data on the total
resistance reduction of ship hulls amount to 10% and more in total. This is an order of magnitude
above the potential for further improvements in wave resistance reduction a very clear motivation
towards extended research on the technologies presented in this paper.

7. Outlook

HSVA is involved in or is currently setting up additional research projects dealing with passive
and active boundary-layer flow control. The EU project AIRCOAT (May 2018-April 2021) deals with
a bionic principle of drag reduction, too: a team of ten European science and industry experts led by
the Fraunhofer Center of Maritime Logistics and Services (Fraunhofer CML) is currently developing
a passive air lubrication technology that utilises the biomimetic Salvinia effect, Barthlott et al. (2010).
Nature has developed this effect through evolution, which allows the Salvinia plant, a fern floating on
the water, to breathe even when submerged by maintaining a permanent layer of air. This ability
builds on a complex surface composed of super-hydrophobic and hydrophilic structures. The project
partners are implementing the Salvinia mechanism on a self-adhesive foil system which is able to trap
air on surfaces in water. A ship equipped with such an AIRCOAT foil will produce a thin permanent
air layer reducing the overall frictional resistance significantly while at the same time acting as a
physical anti-fouling barrier between the water and the hull surface. The project is a prime example of
a biomimetic application where scientists and engineers learn from nature. The potentials of the
AIRCOAT project are enormous: initial estimates indicate that the AIRCOAT technology may reduce
the main-engine fuel oil consumption and exhaust gas emission by at least 25%. The major advantage
over existing air lubrication technologies is that the ship hull is passively lubricated, i.e. no energy for
operation is needed. Also, if successful the AIRCOAT refit technology will be immediately applicable
to the entire fleet in the form of a foil system. Through a combination with the latest self-adhesive foil
technology, AIRCOAT has the potential to revolutionise the maritime coating sector, promising a
ground-breaking future energy-efficiency and emission-reduction technology.

Passive and active boundary-layer separation control is another path of research and development in
. Nature gives again inspiration: the flippers of humpback whales feature a wavy
leading edge formed by so-called tubercles, Aftab et al. (2016). These -
vrability through a postponement of flow separation to larger angles of attack. This passive principle
can be transferred to ship rudders, stabiliser fins or highly loaded propeller blades in order to increase
the lift and the lift-to-drag ratio. Even larger gains in lift at almost no drag penalty can be expected
from active separation control. The focus is here on fluidic oscillators for boundary-layer momentum
enhancement and energisation, Kim et al. (2017). These devices are an elegant, mechanically simple
way of postponing the angle of stall to very large values and can be beneficially applied to hydro-

73
dynamic surfaces prone to separation such as ship rudders and manoeuvring ship hulls. The main
benefit of fluidic-oscillator based active flow control lies in the possibility of turning off the system
when not needed a clear advantage over vortex-generator fins which permanently create drag.
Fluidic oscillators have indeed been successfully used for separation control on aircraft rudders, Lin et
al. (2016).

Apart from a continued development of the present and of novel friction-reducing technologies, the
assessment and validation tools need further attention. Although established EFD and CFD
procedures for the flow analysis around ship hulls are available, including the determination of the
various resistance components, the incorporation of friction-reducing measures into these procedures
is not at all straightforward: the theoretical challenges ahead include the lack of scaling laws e.g. for
model tests with micro bubbles and the incorrect boundary-layer thickness owing to the violation of
-layer resolution in practical
CFD models or in the challenges associated with fluid-structure interaction and many more multi-
physics aspects.

Acknowledgements

In this paper, HSVA presents results from several publicly funded research projects: FLIPPER and
HAI-TECH received funding from the Federal Ministry for Economic Affairs and Energy of Germany
(BMWi). The European Commission funded the projects TARGETS, eSHaRk and AIRCOAT. HSVA
gratefully acknowledges these funding bodies for their financial support. The authors thank Dr.
Jochen Marzi (HSVA) for giving inspiration to the basic theme of this paper.

References

AFTAB, S.M.A.; RAZAK, N.A.; MOHD RAFIE, A.S.; AHMAD, K.A. (2016), Mimicking the hump-
back whale: An aerodynamic perspective, Progress in Aerospace Sciences 84, pp.48-69

BARTHLOTT, W.; SCHIMMEL, T.; WIERSCH, S.; KOCH, K.; BREDE, M.; BARCZEWSKI, M.;
WALHEIM, S.; WEIS, A.; KALTENMAIER, A.; LEDERER, A.; BOHN, H.F. (2010), The Salvinia
Paradox: Superhydrophobic Surfaces with Hydrophilic Pins for Air Retention Under Water,
Advanced Materials 22(21), pp.2325-2328

BECHERT, D.W.; BRUSE, M.; HAGE, W.; VAN DER HOEVEN, J.G.T.; HOPPE, G. (1997),
Experiments on drag-reducing surfaces and their optimization with an adjustable geometry, J. Fluid
Mech. 338, pp.59-87

CARPENTER, P.W.; GARRAD, A.D. (1985), The hydrodynamic stability of flow over Kramer-type
compliant surfaces. Part 1. Tollmien-Schlichting instabilities, J. Fluid Mech. 155, pp.465-510

GAD-EL-HAK, M. (1996), Compliant Coatings: A Decade of Progress, Applied Mechanics Review

49(10), part 2, pp.S147-S157

KIM, J.; MOIN, P.; SEIFERT, A. (2017), Large-Eddy Simulation-Based Characterization of Suction
and Oscillatory Blowing Fluidic Actuator, AIAA Journal 55(8), pp.2566-2579

LIN, J.C.; ANDINO, M.Y.; ALEXANDER, M.G.; WHALEN, E.A.; SPOOR, M.A.; TRAN, J.T.;
WYGNANSKI, I.J. (2016), An Overview of Active Flow Control Enhanced Vertical Tail Technology
Development, Tech. Rep. NASA 20160007653, pp.1-13

MADAVAN, N.K.; DEUTSCH, S.; MERKLE, C.L. (1984), Reduction of turbulent skin friction by
microbubbles, Phys. Fluids 27(2), pp.356-363

SCHMID, P.J.; HENNINGSON, D.S. (2001), Stability and Transition in Shear Flows, Springer

74
SCHRADER, L.-U.; MARZI, J. (2017), A Novel Power-Saving Device for Full-Block Vessels, 20th
Numerical Towing Tank Symposium, Wageningen, pp.205-210

SCHRADER, L.-U. (2019), Passive Drag Reduction via Bionic Hull Coatings, J. Ship Research

SHELL; SILVERSTREAM (2015), Silverstream Air Lubrication Technology Delivers Significant

Energy Savings, [Link]
media-releases/[Link]

WILKE, Y.; STENZEL, V.; PESCHKA, M. (2010), Haifischhaut für Flugzeuge, Schiffe und Wind-
energieanlagen, [Link]
presse/2010/pressetext_2010_haifischhaut.pdf

75
 ISO 19030: Onboard Monitoring System for Real-time Performance
Feedback, Prediction and Optimization
Paolo Becchi, CETENA, Genova/Italy, [Link]@[Link]

Abstract

The evaluation of the effective ship power performances during commercial operation and in particular
after main maintenance events represents an important feedback for the owner, making it possible to
optimize the ship efficiency and management. For this reason, CETENA recently improved its own Ship
Performance Monitoring System including what prescribed by the international procedure ISO 19030,
that is the evaluation of both the effective ship performances (speed and power corrected data) and the
prescribed Performances Indicator (KPI). The system provides to the crew and to the owner indication
of the effect of maintenance events on ship performances together with all the other parameters already
included in the real time data acquisition. The software has been implemented keeping into account the
possibility for further improvements in order to guarantee a quite fast and reliable maintenance any
time the international procedure will be updated. Because of the application of S&P data corrections
requires environmental conditions together with the navigation ones, the system is directly connected
with the remote CETENA own weather forecast system that gives the possibility to know all data needed
(wind, sea state, current, depth). The ISO 19030 procedure has been also included in the Sailing
Assistant Module of the system aimed to optimize future voyages. The system will optimize the voyage
on a minimal consumption basis with the ship affected by reduced environmental-related added power.
The application of ISO 19030 procedure in the system thus allows to take under control ship
performances and management and brings about a reduction of fuel and pollutant emissions.

1. CETENA

In the last years CETENA has implemented its own Performance Monitoring System, designed for the
automatic check and storage of data related to the performances of the ship in navigation. The number
and type of data to be acquired can be modified for any ship depending on the signals available onboard
or from the arrangement of custom sensors. The system has been developed in order to continuously
monitor the data and provide a real time feedback to the crew. In addition, if an internet connection is
available the system will automatically send all collected data to the owner headquarter making it
possible for the technical department to check the ship performances in real time.

Fig.1: Performance Monitoring arrangement on wheelhouse console

Together with hardware components and custom sensors, the system is structured by the following two
software modules:

Data acquisition module (Core), that is usually located on a standalone machine connected to
custom sensors and ship automation. This PC is usually an autonomous industrial PC, directly
connected to the ship LAN, too.
Repeater consoles (Repeater). The repeater console is aimed to load the collected data from the

76
 core module and to show to the user all the information needed for the evaluation of the current
ship performances. There is no restriction for the repeater console that can be arranged onboard
(wheelhouse, chief engineer cabin, ECR, ...).

Fig.2 shows
consoles.

Fig.2: Usual "Performance Monitoring" system layout

The Performance Monitoring system (PM) collects all data needed for the ship performance analysis
and any other data required by the crew/technical dept. In detail, the collected data concern both the
sailing conditions:

time and location

speed over ground, course over ground and heading
propulsive conditions (propeller pitch, shaft rate of revolutions, torque and power)
environmental conditions (wind state
ship loading conditions (draft, trim, displacement)
anti-roll stabilizer fin status (on/off) from automation system
rudder angle
ship operating status (at sea, manoeuvre, docking/anchor) by means of a dedicated signal or by
a proper association of ship automation signals

and the energy generation aspects:

fuel consumption for main engines and diesel generators

current status of MMEE and DDGG
electric power generated by DDGG
electric power generated by Shaft Generator

Fig.3 represents the data collector module of the system, aimed to collect the above-mentioned data
and to show them through the repeater console. The window is subdivided into different areas, each
one is aimed to show data related to different aspects: navigation, environment conditions, analog chan-
nels, shaft, wave radar and extra signals acquired. The system stores the data collected in text files, with
a data record every 5 minutes and a single file per day.

time feedback about the propulsive efficiency depending on the current trim measured. This module
requires some preliminary tests at sea to identify the referenced speed and power curves to be used for
direct estimation. The module provides a direct indication of the efficiency of the trim and a difference
from the best trim curve computed at the same ship displacement. Fig.4 shows the user interface of the

77
Optimum Trim module, including the chart of the current performance (white line) and the two thresh-
olds defining the good, the acceptable and the bad conditions areas (green, yellow and red area).

Fig.3: Data collector module

Fig.4: Optimum Trim module console

Fig.5: Analysis software console

78
The system is usually required for propulsive performance monitoring during ship life, the evaluation
of ship management and fuel consumption. In order to make easy this kind of analysis, a custom
software has been developed for the analysis of the collected data. The user has the possibility to see
history chart of the acquired signals, compare trips performed in different periods (like before and after
dry-dock activity) and save the analysis performed in a worksheet format. Fig.5 shows the main console
of the analysis software, on which all the data collected and the trips performed can be filtered and
rformances.

2. Computation module for ISO 19030 correction

implemented by including a computation

module aimed to analyze the ship performances in accordance with the international procedure ISO
19030. This normative gives indications for the evaluation of the propulsive efficiency increase due to
maintenance events as dry-dock, hull/propeller cleaning, reblading. From this point of view the
propulsive efficiency is strictly related to a well-known correlation involving many elements related to
the hydrodynamics of the propulsion, that are: hull roughness/fouling, propeller, displacement, trim and
environment conditions. The scope of the ISO 19030 is to define a practical method for the
measurement of ship performances, their analysis and correction and to evaluate a practical set of
indicators related with the real efficiency increase. From the point of view of data collection and
analysis, this normative looks to be very close to the ISO 15016 that is aimed to evaluate the real
propulsive performances for ship at official sea trials, and hence including many corrections like water
density, current effects through multiple runs, and added wave resistance. As the ISO 19030 aims at
the data monitoring of ship in operation, at the moment the ISO 19030 prescribes only few corrections
like wind resistance effects, shallow water and so on, but it cannot be excluded that further editions
might recommend other correction algorithms. Despite what recommended in ISO 15016, the system
used for the performance monitoring of ships of different type, size and age in operation must be
completely autonomous. It includes not only the detection of a stable condition for data recording but
also the storage, backup, data transmission and analysis. The ISO 19030 international normative is
structured into the following main steps:

Monitoring of the sailing and propulsive condition

Data recording, storage and analysis (including correction and performance value computation)
Estimation of the four Performance Indicators (PI)

2.1. Virtual Weather Station for environment condition data

The evaluation of the ship performance values and hence the performance indicators, requires the
acquisition of data related both to ship sailing and local environmental conditions. As previously
described, the current version of the normative prescribes only corrections for the effects of
displacement, shallow water and added resistance due to the wind. For this reason, at the moment the
use of a custom anemometer and the eco-sounder may be enough for the evaluation of the real ship
performances. However, a more accurate investigation of the weather conditions will lead to perform a
more detailed analysis of data, including information about the current magnitude and direction and the
sea state (significant wave height, direction and period). For this reason, CETENA has developed its
own Virtual Weather Station (VWS) aimed to provide indication of past, current or expected conditions
for ships sailing in any ocean or sea. not directly interfaced with
instrumentation/sensors measuring the weather conditions, but it collects data from many
weather/forecast services and rebuilds a single database with the more accurate spatial resolution and a
time step equal to 1 hour. The considered data set is the result of simulations based on numerical models
used for forecast and finally tuned with local direct measurements. For each sea/ocean the data set
needed for a complete description of the weather conditions are checked in advance, and made
consistent with the database units and entity already considered. The VWS is currently configured in
order to provide the following entities:

79
 Water depth
Sea state
- significant wave height
- wave direction from
- wave period
- stokes drift components
Current
- Eastward component
- Northward component
Wind
- Eastward component
- Northward component

Furthermore, the current component can be provided at different drafts, in order to identify the more
significant effect on the ship motion. All the water entities describe the average value on one hour time
step. Otherwise, the wind data are related to an average 6 hours because of the lack of more accurate
data sets. Anyhow, any ship with the Performance Monitoring System is also equipped with at least
one anemometer that provides really accurate data, every minute averaged.

The decision of CETENA to develop a service for the weather condition provision, is based on its
experience on both sea trials and computational fluid dynamics. Some years ago, CETENA was asked
to perform S&P sea trials on a ship in service, and then without the possibility to carry out the traditional
multiple runs. In that occasion it has been decided to correct the ship speed using current data coming
from a meteorological service. Thanks to the results have been obtained, it was possible to reduce the
spread of the ship speed values from 0.5 knots up to 0.3 knots; this value is considered acceptable by
all other international partners involved in the project.

The empowerment of the VWS system aims to query and provide the required data directly through
other own systems requiring environmental conditions for their needs. Together with the Performance
Monitoring system, also the software aimed for the Speed&Power sea trials analysis consistently with
ISO15016 has been developed including the possibility of a direct connection with VWS. So, this
system has not a direct user interface for the majority of the services provided. Anyhow it is also
possible to query the weather database for any other reasons by using a custom user interface, Fig.6.
This console gives the possibility of a direct investigation of weather conditions. Furthermore, if the
ship sailing conditions (speed and heading) are detailed for each required point, the apparent environ-
ment conditions will be automatically computed showing apparent wind, wave magnitudes and
direction.

Fig.6: VWS main console

80
2.2. Performance Monitoring process

From the point of view of the Performance Monitoring System, the computation module automatically
proceeds with the sailing condition monitoring, run measuring and storing. As previously mentioned
and about the environment conditions, the current version of normative requires only wind state, that
can be acquired through an anemometer arranged onboard and directly connected to the acquisition
system. However, if further revisions of the recommended procedure will require additional weather
data, thanks to internet connection the system will submit the measured run data to the VWS in order
to firstly characterize them with proper environment conditions, and then finally compute the corrected
propulsive values together with the Performance Indicator. Fig.8 represents the flowchart of the
Performance Monitoring from the sailing condition monitoring to the evaluation of the Performance
Indicators, represented in Fig.7 The
page shows the four indicators prescribed by the normative and described in Table 1.

Table 1: Basic hull and propeller Performance Indicators

Dry Dock Determining the effectiveness of the dry-docking (repair and/or retrofit activities)
Change in hull and propeller performance following present out-docking (evalu-
ation period) compared with the average from previous out-docking (reference
periods)
In Service Determining the effectiveness of the underwater hull and propeller solution (in-
cluding any maintenance activities that have occurred over the course of the full
dry-docking interval)
The average change in hull and propeller performance from a period following
out-docking (reference period) to the end of the dry-docking interval (evaluation
period)
Maintenance Trigger Trigger under water hull and propeller maintenance, including propeller and/or
hull inspection
Change in hull and propeller performance from the start of the dry-docking inter-
val (reference period) to a moving average at any chosen time (evaluation period)
Maintenance Effect Determining the effectiveness of a specific maintenance event, including any pro-
peller and/or hull cleaning
Change in hull and propeller performance measured before (reference period)
and after (evaluation period) a maintenance event.

Fig.7: Performance Monitoring - Repeater console: Performance indicator window

81
 Sailing condition
monitoring

Constant
condition

Data measuring data storage

Data correction

Performance Values computation Performance

Values log

Manteinance
events log

Performance Indicators computation

Fig.8: Performance Monitoring System flowchart

For each indicator, all the Performance Values are plotted in accordance with the following
representation:

the grey points are the PV not considered for the indicator computation
the red points represent the PV included in the reference period
the blue points represent the PV related to the evaluation period

Furthermore, the console shows some further information:

the total number of PV computed

the overall PV value, maximum and minimum
the last maintenance event set

If required, the system can send all the computed data to the owner headquarters in order to make it
possible to check the ship performance indicator and performance and change the maintenance
scheduling if necessary. At the moment of the drafting of this paper, the ISO 19030 had been released
in beta version and the tests were already in progress on a merchant ship used as testcase.

3. Sailing Assistant module for voyage optimization

The integration between the ISO 19030 module and the VWS service gave the opportunity for an
additional module, aimed at the optimization of a single voyage from the point of view of the required
propulsive energy. The idea of this system consists in the evaluation of the added resistance due to the
expected weather condition during a voyage and the optimization of both the waypoints location and
the ship power configuration. All that allows to respect the required estimated time at arrival and
minimize the total propulsive energy needed. The necessity to evaluate all added resistance components
according to the environment conditions has brought a modification of the ISO 19030. Furthermore,

82
the propulsive performance of the ship must be characterized by the speed-power curve and power
setting configurations, and the VWS must also be able to provide weather forecast at least up to 36
hours in advance.

The ISO 19030 has then been implemented in order to perform a more accurate prediction of the added
power and speed variation, and it has been included in an optimization algorithm together with the
request of weather forecast for each waypoint/solution analyzed. The voyage must be configured in
advance, just before leaving and include the departure time, the expected time at arrival, all the
waypoints describing the trips of the whole voyage is structured by and the maximum ship speed value
on each trip, if needed. The configuration of the voyages can be directly performed through a custom
user interface, Fig.9, that prepares the configuration files and uploads them in the Performance
Monitoring System. The console makes it possible to configure all the voyages, stating the waypoints
representing the trips in which a voyage is subdivided, and the maximum speed admitted on a single
trip, if necessary. Furthermore, it is also possible to compute the raw scheduling of each voyage by
setting the actual time at departure (ATD) and the expected time at arrival (ETA). In this way, the
console shows the average speed to be set during the unrestricted trip in order to respect the arrival
time.

The console is equipped with four main pages, aimed to define:

The maximum speed allowed depending on the experienced sea state. This setup must be de-
fined one time by the owner and it will be considered in order to adjust the computed ship speed
depending on the sea state expected on each trip
Voyages configuration, that gives the opportunity to define all the voyages the ship has to do,
as previously described
The extra-propulsive load, that makes it possible to define the power amount required by the
propulsive engines for any other business. This setup is really important for ship with diesel-
electric propulsion, because the diesel generator shared with all the other power needs. The
amount of power required during the day is important for the overall definition of the optimum
engine configuration, and hence of the minimum fuel consumption.

Fig.9: Sailing Assistant configuration console voyages configuration page

The approach on which the sailing assistant is based on consists in the evaluation of the added resistance
due to weather conditions along each trip, the evaluation of the power configuration, that is required to
respect the overall power request at the minimum consumption point and then the modification of the
waypoints coordinates in order to minimize the fuel consumption. The following figure represents a
test performed on a rectangular sea area where the distribution of the wave height (and hence the sea

83
state) is represented by two Gaussian distributions, while the wind is constant. The white line represents
the initial voyage that passes through the two high wave areas; the yellow line represents the optimized
voyage. location in order to define trips
characterized by lower expectation of added resistance.

Fig.10: Sailing Assistant voyages optimization simulation

4. Conclusions

CETENA Performance Monitoring system has been developed with the objective to monitor all the
relevant propulsive parameters on board and provides ad hoc processed information to increase energy
efficiently
systems together with ad hoc sensors, and is able to filter and integrate data, display all relevant data to
ship master and provide customized voyage reports according to customer needs.

Recently, the PM system has been implemented with the addition of the ISO 19030 computation
module, that is aimed to compute and plot the prescribed performance indicators showing the propulsive
efficiency increase due to maintenance activity. Furthermore, another module under finalization has
already been developed for the fuel consumption optimization of any voyage, keeping into account the
expected weather conditions in accordance with the international normative. These two PM skills are
directly connected with the CETENA Virtual Weather Station that is aimed to provide a service for
the evaluation of historical or expected weather conditions in a defined time and geographic
coordinates.

The new version of the CETENA Performance Monitoring system provides powerful and promising
skills for both the evaluation of the current ship propulsive performances and the optimization of
voyages and ship management for what concerns the fuel consumption.

84
References

ISO 19030-1 Ship and marine technology Measurement of changes in hull and propeller
performance Part 1: General principles Edition 2016

ISO 19030-2 Ship and marine technology Measurement of changes in hull and propeller
performance Part 2: Default method Edition 2016

ISO 19030-3 Ship and marine technology Measurement of changes in hull and propeller
performance Part 3: Alternative methods Edition 2016

85
Real Ship Maritime Big Data Analysis for Prediction of Fuel Consumption
Miyeon Jeon, Pusan National University, Busan/ Republic of Korea, myjeon0725@[Link]
Yoojeong Noh, Pusan National University, Busan/Korea, yoonoh@[Link]
Kyung Hwan Jeon, POSSM, Busan/Republic of Korea, khjeon@[Link]
Sang Bong Lee, Lab021, Busan/Republic of Korea, sblee@[Link]

Abstract

ig data pre-processing improves the data quality and also obtains the feature selection through rule-
based data imputation, denoising, clustering, and compression.

The proposed big data analysis

can be used to plan the navigation strategies and improve energy efficiency for real maritime and ship
data with intelligent decision support capabilities.

[Link]

A smart ship is actively being studied as the next generation ship, which is capable of effective ship
management such as engine monitoring, route optimization, ship diagnosis and remote control in marine
and onboard with development of Internet of Technology (IoT). A big data technology including
collection, processing, cleansing, and analysis of a large-scaled data is key-element of the smart ship to
support the

Fig.1: Smart ship in shipbuilding and maritime industry

86
 Fig.2: Overall scheme of a big data analysis for smart ship

As shown in Fig.2, after data including engine and navigation information are collected from onboard
ship using digital acquisition system (DAQ), a large-scaled data is stored in a data storage. These data
through various machine learning models and data
analytics methods. The framework enables to allow operators to use machine learning and analytics and
amplify sophisticated intellect through cognitive service, and then, dashboards is designed to visualize
and make insight of ship and maritime data. An intelligent system helps people or automated system to
act through these decision-making systems.

Data analysis of ship performance monitoring for the smart ship is especially has been expanded to
enable to improve energy efficiency and eco-friendly operate the ship from the relationship with ship
, weather
consumption.
and navigation, it is important to develop the prediction model for
performing big data analysis of the ship and maritime data. Likewise
consumption can be also
consumption and maximizing the fuel efficiency.

In previous studies, the prediction of ship performance including

studied by many researchers. A
process was performed by Yuan and Nian (2018)
by using artificial neural network in Jeon et al. (2018). Soner et al. (2018) used a decision tree model,
which is a decision-making tool using tree structure,
However, the previous studies have not used real ship and maritime data for
consumption, and used statistical method or machine learning method has been applied only for small
amount of real ship maritime data. Thus, these methods have some limitations to be generally used for
large amounts of data including various errors and bias.

In this study, big data analysis framework based on real ship and maritime data is proposed to predict
using various machine learning models with optimized hyperparameters. In the
framework, original data with various types of errors collected from bulk carriers are refined through
data imputation based on rule-based policy, data cleaning, clustering, and data compression. The
proposed framework is validated using large scale data in real ship marine field and meta-learning using
several regression models is proposed for accurate prediction of real ship and maritime data.

87
2. Methodology

This framework consists of two parts: data preprocessing for improving data quality and data regression
process for using meta-learning. Data preprocessing include a rule-based data imputation, data
clustering, data denoising, and data compression for feature extraction. After the data preprocessing is

regression. The data regression includes initial data regression process to make three machine learning
models such as Generalized Linear Model (GLM), Gradient Boosting Regressor (GBR), and
Multivariate Adaptive Regression Splines (MARS), and Bayesian optimization process is used to tune
hyperparameters of each machine learning model. Then, the meta regression process is performed to
combine three machine learning models to one meta model.

As using meta-regression model through this framework, each machine learning model has different
In this study, meta-regression
models yields more stable predictions results each model alone, and this regression process can be
successfully worked for accurate predictions of fuel consumption in real ship maritime data.

Fig 3: Big data analysis framework of prediction of

3. Case Study

3.1. Data preprocssing

Real ship maritime data are used to validate the proposed framework in this case study. The time series
data sets are collected once every ten seconds for three months from a bulk carrier. A detailed vessel
information length, beam, and maximum rated speed, is skipped for security reasons of
the shipping company. The used data have 41 variables about ship operation, navigation, ship state, and
weather condition as follows.

Engine Operation: Main engine fuel consumption, engine power, shaft speed, etc.
Navigation Speed: Speed Through Water(STW), Speed Over Ground(SOG), Ship heading, etc.

88
 Ship state: Forward draft, After draft
Weather Condition: Relative wind speed, Relative wind direction, Wave speed, etc.

The collected raw data xraw includes the corrupted data such as noise, error and bias data. To refine the
corrupted data, data imputation process is performed to remove the outliers and select the needed
operation region where used data set in this case study include abnormal data occurred while vessel
moored or anchored. Since the purpose of this study aims to accurately predict fuel consumption
the data sets also include vessel navigation data during
navigation at the sea. Table I shows the rule-based policy for data imputation.

If the collected datum is not in range with lower and upper bound of limits defined in the policy, datum
is considered moored ship condition, thus, this datum is imputed as a value defined in the solution. In
this study, two input variables, shaft revolution and Speed Over Ground (SOG), are used for data
imputation to conserve a large-scale data though rule-based policy as shown in Table 1. In the event of
docking or anchoring in these data, the data sets outside of the range in Table 1 were replaced with the
ones at a low speed to remove abnormal data, and only dataset with high-speed operation ximp were used
Figure. 4 shows the scatter plot of SOG according to shat revolution
and portion of given data sets either at navigation or moored status.

Table I: Rule-based policy for data imputation

Code Input variables Range Error Solution Category

-vLB< v < vUB v >vUB , v < vLB Ship

Moor1 Speed Over Ground(SOG) v=0
Speed

-xLB< x < xUB x >xUB , x < xLB Engine

Moor2 Shaft Revolution x=0
Operation

After the data set ximp are refined, features are extracted to classify high frequent draft operation regions
in data clustering process. In this study, the k-means clustering algorithm, which is a method of
unsupervised learning, was used to separate the two subsets of laden and ballast vessels using forward
draft and aft draft affecting as shown in Fig.5.

The number of cluster K needs to be selected by users through K-mean clustering, and silhouette
analysis is performed to measure the performance of the clustering by calculating the separation
distance between the obtained clusters. The silhouette plot displays a measure of how close each point
in one cluster to points in the neighboring clusters, and thus, its value can assess the appropriate number
of clusters. In Fig.6, when the number of cluster K is two, the data sets are clustered into two subsets of
laden and ballast vessels and their silhouette values are more than 0.7, which means that the data
clustering is well performed and data are distributed in a dense region. A large scale data sets, thus, can
be effectively classified and used to perform regression analysis through data clustering process. In this
study, the subsets xladen are considered as laden
consumption.

In the next step, data denoising process is performed to remove the outliers in the resulting classified
subset (xladen) and smooth them using Kalman filter. The Kalman filter is a method of filtering data
based on a priori state estimated from previous times and posterior state using update information as
combining with current observation information and priori state. Figure 7 shows the plot of the denoised
data for SOG.

The initial and variance values of the parameters in Kalman filter used the mean and variance values of
the dataset, and the covariance matrix of progress noise and measured noise has 100 and 0.01 as default
values, respectively. The denoised data follows the trend of previous data and their curve become

89
smoothen in rapidly increasing or decreasing region. All data related to engine operation, navigation,
and weather condition were cleaned or smoothed through de-nosing preprocessing as well as the SOG
data.

(a) SOG according to shaft revolution (b)

Portion of data at navigation and moored status
Fig.4: Data imputation process for selection of navigation dataset

Fig.5: Data clustering of draft following the number of cluster K

Fig.6: Silhouette values for K = 2 Fig.7: Data denoising of SOG

However, the number of input variables is still high to generate machine learning models, so that it need
to be reduced. In the data compression process, the denoised data xfilter using Kalman filter is compressed
to reduce the dimension of data, and correlated variables are combined through Principal Component
Analysis (PCA) and correlation Analysis. As results of PCA and correlation analysis, the input variables
are reduced to four variables such as shaft torque, main engine power, shaft revolution, and speed
through water

90
3.3 Regression model of ship fuel consumption by using meta-regression

With preprocessed dataset xcorr,

using various machine learning models. All data are divided into training data sets for learning model,
valid data sets for tuning the parameter of regression model and test data sets
performance with ratio [Link], respectively. Test data sets are split into sequential time-series data sets
, and training and valid data sets are
randomly selected.

In this study, the predictive model is generated from the machine learning regression analysis:
Generalized Linear Model (GLM), Gradient Boosting Regressor (GBR), and Multivariate Adaptive
Regression Splines (MARS). It is because they yield the most reliable and accurate prediction results
as results of validation tests for various machine learning models, which are widely used in ship and
maritime data analysis.

The GLM model is one of additive models using different basis functions, which is based on an
assumption that input variables follow normal distribution. The GLM model is a non-parametric model
and has the advantage of expressing nonlinearity as reason that each input variable can have a different
type of basis function. In general, since ship and maritime data have highly nonlinearity between input
and output variables, the GLM model is known to have the high accuracy for predicting the ship
performance.

The GBR model is an ensemble model combined with weak learning predictive model such as decision
tree by using error correction and tuning hyperparameter of next models. Accordingly, the GBR model
can possibly generate regression models with extremely fast computations or extremely high accuracy
depending on hyperparameters. In ship maritime big data, this method has some merits in terms of
having characteristics of the decision tree as insight of deterministic method and having extremely high
accuracy.

The MARS model is a predictive model using a linear combination of base functions in a non-
parametrical way. Since MARS model automatically models nonlinearity and interaction between
variables, it has advantages especially for high dimensional data with high nonlinearity. Further, it can
prevent generating overfit models, and thus, it can be generally used for ship maritime big data.

Each parameter has different hyperparameters and its performance could be different depending on
hyperparameter values, so that it is necessary to optimize the hyperparameters of each model. In this
study, Bayesian optimization method is used. The Bayesian optimization repeatedly performs the
exploitation and exploration steps to find the optimum hyperparameters. The exploitation step is a
process of finding the most uncertain and the most likely to be improved observation point by
minimizing the acquisition function based on the observed hyperparameter values. The observation
points are updated by performing model training using evaluation function such as R2 or Mean Squared
Error (MSE) in exploration step. When the stopping criteria are satisfied in the loop of exploitation and
exploration step, the optimal hyperparameter with the highest evaluation function is obtained. Bayesian
optimization method is more often used than other typical methods of tuning hyper parameters such as
grid search and random search because the optimization process is very efficient especially for high-
dimensional hyperparameters. For Bayesian optimization, the lower and upper bounds for each
hyperparameter are given in Table II.

As the optimum parameters for each machine learning model are extracted, it is possible to generate
more . Even though optimized
hyperparameters could improve accuracy of models, it cannot still guarantee reliability and accuracy of
each model because machine learning models have different characteristics depending on data quality,
data quantity, dimension of input and output variables, and their linear/nonlinear relationships.

91
For example, GLM model has the advantages of expressing nonlinearity, but it has the disadvantages
of having difficulty to overfit some models. In the GBR model, although high accuracy is mostly
observed, the GBR model sometimes yield high errors in modeling highly nonlinear data in a
deterministic way. The MARS models is simple to use and easy to be generalized, but it might have the
relatively large errors for highly nonlinear data.

Thus, a meta regression needs to be performed to guarantee reliability and accuracy of models
regardless of data errors, dimensions, or etc. The meta regression is an ensemble model that combines
multiple machine learning models to robustly
big data. Each machine learning model is obtained using valid data, and then, a linear regression is
performed to get weights on each machine learning model. After the meta regression is performed, the
predicted results for fuel consumption are obtained, and then, big data analysis process for predicting
now finished.

Table II: Search domain for Bayesian optimization

Algorithm Hyperparameter Lower bound Upper bound
Number of splines 10 100
Generalized Linear
Spline order 1 9
Model(GLM)
Smoothing parameter 0.0001 1

Gradient Boosting Maximum tree depth 100 1000

Regressor(GBR) Minimum samples leaf ratio 0.001 0.5
Minimum search points 10 5000
Multivariate
Adaptive Thresh 0.001 0.1
Regression Penalty parameter 1 10
Splines(MARS)
Maximum order 1 5

4. Results and Discussion

The results of the regression analysis are presented to verify the accuracy of each machine learning
model by calculating error between the measured (target) data and predicted data with 4 error
measurements: Mean Absolute Percentage Error (MAPE), Relative Root Mean Square Error (RRMSE),
Coefficient of residual Mass (CRM), and Coefficient of determination(R2) as shown in Table III.

Each error measure represents various characteristics of errors between the target data and predicted
data. The MAPE calculates the variance of errors with percentage, and the RRMSE measures the degree
of the error deviation with percentage where errors of less than 10% indicate the prediction model is
reasonably accurate, Despotovic et al. (2016). The CRM is a measure of the tendency of the model to
overestimate or underestimate the target data. If the CRM value is close to zero, it means that the model
accurately predicts the target data. If it is larger or less than zero, the model underestimates or
overestimate the target data, respectively, Li et al. (2013). The R2 is a measure of how close the data
are to the fitted regression line. Values of each error measure are estimated for three machine learning
models and ensemble model to verify the accuracy of each model for the valid data set and the test data
set.

The results of regression models for different machine learning techniques are repeatedly obtained 50
times from the same training and valid data sets before the hyperparameters are adjusted, and then, the
variation in the initial model is examined and the accuracy of each model is compared.

92
 Table III. Error measures
MAPE RRMSE CRM R2
n n n n
1 2
yit yi*
2
yit yi* yit yi*
100% n
yit yi* n i 1
i 1
100
i 1 i 1 1 2
n yit 1 n n n
1 n
i 1
yit yit yit yit
n i 1 i 1 i 1 n i 1

yit : the true value of i-th datum, yi* : the predicted value the true value of i-th datum

The results were presented using box plots for each model as shown in Fig.8 (a)-(d). The MAPE and
RRMSE values in Fig.8 (a) and (b) show similar performance in estimating errors, and all machine
learning models has good fit to actual data because MAPE and RRMSE have errors of less than 5% and
10%, respectively. The GLM model has the smallest variation of errors, and thus, it shows the best
accuracy among all machine learning models when comparing all error measure values. On the other
hand, the GBR model has the largest variation of errors depending on hyperparameters, and MAPE,
RRMSE, CRM, and R2 indicate the GBR has the worst accuracy among all models. The MARS model
and GLM model has similar accuracy, but MARS model has a disadvantage in the large variation of
the errors comparing with the GLM model. Unlike GBR, two models tend to underestimate the fuel
consumption based on CRM values, so that they can conservatively estimate it in the economic
evaluation of ship operation. Although MARS model has less accuracy than GLM in this case study,
but the computational time needs to be taken into account in order to analyze real time ship and maritime
data. Therefore, MARS model is still advantageous over other models in practical aspects.

The error values of the initial model and optimized models with tuned hyperparameters are compared
in Table IV to check how much Bayesian optimization improves the model accuracy where the data
sets with the highest R2 value were selected. The parameter tuning is performed in the valid data sets,
and the test data set is used to verify the accuracy of each model. Thus, most models have the lowest
error values for the valid data sets rather than for the test data sets. As shown in Table IV, the model
accuracy is improved by parameter tuning using Bayesian optimization for all three models. In
particular, the GBR model was the least accurate in Fig.8, but its accuracy is significantly improved

93
after the parameter tuning because it is the most sensitive to hyperparameter values among all three
models. On the other hand, the GLM and MARS model has slightly increased the accuracy due to less
sensitivity to the hyperparameter values. However, it is still difficult to select an appropriate model for
general ship data because each model could have different performance according to data quality and
quantity, data preprocessing process, etc. Thus, it is necessary to derive a model with consistently good
accuracy regardless of data state.

Table IV: Comparison of error values before and after Bayesian optimization

MAPE RRMSE CRM*1000 R2

Valid Test Valid Test Valid Test Valid Test
Initial 0.4138 1.1897 0.6863 1.6142 -0.0468 -8.9344 0.9930 0.8285
GLM
Optimum 0.3652 0.9568 0.6270 1.4834 -0.0573 -4.3303 0.9942 0.8554

Initial 2.7433 1.9813 4.0192 2.4825 0.2266 8.0011 0.7091 0.5323

GBR
Optimum 0.0265 0.7254 0.1095 1.2508 0.0043 -4.9935 0.9998 0.8972
Initial 0.9868 1.2740 1.5195 1.9734 -0.0980 -6.7485 0.9623 0.7383
MARS
Optimum 0.4361 0.9349 0.7270 1.5200 -0.0412 -7.5679 0.9922 0.8481

Fig.9 shows the results of GLM, GBR, MARS, and meta regression models with tuned parameters for
estimating the ship
whole, but the predicted data differ greatly from the target data in some regions. The GBR model shows
a tendency to increase or decrease rapidly for a certain region, but it shows accurate prediction for most
target data. The MARS model better describes nonlinearity of target data than other models, and the
meta model accurately predicts the target data similar to the most accurate GBR model.

94
Numerical comparison results of each model accuracy show that the GBR model and the meta model
are the most accurate, and the MARS model shows the most conservative prediction for fuel
consumption. Although the meta model shows the most similar results with the GBR model, it still has
similar characteristics to the MARS model, which tends to overestimate target data and well describe
nonlinearity.

parameter tuning, but the meta regression model can show more robust performance rather than using
a single model for large amount of real ship data including many errors and missing data.

Table V: Error values for GLM, GBR, MARS and meta regression models
MAPE RRMSE CRM*1000 R2
GLM 0.9568 1.4834 -4.3303 0.8554
GBR 0.7254 1.2508 -4.9935 0.8972
MARS 0.9349 1.5200 -7.5679 0.8481
Meta 0.7241 1.279 -5.0213 0.8976

5. Summary & Conclusions

In this study, big data analysis based on real ship maritime data
engine fuel consumption through data preprocessing and regression analysis. In the data preprocessing,
ship maritime big data were refined by sequential data preprocessing including data imputation using
rule-based policy, clustering, denoising and expansion. The regression analysis uses a meta(ensemble)
model combining three machine learning models (GLM, GBR, and MARS) to obtain a more stable and

machine learning model is more refined to improve its accuracy through Bayesian optimization of its
hyperparameters, and also contributes to improve the accuracy of the meta model.

Although the meta model requires the time-consuming process with large computations including
hyperparameter tuning and meta-regression, it is expected to be more accurate and reliable in predicting
ship performance, which leads to operation and management. From this study, the
predicted fuel consumption can be used to support intelligent decision-making in real ship big data
analysis, and thus, optimize the economic strategy on ship for future smart ships.

Acknowledgement

This work was supported by the National Research Foundation of Korea (NRF) grant funded by the
Korea government(MSIP) through GCRC-SOP (No. 2011-0030013), Korea government(MSIT) (No.
2018R1D1A1A02086093), and National Innovation Cluster Program (P0006887, Build on Cloud
Intelligence Platform based Marine Data) funded by the Ministry of Trade, Industry & Energy (MOTIE)
and Korea Institute for Advancement of Technology (KIAT).

References

JEON, M.; NOH, Y.; SHIN, Y.; LIM, O. K.; LEE, I.; CHO, D. (2018), Prediction of ship fuel
consumption by using an artificial neural network, J. Mechanical Science and Technology 32(12),
pp.5785-5796

DESPOTOVIC, M.; NEDIC, V.; DESPOTOVIC, D.; CVETANOVIC, S. (2016), Evaluation of

empirical models for predicting monthly mean horizontal diffuse solar radiation, Renewable and
Sustainable Energy Reviews 56, pp.246-260

95
LI, M. F.; TANG, X. P.; WU, W.; LIU, H. B. (2013). General models for estimating daily global solar
radiation for different solar radiation zones in mainland China, Energy Conversion and Manage-
ment 70, pp. 139-148

SONER, O.; AKYUZ, E., & CELIK, M. (2018), Use of tree based methods in ship performance
monitoring under operating conditions, Ocean Engineering 166

YUAN, J.; NIAN, V. (2018). Ship Energy Consumption Prediction with Gaussian Process Metamodel,
Energy Procedia 152, pp.655-660

96
 A Short History of Hull Cleaning and Where Do We Go Now
Simon Doran, HullWiper, Dubai/United Arab Emirates, [Link]@[Link]

Abstract

This paper describes in brief how traditional hull cleaning methodology has developed from Ancient
times to the now Robotic and Semi-Autonomous 21st Century, what the associated risks are when hull
cleaning is carried out and what
general safety and the Ocean Environment.

1. Introduction to Hull Cleaning

Around the world upwards of USD$5.7 billion is spent every year to prevent and control marine fouling
and we all know now that marine Biofouling is associated with the largest percentage of invasive issues,
while Ballast Water is the second largest. Invasive Species are now an environmental emergency and
as you will know the IMO call
economic well-

Starting in Ancient Egypt. The most famous example of their shipbuilding skills was the Khufu ship, a
vessel 44m in length. An example was found intact in 1954, entombed at the foot of the Great Pyramid
of Giza, and carbon dated to around 2500 BC, made from cedar wood brought from region that is now
Lebanon. The first recorded treatment for ship efficiency comes from an Aramaic scroll dated about

evenly applied to the vessels sides so that she may speed through the blue waters freely and without

97
By the 3rd century the Romans and Greeks were leading the way with shipbuilding and sea exploration,
and it is recorded that they were using tar and wax to coat their ships bottoms.

Between the 8th and 11th century, the rampaging Vikings are recorded as being vigilant in cleaning the
hulls of their ships for speed and efficiency of sailing and with it came the use of pitch, oil, resin and
tallow for hull treatment. Between the 13th and 15th centuries, the Chinese were pushing exploration
from the East on bigger and faster ships and they had the hulls of their ships coated with lime mixed
with poisonous oil to protect the wood from worms. Christopher Columbus also recorded similar
attempts to deal with this invasive problem for his ship, the Santa Maria, stating all ships bottoms were
covered with a mixture of tallow and pitch in the hope of discouraging barnacles and teredo (the

This was the same for Buccaneers and Pirates who recognized the benefits of a clean ship. Having
ed they knew that they
were then faster in the water so they could the catch commercial vessels trying to run away, or outrun
Navy vessels sent to sink them.

Lord Horatio Nelson was a supreme tactician, even though the admiralty did not always agree with his
unconventional tactics. Prior to the battle of Trafalgar, he is known to have ordered his entire fleet of
27 vessels to be hull cleaned. Since he was at a numerical disadvantage in fleet size (with only 27
against the 33 French and Spanish ships), he knew he needed the advantage a clean hull gave him in
speed to even the odds. He then used his fleet s superior speed and agility to famously maneuver into
two columns directed perpendicular against the French and Spanish and delivered a decisive
, the result being 22 lost French and Spanish ships to not a single British loss. So, the speed
advantages of a clean ship were known then as they are today.

Charles Darwin raised the first questions about the risks in transfer of invasive species attached to ship
hulls when in 1836 sailing on
. It
was around these times that ships were starting to advance from wood and sails into iron and steam.

The finest example of this was the SS Great Eastern, an iron sailing steamship designed by Isambard
Kingdom Brunel. When she was built in 1858, she was by far the largest ship ever at 211m (and with

98
a carrying capacity of 4,000 passengers to be transported from England to Australia, without the need
for refueling). Over the next century, all sorts of antifouling paints were used, with varying ingredients
such as lacquer, powdered iron, red lead, tannin, shellac, hot plastic paints then cold plastic paints, to
name a few, and by the 1960s the leaching antifoulings were introduced.

The most effective and well-known ingredient used in the anti-fouling paints, TBT, was eventually
banned worldwide in 2003

Today the marine coating business is a multi-billion-dollar not all

plain sailing as the coating industry faces challenges, with some quarters asking for heavy metals to be
removed from all coatings. So even here new solutions are needed.

But despite this long journey and all of these advancements, the problems that have been around for
thousands of years still exist today, and no clear and present solution is in place which is agreed
globally.

- Biofouling is still an invasive species risk.

- Biofouling still impacts speed and performance

Added to the above, hull cleaning and hull cleaning methodology is/was woefully behind. Traditional
cleaning goes on in many locations. This consists of Divers with hand held scrubbers, or driving
underwater brush carts, marine fouling is aggressively removed from the submerged areas and the
coatings are damaged to varying degrees dependent on the cleaning system and operator controls. The
brush pressure during cleaning can additionally remove upwards of 100mcrons of coating in one clean.
These cleaning systems are effective to a degree however most, if not all, lack fouling collection
systems, and as a result increase inherent risks to both the environment and to life. The fouling is
scrubbed off the submerged area along with heavy metals and allowed to fall within the tidal stream.
In the past two years at least 4 x divers have been killed whilst hull cleaning.

The importance of environmental standards and HSSE has dramatically increased within corporate
culture to the point where the marine industry now considers this to be the norm or is it? Some ship

99
owners, even Oil majors with very robust corporate cultures who set the highest of standards, are often
still driven by the cheapest solution.

By 2003 the first Remotely Operated Vehicle (ROV) with marine fouling capture technology was
introduced to commercial shipping and ROV hull cleaning became an option, whilst at first limited,
now there are known to be 6 x commercially viable ROV Hull Cleaning service providers and more
than another 5 x companies undergoing development and testing. Hull Cleaning with marine fouling
capture systems are here now, and some of these systems do not require divers to drive them, as they

HullWiper and others similar to HullWiper are becoming more and more an option and are diver-free,
cost-effective and environmentally friendly ROVs. Most ROVs are equipped with water jets and
designed to clean from 1000m² up to 2000m² per hour. Utilizing salt water under variable pressures as
the cleaning medium Collects pollutants removed from the vessel s submerged areas for ashore disposal
in an environmentally approved and eco-friendly manner and can clean ships whilst taking bunkers or
during loading/discharging cargo, saving valuable time for vessels

Hull Cleaning is seen as a necessary evil but regulations are now changing, by prohibiting hull cleaning
in ports or restricting divers to daylight hours only, this is proactively influencing development of ROV
hull cleaning with fouling capture systems as an added requirement.

State, federal and international regulators will positively affect the hull cleaning industry and if more
biofouling regulations are implemented and enforced, hull cleaning with fouling capture units may
become the only acceptable option.

Add to the pot, the IMO GloFouling Partnership project, launched in November 2018, is focused on
dealing with the transfer of invasive aquatic creatures on ship hulls. This is aligned with the Ballast
Water Management convention; however, the Ballast Water convention took the best part of 20 x years
to complete and has/will cost vessel owners millions to comply with. The IMO will potentially complete
the GloFouling directive in 2023, which will then help to push stricter regulations worldwide.

100
Secondly BIMCO are progressing forwards to bring together different stakeholders including
HullWiper and our friendly competitors to provide input on new hull cleaning standards that will enable
ports to understand who has been cleaned properly by who and all associated data. This will help in
providing a filter to the relevant authorities to check more effectively and will align well with SEEMP
- the Ship Energy Efficiency Management Plan released by the IMO in 2013.

So what does the future look like for Hull Cleaning and biofouling?

Once upon a time, fossil fuel engines were the future, now today they are a problem and cost so much
to run. Global trade was the future; but no one anticipated that invasive species would wipe out fishing
industries and cost governments billions each year. Antifouling paints were the solution, but no one
fully understood the impact of TBT and or heavy metals deposited into the oceans. For coatings what
is the future? Will there be pressure for strictly, an organic, non-metal compound coating on ships hulls
which removes the threat of heavy metals in the oceans?

101
No one really knows what options will be available in the future and what problems they will be bring,
but ROV Hull Cleaning such as HullWiper amongst others is one solution that gets to the root of the
problem which can have a positive impact.

- Together we can save you money,

- Together we can help save the environment
- Together we help save lives.

102
 A Digital Business Model for Vessel Performance Monitoring
Serena Lim, Ascenz, Singapore, [Link]@[Link]
Roger Teo, Ascenz, Singapore, [Link]@[Link]
Teck Chong Sia, Ascenz, Singapore, [Link]@[Link]

Abstract

Hull and propeller performance monitoring is an important aspect of the overall vessel performance
monitoring and management, which is driven economically towards optimised fuel consumption and
environmentally towards a reduced carbon footprint throughout the operational life of a vessel.
Research and development is carried out continuously to improve operational performance, for
example to increase reliability of measurements in terms of sensor technology and data integration as
well as understanding the effects of vessel draft, trim, speed, seaway and wind. Benchmarking
capabilities are also being constantly improved to enable users to apply standards for vessel
performance monitoring such as the development of ISO 19030. These multi-directional research
approaches for optimising vessel operations, using new sensor technologies and applying machine
learning techniques can result in the slow adoption of new vessel performance solutions. We therefore
propose here adoption of a digital business model for vessel performance monitoring and to provide a
clear direction to drive the industry towards a sustainable and successful digital ecosystem, which
delivers solutions more rapidly than traditional approaches. The benefits and contribution of different
players in the industry will be highlighted and the digital business model will be discussed as the sector
moves towards Industry 4.0. Results and analysis will be presented to demonstrate the state of
implementation of various standards as part of the business model and to highlight specific development
needs.

1. Introduction

As the shipping industry moves towards an era of increasing digitalization, a deeper awareness of
available resources and solutions in the market to support this trend is important. An ecosystem
describes a community of interacting organisms and their physical environment, in a broader business
context, it can be viewed as a complete network of interconnected systems and businesses. A digital
ecosystem is one in which systems are digitally connected, and the success or failure of each component
will affect those systems to which it is connected. A successful digital ecosystem in the maritime
industry is a distributed, adaptive, open socio-technical system with properties of self-organisation,
scalability and sustainability, inspired by natural ecosystems, Zhu (2015). Digital ecosystem models are
informed by knowledge of natural ecosystems, especially for aspects related to competition and
collaboration among diverse entities.

The root of a successful digital ecosystem lies in the availability of high-quality data and information
that can be generated from databases. Data can be presented qualitatively or quantitatively; the quality
of the data depends of the intended use of data in operations, decision making and planning. Data
accuracy, compatibility and completeness remain a major problem for most companies and industries,
since datasets that are flawed or corrupted can lead to wrong information and incorrect forecasts being
made, Crandell (2017).

Data Acquisition Systems (DAS) have been installed on vessels to acquire high frequency data which
record the performance of the vessel. These data can be used to develop standards, improve efficiencies
of operation and used to develop advice on future designs. The European Union Monitoring, Reporting
and Verification system (EU MRV) and International Maritime Organization Data Collection System
(IMO DCS) were developed to reduce the greenhouse gas (GHG) emissions of shipping activities,
focusing on ships operating within the EU area and in international shipping operations respectively,
EU (2015), IMO (2016). These reports also provide useful inputs and guidance for policy making. The
data collected can also be used to measure improvements in the energy efficiency of ships as part of the

103
Ship Energy Efficiency Management Plan (SEEMP), IMO (2012), which feeds the data into Energy
Efficiency Operational Indicator (EEOI) to benchmark ship performances, IMO (2009). The data from
a DAS are also used to influence ship designs and retrofitting, Peri (2016).

The performance of ships during different operations can be evaluated in real time. Vessel performance
monitoring (VPM) products and services can be viewed in terms of three categories. The first is a Data
Acquisition System (DAS) which logs data collected from sensors on a common database, which can
be viewed on-board the vessel, and telemetry data to shore for real time display of data, Rødseth et al.
(2016). The second is added information that can be provided through monitoring sensors, such as total
fuel consumed in each operation, efficiency of machinery that is monitored, Perera and Mo (2017).
The third is advanced analysis to generate additional information from the available databases, such as
for maintenance base monitoring purposes, hull degradation and vessel performance monitoring. The
third category could also include services with advance data processing such as using machine learning
techniques and artificial intelligence, which could lead to advisory systems, Coraddu et al. (2019).

Preparation for the development of a digital ecosystem framework for ship performance monitoring is
proposed. A background understanding for stakeholders is also provided so that choice of VPM services
and the expected outputs of such systems when choosing providers with different business models can
be facilitated. Areas where there is a need for further development towards an ecosystem driver, which
can provide a cyber physical network (Industry 4.0) for the VPM system are also highlighted.

2. Digital business model framework

A digital business model is therefore proposed for vessel performance monitoring to provide clear
direction and to drive and guide the industry towards a sustainable and successful digital ecosystem,
which delivers solutions more rapidly than traditional approaches. The benefits and contributions of
different players within the digital ecosystem will be highlighted and the digital business model will be
discussed, as the sector moves towards Industry 4.0, a cyber-physical system that connects automation
and data exchange in manufacturing technologies.

The framework proposed for a digital business model for vessel performance monitoring (VPM), shown
in Fig.1
An understanding of the types of digital
business models will help develop a more holistic VPM that serves the needs of the shipping industry
in terms of data acquisition systems (DAS) design and development, as well as data processing and
analysis. In addition, users of such systems can be informed of the type of business models that are
available in the industry and can make the informed decisions when choosing suitable providers.

Fig.1 Business models for the digital era

104
Different business models can be derived from two critical dimensions, namely, the knowledge of end
consumers and the business design. The business models depend on the extent of knowledge of the end
consumer needs, from a lower end of partial familiarity towards a higher
level of complete understanding. The business model also depends on the business design, whether to
focus as a value chain business, such as those which add value to the customers and provide for example,
training and after-sales services, or to further develop into an ecosystem to create a network of
interconnected systems. These dimensions combine to form four business models for creating value,
these are businesses as a supplier, a modular monitoring system provider, a multichannel provider or
an ecosystem driver.

2.1 Supplier

Suppliers, depicted at the lower left quadrant of Fig.1 typically have little knowledge of preference for
selecting their end customer, and usually sell the product through distributors or wholesale to businesses
that use the supplied product or services. The product can usually be mass produced, or the service can
be duplicated with repetition. In order to be sustainable as a supplier, efficiency and price plays a big
role if substitutes can be easily found in the market. This type of business model must continue to
provide reliable products/services and innovate to stand out from other suppliers.

Examples of such businesses in the vessel performance monitoring sector are sensor makers, where the
sensors are designed to measure specific parameters and able to transmit data in a digital form. These
include fuel meter makers and torque meters, sensor makers are also good examples of suppliers of
products that are important components of the VPM system. Data storage companies and businesses
that specialise in machine learning using data collected are examples of suppliers of services that
support the VPM system.

The benefit of being a supplier business is that specialised product/services can be developed. The
ability to mass produce components reduces the cost of product/services. The downside of suppliers is
that often the product/services including analytics or diagnostics are available for the part supplied and
do not generally deliver system wide integrity. For example, companies that provide machine learning
solutions to data can provide statistical reliability of the methodology used but often cannot validate
data sensibility of the source of data.

2.2 A modular monitoring system

A company that adopts a business model approach to provide modular monitoring systems lies in the
lower right quadrant of Fig.1. Such a business model offers a wide range of distinct solutions. The
solution of plug-and-play systems can usually work with several different partners. However, such
systems are limited to specific functions and may not be freely integrated for information transfer.

Examples of such business models adapted to provide VPM systems include companies that only deal
with specific type of monitoring system. The company might only specialize in fuel consumption
monitoring, where a line of products is available to accommodate all sizes of pipelines, all system
configurations and different fuel types. However, due to the complexity of ship systems which are made
up of interconnected systems such as the hot water system, sea water system, fresh water system,
propulsion system, fuel oil system etc., solely relying on such plug and play fuel monitoring systems
does not allow the provision of a wider picture of vessel performances, especially when data errors in
the fuel monitoring arises from different systems such as faults in the fuel injection or faults in engine
misalignment. To be able to measure overall vessel performance, an integrated modular monitoring
system that collects data under one single platform and is able to carry out an overall analysis is needed.

The benefit of such a business model is that a fast set up system is achievable, and specific information
can be connected to the DAS within a short period of time. However, additional advanced analysis is
usually not available and the data output is usually in a rigid format.

105
2.3 Multichannel business provider

A multichannel provider lies in the upper left quadrant of Fig.1. This type of business usually has a
good knowledge of customer needs and has direct relationships with a wide customer database. Products
and services of multichannel businesses are often available online with a friendly user interface.

Examples of a VPM multichannel business provider are those that provide products and services using
a more rounded approach. These companies usually offer sensor selection advices, approved and
reliable sensor installations, DAS to communicate with the sensors, systems for visualisation of data
on-board and on-shore. The level of VPM services can usually be customised and adapted depending
on the available budget and system type. Protected and safe long term and short term data storage
solutions usually accompany such products. Advanced analysis of data using different techniques,
including machine learning techniques are usually offered. More mature companies that adopt such
models also provide advice and consultancy services to support the analysis to improve vessel
performance.

The benefit of such systems is that system assurance and accountability of system integrity is usually

Training to use the VPM system are usually available for customers to enhance the added value of the
system installed. Such businesses however come with higher costs because additional research and
innovation is usually carried out to meet the need of the customers, changes in market and for
customised solutions. More specialised engineers, software designers, and data analysts are usually
needed to support such a business model.

2.4 Ecosystem driver (Industry 4.0)

The VPM system, like the maritime industry is moving towards cyber physical system integration
(Industry 4.0), where the business model of an ecosystem driver is desirable. Such a business model
provides a platform that connects customers needs and inter-connects all products and services available
in the industry. Such a system is constantly updated with new technology and services that can be
offered.

A business which acts as an ecosystem driver has a certain level of smartness embedded which advises
users of the options and services that are available that best fit their needs. For example, a monitoring
system with machine learning abilities can detect the need for hull cleaning of a certain vessel. Different
available options to meet and solve such needs will be suggested to the customers. When the option of
a self-cleaning robot is selected, the robot will be sent to solve the problem, where the location of the
ship is known, and the extent of cleaning required is known. Or when there is a faulty sensor, a database
of suitable sensors that can be used as a replacement will be available almost immediately. Although
industry 4.0 sounds futuristic, it is rapidly becoming a reality and systematic steps must be taken in the
marine industry to ensure that this is achieved.

Such an ecosystem requires sharing of reliable information on a large database, where data from an
engine manufacturer is available for integration with other plug-and-play devices for data processing.
This business model relies on the provision of reliable databases by different players. The maritime
industry is moving towards such systems, where the data and system security of such systems is still
being investigated. The creation of such an ecosystem will none the less promote the creation of new
businesses and form new social networks, Bock et al. (2017). Successful digital leaders in the industry
have proven to have systematic data acquisition strategies and stand out based on the data platform
created, to allow successful prediction and decision making based on reliable and validated data.

2.5 Summary of digital business framework

A good understanding of the type of digital business model selected will also allow better allocation of

106
networking activities to participate in, and technology to invest in. The ecosystem driver aims to connect
systems and connect processes to satisfy supply and demand in the industry. The characteristics,
benefits, limitation and examples of these four digital business model frameworks introduced can be
summarized in Table 1. The ecosystem (industry 4.0) encourages interaction of customers, partners,
adjacent industries and government. Allows all users to tackle cyber security at all levels. A business
ecosystem delivers products and services through both competition and cooperation. A healthy
ecosystem inspires collaboration which aims to provide mutually beneficial results to all parties
involved, encourages diverse thinking and innovation whereby the industry can benefit from provision
of quick solutions. The idea is to create a collection of flexible services that can shift around and quickly
be adapted to the ever-changing needs of customers.

Table 1: Characteristics, benefits, limitations and examples of digital business models

Modular Multichannel Ecosystem driver
Supplier
monitoring system provider (Industry 4.0)

Stands out as big

Provides a branded
Characteristics

Mass production of Plug and play players in the

platform which
offerings, usually product or services industry, usually
ensures an excellent
sold to distributors that are easily meets customer
multifaceted
and subjected to adaptable to the needs and provides
customer
commoditization wider ecosystem integrated value
experience.
chain

Provides system
Constant innovation
integrity
Specialised of product is needed Allows plug-and-
product/service to keep up with play third party
Knowledge to
Benefits

market competition product

customize product
Low cost producer
and services to meet
Fast set up system, Information sharing
s
Incremental able to integrate between customers
innovation with common and providers
Provides training to
sensors
maximise use of
product

Usually only
Limitations

Provides non- Higher costs

provides analytics
specific data associated with
on product/service System security
analysis or single intensive research
developed and not
goal data analysis and innovation
as an overall system

Sensors makers
VPM system with
DAS for individual
offers of solutions
Examples

Data storage system. DAS system with

from other sectors
providers advanced analytics
(i.e. robotics,
Offers generic data and advisory systems
blockchain, new
Machine learning analysis
technologies)
companies

3. Challenges in moving towards an ecosystem of interconnected VPM systems

A review of the different digital business models available shown in section 2 enhances our

107
understanding of components needed to achieve an ecosystem of connected VPM systems, where the
knowledge of end users is important and that an ecosystem needs to be established. The challenges of
moving towards the ecosystem driver business model are now presented so that steps can be taken to
overcome these challenges. In order to develop a holistic vessel performance system, good management
and the centralised use of information from different parties is required. The development must also be
inclusive of knowledge from the seafarers who have first-hand experience at sea and the naval architects
and marine engineers who have design experience, with a good knowledge base of structures,
hydrodynamics, material science and engine performance to check for data sensibility range and data
validity before data analyses is carried out, and to explain results from analysed data. There is also the
need for mathematicians, and computer scientists who understand coding and development of software
and cyber security. To achieve an advisory system, the system must also be properly informed of the
policies of different parties, both local and global. A successful ecosystem driver draws power from the
diversity of different skilled individuals.

All knowledge and expertise from different aspects relating to the development of VPM systems must
be integrated to increase reliability of measurements in terms of sensor technology, data integration as
well as understanding the effects of vessel draft, trim, speed, seaway and wind. The VPM system should
be able to provide services to improve operational performance and be constantly updated to comply
with standards and be adaptable to changes of technology. These data can be processed using
deterministic approaches such as using industrial standards, using statistical approaches and using
various machine learning techniques. Benchmarking capabilities using standards are being constantly
improved to enable users to apply standards for vessel performance monitoring such as the development
of ISO 19030 and applying data filters using ITTC recommendations.

This multi-directional research of optimising vessel operations, sensor technology and applying
machine learning techniques can result in the slow adoption of vessel performance solutions. Research
is a core activity which enables the adoption of new technology and the creation of a resilient system
with regularly updated security systems.

Data reliability and data standardization remain one of the major challenges for a successful VPM
ecosystem. The sharing of information is not yet readily transparent, data collected from vessels are not
synchronised and not readily comparable between different ships and fleets. A multichannel VPM
system can take the lead in ensuring reliable data sharing is available, which will allow delivery of
solutions more rapidly than traditional approaches. Tools must also be developed to measure
performance, for research, for improvements, and to identify key performance indicators.

The digital business model adapted for vessel performance monitoring provides an indication that

towards a sustainable and successful digital ecosystem. Using the data from vessel does not focus on
one component but looks at the overall performance of the vessel. Understanding the business models
allows providers to achieve long term investment potential.

4. Reliable databases for further development

Most vessel performance monitoring systems available on the market to date can act as a modular
monitoring system or a multichannel provider. The VPM system design will play an important role in
the transition of the marine industry into Industry 4.0, where reliable databases are required to feed into
the ecosystem. An example of a multichannel provider is
system, particularly in Asia, and is demonstrated here. In addition, the deliberate efforts taken by the
company to encourage the move towards an ecosystem driver business model are highlighted.

4.1 A structured and reliable database

Standardisation of data architecture is one of the key challenges for industries to move towards an
ecosystem driver model, Weyer et al. (2015). Fig.2 shows a methodology adopted for ships to ensure a

108
structured and reliable database is available for use and for integration with other databases. Row A
shows the categorisation of the database, row B shows the required qualitative and quantitative time
stamps of each data set, row C shows the independent data validation step needed for each category and
row D shows examples of analysis that can be carried out in each category. The categories of data can
be broken down into data output from individual sensors, data from an individual system and integrated
systems within the ship and immediate data collected from external systems. The on-board data
acquisition unit and computer can process these different data categories. The time stamp of each
parameter is recorded and the unit is standardised to enable further integration of this database into
external databases.

Fig.2 Developing a structured and reliable database

Independent data validation is required for each category to ensure that the data is reliable when carried
forward for further analysis. After independent data validation, each category will be able to provide
separate information on the ship. A detailed example is further shown in Fig.3. For example, in an
individual sensor category, such as a Coriolis flow meter, the data collected from an individual sensor
can be analysed to provide information about the health of the sensors and to inform operators if any
calibration is required, or if the sensor integrity has been compromised. The vector and scalar quantity
of the parameter monitored can be derived, and information such as flow condition and temperature
changes at the location of the sensor can be monitored. Analysis that can be derived from an individual
system such as the fuel consumption system can be used to inform the actual fuel consumption during
each operation and allows understanding of the system health and assurance of the system. The data
can also be used to further inform the estimated emissions.

When this system is integrated with other ship systems such as the propulsion system and the engine
system, analysis can be carried out to inform the efficiency of the machinery. Further detailed analysis
of frequency, noise and vibration can be used for fault detection. This information can be used to
understand system compatibility and test the possible boundaries for system improvement. Higher
levels of analysis can be carried out using data from integrated systems, such as to understand the energy
flow throughout the entire vessel under different conditions and to provide recommended optimum
operational advices. An example of a common external system integration is the metocean data which
is useful when directly integrated within the ship data system, to provide instantaneous operational
analysis and forecasted routing to ensure safety, comfort and minimise fuel consumption.

A multichannel provider must have a good grasp of the market and customer needs and is able to provide
a structured and reliable database of a ship. This can be done by standardising the data collection process
and by interaction with suppliers and customers. Every category must be validated to ensure that the
ultimate ship database is reliable and readily available for integration with other databases. The outlook

109
of an ecosystem driver is the integration of different databases to provide a well connected cyber
physical system to provide an overall cost optimisation, ensure sustainable growth of maritime
businesses, using data for policy making and to tackle climate change. A systematically thought-out
data collection process will generate a reliable database that can be used to make predictions and
business decisions as well as ensuring market security.

Fig.3 Data integration of individual sensors required for the development of an ecosystem driver
business model

Fig.3 shows that the data collected from individual sensors makes up the fundamental data for individual
systems and is used for integration between different systems. The data collected under a single data
acquisition unit is time stamped and sent to an on-shore database, contributing to the fleet database,
which will be able to feed into the ecosystem driver for integration with external databases such as port
and market databases.

4.2 Database of ship operational parameters

Prior to analysis carried out on vessel data, a good understanding of each data source and also each data
validation process must be carried out. An example of data validation for an individual system such as
a mass flow meter is to compare the measured fuel density, temperature and flow rate of the sensor.
Fig.4 shows an example of data collected from a Coriolis fuel flow meter that is measuring marine gas
oil. The data from the sensor shows that the density of the fuel oil is approximately 840kg/m3, which
falls within a confidence limit of the expected value. Further validation can be carried when required
from the bunkering notes of the vessel. The fuel temperature at the supply line moving through the
sensor shows a steady temperature of approximately 40 C. This information indicates that the sensor is
providing sensible data and is working in a healthy manner. Boundary conditions can be set to enable
alarm triggers when deviation of data from expected levels are recorded, to trigger further investigation
into system health. The data collected from sensor can also be used to indicate sensor tampering or
situations where the fuel in the pipeline is not a single-phase flow.

110
 Fig.4 Example of data validation for an individual sensor

Vessel speed over ground (SOG) is derived using global positioning system (GPS) on-board the vessel
whilst vessel speed through water (STW) can be obtained using different types of approach such as
using electromagnetic, pitometer and Doppler methods. Fig.5 shows histograms that represents the
difference in data for SOG and STW using two different sets of data. Table 2 shows the comparison of
distribution of data. In reality, not all vessels have a speed through water sensor. Fig.5 and Table 2 show
that there could be obvious differences when using SOG and STW to analyse vessel hull and propeller
performances.

Fig.5 Comparison of vessel speed data: speed over ground and speed through water

Table 2: Distribution of data using measurement of SOG and STW

Data set Mean Standard deviation
13.23 1.162
1 (left)
12.70 0.752
13.05 1.148
2 (right)
13.06 0.917

Once all the parameters logged in the data acquisition system are validated, analysis can be carried out
using the full data set. An example of useful further analysis is to understand the power needed to move
the vessel through the water. The data used to plot Fig.5 were filtered for vessel operations and only
considered when the vessel is in transit operation. One month worth of data is shown in Fig.6. A
continuous monitoring of such data trends can be used to estimate the hull and propeller degradation
through time, whereby losses in hull and propeller efficiency can be calculated. In addition, the fuel
consumption by the engine to generate shaft power is also monitored. The fuel consumption and shaft

111
power comparison can be used to calculate the specific fuel oil consumption of the engine, this value
can be monitored to observe the changes in engine efficiency and acts as a tool to prompt maintenance
services and measure changes of engine efficiency after maintenance works are carried out.

Fig.6 Example of shaft power vs vessel speed through water

5. Conclusions

The sustainable way of moving forward towards an ecosystem driver business model for interconnected
VPMs is to carefully but quickly bring together and integrate the expertise of people with emerging
technologies. A sustainable digital business model can improve performance of operations, provide
economic incentives for different stakeholders, and even provide data-bases which provide evidence to
underpin and justify advice given on shipping conduct within a complex and ever changing political
and environmental industry landscape. From the examples given, it is clear that a well thought through
digital ecosystem business model can be further enhanced by using carefully structured databases,
improving the quality and performance of individual ships, the entire fleet, and the businesses
themselves. The next step is to take advantage of reliable databases developed from VPM systems and
to integrate them safely and efficiently with other systems such as port authorities and systems within
the supply chain of trades and services.

References

BOCK, R.; IANSITI, M.; LAKHANI, K.R. (2017), What the companies on right side of the digital
business divide have in common, Harvard Business Review

CORADDU, A.; LIM, S.; ONETO, L.; PAZOUKI, K.; NORMAN, R.; MURPHY, A.J. (2019), A
novelty detection approach to diagnosing hull and propeller fouling, Ocean Engineering

CRANDELL, C. (2017), The problem with data is too much the wrong stuff, Forbes, [Link]
[Link]/sites/christinecrandell/2017/06/26/the-problem-with-data-is-too-much-of-the-wrong-
stuff/#1d429a12d5f4

EU (2015), Regulation (EU) 2015/757 of the European Parliament and of the Council of 29 April 2015
on the monitoring, reporting and verification of carbon dioxide emissions from maritime transport, and
amending Directive 2009/16/EC, Official Journal of the European Union, L 123/55, 19.5.2015

IMO (2009), Resolution MEPC.1/Cric.684, Guidelines for the voluntary use of the ship energy
efficiency operational indicator (EEOI), International Maritime Organization, London

112
IMO (2012), Resolution MEPC.213(63), 2012 Guidelines for the development of a ship energy
efficiency management plan (SEEMP), International Maritime Organization, London

IMO (2016), Resolution MEPC.278(70) Data collection system for fuel oil consumption of ships, IMO
Doc MEPC.70/18/Add.1., International Maritime Organization, London

PERERA, L.P.; MO, B. (2017), Marine Engine-Centered Data Analytics for Ship Performance
Monitoring, J. Offshore Mechanics and Arctic Engineering 139(2), p.021301

PERI, D. (2016), Robust Design Optimization for the refit of a cargo ship using real seagoing data,
Ocean Engineering 123, pp.103-115

RØDSETH, Ø.J.; PERERA, L.P.; MO, B. (2016), Big data in shipping - Challenges and opportunities,
15th COMPIT Conf., Lecce

WEYER, S.; SCHMITT, M.; OHMER, M.; GORECKY, D. (2015), Towards Industry 4.0 -
Standardization as the crucial challenge for highly modular, multi-vendor production systems, IFAC -
Papersonline 48(3), pp.579-584

ZHU, P. (2015), Digital Master: Debunk the Myths of Enterprise Digital Maturity, Lulu Press

113
 Distributing Real-Time Measurements of
Speed Through Water from Ship to Shore
Gunnar Prytz, Miros, Asker/Norway, gp@[Link]
Rune Gangeskar, Miros, Asker/Norway, rg@[Link]
Vemund Svanes Bertelsen, Miros, Asker/Norway, vsb@[Link]

Abstract

This paper describes how latest generation radar-based remote sensing solutions give high-quality
information about ocean wave parameters such was wave heights, directions and periods, surface
current magnitude and direction and ship Speed Through Water (STW). Combining the sensing
technology with Internet-of-Things technologies allows making accurate sea state data available in
real-time both onboard and onshore. This enables significant improvements in such applications as
hull fouling estimation and speed optimization. This paper will present such a solution in detail together
with some examples from testing on vessels.

1. Introduction

The shipping industry is currently undergoing a transformation due to digitalization. A strong focus on
cost of operations, operational efficiency and on the environmental aspects associated with shipping
are some of the main driving forces behind this development.

Situational awareness is a necessary ingredient in the digitalization process. One area that has seen
considerable improvements recently is within real-time sea state measurements. Recent developments
within radar-based technologies have given access to accurate sea state data that can be used to optimize
ship operations, Gangeskar (2017,2018a,2019), Gangeskar et al. (2018). State-of-the-art radar-based
sea state measurements can measure both ocean waves and ocean currents accurately under widely
varying conditions and with high availability, reliability and accuracy.

Sea state has a significant impact on ship performance. This holds true for both ocean waves and ocean
currents. There are intricate relationships between waves and ship performance requiring advanced
models that take into account such factors as 3D hull properties and loading conditions. The situation
is somewhat simpler when it comes to ocean currents. Currents coming against the direction of ship
motion means that more water needs to be displaced per time unit compared to a situation with no
current. Similarly, currents travelling in the direction of ship motion means that less water needs to be
displaced per time unit. Hence, the current component going in the direction parallel or antiparallel to
the vessel heading has a major influence on vessel performance. Currents travelling perpendicular to
the ship motion might also lead to a need to spend energy to counter the forces inflicted by the currents.
Thus, the presence of ocean currents has a profound influence on the performance of the vessel.

The Speed Through Water (STW) parameter is the vessel speed with respect to the water. STW is equal
to the Speed Over Ground (SOG) when there is no ocean current present. SOG is easily measured by
means of a GPS receiver. STW, however, has not been easily measured in an accurate and reliable way
until now, Gangeskar (2019).

Ocean surface current measurements from moving vessels by traditional underwater (in-situ)
instrumentation are associated with challenges and data heavily influenced by noise. Systems
measuring the speed through water (STW) are equally influenced by similar disturbances affecting the
vessel speed log, Antola et al. (2017), Baur (2016), Bos (2016), Fritz (2016). Wave measurements from
underwater instrumentation are only available on rare occasions. The following items are relevant for
both acoustic Doppler current profilers (ADCPs), Flagg et al. (1998), King et al. (1993), New (1992),
and other instruments based on traditional in-situ measurement principles.

114
 Underwater equipment generally involves installation and maintenance procedures being both
time-consuming and expensive.
Underwater equipment is exposed to fouling, Carchen et al. (2017), Goler et al. (2017), Kelling
(2017).
Measurements are disturbed by air bubbles, turbulence, and inhomogeneous hydrodynamics
caused by the vessel motion and propellers, Bos (2016), Carchen et al. (2017), Brown et al.
(2001).
Measurements are disturbed by other instruments, for instance acoustic echo sounders and ves-
sel speed logs.
The surface current itself is considerably affected by the vessel motion.
Sensors are frequently inadequately calibrated, Antola et al. (2017), Bos (2016), giving sys-
tematic errors in certain speed ranges, Antola et al. (2017).

A vessel has an optimal speed which in simple terms depends on the speed vs. fuel relationship of the
vessel and the efficiency characteristics of the propulsion configuration (e.g. the propellers). Ocean
currents of up to several knots can exist on the oceans which means STW might be quite different from
SOG. It is therefore STW and not SOG that should be used as the basis for vessel performance
calculations, i.e. how fast is the vessel moving with a given supply of fuel. Thus, STW is a very
important parameter in ship performance optimization.

There are several vessel applications that will benefit from accurate STW measurements. Hull
performance is often analyzed by investigating the amount of fuel consumed at a given speed. Hull
fouling will lead to increased friction and thus increased fuel consumption at a specific speed. This is
typically based on SOG measurements from a GPS or heavily filtered STW measurements from
underwater sensors. Hull cleaning can be a very expensive process and thus it is important to estimate
the actual hull condition as accurately as possible. Thus, accurate STW measurements can be used to
improve planning of hull cleaning or to investigate the effectiveness of hull cleaning procedures or hull
coatings. Related use cases might be related to performance degradation of parts of the drivetrain, e.g.
the propellers.

While hull performance estimations can be made in retrospect with historical data of medium to low
time resolution, there is another very important application that benefits from having access to real-
time STW measurements. A vessel has an optimal speed where the fuel consumption is the lowest. The
optimal speed is measured relative to the water, i.e. accurate STW measurements are required.
Whenever possible, there is a significant potential for fuel savings by making sure that the vessel STW
is optimal. The fuel savings potential can range up to tens of tons per day for large vessels in areas with
currents of 1-2 knots. Due to the accuracies required, it is in most cases not feasible to rely on theoretical
models of surface current. The actual speed optimization can be done either manually by the crew or
automatically by an autopilot system.

Traditionally, speed control or autopilot systems have been based on GPS input as the STW sensors
have not been reliable enough. With the recent STW solution from Miros it is now possible to have
access to STW measurements that are reliable and accurate enough to be used in real time for speed
optimization, Gangeskar (2018,2019), [Link]
[Link].

Traditionally, information from sensors and automation systems onboard ships have been available
mainly for local use by various onboard systems and users. Remote connections have been limited in
bandwidth and functionality, complex to install, manage and use and connected to highly proprietary
platforms with limited usability for end customers. This no longer needs to be the case. An abundance
of cloud platforms, modern communication technologies and Internet-of-Things solutions makes it
considerably easier to build end-to-end solutions that are cost-efficient and easy to use. A powerful
example of such a technology platform is Microsoft Azure, which offers a very wide set of services and
functions to enable seamless integration of sensors, data handling, processing, visualization and

115
distribution, both locally (i.e. on the Edge) and remotely (i.e. in the Cloud). Particularly, the strong
combination of Edge and Cloud computing, often referred to as hybrid computing, means that Microsoft
Azure is a very attractive platform to build applications related to the Internet-of-Things and
digitalization.

In the rest of this paper, we shall focus on describing the system based on imaging X-band radar that
can provide reliable STW measurements. Furthermore, the results from a verification study onboard a
vessel will be presented in detail. Finally, the integration aspects will be discussed with focus on how
modern IoT technologies can simplify the distribution of STW data from ship to shore.

2. Measurement principle for STW based on imaging X-band radar

Wavex provides current measurements with high accuracy, Gangeskar (2018a,b,c). Measuring the
STW has much in common with measuring currents, and the two measurements are generally based on
the same physical principles. The major difference is what the measured water speed is referred to: the
vessel when measuring the STW, and a fixed position when measuring currents.

The vessel velocity through water and current velocity are related through:

( )

where is the vessel s velocity over ground. Therefore, obtaining reliable current measurements
implies that also STW measurements will be reliable, as they are related to each other (at the same
depth) through the speed over ground (SOG), which can easily be extracted from GPS data.

Fig.1 shows the basic components in a Wavex system on a moving vessel. Specialized, DNV type
approved hardware is connected to the analog video signal output from a marine navigation X-band
radar. This hardware digitizes the analog radar video and outputs a radar image timeseries. Each radar
image includes a sector covering the STW measurement area.

Digitized images can also be acquired directly from radars with digital data output, commonly known
as IP (Internet Protocol) radars. This eliminates the need for additional digitalization hardware.

The Wavex system requires certain radar image meta-data from a GPS and a gyro compass.

To provide STW estimates, all required data are collected, synchronized and processed on the system
computer.

Optimum STW measurement performance requires radar images with sufficient spatial resolution. The

determined by the radar antenna beamwidth. For optimal accuracy, the radar should be operated in short
pulse mode. (If a solid-state X-band radar, utilizing pulse compression techniques, is used, the spatial
resolution in the STW measure area can be sufficient without compromising the radars navigation
performance.) In addition, a wind speed of at least 2 3 m/s is required. At this wind speed, the sea
surface gets sufficiently rough to create sufficient electromagnetic backscatter, Skolnik (1980). Gravity
waves modulate the ocean surface backscatter. A radar image with a clearly visible wave pattern is
shown in Fig.2.

Wavex bases its measurements on radar images covering local areas of interest, in a reasonable distance
from any disturbing structures, including the vessel hull. Fig.3 shows how the STW measure areas are
extracted from the radar images. The measure areas are called Cartesian image sections and are defined
during system commissioning through software configuration. Dedicated algorithms process these
images to provide the user with real-time STW data. The measure areas can be changed by software
reconfiguration at any time.

116
Fig.1: Schematic diagram of system based on imaging X-band radar

Fig.2: Imaging radar

117
 Fig.3: How Cartesian image sections for STW estimates are extracted from a radar image.

Fig.4: Cartesian image section time series are transformed into a wavenumber-frequency spectrum.

Fig.4 illustrates how 3-D fast Fourier transforms (FFTs) are applied to time series of Cartesian images,
giving 3-D spectra with information about the power present at various wavenumbers and frequencies,
Young et al. (1985). Various sorts of noise filtering are applied before STW is estimated from the
wavenumber-frequency spectra using an improved method developed by Miros. The method is, as
previously known methods, based on our already existing knowledge about the relation between
wavenumbers and frequencies of ocean gravity waves for zero current, i.e. the dispersion relation, Pond
et al. (1983):

2
0 g k tanh k d

where 0 is the wave frequency, k is the wavenumber vector, d is the water depth, and g is the gravity
of Earth. If there is a surface current U relative to the radar, a Doppler frequency shift is introduced in
the wave frequency:
0 k U

This Doppler shift causes the energy in the 3-D spectra frequency planes to be located on ellipses, rather
than circles. Based on the power distribution in the wavenumber-frequency spectra, the current vector
can be estimated.

118
Miros has recently developed further improvements to the method used for estimating ocean surface
currents from X-band radar images. This includes an improved method utilizing the full power
distribution properties, improved motion compensation, as well as several improvements increasing
performance under conditions with high current speeds and low signal-to-noise ratios. The method also
includes various functionalities to automatically detect and tag data with respect to quality.

3. Pilot verification of speed through water functionality at Arctic Lady

A number of Wavex pilot systems have been installed on various vessels using various sorts of X-band
radars. The system reliability and the accuracy of radar-based STW measurements have been examined
and verified by comparing with theoretical models and standard speed logs over large geographical
areas in a wide range of weather conditions and sea states.

3.1. Data acquisition

The following examples, based on data acquired from the LNG carrier Arctic Lady, were published in
Gangeskar (2019) in agreement with
of data were made available from their travels between Hammerfest in Norway and Marseilles in
France. In addition to Wavex measurements, simultaneous data were acquired from the on-board
acoustic speed log, from the Norshelf model by the Norwegian Meteorological Institute, and from the
Irish Marine Institute Northeast Atlantic Model. Fig.5 shows the route during a period of simultaneous
data from all sources, from September 15 to October 31.

The acoustic speed log on Arctic Lady is a JLN-550 Doppler Sonar (SDME) provided by JRC,
[Link] which is an advanced and widely
used instrument for measuring the STW from vessels. It is a two-axis, four-beam pulse Doppler Sonar
with optional rate of turn gyro, operating at 2 MHz (for water tracking), measuring a few meters below
the hull bottom. Information about the STW longitudinal component is obtained via the VBW (dual
ground/water speed) NMEA string. This is the STW component parallel to the vessel heading, with
positive values when the vessel moves forward relative to the water.

The Norshelf model, Röhrs et al. (2018), provides ocean current data for the Norwegian Shelf Sea.
The model has been set up at the Norwegian Meteorological Institute (MET) and includes the Skagerak
in the southeast, the northern parts of the North Sea, the shelf sea off western Norway including the
shelf slope, and parts of the Barents Sea in the north, that is, a considerable part of section 1 of the
route, Fig.5. The model provides data at a horizontal resolution of 2.4 km at a temporal resolution of 1
hour. Data are available from MET Norway Thredds Service, [Link]
[Link], in NetCDF files. Model data representing 5 m depth were chosen because this is close
to the effective measurement depth of the radar-based system.

The dataset Irish Marine Institute Northeast Atlantic Model provides surface current vectors for the
Irish waters in the northeast Atlantic, Dabrowski et al. (2016), that is, a considerable part of section 2
of the route, Fig.5. The ROMS hydrodynamic model (Regional Ocean Modeling System) uses a mean
horizontal resolution of 1.9 km and provides data at a temporal resolution of 1 hour. Data for the last
week are available via Thredds and ERDDAP servers in various formats, [Link]
erddap/griddap/IMI_NEATL.html. Older data were ordered from and delivered directly by the data
steward at the Irish Marine Institute in Matlab format.

3.2. Statistics and time series

Current data from the models were extracted at times and positions of interest, indicated by the route
in Fig.5, using time and position data from the vessel and linear interpolation. This is partly similar to
what was done in Gangeskar (2018a)
As already stated above, the accuracy of STW measurements is closely linked to the accuracy of current
measurements, defined by (1). Hence, for convenience, as the models and the Wavex system already

119
provides current data, we chose to consider current values as the basis for statistical measures (Table
I). The STW longitudinal component output from the speed log was simply converted to the current
longitudinal component using (1).

Fig.5: Route from September 15, 2018 to October 31, 2018, indicated by red lines in Google Earth

120
In order to compensate for different averaging strategies, to make statistical comparison more balanced,
to smooth out any minor temporal offsets between various data sources, and to make measured data
more comparable to model data, an additional temporal averaging of measured data was performed
before calculating statistics. For this purpose, a 40- was applied to the time
series. Mean and root-mean-square (RMS) deviations between individual data sources were calculated.

Fig.6 provides an overview of longitudinal current components and STW during the entire route shown
in Fig.5. No additional averaging is applied to these data. Data are missing in three periods because
most of the measurement systems were turned off when the vessel was at rest in Hammerfest (Norway),
Marseilles (France), and Saint-Nazaire (France). Apart from these periods, the data capture is complete.
The rate of defined STW data from the Wavex system is 99.94 % during the periods with available
radar images. In the following, we will look further into the details for a couple of shorter periods.

Fig.6: Overview of longitudinal current components and STW during the entire route shown in Fig.5
No additional averaging.

Table I: Deviations between longitudinal current components from radar-based system, speed log, and
models, based on all available data

Fig.7 shows five days of current and wind data from a period covered by the Northeast Atlantic Model.
The wind speed varies from 0 to 15 m/s, and the surface current in the area is dominated by the tidal
contribution, making it easy to visually observe the agreement between model data and measurements.
Currents in this area are more homogeneous and stable, with less eddies and stronger tidal dominance,
than for instance in the region covered by the Norshelf model. This may make this model more accurate,
and it makes comparison easier, because different averaging strategies and possible remaining temporal
and spatial offsets will make less influence on the results. Table I (based on all available data) shows
that measurements agree slightly better with the Northeast Atlantic Model than with the Norshelf
model.

121
Fig.8 shows the longitudinal current component and the STW during the same period, without any
additional averaging. It is evident that the radar-based system produces considerably smoother data
than the speed log. The reason for the varying amount of noise observed in speed log data is not known.
It is also clear that the speed log measurements are systematically erroneous, with an offset of approxi-
mately 0.5 m/s. This can also be observed from the statistics in Table I (despite additional averaging
before calculating statistics), in which data from the radar-based measurements are considerably more
consistent with model data (comparing green and red columns). Current magnitudes in the range 0 0.5
m/s are expected in this region and period. In the context of fuel optimization, the observed offset in
speed log data could mean an additional fuel cost corresponding to tens of tons of fuel a day for one
ship.

Fig.7: Time series of current and wind data from Arctic Lady, during a period covered by the Northeast
Atlantic Model; radar-based compared to Northeast Atlantic Model.

Fig.8: Time series of longitudinal current components and speeds, during a period covered by the
Northeast Atlantic Model; radar-based compared to Northeast Atlantic Model, speed log, and
GPS (partly covered by radar-based). No additional averaging.

Fig.9 shows the longitudinal current component and the STW during a period covered by the Norshelf
model, without any additional averaging. The wind speed varies from 1 to 19 m/s. Less homogeneous
currents and more local eddies, combined with different averaging strategies, make comparison more
difficult, because both model data and measurements vary relatively quickly with position and time.
Still, clearly the radar-based system produces considerably smoother data than the speed log; the speed
log data are influenced by an offset-like error, though the covariance between the two sensors looks
relatively consistent. Current magnitudes in the range 0 0.5 m/s are expected in this region and period.

122
Some possible explanations for observed deviations between various data sources are:
differences in spatial and temporal averaging strategies;
differences in effective measurement depth;
minor temporal offsets between various data sources;
inaccurate environmental data input to models;
finite resolution and accuracy in models;
measurement errors in sensors.

Fig.9: Time series of longitudinal current components and speeds, during a period covered by the
Norshelf model; radar-based compared to Norshelf model, speed log, and GPS. No additional
averaging.

4. Making STW data available onboard and onshore

Easy access to vast amounts of data from a multitude of sources is currently driving a wave of
innovation that impacts how vessels are designed, built, operated and maintained. Processes that used
to be largely based on manual observations and retrospective analysis based on incomplete data sets
can now be improved and automated with the access to detailed, reliable and accurate data. Modern
technologies make sure that the information can be made available both on the vessel and onshore and
thus enables a wide range of improvements.

In order to support the various application use cases mentioned above, Gangeskar (2019), there is a
need to have a flexible solution that allows easy access to the STW data both onboard the vessel and
onshore. The onboard requirement is particularly related to real-time usage for speed optimization. The
onshore requirement is related to hull and propeller performance estimations and the optimization of
hull cleaning activities. In addition, there are use cases for fleet management, including comparing and
optimizing vessel fleets, as well as reporting.

4.1. Access to STW data onboard

The STW data from a system such as the Miros system discussed in this paper can easily be made
available onboard the vessel either on dedicated displays, on web displays or via integration into a 3rd
party system onboard, Miros AS (2017). Traditionally, such integration has been based on simple
transmission of NMEA (National Marine Electronics Association) formatted data on serial links or
embedded in TCP or UDP transmissions on an Ethernet connection. A UDP transmission can be seen
as a simple push type of communication whereas a TCP connection can be seen as a pull connection as
it has to be initiated by the receiver. Sending NMEA data over a serial or Ethernet connection is still a
very common way to integrate sensors and systems.

123
Modern technologies facilitate integration between sensors and systems. One common technology
found in this domain is the MQTT (Message Queuing Telemetry Transport) protocol. MQTT is a
publish-subscribe type of protocol where a sensor (MQTT client) can send data to a server (often called
an MQTT broker). The broker is then responsible for distributing the information. Any MQTT client
can both send and receive data from the broker. MQTT is one of several commonly used IoT protocols.
One of the major advantages of MQTT over legacy solutions is that the sensor does not have to know
who the receivers of the data are, it only needs to relate to the broker. This increases reliability as the
remaining system continues to work when a client (receiver) goes down or has intermittent connectivity.
MQTT communication can be set up to buffer data in case of connectivity issues or other periods with
downtime on the receiver side. MQTT is a bandwidth-efficient protocol. The core MQTT protocol is
using TCP ports 1883 and 8883 which typically might be outbound blocked by firewalls. A good
solution is therefore to use MQTT over Websockets which is using TCP port 443. This port is
commonly open outbound or can easily be opened as it is typically used by many secure services based
on TLS (Transport Layer Security) communication, e.g. secure https websites, online banking etc.

Traditionally, it has been challenging to get access to real-time data from distributed assets. This has
been due to many factors including limited connectivity, lack of suitable protocols, lack of suitable
interfaces to send and receive data from and lack of platforms that can handle data efficiently and
seamlessly. Particularly, for seagoing vessels the lack of connectivity with sufficient and reliable
bandwidth has been a serious hinderance. This has changed in recent years due to a number of factors,
including:

satellite connections with reliable and cost-efficient connectivity;

efficient and modern communication protocols suitable for real-time transmission of telemetry
data across the internet;
data platforms that can handle large amounts of incoming data in a cost-efficient manner;
scalable and flexible processing platforms that can process incoming telemetry data;
security solutions utilizing mechanisms such as Public Key Infrastructure (PKI) encryption
(e.g. https);
authentication and authorization mechanisms based on Active Directory.

4.2. Access to STW data onshore

The STW data from a system, such as the Miros system discussed in this paper, can easily be made
available onshore via web displays or via integration into 3rd party system using push or pull
functionality. The described solution is based on using Microsoft Azure to collect, store, visualize and
distribute the data from the vessels in a secure manner.

The communication of STW data from the vessel to Microsoft Azure is made via secure communication
using modern protocols such as MQTT over Websockets, as described above. The STW system
onboard the vessel will initiate a secure connection to Microsoft Azure. Both sides (the vessel STW
system from Miros and the Miros environment in Microsoft Azure) is authenticated and authorized to
avoid any possibility for illegal access to data or tampering with data. Communication is established
outbound from the vessel, thus there is no need to open inbound ports in firewalls. Depending on the
vessel network configuration, there might be a need to open outbound firewall ports. If needed this can
be set up to only allow communication with Microsoft Azure to avoid any other services to utilize this
outbound open port.

On the receiving side the access to data is governed by Microsoft Azure security mechanisms based on
Active Directory. This means that only authorized personnel will be able to access the data via
download mechanisms or web displays. Furthermore, any automated data transfer will be secured in a
similar fashion.

124
Utilizing Microsoft Azure means that the STW data can be combined with other types of information,
such as data from other types of instruments and systems, sea state forecasts from weather providers
etc. It is also easy to get an overview of the status and history for a fleet of vessels.

Device management can also be supported by utilizing the functionality found in Microsoft Azure. In
this way it is possible to remotely configure the STW system and provide firmware update features.
This dramatically simplifies the commissioning and maintenance of the solution. With a transparent
solution it is straightforward to identify possible issues related to the configuration or the maintenance
of the physical equipment. This results in improvements in data quality and data availability. Software
updates and configuration changes can be implemented in several ways. One common use case is to
trigger a software update or configuration change from a remote location, i.e. by the equipment vendor
(Miros in this case). Microsoft Azure then makes sure that this change is applied in the most seamless
way. Software updates are then downloaded from the vessel via https communication which is again
authenticated, authorized and encrypted.

The various options for how to integrate the Miros STW solution is shown in Fig.10, starting with using
the STW solution as a standalone system onboard, via integrating the STW solution into an onboard
system to integrating the STW solution to an onshore party via Microsoft Azure. The onboard system
could be a vessel performance system or the vessel control system containing functionality such as an
autopilot or automated speed optimizer.

Fig.10: Various options for how to use the Miros STW solution. a) Miros STW system used as a
standalone solution onboard a vessel, b) Miros STW system integrated into vessel performance
system onboard, c) Miros STW system integrated with an onshore customer system via the
Miros Cloud, d) combining options b) and c).

5. Conclusion

Information about surface currents and STW is of great value for many purposes, for instance as input
to fuel optimization systems and hull performance estimation (detection of fouling). Thanks to
considerable work and progress within the field of radar remote sensing during the recent decades, such
ocean surface measurements can now be performed with a high reliability and accuracy using radar
sensors. Hence, challenges like data heavily influenced by noise and costs related to installing and
maintaining traditional underwater equipment can be avoided. By means of radar remote sensing
techniques, the user can measure the current and the STW in the water of interest, sufficiently far away
from structures and the chaotic conditions close to a vessel hull that would otherwise disturb the
measurements. Combining the sensing technology with technologies from the Internet-of-Things
domain means that the data can be made easily and securely available anywhere in real time.

125
Acknowledgements

Miros would like to thank Höegh LNG and the crew of Arctic Lady for making available and approving
data to be used in this work and paper. Miros would also like to thank the Irish Marine Institute and
their data stewards for well-organized model output provision, and the Norwegian Meteorological
Institute for making model data available from their servers.

References

ANTOLA, M.; SOLONEN, A.; PYÖRRE, J. (2017), Notorious speed through water, 2nd HullPIC, Ul-
richshusen, pp.156-165

BAUR, M. (2016), Acquisition and integration of meaningful performance data on board - challenges
and experiences, 1st HullPIC, Pavone, pp.230-238

BOS, M. (2016), How MetOcean data can improve accuracy and reliability of vessel performance
estimates, 1st HullPIC, Pavone, pp.106-114

BROWN, J.; COLLING, A.; PARK, D.; PHILIPS, J.; ROTHERY, D.; WRIGHT, J. (2001), Ocean
Circulation, The Open University, Milton Keynes (130 pages)

CARCHEN, A.; PAZOUKI, K.; ATLAR, M. (2017), Development of an online ship performance mon-
itoring system dedicated for biofouling and anti-fouling coating analysis, 2nd HullPIC, Ulrichshusen,
pp.90-101

DABROWSKI, T.; LYONS, K.; CUSACK, C.; CASAL, G.; BERRY, A.; NOLAN, G. D. (2016),
Ocean modelling for aquaculture and fisheries in Irish waters, Ocean Sci. 12, pp.101-116

FLAGG, C. N.; SCHWARTZE, G.; GOTTLIEB, E.; ROSSBY, T. (1998), Operating an acoustic Dop-
pler current profiler aboard a container vessel, J. Atmospheric and Oceanic Technology 15, pp.257-
271

FRITZ, F. (2016), Practical experiences with vessel performance management and hull condition mon-
itoring, 1st HullPIC, Pavone, pp.128-136

GANGESKAR, R. (2017), Automatically calibrated wave spectra by the Miros Wavex system Accu-
racy verified, Miros AS, Asker

GANGESKAR, R.; PRYTZ, G.; SVANES BERTELSEN, V. (2018), On-Board Real-Time Wave &
Current Measurements for Decision Support, 3rd HullPIC, Redworth, pp.223-233

GANGESKAR, R. (2018a), Verifying High-Accuracy Ocean Surface Current Measurements by X-

Band Radar for Fixed and Moving Installations, IEEE Trans. Geosci. Remote Sens. 56/8, pp.4845-
4855

GANGESKAR, R. (2018b), Ocean Surface Current Measurement by the Miros Wavex System, Miros
AS, Asker

GANGESKAR, R. (2018c), Surface Current Measurements from Moving Vessels by Wavex Verified,
Miros AS, Asker

GANGESKAR, R. (2019), Measuring the speed through water by Wavex®, Miros AS, Asker

GOLER, H.; BOSKURT, K. (2017), Speed loss analysis of high-speed ro-ro vessels coated with new
antifouling technologies, 2nd HullPIC, Ulrichshusen, pp.173-193

126
KELLING, J. (2017), Ultrasound-based antifouling solutions, 2nd HullPIC, Ulrichshusen, pp.43-49

KING, B. A.; COOPER, E. B. (1993),

antennas with heading from conventional gyrocompass measurements, Deep Sea Research Part I:
Oceanographic Research Papers 40/11-12, pp.2207-2216

NEW, A. L. (1992), Factors affecting the quality of shipboard acoustic Doppler current profiler data,
Deep Sea Research Part A: Oceanographic Research Papers 39/11-12, pp.1985-1996

POND, S.; PICKARD, G. L. (1983), Introductory Dynamical Oceanography, Butterworth-Heinemann,

pp.207-219

RÖHRS, J; SPERREVIK, A. K.; CHRISTENSEN, K. H. (2018), NorShelf: An ocean reanalysis and

data-assimilative forecast model for the Norwegian Shelf Sea, Norwegian Meteorological Institute

SKOLNIK, M. I. (1980), Introduction to radar systems, McGraw-Hill, pp.474-482

YOUNG, I. R.; ROSENTHAL, W.; ZIEMER, F. (1985), A three-dimensional analysis of marine radar
images for the determination of ocean wave directionality and surface currents, J. Geophys. Res. 90/
C1, pp.1049 1059

127
 Exploring the Effect of Vessel Performance Information Barriers
on Decision-Making Practice: A Time Charter Application
Jean-Marc Bonello, UCL Energy Institute, London/UK, [Link].15@[Link]
Tristan Smith, UCL Energy Institute, London/UK

Abstract

Several technological solutions are available to improve vessel efficiency. However, despite short
payback periods, uptake remains low due to several market barriers and failures. Information related
barriers in particular, introduce uncertainty during technology appraisal through poor performance
data quality and use. Furthermore, information asymmetry can exist between stakeholders regarding
vessel performance more generally, and result in a challenge identifying the efficiency and performance
of a ship relative to its peer group or competitors for a charter. This paper explores these problems in
practice and attempts to evaluate the effect of information related shortcomings related to vessel
efficiency and decision making in deep sea cargo shipping. The work is aimed towards creating an
evidence base for targeted commercial tools and policy designed to promote and reward vessel
efficiency by bridging the information gap between technical and commercial sides of shipping.

1. Introduction

After many years of debate, the International Maritime Organisation (IMO) set the course for the
reduction of greenhouse gas (GHG) emissions from shipping by at least 50% of 2008 levels by 2050 at
MEPC 72 (IMO MEPC, 2018). Efficiency is one of the routes by which ships can reduce their emissions
which can be achieved through operational or technological interventions, Bouman et al. (2017).
Despite efforts from the IMO Maritime Environment Protection Committee (MEPC) to increase energy
efficiency (EE), the uptake of these technologies is still found to be low even for measures with short
payback periods, Wang et al. (2010), Faber et al. (2011), Poulsen and Sornn-Friese (2015), Rehmatulla
and Smith (2015a). (The implementation of the mandatory Energy Efficiency Design Index (EEDI),
Ship Energy Efficiency Management Plan (SEEMP) and voluntary Energy Efficiency Operational
Index (EEOI) put in place by the MEPC have not been seen to improve efficiency and some studies
finding vessels being designed less efficiently since the implementation of these measures, Stevens et
). This EE gap in shipping between the hypothetically achievable
and the actual attained vessel performance has been most commonly interpreted through classical
economic barrier theory, Rehmatulla and Smith (2015b), Johnson and Andersson (2016).

Information barriers are consistently found to be one of the more significant market failures hindering
decision-making for EE technology (EET) uptake. The lack of access to the right information, low
quality data and, paradoxically, information overload makes technology appraisal challenging, Eppler
and Mengis (2004), Jafarzadeh and Utne (2014), Rehmatulla and Smith (2015b). Another layer of
complexity is added due to stakeholder relationships and influencers in shipping involved at various
phases of decision-making. Information asymmetry causing lack of transparency and split incentives,
intrinsic to shipping chartering practices, increases uncertainty thus discouraging EET uptake,
Agnolucci et al. (2014), Poulsen and Sornn-Friese (2015), Poulsen and Johnson (2016). This is
particularly the case in the time charter (TC) market where the charterer incurs bunker costs that are
determined by the condition and performance of the vessel which they have limited information about.
A consequence of this is that more efficient ships are not seen to be rewarded on the market for their
better performance, Tamvakis and Thanopoulou (2000), Prakash et al. (2016), Adland et al. (2017),
thus creating a two-tier market which plays in favour of worse performing vessels creating the
theoretical market for lemons, Akerlof (1970).

This paper sets out to:

1) Identify a number of sources of information barriers and asymmetries between owners and
charterers with regards to vessel performance

128
 2) Propose a game theoretic framing of the charterer-owner dynamic at the pre-fixture stage to
capture information asymmetry for a TC scenario
3) Set up a quantitative model to represent the charterer-owner dynamic
Through the evidence produced by achieving the above, this paper aims to determine the effect of
uncertainty due to identified information asymmetry and barriers in economic terms for both parties.

A discussion on information barriers related to vessel performance is followed by brief a review of

current chartering practices. This leads to the proposal of the game-theoretic framework that is used to
build a probabilistic model to quantify the uncertainty different stakeholders are exposed to and
associated financial risks. Readers should note that term energy efficiency and vessel performance
are used interchangeably in this paper. This is since a performance model looks at energy use therefore
a vessel with better performance is analogous with a vessel being more efficient.

2. Information barriers in vessel performance

Performance modelling and monitoring is fundamental to any ship owner and has gained much higher
importance in the last decade with significant human and financial resources dedicated to it. This is
mostly driven by shrinking margins (due to increased fuel costs and oversupply of tonnage in some
markets) and more recently, due to the advent of high frequency data acquisition systems and access to
relatively cheap data processing power. Technical departments and companies are increasingly seen to
be developing complex proprietary models using statistical analysis, multivariate regression and
artificial neural network based models to model and predict vessel performance. Haranen et al. (2016)
give a good summary of the different approaches available with some recent applied examples being
(2018); Yoo and Kim (2019).

With the increasing complexity of these methods and a lack of standardization, comes the penalty of
reduced transparency where black box models for specific vessels are developed which cannot be used
to communicate vessel performance externally. Although ISO 19030, ISO (2016), provides some basis
for standardisation, it is increasingly used market ISO-compliant performance monitoring solutions
which go beyond the standard including proprietary additions. This creates opacity that occurs
especially when third-parties are contracted to undertake performance monitoring where the ship owner
may not understand fully the way performance is being measured and modelled due to lack of expertise
and resource.

The solution that shipping has embraced for years is the use of speed-consumption benchmarks for
vessels. These simple relations are fundamental to daily commercial operations and are the widely
accepted method of communicating vessel performance outside of the technical
department. Once again there is no standardised procedure for creating these curves which are compiled
at the discretion of the owner most commonly based on noon reports. DNV-GL (2015) found that half
the respondents (N=80) of a survey only keep manual daily reading of data, ranging from fuel
consumption to metocean conditions. The use of low frequency data has been shown to lead to high
uncertainty when assessing performance, Aldous et al. (2015), Lund and Gonzalez (2017), due to
associated measurement uncertainty and averaging errors.

The four basic measurements that are required for compiling this curve are speed through water, fuel
consumption, draught and weather condition. Speed logs have a relatively good precision of around 1%
however they do suffer from bias and drift, Aldous et al. (2015). In the absence of flow meters, fuel
consumption is measured by tank soundings which carry a significant uncertainty due to various factors
including inaccuracies in tank geometries, human error, and trim correction factors leading to an
precision of around 5% on average, Hunsucker et al. (2018). With manual fuel consumption
measurement, human error is difficult to avoided altogether and also some incentive to misrepresent
daily readings in order to cover for operational decisions taken by the master or chief engineer which
they may not want to have on record. While draught sensors are notoriously unreliable due to various
factors, this problem is circumvented by the parametrisation of curves for ballast and laden conditions.
Similarly for weather, curves are compiled using data up to a particular threshold (nominally BF4)

129
above which is considered to be heavy weather operation. While this has been shown to be effective to
allow comparison of performance below the weather threshold, Hudson and Daniels Galle (2017), it
leaves the charterer with no indication of vessel performance in rough weather which can make up for
a significant amount of the operating profile on certain routes.

Once some basic filtering is undertaken to remove clear outliers stemming from instrumentation or
human error, heavy weather days are removed and the data set is split to represent the two loading
conditions, the amount of data loss usually results in a very small number of data points. A quadratic
fit, Bialystocki and Konovessis (2016), is usually found to be the best fit for this relation, Fig.1, for
which some basic fitting parameters can be determined. In the example below, an uncertainty of ±2.5
TPD (at a 95% confidence interval) is associated with any consumption estimated using the curve. Thus,
when such a curve is used to determine the warranted performance on a charter party at a given speed,
the uncertainty related to the modelling technique used (in this case the confidence interval for the
quadratic curve) is omitted.

Fig.1: Typical speed-consumption curve [laden, <=BF4]

3. Information asymmetry: stakeholder dynamics in time chartering

Most vessels engaged in deep-sea shipping are owned and operated by different parties who come to a
contractual agreement defining the terms and conditions under which a charter is undertaken. In TC,
the vessel is hired out for a fixed period of time on a $/day charter rate during which the owner is
responsible of all operating costs while the charterer incurs voyage costs, Stopford (2009). This sets up
a split incentive barrier where, any efficiency investment undertaken by the owner with be beneficial to
the charterer. The chartering routine can be described in five discrete steps; pre-fixture, fixture,
execution of charter, post-fixture and claims handling, Plomaritou and Nikolaides (2016). At the pre-
fixture stage orders and positions are matched, usually via brokers, which then leads to a series of
negotiations via offers and counteroffers between the owner and charterer.

Once an agreement is reached, all the details of the charter are made official through the signing of a
charter party contract defining the fixture, Gorton et al. (2009). The role of the charter party is
fundamental as it sets up an agreement on terms with an understanding that each party commits to
financial obligations under a legally binding contract, Assimenos (2017). These contracts are based on
well-established standard forms issued by institutions such as BIMCO and NYPE or, in the case of
larger companies chartering in, compiled inhouse making them more specialized to the requirements of
the charterer (SHELLTIME and EXXONTIME being some common examples of the latter). Amongst
the clauses agreed upon in a TC party is the warranted performance that is defined at a loading condition
and a within a weather threshold. A typical speed and consumption clause would specify the warranted
vessel performance as follows:

ASBA (1981)

130
 Pre-fixture Charter execution

Operational
Position orders
Broker Owner

Charter
party
Master

Broker Charterer
Order Instructions

Fig.2: Typical TC stakeholder interaction

Given that the charterer is responsible for voyage cost, the largest portion of which is cost of bunkers
posing the biggest commercial risk during the execution of the fixture, it is in their interest to have the
best information possible related to vessel performance. The above clause leaves the charterer with
limited possibility to evaluate their exposure to risk due to operation in rough weather, sailing under
any draught other than fully laden or at any speed other than the warranted velocity. The latter is
particularly relevant considering the increasing popularity of slow steaming being instructed by the
charterer for commercial reasons. Moreover, under English Law, this clause is not understood as a
guarantee over the whole charter period but implies that the vessel is capable of the quoted performance
at the time of negotiation and fixture, Gorton et al. (2009). Given the speed under which the above
negotiations are conducted and the fact that they are treated as standard practices, there is little
scrutineering of this clause which raises difficulties in the post-fixture phase when performance claims
can be filed by the charterer. This is not normally a part of the negotiation process which is mainly
focused on the charter rate and other logistical such as delivery time and location, Veenstra and van
Dalen (2011).

Additionally, there is no way for the charterer to ensure that the vessel performance that is being quoted
is representative of the performance they are to expect in operation. Veenstra and van Dalen (2011)
were able to show that owners strategically change the warranted performance from one charter to
another and were being fixed at a performance worse (higher consumption at a lower speed) than their
design performance. Anecdotally, in discussions with industry players, a tendency for owners to warrant
lower performance than observed in practice was encountered in several cases. (One exception that has
been encountered by the author is when owners are entering their vessel into a charter pool. Owners
tend to quote their vessels performing better as some charter pools allocate higher proportions of profits
for higher efficiency.) This information asymmetry regarding vessel performance leads to a significant
level of uncertainty
commercial risk exposure due to bunker cost. During the charter commences, noon reports are
communicated to the charterer who can then start to assess performance only after having committed
to the fixture. The degree of transparency in these reports can also be dubious as there are clear
incentives for misreporting consumption, Poulsen and Johnson (2016).

Another legacy term

number is specified in a charter party. Gorton et al. (2009) explains that this is generally accepted to
which in practice is seen
to be interpreted as a 5% margin on the warranted consumption and a 5% or 0.5kn margin on warranted
speed, Coghlin et al. (2014). This is invariably a point of contention when speed or consumption claims
are taken up to arbitration which gives the owner a margin for underperformance.

Having had a closer look at the specific information related barriers and asymmetries in the previous
two sections, the elements discussed are summarised in Table I along with the associated artefacts to
be included in the framework designed to assess the uncertainty propagation due to each. The selection
of these elements strives to capture both the epistemic uncertainty associated with the information

131
barriers but also the aleatoric uncertainty stemming from the information asymmetry. The bottom line
is that when estimating the possible bunker cost over a fixture, charterers are not provided with
information about the possible range of the cost (through information barriers) as well as the accuracy
of the value they are given (through information asymmetry).

Table I: Information barrier categories and artefacts

Barrier category Barrier artefact
Data quality and acquisition Data collection frequency
method Instrumentation and measurement error
Performance modelling Performance modelling (speed-consumption curve based)
Communication of vessel performance between owner and
Information asymmetry
charterer

4. Proposed game theoretic model

In order to capture some of the TC dynamics described above including both information barriers and
asymmetry, a simple game theoretic model is proposed that has been previously used to frame similar
problems. Game theory (GT) is originally a mathematical framework that has been embraced by
economics and also environmental disciplines to describe strategic behavior between different
stakeholder and assess the outcome for each based on the effect of their decisions on each other, von
Neumann and Morgenstern (1944). Information asymmetry is tackled in a particular branch of GT
which uses principal agent theory to describe problems of adverse selection (where one party is better
informed than another before entering a contract) and moral hazard (where one party is better informed
about the actions that will take place after signing a contract). More information can be found in
Rasmusen (2007). Several studies related to energy efficiency have identified the owner-charterer
dynamic as a principal-agent problem, IEA (2007), Kontovas and Psaraftis (2013), Rehmatulla (2014)
expressing it through GT, Bergantino and Veenstra (2002), Suh and Park (2010), Psarros (2016).

Crewing &
Charterer operation
Owner

(Principal) (Agent)
Daily charter rate

Bunker cost

Fig.3: Principal-agent relationship in TC (adapted from Rehmatulla and Smith (2015a)

Thus, drawing on the examples above, an adverse selection game is proposed as an appropriate model
of charterer/owner dynamics and set up with the owner acting as the agent that has better information
regarding the vessel performance then the charterer who is the principal Fig.4. The game is envisaged
to represent the fixture stage when the owner and charter agree the terms in the charter party including
the warranted performance specified in the performance clause. The charterer will circulate an offer
which an owner will fulfill with their vessel which they can accept or reject. It is assumed that no
negotiation takes place because to the best of the authors knowledge, the performance clause is normally
not negotiated upon. (Large companies that charter in a high volume of vessels may build up an internal
database based on historic performance which will allow them to compare warranted performance
across owners, sister ships and peers which is not possible for smaller charterers.)

High level
Accept

Contract
Charterer Owner
Nature

PRINCIPAL AGENT Reject

Low level

Fig.4: TC fixture represented as an adverse selection game

132
In this case, nature sets the actual performance of the vessel which the charter has no way of knowing
except through the information provided by the owner in the performance clause. It is assumed that the
owner accepts the contract if the conditions of the contract reach some level of expected utility for both
parties. The charterer assesses the contract with some level of trust based on the expectation of
transparency with regards to vessel performance. Thus, assuming the extremes of these player types,
the owner can be modelled as acting with high or low transparency while the charterer can have high
or low trust. This allows to set up different scenarios based on the four possible behavior combinations
to assess the effect on the utility for both players (in this case, operating profit) under the four different
scenarios proposed can be evaluated to analyse the outcome of each case. In the next section, this
framework is used to constrain a quantitative model that sets out to illustrate the risk due to uncertainty
that parties are exposed to when taking chartering decisions under uncertainty regarding vessel
performance. The assumptions regarding vessel performance information use and communication for
these four base cases are described in Table II. From the observations in current chartering practice
presented in Sections 2 and 3, the business-as-usual (BAU) is understood to be Case 2.2 which implies
low transparency from the owner and low trust by the charterer. The three other counterfactuals are set
up to represent the other possible combinations and associated assumptions.

Table II: Description of modelling scenario assumptions

Case 2.2
Case 1.1 Case 1.2 Case 2.1
(BAU)
Trust High Low High Low
Transparency High High Low Low
Performance Speed-consumption curves based on noon report data up to BF4
modelling weather threshold
Measurement
Considered Not considered
uncertainty
Warranted
Based on historic experience Strategically compiled
performance
Provision of speed-con curve
Performance Provision of warranted speed-
including associated confidence
communication consumption at laden condition
interval
Deterministic: Based on
Probabilistic: Based on speed- warranted performance
Charterer bunker
consumption curve and expected communication and expected
estimate method
probability distribution of speeds operational profile during
fixture
Charterer bunker
None Included None Included
estimate margin

Charterer Owner

5. Quantitative modelling

A utility function to evaluate each players payoff under the different scenarios must be devised. A
common approach looks at cashflows for owners and charterers to define operating profits over a fixture
period which can be defined as (Factors such as historic relationships, location of vessels and market
conditions are not considered. A transactional relationship is assumed taking the charter as a one-off
event not being affected by any past charters and not influencing future ones. Brokerage fees on both
sides are also not included.):

133
Where for a charter period T days long, is owner operating profit, Rt is daily charter rate, is the
average daily operating costs, is charterer operating profile, is freight revenue from transport
work, is voyage costs. It should be noted that the cost of bunker is normally in integral part of
voyage cost however in this case it is being represented separately for clarity as this work focuses on
energy efficiency. The notation E() is used to denote an expected term that can only be estimated when
the vessel fixture is agreed. It is assumed that both players are rational and will strategically act to
maximise their respective operating profit.

Operating profit for the owner can easily be evaluated as a good knowledge of operating costs is
assumed and the revenue form the charter is fixed. The charterer payoff however carries uncertainty in
bunker cost and freight revenue which stems partially from market conditions (future utilisation, freight
rates and bunker cost) and factors affecting bunker consumption (vessel performance, hull condition,
weather). The owner is exposed to some risk stemming from performance claims that the charterer may
raise based on the warranted consumption agreed upon in the charter party. However, at the pre-
chartering stage neither the owner nor the charterer can include these payouts in their assessment of
costs throughout the fixture. Only at the end of the charter period T the realised operating profit can be
determined as:

Where is the total value of any performance claims over the charter period T. The total value and
probability of occurrence of performance claims is directly related to how conservative the owner is in
warranting the vessel. Following the equations above, the financial risk that the charter is exposed upon
fixing a vessel can be described as the difference between the expected and the realised operating profit.
Linking back to the scenarios described in Table II:, the four cases infer different information available
to evaluating the term under the different actor types, levels of transparency and trust which in
turn leads to a different evaluation of .

An applied case study is presented for a 50,000DWT chemical/product tanker for which a three-year
long continuous monitoring (CM) data set (10-minute frequency) was made available. This vessel was
used to set up a counterfactual game to illustrate the effect of the information barrier artefacts presented
in Table I.

5.1. Data collection and performance modelling

The data was resampled over 24 hour periods to simulate noon-reporting and, given draught was not
part of the data set provided, Automatic Identification System (AIS) data was used to determine if the
vessel was in ballast or laden condition. After outlier removal, speed trough water (STW) and fuel
consumption (TPD) values were used to compile speed-consumption curves for both loading conditions
for fair weather conditions up to BF4. In order to simulate the measurement errors associated with noon
reporting, a random error (following a Beta distribution) of 5% and 1% was applied to TPD and STW
respectively as specified by Aldous et al. (2015). This yielded a quadratic relationship with an
associated confidence interval that could be applied to any value estimated using the performance
model.

134
5.2. Communication of performance

Based on the historic vessel performance, the warranted TPD and speed value set by the owner is to be
determined. Upon observation of the speed distribution when operating in the laden condition Fig.5:,
the mean velocity at around 12kn which implies that the master strives to keep to this velocity for most
of the time. Assuming a 12.5kn warranted speed, the performance model suggests a consumption of
17.8 ± 2.5 TPD (at a 95% confident interval). Assuming a conservative owner the performance is
strategically quoted to be 18.5 TPD at 12kn laden for fair weather up to BF4 following findings by
Veenstra and van Dalen (2011).

With regards to the weather condition, Fig.5: illustrates that the vessel spends around 45% of time
sailing above BF4 which falls outside of the warranty. The histogram also suggests the speed of the
vessel is not affected by weather condition indicating that higher fuel consumption is expected due to
the rougher weather.

Fig.5: Speed distribution for laden condition

5.3. Expected bunker consumption estimate

Once the charter is presented with the warranted performance, they can use this to estimate the expected
profit from the charter having some experience of operating similar vessels and knowledge of the
expected operating profile over the fixture period. In order to populate Equation 5.2, some assumptions
must be made which are presented in Table III.

The upper bound bunker cost that the charterer expects based on the warranted performance assumes
maximum operation throughout the fixture period with some margin for rough weather operation. This
method would represent an extremely risk-averse charterer or one with no previous experience.

Table III: Baseline assumptions for fuel consumption estimate

Parameter Symbol Value Description Source
Charter rate for peer group for a one-
Charter rate Rt 13,500 (Clarksons, 2018)
year fixture in $/day
Freight rate 50 Freight rate for Easychem cargo in $/t (Clarksons, 2018)
Typical parcel size of cargo from
Parcel size 38,000 (Clarksons, 2018)
previous voyages in t
Average number of voyages undertaken
Voyages (pa) 6 Historic data
per annum since 2014
Daily operating cost based on OPEX
OPEX 6300 (Clarksons, 2018)
index
Load 0.6 Ratio of laden and ballast time Historic data

135
 utilisation
Time Utilisation of vessel, sailing days per
0.85 Historic data
utilisation annum
Bunker cost for IFO 380 at Port of
Bunker price 450 (Clarksons, 2018)
Rotterdam (Q1 2018) in $/t
Weather Additional margin to account for rough
0.05 (Gorton et al., 2009)
margin weather
Additional bunker margin to represent a
Low trust
0.05 conservative approach under low trust Author assumption
margin
scenario
Ratio of mean bunker consumption for
TPD ratio 0.9 ballast condition compared to laden Historic data
conditions
As derived in previous section in t
Warranted
18.5 [Based on speed of 12.5kn with an Historic data
TPD
additional 0.5t]
Warranted
12 As derived in previous section in kn Historic data
Speed

The above assumes that consumption is not sensitive to loading condition and all sailing is conducted
at the warranted speed. Assuming a charterer with some previous experience, a slightly more
sophisticated bunker estimate may be represented as follows:

Equation 5.6 accounts for expected utilisation, time spent in ballast and laden condition and decreased
consumption for the laden condition and is used as method for charterer consumption estimation in the
low transparency scenarios.

In the high transparency case, it is assumed that the speed-consumption curves for both ballast and laden
conditions are shared with the charterer including the associated model uncertainty (standard deviation
for a 95% confidence interval). This allows the charterer to use a probability distribution of expected
speeds rather than relying on the warranted speed which allows for a probabilistic estimate to be created
rather than a deterministic one. The speed consumption curves used include all weather conditions in
order to account for increased consumption due to rough weather operation and are used for the high
transparency scenarios.

5.4. Expected operating profit estimate

The bunker consumption estimations as described are used in Equation 5.2 which will be evaluated for
a one-year fixture. The expected freight revenue can be defined as:

For the case presented in this paper, the values provide in Table III: are used to determine freight
revenue and charter cost. Voyage costs are approximated using costs adapted from Stopford (2009)
to be equal to half the bunker cost .

6. Results

The model described is run for a 12 month long fixture and Fig.6: illustrates the preliminary results for
the baseline parameters presented in Table III: showing expected bunker consumption and charter
operating profit. The attained performance estimate for comparison is obtained from historical data
for an equivalent operating profile and is presented in Table IV: along with the modelling results.

136
 6.1. Charterer perspective

Bunker consumption estimates move further away from the baseline historic performance as
transparency and trust decrease as expected which is also reflected in expected operating profits. With
single point performance communication in the low transparency cases, only a discrete value can be
estimated compared to the probabilistic range from the higher transparency case. This may be a driver
for charterers to take a more conservative approach to estimating expected operating profits and take
decisions based on the worst perceived outcome due to lack of information. In the high-performance
case, even through only noon reports and simplistic performance modelling was used, an accurate
bunker estimate can be observed with the mean being within 3% of the historic baseline.

Fig.6: Bunker and charterer operating profit estimates under different transparency regimes (Results of
Case 1.2 are not show for clarity. The distributions are similar to those for Case 1.1 with means
and standard deviations as specified in Table 6 1.)

Table IV: Bunker and charterer operating profit estimates under different transparency regimes
% of % of
Case Description
Historical Historical
Attained performance
Historical 2.63*
estimate 1.74* 1 2.39* 1
High transparency, high 1.68 2.44
1.1 2.63
trust (0.048)** 0.97 (0.032)** 1.02
High transparency, low 1.49 2.56
1.2 2.63
trust (0.048)** 0.86 (0.032)** 1.07
Low transparency, high
2.1 2.63
trust 1.43 0.82 2.60 1.09
Low transparency, low
2.2 2.63
trust 1.24 0.71 2.73 1.14
*Baseline attained values as defined in Equation 5.4 rather than expected. **Mean (standard deviation)

Fig.7 illustrates the effect on bunker cost estimate based on variations in warranted speed and strategic
warranted TPD changes under the low transparency regime. At no change, the warranted TPD is equal
to the actual TPD obtained from the speed-consumption model derived from noon reports. The accom-
pa¬ny¬ing parallel lines describe the variation expected by strategic changes in TPD in steps of 0.5 t.
Given the quadratic nature of the relationship, the bunker estimate increases at a higher rate as warranted

which allows for a 1kn window around the warranted speed creating a significant amount of uncertainty
when attempting to estimate bunker cost over a fixture.

Recalling the assumption that the charterer is a rational actor looking to maximise operating profits, the
results show that this is the least optimal option compared to the baseline for various reasons. Firstly, it
effectively ties up a significant amount of cash that is expected to be spent on bunker which will not
which becomes a significant issue in tight markets when positive cashflows become paramount in day-

137
to-day operations. Alternatively, that money could be put towards better use through investments or
towards chartering another vessel with a higher charter rate but better performance.

Fig.7: Effect of owner strategic warranty declaration on bunker consumption estimate

Secondly, the results in Fig.6: show that the owner has a large margin within which they are still within
the warranted performance which can be used strategically in various ways due to the lack of
information on the charterer side. This may lead to reduce the efforts of the master and crew to operate
the vessel efficiently as well as reducing the urgency for hull maintenance or other interventions that
will have a positive effect on efficiency. Alternatively, an opportunity is provided for misreporting of
fuel consumption by the master to cover deviations from the agreed instructions or any skimming of
bunker by the crew to be sold on the black market for which Poulsen and Johnson (2016) found
evidence in their interview-based study. (This situation gives rise to another two instances of the
principal-agent problem: i) a second level problem between the owner and the charterer with respect to
the provision of fuel consumption information during the charter and ii) one between owner and the
master on board with regards fuel consumption reporting. These provide fertile ground for further
research.)

6.2. Owner perspective

On the owner side, the expected operating profit is only affected by the reported future operation of the
vessel and any related performance claims. The number of days that are liable for claims is dependent
on the warranted performance as selected by the owner and on the reported performance once the vessel
is in operation which can be used strategically by the owner to their advantage. Fig.8 illustrates the
extent to which strategic warranted performance selection can virtually absolve the owner of risks
associated with claims. For the dataset used in this study, only 45% of the days used were within the
performance clause with the remaining days being at a speed higher than warranted, over BF4 or in
clear breach of warranty. Only 1% of these days were in clear breach of warranty which is negligible
in terms of claim value explaining why the in Table IV: is constant. As warranted performance
is increased (vessel made more efficient), the liability for performance claims increases in likelihood
and value. Veenstra and van Dalen (2011) suggest that the strategic changes in warranted performance
are not bargained upon at the negotiation phase but are used to build a safety margin to protect against
performance claims. Fig.8
performance clause as the number of days liable for performance claims decreases significantly as
warranted TPD increases (or warranted speed decreases).

On the owner side, higher transparency does not have a direct benefit to operating profit as modelled
above however it can be used as a strategy to increase the likelihood of winning more charterers at a
possibly higher rates as charters are empowered to make more informed decisions when leading to
better operating profit estimates. This may become especially relevant in periods of higher bunker cost
or weak markets where bunker consumption contributes to additional costs.

138
 Fig.8: Percentage of days sailing liable to performance claims and prospective claim value

7. Conclusion

This work proposes a novel way of bringing together the technical and commercial regimes in vessel
chartering which are usually treated in isolation both in academic literature and in industry. The use of
GT has been shown to have potential as a good framework to build an analytical model that
transparently considers the perspective of both charterer and owner and their expected economic
outcomes based on their respective strategies. By tackling information barriers and asymmetries
specifically, this techno-economic model can empower charterers and owners to quantify the effect of
these market failures when making chartering decisions and policy makers to test the potential outcome
of measures by also considering the behaviour of industry players. Due to the flexibility offered by the
quantitative modelling, it can be developed to test the outcome for various market scenarios,
implementation of standardisation or policy, different performance modelling and data collection
approaches and performance-based contracting.

The preliminary results presented lead to the conclusion that, in the current scenario, information
barriers and lack of trust regarding vessel performance leave charterers with a large uncertainty when
determining the commercial viability of a fixture. The strategy chosen to overcome this is being
conservative which in turn provides owners with little or no incentive to operate vessels more efficiency
or, in the long term, invest in more efficient tonnage especially as these have not been seen to attract
higher charter rates. Under a higher transparency regime and increased trust, it is proposed that both
parties may stand to make smarter business decisions in the short term whilst also promoting a long-
term shift towards more efficient shipping through both technological and operational means.

The next step in this research involves running systematic sensitivity analysis to assess the impact
changes in the input parameters and player dynamics have on the outcome for charterers and owners
respectively. This will be followed by the addition of another layer of complexity by considering the
addition of energy efficiency technologies, energy savings clauses and performance contracting
assessing the effect of sharing the associated risks and rewards under the uncertainty.

References

ADLAND, R. et al. (2017), Does fuel efficiency pay? Empirical evidence from the drybulk timecharter
market revisited, Transportation Research Part A: Policy and Practice 95, pp.1 12.

AGNOLUCCI, P.; SMITH, T.; REHMATULLA, N. (2014), Energy efficiency and time charter rates:
Energy efficiency savings recovered by ship owners in the Panamax market, Transportation Research
Part A: Policy and Practice 66(1), pp.173 184.

139
AKERLOF, G.A. (1970), The Market for " Lemons ": Quality Uncertainty and the Market Mechanism,
The Quarterly Journal of Economics, 84(3), pp.488 500

ALDOUS, L. et al. (2015), Uncertainty analysis in ship performance monitoring, Ocean Engineering
110, pp.29 38

ASBA (1981) ASBATIME Type 81 Time Charter Party Agreement, [Link]

asbatime_nype_81.pdf

ASSIMENOS, N. (2017), Commercial Operations Management, Shipping Operations Management,

Springer

BAL BESIKÇI, E., KECECI, T., et al. (2016), An application of fuzzy-AHP to ship operational energy
efficiency measures, Ocean Engineering 121, pp.392 402

O., et al. (2016), An artificial neural network based decision support

system for energy efficient ship operations, Computers & Operations Research 66, pp.393 401

BERGANTINO, A.S.; VEENSTRA, A.W. (2002), Principal agent problems in shipping: An

investigation into the formation of charter contracts, Int. Assoc. Maritime Economists Conf., Panama
City

BIALYSTOCKI, N.; KONOVESSIS, D. (2016),

speed curve: A statistical approach, J. Ocean Eng. and Science 1(2), pp.157 166

BOCCHETTI, D. et al. (2015), A Statistical Approach to Ship Fuel Consumption Monitoring, J. Ship
Research 59(3), pp.162 171

BOUMAN, E. A. et al. (2017), State-of-the-art technologies, measures, and potential for reducing GHG
emissions from shipping A review, Transportation Research Part D: Transport and Environment 52,
pp.408 421

CLARKSONS (2018), Clarksons Ship Intelligence Network

COGHLIN

DNV-GL (2015), Energy Management Study 2015 Energy efficient operation what really matters.
[Link]

EPPLER, M.J.; MENGIS, J. (2004), The concept of information overload: a review of the literature
from Organization Science, Accounting, MIS, and related disciplines, The Information Society 20(5),
pp.325 344

FABER, J.; , M. (2017), Estimated Index Values of Ships 2009-2016. Analysis of the Design
Efficiency of Ships that have Entered the Fleet since 2009, Delft, [Link]
pdf/publications/2017_07_CE_Delft_7L97_Estimated_Index_Values_of_Ships_2009-2016_def_2017
.pdf

FABER, J.; BEHRENDS, B.; NELISSEN, D. (2011) Analysis of GHG Marginal Abatement Cost
Curves, Delft, [Link]
pdf

GORTON, L. et al. (2009), Shipbroking and Chartering Practice, Informa

HARANEN, M. et al. (2016), White, Grey and Black-Box Modelling in Ship Performance Evaluation,

140
1st HullPIC, Pavone, pp.115 127

HUDSON, D.A.; DANIELS GALLE, G.P. (2017), Investigation of Ship Motions and Fuel
Consumption with respect to Charter Party Agreements, Shipping in Changing Climates Conf.

HUNSUCKER, J.T. et al. (2018), Uncertainty Analysis of Methods Used to Measure Ship Fuel Oil
Consumption, Fort Lauderdale, [Link]
in_Ship_Fuel_Oil_Consumption.pdf

IEA (2007), Mind the Gap: Quantifying principal-agent problems in evergy efficiency, Paris

IMO MEPC (2018), Reduction of GHG Emissions from Ships, MEPC 72/WP.5, IMO, London

ISO (2016), ISO19030-1 Ships and marine technology - Measurement of changes in hull and propeller
performance - Part 1: General principles, Geneva

JAFARZADEH, S.; UTNE, I.B. (2014), A framework to bridge the energy efficiency gap in shipping,
Energy 69, pp.603 612

JEON, M. et al. (2018), Prediction of ship fuel consumption by using an artificial neural network, J.
Mech. Science & Technology 32(12), pp.5785 5796

JOHNSON, H.; ANDERSSON, K. (2016), Barriers to energy efficiency in shipping, WMU J. Maritime
Affairs 15(1), pp.79 96

KARLSEN, R.; PAPACHRISTOS, G.; REHMATULLA, N. (2018), The diffusion of wind propulsion
technologies in shipping: an agent-based model, 9th Int. Sustainability Transitions Conf., Manchester,
pp.1 28

KONTOVAS, C.A.; PSARAFTIS, H.N. (2013),

case of Principal-Agent Problems in operational emission reduction measures, IAME Conf., Marseille,
pp.1 20

LUND, B.; GONZALEZ, C. (2017), The Big Data Concept in a Performance Monitoring Perspective,
2nd HullPIC, Ulrichshusen, pp.76 88

NEUMANN, J.v.; MORGENSTERN, O. (1944), Theory of Games and Economic Behavior, Princeton
University Press.

PLOMARITOU, E.; NIKOLAIDES, E. (2016), Commercial Risks Arising from Chartering Vessels, J.
Shipping and Ocean Engineering 6, pp.261 268

POULSEN, R.T.; JOHNSON, H. (2016), The logic of business vs. the logic of energy management
practice: Understanding the choices and effects of energy consumption monitoring systems in shipping
companies, J. Cleaner Production 112, pp.3785 3797

POULSEN, R.T.; SORNN-FRIESE, H. (2015), Achieving energy efficient ship operations under third
party management: How do ship management models influence energy efficiency?, Research in
Transportation Business and Management 17, pp.41 52

PRAKASH, V. et al. (2016), Revealed preferences for energy efficiency in the shipping markets,
London

PSARROS, G.A. (2016), Energy Efficiency Clauses in Charter Party Agreements: Legal and Economic
Perspectives and their Application to Ocean Grain Transport, Springer

141
RASMUSEN, E. (2007), Games and information: An introduction to game theory, Blackwell

REHMATULLA, N. (2014), Market Failures and Barriers Affecting Energy Efficient Operations in
Shipping, Thesis, University College London

REHMATULLA, N.; SMITH, T. (2015a), Barriers to energy efficiency in shipping: A triangulated

approach to investigate the principal agent problem, Energy Policy 84, pp.44 57

REHMATULLA, N.; SMITH, T. (2015b), Barriers to energy efficient and low carbon shipping, Ocean
Engineering 110, pp.102 112

STEVENS, L. et al. (2015), Is new emission legislation stimulating the implementation of sustainable
and energy-efficient maritime technologies?, Research in Transportation Business and Management 17,
pp.14 25

STOPFORD, M. (2009), Maritime Economics, Routledge

SUH, S.; PARK, N. (2010), The Charter Fixing Negotiation Procedure with Asymmetric Impatience in
a Game Theory Framework: Case Studies in Coal and Ore Transactions, The Asian J. Shipping and
Logistics 26(2), pp.247 261

TAMVAKIS, M.N.; THANOPOULOU, H.A. (2000), Does quality pay? The case of the dry bulk
market, Transportation Research Part E: Logistics and Transportation Review 36(4), pp.297 307

VEENSTRA, A.W.; VAN DALEN, J. (2011), Ship Speed and Fuel Consumption Quotation in Ocean
Shipping Time Charter Contracts, J. Transport Economics and Policy 45(1), pp.41 61

WANG, H. et al. (2010), Reduction of GHG emissions from ships: Marginal abatement costs and cost-
effectiveness of energy-efficiency measures, MEPC 61/INF. 18, IMO, London

YOO, B.; KIM, J. (2019), Probabilistic modeling of ship powering performance using full-scale
operational data, Applied Ocean Research 82, pp.1 9

142
 Techniques for the Automated Detection of Anomalies and Assessment of
Quality in High-Frequency Data Collection Systems
Carlos Gonzalez, Kyma, Bergen/Norway, cg@[Link]
David Lara Arango, Webstep, Bergen/Norway, [Link]@[Link]

Abstract

During the last few years, the advent of big data techniques and their benefits for obtaining valuable
insights enabling better decision-making, has been a trend topic in all the industries. One of these
benefits has been an increased capability to detect patterns, anomalies and confidence boundaries for
different phenomena; which has significantly improved the overall reliability in various processes
across multiple industries. At the same time, in the shipping industry, questions have arisen about the
reliability of the data gathered automatically from sensors. This is of course vitally important for the
industry; for each of the values used to carry out various operational and managerial processes, it is
necessary to consider the accuracy of the sensors, the calibration processes and any inherent errors in
the measurement process. In recent years, many companies in the shipping industry have decided to
invest more resources in the digitalization of their ships; with the goal of gathering valuable data which
will lead to better decisions and, in consequence, give an advantage over their competitors. This has
been resulted in an improvement of the quantity and quality of the telemetry on board their vessels. By
applying machine learning and statistical analysis techniques, this paper shows the anomalies detected
and the accuracy of datasets, collected on board several ships, containing data collected every 15
seconds (this being the frequency defined in ISO 19030).

1. Introduction

The maritime industry is

operations; maximizing benefits by reducing the operational costs. One of the measures used to
optimize ship performance is by achieving better fuel efficiency, since fuel consumption may account
, Bialystocki and Konovessis (2016). Typically, three strategies
are being used for improving ship efficiency, Wan et al. (2018):

Technical solutions: these are technical means to increase the ship performance, such as better
hull design, waste recovery system, etc. This strategy may involve large investment by the
shipping companies
Operational solutions: these are normally defined in the Ship Energy Efficiency Management
Plan (SEEMP). These solutions are taken by the ship operator, and normally do not require a
high level of investment. Some examples of operational solutions are the adoption of slow-
steaming, trim optimization, etc. In order to optimize any operation, it is critical to have data
available to support the changes implemented.
Market-Based solutions: This is the most controversial solution since the potential benefits are
not universally accepted. This solution carries more challenges which are covered by different
authors, Wan et al. (2018), Shi (2016).

On this study, we paid attention to the operational solutions which may be regarded as the most cost-
efficient. One of the most topical operational solutions adopted is that of digitalization and data analysis.
In this regard, shipping companies are investing a lot of resources in digitalizing their business
structures anticipating seeing the benefits of such investments. It has been pointed out that traditional
ways of analyzing the ship performance by using low frequency data (noon reports) might be not good
enough, Aldous et al. (2013), due to higher uncertainty associated with such low frequency data. The
advent of big data techniques and their benefits for obtaining valuable insights enabling better decision-
making has been and remains a hot topic across the industry. The companies need to manage the data
collected and get insights from that.

143
Nowadays, vessels are using more and more sensors to measure and capture an extensive set of data
including from navigation, cargo, machinery and auxiliary systems. The continual improvement in
quality and accuracy of such sensors giving very high level of precision in the measurements and data
transmission, Gonzalez et al. (2018).

As part of their digitalization strategy, the shipping companies have realized they need systems to gather
and store automatically the various inputs from those sensors, instead of overloading the crews with
additional manual reporting requirements. One way of solving this, is by installing Continuous
Monitoring (CM) and data acquisition systems. These tools will automatically supper better decision
making and reporting, with more transparency. However, in order to retain the advantage of CM system,
priority must be given to monitoring the quality of the input. Otherwise their results become valueless.
One typical example about the benefits and challenges of using high-frequency data and CM systems
can be found when an overconsumption is analyzed. The reason for such overconsumption could be
found in a decrease of the vessel performance. However, a sensor malfunction, wrong or missing
manual recordings, adverse weather, or other external factors can also cause it. Therefore, automatic
data collection, filtering, repeatability and transparency in a performance monitoring system are critical
elements for the credibility of the CM system. Once the input quality is assessed, the combination of
displaying instant performance values together with investigating the long trend of important key
performance values are keeping the crew and the management continuously and accurately updated on
status, Hagestuen et al. (2016).

Besides the beforementioned, the ship-to-shore communication and internet availability on board has
been significantly improved, making it easier to transfer data from ship to shore. As such, it helps the
shore staffs to perform deeper data analysis and even anticipate to a problem on board.
Before 2025, many ships, systems, and components will be linked to the Internet, making them
accessible from almost any location in the world. Maritime connectivity will advance significantly,
forcing to the industry to be ready for using and getting insight from such huge volume of data and at
such high velocity, [Link]

One factor to be considered as essential when the companies are defining their digitalization strategies
and data management plans is, what they need to monitor to get valuable results.

required parameters to
monitor the ship performance. In this regard, ISO, by means of the standard 19030, is trying to
homogenize and define the required parameters and methods to evaluate the hull and propeller
performance in a standard way.

One of the bases for the ISO 19030 is the automatic high-frequency data collection from sensors on
board. This is a delicate issue due to the controversies within the industry about the reliability of the
data measured by sensors. Adding more complexity to that problem, the growing volumes of data and
the increasing number of source systems can lead to possible data errors, duplicates, missing values,
incorrect formatting and contradictions in data analysis, Azeroual et al. (2018).

In this study, we focus on the automated detection of anomalies and assessment of quality in the high-
frequency data collected by continuous monitoring system. The technique used for this study is a
machine learning technique called Long-Short-Term-Memory (LSTM) neural networks, which are
applied on four ships with four datasets formed by high-frequency data (15 s) according the ISO 19030
standard. Other metrics applied on statics are being applied in order to show the quality of the datasets.

2. Methodology and data acquisition

The methodology used is based on the collection of high frequency in-service data automatically from
on board the by a Continuous Monitoring system (CM) system installed on four ships. The CM collects
data from different sensors (Table 1) at 15 s intervals which is averaged into hourly readings for
increasing the performance in the data quality assessment.

144
 Table 1: Parameters and sensors used on this study
Variable Sensor Error Associated
Shaft Power Torque-meter 0.5 % (a
Shaft RPM Torque-meter 0.1 (a
ME Fuel consumption Mass flow meter 0.1-0.2 % (b
Speed Through the water Doppler log 3 % (c
Speed over ground GPS 5 % (c
Draft astern and forward Draft sensors +0.1 m (c
Rudder Angle Rudder Angle Indicator N/A
Ship heading Gyrocompass N/A
Depth of water Echo-sounder N/A
Wind relative speed and direction Anemometer N/A
(a
: Kyma Power Meter, (b: Hunsucker et al. (2018), (c : ISO (2016)

ISO 19030 defines two levels of parameters, primary and secondary parameters. Primary parameters
are two and they are the Shaft Power and the Speed through water. In Table 1, the primary parameters
are in green font and the secondary in blue fonts.

Once the high-frequency data is collected on board by the CM system, it is stored in a database and
then the data is averaged into hourly values. Following to this averaged, the data is filtered eliminating
outliers and, as a final step to assess the data quality, machine learning technique is applied on the data
sets in order to detect anomalies on the inputs from sensors. The whole process is described in Fig.1.

Fig.1: Methodology and data acquisition flow diagram

On this study, we only focus on data quality assessment and anomalies detection for the parameters
Speed Through Water, Shaft Power, Shaft RPM and Speed Over Ground.

2.1. Data Filtering applied on the raw data

The raw data collected by the CM system, is automatically filtered as it is shown in Table 2, to avoid
outliers. These conditions are the only filters applied on the datasets subject of this study.

145
 Table 2: Data filtering applied on datasets
Condition Output
If Shaft RPM < 2 rpm SOG = 0
STW = 0
SP = 0
ME Fuel = 0
If Shaft Power < 100 kW SOG = 0
STW = 0
RPM = 0
ME Fuel = 0
SOG: Speed Over Ground
STW: Speed Through Water
SP: Shaft Power

2.2. Data quality assessment

There are multiple approaches to assess data quality. These approaches can be differentiated based on
the level of control one has over the data generation process. In this sense, data quality assessment can
be either experimentally based or real-performance based. In this paper we focus on the latter. Among
the available real-performance based quality assessment approaches, we make a distinction between
two modelling paradigms, statistical modelling and machine learning modelling (also known as
algorithmic modelling; see Breimann (2001) for a thorough discussion on these two modelling
approaches from an ontological point of view). Statistical modelling is based on mathematical proofs
and widely accepted. Statistics collect, arrange and analyse data sets, providing information about the
instruments used for gathering the data. However, they frequently fail when dealing with complex and
highly nonlinear, Karlaftis and Vlahogianni (2011).

Machine learning modelling (ML) approaches are more flexible; include learning processes, offer better
performance processing outliers and mitigating noisy inputs, Ma et al. (2015). These approaches are
currently accepted as a method to manage high dimensional and non-linear relationships.

In this study, we have decided to assess the data quality of our inputs (values from sensors) using the
Machine learning. Specifically, we make use of a type of Deep learning model, which specializes in
dealing with time-series data, namely Long Short-Term Memory (LSTM).

LSTMs belong to a class of neural networks currently referred to as Recurrent Neural Networks
(RNNs). RNNs are neural network-based models that are particularly geared towards processing time
series type data. A neural network is a group of interconnected nodes (neurons) organized in layers.
Each layer of neurons receives input from the previous layer such that each of the neuron s output is
the result of a non-linear combination of the weighted input it receives from neurons in the previous
layer. There -linear
outputs will be. In this sense, we can roughly categorize neural networks based on this metric (neurons
s rise to the nowadays quite popular Deep

there is no universal consensus on how many layers are needed in order to consider a neural network to
be deep, the difference between a deep neural network and a shallow (not deep) neural network is often
depicted as in Fig.2.

The interaction between neurons across layers is regulated by activation functions. Activation functions
take in the inputs from previous neurons and produce a non-linear output. Some examples of these
functions are Sigmoid, RELU and Tanh, see Mohammed et al. (2018) for an application-based
discussion on these functions.

146
 Fig.2: Non-deep and deep neural networks

Once all the neuron layers have been activated through a succession of activation functions, the outputs
of the output layer are contrasted with a real variable. In our case, the output layer is contrasted with
the measurements we have for each of the variables we have included in this study. By performing such
contrast, a prediction error is calculated, and departing from this error, the weights of all neurons are
changed such that this error is minimized.

This process is often referred to as back-propagation. Specifically, the error is treated as an objective

carried out using multiple optimization algorithms, which are often based in the mathematical process
of gradient descend, Friedman (2011). In order to use neural networks to process time series data, it is
necessary to somehow keep track of time and transfer learning from one-time instance to another i.e. it
is necessary to include memory into the neural network. RNNs do exactly that, they give memory to
neural networks by creating stacked, sequential architectures that can capture time, Georgiopoulos et
al. (2011). While creating such architectures gives neural networks the ability to deal with time-
dependent data, it also creates non-trivial mathematical challenges to ensure that the back-propagation
process is able to minimize the prediction error, Hochreiter (1998). In order to solve these challenges,
a type of neural network with a more sophisticated memory mechanism that received the name of Long
Short Term Memory (LSTM). The models we design and implement of this paper are LSTM neural
networks. For the LSTM model used in this study, dependent variables have been defined in order to
train the model to predict the results and see how far or close these are from the real values, assessing
the data quality by determine the R-square of the series for each input from sensors. The LSTM
approach is well explained on different works, Ma et al. (2015), Bodén (2002), Jaeger (2002,2012),
Ordóñez and Roggen (2016). In order to train and validate the LSTM model, the data was split into two
sets: the training set and the test set. The training set was used to train the LSTM, while the test set was
used for prediction only. The variables defined for the training model are shown on the Figs.3-6.

Fig.3: Fig.4:

147
Fig.5: Fig.6:
STW: Speed Through Water SOG: Speed Over Ground

For the numerical analysis, three metrics are used on this study in order to assess the data quality of the
inputs. These are:

Mean Absolute Error (MAE): it is a measure of difference between two continuous variables.
It is one of the most common metrics used to measure accuracy for continuous variables. MAE
measures the average magnitude of the errors in a set of predictions, without considering their

and actual observation where all individual differences have equal weight.
Mean Square Error (MSE): The mean squared error tells you how close a regression line is
to a set of points. It does this by taking the distances from the points to the regression line (these

signs. It also gives more weight to larger differences. It is called the mean squared error as you
are finding the average of a set of errors.
Coefficient of Determination (R-Square): R-
assess the goodness of fit of our regression model. In R-squared we have a baseline model

to predict the value of dependent variable Y. Instead it uses the mean of the observed responses
of dependent variable Y and always predicts this mean as the value of Y. Any regression model
that we fit is compared to this baseline model to understand its goodness of fit.

2.3. Anomalies detection

In this study, anomalies are defined as a function of predictions and real values. More specifically, the
difference between real and predicted values i.e. the prediction error, is used as a base for determining
whether a particular observation in the data can be considered an anomaly. In this sense, an anomaly is
identified as such if the error term exceeds one of a given set of upper and lower thresholds where these
mean and standard deviation values across a specific time span. In
this study, we defined four main parameters, namely a rolling window (time span), and three multipliers
for the standard deviation of the error. In this specific study, we used a rolling window of 3 hours (i.e.
predictions for a specific time are made using data from the time in question together with data from up
to three hours before), and multipliers (K) of 1.5, 1.75 and 2.0 for the standard deviation. The upper and
lower thresholds for anomaly detection are therefore written as shown in the Eq.(1):

148
K takes the values of 1.5, 1.75 and 2.0. This results in three pairs of thresholds, one for each K. An
anomaly is defined as an observation that exceeds at least one of such thresholds by being either lower
poor higher than the threshold. This should provide an acceptable basis for anomaly detection, provided

are normally distributed with mean zero).

While this methodology directly allows to classify anomalies according to their severity (e.g. anomalies

, K=1.75 or the
K=1.5 respectively), we decided to plot their severity directly instead. Therefore, anomalies are depicted
in our graphs results if an only if they have exceeded the lowest threshold (K=1.5) and their severity
(e.g. their error term value) is plotted as is in red.

3. Vessel Characteristics

The data from four ships have been used on this study. The ships characteristics are shown in Table 3.

Table 3: Ships characteristics

Vessel No Group Ship type Construction Year Built in DWT
1 COT 2016 DSME 300 000
A
2 COT 2016 DSME 300 000
3 LNG carrier 2017 DSME 95 785
B
4 LNG carrier 2017 DSME 95 785
The ships are sister ships between each ship type, therefore ship 1 2 (group A) and the vessels 3 4
(group B) are sister between them.

4. Results

The data quality and detection of anomalies in the inputs from sensors are detailed graphically and
numerically for each group of ships in this section. It is shown on Figs.7 and 8 and in Table 4.

Fig.7: Data prediction and anomalies for vessel group A

149
 Fig.8 Data prediction and anomalies for vessel group B

Table 4: Data quality Numerical results

Group Metric* Shaft power (1 Shaft RPM Speed over ground Speed through water
MSE 31016.1 0.08 0.02 0.03
A MAE 15.03 0.02 0.01 0.02
R-squared 0.99 0.99 0.99 0.99
MSE 13636.2 0.03 0.01 0.01
B MAE 7.33 0.01 0.08 0.01
R-squared 0.99 0.99 0.99 0.99
(1 .

The data quality on these two ships (vessel group A) is good due to the prediction calculated by the
LSTM model (blue line) is very close to the real values measured by the sensors (green lines) for the
four parameters. This means that the data quality for this vessel group and these parameters is fairly
good as far as real performance tests enable us to determine. A full assessment of the data quality
requires experimental tests to complement these findings.

On the graphs, the red vertical-lines represent when the difference between the predicted and real values
is significant (anomalies), and the magnitude of the red lines shows how big the difference between
both values is. Then, checking the number and the length of these red lines determine the number and
magnitude of the anomalies detected. The fewer red lines the less anomalies found.

For the vessel group A, there are not so many anomalies and they are relatively short, indicating that
the sensors reliability is quite good. For the vessels group B, the data quality is also fairly good, going
the predicted and real values very close. And, very few anomalies were detected by the model, therefore
the sensor data is accurate enough to provide good data analysis.

Besides of the graphical data, numerical analysis is done by checking out the three metrics selected
and explained on the section 2.2. The results are presented in Table 4.

150
Concentrating only on how goodness fit is for the variables, it is demonstrated that the four parameters
for these four ships analysed are very good, been the R-square values
condition). Comparing both groups, the data quality for the group B is slight better than the quality on
vessel group A, according to the metrics MSE and MAE. The R-squared is the same for all the
parameters.

5. Conclusions

a) Confidence of anomalies are very related to the quality of the predictions.

b) On this study, only one data set is used for training and another one is used only for prediction.
It is reasonable to expect that more data, for both training and testing, would generate better,
more accurate models.
c) The data quality on the two primary parameters defined by ISO 19030 is fairly good for the
ships analysed
d) The data quality for the other two parameters analysed, Ship Over Ground and Shaft RPM is
also good enough.
e) The number of anomalies detected is not so high, suggesting a good quality of the sensors
installed on board
f) The ships are operating on real situations; therefore, the data collection is done under an
uncontrolled environment.
g) Due to the uncontrolled environment, it is not possible (on this study) to determine if the
anomalies are caused by sensor bias or because the actual ship operation.
h) A data quality assessment is essential for validating the further data analysis.
i) Detection on sensor anomalies can also be used for making an automatic sensor status-check.
j) The current model used can be extended in complexity to assess many more components,
providing additional results and insights about the data quality and sensor quality.
k) For future work, data analysis of other ship performance parameters should be considered for
getting the whole data quality assessment of the data sets.

References

ALDOUS, L.; SMITH, T.; BUCKNALL, R. (2013), Noon report data uncertainty, Low Carbon Shipp.
Conf., pp.1 13

AZEROUAL, O.; SAAKE, G.; SCHALLEHN, E. (2018), Analyzing data quality issues in research
information systems via data profiling, Int. J. Inf. Manage. 41, pp.50 56

BIALYSTOCKI, N.; KONOVESSIS, D. (2016),

speed curve: A statistical approach, J. Ocean Eng. Sci. 1/2, pp.157 166

BODÉN, M. (2002), A guide to recurrent neural networks and backpropagation, Dallas Proj. SICS
Tech. Rep. 2, pp.1 10

BREIMAN, L. (2001), Statistical Modeling: The Two Cultures, Stat. Sci. 16/3, pp.199 231

FRIEDMAN, J.H. (2011), Greedy function machine: A gradient boosting machine, Statistics 29/5,
pp.1189 1232

GEORGIOPOULOS, M.; LI, C.; KOCAK, T. (2011), Learning in the feed-forward random neural
network: A critical review, Perform. Eval. 68/4, pp.361 384

GONZALEZ, C.; LUND, B.; HAGESTUEN, E. (2018), Case Study: Ship Performance Evaluation by
Application of Big Data, 3rd HullPIC, Redworth, pp.104 119

HAGESTUEN, E.; B. LUND, B.; GONZALEZ, C. (2016), Continuous Performance Monitoring A

151
Practical Approach to the ISO 19030 Standard, 1st HullPIC, pp.49 61

HOCHREITER, S. (1998), The Vanishing Gradient Problem During Learning Recurrent Neural Nets
and Problem Solutions, Int. J. Uncertainty, Fuzziness Knowledge-Based Syst. 6/2, pp.107 116

HUNSUCKER, J.T.; PRZELOMSKI, D.; BASHKOFF, A.; DIXON, J. (2018), Uncertainty Analysis of
Methods Used to Measure Ship Fuel Oil Consumption, January

ISO (2016), ISO 19030 - Part 1, ISO, Geneva

JAEGER, H. (2002), A Tutorial on Training Recurrent Neural Networks, Covering BPPT, RTRL, EKF
, German National Research Center for Information Technology

JAEGER, H. (2012), Long Short-term Memory in Echo State Networks: Details of a Simulation Study,
Jacobs University

KARLAFTIS, M.G.; VLAHOGIANNI, E.I. (2011), Statistical methods versus neural networks in
transportation research: Differences, similarities and some insights, Transp. Res. Part C Emerg.
Technol. 19/3, pp.387 399

MA, X.L.; TAO, Z.M.; WANG, Y.H.; YU, H.Y.; WANG, Y.P. (2015), Long short-term memory neural
network for traffic speed prediction using remote microwave sensor data, Transp. Res. Part C Emerg.
Technol. 54, pp.187 197

MOHAMMED, M.A.; NAJI, T.A.H.; ABDULJABBAR, H.M. (2018), The effect of the activation
functions on the classification accuracy of satellite image by artificial neural network, Energy Procedia,
pp.164 170

ORDÓÑEZ, F.J.; ROGGEN, D. (2016), Deep convolutional and LSTM recurrent neural networks for
multimodal wearable activity recognition, Sensors (Switzerland) 16/1

SHI, Y. (2016), Reducing greenhouse gas emissions from international shipping: Is it time to consider
market-based measures?, Mar. Policy 64, pp.123 134

WAN, Z.; EL MAKHLOUFI, A.; CHEN, Y.; TANG, J. (2018), Decarbonizing the international
shipping industry: Solutions and policy recommendations, Mar. Pollut. Bull. 126, pp.428 435

152
 Study on Additional Ship Resistance due to Roughness using CFD
Anders Östman, SINTEF Ocean, Trondheim/Norway, [Link]@[Link]
Kourosh Koushan, SINTEF Ocean, Trondheim/Norway
Luca Savio, SINTEF Ocean, Trondheim/Norway

Abstract

The additional resistance due to roughness is studied by means of CFD simulations. The KVLCC2 hull
at full-scale Reynolds number is considered as a test case. A wall function formulation is used to model
the rough wall turbulent boundary layer, where the roughness function is based on data from towing
flat plates coated with paint of similar roughness as for the full-scale vessel. The additional resistance
for coatings with various roughness heights is studied, with roughness heights ranging from less than
-cost reduction of frictional resistance is
investigated. High-quality paint coating (with low roughness) can be applied at given locations where
the skin friction is high, while using cheaper coating and application procedures (resulting in larger
surface roughness) at other locations where skin friction is of less importance.

1. Introduction

Resistance due to fouling and poorly applied antifouling coating can have a significant contribution to
the total resistance of a ship. This is especially true for ships operating at low Froude numbers, where
skin friction resistance is the dominating component of the hydrodynamic resistance and could account
for 60% or more of the total resistance. Reliable estimations of added resistance due to rough hull
surfaces are important in order to be able to preform speed prediction of the vessel. Also, insight into
the relative importance of roughness at different parts of the hull can be used to give guidance on where
and how to apply antifouling coating on the ship.

Numerical simulations of rough surface friction drag are traditionally based on roughness functions that
relies on finding an equivalent sand grain height that fits Nikuradse (1933) pipe flow experiments,
examples can be found in Vargas and Shan (2016) and Deminel et al (2014).

In the present paper a different approach is applied, the roughness function is based directly on
experimental resistance test of the specific surface coating. The roughness function is derived based on
experimental flat plate towing tank tests. Plates with various surface roughness were towed at constant
speeds, Savio et al (2015), and the results were post processed using methods proposed by Granvile
(1987). The roughness levels of the different plates were related to typical real applications processes
used in the marine industry. Results from these tests were implemented in customized rough wall
functions in the OpenFOAM flow solver, and presented in a validation study, Östman et al (2017). The
validation showed very good agreement of computed flat plate resistance against experiments.

The method is in the present paper applied on a full-scale ship hull. The coatings and roughness models
developed in the 2D validation study is used to model the rough surface ship hull. The computed total
friction resistance of the various coatings is compared. Also, the effect of applying the highest quality
coating at limited areas, selected based on the computed friction coefficient, is studied.

2. Formulation of the roughness wall function

The implemented physical model that is used to model rough surfaces in the CFD simulations relies on
measured flat plate resistance experimental results. Flat plates with various surface roughness were
towed at constant speed while resistance was recorded, Savio et al (2015). The roughness on the plates
was due to paint applied on the surface of the plates with various quality of application process. The
aim was to mimic typical real application processes used in the marine industry. Three roughness levels
(denoted A, B and C with increasing order of roughness) were considered. Roughness level A represents

153
an optimal new build or full blast dry docking application of the paint. Roughness level B corresponds
to dry dock situation with some underlying spot repair roughness and poor coating application of the
paint. Finally, the plate with the most severe roughness was denoted level C, which could simulate an
extreme case with severe underlying roughness accumulated from several dry dockings and very poor
application of the paint. In addition, a set of smooth blank plates were used in order to have a reference
to the theoretical smooth boundary layer friction drag.

The measured drag was post-processed following methods proposed by Granvile (1987) and presented
in terms of inner variables, Fig.1 velocity profile in the
logarithmic part of the boundary layer as a function of the non-dimensional roughness height, k+, where
k+ is defined by k+ = kU is

Reynolds number for the surface roughness in the boundary layer. The value of typical roughness
height, k, is found from a statistical analysis of the actual rough surface and defined as the rms (root
mean square) of absolute heights of the surface, and denoted Sq in the following. The statistics of the
surface is found from analyzing high-resolution laser scanns of imprints of the surface. The measured
rms roughness height of the plates is presented in Table 1. Visualisations of the surface from the laser
scan is shown in Fig.2.

Fig.1: Presentation of the experimental data in terms of inner variables

Fig.2: Visualization of surface scans of the plates

154
Table 1: Measured root mean square of absolute heights of the surface roughness (Sq) of the plates

The towing test results of the rough plates was used to derive the roughness function that is implemented
in the CFD solver. The experimental results, Fig.1, show that each plate has a linear relation between
velocity shift and log(k+). Based on this observation, the idea of developing dedicated roughness
functions for each surface coating was considered.

The velocity profile in the log law region is described by the equation, Cebeci and Bradshaw (1977):

where =0.41 is the von Karman constant and E is a constant which equals 9.8 for smooth walls. For
rough walls the velocity profile is switched downward in the logarithmic region. This can
mathematically be expressed by substituting E with a modified variable E' defined as

Where the roughness function, f, can be found directly based on experimental results of the velocity
shift ( +). The procedure on how to estimate f directly from measurements are described in the
following. Inserting the expression given in Eq. (2) into Eq. (1) gives:

+
The last term in the equation is the velocity shift,

+
is defined to be positive when the velocity profile is shifted downwards. The roughness function f
can now be found directly from Eq. (4):

As seen in the experimental results for the velocity shift that are presented in Fig.1 as a function of k+,
it is evident that a logarithmic fit can be found for each plate. An expression of the velocity shift can
be formulated as:

where a0 and a1 are constants of the curve fit. The best fit for different plates are shown in Fig.3.

4. CFD solver

The flow solver used is the simpleFOAM mono fluid solver included in the OpenFOAM package. The
solver solves the steady-state fluid flow using the SIMPLE algorithm, Ferziger and Peric (2002). The
k-omega SST turbulence model is used to model turbulence. The flow over the rough surfaces is
modeled by means of modifying the smooth wall function as described in the previous section.

155
 Fig.3 Curve fit of measured velocity shift for the plates

5. Flat plate validation study

The flat plate validation study, presented in Östman et al (2017), is in the following shortly summarized.
The problem is simplified in the CFD analysis by neglecting wave generation and end-effects of the
towed plates. This is done by solving the equations for a mono-fluid flow field in a 2D dimensional
flow domain. Separate meshes were generated for each speed, the meshes were generated with a target
for the near wall mesh spacing that results in y+ , Fig.4.

Fig.4: Mesh in the flow domain and boundary conditions

Fig.5: Skin friction resistance coefficient. Comparison of CFD results using the direct roughness
function formulation Eq.(5) against experimental results.

156
The simulations were performed for the same speeds as tested in the towing tank, U=3, 5, 7 and 9m/s.
The computed results are compared against experiments in Fig.5. The comparison is very good for all
plates. The implemented roughness function is thus able to accurately model the behavior of the rough
surfaces.

5. Full scale ship hull simulations

Simulation of the full scale KVLCC2 hull is performed at 15 kn. The length between perpendiculars of
the vessel is 320 m, resulting in a Reynolds number of 2.08*109 and Froude number 0.137. The same
rough hull coatings as in the flat plates experiments is assumed. At this low Froude number, wave
resistance is of less importance, while skin friction resistance being the dominant resistance component.
The purpose of the present study is to quantify the increase in resistance due to hull surface roughness.
It was therefore decided to simplify the simulation setup by replacing the free surface with a fixed slip
surface. The motivation for this simplification is: (i) Changes in wave resistance due to hull surface
roughness is assumed to be very small. The wave pattern, and hence, the wave resistance is not expected
to be influenced by the hull surface roughness. (ii) We are only interested in the difference in resistance
due to hull roughness, thus, the actual level of resistance is of less importance as long as the difference
in resistance is captured by the simulation setup. (iii) The wave resistance is anyhow a small component
of the total resistance due to the small Froude number.

The rough surface was modelled using the direct roughness function with the same parameters as for
the flat plate simulations. The 3D volume mesh was generated using the HEXPRESSTM grid generator.
Illustrations of the flow domain and mesh on the hull surface in the symmetry plane are shown in Fig.6
and Fig.7. The mesh is refined in the boundary layer in order to accurately capture the boundary layer
profile. The aim is to have a y+ value in the range 40-100 at the cell centre in vicinity to the hull surface.
Based on boundary layer theory the size of the first cell normal to the hull surface is chosen as 0.6 mm.
The total number of grid cells was approximately 6.2M.

Fig.6: Illustration of mesh on the hull surface and in the symmetry plane

Fig.7: Mesh on the hull surface and in the symmetry plane in vicinity to the hull

157
The computed wall shear stress on the hull for the different surface coatings is presented in Fig.8. For
simplicity, the surfaces are denoted to have different coatings, this is strictly not correct, all surfaces
are treated with the same coating, but using different quality of the application process, resulting in
different surface roughness. However, the term "coating" is in the following used to simplify the presen-
tation.

As expected, the wall shear increases with increasing surface roughness. Moreover, the increase in
shear stress is seen on the entire hull surface. As was also observed in the flat plate simulations, the
smoothest coating is very smooth, which results in a shear stress that is very similar to the stress on the
hydraulic smooth reference surface. The largest values of wall shear stress are seen in areas with accel-
erations in the flow, such as around the shoulder and at the bilge in the bow area. The computed friction
resistance coefficients are compared in Table 2. The increase in resistance for coating A, compared to
the smooth surface, is only 0.9%. Coating B results in about 11% increase, while the increase in re-
sistance for coating C is 24% compared to the smooth surface.

Fig.8: Computed wall shear stress at the hull surface for the different coatings

In an attempt to investigate the relative importance of surface coating quality at different parts of the
hull, the hull surface was split based on shear stress threshold values. Using coating C as a basis, areas
with shear stress exceeding 60, 65 and 70 Pa were identified. The hull surface was thereafter split in
two parts based on these threshold values. The splits are shown in Fig.9. The red part illustrates areas
with high wall shear stress. The size of the area of the high shear part of the hull depends on the
threshold value. The area of the part of the hull where w > 70 Pa is 134 m2, this corresponds to
approximately 0.5% of the total wetted hull surface area. Threshold w > 65 Pa results in an area of 798

158
m2 (3%), while w > 60 Pa results in an area of 2433 m2 (9%). Simulations was thereafter performed
using the high quality coating A at the part of the hull where the shear stress exceeds the threshold
value, while the remaining part of the hull has the rough coating C. The computed wall shear stress on
the hull surface is also shown in the figure. The shear stress is significantly reduced at areas where the
smooth coating is applied.

Fig.9 Above: The hull surface splitted based on a wall shear stress threshold. Below: Computed wall
shear stress at the hull surface. Simulations using different coatings. Surface coating C is applied
on the grey part of the hull, coating A is applied on the red part.

Table 2: Computed friction resistance coefficient of hull with different coatings. The increase relative
to the smooth hull is also presented.

The computed friction resistance coefficients are presented in Table 2. The resistance is reduced for the
hull with mixed coatings compared to the hull with coating C. However, the reduction is not very large.
For Split 1, where only 0.5% of the hull is treated with coating A while the rest of the hull consist of
coating C, the increased resistance compared to the smooth reference is 23.8%, instead of 24% for the
hull with coating C on the entire wetted surface. This corresponds to approximately 1% "reduction of

159
increase" (0.2% of 24%). That is, by treating 0.5% of hull with a high-quality coating, the increase in
resistance is reduced by 1%. By increasing the area of the part which is coated with coating A the
resistance is further reduced. For Split 3, where 9% of the hull is treated with coating A, the increase in
resistance is reduced to 21.4 % compared to the smooth surface hull simulations. The computed re-
sistance of the various surface coatings is also compared in Fig.10.

Fig.10: Computed increase in viscous resistance for the different coatings

6. Conclusions

The additional viscus resistance due to surface roughness on a full-scale ship hull has been studied
using CFD simulations. Three different rough coatings were modeled, the rough surfaces correspond
to realistic hull surface conditions found in the marine industry. Parameters in the numerical
implementation of the roughness function, which is used in the turbulent wall function, relies on towing
tank experiments conducted on coated sample plates.

The simulations showed, for the roughest coating, an increase in viscous resistance of 24%, compared
to a smooth hull surface. The potential of low-cost reduction of frictional resistance was also
investigated. When a low roughness coating was applied at locations where the shear stress is high,
while the rest of the hull had a high roughness coating, the resistance was reduced compared to having
the same rough coating on the entire hull. However, the reduction of viscous resistance was not very
large. When 9% of the hull has a low roughness coating, while the rest of the hull is coated with the
roughest coating, the increase in viscous resistance was computed 21.4%, instead of 24%, which is the
increase in resistance when the entire hull is coated with the roughest coating.

Acknowledgements

Part of this work was -toxic cost-efficient environment-friendly

References

CEBECI, T.; BRADSHAW, P. (1977), Momentum Transfer in Boundary Layers, McGraw-Hill

DEMIREL, Y.K.; KHORASANCHI, M.; TURAN, O.; INCECIK, A. (2014), A CFD model for the
frictional resistance prediction for antifouling coatings, Ocean Engineering 89, pp.21-31

FERZIGER, J.H.; PERIC, M. (2002), Computational Methods for Fluid Dynamics, Springer

GRANVILLE, P.S. (1987), Three Indirect Methods for the Drag Characterization of Arbitrarily Rough
Surfaces on Flat Plates, J. Ship Research 31, pp.70-77

160
GRIGSON, C.W.B. (1992), Drag losses of new ships caused by hull finish, J. Ship Res. 36, pp.182
196

NIKURADSE, J. (1933), Laws of flow in rough pipes, NACA Technical Memorandum 1292

ÖSTMAN, A.; KOUSHAN, K.; SAVIO, L. (2017), Numerical and Experimental Investigation of
Roughness Due to Different Type of Coating, 2nd HullPIC, Ulrichshusen

SAVIO, L.; BERGE, B.O.; KOUSHAN, K.; AXELSSON, M. (2015), Measurements of added re-
sistance due to increased roughness on flat plates, 4th Int. Conf. Advanced Model Measurement Tech-
nology for the Maritime Industry, Istanbul

VARGAS, A.; SHAN, H. (2016), A numerical approach for modeling roughness for marine applica-
tions, FEDSM2016-7791, ASME 2016 Fluids Engineering Division Summer Meeting, Washington

161
 Influence of Data Sources on Hull Performance Prediction
Daniel Schmode, DNV GL, Hamburg/Germany, [Link]@[Link]
Ole Hympendahl, DNV GL, Hamburg/Germany, [Link]@[Link]
Francesco Bellusci, Scorpio, Monaco, fbellusci@[Link]

Abstract

Hull Performance assessment is done based on very different data sources. Basic noon reporting is
already implemented on all vessels. Specific reporting software is guiding crew and reducing
reporting mistakes. Complex auto-logging systems reduce the human factor and deliver higher data
frequencies. All sources have pros and cons. Here we present ways to combine these sources with 3rd
party data. Best combination of sources with respect to value to cost ratio is differing depending on
vessels equipment and operation. A scoping exercise is always recommended before engaging in any
project to properly define objectives (improve commercial performance, improve technical perfor-
mance, etc.) and invest as consequences.

1. Introduction

All commercial vessels report data in different formats. Some are also equipped with automatic data
acquisition systems. Owners, managers and operators are interested in assessing vessel performance.
The quality of the performance assessment depends on the input data which differs in quality and
quantity.

We are following the discussion arguing either to do analysis based on simple noon or on high-end
auto-logged data. In reality, things are not just black or white; we see many shades of grey. In the
following, we discuss the pros and cons of different data sources as well as options to combine them
to get better results.

2. Available data sources

We see a wide spectrum of reporting regimes on-board vessels today. They are driven by different
stakeholders, such as vessel owners, technical managers, operators, routing companies, etc. Because
stakeholders have different aims, reporting systems vary in scope and technology. Often vessels are
required to report similar information to different stakeholders through different systems at the same
time.

Below we try to categorize the reporting systems we have seen. Here we categorize from the hull
performance prediction perspective. Thus, we might skip information relevant for other tasks. For
other purposes, a different categorization might be more suitable.

2.1 Noon reporting

All commercial vessels do noon reporting due to SOLAS requirements. Typical noon reports contain
time, position, speed, draft/trim, weather and fuel RoB/consumption data. Noon reports are sent every
noon. Thus, their frequency is once in 24h. In case the time zone changes, the period between two
reports can be either 23 or 25 hours. This needs to be accounted for if consumption rates are
calculated. Most noon reports include the current operational status (sea/port), but do not differentiate
consumptions by operation modes like sea passage, maneuvering or anchoring.

Data format varies from simple free text email to complex spreadsheet forms. Free text email is hard
to process for a computer, since small differences (e.g. comma instead of decimal point) can irritate
the parsing algorithm. Spreadsheet forms with automatic format validation overcome the issue.

162
Operational modes are not separated; thus, consumption cannot be mapped to operational modes. This
is a substantial intrinsic inaccuracy. Typically, crews do not get feedback on the reporting quality,
which can lead to inaccurate reporting. Manual reporting is open to misuse, typically for fuel
consumption, as to justify inefficiencies or increase efficiencies.

Once forms are developed, documented and deployed, there are no further costs related to data
recording. Due to a lack of structure in the data, parsing and analyzing the data might cause high
costs.

2.2 Event reporting

Some vessels have implemented an event-based reporting using complex spreadsheet or html forms or
specialized reporting software. Typical event reports contain time, position, speed, draft/trim, weather
and fuel RoB/consumption data. Often also basic voyage planning (next port, ETA, distance to go, ...),
Cargo and SOF related data, engine RPM and running hours, and lube oil and sludge RoBs is reported
precisely. Often also detailed fuel quality is reported from bunker delivery notes including sulfur
content and LHV.

Event reports are sent at least every noon. Thus, the minimum frequency is once in 23h/24h/25h. If
the vessel changes operation mode (sea passage, maneuver, port) a specific event is reported. Typical
events are: arrival/departure, begin/end of sea passage (BOSP/EOSP), begin/end of anchoring/
drifting. Advanced systems also allow having special events for bunkering, sounding, disposal, cargo
operations, etc.

Event reporting systems differentiate between port, sea passage, maneuvering, drifting, anchoring,
ballast, laden. Consumptions/RoBs are recorded for every change of operational mode.

Data format varies from complex spreadsheet/html forms to specialized reporting software with RoB
bookkeeping and plausibility checking.

Operational modes can be analyzed separately. Systems with built-in plausibility checks avoid typos
or unit mismatch. Risk of involuntary or voluntary error is always present, a vessel tracking system
service tends to reduce certain inaccuracies.

When 3rd party software is used typically some license costs for reporting and aggregation of events.
Analyzing the results is then straightforward.

2.3 Snapshot reporting

Event reporting allows reporting average numbers (e.g. speed, power, wind) between the events.
These numbers are not always constant over the reporting period. With respect to hull performance
evaluation, averaging over a 24h period introduces substantial noise. Thus, DNV GL introduced the
concept of snapshot reporting. Here figures are averaged over a short period (15-30min) and reported
in a special snapshot event. Crews are advised to do this measurement when weather and vessel are in
steady state.

Typically, GPS data, heading, speed log, draft/trim, wind, RPM, shaft power, fuel flow and often M/E
parameters like T/C speed, exhaust gas temperature, etc. On sea passage, reports are created 1-3 times
per day, and only on sea passage.

The reports are transmitted as spreadsheet or reporting software. Trained and motivated crews
typically deliver good quality data which is sufficient for hull performance evaluation. There are no
additional cost involved, just crew time to carry out measurements.

163
2.4 Automatic data acquisition systems

Especially modern and complex vessels have many sensors connected to an automatic data acquisition
system. Typical sensors are GPS data, gyrocompass for heading, speed log, draft/trim, wind, RPM,
shaft power, fuel flow and often M/E parameters like T/C speed, exhaust gas temperature, etc. Fuel
quality is usually not available. If the fuel flow measurement is based on volume, density and
temperature are required but usually not measured.

Typically, the data aggregation period is 1 to 15 minutes. Automatic data acquisition systems do not
differentiate between operational mode (port, sea, etc.). Data format varies from text/csv (comma
separated value) files to compressed binary formats such as hdf5. The quality depends on sensor
maintenance/accuracy. The same applies for manual reading, but we see that through manual
reporting broken or imprecise sensors are detected and fixed earlier. Draft sensors do not give reliable
results in many cases due to their working principle based on static water head. Here manual readings
or loading computer data is more reliable.

Sensors and data loggers in the maritime environment are complex technology. They require
substantial CAPEX and OPEX. Nowadays also simple solutions are available at a lower cost which
are only connected to an existing NMEA bus. These might lack relevant channels such as power and
consumption.

2.5 Third-party sources

Available data is weather (wind and waves), currents (sea and tidal), water temperature, water depth
and AIS signals. Data is collected from once per minute (AIS close to shore) to once per 6h (weather
models) and comes through APIs, with weather data in complex binary formats. Weather models
deliver good quality hindcast. Close to the coast, substantial deviation might occur due to local
weather effects and tidal currents. AIS is a cheap auto logging of the GPS signal. The quality is very
good. Many weather sources are available free of cost. Converting and interpolating to vessel position
is complex. For AIS data a subscription is required.

3. Requirements for vessel reporting schemes

In the classical commercial setup, a vessel owned by one party (owner) is managed by another
(manager), who is responsible for the maintenance of the ship. The vessel is chartered by the owner to
an operator, who usually pays the fuel. The different parties have different requirements which we
summarize here. The list is by far not complete, but we focus here on what is motivating the reporting
in many cases.

3.1 Requirements from the ship owner

Ship-owners need to have a clear view on the technical performance of their vessel in order to

a) define a competitive C/P description that has a low risk of successful claims,
b) make correct investment decisions e.g. for retrofitting and high-performance coating,
c) benchmark the technical manager,
d) ensure asset protection.

We also see a growing demand for transparency related to vessel efficiency from financial
institutions.

3.2 Requirements from the ship manager

The ship manager works on behalf of the owner and needs to support him with regards to his

164
requirements listed above. He has to educate and support the crew on a day-by-day basis to operate
the vessel in the most efficient way. The focus areas, which require daily monitoring, are speed, trim,
A/E and boiler utilization. Also, long term degradation processes such as propeller, hull and engine
fouling need to be tracked.

3.3 Requirements from the operator

The operator is responsible for the transport of cargo or passengers and the interaction with the
surrounding supply chain. His priority is typically to maintain a schedule or meet a laycan. His
highest cost is often the fuel-cost. Thus, he is closely checking C/P compliance and is requesting the
crew to report consumption, speed and weather. C/P vessel description and guaranteed performance
are anyhow generic (up to Beaufort 5, no matter wind direction and current, based on SOG) thus
getting affected by external factors not depending on the owner/manager. Often another 3rd party such
as weather routing providers are involved to perform post voyage calculations which can then lead to
a claim.

3.4 Requirements for MRV

Vessels (above 5000GT) trading to, from or inside the European Union need to hand in basic voyage,
consumption and cargo data for those voyage legs. Different to commercial C/P reporting, where only
the sea passages are separated and analyzed, EU MRV separates port and under-way consumption.
With standard noon reporting as described above, this is not possible because sea and port
consumptions are mixed. Also, some traditional C/P reporting schemes only require reporting events
at begin/end of sea passage. (Sometimes the BOSP/EOSP are named arrival/departure in those
schemes. MRV in contrary refers to arrival/departure at times were the vessel reaches or leaves the
berth.) These schemes also do not comply with MRV requirements because maneuvering and port
times are not separated. Systems purely relying on automatic data acquisition systems usually do not
have all the required information. Cargo data and precise arrival/departure times need to be reported
manually.

3.5 Requirements for IMO DCS

IMO DCS requirements are similar to EU MRV but data needs to be reported for all voyage legs, not
only those to or from EU ports but without specifying cargo figures.

3.6 Overlap of requirements

On most vessels all requirements described above need to be met. Most reporting solutions do not
meet all requirements. Also, stakeholders tend to insist on their in-house system to be used. Thus,
crews often must report the same data via multiple systems.

4. Data Quality

The accuracy of the analysis based on reported data depends on the quality of the reported data. In the
following, we share our subjective view on some typical issues and how they can be mitigated.

4.1 Sensor quality

Most sensors need maintenance, calibration and regular plausibility checking. In case the crew reports
data manually, sensor failure can be detected by the crew. Reporting software may facilitate plausibil-
ity checking. When automatic data acquisition systems are used, we recommend establishing an
automatic plausibility checking as well. These checks should include missing signals, physical range
limits, and plausibility against baselines. For example, the measured torque and rpm can be checked
against a curve taken from the shop test. Here some tolerance needs to be applied due to the influence
of added resistance and fouling.

165
To identify the faulty sensor, data points that have substantial deviation can be checked against
measured consumption. If RPM and consumption are close to the shop test curve we conclude the
torque measurement inaccurate, if torque and consumption are close to shop test we would conclude
the RPM measurement is wrong. Fig.1 illustrates an example where torque is measured inaccurately
for a certain period.

Fig.1: RPM, power and consumption in relation to baseline data

Even well-maintained sensors have certain uncertainties. For example, draft sensors based on
hydrostatic pressure perform well in port but are biased by hydrodynamic effects at sea.

4.2 Human factor

The operator orders the speed (maybe indirectly by defining an ETA) which has a huge impact on the
consumption. Other measures that influence fuel consumption are in the responsibility of the technical
manager. To ensure that the vessel is managed efficiently, the operator and the manager agree on a
target consumption which is defined in the charter party agreement. In container shipping these are
speed/consumption curves for 2-3 drafts. In tanker and bulker shipping, it is a table with speed, draft
and consumption points. Additional weather limits are defined which are considered good weather .

166
After each voyage the reported consumption is compared with the consumption described in the
charter party agreement. Here only good weather days are considered. If the reported consumption
exceeds the described consumption the operator can claim some money for over-consumed fuel, or, if
the ordered speed was not reached, for the lost time.

Traditionally many ship-to-shore reporting systems are designed and applied to facilitate the process
of managing charter party compliance. In cases were vessel performance is poor due to fouling or
missing maintenance, the crew might be tempted to report more wind than observed or less
consumption than measured to avoid a claim from the operator. Extra consumption can then be
recovered during uncontrolled activities (vessel discharging, tank cleaning, bad weather etc.) If we
use this data to predict hull performance, we will overrate the vessels performance, or underestimate
the fouling.

4.3 Comparison of manually reported and automatically logged data

In Fig.2, reported and auto-logged main engine work is plotted over a period of 10 months. The auto-
logged data has been aggregated to the reported events for better comparability. We see that manual
reported work is significantly higher. Applying both datasets to hull fouling evaluation would result in
differences of more than 20%.

Fig.2: Work, reported vs. auto-logged data aggregated to events

In Fig.3, the work is related to the reported consumption by computing a specific fuel oil consumption
(SFOC). We see the auto-logged data scatters around an average of 205g/kWh which, depending on
fuel quality and environmental condition, is a realistic value. The SFOC based on manual reporting
data is constantly 179 g/kWh in the first 3 months. It is similar to auto-logged data in the next 2
months and, finally, in the remaining 6 months it is constantly 170 g/kWh. We suspect here the crew
did not have access to the torque measurements and has calculated the work based on the measured
fuel consumption. The basis for this calculation has been an assumed SFOC of 179 g/kWh in the first
period. Then, after a crew change, the new chief engineer has calculated with an assumed value of
170 g/kWh.

When looking into speed/power relation for hull performance evaluation, we need to exclude periods
where the vessel is accelerating (speeding up or speeding down). In manual reporting we only see
average rpm and average power. Here acceleration phases cannot be excluded. In pure auto-logged
data, the BOSP and EOSP times are not available. Here we need to check derivatives of e.g. rpm.
Fig.4 shows auto-logged RPM over 4 days in time. Grey dots mark periods out of the sea passage

167
(identified by manual reporting). Green dots mark the periods identified as constant rpm. Blue dots
mark data which has been identified as acceleration phase base on the time derivative of rpm. We
recommend applying such kind of filtering on auto-logged data to reduce scatter on hull performance
evaluation.

Fig 3: Specific fuel oil consumption, reported vs. auto-logged aggregated to events

Fig.4: RPM over time. Grey are periods out of sea passage, blue dots are detected acceleration phases,
green points indicate steady-state

Fig.5: Auto-logged apparent wind plotted over data from hindcast indicating good correlation

168
Specific focus needs to be put on measured and reported wind, because the added resistance due to
wind has a large impact on the power demand. Different reporting systems have different definitions.
Some use apparent wind (typically measured by automatic data acquisition systems), some manual
reporting systems use true wind since it is often reported in the log book. Some use m/s, some knots,
others use Bft. Some systems do not give clear guidance and it changes with the crew. We
recommend comparing data from the vessel with hindcast data to ensure proper transformation. Fig.5
shows such a comparison with what we consider a good correlation.

5. Hull Performance evaluation

Hull and propeller fouling can cause an increase in consumption by more than 30%. Depending on the
paint, the trading and off-hire periods, the degradation rate is very different. Thus, all stakeholders
have the interest to predict and manage hull condition properly.

5.1 Requirements from hull performance evaluation perspective

ISO 19030 gives clear requirements for its standard method. These cannot be met by many vessels.
Thus, many companies make use of part 3 of the ISO standard and apply similar methods based on
what data is available. Below is a list of the minimum requirements based on our experience:

To assess hull performance, we need speed (GPS and/or log), draft/trim, wind (speed and di-
rection), waves, shaft power, water depth, water temperature.
From one snapshot report per day to one reading per hour is suitable. The benefit of higher
frequency is limited.
The vessel should sail at constant speed, RPM and weather should be constant during the
measurement period.
Every structured format is usable.

5.2 Best possible combination of data sources

DNV GL together with Scorpio has developed a new approach combining high-quality auto-logged
data with event reporting, initially introduced for commercial performance evaluation and various
third-party data sources. We use the following data sources for the main input data:

Speed: Here we take speed through water and over ground from the auto-logging system and
validate it against AIS and hindcast current data. Periods where these sources do not agree,
we exclude from the analysis.
Power: Power is taken from auto-logged torque meter and compared with rpm and consump-
tion. When these values are not in line with the expected relation from the shop test we ex-
clude these periods.
Weather: We compute added resistance based on onboard measured data and compare it with
added resistance computed based on hindcast data. If not in line, we exclude.
Draft: Draft sensors have their limitation at sea. Since values do not change much between
events, we take values from event reporting.

By this combination of data sources, and by intelligent filtering, we achieve a very high evaluation
quality. Fig.6 shows calculated hull performance over a period of one and a half years. The kernel
density function illustrates very low scatter.

169
 Fig.6: Hull performance index over 1.5 years

6. Conclusion

All commercial vessels report performance data to various stakeholders. Content and data quality
vary. When it comes to analyzing the data, this needs to be considered. To assess the vessel
performance in many cases additional data needs to be recorded. Ideally, auto-logged or snapshot data
should be collected. To ensure good data quality we recommend to train crews and establish an
immediate feedback loop for data quality. The best results can be achieved by combining available
data sources.

References

BAUR, M.v. (2016) Acquisition and integration of meaningful performance data on board chal-
lenges and experiences, 1st HullPIC Conf., Pavone, pp.230-238, [Link]
[Link]

BERTRAM, V. (2012), Practical Ship Hydrodynamics, Butterworth & Heinemann

BERTRAM, V. (2016), Added Power in Waves Time to Stop Lying (to Ourselves), 1st HullPIC
Conf., Pavone, pp.5-13, [Link]

BERTRAM, V. (2017), Some Heretic Thoughts on ISO 19030, 2nd HullPIC Conf., Ulrichshusen, pp.4-
11, [Link]

BOS, M. (2016), How MetOcean data can improve accuracy and reliability of vessel performance
estimates, pp.106-115, [Link]

DÜCKERT, T.; SCHMODE, D.; TULLBERG, M. (2016), Computing hull & propeller performance:
ship model alternatives and data acquisition methods, 1st HullPIC Conf., Pavone, pp.23-28, http://
[Link]/[Link]

FREUND, M. (2013), Holistic analysis of onboard consumption and efficiency of the energy systems
of ships, PhD thesis, Helmut-Schmidt University, Hamburg, [Link]
hsu/volltexte/2013/3004/pdf/2012_Freund.pdf

JONSSON, S.; FRIDRIKSSON, H. (2016), Continuous estimate of hull and propeller performance

170
using auto-logged data, 1st HullPIC Conf., Pavone, pp.332-338, [Link]
pdf

SCHMODE, D. (2017), Influence of Noise and Bias on the Uncertainty of Data-Based Hull
Performance Prediction, 2nd HullPIC Conf., Ulrichshusen, pp.19-24, [Link]
[Link]

ISO 19030-1:2016 Ships and marine technology Measurements of changes in hull and propeller
performance, International Organization for Standardization, Geneva

171
 Full-Scale Measurement of a Flow Field at Stern using Multi-Layered
Doppler Sonar (MLDS)
Tomoki Wakabayashi, Yasuhiko Inukai, Japan Marine United Corporation, Yokohama/Japan,
{wakabayashi-tomoki, inukai-yasuhiko}@[Link]
Takeshi Yonezawa, Norihide Igarashi, MTI, Tokyo/Japan,
{takeshi_yonezawa, norihide_igarashi}@[Link]
Masahiko Mushiake, Satoshi Kawanami, Furuno Electric, Nishinomiya/Japan,
{[Link],[Link]} @[Link]

Abstract

This paper introduces the ongoing joint research project on full scale measurement of a flow field at
stern with Multi Layered Doppler Sonar (MLDS). Full scale measurement was conducted on a 14,000
TEU container ship with multiple MLDSs and the interim results are shown here. It was confirmed
that the full scale Computational Fluid Dynamics (CFD) agreed well with the measurements in wide
region. And our new plan to apply MLDSs on VLCC is shown as well.

1. Introduction

A flow field around a ship at full scale is different from that at model scale due to so-called scaling
effect even though the geometry of the model is completely scaled. While a model test is usually
performed to know a ship performance and is a well-established technique, the scaling effect is still
not fully understood. A direct calculation with full scale Computational Fluid Dynamics (CFD) is
expected to solve this problem. If the technique of the full scale CFD is established, it is also expected
to enhance the practical applications of an optimization of a hull form, an energy saving device and a
propeller at full scale to reduce GHG emission. However, verification and validation works for full
scale CFD are not enough at this moment. The number of full scale measurements, especially related
to a flow field, is very limited.

In this background, Nippon Yusen Kaisha (NYK), MTI Co, Ltd. (MTI), Furuno Electric Co. Ltd.
(FURUNO) and Japan Marine United Corporation (JMU) started a joint research project on full scale
measurement for 14,000 TEU container ships using Multi-Layered Doppler Sonar (MLDS). MLDS
developed by FURUNO and MTI is an acoustic Doppler sonar capable of measuring relative water
velocity at multiple arbitrary depths along ultra-sonic beams, [Link]
161/upload/[Link]. A full-scale measurement using a MLDS was conducted in 2017 as the world
first application to a flow field at the stern. It was confirmed that the MLDS works well and the
fruitful result was reported at HullPIC 2018. Subsequently, we continue this project and installed
three MLDSs on the sister ship in order to enlarge the measurement area than that in 2017. In this
paper, a quick report of the measurement is presented. And our new plan to apply MLDSs on VLCC
is shown as well.

2. Full scale measurement of a flow field at stern of 14000TEU container ship

2.1. Particulars and measurement location

The subjected ships in the measurement in 2017 and this time have same hull form with particulars
shown in Table 1. Hereafter, we call them Ship-A and Ship-B respectively. Both are built by JMU and
operated by Ocean Network Express Ltd. We plan the measurements on SHIP-B two times. The first
was performed during a voyage from Southampton Port to Suez Canal in February 2019 with 1m
deeper draft than that in the measurement on Ship-A . The second is scheduled in summer 2019.

Table 1: Particulars of 14000TEU container ship

Length overall 364m Depth 29.5m

172
 Breadth 50.6m Summer load draft 15.8m
2.2. Measurement system

Fig.1 shows the configuration of MLDS equipped on SHIP-A and SHIP-B. The commercial Doppler
Sonar, Model: DS-60, FURUNO 2019, developed by FURUNO in 2010 is used in the system. The
detail is described in Inukai (2018). The differences between both ships are shown in Table 2. The
aims of the change from SHIP-A are as follows,

Fig.1: Configuration of system for MLDS equipped on 14,000 TEU container ships

1) Expansion of measurement area

The number of MLDSs increases from one to three (TD1, TD2, TD3) to measure the velocity
in the wider area than SHIP-A. The locations of MLDSs and measuring area on SHIP-B are
shown in Fig.2. The location of the MLDS on SHIP-A was between TD1 and TD2 of SHIP-B.
By using three MLDSs, it is possible not only to enlarge measurement area but also to obtain
three velocity components at the intersection of ultrasonic beams transmitted from three
MLDSs.

Additionally, the number of beam directions in which velocity can be measured increases
from six to twelve as shown in Fig.3. It is realized by rotation of the transducer by 90 in the
Gate Valve. The solid lines show the original directions, i.e. SHIP-A, and the dotted lines
shows directions added for SHIP-B. The number of measurement layers in depth direction is
also increased from nine to eighteen by changing the software.

2) Reduction of the interaction between beams

MLDS transmits beams in three directions (beam 1, 2, 3 or 4, 5, 6) at the same time. It was
found in the measurement on SHIP-A that when one beam hits the propeller or the hull, the
velocities in the other two directions were affected by it. Thus, we limit the number of ultra-
sonic beams transmitted simultaneously from three to one in order to eliminate interaction
between beams.

Table 2: Measurement system equipped to SHIP-A and SHIP-B

SHIP-A SHIP-B
Number of MLDS equipped at stern 1 3
Number of directions ultrasonic beams are transmitted by each transductor 6 12
Number of measurement layers in depth direction of each ultrasonic beam 9 18
Number of ultrasonic beams transmitted at same time 3 1

173
 Fig.2: Measurement range with three MLDSs

Fig.3: Directions ultrasonic beam transmitted to

2.3. Measurement results

Full-scale measurement on SHIP-B was carried out from 8th to 10th February 2019 in the Mediterra-
nean Sea. Fig.4 shows an example of measurement result on 8th February. The figures on left show
measurement result when beam 1 was transmitted forward (corresponding to solid lines shown in
Fig.3) and the figures on right show measured result when the transducer was rotated by 90
(corresponding to dotted line in Fig.3). The measurement results of TD1, TD2 and TD3 are shown in
order from the top. Horizontal axis indicates a distance from the transducer to measuring point and
vertical axis indicates the velocity in the beam direction which is normalized by ship log speed

174
measured by DS-60 equipped at the bow. The marks show measurements and the lines show CFD
calculation. The colors of each line correspond to those in Fig.3.

1) Near field

In the range of within about 6 m from the transducers, the measured velocity agrees well with
the estimation by CFD. Fig.5 shows the contour plot of the difference between the
measurement and the calculation within this range. It can be seen that the difference is less
than 2% of ship speed in the most part. On the other hand, the deviation in the frontal area of
TD1 was a little bit larger. The reason of the deviation that the measured velocity was slower
than the calculation is not clear at this moment. We will investigate the quality of both
measurement and calculation through not only this measurement but also the second
measurement scheduled in summer.

Fig.4: Comparison between measurement and CFD for velocity in direction of each beam

2) Far field

In the far field, the difference between the measurement and the calculation became much
larger. Although we intended to transmit only single beam in order to eliminate the interaction
with the other beams, the interaction obviously remained. For example, the velocity in the
beam 1 of TD1 was disturbed in the region 6m far from the transducer as shown in the upper

175
left figure in Fig.3 despite the fact that the beam couldn t hit any obstacles. It can be thought
to be influenced by the other beams which hit hull or propeller. Because it is difficult to form
a beam in completely one direction, there are beams transmitted to unexpected direction no
matter how weak they are. While the signal level from the reflection against small particles is
attenuated as the distance gets large, one from the reflection against hull or propeller remains
strong. Thus, even if the power of the transmitted ultrasonic beam toward the hull or propeller
is small, the influence of the beam is not negligible. We are trying to find a solution to reduce
the influence by the second measurement.

Fig.5: Contour plot for difference between measured velocity and CFD

Fig.6: Comparison of three components of velocity at intersection

176
 3) Three velocity components

Three velocity components u, v, and w in x, y, and z direction, can be derived at the

intersection of the beams transmitted from the three MLDSs. The intersection is plotted as dot
in Fig.2 and Fig.5. Velocities of beams which pass through the intersection are interpolated
with measured data because it is difficult to locate MLDSs so that the beams cross at one
point. Fig.6 shows the comparison of velocities in three components between the measure-
ment and the calculation. Vertical axis indicates the velocity normalized by ship speed. The
sign of u, v and w are positive respectively forward, toward port and upward. The differences
in all the components are less than 2% of ship speed and the calculation agrees well with the
measurements at the intersection.

2.4. Second measurement

The second measurement on SHIP-B is scheduled in summer 2019. A close investigation will be
made over the quality of measurement by increasing the sampling data. We look for a solution against
the problem that the interaction among beams still remains and will try it in the second measurement.

3. Full scale measurement of a flow field at stern of VLCC

Flow field at stern of blunt ship like VLCC is different from that of slender ship like container ship
because of the existence of strong vortex generated from the bilge part. Fig.6 shows examples of
velocity contours calculated by CFD. The upper and lower figures show flows at model scale and full
scale, and the left and right figures show the flow of blunt ship and slender ship respectively. It can be
seen from the figures that the boundary layer of blunt ship is thicker and the scaling effect is larger
than those of slender ship. Because some physical models are used in CFD calculation, e.g. turbulence
model, the proper model should be set up for accurate calculation. The different flow pattern may
require the different model. Thus, the validation of CFD should be made for as many kinds of ship as
possible.

Fig.6: Difference of flow field at stern between blunt ship and slender ship

177
Accordingly, we decided to expand the application of MLDSs to VLCC as the next challenge.
Classification Society of Nippon Kaiji Kyokai (NK) has joined the current research group and we plan
the measurement in 2020.

4. Conclusion and future work

Full scale flow field measurement with three MLDSs has been carried out on a 14000TEU container
ship. This paper shows just quick report because the measurement is ongoing. Following the report at
HullPIC 2018, a good agreement between the measurement and the calculation were confirmed.
Compared to the last report for the measurement with a MLDS, the velocity in wider area could be
measured by installing three MLDSs. The second measurement is scheduled this summer in order to
accumulate more data and investigate qualities of the measurement and the calculation deeply.

In addition to that, we will carry out a full-scale measurement for VLCC. Class NK joined the project
so as to develop the MLDS as widely applicable measuring system. Our goal is to contribute to
reduction of GHG emission by practical application of full scale CFD to the ship design. More
validation data is necessary to achieve the goal. MLDS has a great potential as a measuring system of
flow field at full scale because it has favorable features of being inexpensive and easy to handle.
Therefore, we continue works to establish the technology of MLDS.

References

INUKAI, Y.; SUDO, Y.; OSAKI, H. ; YANAGIDA, T. (2018), Extensive Full-Scale Measurement on
Propeller Performance of 14000TEU Container Ship, 3rd HullPIC, Redworth

178
 Case Study on Performance Improvement using Advanced Hull
Performance Analysis Methodologies
Iason Stefanatos, DNV GL, Piraeus/Greece, [Link]@[Link]
André Kauffeldt, DNV GL, Potsdam/Germany, [Link]@[Link]
Eirinaios Petoumenos, PPG, Athens/Greece, Petoumenos@[Link]

Abstract

DNV GL and PPG performed an assessment of the hull degradation on various vessels when using
energy saving coating systems based on low friction materials. The scope of this exercise was to use
already existing data in various formats avoiding to perform targeted measurement campaigns and,
thus, allowing the methodology to be replicated for any vessel manager irrespective of the data
collection followed. Working with various levels of data qualities and reporting systems enabled DNV
GL to define the boundaries of the methodology steps including the hull degradation calculations. The
method used by DNV GL and delivered already to more than 2,000 vessels either through Trim
optimization (ECO Assistant), Fleet Performance Management (ECO Insight), or both, is based on ISO
19030 but further expanded to cover all conditions in a more granular way by incorporating the latest
developed techniques (phenomenological modelling combined with data-driven methods and machine
learning techniques). On this project, the methodology was applied for different vessel types, including
cruise vessels, LNG carriers, bulk carriers and crude oil tankers over a combined period of 30 years of
operation and using approximately 165,000 data points, gathered through both manual and automatic
data acquisition systems. This paper presents the methodology and indicative results derived from the
performance analysis of two LNG carriers and one cruise vessel.

1. Introduction and project overview

Energy efficiency is an important factor in the life cycle costs of a ship. With rising fuel oil prices, an
increasingly competitive market and new environmental regulations, like the recent IMO Green House
Gas (GHG) strategy aiming for a 50% reduction of CO2 by 2050 compared to 2008 (MEPC.304(72),
2018), improving energy efficiency is vital. The maritime sector is known for sometimes being reluctant
to embrace change and tends to stick with proven concepts and best practices, which have been
developed and applied for a long time. However, times change and today many shipping companies
invest in energy efficiency improvement, whose importance is steadily growing. It is realized that a
holistic view on the vessel performance is important to continuously lift the remaining potential of
performance improvement.

There are various approaches to improving fuel consumption, and hence CO2 emissions. Bouman et al.
(2017), aggregated results from various publications indicating a wide range of solutions covering hull
design, power and propulsion systems, alternative fuels, alternative energy sources, and operation. Fig.1
depicts the potential improvement that could be achieved through each solution. It is observed that there
are solutions of different nature, which are not available or feasible for all ship types.

Hybrid power propulsion for instance is a good solution for small scale of vessels with specific trade
routes, while it would be an infeasible solution for a large oil tanker. Other solutions can provide
significant improvement of CO2 emission but at a very high cost, which make them non-feasible
options, for the time being, e.g. fuel cells. The low hanging fruits were already harvested through
different stages of energy efficiency improvement measures in the last years, Fig.2:

1st wave: slow steaming

2nd wave: trim optimization and other tools
3rd wave: retrofits now the monitoring and controlling gains importance
4th wave: managing operations and performance management

179
Fig.1: CO2 emission reduction potential from individual measures, classified in 5 main categories of
measures, Bouman et al. (2017)

During the first stages, specific energy efficiency topics were addressed, while now the focus shifts to
ensuring that vessels operate efficiently all the time by addressing the whole spectrum of measures
affecting the vessel performance. To achieve this a good and detailed understanding of the vessel is
important. This is where performance management comes into play.

Fuel consumption evaluations of similar sized vessels during the sea passage show that there are huge
differences between the best performer and the average of the peer group, Fig.3. Depending on the
vessel type the differences may account for up to 36%, Kauffeldt and Hansen (2018).

180
 Fig.2: Steps of energy efficiency improvement Fig.3: Remaining potential of performance im-
measures proving measures: class average (green),
best performer (blue)

An important parameter of the vessel performance is hull degradation as well. The hull protective
coating is a key element that determines hull performance and the levels of potential degradation,
having an impact on the Energy Efficiency Design Index (EEDI) and Energy Efficiency Operation
Index (EEOI) of a vessel.

More specifically, on energy saving coating system solutions, according to studies, silicone-based
coatings decrease hull drag resistance, improving the hydrodynamic performance of a vessel, Schultz
et al. (2010). Silicone-based fouling release coatings provide optimum fouling anti-adhesion due to
hydrophobicity and low surface energy properties, Baier (2015), capable to minimize the adhesion of
marine organisms on coated surfaces, Eduok et al. (2017). Analysis shows skin frictional drag reduction
by 9 22%, in different velocities (6.5 22.7 Knot), when compared to SPC technologies, Mirabedini et
al. (2006), along with high durability, antifouling and anticorrosive effectiveness, Marceux et al. (2014,
Peres et al. (2018).

Investigation on the fuel consumption of tin-free SPC coatings and foul-release coatings illustrated that
the latter can decrease speed-adjusted fuel oil consumption by 22%, Corbett et al. (2011). Overall,
improved vessel operational costs, fuel oil consumption and energy savings are to be expected, as
studies suggest, Selim et al. (2017).

In this article, the results of a joined effort between DNV GL and PPG to assess hull degradation of
various vessel types, and different operational profiles, when using PPG energy saving coating systems
based on low friction materials, are presented.

There are various ways to assess the performance of a coating over-time, e.g. towing tank model tests,
CFD calculations, etc. Further to this, hull performance depends on various factors, such as the
environmental and sailing conditions, vessel operational profile, coating application specifications, hull
pretreatment, etc.

In this study, a methodology developed by DNV GL (described in Section 2) was applied. This is
relying on the technical specifications of the vessel, i.e. design data and operational data. This enables
to efficiently applicate the method to a large number of vessels without the need of costly and time-
consuming installations and infrastructures, e.g. sensors, experimental setup, or physical model
building.

By using already existing data in various formats it was additionally avoided to perform elaborated
measurement campaigns. Thus, the developed methodology can be easily applied to any vessel,
irrespective of the data collection process defined by the managing company. Due to the fact that
various companies utilize different data acquisition methods and reporting systems, data of a wide range

181
of quality were received and used for calculations. Access to these big data source enables DNV GL to
define the boundaries of the methodology steps including the hull degradation calculations and its filter
settings, thus providing more accurate quality analysis.

The project included studies of four series of different sister vessel groups; cruise vessels, LNG carriers,
bulk carriers, and crude oil tankers. The owning and managing companies of the vessels are located in
different geographical regions while the vessels trade globally, under different operational profiles and
conditions. Section 3 describes the detailed specifications of this project along with the key results.

2. Methodology

From the point of view, performance management covers a wide range of aspects and
considers the ship as a holistic system. This approach can become very complex and involve monitoring

approach. However, most performance analyses consider the following main aspects:

Efficient main engine operation (power, fuel oil consumption (FOC), RPM, etc.)
Efficient management of auxiliary engines (combination, power, FOC, etc.)
Controlling of other auxiliaries (boilers, separators, etc.)
Hull and propeller performance considering:
Resistance
Weather
Propulsion
Trim
Hull degradation
Voyage management (Optimized routing, consumptions, weather, speed, legs)
Bunker statistics
Port/supplier rating
Emissions and disposals
Maintenance and survey interval optimization

This list is not exhaustive and shall only demonstrate the complexity of the topic. The paper focuses on
hydrodynamic performance, considering resistance, propulsion, hull degradation as well as propeller
effects.

Vessel performance assessment cannot provide safe conclusions by simply plotting the measured power
demand during operation over time. The scatter observed in such plots is dominated by speed and draft
variations in operation, which have a much bigger impact on the power demand than the rather small
effect of hull degradation. Hence, to identify trends in hull degradation, it is necessary to normalize the
power measurements to eliminate the effects of different operating conditions regarding speed, draft
and trim. Basis of such a normalization could be the ideal power demand determined for the same
condition. It can be used as a reference, representing the theoretical power demand of the vessel at this
condition. Measured power and reference power demand together allow conclusions about the hull
performance and, therefore, the hull degradation.

Simply put, looking at the impact of the hull, a performance analysis compares a measured power
demand for a certain operating condition to a reference value valid for that condition.

Here, the relevant condition is mainly defined by the following parameters:

Operating conditions:
Draft
Trim
Vessel speed

182
 Environmental conditions:
Wind (force and direction)
Sea state (height and direction)
Swell (height and direction)
Temperature (air and water)
Current (speed and direction)
Water depth

By comparing the measured with a reference power of a comparable condition, a performance index
can be deduced, which enables operators to easily infer conclusions about the vessel performance.

Many factors influencing performance make it difficult to distinguish effects and hence understand the
parameters that drive performance. Therefore, for a meaningful and accurate performance analysis it is
crucial to have a matching reference condition for all measured operating conditions. Hence,
comprehensive baselines covering the entire operating range regarding draft, trim and speed are
prerequisite to get the most out of the performance analysis. Or the other way around, the further an
existing reference condition must be extrapolated to meet a measured operating condition, the less
accurate the result of the performance analysis would be.

2.1 The right baseline for the right ship

For assessing ship performance, speed-power curves of the individual vessels serve as a reference. They
reflect the calm water power demand of the corresponding vessels under ideal conditions:

Clean hull
Deep water
No wind
No waves

There are several well-known ways to determine speed-power curves for a vessel, which can serve as
reference in a performance analysis. As summarized in Fig.4, each approach has its own advantages
and disadvantages:

Empirics:
Based on generalized equations this approach is not intended to fit a specific vessel. Neglecting
the individual characteristics of a vessel, this approach delivers a general idea of the magnitude
of the power demand only.
Pro: Very quick and therefore cheap
Contra: Inaccurate and not suited for ship specific comparison due to neglecting of
individual vessel characteristics
Model scale measurements:
A well-established approach, which is based on measurements in model scale. Since these
measurements are conducted under laboratory conditions, the accuracy of the model scale re-
sults is generally high.
Pro: The widely accepted approach with its accurate results is well suited for the
comparison of different hull shapes.
Contra: Building of a scale model, preparation and execution of each test is time
consuming and expensive. Due to uncertainties when extrapolating to full-scale, the
results are not fully consistent across different test facilities. Possible scaling effects
are not considered.
Full-scale measurements:
General accepted approach to validate the performance predictions for a newly delivered ship.
With additionally installed measuring equipment the power results are very accurate. However,

183
 the measurement quality strongly depends from the environmental conditions. It is not the first
choice for comparative measurements.
Pro: Accurate power results without scale effects.
Contra: Often only conducted at ballast draft and under non-ideal conditions. The
correction of the environmental conditions as well as the extrapolation to the design
draft lead to uncertainties. Depending on the used correction and extrapolation methods
the results are not fully consistent across different test facilities or yards.

Fig.2: Methods of power demand analysis Fig.3: Differences in power prediction

Computational Fluid Dynamics (CFD) simulations:

CFD has become a well-established tool for relative comparison of different designs, configu-
rations and operating conditions. The absolute accuracy still depends on the CFD setup and
available resources.
Pro: A fast and effective approach to conduct full-scale simulations and, thus, become
independent of scaling effects. Well suited tool for comparative studies.
Contra: There are still substantial computing capacities required to conduct large-scale
studies efficiently. The absolute accuracy depends on the CFD setup and the experience
of the user.

Probably none of the common methods is perfectly suited to meet all requirements of the maritime
industry. Fog.5 visualizes the mentioned inconsistencies of absolute power predictions, which not only
vary between model test and CFD but also between different model test facilities. In some cases,
differences of up to 10% are reported between different model test facilities using the same model.

When generating baselines, it is indispensable to cover all relevant operating conditions of a vessel
regarding draft, speed and trim to get reliable performance information. More precisely, the operational
profile must be closely covered by the reference simulations to minimize the uncertainties due to
extrapolation from the available reference to the recorded operating condition. This means that a large
amount of reference conditions is required to produce accurate power baselines.

Figs.6 and 7 show a typical operating condition of a large container vessel (red cross) relative to the
reference points based on different data sets. Fig.4 presents an example with a dense grid of reference
points based on comprehensive CFD simulations. A matrix of 7 trims, 11 speeds and 8 drafts results in
616 reference points covering the operational profile of the vessel. Fig.5, on the other hand, shows a
reduced data set based on a standard model test of the same vessel. Here, the matrix contains only 3
drafts (scantling, design and ballast), 1 trim and 6 speeds resulting in 18 reference points
inhomogeneously distributed over part of the operating range. The example with the reduced data set
shows impressively the distance between the measured operating condition and reference points.

184
 Fig.4: Interpolation of a measured conditions using a dense grid of CFD-based reference points

Fig.5: Extrapolation to a measured condition using a reduced data base referring to a standard model
test with ballast, design and scantling draft available

For container vessels with their extremely large operating range far from design condition regarding
draft, trim and speed, Fig.8, the use of CFD approaches is strongly recommended. Today, standardized
CFD setups ensure high comparability. The application of state-of-the-art fully viscous, free surface
CFD approaches using Volume of Fluid (VoF) and Reynolds averaged Navier Stokes equations
(RANSE) in combination with standardized prediction methods as well as full-scale computations (no
scale effects), provide reliable results. Beside the significant technical enhancement, the constantly
growing computing power and the resulting reduction in lead time and cost means that large numbers
of simulations can be conducted efficiently a crucial factor when developing comprehensive baselines
for performance analysis of container vessels.

However, the operating profile of many vessel types is much less extensive and easier to predict than
that of container ships. Bulk carrier and tanker for example, mainly operate at ballast or laden condition
reducing the draft-trim combinations significantly.

Cruise vessels even operate at one draft only. In addition, the speed range of these vessel types is
significantly reduced compared to that of container vessels. Thus, it is not necessarily required to use
the same comprehensive CFD baselines for all types of vessels. Often, it might be more efficient to use
existing model test and sea trial results to create appropriate baselines for certain vessel types.

185
 Fig.6: Typical operating profile of a large container vessel far from design condition

2.2 Calculation of the Hull Performance Index

The hull performance index is based on:

(1)

The reference power reflects the power demand under ideal conditions, whereas the corrected measured
power is based on the measured power corrected for the influence of the weather. The weather
correction is mostly based on empirics and often includes the following corrections:

wind
sea state
swell
water temperature
current

Using Eq.(1) to determine performance means, that the vessel performs at 100%, if the corrected
measured power is equal to the reference power. If the corrected measured power is twice as large as
the reference power, the vessel only performs at 50%. Hence, using this performance index it is
expected that the performance is ideally close to 100%.

Plotting the calculated HPIs of all reported events over time results in a hull degradation trend, which
is the key parameter for deciding whether the next hull cleaning is advised. It may also provide insight
on the performance of various hull coatings and may support the decision process for the best suited
product depending on the characteristics of the operating area and profile. Depending on that, the
corresponding decisions can vary significantly.

In general, the HPI plots tend to scatter more or less strongly. Besides the data base quality, the scatter
could also be the result of uncertainties with respect to long-term measurements during normal
operation. For example, if the measuring equipment is not calibrated on a regular basis, the accuracy of
the reported measurements decreases with time. In addition, reporting environmental conditions is
obviously still a challenge for many crews causing uncertainties in the weather correction.

2.3 Corrections and filter settings as main differentiator

The ISO 19030 is intended to provide a methodology for assessing changes in hull performance. It is
organized in three parts, where part 2 describes the default method, which stipulates high-frequency
data capturing. Changes in hull and propeller performance are quantified in terms of KPIs, which are

186
designed to evaluate the effect on performance over time of different maintenance, repair and retrofit
activities. As stated in ISO19030
performance in order to, for instance, compare different ship types, but only to compare performance
changes over time for a given ship.

The weather corrections in ISO19030 are relatively simple. There are no explicit corrections for wind
speed and swell provided. Instead, the default filter recommendations strictly limit the wind speed,
which is corrected based on generic wind corrections with wind force coefficients taken from open
literature, potentially leading to increased inaccuracy of the attained results.

The DNV GL method follows the default method in ISO 19030 Part 2 up to an extent, but deviates in
other elements using the leeway given in Part 3. The main differences between DNV GL approach and
ISO 19030 methodology can be summarized as follows (details can be found in Schmode et al. (2018):

No certain data frequency is required in order to use all available data, irrespective of the data
collection process, but always applying data quality assessment criteria.
A correction for sea water temperatures is applied to reduce scatter and increase accuracy.
Vessel group specific weather corrections make the performance index less weather dependent.
Smaller scatter equates better accuracy.
The application of smart wind filters leads to many more surviving data sets without global
loss of accuracy.
The speed dependency in the performance indicator is reduced to give more realistic and less
speed dependent values.
A machine learning correction is applied to account for effects not modelled.

On top of above, AIS data are used as well

conditions.

Apart from the calculation of baselines based on design data, another set of baselines was calculated
based on reported data from the first year of the studied period, as per ISO 19030. The design baselines
were modified to take the actual performance of the first year into account.

Finally, it should be noted, that the results were normalized based on reported data from the first year
of operation, i.e. first year of operation is equal to 100% performance.

3. Case studies and indicative results

The methodology described in Section 2 was applied for various vessel types, including cruise ships,
LNG carriers, bulk carriers and crude oil tankers. A combined period of over 30 years of operation and
approximately 165,000 data points, gathered through both manual and automatic data acquisition
systems. In this paper, the key results of 2 LNG carriers and 1 cruise vessel are described. For the
development of the baselines, data from the sea trials included as presented in Table 1.

Table 1: Sea trial data used for model development

LNG Carriers Cruise vessel
Draft Two drafts One draft
Since all vessels are mainly operated on the Introduction of a draft band according to
provided drafts, no major disadvantages ISO 19030 in which the power extrapola-
due to the interpolation are to be expected tion is allowed. Valid draft band range:
here 7.75-8.57m (± 5% displacement change)

Speed A large speed range is provided (13-22 kn) A limited speed range (22.5kn -25kn)
Range minimizing the need to extrapolate the Extensive extrapolation of lower speeds
results. was required.

187
 Trim No trim information. No trim information.
Vessels operated near level trim and
therefore the missing trim consideration is
acceptable

AIS data were used to identify the trade areas and operational profile of each vessel, as shown in Table
2, Figs.9-11. Key parameters that may affect the hull performance significantly are the sailing speed,
the time spent in anchorage, and the idle time. For this reason, AIS data were used to identify the
average sailing speed, the time spent annually in anchorage (speed less than 8kn) as well as idle time
for more than 15 consecutive days, Table 2.

Fig.9: Trade route of LNG carrier Fig.10: Trade route of LNG carrier 2

Table 2: Operating profile of vessels: trade areas, anchorage, idling time

Vessel LNG Carrier 1 LNG Carrier 2 Cruise Vessel

Winter trade areas Global trade Global trade New Zealand

Australia
Indonesia
Summer trade areas Global trade Global trade Canada
USA
Avg. sailing speed 14.6 kn 15.1 kn 16.1 kn
Anchorage yearly 15.8% 11.4% 33.1 %
Idling time 2 times: 1 time: N/A
(<15 consecutive days) S.E. Asia: 17 days S.E. Asia: 15 days
S.E. Asia: 22 days

Fig.11: Trade route of Cruise Vessel

188
The analysis of the LNG carriers operation demonstrated good hull performance over the studied
period, considering the sailing and idling conditions. More specifically, LNG Carrier 1 Hull
Performance Index (HPI) degraded on an average annual rate of 3.7%, Fig.12, while the average speed
drop was calculated at 1.2%, Fig.13. LNG Carrier 2 demonstrated an average annual HPI rate of -4.0%,
Fig.14, and an average annual speed drop of 1.2%, Fig.15.

Table 3 summarizes the operational conditions and hull performance results. The negligible perfor-
mance deviation between the two case studies, could be explained by different average speed,
anchorage time and idling for more than 15 days.

Table 3: Overview of sailing pattern and hull performance versus 1st year performance
Avg. Sailing Avg. Sailing Anchorage Annual HPI Speed
Speed, [kn] Draft, [m] per year, rate, [%] reduction,
[%] [%]

LNG Carrier 1 14.6 10.4 15.8% -3.7% 1.2%

LNG Carrier 2 15.1 10.7 11.4% -4.0% 1.2%

Fig.7: LNG Carrier 1 Hull Performance Indicator (HPI) vs time

Fig.8: LNG Carrier 1 Speed deviation vs time

189
 Fig.9: LNG Carrier 2 Hull Performance Indicator (HPI) vs time

Fig.10: LNG Carrier 2 Speed deviation vs time

For the cruise vessel, the analysis was performed using the vessel ideal/design (based on sea trials)
conditions as reference; a speed and power improvement were observed that may be attributed to
improved performance of the new applied coating. More specifically, the Cruise Vessel demonstrated
a speed increase of 0.5% and HPI improvement of 3.2% compared to the sea trials.

Fig.11: Cruise vessel Hull Performance Indicator vs. time (Initial improvement against the sea-trials)

190
Fig.12 Cruise vessel Speed deviation vs. time (Initial improvement against the sea-trials)

Table 4: Overview of sailing pattern and hull performance versus vessel design data
Avg. Avg. Anchorage Speed Speed HPI vs Hull
Sailing Sailing per year deviation reduction sea Perf.
Speed Draft vs sea trials
trials
Cruise 16.1 kn 8.5 m 33% +0.5% 0.5% +3.2% 98.8%
vessel

4. Conclusions

In this article, the joined project between DNV GL and PPG on assessing the hull degradation of vessels
was discussed. The advanced hull performance analysis methodology developed by DNV GL, was
applied on various ship types of different sister vessel groups. Indicative results were presented,
referring to a specific coating system based on low friction materials, developed by PPG.

Analysis was based in different reporting systems. Design data, e.g. sea trials, but also actual
operational data were used to create the baselines and reference periods of the case studies. In addition,
AIS data were utilized, in order to gain a holistic view of each vessel operational profile and conditions.
Indicative results were presented; 2 LNG carriers with constant hull performance over the studied
period. Furthermore, 1 cruise vessel showing speed and power improvement compared to the sea trials,
which could also be attributed to the improved performance of the new applied coating.

References

BAIER, R.E. (2015), Correlations of Materials Surface Properties with Biological Responses, J. Sur-
face Engineered Materials and Advanced Technology 5, pp.42-51

BENDICK J.A.; SCHULTZ M.P.; HOLM E.R.; HERTEL W.M. (2010), Economic analysis of under-
water hull coatings and maintenance, NSWCCD-63-TR-2010/17, NSWC-CD, Bethesda (MD)

BOUMAN, E.A.; LINDSTAD, E; RIALLAND, A.I.; STRØMMAN, A.H (2017), State-of-the-art

technologies, measures, and potential for reducing GHG emissions from shipping A review,
Transportation Research Part D: Transport and Environment 52/A, pp.408-421

191
CORBETT, J.J.; WINEBRAKE, J.J.; COMER, B.; GREEN, E. (2011), Energy and GHG Emissions
Savings Analysis of Fluoropolymer Foul Release Hull Coating, Energy & Environmental Res. Assoc.

EDUOK, U.; FAYE, O.; SZPUNAR, J. (2017), Recent developments and applications of protective
silicone coatings: A review of PDMS functional materials, Progress in Organic Coatings 111, pp.124-
163

KAUFFELDT, A.; HANSEN, H., Enhanced Performance Analysis and Benchmarking with CFD
Baselines, 3rd HullPIC, Redworth

MARCEAUX, S.; BRESSY, C.; PERRIN, F.-X.; MARTIN, C.; MARGAILLAN, A. (2014), Develop-
ment of polyorganosilazane silicone marine coatingsSandra, Progress in Organic Coatings 77,
pp.1919-1928

MEPC.304(72), Initial IMO Strategy on Reduction of GHG Emissions From Ships, 72nd MEPC, IMO

MIRABEDINI, S.M.; PAZOKI, S.; ESFANDEH, M.; MOHSENI, M.; AKBARI, Z. (2006),
Comparison of drag characteristics of self-polishing co-polymers and silicone foul release coatings: A
study of wettability and surface roughness, Progress in Organic Coatings 57 421-429

PERES, R.S.; ZMOZINSKI, A.V.; BRUST, F.R.; MACEDO, A.J.; ARMELIN, E.; ALEMÁN, C.;
FERREIRA, C.A. (2018), Multifunctional coatings based on silicone matrix and propolis extract,
Progress in Organic Coatings 123, pp.223-231

SCHMODE, D.; HYMPENDAHL, O.; GUNDERMANN, D. (2018), Hull Performance Prediction be-
yond ISO 19030; 3rd HullPIC, Redworth

SCHULTZ, M.P.; BENDICK, J.A.; HOLM, E.R.; HERTEL, W.M. (2011), Economic impact of bio-
fouling on a naval surface ship, Biofouling 27 1, pp.87-98

SELIM, M.S.; SHENASHEN, M.A.; EL-SAFTY, S.A.; HIGAZY, S.A.; SELIM, M.M.; ISAGO, H.;
ELMARAKBI, A. (2017), Recent progress in marine foul-release polymeric nanocomposite coatings,
Progress in Materials Science 87, pp.1-32

192
 Application of an ISO 15016:2015 based Method
Performance
Beom Jin Park, KRISO, Daejeon/Republic of Korea, bjpark@[Link]
Myung-Soo Shin, KRISO, Daejeon/Republic of Korea, msshin@[Link]
Gyeong Joong Lee, KRISO, Daejeon/Republic of Korea, gjlee@[Link]
Min Suk Ki, KRISO, Daejeon/Republic of Korea, mski@[Link]

Abstract

ISO 19030 was developed in order to prescribe practical methods for measuring changes in ship
specific hull and propeller performance. ISO 19030 was expected to give a set of relevant
performance indicators for hull and propeller maintenance, repair and retrofit activities. However, as
presented in previous HullPIC conferences, results of ISO 19030 is not yet fully verified to give
reliable performance indicators. On the other hand, ISO 15016 was developed to provide a set of
standard methods to analyse results of speed trials of newly built ships with the accuracy of 2% in
shaft power and 0.1 knots in speed, if all requirements are met. However, as ISO 15016 is intended to
be used for sea trials, which are run on somewhat controlled environment with as low environmental
forces as possible, some part of ISO 15
proposes a new method based on ISO 15016:2015 and presents results of applying to operational
data of bulk and container carriers and identifies benefits over ISO 19030 in ship performance.

1. Introduction

ISO 19030 standard was developed to prescribe practical methods for measuring changes in ship
specific hull and propeller performance. The emphasis was laid on practical methods and many
analysis techniques, such as calculating added resistance due to waves, was not included in the
standard since there was no practical means to collect data necessary to apply such techniques at the
time. Filtering and averaging were introduced instead to deal with environmental effects that was not
accounted for in the standard. The results were
with limitation that it can only be used relatively to compare ship specific changes and unsuitable to
be used to compares different ships, with rather coarse time resolution. (Averaging period should be at
least 3 month or more and typically a year).

journey and often day by day to retain their competitiveness to survive in ever challenging business
environment s averaged
too coarse to be used in decision-making process.

This paper tries to identify possible alternative

practicality of the analysis method. ISO 15016:2015 is the new amended standard of ship speed trial
analysis method. Its accuracy objective of 2% in shaft power and 0.1knots in speed was verified from
many sources during its development process as it is used in contracts and mandatory regulations such
as the reference speed in EEDI calculations. ISO 15016:2015 was also considered during ISO 19030
development and many parts in ISO 19030, such as wind correction, was directly referenced. In this
paper, a new analysis method based on ISO 15016:2015, but retaining practical applicability as in ISO
19030 is proposed and application results to ship operation data is discussed.

2. Analysis methods

2.1. Overall process

The overall process of the new proposed method is basically the same as ISO 19030 and ISO
15016:2015. First, outliers and unwanted data, such as when turning direction or in shallow water is

193
removed by filtering. Then resistance increase due to environmental forces are estimated by the same
method used in ISO 15016:2015. These resistance increases are then used to correct power using
direct power method used in ISO 15016:2015. For comparing purposes, the analysis results are further
corrected to standard displacement.

2.2. Filtering

In ISO 19030, filtering is used to remove outliers and only leave the part of data when the ship is
steadily cruising. However, as noted in Park et al. (2017), filtering in ISO 19030 have unwanted
effects. In ISO 15016:2015, no filtering on data is applied as speed trial is only conducted in very
calm environment and very few measurements are taken.

The proposed method uses only very simple filters to avoid any unwanted effects:
- Remove too low or high speed. Usually the speed range in model test is used as reference and
any speed outside the range is discarded.
- Remove when rudder angle is larger than 5° to remove when changing direction.
- Remove when the ship is operating in shallow water.

2.3. Estimating resistance increase

As in ISO 15016:2015, resistance increase due to wind, wave and difference in water density in
estimated with the same method used in ISO 15016:2015. For wind resistance, ISO 19030 and ISO
15016:2015 basically uses the same method. For wave resistance, it was not included in ISO 19030 as
no practical means to measure wave parameters are not available. However, as noted in Park et al.
(2018) there are publicly available wave data for sources such as National Oceanic and Atmospheric
Administration (NOAA). ISO 15016:2015 includes STAWAVE II and theoretical method for wave
resistance calculation, but since theoretical method requires more detailed ship geometry data and
requires more time for calculation, STAWAVE II, even with its limitation of only calculating waves

Resistance increase due to difference in water density can be calculated from water temperature,
which is also available from weather services even if the ship is not equipped to record the data.

2.4. Power correction

When all resistance increases are estimated, they are used in correcting the measured power value.
The corrected delivered in ideal condition is calculated using Eqs.(1) and (2).

( 1 )

is the required correction for power in watts;

is the total resistance increase in newtons;

is the measured delivered power in the operating condition;

is the propulsive efficiency coefficient in the operating condition;
is the propulsive efficiency coefficient in the ideal condition.

Then, the corrected power is calculated by:

( 2 )

is the corrected delivered power in the ideal condition.

194
Unlike ISO 15016:2015, current correction is not included in the new method. Current correction
method in ISO 15016:2015 assumes that the measurements are taken in the same geographical
location and therefore current can be expressed as periodic function dependent on time. However, for

performance is changing as well. How to account for the current is left as future work at this time.

2.5. Displacement correction

Each leg and journey of the operating ship has different displacements and in order to compare
analysis results, the difference in displacements has to be accounted for. In the proposed method, this
is achieved by first defining standard displacements for typical loading conditions such as laden and
ballast for bulk carriers and 80% or 90% displacements for container carriers and then using
displacement correction method in ISO 15016:2015 to correct power to standard displacements. After
displacement correction, analysis results can be compared to each other and even with model test or
sea trial results, if such data is available for the same loading conditions.

3. Data collection

During ISO 19030 development, particular care was given so that only practically available data is
used in the standard. The proposed method uses the same basic principles. All data is typically
available in most ships. The data used is the proposed method is summarised in Table 1.

Table 1: List of operational data used in the proposed method

Data item Mainly used in Typical source
Speed over ground Wind resistance calculation Onboard measurements
Speed through water Power correction Onboard measurements
Wave resistance calculation
GPS heading Wind resistance calculation Onboard measurements
Gyro heading Wave resistance calculation Onboard measurements
Rudder angle Filtering (changing direction) Onboard measurements
Water depth Filtering (shallow water) Onboard measurements
Weather service data
Shaft power Power correction Onboard measurements
Wind speed Wind resistance calculation Onboard measurements
Weather service data
Wind direction Wind resistance calculation Onboard measurements
Weather service data
Air temperature Wind resistance calculation Onboard measurements
Weather service data
Sea wave height Wave resistance calculation Weather service data
Sea wave period Wave resistance calculation Weather service data
Sea wave direction Wave resistance calculation Weather service data
Swell height Wave resistance calculation Weather service data
Swell period Wave resistance calculation Weather service data
Swell direction Wave resistance calculation Weather service data
Sea water temperature Wave resistance calculation Weather service data
Water density resistance calculation
Static draught Displacement correction Onboard measurements
Port inspection

4. Analysis results

The proposed method is applied to two ships one 176K bulk carrier and one 8600TEU container
carrier. Figs.1 to 4 shows typical analysis results of the bulk carrier and shows those of the container

195
carrier. Unlike ISO 19030, the results are expressed in a point cloud. Its dispersion trends and slope
show much more information of the performance of the ship than one averaged value of ISO 19030.
In all figures, after power correction, the point cloud resembles a line similar to the reference speed
power graph obtained from model test. Basically, the difference in the performance from the reference
condition, similar to the Performance Indicator in ISO 19030 can be identified by the location of the
point cloud in reference to solid line, which represents model test results. But unlike ISO 19030, this
analysis can be done for each leg of a journey or combined data of several legs or journeys, as long as
the loading condition is the same.

14.000

12.000
Delivered Power (kW)

10.000

8.000

6.000

4.000

2.000

0
10 11 12 13 14 15 16
Speed through Water (knots)

Measured Power Corrected Power Model test

Fig.1: Typical analysis results of bulk carrier in ballast condition

16.000

14.000

12.000
Delivered Power (kW)

10.000

8.000

6.000

4.000

2.000

0
9 10 11 12 13 14 15
Speed through Water (knots)

Measured Power Corrected Power Model test

Fig.2: Typical analysis results of bulk carrier in laden condition

196
 30.000

25.000

Delivered Power (kW)

20.000

15.000

10.000

5.000

0
12 14 16 18 20 22
Speed through Water (knots)

Measured Power Corrected Power Model test

Fig.3: Typical analysis results of container carrier in 80% loading condition

30.000

25.000
Delivered Power (kW)

20.000

15.000

10.000

5.000

0
12 14 16 18 20 22
Speed through Water (knots)

Measured Power Corrected Power Model test

Fig.4: Typical analysis results of container carrier in 90% loading condition

Fig.5 shows analysis results of 4 journeys with same departure and destination of the bulk carrier in
ballast condition. The legend shows the dates when each journey is made. There is a noticeable
difference in the performance of the bulk carrier between each journey and journey made in February,
2017 have distinctively better performance than the journey made in July of the same year. Fig.6
shows the same results of container carrier in 90% loading conditions. Unlike the bulk carrier, the
container carrier does not show distinctive difference in performance between April and October,
2018.

197
 14.000

12.000

Delivered Power (kW)

10.000

8.000

6.000

4.000

2.000

0
10 11 12 13 14 15 16
Speed through Water (knots)

2017-2 2017-4 2017-5 2017-7 Model test

Fig.5: Comparison of analysis results of bulk carrier in ballast condition

30.000

25.000
Delivered Power (kW)

20.000

15.000

10.000

5.000

0
12 14 16 18 20 22
Speed through Water (knots)

2018-4 2018-7 2018-10 Model test

Fig.6: Comparison of analysis results of container carrier in 90% loading condition

In ISO 19030, in order to analyse change in performance before and after a dry docking or a
maintenance event, at least 6 months of data (3 months before and 3 months after) is required.
However, in the proposed method, it can be analysed even with only one journey before and after dry
docking as shown in Fig.7 to Fig.9. Fig.7 shows analysis results of one journey before and after dry
docking in laden condition of the bulk carrier and Fig.8 in ballast condition. Fig.9 shows analysis
results of one journey before and after dry docking in 80% loading condition of the container carriers.
They all show noticeable difference from dry docking and may even describe the effects dry docking
more accurately as the time period is very short, within two months for total period.

198
 16.000

14.000

12.000

Delivered Power (kW)

10.000

8.000

6.000

4.000

2.000

0
9 10 11 12 13 14 15
Speed through Water (knots)

Before dry docking After dry docking Model test

Fig.7: Comparison of before and after dry-docking performance of bulk carrier in laden condition

16.000

14.000

12.000
Delivered Power (kW)

10.000

8.000

6.000

4.000

2.000

0
10 11 12 13 14 15 16
Speed through Water (knots)

Before dry docking After dry docking Model test

Fig.8: Comparison of before and after dry-docking performance of bulk carrier in ballast condition

5. Conclusions

This paper proposes a new method based on ISO 15016:2015 that can give more detailed analysis
results on the changes in the performance of a ship than ISO 19030. Specifically, the proposed
method does not need averaging as it takes account of major environmental effects and therefore can
analyse the performance of a ship with data collected in shorter period of time than ISO 19030 and
provide more insights on the performance the ship as it gives more than a single averaged value.

199
 30.000

25.000

Delivered Power (kW)

20.000

15.000

10.000

5.000

0
12 14 16 18 20 22
Speed through Water (knots)

Before dry docking After dry docking Model test

Fig.9: Comparison of before and after dry-docking performance of container carrier in 80% load

While the proposed method shows good promises, more validation is required to many different types
of ships. Also, one major environmental effect that is not accounted in the proposed method still
remains currents. A reliable method to estimate resistance increase will improve the accuracy of the
proposed method further.

Acknowledgements

This research was sponsored by the Ministry of Trade, Industry & Energy (Korea Government) under
the project -ship correlation based on the ship performance
measurement (PNS3370 and -power analysis technology to attain
high energy efficiency and GHG reduction of operating ships under real sea condition .

References

PARK, B.J.; et al (2017), Experience in Applying ISO 19030 to Field Data, HullPIC 2017, Ulrichs-
husen, pp.150-155

PARK, B.J.; et al (2018), Ideas on How to Improve ISO 19030 based on the Results of Applying to
Field Data, HullPIC 2018, Redworth, pp.178-184

ISO 15016:2015 (2015), Ships and marine technology Guidelines for the assessment of speed and
power performance by analysis of speed trial data, Int. Organization for Standardization

ISO 19030-1 (2016), Ships and marine technology Measurements of changes in hull and propeller
performance Part 1: General Principles, Int. Organization for Standardization

ISO 19030-2 (2016), Ships and marine technology Measurements of changes in hull and propeller
performance Part 2: Default method, Int. Organization for Standardization

ISO 19030-3 (2016), Ships and marine technology Measurements of changes in hull and propeller
performance Part 3: Alternative methods, Int. Organization for Standardization

200
 Full-Scale Performance Measurement and Analysis of the Silverstream
Air Lubrication System
Luke De Freitas, Silverstream Technologies, London/UK, [Link]@[Link]
Noah Silberschmidt, Silverstream Technologies, London/UK,
[Link]@[Link]
Takis Pappas, Silverstream Technologies, London/UK, [Link]@[Link]
David Connolly, Silverstream Technologies, London/UK, [Link]@[Link]

Abstract

In this paper, Silverstream Technologies will discuss and present their experience in executing
proprietary Air Lubrication System; the
Silverstream® System. Experiences discussed are drawn from execution of commercial installations on
a chemical tanker and two cruise ships. The intent is not to present the mathematical nuisances of trials
data analytics but to provide an informative script as to the challenges encountered with on-board data
collection and the reality that one defined method is not fit for evaluating performance of Energy
Efficiency Technologies (EETs). The discussions encompass the accuracy of measurement instruments,
methodologies applied in determining in-service performance, applicability of ISO 19030 and
uncertainties that changing environmental factors introduce in full-scale performance monitoring. The
experiences fall against the backdrop of a unique hydrodynamic EET that has the capability of being
switched ON and OFF greatly simplifying the baseline against which performance is measured.

1. Nomenclature

The following words and/or expressions used in this paper shall have the meaning as defined below:

ARU Air Release Unit(s)

ALS Air Lubrication System
BF Beaufort Wind Scale
CAA Air Resistance Coefficient
EET Energy Efficiency Technology
Gross Total reduction in power
HSVA Hamburgische Schiffbau-Versuchsanstalt (Hamburg Ship Model Basin)
hh:mm:ss Timestamp in hours, minutes and seconds
kn Values given in Knots
kW Values given in Kilowatts
LR Lloyds Register
Net Total reduction in power after deduction of compressor power
RPM Revolutions Per Minute
Run Consecutive block of ON and OFF data
SOG Speed Over Ground
STW Speed Through Water

2. Introduction

frictional resistance is the most dominant. Injection of air into the boundary layer (between the
stationary and moving water) can reduce the frictional resistance of the hull, Liem et al. (2013). The
concept of using air lubrication has been considered a viable method for reducing frictional resistance
of ocean-going vessels for decades with the first patent; US5644 A filed in 1848 for increasing vessel
speed, Stevens and Stevens (1848). Many methods have been proposed and tested since then, but no
single method of achieving the desired net energy reduction has been readily adopted by the industry.

201
Silverstream have spent the last decade perfecting the Silverstream® System which has been based on
full-scale testing at research facilities in combination with full-scale testing on three shipboard
installations deployed over the last 5 years. Full-scale testing is essential for proving air lubrication as
accurate scaling of microbubbles and air layers from full to model-scale is practically impossible due
to the widely known Reynolds number scaling problem, Molland et al. (2017). Consequently, caution
is required when inferring full-scale performance from the basis of model tests only. As a result,
Silverstream Technologies have adopted a split development programme encompassing full-scale
testing at HSVA using their cavitation tunnel to experiment and demonstrate the formation of the
he experimentation work allows for measurement
of the reduction of frictional resistance on a plate immediately aft of the ARU. This translates to
reduction in hull surface frictional resistance requiring full-scale trialing to capture the coverage
achieved and adherence of the microbubble carpet formed by the ARU to the hull which underpins the
importance of measuring performance at ship scale.

The Silverstream® System utilises a natural phenomenon known as the Kelvin-Helmholtz instability
as shown in Fig.1, using uniquely designed and patented ARUs to generate microbubbles with minimal
energy input, Johannesson et al. (2015).

Fig.1: Kelvin Helmholtz Instability

A shearing mechanism is created by the passing of fast-flowing water across a stationary air interface
resulting in the formation of microbubbles in the boundary layer which is then carried along the flat
bottom of the hull as the vessel moves through the water. The layer of microbubbles remain in the
boundary layer reducing the frictional resistance of the vessel.

This gives a reduction in shaft power and an increase in speed for a constant RPM as the
thrust/resistance ratio associated with the specific hull reaches a new equilibrium. The increase in speed
can also be translated into a change in shaft power based upon the vessels measured speed/power curve
which provides three (3) possible scenarios for Ship Owners/Operators:

RPM is maintained as constant and benefit of technology is taken as a combination of fuel savings
and increase in speed
Shaft power is maintained as constant resulting in an increase in RPM allowing the full benefit to
be taken as speed gain
Speed is maintained as constant reducing the required engine power and RPM allowing the full
benefit to be taken as fuel savings

The first prototype Silverstream® System was sponsored by Shell and retrofitted to the relatively
unsophisticated 40,000 DWT chemical tanker MT Amalienborg in 2014, Silberschmidt et al. (2016),
(renamed to MT Shandong Zihe) which demonstrated average net savings of 4.3% during extensive
trials post installation with many confirmed double runs exceeding net savings of 6% and with a
maximum of 12%. During these sea trials, 52 runs consisting of 26 double runs were executed over 3
days in near perfect metocean conditions (i.e. flat calm with minimal wind and waves) which is rarely

202
afforded within a commercial installation scenario. Within these trials, care was taken to measure
impact of the bubble carpet upon performance of the ships propeller to remove concerns relating to loss
of propeller thrust and risk of cavitation, neither of which was experienced. Long term in-service
monitoring in the fully-loaded condition then demonstrated savings of 5.1% and the results of both
trials were independently verified by Shell, LR and Southampton University.

More recently, the Silverstream® System was installed on two cruise vessels, one newbuild and one
retrofit for major owners which demonstrated savings in the range of 4-6%, the results of which have
been also been independently verified by LR and HSVA.

These vessels encompass state of the art technologies with modern sensor units, integrated ship
management systems and sophisticated performance monitoring equipment. In theory, given the
availability of these technologically advanced systems onboard geared to monitor all aspect of vessel
performance, analysis of the performance of an EET should be straightforward and simple. However,
the complexity of system integration poses issues in terms of ensuring that sensor signals are provided
uncorrupted since it is inevitable that they will pass through multiple systems before connection to the
control and monitoring system of the EET. Additionally, irrespective of the complexity involved, vessel
performance is still heavily dependent upon simple measurement instruments such as a speed log for
STW which has been seen to provide inaccurate and unreliable signals in many cases. Additionally, the
commercial pressure to return these vessels to service immediately after dry-docking is also such that
execution of dedicated trials like those conducted on MT Amalienborg, Silberschmidt et al. (2016), is
rarely a viable option.

As a result, trials conducted in-service can introduce many issues such that the effort involved in data
collection onboard and performance analysis to determine the simplified net savings effect from ALS
requires on average 100-man days for each vessel.

3. Performance Monitoring & Data Analysis

The complexity of conducting performance trials can also be somewhat daunting as there are various
and navigating traffic

Additionally, changing weather conditions also cause fluctuations in conditions such as wind direction,
speed and current. For many ships, the above is unavoidable as it is often not possible to conduct trials
in isolated conditions outside of ship operations i.e. during dedicated shipyard trials which could
improve accuracy of results.

However, one of the advantages with trials on cruise vessels is the simplification of operating at a single
draught as hull flow lines can change dramatically especially when the waterline on the bow differs in
the presence of a bulb. With small changes in draught (typically in the range of 10-20mm), the hull can
move to non-optimum operating conditions due to additional submergence of the bulbous bow thereby
increasing the residual resistance on the hull and care should be taken when considering results where
there have been any variations in draught, Gorski et al. (2013).

3.1. Analysis Methodology

The Silverstream® System, unlike other EETs, has the added benefit of being able to be switched ON
and OFF which greatly simplifies measurement of system performance. This functionality provides the
added benefit of being able to isolate the impact of the technology from other effects on the ship such
as change in fin stabiliser deployment particularly relevant for cruise ships.

The Silverstream Performance Monitoring System integrates with ship systems collecting high
frequency data from other systems to accurately measure changes in performance between system OFF
and ON conditions as shown in Fig.2.

203
 Fig.2: Silverstream Procedure for Switching System ON/OFF

The settling and measurement periods shown in Fig.2

experience during performance trials which identified that a period of 10-15 minutes is enough to allow

The settling period however may be adjusted up or down during the testing phase as required to increase
accuracy of measurements being taken onboard. Data inputs to the Silverstream computer are logged
upon any change in an input outside the prescribed limits. This is above the required frequency given
in ISO 19030, ISO (2016), as some parameters such as fin stabiliser and rudder angle positions can be
required to be recorded at a higher frequency to sufficiently capture the variation and account for
measurable changes between ON and OFF periods.

The two most critical parameters required to evaluate performance of the Silverstream® System are
shaft power and vessel speed as for a constant RPM, the shaft power required is reduced resulting in an
increase in speed in the system ON condition when compared with the corresponding OFF condition.
Typically, a total gross power reduction of 500-600 kW can be seen at 14-15 kn whilst at 17-18 kn, the
power reduction is usually in the region of 800-1000 kW as seen on large cruise ships. On average, the
ratio between shaft power reduction and equivalent power from the speed increase has been identified
as a 40:60 split but this can vary based on changing conditions. This means that at a constant RPM,
40% of the power reduction is seen at the shaft and 60% of the power reduction is gained from a speed
increase of 0.1-0.7 knots.

Both speed and shaft power data inputs were found to have varying reliability over the period of analysis
which introduced a level of uncertainty in the results as discussed in the next section.

3.2. Data Quality & Accuracy

In this section, some examples of data quality issues and actions taken relating to the main parameters
affecting accuracy of the performance evaluation are provided.

3.2.1 Shaft RPM

Data from the torque meters is usually received at a high frequency and the signal is assumed to be
sufficiently accurate with the expectation that is it routinely calibrated and for simplicity, the calculated
power is usually fed to the onboard datalogger. In a given period during trials, some inaccuracies in the
power was noticed and forensic analysis driven by inconsistency between the two shafts showed that
shaft 2 RPM measurements fluctuated by 1-2 RPM as shown in Fig.3, such that power measurements
derived from these RPM readings were unreliable in comparison to shaft 1 which kept constant
throughout performance testing. Given the averaging of data, it was not obvious that shaft 2 RPM was
fluctuating but instead showed that the readings were constant at one 1 RPM higher leading to warped
results. Only after RPM mea-surements were analysed in detail, the fluctuations of 1-2 RPM became
apparent. Consequently, as we were confident that both engines were balanced, inconsistency in the
data was minimised in this instance by doubling the power from shaft 1 in calculation of the overall
performance to increase accuracy of results pertaining to RPM and shaft power measurements.

204
 RPM Shaft 1 & 2
RPM Shaft 2 RPM Shaft 1
2,00

1,50

1,00
RPM

0,50

0,00

-0,50

-1,00
Timestamp (hh:mm:ss)

Fig.3: RPM Fluctuation over time

3.2.2 Speed Measurement

Experience from several trials also identified that there can be intermittent issues with readings from
the speed log. Fig.4 shows a time plot of STW vs. SOG signals over a 6-hr period whilst at sea which
identified that the STW signal dropped out frequently throughout performance testing.

STW (kn) vs SOG (kn) Fluctuations

SOG (kn) STW (kn)
24
22
20
18
16
14
Speed (kn)

12
10
8
6
4
2
0
-2
Timestamp (hh:mm:ss)
Fig.4: STW vs. SOG Fluctuation over time

This provided the greatest source of inaccuracy as the speed gain provided by the Silverstream® System
makes up approximately 60% of the shaft power savings as discussed earlier. It should also be noted
that if there are no changes in current, the SOG can be more accurate and reliable in measuring the
speed gain achieved between OFF and ON conditions. However, the changes in current within a
measuring period can also affect the accuracy, but if large datasets are available the confidence can be
increased as the variation can be reduced. Hence, if the STW signal is seen to be temperamental, but
SOG and STW are similar, the confidence level achieved can be higher.

205
3.2.3 Wind Speed & Direction

performance with EETs.

level of uncertainty as large and frequent changes in relative wind direction were identified in recorded
values compared to small changes in wind speeds for corresponding times as shown in Fig.5.

Changes in Relative Wind Direction vs. Wind Speed

Relative Wind Direction (deg) Relative Wind Speed (kn)
300 24
22
Relative Wind Direction (deg)

250 20

Relative Wind Speed (kn)

18
200 16
14
150 12
10
100 8
6
50 4
2
0 0
Timestamp (hh:mm:ss)
Fig.5: Wind Direction and Speed Fluctuation over time

These environmental changes play an important part as fluctuations in weather conditions very often
introduce large data scatter in results. It is also worth pointing out that the effect of changes in wind
speed and direction will vary across ship types as wind on a cruise ship will have a larger effect on the
resistance impact than on a fully laden tanker/bulk carrier. CAA for vessels with up to 5 decks can be in
the region of 0.04 0.07, Kristensen and Lützen (2013), whereas on cruise ships with 10-20 decks, the
coefficient will be significantly larger. Additionally, height of the anemometer on cruise vessels are
often located greater than 50m above the waterline thus introducing varied results. Hence, ISO19030
should be used as guidance and applied where applicable for type of ship being considered when filter-
ing out for changes in weather and wind conditions.

4. Data Filtering & Analysis

Data quality also plays an instrumental part in determining performance of EETs as described above.
Fig.6 shows unfiltered ON and OFF data collected onboard a cruise ship during 2 separate sea trials
over a total trial period of 10 days. The large scatter in data can be attributed to changing parameters,
environmental conditions as well as varying levels of accuracy in the data collected from several
onboard instruments and sensors.

In accordance with the Silverst Testing Procedure, data is analysed in 10-15 min
blocks where each block is made up of 10 or 15 x 1 min average data points. Runs are then analysed
and eliminated where there are any significant changes in rudder angle, shaft rpm, wind speed and wind
direction between ON and OFF periods. The time interval is dependent mainly on the ship size/type as
it is based on the time required for the ship to come to a new equilibrium condition following a change
in operation which initiates an effect on the resistance forces acting on the hull whilst sailing in system
ON vs. OFF conditions. Similar changes will also occur when variations in RPM and main engine
power settings are made by the crew. The criteria for filtering runs between ON and OFF conditions in
Table 1 has generally been used within which specific data points that comply with the criteria below
are retained.

206
 Cruise Ship Performance Trials (unfiltered)
Silverstream® System ON Silverstream® System OFF
1,0
0,9
Normalised Shaft Power

0,8
0,7
0,6
0,5
0,4
0,3
0,2
0,1
0,0
12 13 14 15 16 17 18 19 20 21 22
Ship Speed (kn)
Fig.6: Cruise Ship Performance Trials (unfiltered data)

Table 1: Criteria for filtering data between ON/OFF periods

Rudder Angle (rate of change) < 1°(in accordance with ISO 19030-
2:2016 Annex I & J for outliers)
RPM < 1 RPM
Wind Direction < 30°
Wind Speed < 10 kn
Water Depth > 60 m
Max Wind Speed All valid runs < BF 5 wind scale
(change in conditions of 1 BF scale)

Cruise Ship Performance Trials (filtered)

Silverstream® System ON Silverstream® System OFF
1,0
0,9
0,8
Normalised Shaft Power

0,7
0,6
0,5
0,4
0,3
0,2
0,1
0,0
12 13 14 15 16 17 18 19 20 21 22
Ship Speed (kn)
Fig.7: Cruise Ship Performance Trials (filtered data)

In addition to filtering using the above basic criteria, data points which showed inaccuracies due to
errors from instruments (such as the speed log) were discarded to increase the confidence in perfor-

207
mance achieved from the system. In the below example, a total of 131 runs were conducted during
commercial operations which resulted in 76% of the data points (100 runs) being discarded once filter-
ing was applied. Performance of the system was then evaluated using 24% of the data points (31 runs)
which were found to be valid as shown in Fig.7.

Although filtering was carried out to account for all known inaccuracies and changes in conditions, it
is evident that there are still some unknowns in the data remaining which confirms the complexity of
being able to accurately measure and evaluate performance savings for hydrodynamic based EETs even
with the added ON/OFF functionality of the Silverstream® System.

Another factor which should be considered based on Fig.7 is the need for additional data points at
higher speeds i.e. above 18 knots. This can have an effect on the speed/power curve as fitting of the
curve can be influenced by the lower speeds due to the large quantity of data points in the range of 16-
18 kn. Hence, to accurately determine performance impact of the technology at higher speeds, a sepa-
rate speed/power curve should be fit to points at higher speeds if it is not possible to collect additional
data.

5. Discussion

Although the Silverstream® System has the added benefit of switching ON and OFF, there are several
other factors and non-linear environmental parameters which introduce a large scatter of data. These
increase complexity in filtering, analysis and determining performance of the system.

One of the main areas that can introduce uncertainty is the quality of data being received onboard. Data
is received from various instruments and then routed through several onboard systems before being
received in the Silverstream Control and Monitoring System. This increases complexity significantly
as it is often not clear how data is processed before being received. With the large quantity of instru-
ments and sensors providing data and not being able to check the accuracy of each one, it is often
difficult to determine the quality of data being received.

Data in 15 min blocks has been filtered out to remove inaccuracies and separated into the main catego-
ries as shown in Fig.8.

Overview of Invalid Runs

25

20
Frequency

15

10

0
Environmental Conditions 4 3 0 1 3 1 5 9 0 4
Rudder Angle 1 3 2 2 0 5 0 0 20 4
RPM 1 0 0 1 2 2 0 0 5 3
Sensor Inaccuracy 0 3 7 7 2 0 0 0 0 0
Type of Inaccuracy
Fig.8: Cruise Ship Performance Trials (filtered data)

Most of the data scatter can be attributed to unavoidable environmental factors. However, another major
contributor to data inaccuracy is the unreliability of data received from onboard instruments which is

208
often only identified after testing has been completed and data is interrogated during analysis.
Despite the above, the Silverstream® System offers an advantage that can increase confidence in the
accuracy attained as savings can also be verified by analysing runs in near ideal environmental
conditions with stable parameters. Hence, any miscalibration or inconsistency of signals over time can
be cancelled out by comparing consecutive ON and OFF performance within a specific time interval.

6. Recommendations

To improve the accuracy and confidence in savings attributed to EETs, it is recommended that the
following steps are taken:

Use of reliable and calibrated torque and thrust meters.

Regular checks of all data feeds and measurement accuracy to be considered with focus on speed
logs and shaft torque meters.
Utilise signals directly from the sensors/instruments (if possible) thus avoiding any processing or
rounding of values and loss of accuracy in transmission.
Greater fidelity in data capture, quality of signals and data collection may demonstrate increased
performance.
Real time analysis of data collected during performance trails to check that data signals being re-
ceived from onboard instruments are providing accurate readings in adequate frequency.
Utilising hindcast metocean data to cross check measurements at ship where data permits.
Long term analysis can serve to give further confidence in sea trials results, but care should be
taken not to mix data from different conditions such as ballast/laden draughts and the issue of hull
degradation over time should also be considered.

The above points would help to address some of the uncertainties encountered in performance trials
with the hope of having greater confidence in performance results measured onboard.

7. Conclusions

Although the Silverstream® System has the distinct advantage over other hydrodynamic based EETs
in that it can be switched OFF allowing a baseline to be established in a significantly more controlled
manner, measuring and demonstrating performance can be extremely complex. This can be attributed
to changes in ship parameters, environmental conditions and sensor/instrument inaccuracy because of:

Variations in rudder angle, RPM, fin stabiliser positions due to ship operational schedule and nav-
igating in shipping lanes.
Inability to conduct dedicated trials outside of commercial operations in near ideal conditions

large scatter of results.

Inaccurate measurement signals recorded from onboard ship instruments.
Reduced data quality and clarity as data is processed, received from various instruments and then
routed through many onboard systems.
Increase in data scatter and uncertainty when different draughts are being considered.
Lack of data points at certain speeds can introduce inaccurate curve fitting through these points.
SOG measurements over the years have become extremely accurate while the STW technology
available has remained relatively stationary and does not provide accuracy to the desired degree
increasing the uncertainty in recording accurate STW.

Additionally, ISO 19030 can offer a sound base document for evaluating performance but there is the
need to differentiate between reported results based on the sophistication introduced and going forward
the following should also be considered:

209
 One should not underestimate the effort and quantity of data required to analyse/conduct trials and
that a single trial with a couple of runs is unlikely to provide a true indication of an EETs perfor-
mance.
ISO 19030 and basis of data filtering is currently the only viable means of executing analysis per-
formance of an EET.
Time should be spent to check and verify that sensors are operable and reading accurately, a re-
commendation which is already encompassed within ISO 19030.
Investing in technology to improve efficiency onboard requires robust sensors and systems which
can be sufficiently integrated with EETs to provide a reliable route of measuring performance and
advancing the EET to provide additional benefits to both the owner and the environment.

Considering the above factors and through our experience of performance testing, achieving high
confidence levels requires large sets of high frequency data with minimal instrument/sensor inaccuracy
which can be difficult to achieve on many occasions especially during ship trails in commercial
operations.

References

GÓRSKI, W.; ABRAMOWICZ-GERIGK, T.; BURCIU, Z. (2013), The influence of ship operational
parameters on fuel consumption, Maritime Univ. of Szczecin, Vols. ISSN 1733-8670, p. 6

ISO (2016), ISO 19030: Ships and Marine Technology - Measurement of changes in hill and propeller
performance, Int. Organization for Standardization, Geneva

JOHANNESSON, J.; SILBERSCHMIDT, N.; CLAUSEN, J. (2015), Air Lubrication System and
Vessel Comprising such a System, Int. Patent WO2015/133900a1

KRISTENSEN, H.O.; LÜTZEN, M. (2013), Prediction of Resistance and Propulsion Power of Ships,
TU Denmark, Lyngby, p. 52

LIEM, H.C.; TODA, Y.; SANADA, Y. (2013), Effect of Air Lubrication Method on Frictional
Resistance of Ship, ASEAN Eng. J. 4/A, p.9

MOLLAND, A.F.; TURNOCK, S.R.; HUDSON, D.A. (2017), Ship Resistance and Propulsion,
Cambridge University Press

SILBERSCHMIDT, N.; TASKER, D.; PAPPAS, T.; JOHANNESSON, J. (2016), Silverstream®

System Air Lubrication Performance Verification, Low Carbon Shipping, p.12

STEVENS, R.L.; STEVENS, F.B. (1848), Improvement in apparatus for the increase of the speed of
vessels, USA Patent US5644 A

210
 Autolog Data Processing for Vessel Performance Application
Najmeh Montazeri, Vessel Performance Solutions, Lyngby/Denmark, nmo@[Link]
Jakob Buus Petersen, Vessel Performance Solutions, Lyngby/Denmark, jbp@[Link]
Søren Vinther Hansen, Vessel Performance Solutions, Lyngby/Denmark, svh@[Link]

Abstract

This paper describes the procedure that has been developed to use autolog datasets in the analysis of
hull and propeller performance for ships. The datasets have been received from the sensors onboard
the ships and the data have been validated through different processes. The influence of this data
processing on the scatter of added resistance is investigated and the results of performance trends are
compared with manual (noon) data.

1. Introduction

During the last couple of years, VPS has developed an advanced Vessel Performance Decision
Support system (VESPER) that provides information about fleet fuel efficiency using traditional noon
reports. Noon reports will continue in the shipping industry to be the main source of data exchange
between vessels and operators for quite some time. Therefore, we have to develop robust methods for
performance analysis of noon reports. However, those manual data have the disadvantage, that they
typically cover a period of 24 hours where, in many cases, the important parameters that influence the
propulsion of the vessel are not stable. Moreover, the data are averaged and subject of human
interpretation especially the evaluation of the waves, wind and currents over the noon reports. This,
inherently, has the consequence that the determination of the vessels hull and propeller performance
over time deals with some uncertainty and a relatively long evaluation time is required to determine
the hull and propeller degradation slope over time.

Using autolog data for performance management has been thoroughly discussed in recent years and
more ships are being equipped with sensors. The sensors measure operational data with a very high
frequency. However, the use of autolog data does not necessarily solve the above-mentioned
uncertainty, one issue is for example that the vessel is often experiencing periods of acceleration.
Environmental forces are constantly fluctuating causing constant accelerations. Engine response is
also not constant, most evident is variations in RPM when the engine is not running at optimum
conditions (engine overload). Since most methods for analysis of performance data are based on the
assumption of quasi-steady state forces, such accelerations may cause scatter in the results when
analyzing over shorter periods of time. Moreover, faulty or drifting sensors can significantly mislead
the performance evaluation. Therefore, sensor reliability becomes a major issue for the analysis.

Towards the industry prospects, VPS has started to investigate how the VPS performance
management platform VESPER can be improved to reasonably support autolog data. In order to get a
valuable and efficient assessment of the vessel performance based on autolog data, filtration,
averaging and plausibility checks are important. The objective of this work (in the scope of iSea
project funded by EU program, [Link]
2/eurostars-cut-off-5/isea-intelligent-telematics-and-safety-solution-for-sea-vessels) is to develop the
algorithm that determines stable periods without significant accelerations and faulty signals. If
successful, we will be able to establish the performance of vessel faster and more reliable than using
traditional noon reports or averaged-unfiltered autolog data sets. On the other hand, using noon data
for performance management is still unavoidable. So, we will try to combine the results of noon and
autolog data. The evaluated performance indicators of the vessel are expected to be less scattered and
more reliable. Further, we expect to learn more about the influences of the environment acting on the
ship itself and to learn more about the baseline ship model itself with more accurate stable periods
from high frequency autolog data. Information that we can learn from and utilize in our daily analysis
of the less precise noon data.

211
The method of stable period detection takes its origin in an approach for automatic fault diagnostics,
Lajic (2010).

2. Input parameters

Multiple measurements/signals are received from different sensors. Depending on the performance
indicator that needs to be analysed, the different signals are optional to be used in the filtration
procedure. In this paper, we focus on hull and propeller performance. Therefore, the following
channels are used:

Speed over ground

Logged speed
Ship heading
Wind speed
Wind direction
Fuel consumption from fuel flow meter
Shaft RPM
Power from torsion meter

For many vessels, the draft data are not provided. So, in the analysis, the drafts are transmitted from
the noon reports in the same periods. A good example of how noon data and autolog data can
supplement each other. It should be noted that more channels can be used, for instance, water
temperature, turbocharger RPM, rudder angle, etc, if they are available.

A table of parameters is predefined for each vessel as filtration criteria. The parameters that are used
to define the stable periods depend on the available sensors and the ship characteristics. Therefore, the
parameters set needs to be defined for each vessel type and/or vessel segment individually. The
optimization of the parameters is not automated yet. Therefore, verification based on visual
observations of samples of stable periods that come out of the algorithm is required for each dataset.

The signals are filtered preliminarily based on physical range checks. In addition to filtration of non-
physical outliers, the algorithm looks for spikes. A spike is recognised as a single point, the deviations
of which from the previous and the next points are higher than X% of the predefined [max min] val-
ue for that sensor. X is considered as a parameter.

The following parameters are used as criteria for individual time series:

1. Minimum mean (min mean)

2. Maximum mean (max mean)
3. Maximum standard deviation (max std)
4. Mean deviation threshold
5. Standard deviation threshold

Each parameter corresponds to a check that is applied on the time series. Every check can be
included/excluded through a flag for individual signals.

3. Filtration algorithm

The first step is to go through the data, point by point. The window size (the period that is considered
at each step) is defined as another parameter. The mean value and the standard deviation of the first
window is calculated. Then the
checked to be less than
(parameter 3). If any of the checks fail, the next window is looked at. The first window the
statistical parameters of which, passes the above criteria is considered as the reference window.

212
Then the next window is considered, and the mean and standard deviation are calculated. We wish to
find periods when the ship responses are stable according to our predefined criteria. In order for time
series to be stationary, the mean value and standard deviation have to remain the same as the
reference window over time. Therefore, the theory of probability of detection of a change in the mean
and standard deviation is utilized, Lajic (2010). In this step, mean deviation threshold and standard
deviation threshold (parameters 4 and 5) are used to compare the statistical parameters of the current
window with the reference window.

An output matrix is defined with an equal size to the number of data points in time series. If any of the
stability criteria checks fails, the elements in the output matrix become red . In this case, the
algorithm resets the reference statistical parameters in the next step. In other words, the 3 former
checks will apply on the next windows until a new valid reference window is found. If the
measurement point passes through all checks, all points in the specific window become green in the
output, meaning that the window is a stationary period.

Since the outliers are removed from the time series, the time gap between every two consecutive time
stamps are checked. If this time gap is longer than 20% of the window size, the output becomes red .

The above procedure applies on every signal from section 2.

4. Averaging

When stationary periods are detected for all measurements, we focus on

which we define as a period when all the time series are simultaneously stationary. In those periods,
the average of individual time series is calculated over those periods. If the mutual period is longer
than the output sampling rate (set as a parameter), the average is calculated over that specific period.
For instance, if the output sampling rate is 60 min and the mutual stationary period is 130 min, the
average value is calculated every 60 min and therefore, 3 average points will be calculated for each
sensor. Correspondingly, the report durations for the new points will be 60, 60 and 10 minutes.

The averaged values and the corresponding report durations are stored in the database. Another

value, the average points are removed from the data.

Beside the detection of stable periods, there is an algorithm to calculate the average of every sensor
measurement every Y hours without doing any filtration. Y is another arbitrary parameter defined by
the user. This averaging of autolog data is, today, a common method in the industry. The 24-hours
averages are used in section 8.

5. Case study

The case study in this paper is a tanker, Table 1.

Table 1: Vessel properties

Property Value
LOA 244.60 m
LPP 233.00 m
Beam 42.00 m
Design Draft 12.00 m
Design Deadweight 107640 t
MCR 15260 kW

The sampling frequency of the input data is 2 minutes.

213
6. Stable period results

Fig.1 shows an example of stable time series detection. The blue lines show the input data. The green
points indicate the detected stationary time series for individual signals. The black points appear when
a change is detected in the mean or standard deviations of the stationary time series. The red points
are the average points over the period of mutual stable time series. In this example, 5 different signals
have been considered for the analysis.

Fig.1: Stable period detection and average evaluation

The average values of stable periods are stored in VESPER for analysis and calculation of
performance trends. The following sections show some of the findings.

7. Validation analysis

In order to assess if the dataset is valid and worthy for performance analysis, we need a validation
process from a physical point of view. The parameters relative to each other are checked based on
physical boundaries or a range perspective. The ship model (baseline) determines what is possible and
what to expect for the boundaries of the parameters.

Fig.2 compares the SFOC values with the shop test results. It can be observed that autolog data and
noon data are quite conforming and the values and the scatter are reasonable. There are other technical
validation procedures that can be carried out manually regarding the engine related data, but they are
out of the scope of this paper.

The data also get through the validation engine that is designed in VESPER. There are many rules that
can be set by the user to check the validity of data depending on the point of interest. For the
considered autolog dataset in this paper, rules for missing fields and range checks of the sensor
measurements are set up. The output of the validation engine will help to detect and diagnose
problems in the received data, i.e. sensor reliability.

In Fig.3, the data quality is assessed through the VESPER validation engine. Red output shows the
amount of failed data point through a specific rule. Yellow output shows the number of warnings
coming out of a specific rule. The bars on the left show the total percentages of errors and warnings

214
among all rules for the specific vessel. The bars on the right show the results per rule. It can be
concluded that the data quality is very promising in this case. The draft rule has triggered 100%
because the drafts are missing in the autolog data and they are enriched with noon data in the analysis,
as mentioned before.

Fig.2: SFOC values vs the model

Fig.3: Validation output for autolog data set

Through this validation check, the users can find out whether or not the autolog data are worthy for
analysis and if not, they can recognize the problems and investigate the reasons and possibly resolve
them. It is noteworthy that during our studies, we have found a couple of vessels for which, the
autolog data included invalid values from sensors. In that case, the results of performance evaluation
have to rely on noon results only.

8. Performance trends- Comparison with noon reports

In this section, the performance trend results are shown. In the previous section, it is shown that the
data set has very good quality and delivers sensor values within acceptable ranges. For comparison
purpose, beside the stable periods, we also make the analysis based on the unfiltered autolog data
which are averaged every 24 hours and compared with the conventional noon data. In this section, we
use three different input data for deriving the added resistance:

1. Autolog results using stable periods

2. Autolog results (unfiltered) using 24 hours averages
3. Noon reports

215
 Added Resistance [%]
Added Resistance [%]
Added Resistance [%]

Fig.4: Comparison of added resistance (Logged speed-TorsionMeter) between unfiltered autolog

(top), stable autolog (middle) and noon (bottom) data

216
 Added Resistance [%]
Added Resistance [%]
Added Resistance [%]

Fig.1: Comparison of added resistance (Observed speed-TorsionMeter) between unfiltered averaged

autolog (top), stable autolog (middle) and noon (bottom) data

217
Added resistance is a major performance indicator in VESPER. This value represents the relative dis-
crepancy between the actual and the expected resistance. The expected resistance is calculated based
on the propulsion model for the specific vessel plus the calculated resistance due to the environmental
conditions/weather. The actual resistance is calculated based on the measured fuel/power. The added
resistance trend indicates the level of fouling/damage on the hull and propeller over time since the last
event on the vessel. The trend evaluation is started over, once an event/treatment is registered.

VESPER can calculate the added resistance trend based on different methods. Here, we show only the
best results which are based on power from torsion meter and the vessel speed (both observed speed
and logged speed). The added resistance based on logged speed and torsion meter was found to have
the lowest scatter and the results are very close to noon data. Fig.4 and 5 show the trends of added
resistance using torsion meter-logged speed and torsion meter-observed speed, respectively. The
trends are calculated based on the P-Rate method which is a modified linear trend depending on the
prediction reliability. As mentioned before, for the same vessel within the same period, we extracted
also noon reports to calculate added resistance. The results are shown in these figures for comparison.
As mentioned above, the unfiltered data are averaged every 24 hours. The filtered (stationary) data are
averaged every 1 hour.

Tables 2 and 3 show the resulting added resistances from Figs.4 and 5. Two trends are considered
before and after the propeller polish on 14/7/2018. The prediction
reliability function is a VESPER internal function which depends on both the number of valid reports
within the trend period and the scatter (STD) around the least squared fitted trend line. The more data
points and the less scatter in added resistance, the higher prediction reliability would be obtained. In
both tables, by using the logged speed, the magnitudes of added resistance, which is the performance
of vessel at the evaluation date, for filtered autolog data are very close to noon dataTable 1. The
scatter (STD) is significantly decreased by use of filtered autolog data, approximately half the
standard deviation of the noon reports and significantly more data points. The prediction reliability is
20-30% higher than noon as the number of points are higher and the scatter is lower.

Table 1: Comparison between results of autolog data and noon reports- Added resistance based on
torsion meter- Trend before propeller polish on 14/7/2018
Method Source Valid Prediction Prediction P-Rate STD
reports Reliability reliability corrected of
Score trend AR% AR%
Logged Autolog (avg. 24 h) 76 25 Low 42 39
speed Autolog (Stable periods) 449 80 High 40.2 9.3
torsion Noon 74 53 Medium 39.2 22
meter
Observed Autolog (avg. 24 h) 76 25 Low 35.7 40
speed- Autolog (Stable periods) 449 77 High 36.4 17
torsion Noon 74 55 Medium 33.5 22
meter

Table 2: Comparison between results of autolog data and noon reports- Added resistance based on
torsion meter- Trend after propeller polish on 14/7/2018
Method Source Valid Prediction Prediction P-Rate STD
reports Reliability reliability corrected of
Score trend AR% AR%
Logged Autolog (avg. 24 h) 16 5 Low 51 36
speed Autolog (Stable periods) 41 43 Medium 37.2 6.5
torsion Noon 30 23 Low 37.8 22
meter
Observed Autolog (avg. 24 h) 16 1 Low 44.7 36
speed- Autolog (Stable periods) 41 20 Low 18.6 19
torsion Noon 31 0.4 Low 46.4 71
meter

218
By using the observed speed, the added resistance value just before the propeller polish is very close
between stable autolog and noon reports. But at the end of the next period (after the propeller polish),
the added resistance is much lower than for the noon data.
However, reliability of the noon data is skeptical here which can be seen by very low prediction
reliability (0.4) and the the very high scatter around the trend line (71) in the added resistance. The
scatter from observed speed and torsion meter is twice compared to the results from logged speed and
torsion meter which could be due to the ocean currents and, for once, a well-functioning speed log.

Using stable periods based on logged speed seems in this case to be significantly superior to both 24
hour average autolog data and noon reports with respect to standard deviation and prediction
reliability. The performance evaluation seems to be consistent between the three methods in the first
period, whereas the three methods are less consistent in the second period. VPS will further
investigate the reasons for this in the future. Reduction of scatter is important, but we must not forget
the main objective, which is to determine the performance of the vessel.

9. Summary and Conclusion

VPS has developed a tool for analyzing ship-based autolog data to be integrated into the VESPER
performance application. The target is to investigate how the sensors that are already invested in and
installed onboard vessels all around the world can be utilized for performance management. Human
errors in the manual data can somehow be compensated by combining the noon data with reliable
autolog data.

The parameters set for stable period detection should be vessel specific and still needs to be fine-tuned
manually. The stable periods will be identified based on the theory of probability of detection of a
change in the mean and standard deviation, Lajic (2010). The average signals are stored to be
transferred to the analysis platform. The program is also capable of calculating the average
measurements over the requested period without any filtration. This gives the user the ability to
compare the results of filtered and unfiltered data.

Validation engine in VESPER needs to be run to check the data quality. If the data are not in the
expected physical ranges, the sensors need to be inspected or calibrated. The data from the tanker that
is used in this paper show, relatively, a very good data quality.

Hull and propeller performance is analyzed based on the filtered and unfiltered autolog data. The
method of stable periods is superior to noon data and unfiltered 24 hours-averaged autolog data.
However, the results of performance evaluation are not necessarily the same as noon data, which
indicates that more thorough investigation is required. In general, the logged speed has been found to
give more consistent performance compared to noon results for this vessel.

In the , [Link] VPS will continue the work for

further investigation of autolog-based performance analysis and also improvements of modelling in
VESPER.

References

LAJIC, Z. (2010), Fault-tolerant onboard monitoring and decision support systems, PhD thesis, TU
Denmark, Lyngby, pp.50

219
 When ISO19030 Fails: Utilizing Basic Machine Learning/Data Fitting
Methods for Performance Analysis without Reference Data
Ditte Gundermann, Hempel, Copenhagen/Denmark, DIGU@[Link]
Danny McLaughlin, Hempel, Copenhagen/Denmark, DaMcL@[Link]

Abstract

ISO 19030 compares in-service data with reference data such as sea trials and model tests. Non-
conclusive results are common when there is not enough reference data to match the in-service data
or when the filters have removed a large part of the in-service data. An alternative analysis method
based on a multiple linear regression model is tested for different ship and data types. The linear
model is fitted to in-service data only and requires no reference data. The predicted speed power
curves are evaluated against available reference curves, and the predicted performance trend in time
is compared to the performance trend found by the ISO 19030 method. Despite some limitation and
uncertainty, the model is found to give good prediction of the power at different loading conditions
and at different points in time. Some suggestions for improving the uncertainty are given.

1. Introduction

The ISO 19030 standard on performance monitoring describes how to evaluate the hull and propeller
performance of a ship by comparing data measured on board the ship (in service data) with reference
data such as sea trials or model tests. In the common case that the reference data cannot be related to
most operational conditions of the vessel, the ISO19030 method gives no or inconclusive results.
Also, even with adequate reference data, there are other aspects of the ISO 19030 standard analysis
that can give rise to misleading results. One of these aspects is the well-known speed (and draft)
dependence of the performance values, Bertram (2017), which can be quite significant and give false
impressions of performance variations, Schmode et al. (2018). Using a mathematical model of the
ship at reference condition can help solve these problems by expanding the reference data
significantly and by normalizing the performance values to a single condition (draft/trim/speed), but
such models often require the same information in terms of reference data as the ISO standard, and
are thus not always a possibility. Another limitation of the results of the ISO19030 method is that
they describe relative performance changes over long periods, rather than actual performance at a
specific time. Further, it does not specify how to get actual consumption/power curves for different
drafts or at a given point in time.

The increase in the availability of high frequency data has provided new possibilities for utilizing
mathematical algorithms for analyzing ship performance based on in-service data alone. Many
different methods and algorithms exist and are used also by some performance monitoring service
providers who specialize in these methods, but ready-made tools are available in many open source
and commercial software packages making it increasingly possible for a wider range of users to
utilize machine learning for data analysis.

Apart from not needing reference data, an advantage of such methods is the prediction of the
performance and consumption/power requirement at any time, speed, draft and weather condition. A
disadvantage compared to the ISO 19030 method is obviously the lack of transparency and the risk of
receiving misleading results due to lack of understanding of interdependencies in the data.

With the aim of determining the hull and propeller performance of a vessel using only in-service data,
we examine the feasibility of using a multiple linear regression model for the dependency of power
on different input variables. Although linear models may be considered less accurate than other
machine learning methods for vessel performance analysis, Pedersen (2014), the advantage of linear
models are that they are simple, give a continuous output and make it possible to interpolate and
extrapolate outside of the measured data ranges.

220
The linear model has been tested on data of different frequency ranging from noon reports to high
frequency auto-logged data from 6 ships of different types. The vessels type and data frequency are
shown in Table 1.
Table 1: list of vessels and data used in this study

LNG Tanker 1/15 sec autologged

VLCC 1/5 min autologged
Passenger Carrier 1/5 min autologged
Crude Oil Tanker 1/24 hour manual
Container A 1/24 hour manual
Container B 1/24 hour manual

2. Method

The linear model can be described by the following general expression:

Where x, y, z... are different input variables that affect the output f, and C0,Cx,Cy,Cz are constants.
The number of input variables depend on the given case (type of vessel, data availability and data
frequency). Most significant variables are usually speed, draft, weather and time. Other inputs could
be waves, trim and sea water temperature. In order to fit the model to the data, the data has to be
linearized. In the case of using speed (v), draft (d), weather (w) and time (t) as input the equation
becomes:
.

where p0, v0, d0, and w0 are normalization factors.

The model is applied to in-service data of different frequencies, ranging from noon data to high
frequency auto-logged data. In the case of high frequency data some filtering has to be done on the
raw signals in order to remove non-steady state data points. This is done to minimize autocorrelation
effects. Fig. 1 shows an example of the filtering.

Fig. 1 Raw (red) and post-filtering (orange) speed data. Left: selection of time series data. Data
measured while the ship is accelerating is not used for modelling. Right: speed power plot of
entire raw and filtered data sets. Times when the vessel was accelerating are seen in the low
speed/power range, where the scatter is much higher than at higher speed/power values.

221
The linear expression is then fitted to the data to find the coefficients C0,Cv,Cd, and Cw. Once the
coefficients are found, the predicted power for any input variable combination can be found. This
means that we can for instance plot the time development of power for a certain speed and draft in no
wind, or we can find speed power curves for any draft at any time and for any weather (wind speed
and direction) condition.

3. Results

3.1. Model prediction vs measured

Fig. 2 left shows a time series plot of measured and predicted power for the entire available time
period for all the vessels (after filtering described in section 0). Fig. 2 right shows the same data in a
scatter (correlation) plot. The closer the points are to the red line, the better match between predicted
and measured power. The values are for most data sets evenly distributed around the red line,
showing that the model on average predicts the power well and that the use of a linear relationship
between input variables and the dependent variable power is acceptable. There is some scatter in the
measured power that is not captured by the model. This is also seen in the time series plots, where
some measured values are not well reproduced by the model. This probably reflects the fact that not
all relevant input variables are included in the model. Perhaps the most significant effect not captured
by this model is that of waves. Table 2 shows the coefficients of determination (R2) and the root mean
squared error (RMSE). The R2 values for the noon data-based models are generally lower than the
values for the auto-logged data based models. Similarly, the RMSE values are generally higher for
noon data than for auto-logged. This shows - not surprisingly - that the model performance is better
for high frequency auto-logged data.

LNG Tanker

VLCC

222
 Passenger Carrier

Crude Oil Tanker

Container A

Container B

Fig. 2: Measured and predicted power for all vessels. Left: time series plot, right: correlation scatter
plot. Transparent dots are used for high frequency data, while solid dots are used for noon data
in the correlation scatter plot.

223
 Table 2: Model performance measures for all ships
Vessel Dry Docking
Interval Measure
R2 RMSE
LNG Tanker 1 0.9598 0.1036
VLCC 1 0.9598 0.1505
Passenger Carrier 1 0.9778 0.0878
Passenger Carrier 2 0.9772 0.0857
Crude Oil Tanker 1 0.9415 0.1853
Container A 1 0.9588 0.1353
Container A 2 0.8902 0.1586
Container A 3 0.8703 0.1659
Container B 1 0.9160 0.1901
Container B 2 0.9338 0.2426

LNG Tanker Crude Oil Tanker

VLCC Container A

Passenger Carrier Container B

Fig. 3: Measured (red) and corrected power (blue) for most common draft and speed, calm weather.
Predicted trendline in turquoise). The auto-logged data on the left-hand side has been averaged
to 24 hours, while the noon data on the right are untreated.

224
 LNG Tanker Crude Oil Tanker

VLCC Container A

Passenger Carrier Container B

Fig. 4: Predicted speed power curves for different loading conditions and times in calm wind
conditions. Reference curves are shown for comparison (dashed curves). Different colors refer
to different loading conditions while different line styles in same color refer to different dry
docking intervals.

3.2. Time series plots

After the dependencies (coefficients) of power on the input variables have been found, it is possible to
determine the theoretical predicted power at any time, draft, speed and wind condition. Fig. 3 shows
time series plots of all the measured power values and the predicted trendline for the most common
draft and speed, at calm wind conditions. The predicted trendlines shows a realistic trend of slowly

225
increasing power over time, reflecting hull and propeller performance deterioration over time. The
effects of dry dockings are also seen as a discontinuous shift in the trendlines. One data set (bottom
right) shows a declining power (indicating increasing performance) over time, which could reflect
that events such as hull cleaning and propeller polishing activities which are not taken into account.
The plots also show the measured power values after correcting for the variation in draft, speed and
wind. The measured power has been corrected to the same draft, speed and wind condition as the
predicted line, and agrees well with the trendline (as it should).

LNG Tanker Crude Oil Tanker

VLCC Container A

Passenger Carrier Container B

Fig. 5 Predicted speed power curves for one loading condition (blue lines) at time 0 compared to
measured data points at the same loading condition (orange or light blue dots). Reference
curves for the same loading condition are also shown for comparison.

226
The correction reduces the variation in the power significantly for most data points, but there are
some outliers among the corrected values showing higher values than the measured. Again, the scatter
is likely due to the power-dependent physical phenomena not taken into account in the model, such as
the effect of waves.

3.3. Speed-power plots

One purpose of using the linear model was to produce speed power curves for any condition and at
any point in time. Fig. 4 shows the predicted curves for each ship for one to three loading conditions
at the beginning of the dry docking or data period.

In order to evaluate how realistic these predictions are they are compared with the available reference
curves (sea trial or model test) for the same loading conditions. The predicted curves are similar in
shape to the reference curves for most ships, but are shifted up reflecting a higher power consumption
compared to the new build ship. This agrees with well with expectations.

A few of the predicted curves in have a different speed power relationship (shape of the curve) than
the reference curves at high speeds. When considering the speed range of the measured data in Fig. 5,
this difference in shape probably arises from lack of data in the high-speed range.

In Fig. 5 the predicted curves are shown together with measured raw data points at the same loading
condition (± 0.5m of draft). The predicted curves appear to represent the data points quite well in the
measured speed range.

3.4. Comparison of results with ISO 19030

A final comparison is done to an ISO 19030 standard analysis for the single ship where the ship is
actually trading inside the reference speed and draft range. Fig. 6 shows ISO 19030 defined Power
Performance Values for the passenger carrier together with linear regression lines through those
points. The trendlines from the linear model shown in purple are very similar to the regression lines
through the ISO 19030 Power performance values.

Fig. 6 ISO 19030 Power Performance Values (dark blue dots) and linear regression trendlines
through the points (blue and red). The purple lines are predicted time development of Power
Performance from the linear model.

227
4. Conclusion

A linear model was tested on in-service data of different frequencies for a number of vessel types. he
results of the model were evaluated by:

Comparing the predicted and measured power values. The agreement between measured and
predicted power values is generally better for auto-logged data.
Evaluating the time development found from the model and dry-docking effects of predicted
power. All vessels (except possibly one) shows realistic time developments.
Comparing speed power curves, calculated from the model results, to available reference
curves. The predicted curves were found to agree well with the reference curves in the meas-
ured speed range for all ships.
Comparing speed power curves, calculated from the model results, with measured data at
some loading condition. The predicted speed-power curves agreed well with measured data
for all ships.
Comparing the predicted time development of power deviation to a linear regression line
through ISO 19030 defined Power Performance Value data. The linear model trendline is
very similar to the trendline of the ISO 19030 Power Performance Values.

The model was found not to work in cases not shown in this report, where:

Ship is trading within a narrow speed range (especially for noon data).
There is strong correlation between input variables. If for instance ship is systematically re-
ducing or increasing speed over time.

Besides from these cases, in order for the model to work, some amount of data have to be available.
This means that it can only be used after the ship has been trading for some time in various loading
conditions and speeds. In order to improve the uncertainty of the model, an obvious suggestion would
be to include more input variables. Waves, trim and water temperatures are probably the most signifi-
cant.

Despite the limitations and uncertainties, the linear model can be useful first of all for estimating
speed-power curves at different loading conditions, and second for monitoring long-term performance
trends. In cases where reference curves are not available, or the ship operates outside the speed-power
range of the reference curves, the ISO 19030 standard analysis is not applicable and the linear model
is a good alternative. It can also be used to supplement the ISO 19030 method, because of the possi-
bility for producing speed-power curves for any loading condition and at any point in time. This is
very useful for operational purposes, where the alternative is often to divide data into two different
loading conditions, remove the points with wind above some limit and then fit a speed-power curve
for each condition at certain time intervals. The linear model uses all the data for fitting rather than
dividing into different groups and fitting fewer data separately with higher resulting uncertainty.

References

BERTRAM, V. (2018), Some Heretic Thoughts on ISO 19030, 2nd HullPIC Conf., Ulrichshusen

ALDOUS, L.G. (2015), Ship Operational Efficiency: Performance Models and Uncertainty Analysis,
PhD Thesis, University College London

PEDERSEN, B.P. (2014), Data-driven Vessel Performance Monitoring, PhD Thesis, Technical Uni-
versity of Denmark

SCHMODE, D.; HYLPENDAHL, O.; GUNDERMANN, D. (2018), Hull Performance Prediction

beyond ISO 19030, 3rd HullPIC Conf., Redworth

228
 Application of Sensor Fusion to Drive Vessel Performance
Angelos Ikonomakis, Maersk Line Fleet Performance, Copenhagen/Denmark,
[Link]@[Link]
Roberto Galeazzi, DTU Electrical Engineering, Copenhagen/Denmark, rg@[Link]
Jesper Dietz, Maersk Line Fleet Performance, Copenhagen/Denmark, [Link]@[Link]
Klaus Kähler Holst, Maersk Digital Research, Copenhagen/Denmark, [Link]@[Link]
Ulrik Dam Nielsen, DTU Mechanical Engineering, Copenhagen/Denmark, udn@[Link]

Abstract

emissions due to lower propulsion power demands. Fine tuning of the underlying hydrodynamic mod-
els is key to the reliable forecasting of fuel consumption. speed-
through-water (STW) is paramount to estimate the actual resistance of the vessel. The paper presents
a feasibility study about the use of sensor fusion methods for real-time estimation of STW based on
inertial measurements of ship motions and measurement of sea current. By combining a purely kine-
matic model together with linear Kalman filtering, the paper addresses the challenge of designing an
optimal STW estimator by detailing the fundamental design choices. The proposed STW estimator is
verified on simulated data and tested with measured data from an in-service container vessel.

1. Introduction

1.1. Background and motivation

At sea the speed of a moving craft is measured either relative to the seabed (speed over ground -
SOG) or relative to the water flowing past the hull (speed through water - STW). Both of these speed
types apply in modern navigation systems. The STW is a crucial source of data for the performance
monitoring of a container vessel. Hull and propeller fouling estimation derive from the knowledge of
the STW, hence its accurate observation is a key element towards energy efficiency. Traditionally,
maritime speed logging devices use either of the following measurement principles to obtain the
STW: water pressure, electromagnetic induction, or the transmission of low frequency radio waves
Tetley and Calcutt (2007). The latter refers to as the Doppler velocity log (DVL), which is the speed
log used by the vessels dealt with in this paper. A range of environmental factors can sometimes
influence the accuracy of the measured speed log, Litton (1998). In addition to that, speed log
manufacturers always provide distinct information concerning the possible accuracy.

The speed log measurements cannot be categorically trusted, Griffiths and Bradley (1998). Occa-
sionally, offsets occur when measured STW is compared with computed STW, obtained through post-
processing of hindcast and propulsion data. This observation, along with other influences, such as
trim, aeration, sensor fouling and water clarity, are factors that diminish the confidence of the
measured STW. Therefore, the availability of tools for reliably estimating the STW based on either
already available measurements or measurements deriving from potentially installed sensors, is of
interest to fleet managers for the optimization of fleet performance. The scope of this work is to
estimate STW with an accuracy significantly higher than that of the signal measured by the Doppler
velocity log. This is achieved by fusing both onboard inertial measurements and external data into a
kinematic model of vessel motions. The virtual sensor can be used as input for associated applications
such as data-driven fuel tables, trim optimization and accurate hull cleaning predictions.

1.2 State-of-the-art

The analysis of the data provided by onboard sensors is a complex and necessary task, in order to get
is always
some uncertainty on the measurements, hence it is necessary to analyze each sensor with respect to its

229
type. Besides the manufacturer information on the sensor uncertainty, several uncertainty analyses
have been published. With regard to the ITTC Powering Prediction Method, ITTC (2002), an analysis
has been conducted on the uncertainty assessment of the performance parameters during sea trials,
ITTC (2005). The results of the trials include errors due to measurements, hull form production and
corrections of environmental conditions.

It is noted in the literature, ITTC (2005), that, as speed increases, both bias and precision errors
decrease. Bias errors are 5% for all speed range and 3% at the design speed, while the precision errors
are 9% at lower speeds and 7% at the design speed. Concerning STW, there is a claim from Boom et
al. (2013) which states that the speed log is one of the most inaccurate onboard measurement devices.
Bos (2016) identifies situations where STW data can result in misleading performance trends. These
statements have to be investigated and different values of uncertainty should be tested for STW before
reaching conclusions.

Numerous researchers have pinpointed the various difficulties in STW estimation including Pyörre
(2012) and Antola et al. (2017). Antola et al. (2017) proposed a Bayesian approach to the STW
estimation combining an extended Kalman filter with a mixed kinematic-kinetic model of the vessel.
Despite the very promising results showed through validation on almost 200 vessels, three main
elements are pointed out that could impact negatively the STW estimate: (i) the model assumes
(ii) to exploit measurements of shaft torque it is
assumed that the vessel STW equals the propeller speed of advance, thereby neglecting wake effects
that for ships with a single propeller may account for reductions in the range of 20% to 45% for the
speed of advance; (iii) the vessel accelerations due to wave motion are neglected. Power-resistance
curves are available for each operating vessel; however this data, usually obtained through model
tests, reflects the characteristics of the vessel as new and does not necessarily account for changes in
the hull due to retrofits or simply aging. For a given shaft rotational speed switching the speed of
advance of the propeller with the vessel STW gives rise to a larger advance ratio, which may
determine a systematic mismatch between measured and modelled shaft torque. For a vessel sailing in
waves a thrust-resistance balance to achieve zero surge acceleration is rather unrealistic since it
demands the thrust/torque/shaft speed control systems to completely compensate for wave-induced
speed variations. Therefore, a vessel in seaway can experience substantial accelerations that, at least
elements can potentially introduce systematic errors in the

1.3 Novelty and contribution

This paper addresses the reliable estimation of the STW within the linear Kalman filtering framework
by combining a purely kinematic model of vessel dynamics with inertial measurements of
position, velocity and acceleration as well as velocity of the sea current. The vessel kinematic model
is favored because it does not require partial or full knowledge of vessel hydrodynamics or propulsion
characteristics. The available STW measurement is not considered in the estimation, while it is used
for the evaluation of the estimated STW. Although the adopted model may be used to estimate the
speed through water for a vessel sailing in an arbitrary sea state, this feasibility study focuses on calm
water condition to determine the potential estimation performance in absence of major disturbances
affecting the vessel dynamic behavior.

The paper details the modeling procedure towards the design of an optimal estimator of STW and
illustrates a systematic approach for the evaluation of the obtained estimate. The STW estimator is
evaluated both on simulated and full-scale data from an in-service container ship. Preliminary results
confirm the possibility to improve the knowledge of the STW with respect to the DVL measurement.

2. Problem Formulation

For a correct formulation of the problem the following assumptions are made:

230
Assumption 1: The motion of the vessel in the horizontal plane is described by three degrees-of-
freedom (DOF), namely surge, sway and yaw.

Assumption 2: Two reference frames are used to represent the motion of the vessel: the North-East-
Down (NED) coordinate frame is the inertial frame used to describe the pose of the vessel; the
body-fixed coordinate frame is the non-inertial frame fixed to the vessel used to describe linear
and angular velocities.

T
Assumption 3: The ocean current in the inertial frame is constant and
irrotational, i.e. .

Assumption 4: The vessel is equipped with a GNSS (global navigation satellite system) receiver
providing synchronous measurements of vessel position and speed over ground at the rate
. The position measurement in the North and East directions is affected by zero mean
white Gaussian noise with variance , i.e. and . The SOG measurement
is affected by zero mean white Gaussian noise with variance , i.e. . It is further
assumed that the noise sources , and are uncorrelated among each other.

Assumption 5: The vessel is equipped with a tri-axis accelerometer providing measurements of linear
accelerations along directions parallel to the axes of the body-fixed frame at the
rate . Each measured acceleration is affected by zero mean white Gaussian noise with
variance , i.e. with . The three noise sources are uncorrelated among
each other.

Assumption 6: The vessel is equipped with a compass providing a measurement of heading at the rate
. The heading measurement is affected by zero mean white Gaussian noise with variance
, i.e.

Assumption 7: A prediction of the sea current velocity vector is available through an external
provider at a rate . Each predicted component of the velocity vector is subject to zero
mean white Gaussian noise with variance , i.e. with . The two noise
sources are uncorrelated between each other.

Remark 1. The noise sources impacting the GNSS measurements of position and SOG are strongly
correlated when the SOG is computed by differencing two consecutive position measurements. This
is the case in low-end receivers that are usually not adopted in motion control applications of marine
crafts. Medium to high-end receivers provide the SOG measurement based on either the Doppler fre-
quency shift of the received signal due to vessel-satellite relative motion or the time differenced carri-
er phase, Freda et al. (2015), Gaglione (2015). Here the uncertainty of the SOG depends on the quali-
ty of the phase measurements (affected by measurement noise and multipath) as well as on the accu-
racy of the computed position, although to a minor degree. Hence in this case the noise source
can be assumed uncorrelated with the noise affecting the position measurement.

Let be the pose vector in the frame and be the veloci-

ty vector in the frame. Then the 3-DOF maneuvering model reads, Caharija et al. (2012), Moe et
al. (2014).
(1)

(2)

231
where is the total mass given by the sum of the rigid body contribution and the added
mass/inertia; is the sum of centripetal and frictional forces; is the
vector of generalized forces and moments; is the relative ship
speed, i.e. the speed though water. is the velocity vector of the current in the
frame. The transformation of coordinates between the reference frames and is achieved
through the rotational matrix

(3)

Based on the Assumptions 4 7, the output model is given as

(4)
(5)
(6)
(7)
(8)
(9)
(10)
(11)

where the subscripts m and p respectively.

Problem statement: Consider a ship sailing in calm water along a piecewise continuous path with
SOG and heading . The vessel is subject to an ocean current with speed
. Exploiting asynchronous inertial measurements of vessel motion and predictions of the

STW .

3. STW Estimator Design

Sensor measurements are noisy and their accuracy proportionally relates to cost. To overcome the
limitations introduced by each individual measurement system, in multi-sensor applications the over-
all quality of the information is improved by using sensor fusion methods. In Elmenreich (2002), sen-
sor
resulting information is in some sense better than would be possible when these sources were used
or fusion methods are the Weighted Least Squares, the
Maximum Likelihood, the Maximum Posterior, the Particle Filter, and the Kalman Filter, Castanedo
(2013). In this paper the Kalman filter is adopted because in the presence of uncertainty it provides
the optimal estimate of the quantities of interest in the sense of minimum variance.

Designing an estimator of STW based on the maneuvering model (1)-(2) requires detailed knowledge
of the vessel hydrodynamic characteristics, which in general might be unavailable or simply outdated
due to aging of the hull or hull modifications. Therefore, robustness of the estimation could be
achieved only through the parallel estimation of and model parameters. This will obviously in-
crease complexity of the estimation scheme.

An alternative approach, which conjugates both simplicity and robustness, based on only the vessel
kinematics (1) and exploiting the available measurements of ship motion and predictions of current
velocity. Since an estimate of STW is sought, rather than its projection onto the body axes, and ,
then the estimation problem can be formulated in terms of traveled distance and cruising speeds,
and .

232
Let be the state estimation vector, where is the estimate of the traveled
distance and is the estimate of the speed of the current . Then the
discrete time state space model at the core of the estimator is given by

(12)

(13)
(14)
(15)

where is the sampling time, is the total ship acceleration, i.e.

, and is a zero mean white Gaussian noise source with variance .

Remark 2: Eqs. (14)-(15) are valid only when the vessel sails with constant heading, i.e. . Dur-
ing heading alterations the equations should also account for fictitious accelerations due to centripetal
forces, Fossen (2011), [Section 8.3].

Remark 3: Based on the direct measurements (10)-(11) the indirect measurement of total ship accel-
eration is given by , which is a stochastic process whose characteristics are in general
non-Gaussian. When the acceleration components and are both zero, then will be a Rayleigh
distributed random process. However, when the acceleration along the surge and/or sway directions is
different from zero then the random component of is still well approximated by a Gaussian distri-
bution. For the sake of simplicity and in order to apply standard results in Kalman filtering, it is as-
sumed that the measured total ship acceleration can be approximated as , where is
zero mean white Gaussian noise with variance . Hence the total ship acceleration in Eqs. (12)-
(14) is given by .

Based on the output model (4)-(11) and considering asynchronous measurements the following meas-
urement models are defined

(16)
(17)

where is the measured traveled distance, is the predicted sea current speed,
and is a counter.

The measurement noise vectors and are described by inde-

pendent and uncorrelated Gaussian distributed stochastic processes, i.e. with di-
ag{[ , , ]} and with diag{[ , ]}.

Remark 4: The measurement of the travelled distance d is indirect since it is computed from the
direct GNSS measurements and . Similarly, the measurement of the speed of the current is
indirectly computed based on the predicted . Both and are in general non-Gaussian stochas-
tic process due to the nonlinear processing of the white Gaussian noise affecting the direct measure-
ments. Applying the same line of reasoning of Remark 3, the measurement models (16)-(17) are
adopted

Remark 5: Deviations of the true measurements from the proposed approximations in Remarks 3 and
4 will determine a poorer performance of the Kalman filter with respect to the theoretical expectation.

233
In particular, the estimation error covariance will be larger than the theoretical expected value and
correlation may appear in the estimation error.

Based on the estimator model (12)-(15) and measurement model (16)-(17) the discrete time state tran-
sition matrix , the input matrix , the process noise input matrix and the output matrices and
can be set up:

(18)

(19)

The last step towards the implementation of the STW estimator based on linear Kalman filtering is to
define the process noise covariance matrix . Given the estimator model (12)-(15) two process noise
sources are identified, namely and , which are assumed to be uncorrelated between each other.
Therefore, the process noise covariance matrix is given by diag{[ , ]}. Whilst the variance
of the process noise component as well as the variances of the measurement noise are given by
the noise characteristics of the individual sensors, the variance is a tuning parameter for the Kal-
man filter that can be used to determine the trade-off between smoothness of the estimation and
promptness to changes in the operational conditions of the system to perform estimation on. The line-
ar optimal STW estimator in its predictor-corrector formulation reads Lewis et al.(2017) [Section 2.3]

Time update
(20)
(21)

Measurement update

(22)
(23)

(24)
(25)

where , is the Kalman gain. During the time update the state predic-
tion and the a-priori estimation error covariance are computed based on the linear time invari-
ant discrete time model and the process noise covariance When a new measurement be-
comes available, the state estimate and the a-posteriori estimation error covariance are updated
based on the measurement model , the Kalman gain , and the measurement noise covariance

4. Implementation on Simulated Data

The simulations are carried out using the Marine Systems Simulator (MSS), Fossen (2011). The simu-
lator is a Matlab/Simulink library for marine systems that includes models for ships, underwater vehi-
cles, and floating structures, [Link] The library also contains guid-
ance, navigation, and control functionalities for real-time simulation. The library has been translated

234
to Python and then focused on the model for coupled motion of steering and rolling of a high-speed
container ship, introduced in Son and Nomoto (1981). The simulation study case represents a L = 175
m vessel, which travels with variable SOG , subject to a sea current coming from astern with con-
stant speed .

The simulation environment has been configured to run with an integration time step of 0.01 seconds,
in order to mimic the continuous time nature of the vessel motion and enable further resampling of the
data due to simulated measurement processes. To verify the ability of the designed estimator to con-
verge in mean value towards the true value as well as to track time-varying behaviors, the vessel has
been initialized with and the shaft angular velocity , which gives rise
to a non-stationary condition when is increasing or decreasing over time.
The measurement process has been set up by simulating sensors with sampling rates and noise charac-
teristics as stated in Table 1.

Signal Frequency [Hz] Noise [std]

0.033 [30s] 2.5 [m]
0.033 [30s] 2.5 [m]
0.033 [30s] 0.065 [m/s]
0.008 [20min] 0.2 [m/s]
0.033 [30s] 0.05 [m/s2]
0.033 [30s] 0.05 [m/s2]
Table 1: Sampling frequency and noise intensity (1 ) of each sensor used by the estimator

Fig.1 shows the estimate of the speed through water (virtual STW). The true STW obtained from
the simulation is indicated by the dashed-blue line. It is evident by checking the log lines that the data
is asynchronous meaning that it resembles a real-world scenario where the measurement of the current
speed is acquired with a different frequency than the SOG measurement from the GPS sensor.

Fig.1: Estimation of speed through water in comparison to other speed signals obtained in simulation

Simulated results show that the designed estimator performs well in reconstructing the STW: although
there is a noticeable phase lag, in only a few iterations it is visible that despite the noise and the asyn-
chronicity of the sensor rates, the STW estimator converges in mean value to the true value of the rel-
ative ship speed .

235
 Fig.3: Virtual STW residual within 1 , 2 and 3 confidence bounds

Fig.3: Autocorrelation function of virtual STW residuals. Lags are displayed in x-axis.

To qualify and quantify the performance of STW estimator an analysis of the estimation error
is carried out. Fig.2 illustrates the estimation error with the 1 , 2 and 3 confidence
bounds. Fig.3 shows the autocorrelation function of the estimation error.

Under ideal conditions, if the STW estimator correctly reconstructs should

be a white noise process with zero mean and bounded small variance. Figs.2 and 3 show that the re-
sidual is well within the 3 confidence bound and that the autocorrelation falls within its confidence
bound as soon as the stationary condition is achieved. Therefore, the estimation error can be con-
sidered a white sequence, meaning that the designed estimator reconstructs the dynamics of interest.

5. Implementation on full-scale data

The data originates from a 10,500 TEU container vessel owned and operated by Maersk Line. Only
relevant vessel information and values are included. Fig.4 illustrates the vessel route and speed profile
for a 1-month sample data of the vessel. It can be noticed that along the route there are periods when
the speed is switching from high to low values and vice versa. Sea currents affect the STW either pos-
itively or negatively according to their angle of attack. When , it refers to head currents.

236
 Fig.4: Vessel service route with measured speed-through-water (left) and sea current speed (right)

When measuring and analyzing data, it is desirable that the underlying measured process is wide sense
stationary, i.e. its statistical properties are time invariant. In practice, the measured process and, there-
by, the collected data will often display non-stationary features over certain periods of time. Changes
in environmental or operational conditions over time, such as variations in current speed magnitude
-
stationary data. It is possible, though, to find cases where the ship can be assumed to operate in a sta-
tionary condition.

An inspection on the whole 1-month-dataset has been conducted to ascertain how stationary the STW
sensor is overtime; always keeping in mind to maintain the mean draught in the same level and the
trim close to zero (even keel). Additionally, the sea state was loosely inferred by checking both roll
and pitch angles so as to focus on calm water conditions. Moreover, when visualizing the sea current
speed and the speed over ground , it is clear that their patterns do not always match, which is
probably due to the fact that sea currents are not accurately predicted, especially in open waters. By
checking Fig.5 one can see that the two signals have a strong match only in days 18-19.

Fig. 5: SOG vs predicted sea current speed, within the 1-month-dataset. Index refers to day number.

237
Thus, the estimator is applied to data collected while the vessel passed through the English Channel,
which refers to day 19 of the dataset. Fig.6 shows location and direction of the vessel when operating
on this specific trip. Based on the small amplitude of the roll and pitch angles it is assumed that the
vessel sailed in calm weather conditions. The is considered constant for most of
the time (the variance of heading angle is var{ } 3 deg2 in between indicated turns in Fig.6). Note-
worthy that for the selected vessel the only measurement of acceleration is not provided by an inertial
measurement unit, but it is given by processing the GNSS outputs. This introduces correlation be-
tween the measurements used in the Kalman filter, which as stated in Remark 5 potentially determines
a poorer performance of the estimator.

Fig. 6: On top the trip of interest drawn on a map. On the bottom, map is connected with the sea state
characteristics (roll, pitch on the left y-axis and true heading on the right y-axis).

Fig.7 shows the speed profiles of the "trip of interest". It is evident from the measurements that the
STW is not stationary and there are strong and changing sea currents.

238
 Fig. 7: Speed profiles of the "trip of interest"

The STW estimator has been applied to a synchronous sensor rate implementation and then to an
asynchronous one, similar to the simulation implementation where the predictions of the sea current
speed were invoked at a lower frequency than the rest of the signals. The synchronous implementation
outcome is shown in Fig.8. The estimated STW (virtual STW in the plot) shows some significant
deviations from the measured STW, up to approximately 0.5 m/s, and in general seems to better ac-
count for variations in current speed.

Fig. 8: Estimation of the STW (virtual STW) throughout the "trip of interest" based on the syn-
chronous implementation of the measurement model

In Fig.9 the asynchronous implementation of the filter is depicted, when current speed is provided
every 20 minutes, which is standard for most of the hindcast data providers. This time the estimated
STW (virtual STW in the plot) shows a more fluctuating behavior induced by the more sporadic
knowledge of the current speed. The jerky variations are related to each measurement update per-
formed by the estimator whenever a new prediction of the current speed is available. Despite the re-
duced smoothness, the estimated STW is in line with the results obtained through the synchronous
implementation both in terms of maximum deviations from the measured STW and the overall trend
of the estimate.

239
 Fig. 9: Estimation of the STW (virtual STW) throughout the "trip of interest" based on the asyn-
chronous implementation of the measurement model.

6. Discussion

Throughout the paper, the methodology, the model design, the application as well as the results, have
all been described and meticulously assessed. The outcome of the implemented filter is highly de-
pendent on the input parameters, i.e. the auto-logged vessel data. On a literature basis study, the indi-
cated total of errors can be expected up to 9% for the auto-logged data, ITTC (2005). However, it is
believed that precision errors are limited, due to the extended data processing.

The Kalman filter is a mathematical tool that conjugates the knowledge about physical phenomena
formalized in a model with measurements of quantities directly or indirectly related to that phenome-
na in order to provide estimates of unmeasured states as well as to improve the signal-to-noise ratio of
the measured quantities. Hence its performance is tightly coupled with the accuracy of the model and
the quality of the available measurements.

In order to provide a simple solution that could be used as reference for future studies, some assump-
tions have been made concerning linearity of models, Gaussianity and uncorrelation of noise sources.
However, these assumptions are not always satisfied in reality giving rise to performance deteriora-
tion of the estimation process. Accounting for correlation among the available measurements can po-
tentially improve the estimation in terms of lower estimation error covariance, however this will come
at the cost of a somewhat more complex implementation.

The STW estimator was tested on simulated data, before being applied to full-scale data. The availa-
bility of high-fidelity simulators able to produce numerical data capturing the phenomena of interest
facilitates the design and tuning of the Kalman filter, especially when the true value of the variables to
be estimated is in reality unknown. In the addressed study case a measurement of the STW was actu-
ally available but considered untrustworthy, thereby it was not included in the estimator design. De-
spite the simplifying assumptions, the evaluation of the designed STW estimator on the full-scale data
is positive and it shows the feasibility of using a pure kinematic model to estimate a STW signal,
which appears more aligned with the other inertial measurements than that one provided by the DVL.

7. Conclusion

A kinematic-based linear Kalman filter was designed to compute an estimate of the STW based on
onboard inertial measurements and external hindcast sea current measurements of a large container
vessel sailing in calm water. The analysis of the results obtained by processing full-scale measure-
ments indicates an improvement of estimated STW with respect to the measurement provided by the

240
However, these results are preliminary and an extended investigation should be conducted by e.g.
broadening the analysis to various container vessels of different characteristics, including more com-
prehensive and extended analyses made during sailing in seaway.

References

ANTOLA, M.; SOLONEN, A.; PYÖRRE, J. (2017), Notorious speed through water, 2nd HullPIC
Conf., Ulrichshusen, pp.156-165

BOOM, H.V.D.; HUISMAN, H.; MENNEN, F. (2013), New guidelines for speed/power trials: Level
playing field established for IMO EEDI, SWZ Maritime, Schip en Werf de Zee Foundation, Rotter-
dam

BOS, M. (2016), How MetOcean data can improve accuracy and reliability of vessel performance
Estimates, 1st HullPIC Conf., Pavone, pp.106-114

CAHARIJA, W.; CANDELORO, M.; PETTERSEN, K.Y.; SØRENSEN, A.J. (2012), Relative veloci-
ty control and integral LOS for path following of underactuated surface vessels, IFAC Proc. Vol.
45.27, pp.380-385

CASTANEDO, F. (2013), A review of data fusion techniques, The Scientific World J. 13

ELMENREICH, W. (2002), An introduction to sensor fusion, TU Vienna, 502

FOSSEN, T.I. (2011), Handbook of Marine Craft Hydrodynamics and Motion Control, John Wiley & Sons

FREDA, P.; ANGRISANO, A.; GAGLIONE, S.; TROISI, S. (2015), Time-differenced carrier phases
technique for precise GNSS velocity estimation, GPS Solutions 19.2, pp.335-341

GAGLIONE, S. (2015), How does a GNSS receiver estimate velocity?, Inside GNSS, pp.38-41

GRIFFITHS, G.; BRADLEY, S.E. (1998), Correlation speed log for deep waters, Sea Technology
39, pp.29-36

ITTC (2005), Final Report and Recommendations to the 24th ITTC, Manoeuvring Committee et al.,
24th ITTC, pp.137-198

ITTC (2002), Final report and recommendations to the 23rd ITTC, Propulsion Committee et al., 23rd
ITTC

LEWIS, F.L; XIE, L.; POPA, D. (2008), Optimal and robust estimation: with an introduction to sto-
chastic control theory, CRC press

LITTON (1998), Srd-500 dual axis doppler speed log. Operation Manual, Litton Marine Systems

MOE, S.; CAHARIJA, W.; PETTERSEN, K.Y.; SCHJØLBERG, I. (2014), Path following of under-
actuated marine surface vessels in the presence of unknown ocean currents, IEEE American Control
Conf., pp.3856-3861

PYÖRRE, J. (2012), Overcoming the challenges in vessel speed optimisation, HANSA 149, pp.130-135

SON, K.H.; NOMOTO, K. (1981), On the coupled motion of steering and rolling of a high speed con-
tainer ship, J. Society of Naval Architects of Japan 150, pp.232-244

TETLEY, L.; CALCUTT, D. (2007), Electronic navigation systems, Routledge, pp.45-87

241
 Using CFD to Predict Ship Resistance due to Biofouling, and Plan Hull
Maintenance
Abel Vargas, Hua Shan, Eric Holm, Naval Surface Warfare Center, Carderock Division, West
Bethesda/USA, [Link]@[Link]

Abstract

Computational fluid dynamics (CFD) was used to investigate the impact of biofouling roughness on
ship resistance, for the DTMB 5415 hull form at full scale. The ship hull form was divided into sections
to allow for the evaluation of both homogeneous and heterogeneous distributions of roughness. Total
resistance CT, and frictional resistance CF, associated with biofouling increased with hydrodynamic
roughness (as represented by equivalent sand grain roughness ks). Local values of CF varied across the
hull with higher values observed at the bow. Ranking of the hull sections in terms of CF was bow >
sides > flat bottom > stern > transom when scaled by the proportion of the wetted surface area of the
hull accounted for by the section of interest. Several scenarios associated with heterogeneous
accumulation of biofouling were investigated, including increased growth at the waterline, and
sequential cleaning of the hull. The benefits of carrying out partial cleaning of the hull (for example,
bow + sides, bow + sides + flat bottom) were observed to depend on the initial biofouling condition.
As biofouling roughness increased, cleaning of more of the hull (from bow to stern) was required to
attain a particular reduction in CT. These results demonstrate the potential use of CFD to efficiently
evaluate alternative biofouling control strategies.

This paper is provided for information only and does not constitute a commitment on behalf of the U.S. government to
provide additional information on the program and/or sale of the equipment or system.

1. Introduction

millennia,
WHOI (1952), Townsin (2003). The ability to accurately predict the impacts of biofouling, and thus
evaluate the potential benefits of biofouling control strategies is, however, still in development. Alt-
hough powering trials or towing tests have been used since the 1920s (at least) to quantify the effects
of biofouling on vessel operations, the attached communities (or other sources of drag-producing rough-
ness) would now be considered to be poorly described (if described at all), or the extent of biofouling
to be well beyond that which would currently be tolerated, e.g. Visscher (1927), Davis (1930), Izubuchi
(1934), Kempf (1937), Kan et al. (1958). More recently, Townsin et al. (1981), Lewthwaite et al. (1984);
Hundley and Tsai (1992), trials have been linked to detailed inspection of the hull, substantially im-
proving understanding of the relationships between particular types of biofouling or roughness and their
effects on vessel performance.

Although ship trials, when combined with thorough inspection of the hull, provide the best information
on the impact to ship performance of biofouling, such efforts are expensive, can require taking a ship
out of service, and are limited in the number or diversity of cases that can be investigated in a reasonably
short amount of time. Mathematical approaches may provide an efficient alternative. Schultz (2004,
2007), Schultz et al. (2011) used procedure to predict the effect of roughness (in-
cluding biofouling) on the resistance of two classes of naval vessels, (see Oliveira et al. (2018), for a
model commercial vessel). This procedure assumes that biofouling is homogeneously distributed across
the hull surface, and thus cannot be used to investigate impacts associated with heterogeneous rough-
ness distributions, such as may occur if accumulation of biofouling varies with location on the hull
either due to biological processes (for example, differential attachment of propagules, growth, or mor-
tality) or management practices, e.g. targeted hull cleaning, Hundley and Tsai (1992) or use of different
biofouling control coatings on different parts of the hull, Swain and Lund (2016). Townsin et al. (1981)
and Monty et al. (2016) presented calculation methods that allowed for heterogeneous biofouling along
the length of the hull. Townsin et al. (1981) used their method to demonstrate outsized benefits for
reducing roughness on the forward quarter of a containership, much as Hundley et al. (see Hundley and

242
Tsai (1992)) showed using targeted hull cleaning combined with detailed biofouling assessments and
powering trials.

Recently, computational fluid dynamics (CFD) methods have been applied to the hull biofouling prob-
lem, Leer-Andersen and Larsson (2003), Demirel et al. (2014, 2017). These methods offer certain ad-
vantages over those of Granville (1958), Townsin et al. (1981), and Monty et al. (2016) in that they
take into account details of the form of the hull, at the cost of being much more computationally inten-
sive. Leer-Andersen and Larssen (2003) and Demirel et al. (2014,2017) used CFD to evaluate the im-
pact of coating and biofouling roughness distributed homogeneously on a ship hull. Herein an alterna-
tive CFD formulation, to modelling the frictional resistance of a rough ship hull, is employed to exam-
ine the effects of both homogeneously-distributed biofouling and heterogeneous accumulations, such
as might be associated with differential growth of biofouling or targeted hull maintenance strategies.
The results demonstrate the potential for use of CFD to evaluate the efficacy of alternative biofouling
control strategies in terms of effects on ship performance, or to plan hull maintenance activities so as
to extract the greatest benefit.

2. Computational Fluid Dynamics

Computational fluid dynamics (CFD) was carried out in NavyFOAM, Shan et al. (2011), an integrated
CFD package based on OpenFOAM, and developed at Naval Surface Warfare Center, Carderock Di-
vision under funding by the Department of Defense High Performance Computing Program (HPCMP)
CR . Details of this solver, modifications made to the turbulence
model to account for roughness due to biofouling, and validation of the approach, can be found in
Vargas and Shan (2017). Within the turbulence model, biofouling roughness is represented by the
equivalent sandgrain roughness height, ks. In the simulations described below, the ks values used were
obtained from Schultz (2007), although additional values associated with various types and configura-
tions of biofouling have been determined experimentally, e.g. Monty et al. (2016), Murphy et al. (2018),
and can be readily employed in the simulations.

2.1. Computational Domain: DTMB 5415

The DTMB 5415 hull form was chosen to carry out CFD at full scale. This hull form resembles that of
a US Navy surface combatant. Details of the hull form (see Olivieri et al. (2001), Larsson et al. (2014)
include the following: length between perpendiculars (Lpp) = 142 m; wetted surface area = 2976.2 m2;
and Froude Number (Fr) = 0.28. The hull was tested at a draft of 6.15 m with a fixed sinkage and trim
of -1.82 x 10-3Lpp and -0.108°, respectively, and without any appendages. Fluid properties of sea water
included density = 1026.02 kg m-3, and kinematic viscosity = 1.1892 m2/s. After carrying out a study
of grid sensitivity, a grid size of 10 x 106 cells was chosen. The computational domain, discretization
methods, and numerical schemes for all simulations were as described previously for model-scale eval-
uations, Vargas and Shan (2017).

3. Results

3.1 Homogeneous Biofouling

As a first step in evaluating the impact of biofouling on resistance, homogeneous biofouling roughness
was added to the hull form. For small values of biofouling roughness ks less than 1000 m, such as
those associated with slime or biofilm biofouling, the total resistance coefficient CT increases rapidly,
Fig.1, Table 1. As larger or rougher biofouling (ks > 1000 m) is added to the hull, the slope of the
relationship between ks and resistance decreases, Fig.1, Table 1. Numerical simulation allows for the
estimation not only of the total resistance of a rough ship hull (RT), but also components of the total
resistance associated with frictional (RF) and residuary resistance including pressure and wave re-
sistance (RR) (Vargas and Shan 2017). The force components can be written in dimensionless coeffi-
cients defined as follows:

[1]
243

[2]
w U is the ship speed, and S is the wetted surface area of the hull. In the
current computations, appendage drag and still air drag are not added to CT, and only CF and CR are
used to compute the total resistance (CT = CF + CR).

Residuary resistance CR is relatively unaffected by the presence of biofouling roughness on the hull.
Instead, variation in CT is driven by increases in frictional resistance CF with ks, Fig.1. Based on bio-
fouling conditions and associated values of ks from Schultz (2007), accumulation of bacterial biofilms
(light slime, heavy slime) results in increases in CT of 15%-30% over the hydraulically smooth condi-
tion, and 10%-25% over the as painted condition, Table 1. The worst case investigated by Schultz
(2007), heavy calcareous biofouling, increases CT by approximately 93% relative to the hydraulically
smooth condition, and 85% relative to the as painted condition, Table 1.

Fig.1: Change in resistance with biofouling roughness ks, for homogeneously-distributed biofouling
roughness. CT = total resistance coefficient; CF = frictional resistance coefficient; CR = pressure
resistance coefficient.

244
Table 1: Impact of roughness on total resistance CT, relative to a hydraulically smooth hull and a hull
painted with a typical antifouling coating
Biofouling/Roughness % Change in CT % Change in CT
Condition Smooth Hull as Reference Painted Hull as Reference
Hydraulically-smooth hull - -
Typical as-applied antifouling
4.35 -
coating
Light slime 15.53 10.72
Heavy slime 30.86 25.41
Small calcareous biofouling 45.36 39.31
Medium calcareous biofouling 64.36 57.51
Heavy calcareous biofouling 92.82 84.78

Fig.2: Distribution of local skin friction values for homogeneous biofouling roughness of varying ks,
for a ship speed of 20 kn.

245
Fig.3: Distribution of local skin friction values for homogeneous biofouling roughness of varying ks,
for a ship speed of 20 kn. CF is rescaled from Figure 2 in order to visualize spatial variation in
skin friction for larger values of ks.

Fig.4: Division of the DTMB 5415 hull form into sections

Values of local frictional resistance (Cf) defined as:

where w is the wall shear stress, is the fluid density, and U is the ship speed, as associated with
biofouling roughness, are heterogeneously-distributed across the hull, Figs.2 and 3. The contours indi-
cate that various sections of the hull contribute differentially to the overall impact of biofouling rough-
ness on resistance. This phenomenon has been demonstrated previously in ship trials using sequential
hull cleaning (for example, Hundley and Tsai (1992), and computationally by Townsin et al. (1981),
but CFD provides the opportunity to evaluate the nature and magnitude of these effects for any case of
biofouling roughness which can be represented by ks (as opposed to the specific cases associated with
ship trials, for example).

Unlike a flat plate where high skin friction is concentrated at the leading edge and then monotonically
decreasing along the streamwise direction, Vargas and Shan (2016), the curvature of the hull changes
the local velocity thus the wall shear stress. This change in wall shear stress is reflected in the skin
friction contours in Figs.2 and 3, where changes in Cf not only occur in the streamwise direction, but
also from the waterline to the keel. Skin friction is higher at the bow for all roughness conditions, but
as ks increases, regions of high frictional resistance extend further aft and beyond midships, Fig.2 and
3. Minimal change in skin friction is observed near the stern section, even at high values of ks.

246
The impact of biofouling roughness on frictional resistance (CF), for individual ship hull sections, was
investigated by dividing the hull into parts or sections (bow, sides, flat bottom, stern, transom) corre-
sponding to the inspection plan used by divers to evaluate the biofouling condition of in-service US
Navy vessels, Fig.4. The hull sections were created by assigning the faces of each grid cell that were
inside a specified bounding box which intersected the hull into the corresponding section. The bounding
boxes were defined by hull locations defined in diver inspection plans. The dividing of the hull was
carried out during the pre-processing phase of the CFD setup. The resulting hull sections accounted for
the following percentage of the total wetted surface area of the hull: bow 16%; sides 42.9%; bottom
18%; stern 22.9%; and transom 0.2%.

Fig.5: Frictional resistance (as CF) associated with sections of the DTMB 5415 hull form, for varying
levels of biofouling roughness (ks). A. Absolute values of CF for each section. B. Values of CF
scaled by the proportion of the hull wetted surface area accounted for by each section.

247
Frictional resistance generated by biofouling roughness on the sides of the hull is much larger than that
for the other hull sections, while biofouling of the transom generates no, or very little, frictional re-
sistance at all, Fig.5A. Biofouling of the bow, flat bottom, and stern produce comparable absolute val-
ues of CF at a given level of ks, Fig.5A. If, however, frictional resistance is scaled by the proportion of
the wetted surface area of the hull accounted for by the section of interest, Fig.5B, the importance of
biofouling of the bow to ship-wide frictional resistance stands out. On a per unit area basis, biofouling
pact on resistance than any of the other hull sections. As well, the
rate of increase in CF from one biofouling condition to another is greater for the bow than any other
section; a given change in ks results in a greater change in CF for the bow than the other hull sections
(Figure 5B). The rate of change in CF with ks for the sides and flat bottom are comparable once biofoul-
ing progresses beyond the light slime condition. These scaled results match predictions which may be
drawn from the contour plots of local values of frictional resistance, Figs.2 and 3. Regardless of the
biofouling condition (as represented by ks), the ranking of the hull sections in terms of CF remains bow
> sides > flat bottom > stern > transom, on a per unit area basis, Fig.5B.

3.2 Heterogeneous Biofouling Case Study Effect of Biofouling at the Waterline

The ability to divide the hull form into sections further enables the use of CFD to investigate the impact
on total resistance of heterogeneous accumulation of biofouling roughness. In order to demonstrate this
capability, an evaluation of the effects of biofouling at the waterline was undertaken. Accumulation of

factors including the availability of light (affecting growth of macroalgae) and the presence of degraded
or ineffective antifouling coatings (as a result of abrasion or other types of physical damage, or the
harsh environment represented by the air-water interface). A ks value equivalent to light slime (ks =100
m) was applied homogeneously to the DTMB 5415 hull form. Biofouling of varying roughness was

approximately 23% of the total wetted hull area. Total resistance CT was estimated, and compared to
corresponding values for biofouling distributed homogeneously across the hull (Figure 6). Results in-
dicate that increasing roughness at the waterline from a ks value equivalent to light slime to one equiv-
alent to medium calcareous fouling increases ship-wide CT by 10%, to a level comparable to that for a
hull supporting homogeneous biofouling roughness equivalent to heavy slime (Figure 6).

Fig.6: Effect of biofouling roughness (as represented by ks) on the total resistance coefficient CT, for
biofouling at the waterline and over the entire hull.

248
3.3 Heterogeneous Biofouling Case Study Effect of Cleaning Individual Hull Sections

The divided hull form and CFD can also be used to evaluate the benefits of hull cleaning. In this case,
a ks value equivalent to medium calcareous fouling (ks =3,000 m) was applied homogeneously to the
DTMB 5415 hull form. The impact of cleaning the bow, sides, flat bottom, or stern was then quantified
by reducing the ks value for that section to that for light slime, while keeping biofouling roughness on
all other sections constant.
to frictional resistance when biofouling roughness is homogeneously distributed across the hull. Clean-
ing of the sides produces the greatest percentage reduction in CT, Fig.7. When scaled, however, by the
proportion of wetted surface accounted for by the cleaned section, cleaning of the bow results in the
greatest reduction in resistance. Hundley (see Hundley and Tsai 1992) demonstrated a similar effect
using ship powering trials paired with partial/sequential cleaning of the underwater hull, and Townsin
et al. (1981) by computation.

Fig.7: Effect of cleaning of individual hull sections on the percentage change (reduction) in total re-
sistance CT, and percentage change in CT scaled by the proportion of the hull wetted surface area
accounted for by each section. Cleaned hull sections are assumed to support biofouling by light
slime.

3.4 Heterogeneous Biofouling Case Study Effect of Sequential Cleaning of the Hull

The divided hull form and CFD were further employed to investigate the benefits, in terms of reduced
resistance, of cleaning multiple sections of the hull in sequence. The change in resistance with the
cleaning of additional hull sections was quantified for three values of ks, corresponding to small (ks
=1,000 m), medium (ks =3,000 m), and heavy calcareous fouling (ks =10,000 m), applied homoge-
neously to the DTMB 5415 hull form. As with the case for cleaning of individual hull sections (see
Section 3.3 above), hull cleaning was assumed to reduce ks on the affected sections to that representative
of light slime. For all biofouling conditions, total resistance CT decreased with the cleaning of additional
hull sections, Fig.8A. When scaled by the proportion of wetted hull surface area cleaned, however, the
impact of cleaning on CT decreased with each additional section, Fig.8B, reflecting the distribution of
local skin friction observed previously, Fig.2 and 3. The greatest reduction in CT (on a per unit area
basis) was observed with cleaning of the bow, and no change in this proportional reduction was appar-
ent beyond cleaning the bow and sides, Fig.8B. The pattern of change in CT itself, with sequential

249
cleaning of hull sections, suggests that the pattern of benefits accruing from hull cleaning depends on
the initial biofouling condition, Fig.9. For hulls fouled by small or medium calcareous fouling, cleaning
of the bow and sides reduces CT to a value equivalent to that of a hull supporting homogeneous coverage
of heavy slime. For hulls fouled with heavy calcareous fouling this level of total resistance can only be
attained after cleaning of the bow, sides, and stern, Fig.9.

4. Discussion

Previously, full-scale predictions of ship performance penalties associated with the accumulation of
biofouling have been generated from limited ship powering trials, e.g. Hundley and Tsai (1992), or
have been calculated using boundary layer similarity law methods, e.g. Schultz (2007), Schultz et al.
(2011), Oliveira et al. (2018). Computational fluid dynamics methods, e.g. Demirel et al. (2017), offer
an alternative approach which allows the details of both the hull form, and the distribution of biofouling
roughness across the hull, to be taken into account. Together, these features may allow for more
accurate estimation of performance penalties. The ability to represent spatial variation in biofouling
growth may be particularly important. Results presented above demonstrate that local values of
frictional resistance differ across the hull, even for hydrodynamically-smooth hulls. These effects may
be amplified, or diminished, if distribution of biofouling roughness is spatially correlated with the ship
hull features that contribute to the variation in local skin friction in the smooth condition for example,
if more hydrodynamically-rough biofouling were to preferentially develop on the bow. Under such
scenarios, representation of the biofouling on the entire hull by a simple mean roughness value could
result in either under- or over-estimation of performance penalties. Methods such as that of Monty et
al. (2016) may to some extent mitigate this problem.

The ability of CFD to account for spatial variation in roughness enables the evaluation of impacts to
ship performance of either natural processes or control practices that generate heterogeneity in
occurrence of biofouling (for example, see Sections 3.2-3.4 above). Increased biofouling of the
waterline is a common problem that may be amenable to control through targeted in-water hull
cleaning. Simulation of waterline biofouling for the DTMB 5415 hull form suggests that accumulation
of medium calcareous fouling in this zone increases ship-wide CT by 10%. The potential benefit of
reducing this penalty by targeted cleaning can be estimated by comparing the cost of the cleaning effort
to the value of the fuel saved through reduction in the resistance associated with the biofouling. Such
an analysis may not be possible using similarity law methods that base their calculations on
homogeneous roughness. Simulations of sequential hull cleaning indicate that full hull cleaning may
not be required to recover acceptable vessel performance such as that associated with typical biofilm
growth. CFD, however, also shows that the details of this result depend on the initial level of biofouling;
the presence of greater biofouling roughness requires more of the hull (from bow to stern) to be cleaned
to attain a given reduction in resistance.

Comparable results to those described above have been obtained from ship powering trials, e.g. Hundley
and Tsai (1992). Computational fluid dynamics, however, allows the simulation of a greater range of
biofouling or maintenance scenarios without the expense of ship trials, the need to wait for a particular
configuration of biofouling to develop on the hull, or the need to take a vessel out of service. Thus,
CFD can form the basis for more efficient evaluation of the costs and benefits of alternative, novel,
biofouling control strategies.

Acknowledgements

This research was funded by the Naval Surface Warfare Center Carderock Division as part of the Naval
Innovative Science and Engineering (NISE) program under the direction of K. Michalis. Many thanks
to J. Gorski, P. Chang, and S. Aram for providing useful insights and technical support. The rest of the
-associated Roughness and

250
Fig.8: Effect of sequential cleaning of the hull on change in total resistance CT. A. Percentage change
(reduction) in CT with cleaning. B. Percentage change (reduction) in CT with cleaning scaled by
the proportion of the hull wetted surface area accounted for by the cleaned section(s). Results
plotted for three different levels of biofouling roughness. Cleaned hull sections are assumed to
support biofouling by light slime.

251
Fig.9: Effect of sequential cleaning of the hull on change in total resistance CT, for three different levels
of biofouling roughness. Cleaned hull sections are assumed to support biofouling by light slime.

References

DAVIS, H.F.D. (1930), The increase in S.H.P. and R.P.M. due to fouling, J. American Society of Naval
Engineers 42, pp.155-166
DEMIREL, Y.K.; KHORASANCHI, M.; TURAN, O.; INCECIK, A.; SCHULTZ M.P. (2014), A CFD
model for the frictional resistance prediction of antifouling coatings, Ocean Engineering 89, pp.21-31
DEMIREL, Y.K.; TURAN, O.; INCECIK, A. (2017), Predicting the effect of biofouling on ship
resistance using CFD, Applied Ocean Research 62, pp.100-118
GRANVILLE, P. S. (1958), The frictional resistance and turbulent boundary layer of rough surfaces,
J. Ship Research 2, pp.52-74
HUNDLEY, L.L.; TSAI, S.J. (1992), The use of propulsion shaft torque and speed measurements to
improve the life cycle performance of U.S. naval ships, Naval Engineers Journal 104(6), pp.43-57
IZUBUCHI, T. (1934), Increase in hull resistance through ship bottom fouling, Zosen Kiokai 55
KAN, S.; SHIBA, H.; TSUCHIDA, K.; YOKOO, K. (1958), Effect of fouling of a ship's hull and
propeller upon propulsive performance, International Shipbuilding Progress 5, pp.15-34
KEMPF, G. (1937), On the effect of roughness on the resistance of ships, Trans. INA 79, pp.109-119
LARSSON, L.; STERN, F.; VISONNEAU, M. (2014), Numerical ship hydrodynamics: an assessment
of the Gothenburg 2010 workshop, Springer
LEER-ANDERSEN, M.; LARSSON, L. (2003), An experimental/numerical approach for evaluating
skin friction on full-scale ships with surface roughness, J. Marine Science and Technology 8, pp.26-36
LEWTHWAITE, J.C.; MOLLAND, A.F.; THOMAS, K.W. (1984), An investigation into the variation
of ship skin frictional resistance with fouling, Trans. RINA 127, pp.269-284
MONTY, J.P.; DOGAN, E.; HANSON, R.; SCARDINO, A.J.; GANAPATHISUBRAMANI, B.;
HUTCHINS, N. (2016), An assessment of the ship drag penalty arising from light calcareous tubeworm

252
fouling, Biofouling 32, pp.451-464
MURPHY, A.K.; BARROS, J.M.; SCHULTZ, M.P.; FLACK, K.A.; STEPPE, C.N.; REIDENBACH,
M.A. (2018), Roughness effects of diatomaceous slime fouling on turbulent boundary layer
hydrodynamics, Biofouling 34, pp.976-988
OLIVEIRA, D.; LARSSON, A.I.; GRANHAG, L. (2018), Effect of ship hull form on the resistance
penalty from biofouling, Biofouling 34, pp. 262-272
OLIVIERI, A.; PISTANI, F.; AVANZINI, A.; STERN, F.; PENNA, R. (2001), Towing tank
experiments of resistance, sinkage and trim, boundary layer, wake, and free surface flow around a
naval combatant INSEAN 2340 model, IIHR Technical Report 421
SCHULTZ, M.P. (2004), Frictional resistance of antifouling coating systems, Journal of Fluids
Engineering 126, pp.1039-1047
SCHULTZ, M.P. (2007), Effects of coating roughness and biofouling on ship resistance and powering,
Biofouling 23, pp.331-341
SCHULTZ, M.P.; BENDICK, J.A.; HOLM, E.R.; HERTEL, W.M. (2011), Economic impact of
biofouling on a naval surface ship, Biofouling 27, pp.87-98
SHAN, H.; DELANEY, K.; KIM, S.E.; RHEE, B.; GORSKI, J.; EBERT, M. (2011), Guide to
NavyFOAM V1.0, NSWCCD-50-TR-2011/025, Naval Surface Warfare Center Carderock Division,
West Bethesda
SWAIN, G.; LUND, G. (2016), Dry-dock inspection methods for improved fouling control coating
performance, J. Ship Production and Design 32, pp.186-193
TOWNSIN, R.L. (2003), The ship hull fouling penalty, Biofouling 19(Supplement), pp.9-15
TOWNSIN, R.L.; BYRNE, D.; SVENSEN, T.E.; MILNE, A. (1981), Estimating the technical and
economic penalties of hull and propeller roughness, SNAME Trans. 89, pp.295-318
VARGAS, A.; SHAN, H. (2016), A numerical approach for modeling roughness for marine
applications, ASME 2016 Fluids Eng. Division Summer Meeting, Washington DC

VARGAS, A.; SHAN, H. (2017), Modeling of ship resistance as a function of biofouling type,
coverage, and spatial variation. 2nd HullPIC, Ulrichshusen

VISSCHER, J. P. (1927), Nature and extent of fouling of ships' bottoms, Bulletin of the US Bureau of
Fisheries 43, pp.193-252
WHOI (1952), Marine fouling and its prevention, Woods Hole Oceanographic Institution, US Naval
Institute, Annapolis, MD, USA

253
 Reducing the Resistance of Container Ships with a
FLUME® Roll Damping Tank
Bastian Marquardt, Alessandro Castagna, Amirhossein Rahimi, Aaqib Gulzar Khan,
Johanna Marie Daniel, Christian Voosen, Nikhil Mathew, Hoppe Marine, Hamburg/Germany,
[Link]@[Link]

Abstract

Hoppe Marine GmbH, an international specialist for anti-rolling tank systems having over 60 years
of expertise, designs and outfits all kind of vessels with passive damping systems. The company is
taking part in the Joint Research Project HERMes (Improvement of the Harmonic Excitation Roll
Motion Procedure). This project is cooperation with the Hamburg Ship Model Basin (HSVA), the
Hamburg University of Technology and the Peter Döhle Schiffahrts-KG. The project is founded by the
Federal Ministry for Economic Affairs and Energy. Within the framework of this project, numerous
model tests and CFD simulations have been carried out for a modern 9,000 TEU container vessel
design with and without the passive roll damping system. Additionally full-scale measurements on a
5000 TEU container vessel of Peter Döhle Schiffahrts-KG are still ongoing. The aim of this project is
the investigation of roll damping with respect to different roll damping devices and their influence on
the roll motion behavior as well as resistance of the ship hull. New possibilities to quantify benefits of
having a passive roll damping system aboard of modern vessel have been used and developed. This
paper focusses on the investigation of influences of a passive roll damping tank on the reduction of
ship resistance and consequently the possible fuel savings. The advantage of a passive roll damping
tank over conventional bilge keel setups will be shown through the example of a modern container
vessel.

1. Introduction

Passive roll damping systems are distributed by Hoppe Marine under the brand name FLUME® Roll
Stabilization Systems. The main purpose of installing a FLUME® Roll Stabilization System is the
reduction of roll motion and the associated accelerations aboard a vessel. This is consequentially
leading to more crew comfort & safety and a reduction in forces, increasing the cargo capacity and
reducing the risk of cargo damage and loss. However, the reduced roll motions due to the FLUME®
tank also have a direct influence on the resistance of the vessel and this influence is quantified in this
paper. Beforehand a short explanation about the different damping possibilities and the functionality
of a FLUME® tank is given.

Generally, the natural damping of a vessel for the roll degree of freedom is low and thereby
necessitating the use of additional equipment to enhance the roll damping. The most common way to
achieve the additional roll damping is the use of bilge keels. However, the biggest disadvantage of the
bilge keels is that they have a negative effect on the resistance of the vessel and usually offer a
marginal roll reduction of 5-20% only. Active devices such as fin stabilizers can be used to achieve a
much higher damping, but they are quite expensive in purchase, operation and maintenance, making
them irrelevant for container vessels or commercial shipping in general. Since, the FLUME® Roll
Stabilization System is passive and has no moving parts, no maintenance and operational costs occur.
Moreover, a FLUME® Roll Stabilization System does not increase the resistance and at the same
time there is a significantly higher reduction in roll of the vessel. This roll reduction is usually 30-
70% in resonance and depends on the design and the operational conditions of the vessel. Parametric
roll can be even completely eliminated.

2. FLUME® Roll Stabilization System

The FLUME® Roll Stabilization System is essentially a tank running across the beam of the ship
containing ballast water that can flow from one side to the other, driven by the roll motion of the

254
vessel itself. The damping effect is based on the existence of a travelling wave that has a phase lag
with respect to the roll motion of the vessel, providing a counteracting moment. Ideally the water
motion has a delay of a quarter cycle behind the roll motion of the ship, which itself is delayed about a
quarter cycle behind the wave excitation at resonance, Journée and Massie (2001). Thus, the liquid in
the tank will directly oppose the upward buoyant force of the wave. This is reducing the roll motion as
shown schematically in Fig.1.

Fig.1: FLUME® tank working principle

The liquid level inside the tank can be changed in order to tune the tank to the different loading
conditions of the ship. The internal wave is travelling faster with higher water levels, thus a tank must
be designed and operated in a way that the water depth and consequently travelling wave speed is in
accordance with the roll period of the vessel. This ensures that the roll resonance period of the tank is
the same as that of the ship for the optimum damping effect. In theory a FLUME® tank can be
installed anywhere on the vessel. However, there are some optimum positions that minimize the
impact on the cargo intake and the general arrangement of the vessel thereby making the installation
more economic and efficient. These optimum positions are shown in red in Fig.2. FLUME® tanks
have been installed or investigated aboard different container vessels in all these positions by Hoppe
Marine. The tanks can be installed for a newbuilding and can also be retrofitted.

FLUME® tank

Fig.2: Optimum positions for Fig.3: FLUME Tank on a modern Container Ship
installation

Fig.3 shows a practical FLUME® Roll Stabilization System installation aboard a modern 15500 TEU
container vessel. Hoppe Marine has designed and installed systems on 52 modern container vessels
varying from 2100 TEU to 20000 TEU in size. The FLUME® tanks have proven to be a very
effective solution especially for ULCVs with a large variation in GM values in many operational
scenarios.

3. Influence of FLUME Tank on Ship Resistance

The main purpose of a FLUME® tank on container vessels is to reduce the roll motions in order to
increase the vessel capacity, comfort and safety. However, this reduction in roll motion reduces the

255
resistance of the vessel as well. The reduction in roll and resistance of the 9000 TEU container vessel
that was especially designed for the research project was quantified in a specialized model test
program. These model tests were carried out by one of the partners of the joint research project, the
Hamburg Ship Model Basin (HSVA). The model was self-propelled and equipped with bilge keels
along with the FLUME® tank. Four different bilge keel sizes were tested: 0 mm (no bilge keel), 200
mm, 400 mm and 800 mm. Within this paper, only the results of no bilge keels and 400 mm bilge
keels in combination with the FLUME® tank are presented. The ship model installed with the
FLUME® tank in the area of the deckhouse is shown in Fig.4.

Fig.4: Model of the 9000 TEU Container Vessel

The model tests conducted are different from usual seakeeping model tests in the sense that the ship
model was not excited by regular or irregular waves. Instead a gyroscopic excitation machine that
oscillates the model at a given frequency and amplitude has been used for these tests. The water
surface is flat and undisturbed prior the tests. The application itself is called HERM test and was
developed as a part of the research project in a cooperation between HSVA and Hamburg University
of Technology (TUHH). The advantage of using the HERM test rather than tests in regular or
irregular waves is that the roll degree of freedom can be studied independently without the influence
of other degrees of freedom. Additionally this procedure offers the possi bility to repeat tests with
almost absolutely identical environmental and physical conditions, offering a great opportunity to
compare different scenarios as e.g. the influence of different bilge keel sizes.

In this model test program different excitation moments are generated corresponding to different wave
heights at 4 different test speeds (0.0, 7.0, 14.0 and 21.0 kn). The vessel was tested always at the
design draft of 13.0 m and with two different GM values of 2.0 m and 7.0 m. The results presented
here correspond to a stability of 7.0 m and a ship speed of 14.0 kn. The excitation period was always
set to the natural roll period of the vessel leading to resonant roll only.

The roll angle is a direct result from the measurements and is not subject to any scaling laws, why it
can be directly used for the results of the full scaled ship. The roll damping coefficients are estimated
based on the conservation of energy approach over one roll period, Handschel and Abdel-Maksoud
(2015), but it is not further important for the results shown in this paper and therefore only mentioned
for completeness. The delivered power is calculated by the mean over one steady excitation level
(constant roll amplitude) and extrapolated to full scale in order to correct the overestimated friction
from the model. The main problem for the precise transmission of the power to full scale is the
different propulsion point. Therefore, from the Froude extrapolated thrust difference between the
HERM test and calm water test the additional resistance is calculated taking the thrust coefficient into
account. The full scale delivered power is estimated using the open water diagram from the 9000 TEU
container vessel estimated from separately conducted propulsion tests, Daniel (2018).

The estimation of the corresponding wave height from the HERM test excitation moment is based on
the calculation of an effective wave slope (EWS). This method was already developed by William

256
Froude and is fairly valid for longer beam sea waves, Tupper (2013). The EWS method is seen to be
sufficient to give a good correlation between the moments generated from the HERM application and
a corresponding regular beam sea wave.

The results from the model tests that are presented here essentially focus on the comparison of the
Fig.5 shows the relationship between roll
amplitude and wave height, Fig.6 shows the relationship between delivered power and roll amplitude,
and Fig.7 shows the relationship between delivered power and wave height, all with and without the
FLUME® tank at a speed of 14 kn.

From these results the influence of the FLUME® tank becomes clearly visible. Fig.5 is indicating that
the reduction in roll angles for the smaller wave heights (< 3.0 m) is more than 80%. This reduction is
around 60% for higher wave heights and is still 40% for largest wave heights of 6.5 m. Fig.6 shows
the delivered power as a result of the roll angle of the ship. There is no difference with and without
the FLUME® tank. This is because for a given value of roll amplitude, the amount of delivered power
is dependent on the resistance created by the hull only. If that certain roll amplitude is occurring with
a roll damping tank being present or not has no influence on the roll motion pattern. The purpose of
the FLUME® tank is to reduce the roll motion, but when the roll motion is fixed, the delivered power
shall remain unchanged for a given draught. During all tests the draught of the ship was same, even
when the FLUME® tank was not active, an equivalent weight was installed on the model to have
always the save immersion of the hull for better comparison.

If Figs.5 and 6 are compared as done in Fig.7, it can be seen that the resistance of a rolling ship can be
reduced significantly, when it is equipped with a FLUME Tank. For the regular waves, at a moderate
wave height of 3.0 m the FLUME® tank reduces the delivered power by about 30%.

Fig.5: Roll Amplitude vs Wave Height at 14 kn Fig.6: Delivered Power vs Roll Amplitude at
14 kn

Fig.7: Delivered Power vs Wave Height at 14 kn

Another major task of the research project was to get these results confirmed in reality, by
investigating on board measurements taken from the M/V JOGELA (Loa=255.40m; 5000 TEU).

257
Unfortunately the measurements have been partly corrupted for some data during the 2 years of
measurement, thus no clear picture could be yet concluded due to the leak of data. Hoppe Marine is
now constantly collecting data and woks on a concept to prove the benefits also for a full-scale vessel.

4. Comparison of the Influence of Bilge Keels and FLUME Tank on Ship Resistance

The results from the HERM model tests indicated that the FLUME® tank reduce the roll motion of
the ship and at the same this reduced roll motion leads to a decrease in the resistance of the vessel.
The common method to reduce the roll motion of the ship is the installation of bilge keels. Bilge keels
dampen the roll motion of the ship to a certain extent, but at the same time they increase the overall
resistance of the ship.

It is not possible to achieve a similar amount of roll damping with bilge keels as that of a FLUME®
tank, unless their size is huge which would mean that they increase the ship resistance drastically.
When the FLUME® tank is installed aboard a vessel there is an additional load on the vessel which
increases the draught slightly. This increase in the draught shall also increase the resistance of the
vessel and thus it becomes imperative to analyze the phenomenon.

A full-scale Computational Fluid Dynamics (CFD) analysis was done by the Hamburg University of
Technology in order to evaluate the additional resistance due to the bilge keel and the FLUME® tank.
The analysis was done for the 9,000 TEU container vessel at a speed of 21 kn and a design draught of
13 m. The presence of the FLUME® tank increased the draught of the ship by 0.1 m. The additional
immersion due to the weight of the tank was considered as a parallel immersion. Here it is not
considered, that in reality in many cases the additional water in the FLUME® tank compensates some
of the ballast water that would be used otherwise.

The results are shown in Figs.8 and 9. Fig.8 shows the variation of delivered power with different trim
angles. The ship has a length of 312.0 m, therefore a trim angle of 0.25° means 1.5 m trim forward.
From this result it is determined that on average for all the considered trim angles, the 400 mm bilge
keels requires an additional power of 0.9 %. Even though the FLUME® tank generates an additional
immersion, it still requires 0.3 % less power than the bilge keels, for the case that the vessel is always
sailing with 21.0 kn. Fig.9 shows the variation of delivered power over draught at 21 kn. It can be
seen that the additional resistance offered by the 400 mm bilge keels remains almost independent of
the draught. It is assumed that this will be also the case for the other trim angles that might occur.

Fig.8: Resistance of a 9000 TEU vessel at 21 kn Fig.9: Resistance of a 9000 TEU vessel at 21 kn
with & without bilge keels and FLUME® with and without bilge keels at different
tank at different trim draught

5. Commercial Evaluation: Reduced Fuel Consumption due to Reduced Roll Motions

In order to quantify the fuel saving potential of a FLUME® tank two things need to be considered.
The magnitude a vessel is rolling at sea during its entire operation profile and the roll reduction that

258
can be achieved with a FLUME® roll damping tank for all these occurring roll scenarios. Fig.10
shows the evaluation of the quantity of rolling of M/V JOGELA for different roll RMS windows of
0.5° during 1.5 years of operation. It can be seen that the vessel is facing significant roll (RMS below
0.5°) almost on a daily basis. Error! Reference source not found. as well gives for different
windows of roll RMS the probability of occurrence. These windows are defined according to the roll
reductions that can be expected for the different occurring magnitudes of roll, in case a FLUME®
tank is present. The possible power savings have been estimated from the results of the HERM model
tests, Fig.6, for these different windows. Multiplying these numbers for the possible savings with the
probability of occurrence of each of the defined windows leads to the final power savings the vessel
could have, in case a FLUME® tank is installed. The sum of these savings gives the overall saving of
1.4%, but the additional resistance due to increase of deadweight is not yet considered.

Fig.10: Distribution of RMS roll angles of the 5000 TEU CV M/V JOGELA
Table 1: Distribution of roll angle occurrence and according savings at these roll scenarios due to
FLUME® tank
RMS probability of Roll Red. Possible Power Savings Power
occurrence Savings (considering occurrence)
0.0°-0.5° 43% 0% 0.0% 0.0%
0.5°-1.5° 46% 40% 1.4% 0.9%
1.5°-3.0° 10% 60% 4.0% 0.4%
>3.0° 1% 30% 5.1% 0.1%
Avg. Savings 1.4%

The fuel costs of a modern 9000TEU Container vessel are assumed to be approximately 16 Mio USD
per year. Consequently the 1.4% less engine power required lead to savings of 224,000.00 USD.
Considering at the same time the resistance increases due to an increase in deadweight for all cases
the vessel is sailing about 0.6%, because the FLUME® tank is constantly in operation, still 0.8% less
engine power are required and the savings are still 128,000.00 USD.

Other potential savings result from the absence of bilge keels presented in section 4. In case an
innovative vessel design makes use of a roll damping tank only and dispense on bilge keels the
required power is further reduced. In the following commercial evaluation the additional savings are
only considered for the 43% of operation time (see Table 1) where the vessel was rolling with an
RMS value below 0.5°. As the absence of the bilge keels has still a positive influence on the power
for the significantly rolling vessel this is a conservative approach, and will lead to a reduction of 1.7%
in case the draft increase due to the FLUME® tank is considered throughout the entire operation time
of the vessel. Assuming that the additional weight of the FLUME® tank is compensated by the
absence of other ballast water, even 2.3% power savings are possible.

259
It is uncontested that these numbers vary for different vessel sizes and operation profiles, but it shows
that an innovative roll damping concept, making use of a FLUME® tank instead of bilge keels offers
vessel owners the possibility to save between 272,000.00 USD and 368,000.00 USD per year for the
example given of a modern 9000 TEU container vessel.

6. Other FLUME Tank Benefits

Although these investigations show that the installation of a roll damping tank has a positive effect on
the fuel consumption of a vessel, it is only one of many positive effects leading to a much more

container vessel are:

Commercial benefits:
Additional container intake (2%-15% per bound, depending on container weight, ship size
and route)
Fewer restows (50%-75% in first harbor of west bound, other scenarios were not investigated)
Reduced fuel consumption and reduced exhaust emissions per container transported
Reduction of cargo damage and insurance claims
Optimization of route planning

Comfort & Safety benefits:

Reduced risk of losing cargo
Elimination of parametric roll scenarios
Increased crew comfort

The installation of a FLUME® tank leads to a more efficient and environment friendly vessel.
Investigations have shown that the ROI (Return of Investment) is less than a year for many vessels,
even if the tank is retrofitted.

7. Classification

For the benefit of fuel saving the classification society does not need to accept the FLUME® tank in
the vessel approval. However, the most valuable benefit for container vessels is the cargo boost
possibility resulting from reduced lashing forces. Some classification societies have a provision in the
calculation of container lashing forces. There is a reduction factor ( _ ) multiplied to the
maximum roll angle similar to the coefficient used for bilge keels ( _ ) The value of _ is
determined performing hydrodynamic calculations on the tanks using model tests or CFD followed by
scatter data evaluations.

References

ABDEL-MAKSOUD, M. (2010), Seakeeping of Ships, Lecture Notes, TU Hamburg

DANIEL, J.M. (2018), Influence of a FLUME tank on the roll damping of a container ship in
interaction with different sized bilge keels, TU Hamburg

HANDSCHEL, S.; ABDEL-MAKSOUD, M. (2015), Improvement of the Harmonic Excited Roll

Motion Technique for Estimating Roll Damping, TU Hamburg

JOURNEE, J.; MASSIE, W. (2001), Offshore Hydromechanics, TU Delft

TUPPER, E.C. (2013), Introduction to Naval Architecture, Butterworth & Heinemann

260
 ISO 19030 2.0: How to Improve
Sergiu Paereli, Jotun, Sandefjord/Norway, [Link]@[Link]
Manolis Levantis, Jotun, Athens/Greece, [Link]@[Link]

Abstract

Introduction of ISO 19030 measurement of changes in hull and propeller performance to the marine
industry, contributed significantly in realizing the importance of performance monitoring and the huge
potentials on fuel and CO2 emissions savings that can be achieved. Although the standard is a powerful
tool for all relevant stakeholders, there are possibilities for further improvements. This paper describes
some approaches suggested for better weather filtering. Among these are: smoothening true wind speed
data before filtering for it; accounting for variability in true wind speed; modifying the cut-off value,
aggregation of true wind speed data, using wave data in performance analysis filter.

1. Introduction

ISO 19030 has been playing in important role in the marine industry for almost 3 years now. Its
publication in 2016 has made it possible for all stakeholders to solve their one of the biggest challenges
lack of measurability. The standard has contributed to the alignment of the parties with regards to not
only using a practical approach of quantifying changes in hull in propeller performance, but also
reducing costs and GHG emissions. Observations show that there is a clear tendency towards higher
investments in premium technologies related to improvement in hull and propeller performance. It

accuracy levels of the savings quantification and transparency of the method.

The majority in marine industry are nowadays collecting data from their ships and, in a way or another,
monitoring the performance of the latter. The way in which vessel performance monitoring is done has
evolved during the years and today continuous monitoring seems to be more and more popular as it
allows people to first and foremost make faster decisions.

Continuous monitoring, as its name suggests, refers to regular collection of data from the ship, and then
processing of this data. ISO 19030 describes very well the steps of continuous ship data analysis, among
these data acquisition, storage, preparation (here referring mainly to filtering, validation and correction),
and calculation of performance indicators. Although, the standard discusses practical and rather simple
guidelines, ship performance monitoring, as such, is a complicated job. Challenges come with the need
of data preparation for analysis and are mainly related to changes in ship operational profile (speed,
loading condition, etc.) and changes in environmental conditions (wind, sea state, currents, etc.). To
isolate changes in hull and propeller performance, two main approaches can be considered. It is either
that one estimates the contribution of changes in other factors and corrects for these, or data should be
filtered for comparable conditions. ISO 19030 describes both approaches.

As mentioned in the standard, variation in loading condition is accounted for by adjusting vessel model
(speed-power reference curves). Wind effect can be estimated using wind correction scheme or data
can be simply filtered for bad weather (wind and sea state). Wind correction is a good approach but
additional data, such as wind tunnel test results, would be needed. These are not always available for
analysis, or not meaningful (e.g. for container vessels). Therefore, the simplest approach in data
preparation, is to filter the latter for similar conditions. This is, of course, to be done after removing
outliers and validation of the dataset.

In ISO 19030 it is suggested that data is filtered for wind speed so that only periods when true wind
speed is below 7.9 m/s (BF 4 or 16 knots) are considered for analysis. As practice shows, this condition
eliminates a significant amount of data from analysis. It has been found that on average about 40% of
data is filtered out.

261
There is no doubt that the higher the true wind speed is encountered by the vessel, the bigger the impact
on apparent performance is. Nevertheless, it remains a question whether the 16 knots standard cut-off
would be ideal. This paper discusses whether the proposed cut-off is justifiable and suggests alternative
ways of filtering data for bad weather.

2. In-service data

Weather is one of the main factors influencing vessels performance. It does not influence the in-service
performance as such, but rather operational performance. Nevertheless, if one is after accurately
quantifying changes in hull and propeller performance in time, he needs to somehow deal with
variations in weather. There are several aspects one would ideally investigate, among these the impact
of wind and sea state (waves and swell). Unfortunately, as of today there are no commonly accepted
devices that could be installed on a vessel in service and that could provide sea state data. The latter,
though, can be obtained from for example noon reports (crew observations). Another source of sea state
data could be observations and model data provided by research institutes. These data are, however,
available on a relatively low frequency and resolution and the accuracy level is questionable. The
common approach today is to rather focus on wind readings and apply filters on true wind speed. In this
way it is expected that wave effects will, to a certain extent, be eliminated as well, since waves
development is generally not as spontaneous as wind development.

In this paper, an evaluation of a dataset from a VLCC was performed and the goal was to better
understand the impact of filtering data for true wind speed on the accuracy of the performance values
(speed deviation). Chosen vessel was equipped with a high frequency data logger, receiving and saving
data every 15 seconds. All necessary parameters for speed deviation computation - speed through water,
shaft power, draft aft and fore were logged. In addition, speed over ground, propeller rpm, fuel
consumption and wind speed and direction were available, so that data consistency could be concluded
to be good and necessary wind filtering to be applied.

From the whole dataset, only the first year of data (after dry-docking) was isolated for analysis purposes.
This has been done to ensure stable speed deviation throughout the whole period. Before speed
deviation computation, data has been filtered using a basic performance analysis filter. This was based
on a certain speed range, shaft power range and mean draft range. In addition, for every simulation, a
filter on true wind speed was applied and the impact of this filter on the final speed deviation over a
one-year period has been evaluated.

Simulation 1: In Simulation 1 true wind speed has been filtered as per ISO 19030. All periods for which
true wind speed was higher than 16 knots have been eliminated from analysis. Calculated speed
deviation points and their average are shown in Fig.1.

Fig.1 Speed deviation (blue points) and flat average (red line)

262
The first observation made was the amount of data left for analysis after applying the filter on the initial
dataset 25.8%. This value lies within normal range, as usually after filtering data for speed, shaft
power, loading condition and wind speed, there is 20-35% data left for analysis. No significant
fluctuations have been noticed in speed deviation and it is concluded to be stable throughout the whole
period. In table 1 several statistical parameters describing the resulted performance values are presented.
These are mean, standard deviation, mean absolute deviation (MAD).

Table 1: Speed deviation statistics for Simulation 1

Mean Standard deviation Mean absolute deviation (MAD)
-6.2 4.60 3.58

Mean is the measure of central tendency. The next two refer to statistical dispersion and talk about data
variability. The reason of looking at both measures describing scatter is the fact that unlike standard
deviation, mean absolute deviation is a robust measure of scale which means that it is less affected by
outliers. standard deviation can be achieved not only by low variance in data, but also
changes in . This is important as by applying difference filters on the same initial 1-
year dataset, one could be left with different amount of data for computation of performance value.
Simulation 1 is taken as reference and the results of the following simulations are to be compared with
results presented in Table 1.

Simulation 2: In this simulation, the same cut-off value for true wind speed as suggested in ISO 19030
has been used. The difference from Simulation 1 is that before filtering data for wind, a very simple
low pass filter has been used for smoothing the signal moving average. When setting the 16-knot
threshold on the raw true wind speed data, one aims at eliminating bad weather (anything strictly
higher than BF4) from analysis. However, it remains a question whether values around this threshold
but slightly above or below are really affecting the final performance values. In Fig.2 a couple of
examples of these doubtful situations are given.

Fig.2: Differences between filters in simulations 1 and 2 (15min moving average)

In example 2 from Fig.2 one could see that some data points that lie within the 16-knot threshold are
kept for analysis and the question is whether they should. Wind trend is clearly higher than accepted
cut-off value and unless values between BF4 and BF5 throughout the whole period are accepted, it
seems that these points can increase the uncertainty of the analysis.

263
On the other hand, as shown in example 1, when the general wind trend is below and/or very close to
the cut-off value, one might consider keeping for analysis those data points that are slightly above the
threshold.

Several scenarios have been considered in Simulation 2: 15-minute , 1h , 6h and 1 day moving
average for true wind speed. These are referred to as S2-1, S2-2, S2-3 and S2-4. The amount of data
left for analysis in Simulation 2 is shown in Table 2. In Table 3 speed deviation statistics are presented.

Table 2: Data left upon filtering in scenarios 1-4 from Simulation 2.

Window size Data left for analysis
15min 25.6%
1h 25.6%
6h 25.7%
1day 25.6%

Figs.3 to 5 give more examples of impact of true wind speed smoothing.

Fig.3: Example 1 - Differences between filters in simulations 1 and 2 (all scenarios: S2-1, S2-2, S2-3
and S2-4)

No big differences in neither amount of data left for analysis, nor statistical parameters have been found
and it is therefore difficult to judge whether evaluated filters have any positive impact on the
performance analysis precision. Nevertheless, Figs.3 to 5 point again to the idea that it is probably not
always meaningful to filter raw true wind speed data but rather first apply a low pass filter on it. There
is a significant amount of data around the threshold of 16 knots that could potentially be part of
performance analysis if not filtered out.

Table 3: Speed deviation statistics for Simulation 2

Window size Mean Standard deviation Mean absolute deviation (MAD)
15min -6.2 4.60 3.57
1h -6.2 4.59 3.57
6h -6.1 4.56 3.54
1day -6.2 4.67 3.62

264
From all three figures, but especially Fig.3, it could be concluded that applying 1-day moving average
on true wind speed does not make much sense. Although, over filtering is not concluded, even if
somewhat expected, applying 1-day moving average on wind does not make it possible capturing
shorter-term fluctuations. Similar percentage of filtered data can therefore be explained by the fact that
or vice-
versa occur and less data retained when moving average has a highly negative or highly positive slope
around 16-knot threshold.

Fig.4: Example 2 - Differences between filters in simulations 1 and 2 (all scenarios: S2-1, S2-2, S2-3
and S2-4)

If wind signal is smoothened by applying 15 minutes moving average, then filtering for the latter leads
to sometimes retaining maybe a bit too much data for analysis. When short-term downward fluctuations
in true wind speed occur, see figure 4 -minute
window size. This might impact the final performance values, especially if those fluctuations are just
some noise. It can indeed happen that true wind speed drops short-term, but then it remains a question
whether the apparent operational performance of the vessel picks up in those periods.

By looking at Figs.3-5, it could be suggested that instead of following one of the 4 tested scenarios, a
combination of them is used. Since using 1-day moving average has been concluded not practical, it is
suggested that one proceeds with both 1-hour and 6-hour moving averages before applying the standard
filter of 16 knots on true wind speed. This is also supported by speed deviation statistics presented in
Table 3. In this way, more data is retained for analysis when wind trend is just below the cut-off value.
Furthermore, in many cases, it is believed that the impact of true wind speed values slightly exceeding
the threshold does not impact negatively the final performance figure.

If this is implemented for this particular vessel, about 23% of data would be left after filtering.

There are cases, though, when this approach does not work as expected and therefore an additional
simulation has been carried out.

265
Fig.5: Example 3 - Differences between filters in simulations 1 and 2 (all scenarios: S2-1, S2-2, S2-3
and S2-4)

Simulation 3: Approach tested in Simulation 3 is very similar to the one used in Simulation 2. The
purpose of running it is to check how filters mentioned in Simulation 2 would behave if one accounts
for the variability in true wind speed. Variability in true wind speed has been evaluated by computing
the moving average of standard deviation. Window size used in this sense corresponds to that used for
smoothening the wind speed signal. Same window sizes as in Simulation 2 were used. The distributions
of moving standard deviations are plotted in Fig.6.

Fig.6: Distributions of moving standard deviations (different window sizes)

As expected, variability in wind speed during 1 day is generally very high. This again confirms that
probably using 1-day moving average of true wind speed as filtering parameter makes no sense. The
distributions of moving standard deviations in case of 15min and 1h window sizes are similar.
Standard deviations of this distributions are 0.78kn and 0.94kn while the means 1.3kn and 1.7kn,
respectively. The variability in wind speed over 6h lies somewhere in between the described
distributions.

By smoothening true wind speed signal (window size 15min and 1h) before using this parameter in
filtering criteria, it is expected that more data is retained for analysis when wind trend is just below the
cut-off value of 16 knots but, at the same time, less data is retained when wind trend slightly exceeds
the threshold. The question of keeping some data points in the first case or filtering them out in the
second case is addressed by evaluating the variability of true wind speed values in respective periods.

266
The threshold for moving standard deviation has been set to the sum of distribution mean and
distribution standard deviation. For scenario S3-1 (using 15min window size) threshold is therefore 2.1
kn, for S3-2 (using 1h window size) 2.6kn and, finally S3-3 (using 6h window size) 4.2kn. The
latter seems a bit to high but, at the same time, if lowering it to 2-2.5kn a lot of data would be filtered
out. Nevertheless, it is decided that all 3 scenarios are plotted. Speed deviation statistics are presented
in Table 4 and some results from Simulation 3 presented in Figs.7 and 8.

Table 4: Speed deviation statistics for Simulation 3

Window size Mean Standard deviation Mean absolute deviation (MAD)
15min -6.1 4.56 3.54
1h -6.1 4.56 3.53
6h -6.1 4.53 3.51

Fig.7: Example 1 - Differences between filters in simulations 1 and 3 (all scenarios: S3-1, S3-2, S3-3)

In Fig.7 is shown a case in which, somewhat more short-term fluctuations in wind occurred. If the initial
1-year dataset is filtered according to ISO 19030 then all the points below 16 knots would be kept for
analysis. On the other hand, it could be seen that the general wind trend from midnight to about 9am is
above the threshold. Imposing variability limits on true wind speed data enables to filter out such dips
which are probably less meaningful. There is no doubt that wind can slowdown but not to forget that
wind is the only logged parameter that reflects the weather overall. If the general trend in wind is higher
than the accepted threshold over enough long periods, sea state changes as well. On the other hand,
fluctuations in sea state condition are not expected to happen as fast as fluctuations in wind speed.

In Fig.8, the opposite case is shown. There is a sudden peak in true wind speed while the overall trend
is below the accepted cut-off value. Furthermore, there are some values which are very close to the
threshold but slightly above. Normal filter from Simulation 1 would not retain such values, while if
averaging true wind speed and then on top impose a variability limit, one could keep more data for
speed deviation computation.

267
Fig.8: Example 2 - Differences between filters in simulations 1 and 3 (all scenarios: S3-1, S3-2, S3-3)

When simulating chosen scenarios, it has been noticed that sometimes S3-1 and sometimes S3-2
provide desirable results. It is difficult to draw a conclusion, even from table 4, on which approach
should be used instead of S1, but probably a combination of the two would work well. It is also to be
noted that somewhat less data is kept for analysis after filtering data in Simulation S3 in comparison to
simulation S2. The amount of retained data is in between 23.5 24.6%.

Simulation 4: If one would like to keep more data for analysis, then one could experiment also with
increasing the threshold of 16 knots. In Simulation 4, it has been decided to still average true wind speed
(window sizes 15min and 1h) and when filtering for it account for variability in data but increase the
cut-off value to 18 knots (scenario S4-1) and 21 knots (scenario S4-2). Statistics for resulted speed
deviation are summarized in Table 5. As expected by increasing the threshold, standard deviation and
MAD increase slightly but no significant differences have been spotted. It is therefore questionable
whether excluding true wind speed values in the range 16-21 knots from analysis is justifiable.

Table 5: Speed deviation statistics for Simulation 4 (S4-1 and S4-2)

Window size Mean Standard deviation Mean absolute deviation (MAD)
15min (S4-1) -6.2 4.58 3.56
15min (S4-2) -6.3 4.61 3.62
1h (S4-1) -6.2 4.58 3.57
1h (S4-2) -6.3 4.60 3.61

Another scenario considered in Simulation 4 is using wind speed and eventually wave height from
research institutes data for filtering purposes. Data is available either from in-situ observations (wind
speed) or models (wave height) and could be used especially in cases when anemometer data is not
trustful. Wind data is available as 6h averages while wave data 3h instantaneous values.

First, it has been checked whether there is a correlation between in-situ/model data (wind and waves)
and data from anemometer. The correlation of weekly moving averages is shown in Fig.9.

268
Since logged data is very well correlated with in-situ data, one could try using the latter for filtering bad
weather. It is to be noted that no averages have been applied to in-situ wind and wave data. Results of
scenario S4-3(only wind), S4-4(only waves) and S4-5(wind and waves) are presented in Table 6.

Fig.9: Pearson correlation

Table 6: Speed deviation statistics for Simulation 4 (S4-3)

Scenario Mean Standard deviation Mean absolute deviation (MAD)
S4-3 -6.3 4.64 3.62
S4-4 -6.0 4.61 3.55
S4-5 -5.5 4.42 3.42

When in-situ wind speed data is used in performance analysis filter, no big changes in speed deviation
are noticed. However, when wave data is included as well, it seems that some more scatter is being
removed. The mean value changes significantly, while standard deviation and MAD become the
smallest among all previous simulations. The amount of data left for analysis upon filtering in scenarios
S4 3-5 ranges between 19.5 31.1% with the smallest values corresponding to S4-5. This points to the
conclusion that wave data could indeed increase the precision of the analysis. Besides this, one could
possibly think of aggregating weather data instead of smoothening the signals by applying a low pass
filter. This has been tried and briefly discussed further.

3. Alternative approach

An alternative approach to the above would be to aggregate true wind values into longer period
averages. The idea is that speed deviation (performance values) is not significantly affected by
temporary fluctuations in true wind speed (ex. gusts), but rather by the overall trend in true wind on a
longer period. In addition, aggregating to lower frequencies than 15 seconds is an indirect way to reduce
the noise from the sensor (anemometer). In order to check the effect of true wind aggregation, the same
initial dataset was used as in previous chapter. Aggregation levels chosen were one minute, 5 minutes,
15 minutes and 1 hour. Padding was selected in order to fill the missing values when projecting true
wind values into 15 seconds frequency. Finally, median was selected for aggregation instead of mean
as median is less sensitive to outliers and as such more representative of the longer trend of true wind.
The effect of wind signal aggregation is shown in Fig.10.

Fig Effect of wind signal aggregation on different scale

In order to measure the impact of this approach on speed deviation, the mean, standard deviation(std) and
mean absolute deviation(MAD) of perfromance values were calculated for each of the selected aggregated

269
frequencies. By applying the above, the number of points left for speed deviation calculations varies to
±0.5%, which means that there is no prefered aggregation frequency which potentially could increase data
retention. A possible explanation for this is that the amount of points which are being excluded and at the
same time included around the threshold of 16 kn is similar. This could also be a possible explanation of
the findings of this simulation. As seen in Table 7, all statistical metrics are suprisingly the same, meaning
that there is no significant difference in mean, standard deviation or mean absolute deviation of perfromance
values.

Table 7: Speed deviation statistics for alternative approach

Period Averages Mean Standard deviation Mean absolute devaition(MAD)
15sec -6.2 4.60 3.58
1min -6.2 4.60 3.58
5min -6.2 4.60 3.58
15min -6.2 4.60 3.58
1h -6.2 4.61 3.58

For validation purposes one more simulation was carried out with a dataset from an LNG vessel using
the same approach. The findings were similar, and all statistical metrics were almost identical
Although further research is needed with several datasets and especially datasets from containers (due
to higher impact of true wind in such vessel type), one may conlcude that short term fluctuations of true
wind (ex. gusts) do not have a significant impact on speed deviation. This conlusion may be used as an
argument in order to increase the threshold of 16knots as imposed by the ISO 19030 standard

Park et al. (2017) mention in their analysis, that once all filtering recommended by the ISO 19030 is
applied, then a statistically significant dataset cannot be realized (approximately 90% of the dataset is
filtered out). This is not always the case, as many points are being filtered out due to the fact that
operational parameters (speed and power) are not within the range of the available external information
(speed-power curves). However, threshold of true wind value of 16 knots has a significant role in the
number of points which are filtered out. Given the above conclusion one may suggest that increasing
the threshold of 16 knots is meaningful and will not have significant impact on calculations. The
question remains; which level of threshold should be introduced in order to maintain similar accuracy
levels in calculation of speed deviation.

To address this issue, a simple exercise was made for the vessels in question (VLCC and LNG). The
simulation was carried out with the alternative approach but a threshold of 18 knots applied. In both
cases, the amount of data retained was on average 12% higher. Mean value was affected by ±0.1% in
VLCC and by ±0.2% in the LNG case. The other statistical measures (std, MAD) were very similar
without a clear indication of improvement. As such, the trade-off between the amount of data and
analysis precision seems to pay off.

4. Summary

This paper discusses the possibility of improving data filtering for weather crucial step in data
preparation stage. Weather is a broad term and can be described by many parameters wind, waves,
currents, etc. However, not all are being logged and available for vessel performance monitoring
purposes. Therefore, paper focuses on primarily wind data filtering and two approaches have been tried
out.

First, wind data signal is smoothened by applying a low pass filter. In this approach several scenarios
were tested. It is the window size which was varied, data variability which was accounted for and cut-
off value which was changed. When attenuating true wind speed signal, 15min , 1h , 6h and 1-day
moving averages have been used. After that wind speed values have been filtered according to the
threshold mentioned in ISO 19030 (16 knots). It has been concluded that 1h and 6h moving averages

270
provide somewhat better results, as in those cases less periods with short-term fluctuations in wind are
kept for further analysis.

If, besides smoothening wind signal, one attempts to account for wind data variability, then even more
scatter can be eliminated. When fixing variability limits on wind speed data, one could filter out either
some more noise in data or, as described in the paper, less meaningful periods. Fluctuations in wind
speed in time are expected but sudden changes in wind do not necessarily mean changes for example
in sea state and, therefore, these probably should not reflect instantly on the performance values. In case
wind speed from anemometer is not available, one could use wind and eventually wave data from either
in-situ observations or models. This has been simulated and results suggest that somewhat more scatter
is eliminated from data. It is believed that there is a two-fold reason for this - filtering data for wave
height and using aggregated wind data. Since lower-frequency wind data (from in-situ observations)
has shown better results, additional simulations with anemometer wind data aggregation have been
carried out.

Wind data has been aggregated to 1min, 5min, 15min and 1h and then used in filtering criteria for
performance analysis. Padding was selected to maintain high frequency of the data and median was
selected in aggregation as it is less sensitive to outliers. Results suggest that wind aggregation has no
significant impact on the final performance values regardless of the selected aggregation level. To
validate this, a different dataset from an LNG was used and the same has been concluded. This was not
fully expected, since aggregation is supposed to reduce scatter in wind signal.

Finally, since both approaches indicate that there is no big impact on performance values, it has been
concluded that by increasing the threshold on true wind speed data from 16 to 18 knots is meaningful.
This increases the amount of data retained by about 12% on average whereas it does not affect
significantly the statistics of the performance values.

References

PARK, B.J.; SHIN, M.S.; KI, M.S.; LEE, G.J.; LEE, S.B. (2017), Experience in Applying ISO19030 to
Field Data, 2nd HullPIC Cof., Ulrichshusen

271
 Effects of Hull and Propeller Cleaning on Propulsion Efficiency of an
Offshore Patrol Vessel
Kevin Murrant, Allison Kennedy, Rob Pallard, Marcel Montrose,
National Research Council Canada, /Canada, [Link]@[Link]

Abstract

The effects of hull and propeller cleaning on propulsion efficiency are evaluated for an offshore
patrol vessel belonging to the Canadian Coast Guard. Dedicated sea trials were conducted between
cleanings to evaluate propulsion efficiency using standard speed and power trial procedures. These
results are compared with efficiency analysis performed on continuous operational data. A hull
recoat was applied in fall of 2018, and the condition of biofouling and coatings observed at dry dock
is discussed.

1. Background

The Canadian Coast Guard Ship (CCGS) Cygnus is an offshore patrol ship that operates out of St.

system, developed by OpDAQ. The system measures shaft torque, shaft speed, shaft power, and
vessel fuel consumption. The National Research Council (NRC) of Canada has a separate Data
Acquisition System (DAS) onboard the CCGS Cygnus since Fall 2015. The NRC DAS stores data
from a number of vessel systems such as the navigation system and propulsion system. The NRC has
also been obtaining OpDAQ data from the CCGS Cygnus since 2016. The data from the NRC DAS
and OpDAQ system is used for the current project to quantify changes in vessel performance as a
result of hull and propeller cleaning.

This report summarizes the propulsion efficiency analysis of the CCGS Cygnus operational data prior
to and subsequent to cleaning the hull and propeller. This data is used to quantify any changes in
vessel performance, specifically the power versus speed relationship. In addition, the vessel fuel
consumption at a given power level will be quantified prior to and post cleaning events.

Three dedicated sea trials were conducted to support this project. Each set of trials is a dedicated
Speed and Power trial and was planned and conducted in accordance with International Towing Tank
Conference (ITTC) guidelines. The first set of trials is a baseline trial to quantify the performance
before the hull or propeller are cleaned. The second set of trials is a post hull cleaning trial to quantify
the performance subsequent to cleaning the vessel hull only. The third trial is conducted post
propeller cleaning and is used to quantify any changes in speed and power performance as a result of
cleaning the propeller.

The result of this project suggests how the power and speed relationship for the CCGS Cygnus
changes after cleaning events within the scope of the trials. It also quantifies how the power and fuel
consumption relationship changes as per the observed data. These changes are quantified using
measured data from dedicated sea trials. This information could be used to support planning and
optimization of vessel cleaning schedules.

2. CCGS Cygnus Vessel Details

the Grand Banks area which

crew change the vessel departs again for Grand Banks to continue patrolling. The vessel has two main
medium speed, diesel engines. The ship particulars of the CCGS Cygnus are outlined in Table 1.

272
 Table 1: CCGS Cygnus Main Particulars
Particular Value
Length (m) 62.4
Breadth (m) 12.2
Draft (m) 4.0
Freeboard (m) 0.9
Cruising Speed (kts) 13.0
Maximum Speed (kts) 16.0
Number of engines 2
100% MCR (kW) ~3000
Number of propellers 1

3. Hull and Propeller Fouled Condition

Prior to cleaning the hull and propeller, a subsea survey was conducted to characterize the level of
fouling present. A guideline from the Royal Navy (2011) was followed for this procedure. The hull
fouling was characterized as a specific type and rated from 0 to 100 to indicate the level of severity.
The hull cleaning was completed by divers when the vessel was docked in Conception Bay, NL on
June 5, 2018. The diving company prepared a report to document the level of fouling on the hull.
When assessing the level of fouling on the hull, the divers divided the vessel into 27 regions. At each
of these regions photos were taken to document the fouling condition. The location of each region is
shown in Fig.1, as taken from the diver s report. Note that this image is not to scale, nor is it a
representation of the CCGS Cygnus. It is used only to indicate general location and quantity of
underwater areas that were surveyed.

Fig.1: Hull fouling characterization locations (from diver s report)

An underwater video was also taken to support characterization of hull fouling. By evaluating the
vessel in situ, through photographs and using the underwater videos, the divers characterized the level
of fouling at each of the 27 locations across the hull. The divers used the Royal Navy (2011) and US
Naval Ships Technical Manual (2006) to define fouling type and rating values so that results would
be consistent with Navy practices. A summary of the hull fouling characterization is provided in
Table 2. All fouling on the CCGS Cygnus hull was noted to be soft. The dominant organisms in the
soft fouling type are slime and grass. The fouling rating included FR 20 and FR 30. This type of
fouling involves advanced slime and grass filaments up to 76 mm long. The percentage of fouling
coverage in each area ranged from 40-100%. It was noted that fouling was located from waterline
down to turn of the bilge with heavier growth present near the waterline.

273
 Table 2: Hull fouling characterization type, rating and percent coverage
Location Fouling Type Fouling Rating Percentage Coverage (%)
1 Soft 30 80
2 Soft 20 50
3 Soft 20 75
4 Soft 20 100
5 Soft 20 80
6 Soft 20 90
7 Soft 20 100
8 Soft 20 50
9 Soft 30 50
9 Soft 20 50
10 Soft 30 70
11 Soft 20 50
12 Soft 20 40
13 Soft 20 80
14 Soft 20 80
15 Soft 20 90
16 Soft 20 90
17 Soft 30 65
18 Soft 30 80
19 Soft 20 90
20 Soft 20 60
21 Soft 20 90
22 Soft 20 100
23 Soft 20 90
24 Soft 20 100
25 Soft 20 95
26 Soft 20 50
27 Soft 20 80

The hull and propeller of the CCGS Cygnus had not been cleaned in two years prior to this project.
The level of fouling present was a result of 2 years of operation. The CCGS Cygnus operates year
round on a two week rotation with a two day layover. Vessels with an off-season or with long layover
periods would likely have more fouling in similar operational and environmental conditions.

Fig.2: Typical propeller pressure face pre (left) and post (right) cleaning

The propeller was also assessed by divers to quantify the level of fouling present on July 18, 2018.
All propeller blade faces were covered in a light to moderate slime which was heavier at the root and
tapered towards the tips. Under the slime the propeller blades were covered with a heavy calcium
buildup. The level of propeller fouling was measured using a ship propeller roughness gage which
characterizes the propeller roughness per the Rubert Comparator scale. The propeller fouling was

274
rated as Rubert scale E. Once polished, the propeller was rated at a Rubert scale A/B. Post polishing
trials were conducted at this polished state. Fig.2 illustrates the pre-cleaning and post-cleaning
condition of a typical propeller pressure face of the CCGS Cygnus propeller. This report includes a
number of images of pre and post cleaned propeller surfaces.

4. Sea Trials

Three separate sets of sea trials were completed. The first was conducted prior to cleaning the hull or
propeller and provides data to use as a baseline. The second set of sea trials were completed after the
hull was cleaned and the third set of sea trials were completed after the propeller was cleaned.

All trials followed the same procedure and occurred at the same location. The trials followed ITTC
2014 guidelines for the completion of speed and power trials. These guidelines outline boundary
conditions as a cutoff point for the completion of such trials. These boundary conditions relate to
location, water depth and environmental conditions and vary based on the vessel size. The specific
trials boundary conditions for the CCGS Cygnus, are summarized in Table 3.

Table 3: Sea Trials Boundary Conditions

Parameter Parameter Detail or Value
Selected location should have minimal vessel traffic and should be sheltered to avoid
Location
wind / wave where possible.
Water Minimum water depth of 52.2 m. Data corrections required for water depths less than
Depth 71.8 m.
Wind shall not be higher than Beaufort 5. Beaufort 5 relates to mean wind velocity
Wind
between 17-21 knots.
Sea State The maximum wave height when derived from visual observation should be 1.2 m.
Areas with known large current variations in time or space should be avoided. Small
Current currents will be corrected for by completing tests in two directions, one upwind and the
other downwind.

Prevalent weather conditions and vessel traffic intensity were considered when selecting a trials
location. The location was selected to be within Conception Bay to reduce the likelihood of heavy sea
states when compared to a location along the normal Cygnus operational route. The location was set
to north of Bell Island since there was relatively little vessel traffic at this location than other areas of
the Bay.

During each trial three or four different power settings were tested. The power settings tested
included 50%, 65%, 80%, and 100% of the main engines Maximum Continuous Rating (MCR). All
tests were completed in two directions: upwind and downwind. A double run at 65% MCR was
conducted once during each set of trials. The double runs completed at 50%, 80% and 100% MCR
were conducted twice, as per the ITTC 2014 guideline. The baseline trials included only three power
settings (65%, 80%, and 100%) as the original plan did not specify runs at 50% power setting. After
analysis of the baseline trial data, it was decided to include runs at 50% power in the subsequent trials
to provide additional context for the higher power data points. It was attempted to perform all trials at
a consistent displacement and as such there were no significant changes in cargo or machinery
between trials.

Fig.3 shows the location of the sea trials. The direction of all trials was along the yellow line, between
the points NRC 1 and NRC 2. This track has a total length of ~10 km to provide space for the high-
speed runs. Each test required 10 minutes of constant rpm, pitch, and speed settings. As such, some
tests were shorter in distance than others. All tests were centered near the subsea acoustic probe
(Autonomous Multichannel Acoustic Recorder AMAR) point in Fig.3. The AMAR point is located
at 47o -52o
of the Bay, which corresponds with the prevailing wind direction. Once a test was completed in one
direction, the vessel would turn around and complete the same test in the opposite direction.

275
 Fig.3: Trials location and direction in Conception Bay, NL

Each trial run involved a period to get up to speed and attain constant settings, a 10 minute constant
setting period, and then a Williamson turn to return vessel to opposite direction for subsequent
testing. The trial trajectory was similar to that shown in Fig.4.

Fig.4: Trial trajectory illustrating the approach for each reciprocal run

4.1 Baseline Trials

Baseline (pre-cleaning) trials were completed on May 23, 2018. The wind and sea conditions during
the morning were higher than the boundary conditions for these tests and as such all tests were
completed in the afternoon when conditions calmed. The conditions during baseline trials are
summarized in Table 4. There were 11 runs completed in total. Two of these were runs at a MCR
setting of 65% (upwind and downwind), four at MCR of 80% (two upwind and two downwind) and
four at a 100% MCR (two upwind and two downwind). There was a repeat test of the first run which
was 65% MCR in the upwind direction. The repeat was conducted because the wave and wind
conditions were higher during the first test of the day than they were during the remainder of the tests.

Table 4: Baseline Trial Conditions

Condition Value
Testing timeframe 13:30 16:15
Vessel forward draft (m) 3.35
Vessel aft draft (m) 4.83
Range in true wind speed (kts) 16 - 22
Range in wave heights (m) 0.5 1.0
Range in swell height (m) 0 0.5
Water temperature (oC) 4.0

276
During the baseline trials the wave and swell heights were estimated by the vessel Captain. These
values were not measured during baseline trials as the wave buoy was not deployed due to morning
weather conditions. The water temperature was also estimated for the baseline trial, using historic
water temperature values from the area. In addition, the estimated water temperature was compared to
water temperature measurements taken from a wave buoy that was located in Holyrood Harbor,
which is not too far from the trials site.

4.2 Post Hull Cleaning Trials

The post hull cleaning trials were completed on July 18, 2018. The weather conditions during post
hull cleaning trials are summarized in Table 5. There were 14 runs completed in total. Four of these
were runs at 50% MCR (two upwind and two downwind), two at a 65% MCR (one upwind and one
downwind), four at 80% MCR (two upwind and two downwind) and four at a throttle setting of 100%
MCR (two upwind and two downwind).

Table 5: Post Hull Cleaning Trial Conditions

Condition Value
Testing timeframe 10:30 14:00
Vessel forward draft (m) 3.05
Vessel aft draft (m) 4.66
Range in true wind speed (kts) 14 - 25
Range in wave heights (m) 0.2 0.4
Range in swell height (m) 0 0.25
Water temperature (oC) 10.2

During the post hull cleaning trials the wave and swell heights were estimated by the vessel Captain.
These values were also measured by a wave buoy during these trials. Estimated values were
compared with those measured. Values estimated were consistently higher than those measured, by
approximately 50%. Measured values are summarized in the trials log as well as in Table 5.

4.3 Post Propeller Cleaning Trials

The post propeller cleaning trials were completed on August 1, 2018. The weather conditions during
post propeller cleaning trials are summarized in Table 6. There were 14 runs completed in total. Four
of these were runs at 50% MCR (two upwind and two downwind), two at 65% MCR (one upwind and
one downwind), four at 80% MCR (two upwind and two downwind) and four at 100% MCR (two
upwind and two downwind).

Table 6: Post Propeller Cleaning Trial Conditions

Condition Value
Testing timeframe 11:45 15:05
Vessel forward draft (m) 3.02
Vessel aft draft (m) 4.72
Range in true wind speed (kts) 4.5 10.2
Range in wave heights (m) 0.3 0.6
Range in swell height (m) 0
Water temperature (oC) 14.9

During the post propeller cleaning trials the wave and swell heights were measured by a wave buoy
deployed for the test.

5. Measured Speed and Power Data

The measured shaft power versus speed through water for each test during each trial are plotted on the
same axes in Fig.5. All collected data points follow a similar relationship that can be fit with an

277
exponential curve. There is less variation in the post propeller trials data when compared to the other
trials results for a given engine setting. This is expected due to the calm wind and sea conditions
during the post propeller polishing trials which were lighter than those for the other two trials.

Fig.5: Uncorrected power versus speed data from each test run during each trial. An exponential
curve fit is shown to illustrate the trends between trials. The uncorrected data does not show a
significant difference between trials in the speed curve.

The measured data was analyzed first using the mean of means method to provide further insight
towards data trends between trials. The mean of means method involves taking the mean of
consecutive double runs at a given engine setting and then taking the mean of those means to
represent the speed and power values at that engine setting. The intent of this method is to eliminate
the unidirectional effects of wind and current under the assumption that these effects will average to
zero. The mean of means for all trials completed at a given engine setting, within each sea trial, were
calculated. The results of shaft power and vessel speed through water for each sea trial were plotted
(Fig.6). Trend lines were fitted through the data for each sea trial.

The differences in the relationships between power and speed for the three sea trials is clearer here.
The trend line relationships between trials change across the speed range. The post-hull cleaning trials
trend line requires approximately 4% less power to attain speeds between 12.5 and 16 knots. In this
same speed range, the post-propeller cleaning trials trend line indicates that approximately 5% less
power is required to attain a given speed when compared to the post hull cleaning trials. These results
indicate that a total of approximately 9% less power is required to attain a given speed (in speed range
between 12.5-16 knots) as a result of cleaning both the hull and propeller. For speeds less than
approximately 12 knots, more power is required to attain a given speed for the post hull cleaning
trials when compared to the baseline trials. This result is unexpected and may be influenced by the
higher level of uncertainty involved in the lower engine setting trials.

278
Fig.6: Means of means for each trial. Each data point is the mean of 2-4 runs at that speed in
reciprocal runs.

It is also noteworthy to discuss the fact that these results are presented in terms of power versus speed
rather than overall resistance versus speed. Overall resistance is more difficult to characterize since it
takes into account the propulsion as well as associated efficiencies (hull efficiency, propulsive
efficiency, and relative rotative efficiency). The power can be roughly calculated by dividing the
resistance multiplied by the vessel speed, by the overall system efficiency. The overall propulsion
system efficiency is a combination of multiple complex factors. For example, the propulsive
efficiency would increase as a result of cleaning the propeller. Another example is that the hull
efficiency, which describes how the water flows around the hull and into the propeller, can affect the
propeller efficiency as a result of cleaning the hull. It is therefore possible that the resistance versus
speed relationship for each set of trials would not exhibit the same performance gains across the
speed range compared to simply comparing power versus speed.

6. Speed and Power Analysis and Results

The speed and power data measured during field trials was analyzed to remove variations due to
environmental differences between trials. This was completed following ITTC guidelines for the
analysis of full-scale speed and power trials, ITTC (2005). This analysis method was complimented
using insight from ISO 15016 when additional guidance was needed. The ITTC guideline requires the
conversion of measured power data to vessel resistance in order to apply certain correction factors to
account for environmental effects and correct data to a common, calm state. The ITTC guideline
describes a method to determine the resistance of each trial run by using the measured torque along
with information from associated propeller curves. A major element of uncertainty in this analysis is
that the open water propeller curves for the Cygnus propeller are not available to support data
analysis. As such, a standard B-Series propeller curve was assumed to be representative of the
Cygnus propeller.

The ITTC guideline provides methods to calculate resistance corrections for: wind, waves, deviation

279
in water temperature and density, water current, shallow water, and displacement variation.
Resistance corrections were calculated for each trial run within each specific sea trial. For all cases,
the resistance correction to account for water current was not calculated since the vessel speed
through water was measured directly. Also, the shallow water correction was not calculated for any
trials since the trials were conducted in deep water. The resistance correction to account for variations
in vessel displacement were provided only for displacements that varied less than 2%. Based on the
forward and aft draft measurements taken at the beginning of each trial, the displacement varied by
approximately 8%, exceeding the range of the correction method. As such, there were no corrections
added to account for displacement variation and the change in displacement is a source of variability
within the results. Note that the forward and aft were estimated by the vessel crew based on draft
marks prior to each trial and were not measured directly. Therefore, the variation in displacement
could differ than the percentage value calculated using the estimated trim values.

The resistance corrections calculated for each trial were subtracted from the trial resistance that was
calculated to reduce the resistance to a calm water baseline which could be used for direct comparison
between trials. The corrected vessel resistance was used to calculate the corrected power. The ITTC
analysis method required estimation of a number of coefficients specific to the vessel used in trials as
well as the estimation of a number of environmental parameters that were not directly measured.
Estimation of these parameters leads to a level of uncertainty in the results. A summary of the
estimated parameters is provided below.

Wake fraction, thrust deduction fraction and propeller relative rotative efficiency. These coef-
ficients can be found from model test results for a particular vessel. Model test data for the
CCGS Cygnus was not available for this data analysis. As such, the commercial software
NavCad was used to model each trial and output the associated coefficients. The measured
and predicted shaft power values compared well (within 10%) and thus the coefficients out-
put from NavCad were deemed as reasonable.
Thrust coefficient and advance coefficient. The ITTC analysis guideline states that the pro-
peller open water thrust and advance coefficients, both required to calculate resistance, are to
be retrieved from propeller open water curves. The CCGS Cygnus propeller open water
curves were not available for this analysis. As such, standard B-Series open water propeller
curves were used to represent the Cygnus propeller. The standard B-Series open water pro-
peller curves were updated to match the pitch (as approximated by NavCad) of each trial run.
Each unique set of curves was then used to retrieve the required data associated with the cor-
responding run. Unfortunately, the actual pitch relating to each test was not known and had to
be approximated based on the pitch percentage which was noted from a gage on the bridge of
the vessel and using NavCad. This added to the uncertainty involved in using the standard B-
Series curve. In addition, the Cygnus propeller is controllable pitch and the standard B-Series
propeller is not. The ratio of hub diameter to propeller blade length is larger for a controllable
pitch propeller than for a fixed pitch propeller.
Wetted surface area. The wetted surface area of the CCGS Cygnus was estimated with
NavCad using input of the vessel main particulars and selection of representative vessel type.
Transverse projected area above waterline. The transverse projected area above waterline of
the CCGS Cygnus was estimated using measurements from the general arrangement drawing
of the vessel and known draft.
Wind resistance correction. The correction for wind resistance was estimated using recom-
mended equations for the calculation of wind resistance.
Wave height during baseline trials. The wave height was not measured during baseline trials.
It was estimated using the measured wind speed and the fetch limited JONSWAP wave spec-
trum. The value of fetch used for the trials was 30 km. These estimates were compared to
measured wave height data from a nearby (Holyrood) wave buoy and the results matched
well (within 10%).
Water temperature during baseline trials. The water temperature was not measured directly
during baseline trials and was estimated based on historic water temperature data during the

280
 same time of year. The estimated value was compared to measured data from a nearby (Holy-
rood) wave buoy and the results were similar.
Water density for all trials. It was assumed that 3.5% salinity was representative of the water
density during trials.
Kinematic viscosity for all trials. The water kinematic viscosity was not directly measured
and was estimated using the water temperature and ITTC Salt Water Property tables.

Preliminary results of the analysis showed that the wind resistance correction the most significant
factor in comparison to the other corrections, for all sea trials. In addition, for the baseline trials and
the post hull cleaning trials, where wind speeds were towards the upper wind speed limits of the trials,
the wind resistance correction was very large in comparison to the bare hull resistance, particularly at
the lower speeds.

The measured speed and power data was analyzed three separate times to correct for environmental
conditions, each using the ITTC 2005 method or a slight variation to the method. The first analysis
approach was conducted strictly to the ITTC guideline. The second and third analysis approaches
were conducted using the ITTC 2005 guideline with a different estimation of wind resistance
correction. The second attempt involved a wind resistance correction estimation using the Fujiwara
method. This method was one of the wind resistance predictors recommended in NavCad, a
commercially available vessel performance evaluation software. The third attempt involved a wind
resistance correction estimation of half the predicted value using the Fujiwara method. Three separate
analysis were completed to illustrate the variation in result that occurs due to different estimations of
wind resistance correction.

6.1 Results Baseline Trials

The results of the three analysis methods for all trials data is summarized in Fig.7. The measured
trials data have higher power per speed than the corrected data for all test cases except the 11 knot
speed. It is expected that the measured power would be higher since the power is being corrected to a
calm condition and less power would be required to attain a given speed in calm seas. The measured
data is very close to the corrected values at 11 knots which suggests that the wind and wave
conditions during these tests were relatively mild. There is not much difference between the corrected
results using the ITTC wind correction and the NavCad (Fujiwara) wind correction, for all tests. The
corrected data using half the Fujiwara wind correction, lies in between the measured data and the
other corrected data.

The measured trials data have higher power per speed than the corrected data for all test cases except
the 11 knot speed. It is expected that the measured power would be higher since the power is being
corrected to a calm condition and less power would be required to attain a given speed in calm seas.
The measured data is very close to the corrected values at 11 knots which suggests that the wind and
wave conditions during these tests were relatively mild. There is not much difference between the
corrected results using the ITTC wind correction and the NavCad (Fujiwara) wind correction, for all
tests. The corrected data using half the Fujiwara wind correction, lies in between the measured data
and the other corrected data.

6.2 Results Post Hull Cleaning Trials

The data corrected using the ITTC wind correction and the NavCad wind correction are similar. The
spread between the measured and corrected data increases with speed for the post hull cleaning trials.

6.3 Results Post Propeller Cleaning Trials

The measured data is very close to the corrected data for the post propeller cleaning trials due to the
mild environmental conditions during the trials. There does appear to be one outlier for the tests at
speed between 11 and 12 knots for which the measured power is below the corrected power values.

281
Fig.7: Corrected speed and power data for all trials. Due to the correction, environmental effect is
reduced and the data points follow more closely the trend.

Table 7: Corrected speed and power data for all trials

Baseline Post Hull Cleaning Post Propeller Cleaning
Speed (STW) Power Speed (STW) Power Speed (STW) Power
(knots) (kW) (knots) (kW) (knots) (kW)
7.6 308 7.9 293
7.9 356 7.5 264
7 276 7.5 298 7.9 291
8 265 8.0 359 7.5 256
10.8 611 11.2 685 11.5 739
11.2 657 11.8 813 11.2 716
13.9 1182 14.2 1305 14.8 1539
14.2 1418 14.9 1541 14.6 1512
13.9 1175 14.3 1275 14.8 1518
14.3 1409 14.9 1575 14.6 1565
16.0 2353 16.0 2477 16.3 2555
16.2 2635 16.3 2694 16.1 2501
15.9 2323 16.1 2481 16.3 2792
15.9 2413 16.4 2766 16.1 2460

6.2 Results Post Hull Cleaning Trials

The data corrected using the ITTC wind correction and the NavCad wind correction are very similar.

282
The spread between the measured and corrected data increases with speed for the post hull cleaning
trials.

6.3 Results Post Propeller Cleaning Trials

The measured data is very close to the corrected data for the post propeller cleaning trials due to the
mild environmental conditions during the trials. There does appear to be one outlier for the tests at
speed between 11 and 12 knots for which the measured power is below the corrected power values.

6.4 Results Comparison of Trials

The corrected power results for each sea trial were plotted against speed on the same plot to illustrate
differences in performance. This was completed for each of the three wind correction approaches
considered. For each analysis method, the results of each trial are relatively similar in terms of the
power required at a given speed. Given this, it is difficult to quantify the gain in power associated
with cleaning the hull and propeller, from this data. The regression lines for each trial are very similar
for all analysis methods used. For each approach, the baseline trial regression line is higher than the
post hull and post propeller cleaning regression lines, at speeds higher than approximately 13 knots.
This indicates that there is a benefit of cleaning the hull and propeller in these speed ranges in terms
of power required to attain a given speed. Below, approximately 13 knots, the regression line for
baseline trials falls below the regression line for the other two trials. This change in regression line
relationship between trials is consistent to the measured data results and may be due to higher
uncertainty at the low speed tests.

The power savings above 13 knots were quantified using the regression line equations from the
NavCad wind correction approach. Between 13.5 and 16 knots an average of 5% less power is
required to attain a given speed after cleaning the hull, when compared to the baseline power
requirements. There is no additional power reduction identified within this speed range as a result of
cleaning the propeller which was unexpected. In fact, the performance after cleaning the propeller, in
terms of power versus speed, is worse in this speed range.

Note that there is variability within the tests conducted at a given throttle setting for a given trial in
terms of speed through water and corrected power. The corrected power for a throttle setting for one
trial often falls within the range of corrected power for the same throttle setting in a different trial. For
example, at a throttle setting of 10 the speed through water varies between 15.9-16.2 knots for the
baseline trials, 16.0-16.4 knots for the post hull clean trials and 16.1-16.3 knots for the post propeller
cleaning trials. For this same throttle setting the corrected power (NavCad wind) ranges from 2272-
2677 kW for the baseline trials, 2462-2823 kW for the post hull clean trials and 2438-2599 kW for
the post propeller cleaning trials. The speed and power values from one trial, fall within the speed and
power range for a different trial for this throttle setting. This is consistent for the other throttle settings
considered and is true for the measured data as well as the corrected. This may be due to variability
between trials (e.g. displacement variation, fouling present) and leads to less reliability in the power
savings quantified using this data.

7. Operational Data Analysis

In addition to the sea trials conducted under somewhat controlled conditions, operational data was
collected on an ongoing basis throughout the time period of the sea trials. From this data, the data was
segmented based on speeds greater than 7 knots, and durations of at least 2 hours. Each segment
meeting these criteria was averaged to provide a mean data point. These data points, color coded by
period between cleaning events, are illustrated in Fig.8.

283
Fig.8: Mean data points extracted from operational data segments greater than 7 knots in speed and 2
hours in duration. Data points are color coded according to the period between cleaning events
they correspond with.

Fig.9: All operational data points throughout time period covering the cleaning trials. Scatter points
occur due to operational activities, equipment and instrumentation fluctuation, and environmental
condition.

284
The complete operational dataset is illustrated in Fig.9. This data covers all recorded data points
throughout the operational period covering the cleaning trials (May-Aug, 2018). The data points are
scattered due to operational activities, such as towing, equipment or instrumentation fluctuations,
such as shaft power, and environmental conditions, such as large wind and waves. The open water
power to speed relationship is visible, but not easily characterized.

To reduce the extraneous data points, the data set was clipped by a threshold of 300 kW about the
approximate power to speed relationship. Fig.10 illustrates the data set with curve fits and comparison
to the mean trial data. From this figure, it can be seen that the relationships are apparent, but the trial
conditions capture the minimum power to speed curve from the operational data scatter. This is due to
the relatively calm conditions under which the trials were conducted versus the typically rougher
conditions encountered by the vessel under normal operation.

Fig.10: Reduced operational dataset covering trials period. Curve fits are applied for comparison. The
mean trial data is shown for comparison with the operational data result.

8. Fuel Consumption and Speed

To identify the fuel curve for the vessel, total main engine fuel consumption versus shaft power was
plotted for each trial. The fuel consumption versus power curve should be consistent for all trials
since both fuel consumption and shaft power are not affected by external parameters such as
environmental condition or hull and propeller condition. The speed attained however, does change as
a result of variation in these external parameters. The measured total main engine fuel consumption
versus power curve for all trials is shown in Fig.11. For each engine setting tested during trials, the
baseline trials and post propeller cleaning trials data aligns well. However, there is an offset when it
comes to the post hull cleaning trials data. For these trials there is less fuel required for a given power
setting, particularly at higher power values. It is expected that there was some mechanical difference
in the fuel measurement system that led to the discrepancy in the post hull cleaning fuel versus power
data. A possibility is that one (or more) of the fuel flow meters surrounding one of the main engines
was bypassed or partially bypassed or blocked during the post hull cleaning trials. However, the
OpDAQ system bypass indicator did not highlight a complete bypass during this, or any of the trials.

285
The baseline and post propeller trials data for fuel consumption versus power are used to define the
general fuel versus power curve for the Cygnus main engines. The post hull cleaning trials data was
not used for this purpose due to the discrepancy from the other trials data. The general fuel versus
power regression equation (regression equation corresponding to Fig.11) was used to calculate the
ng the corrected power values at each engine setting for each
trial. The power values corrected using the NavCad wind correction were used in this analysis. There
is no quantifiable difference between the fuel consumption rate and speed through water for the
different sea trials. This is due to the level of variation in data at single test condition within a given
trial and how data from different trials fall within this variability range. For example, at a throttle
setting of 8.0, the baseline trials speed through water ranges from 13.9-14.3 knots and the corrected
fuel consumption ranges from 373-475 L/h. At this same throttle setting, the post hull cleaning trials
speed through water ranges from 14.2-14.9 knots and the corrected fuel consumption ranges from
393-526 L/h. There is overlap in the speed and fuel consumption variability ranges between trials for
each engine setting.

Fig.11: Uncorrected total main engine fuel consumption rate versus shaft power

9. Condition of Cygnus Hull in September 2018

The CCGS Cygnus was taken into dry dock on September 11, 2018 to perform vessel maintenance.
When in dry dock, the hull was observed to have a relatively high level of fouling on one side in
particular (port). The level of fouling present on the port side was similar to the amount that was
present during the underwater survey conducted in May, 2018, prior to cleaning the hull. Specifically,
there was slime and sea grass covering a large portion of the underwater hull (Fig.12). These
observations were unexpected for two reasons. The first relates to the speed of fouling taking place on
the Cygnus hull. Prior to the May hull cleaning, the Cygnus hull had not been cleaned in two years. It
was anticipated that the level of fouling present in September, just 3.5 months post hull cleaning,
would be much less than that observed during the May survey. Some reasons that could have led to
rapid fouling growth during this short period include the relatively warm temperatures during summer
2018 in the region and a depletion in anti-fouling coating.

286
It is observed that the September hull condition level of fouling differs on the port and starboard sides
of the vessel with the starboard side having a higher level of fouling. The May survey results
indicated that the starboard side of the hull had a slightly worse level of fouling. It was anticipated
that if one side was more fouled than the other in September, it would have been the starboard side to
be consistent with earlier results. Increased fouling on one side of the vessel could result from
frequent docking on one side (fouling occurs more on side subject to sunlight) or lower quality of
anti-fouling coating on one side of vessel. Cygnus Captains were consulted and it was confirmed that
the docking side varies. The anti-fouling paint was noted to be highly depleted in September, on both
sides of the vessel (see Fig.13). This likely played a role in the high fouling accumulation rate during
the summer period.

Fig.12: Images of port side of hull during September, 2018 dry dock. The level of fouling is similar to
that encountered in May, 2018.

Fig.13: Images of Anti-fouling coating after biofouling was removed. The anti-fouling coating is
deteriorated in large swatches across much of the hull.

Note that the anti-fouling paint is the black paint that can be observed in Fig.13. This image indicates
the level of depletion of the anti-fouling paint after the biofouling was removed using a pressure

287
washer in dry dock. The anti-fouling paint adhesion was investigated by brushing it lightly by hand
using a scouring pad. This resulted in the anti-fouling paint flaking off as a result of the brushing. It is
possible that the May and September hull cleaning events enhanced this level of depletion.

The status of the anti-fouling paint was considered further by consulting with the Canadian Navy. A
Canadian Navy biofouling Subject Matter Expert (SME) noted that the level of depletion of anti-
fouling coating on the CCGS Cygnus was not typical given that the anti-coating was applied to the
vessel only two years prior. The Canadian Navy SME indicated that this level of depletion was not
seen on Navy vessels even after 5 years of use. They suggested to check on conditions during
application and noted that application during high humidity levels could lead to faster depletion of
coating. They also indicated that the type of coating used on the CCGS Cygnus was different than
that used on Navy vessels and recommended that the CCG use an alternative coating.

The relatively high level of fouling present in September likely has an impact on the results presented
in this report since there may have been a level of fouling present on the hull during post cleaning
trials. This is particularly true for the post propeller cleaning trials which were not conducted until
August 1, 2018. Unfortunately, there is no way to quantify the amount of fouling present during the
post hull cleaning and post propeller cleaning trials. In general, if fouling was present during the post
hull cleaning trials the analysis would indicate a lower level of power and fuel savings than the actual
values. Also, if more fouling was present during post propeller cleaning trials than during post hull
cleaning trials, the discrepancy between the measured savings potential and the actual savings
potential would differ between trials and be larger for the post propeller cleaning trials. This could
relate to the unexpected result of the post propeller trials found from the ITTC analysis methods.

10. Discussion and Recommendations

The measured results indicated an approximate 4% improvement in terms of required power to attain
a given speed as a result of cleaning the hull and approximate 5% improvement as a result of cleaning
the propeller, for speeds between 12.5 and 16 knots. However, there were variations in the
environmental conditions and condition of the vessel between trials and these differences have an
influence on vessel performance. The wind, wave and water temperature variations between trials
were corrected based on ITTC guidelines. The wind correction was also made using two variations of
the ITTC recommended wind resistance correction method for comparison. The discrepancy between
vessel displacement during the different trials was not corrected for since there was no standard
guideline available to correct for this when the displacement varied by more than 2%. It was
estimated that the displacement between trials varied by approximately 8%. The lhull condition also
varied between trials in that there was likely some level of fouling present during both the post hull
clean and post propeller clean trials. Therefore there is some uncertainty in level of fouling between
the post hull and post propeller cleaning trials. Based on the data that was corrected for wind
(NavCad correction method), waves and water temperature, there is an average of 5% savings in
terms of power required to attain a given speed as a result of cleaning the hull, for speeds between
13.5 and 16 knots. The performance decreases after the propeller polishing in this same speed range
based on this analysis. To correct for wind, wave and temperature variation a number of parameters
(e.g. wind resistance coefficient, propeller pitch for each trial, hull underwater area) had to be
estimated. As a result, there is uncertainty involved in the corrected power values.

In terms of fuel consumption, there appeared to be a measurement error during the post hull cleaning
trials which led to lower fuel consumption rates for a given power setting. This could be a result of
partially closed fuel valve(s) surrounding one of the main engines. Therefore, it was impossible to
quantify fuel savings resulting from cleaning the hull directly from the measured data. The general
fuel consumption rate versus shaft power regression equation was used to calculate the corrected fuel
consumption rates for each trial using the corrected (NavCad wind, ITTC wave and sea temperature)
power values. This resulted in a corrected fuel consumption rate versus speed through water plot for
each set of sea trials. The results for each trial were very similar and there were no quantifiable
differences in the three curves.

288
10.1 Recommendations to Reduce Uncertainty and Gain Result Clarity

There were a number of recommendations identified for conducting a similar study in the future
which would lead to lower uncertainty in the data and more clarity in the results. These are
summarized in point form below.

1. Conduct all tests in very low wind and wave conditions. In this project the baseline trials and
post hull cleaning trials were conducted in similar conditions which were high in terms of the
environmental condition limit. The post propeller cleaning trials were conducted in relatively
mild conditions. The methods to correct for wind and wave conditions lead to uncertainty in
the results since certain parameters need to be estimated. In mild conditions these corrections
are much smaller and therefore less significant.
2. Conduct all tests at the same displacement. Variations in displacement lead to changes in ves-
sel performance. In this project it was attempted to complete all trials at the same draft levels.
However, since the Cygnus is an operational vessel and trials were completed weeks apart
this was difficult to manage. As such, there was a variation in the draft (and displacement)
and the effect of this on the results was not quantified. Trials at the same displacement would
not have this source of variation and would lead to increased confidence in results.
3. Select vessel that has available propeller open water curves, wind tunnel test data and model
resistance test data. This would reduce the number of parameters estimated in the data correc-
tion analysis and lead to lower uncertainty in the results.
4. Conduct tests closer together in time. The three tests involved in this test were completed be-
tween May August of 2018. During this time there was some level of fouling that devel-
oped on the hull between trials and subsequent to the hull cleaning. This leads to a lower lev-
el of confidence in the results since there may have been some fouling present during both
post hull and post propeller cleaning trials. If the trials had occurred closer together in time
(e.g. days apart rather than months) this would limit the potential for fouling to develop on
the hull or propeller between trials.
5. Conduct study on a vessel that has an off-season or longer alongside duration. The CCGS
Cygnus is continuously in operation throughout the year and has a short, 2 day, layover peri-
od between operations. This gives limited time for the accumulation of biofouling and as such
it was expected that the amount of fouling present initially on the Cygnus would be relatively
low. The performance increase as a result of cleaning the hull and propeller would be larger
for a vessel with more fouling in the baseline condition. A good candidate would be a vessel
that does not operate for a portion of the year, during which time fouling would accumulate
faster than during operations.

10.2 Comparisons to Similar Publically Available Data

A brief literature search was completed to compare the results of this study to data available in the
public domain. There were no directly comparable results identified in the literature in terms of
comparable vessel size or initial level of fouling. However there were guidelines identified that
provided insight as to what performance increases could be expected from cleaning the hull based on
different initial levels of fouling (Schultz, 2007). These guidelines are based on model scale drag
measurements and boundary layer similarity law analysis and were made for a mid-sized naval
combatant at two speeds, 15 and 30 knots. Different fouling ratings (FR) as per the Naval Ships
Technical Manual (2006) were used in this study. Table 8 summarizes the results of this study for a
vessel speed of 15 knots in terms of increase in shaft power resulting from different levels of fouling.
The fouled (baseline) condition of the CCGS Cygnus was mostly FR 20 with some areas having FR
30 (~15% of vessel). The corrected (for wind, wave and temperature) results indicate that the baseline
trials required approximately 5 % more power than trials during which the hull was clean, for speeds
greater than 13.5 knots. This is smaller than the 11% estimated increase in power for FR 10-20 as
outlined in Table 8. However, the baseline condition was not a hydraulically smooth surface and was
better described as a somewhat deteriorated coating. Therefore, it is reasonable to expect a lower
power savings when comparing the two conditions.

289
 Table 8: Expected performance changes as a result of hull fouling

NSTM Increase
Rating Description in SHP
0 Hydraulically smooth surface 0%
0 Typical as applied antifouling coating 2%
10-20 Deteriorated coating or light slime 11%
30 Heavy slime 21%
40-60 Small calcareous fouling or weed 35%
70-80 Medium calcareous fouling 54%
90-100 Heavy calcareous fouling 86%

Giorgiutti et al. (2014) conducted a study to investigate the impact of fouling on a crude oil tanker.
This study investigated the effects of fouling on the hull and propeller separately and involved several
sea trials. The data from sea trials was analyzed using ITTC analysis guidelines and complimented
with other recommended methods. The analysis involved corrections for wind, wave, sea temperature
and displacement variation. Details on the displacement during each trial or how this was corrected
for were not provided. In this study the level of fouling at baseline condition was much higher on both
the hull and propeller than that which was present on the Cygnus. The fouling was not rated as per
Naval Guidelines however it was indicated that there was severe hard, calcareous fouling that was
difficult to remove covering the majority of the propeller and underwater hull surface. The savings
resulting from cleaning the hull and propeller were approximately 45 % in terms of reduced power at
cruising speed. This study included both propeller cleaning and polishing.

A Computational Fluid Dynamics (CFD) based study was presented by Demirel et al. (2016) in the
Journal of Applied Ocean Research. This investigation predicted the effect of biofouling on resistance
and power requirements of a container ship based on full scale simulations. These predictions
indicated an increase in power by 18% for the ship fouled with light slime and an increase by 38% for
the ship fouled with heavy slime. There were no sea trials used to compare or validate the CFD
results. However, model test data was compared to the non-fouled predictions and they compared
well.

In general, there is limited comparison data available in the public domain for this type of study,
particularly data resulting from sea trials. The data that does exist can be compared generally but not
directly since the hull forms and initial level of fouling vary. In addition, there are gaps in the
methodologies applied for data analysis and trial corrections for the comparative data that is available
in the literature.

11. Concluding Remarks

The primary goal of this study was to quantify the effects of cleaning the hull and propeller on the
vessel performance in terms of speed and power, for the CCGS Cygnus. The corrected sea trials data
indicated a reduction in power required to attain a given speed by an average of 5% between the
speed ranges of 13.5-16 knots. However, these results were not corrected for variation in
displacement across trials or the presence of slight fouling during the post cleaning trials. The results
compare reasonably to estimations of power increase for a mid-sized Naval frigate for similar
baseline and fouled conditions.

This study provided insight towards steps that could be taken to increase the value of future tests of a
similar nature. These recommendations should be considered when planning future work to increase
the level of confidence in results.

290
Acknowledgement

The authors would like to thank the Canadian Coast Guard for their support in coordinating sea trials,
providing necessary vessel data and prompt response to questions during this work. Thanks to the
Captain and crew of the CCGS Cygnus who executed the sea trials and were cooperative and
supportive towards this research initiative.

References

GIORGIUTTI, Y.; REZENDE, F.; VAN, S.; MONTERIO, C.; PRETEROTE, G. (2014), Impact of
, 25th National Congress of Waterway Transportation, Ship-
building and Offshore, Rio de Janeiro

ITTC (2014), Speed and Power Trials, Part 1: Preparation and Conduct, Int. Towing Tank Conf.,
7.5-04-01-01.1, Revision 1.1.

ITTC (2005), Full Scale Measurements Speed and Power Trials Analysis of Speed/Power Trial Data,
Int. Towing Tank Conf., 7.5-04-01-01.2, Revision 00.

ISO (2015), Ships and Marine Technology Guidelines for the Assessment of Speed and Power Per-
formance by Analysis of Speed Trial Data, International Standard, ISO 15016, 2nd Ed.

ISO (2016), Ships and Marine Technology Measurement of Changes in Hull and Propeller Perfor-
mance Part 1: General Principles, International Standard, ISO 19030-1

, Waterborne Underwater Hull Cleaning of Navy

Ships, Chapter 081, S9086-CQ-STM-010, Revision 5, Published by Direction of Commander, Naval
Sea Systems Command

ROYAL NAVY (2011), Surface Ship Hull Fouling Management, Maritime Acquisition Publication
No 01-026, Issue 02

291
 Information Sharing Concept in Ship Performance Management Systems
Wojciech Gorski, Enamor, Gdynia/Poland, [Link]@[Link]

Abstract

Modern Ship Performance Management Systems (SPMS) automatically collect large amount of
diversified data and usually allow access also for users not physically present onboard. Typical
scenario for this arrangement is ship crew access onboard and technical office onshore. In a most
simple approach both users just have an access to the same dataset. However, providing access to the
same data, does not ensure that SPMS efficiently supports vessel operation. Contrary it may easily
create user confusion due to huge amount and diversity of data. Paper attempts to present means used
to enhance user comprehension of ship data and collaboration between crew and owner office in a
context of taking decisions towards ship performance enhancement. Variety of methods including data
pre-processing and visualisation, transition of data towards information, information encapsulation,

1. Ship data digitalisation

Ship digitalisation is nowadays highly discussed topic. It is expected that process of digitalisation will
reshape of maritime industry in coming years. Review of Maritime Transport provided by UNCTAD
(2018) indicate that digitalisation process will be a key factor in transport optimisation. It however
addresses also major threats connected with data safety and security which must be addressed. Report
indicates also data sharing as the critical aspect since data transformation between systems is limiting
factor for implementation of digital ship concept.

Digitalisation process has been already started by different initiatives described e.g. by Schmode et al.
(2018) but yet have not converged into universal solution. The process, its tools and methods and data
standards will develop and evolve however it is already visible two general approaches. First one is
based on relatively low resources hardware onboard dedicated for data acquisition and transfer to cloud
server. It has limited capability for instant data processing and visualisation thus is intended for
decentralised systems where major processing is on remote server side and therefore rely on continuous
data exchange. More traditional approach assumes high processing capabilities onboard which allows
for data processing, visualisation and user interaction independent on cloud server. This paper presents
SeaPerformer SPMS as an example of latter approach indicating its capabilities for ship digitalisation
through support of user comprehension of data and collaboration between crew and owner office in a
context of taking decisions towards ship performance enhancement.

2. Data sharing

Typical scenario of data sharing between ship and shore consists of single direction of data flow. Data
collected onboard are send to onshore server. This scenario may however result in serious system
limitations. It imposes that complete data set registered onboard must be send each time data are
exchanged even in case some data are not important in daily operation. In case of data which changes
seldom or high resolution data single direction data exchange results in increased data traffic and high
demand for onshore server space. Bi-directional communication allows for data transfer optimisation.
Some data may be excluded from data set periodically send onshore and kept onboard until they are

failure. Until failure occurs these data do not provide useful information and may be retained onboard
without impact SPMS performance. However, in case of failure these data contain critical information
with respect to failure diagnostics and system recovery. Bi-directional communication allows onshore
user to request specific subset of onboard data whenever needed. In most simple form onshore user may
request onboard data based on situation description provided by ship crew. This simple approach
requires crew to recognise failure, provide this information to onshore user (technical office, service

292
support) which in turn requests appropriate data set to be send. This scenario may be however
automated. SPMS may evaluate specific signals (triggers) as indicators of abnormal situation and,
whenever this situation is recognised, automatically request additional relevant data set. This approach

developers. This version of system includes fast signal module which records each engine cylinder
combustion pressure and other relevant information (valves positions) as a function of crankshaft angle.
In order to maintain constant angular resolution of 0.1 degree irrespectively of engine rpm, cylinder
signals are recorded at 10kHz frequency. Resulting data set exceeds few MB per hour of engine
operation and is far too large taking into consideration onboard satellite communication restrictions.
Therefore fast data set is excluded from data package send periodically to onshore server. High
frequency data are kept only in onboard database for some time. Ship crew can use SPMS user interface
to evaluate engine combustion signals and in case of need for assistance, select appropriate subset and
send it onshore for consultancy. Moreover, SPMS allows remote engine monitoring. For this purpose
set of standard engine key performance indicators (KPIs) are calculated based on high frequency
cylinder signals for each engine stroke. These KPIs are send ashore and can be evaluated by competent
personnel in order to detect abnormal situation. In case KPIs indicate situation which needs more
detailed investigation relevant subset of high frequency data can be requested from the ship. Such
request is automatically processed by onboard SPMS and appropriate data packed is send in return.
Onboard SPMS can be also used for automated engine monitoring. Set of rules based on engine KPIs
can be continuously monitored by the system and in defined conditions relevant subset of high
frequency data is send ashore without onboard or shore personnel intervention.

Fig.1: Scheme of data exchange implemented in SeaPerformer

293
Some of data exchange scenarios described in previous paragraph requires onshore user to interact with
SeaPerformer SPMS onboard. Since typically vessels are not online due to limitations imposed by
satellite communication systems or owner internal regulations, interaction with onboard system is
realised with use of emails. General scheme of data exchange used in SeaPerformer is presented in
Fig.1. SeaPerfrormer has been also furnished with remote diagnostic module (RDM) in order to allow
safe channel of communication from shore to ship. RDM implements SMTP and POP3/IMAP client
for receiving commands with data and sending responses. All the data exchanged with use of RDM are
encrypted using combination RSA1024/RSA2048/AES/SHA256 encryption. Regardless the above the
additional transport encryption is recommended between endpoints of email transfer. The onboard
email clients support SSL/TLS when connecting to relevant server. This way data encrypted at source
remains in that state until it reaches the destination. All commands sent to on-board system are encoded
and signed. The on-board system checks the received message against the sender address, proper
encoding, signature, the age, command syntax. Only verified commands are executed. The execution
of commands requires an authorisation command to be executed first. RDM is therefore a great tool for
secure remote diagnostic purposes. Onshore user can send onboard a command for SeaPerformer to
prepare specific subset of data and send it back for further analyses. Although this feature is extremely
useful, RDM can be also used for other purposes. Since, with use of RDM, remote user can get access
to local files at authorised paths general operation system (OS) diagnostic can be handled. System or
specific application log files can be downloaded from onboard computer and analysed onshore. Based
on received information, maintenance commands can be executed remotely. This feature greatly
simplifies OS maintenance minimising a need of IT expert s physical presence onboard. Last but not
least RDM allows for remote update of SeaPerformer software. Encrypted and zipped update files can
be uploaded onboard and update process can be completed remotely without involvement of ship crew.

3. Visualisation and contextualisation of data

Data visualisation comprise one of prime functionalities of SPMS. It allows user access to recorded
data but what is even more important provides context which is required for data interpretation.
Visualisation is also a significant tool supporting transition of data into information i.e. makes data
understandable for user and helps him taking rational decisions. Taking above into consideration it is
largely insufficient to provide user possibility to build graphs with arbitrary data. Usual time trend data
plots although provide access to data, may not be very helpful in providing context and building data
understanding. Furthermore, building user defined graphs is quite involving and requires sufficient level
of experience. Therefore, SeaPerformer provides number of predefined graphs and views which
incorporate data pre-processing (i.e. data cleaning and grouping), contextual selection of graphic
elements types and blending of data elements. Although these graphs are predefined in order to
minimise user effort it allows for certain degree of customisation, especially in case visualisation
combines multiple data sources. Good example of this approach is a map view. It provides geographical
context of vessel operation by default and therefore helps comprehension of data. It also allows for
combination of different data in one consistent view i.e. providing bathymetry or weather data as the
supplementary information to ship operational data. This feature is a powerful tool in building
understanding of physical processes associated with ship operation. Although map view is very helpful
in building context and blending multiple information it does not allow for direct presentation of data
values as in case of typical X-Y plots. Therefore, appropriate data representation are used in map view.
SeaPerformer map view combines colour coding (i.e. value representation by appropriate colour) for
scalar data (i.e. ship speed through the water or engine power), arrow representation for vector data and
text boxes for providing direct access to data values. Furthermore, map view can be enhanced with
supplementary information such as bathymetry overview, navigation aids or extends of emission control
areas (ECA). Such versatile visualisation can be easily overloaded with information and final effect can
hinder understanding of presented information. Therefore, in order to facilitate handling of
visualisation, map view elements are associated with layers which can be managed by user. This way
visualisation can be easily adjusted for the purpose and needs of user.

294
 Fig.2: Different visualisation techniques used in map view

Fig.3: Time graph enhancements supporting data contextualisation

Sophisticated techniques can be also employed in more traditional visualisation such as time trend. This
type of visualisation is probably the most common graph used in SPMS systems however in many cases
does not sufficiently support user in understanding of presented data. An obvious advantage of time
trend is visualisation of signal dynamics and providing easy access to data values either by use of axis
and data grid or textual overlay (so called tool tips). Time trend graphs does not readily support
visualisation of data which values are not numerical (i.e. binary data or system states) or which range
is of different magnitude. Latter may be well handled by multiple Y-axis or re- scaling of data. Non-
numerical data visualisation requires special techniques such as use of alternative graph background
colour. Enhanced data comprehension can be also achieved by combination of different graph styles.
An example of this approach in SeaPerformer is auxiliary engines load graph. This visualisation
combines two principal graph styles; stack area graph for representation of individual genset load with
colour differentiation of each unit, and line graph for indication of maximum output of gensets in
operation. This type of graph is helpful in identification of concurrent use of many gensets resulting in
low efficiency and excessive fuel consumption. Graph arrangement supports this task; colour layers

295
identify number of active gensets, stack area graph provide actual output of gensets in operation, line
graph define total available output. Large gap between line graph and stack graph identify usage of too
many gensets resulting in low efficiency. Inefficient use of gensets is identified onboard and
communicated to the crew. In order to illustrate crew response to the warning, time of unnecessary
concurrent use of gensets is visualised by colour overlay on a graph. Yellow overlay indicate time span
when warning message is presented for the crew until corrective measures are taken (see Fig.3). In case
reaction time is exceeded, overlay changes colour to red giving clear visual indication to supervising
personnel.

4. Pre-processing

Data pre-processing is an indispensable feature of any SPMS if the system is intended to be capable of
transforming data into information. Well known, and already numerously quoted in past editions of
hatically highlights importance of this
process. Handling of unverified data which may include errors, data gaps or data context
misinterpretation will lead to false conclusions. While, to certain extend, data reliability problems can
be overcome by experienced user who based on his past knowledge, abstract thinking and problem
generalisation can detect, neglect or regenerate problematic dataset, automatic handling of improper
data pose serious challenge. In SeaPerformer development of data pre-processing mechanisms is treated
as key feature. Some of advances with this respect have been already presented in past edition of this
conference by Gorski (2017). These concerned data redundancy and regeneration. Both techniques
address problems of data gaps or identified false readings. However, very often reliable identification
of data failure became a challenging task. Onboard equipment used as data source seldom provide self-
diagnostic features and does not deliver clear identification of malfunction (e.g. failure flag) which may
be automatically interpreted by the SPMS. Except obvious case of data gap (i.e. system reads NULL
signal from a data source) identification of corrupted signal is a difficult problem which requires
implementation of advanced processing techniques usually taking into account context of analysed
signal. In order to clearly explain these difficulties two practical examples are discussed in details. Both
concerns vital signals for performance evaluation. First example concerns ship static draft which is
important for determination of vessel operation condition and is crucial in evaluation of required engine
power. Static draft fore and aft is used to determine actual ship trim (her longitudinal inclination with
respect to water plane) which is essential for trim optimisation. Secondly rudder deflection signal is
discussed. This signal is used for identification of ship manoeuvring and allowing appropriate engine
power correction to reference conditions.

Static draft can be read directly from ships loading computer (LC) however interfacing LC is sometimes
challenging. In case LC is not available static draft can be determined with use of draft sensors i.e.
pressure sensors scaled according to static head of the water and corrected for sensors position with
respect to ship bottom. Draft sensors readings shall be however treated with caution since they are
vulnerable to dynamic changes of pressure due to water movement with respect to ship hull.
Furthermore ship changes her position at speed with respect to water plane (dynamic trim and sinkage)
which impacts draft sensor readings as well. Therefore static draft can be read with use of draft sensors
only at ship rest (in port after loading and ballasting procedure is completed). Obtained value shall be
retained in SPMS and used until next port call. This method can provide practical information unless
major ballasting operation is performed. If re-ballasting is done during sea voyage use of this method
e.g. for trim determination is misleading and additional data input such as trim inclinometer is required.
As explained above determination of such basic parameters as ship static draft and trim may be quite
complicated in practical operational conditions. It requires special approach which includes data sources
redundancy and special processing. However, in order to reflect reality of ship operation it must be
noted that draft signal reliability often pose a serious problem. Author experience in use of draft data
collected on more than 200 vessels of various type and equipment reveals frequent problems with draft
sensors. These sensors are prone to failure providing awkward results which, if not recognised, may
lead to problems in performance analyses or trim optimisation. It is often observed that draft sensors
after failure continue to provide data. Therefore, most common and simple detection of malfunction by
recognition of lack of data may not be sufficient. However, since draft signal after sensor failure changes

296
its properties i.e. often exhibit lesser sensitivity and shift, failure detection based on data processing
may be effectively used. For the purpose of draft sensor malfunction detection is based on parallel
evaluation of four draft sensor signals (fore, aft, midship portside and starboard) and speed over ground
signal. Detection algorithm works in two variants depending on ship speed. At higher speeds only signal
lost or rapid shift uncorrelated with speed on any of single draft sensors is detected. At lower speeds
above method is supplemented with calculation of hull trim, list and longitudinal deflection. Values
obtained based on draft sensors are compared with typical operational ranges which allows for detection
of possible failure. Algorithm automatically recognises port operations during which only signal
continuity is verified.

Another signal which often poses difficulties in operation is rudder deflection. This signal is important
since allows for detection of dynamic states during ship manoeuvres and thus enable proper signal
filtering for performance analyses. Rudder deflection is usually taken from synchro transmitter/receiver
as the analogue signal or, in modern equipment, already converted to digital NMEA format. Rudder
deflection is often subject of errors, especially in case of analogue signal transmission. The most
straightforward in processing is signal lost however often situation is more complicated. It has been
observed that many cases rudder deflection transmitter continue to provide data but with significant
error. Two types of erroneous signal patterns have been observed: signal stall and signal shift. Signal
stall results in almost steady (sometime small oscillations can be observed) signal irrespectively of
actual rudder deflection. Signal shift is characterised by appropriate reproduction of signal gradient but
rudder neutral position is not preserved. It is important that SPMS is able to recognise not only rudder
signal lost but also other errors. For this purpose, it should be able to continuously process data during
acquisition process. SeaPerformer implements appropriate algorithms for detection of rudder signal
stall and shift. Due to nature of these errors stall recognition algorithm works in shorter time window
and offset in longer time window. Stall of rudder signal is recognised based on standard deviation.
Small value of standard deviation indicates an error. Rudder signal offset error determination is based
on 3% quantile calculation at extreme ends of signal distribution. Large value of quantile suggests
problem with data source. Error determination based on statistical quantities proved to be efficient
approach (see Fig.4) comparing to method based on signal comparison with expected range of signal
variation since it copes also with signal having false analogue to digital scaling factors.

Fig. 4 Erroneous rudder signal identified by signal processing module

As it has been highlighted in above examples, verification of signal quality and detection of possible
data source malfunctions may involve special computational methods. It is usually not sufficient to
relay on current value of signal or signal gradient. Due to scattered nature of signals registered onboard
signal pre-processing is necessary. These methods include moving window averaging and statistical
calculations (e.g. standard deviation and quantile calculations) on data subsets ranging from dozen of
minutes up to couple of days. Data processing may apply to several thousands of records and therefore
could be computationally demanding. It shall be noted that signal verification supposed to be realised
in parallel to data acquisition and visualisation therefore it must be taken into account in SPMS
hardware resources, mainly computing processor and memory.

297
5. Encapsulation

Data encapsulation, also known as data hiding, is a term used in programming and means mechanism
whereby the implementation details are kept hidden from the user. In a context of SPMS it can be useful
technique which simplifies user interaction and presentation of information. SeaPerformer system
utilises encapsulation both in onboard and in web applications. System notifications can be taken as an
example where this technique is employed. Onboard software processes multiple signals according to
specific algorithms in real time. On this basis system detects situations which require ship crew action.
However instead of presenting relevant signals system encapsulate them into warning symbol (bell icon
as presented on Fig.5) and guidance information. This method allows to focus crew attention and
minimises time to take necessary actions. Web application also uses encapsulation in case of system
notifications but in different way comparing to onboard software. Instead of using bell symbol, warning
information is encapsulated on ship track using colour. Yellow or red colours denote part of the voyage
where unwanted situation occurred. This method allows for quick warning identification and at the
same time provides larger context through map view.

Fig.5: Information encapsulation in onboard application

6. Data hierarchy

Trim optimisation provides an interesting example of data hierarchy concept application. Ship hull trim
is an important factor with respect to fuel consumption. Depending on hull form, selection of favourable
trim setting may result in 3% to 8% savings in fuel consumption comparing to situation when trim is
selected only with respect to stability and strength criteria. Due to complicated hydrodynamic
phenomena of flow around hull geometry optimum trim vary with ship speed and draft and cannot be
determined with use of simple methods. Traditionally hydrodynamic model tests performed in wide
range of speed, draft and trim were employed for this purpose. For each combination of speed and draft,
impact of trim on hull resistance or ship propulsive power was determined. Continuous interpolation of
test results constitute a reference model upon which optimum trim can be determined during ship
operation. Physical model tests can be nowadays substituted with methods of computational fluid
dynamics resulting in lower costs and time of the reference model determination process. Another
promising method for reference model elaboration is utilisation of operational data but this approach
involves sophisticated data cleaning techniques and may be limited by range of parameters at which
vessel is operated. Irrespectively of the reference model origin the method of optimum trim
determination is the same. Initially vessel speed through the water, static draft and trim shall be
determined. Local surrounding of operational parameters is searched for better solution (usually search
is done for trim only while speed and draft is kept constant). If substantial improvement is recognised

298
trim change is recommended. It must be emphasised that trim optimisation process is highly dependent
on proper recognition of actual operation parameters. In case speed, draft or trim is determined
inaccurately it will influence a starting point of optimisation and thus also final result. Unfortunately, it
is difficult to determine all of important parameters for the process. Problems of accurate determination
of speed through the water has been already discussed e.g. by Bos (2017) but also in case of draft and
trim their proper identification is sometime challenging. It should be noted that usually reference model
for trim optimisation is based on static draft and trim while direct measurement of draft (and
recalculation of trim) by draft sensors provides dynamic parameters. These values may differ
substantially especially on higher speeds where dynamic trim and sinkage effects are pronounced.
Furthermore, as already discussed in previous chapter, draft sensors are sensitive for dynamic pressure
around the hull. These effects may falsify readings especially for bow sensors. In order to counteract
mentioned problems SPMS may use multiple sources of data arranged in hierarchy of accuracy and
reliability. SeaPerformer uses following sources for draft and trim determination:

a) loading computer,
b) draft sensor readings in port corrected for actual trim measured by ship inclinometer,
c) direct draft sensor readings at speed corrected for actual trim as in b),
d) direct draft sensor readings uncorrected for dynamic effects.

Static draft and trim shall be determined with use of loading computer. Accurate modelling of hull and
internal compartments allows for determination of draft with error less than few cm. However, LCs are
usually separated systems, difficult to interface with. Furthermore, even in case SPMS is connected to
LC data exchange must be initiated by LC user which is often forgotten. For this purpose, in parallel to
LC data SPMS reads and process data from draft sensors. SeaPerformer monitors draft readings during
port operations and retains values at the completion of loading operations (final static draft at departure).
Static draft may however change during voyage i.e. due to ballasting operations which may not be easily
recognised since draft sensors are burdened by dynamic effects. In order to take account for these effects
ship inclinometer is used. During ship voyage inclinometer measures dynamic trim. This may be
however recalculated to static value based on reference model (it must contain both static and dynamic
trim). In case actual dynamic trim is outside of reference model range and cannot be recalculated to
static value, SPMS can use directly draft sensor readings as they were static ones. The latter is the less
accurate method and shall be only used at low speeds where dynamic effects are not pronounced.

6. Conclusions

As indicated in the introductory note there are two general approaches differing in allocation of
computational resources. Preceding paragraphs illustrate important features of SPMS which increase
data comprehension and supports operational decisions resulting in improvements of ship performance.
Irrespectively of the SPMS concept these features shall be incorporated as they are vital for achieving
required functionality. Both concepts theoretically allow supporting data pre-processing and
visualisation, transition of data towards information, information encapsulation, sharing and hierarchy.
It must be however confronted with existing ship data transfer infrastructure. Nowadays majority of
vessels do not support data streaming. Data are sent in packages and therefore data processing on cloud
server is delayed by data exchange period. It is serious limitation especially in case of signal pre-
processing with respect to possible data source malfunctions. Cloud server-side processing makes it
necessary that results shall be send back onboard in case they require crew attention. This scenario
therefore increases data transfer which is nowadays critical due to costs. Taking this into consideration
solution which secures sufficient computing resources onboard reveal certain advantage until satellite
communication enables data streaming on reasonable costs.

Ship Performance Management Systems can effectively support increase of vessel operation
effectiveness thus enable cost savings and lower environmental impact. However, in order to take full
advantage of the system it must incorporate features that enhance user comprehension of ship data and
collaboration between crew and owner office. These important features include data pre-processing and
visualisation, transition of data towards information, information encapsulation, sharing and hierarch

299
which were in details discussed in this paper. Available data transfer technology implies that solution
allowing for local, onboard data processing pose advantage.

References

BOS, M. (2016), How MetOcean Data Can Improve Accuracy and Reliability of Vessel Performance
Estimates, 1st HullPIC Conf., Pavone

GÓRSKI, W. (2017), Automatic MRV Reporting with Use of Onboard Performance Management Sys-
tems, 2nd HullPIC Conf., Ulrichshusen

SCHMODE, D.; HYMPENDAHL, O.; GUNDERMANN, D. (2018), Hull Performance Prediction be-
yond ISO 19030, 3rd HullPIC Conf., Redworth

UNCTAD, (2018), Review of Maritime Transport 2018,

[Link]

300
 Application of Biotechnology in Antifouling Solutions for Hard Fouling
Prevention
Philip Chaabane, Catherine Austin, Dan Isaksson, Oliver Weigenand, Cecilia Ohlauson,
I-Tech AB, Gothenburg/Sweden, [Link]@[Link]

Abstract

This paper describes the development, subsequent application and success of using biotechnology in
marine coatings for hard fouling prevention purposes. Existing self-polishing and controlled depletion

uptake and in-service performance data for selected case studies are also presented. The paper also
shares insight into I-
materials that make use of Selektope®.

1. Introduction

Since the worldwide ban of the use of tri-butyl-tin (TBT) in antifouling coatings in 2002, suppliers of
marine coatings have faced increasing pressure to offer antifouling products that deliver the same level
of effectiveness as those that previously contained TBT for preventing the build-up of biofouling on
ship wetted hard surfaces.

The banning of TBT use forced the marine coatings sector to reconstruct their formulations to
accommodate different biocides, with copper designated as the favoured candidate, supplemented with
booster biocides. Soon after, biocides faced new regulatory challenges in the shape of the EU Directive
on Biocidal Products (98/8/EC). The effect of this was the number of certified biocides available for
use in antifouling coatings reducing in number to just 12 active substances, including Selektope®.

Neither scheduling penalties, nor increased fuel costs were acceptable to a globalised industry reliant
on just in time delivery. A solution had to be found and, for barnacle fouling, that solution was
Selektope®.

Some two years earlier, in February 2000, biologists at the University of Gothenburg published a
research paper on biofouling in Swedish waters. Researchers had been investigating how a range of
substances that would prevent the settlement of hard fouling when dissolved in seawater could be used.

This research focused on the barnacle Amphibalanus improvisus s of man-made

Dahlström
et al. (2000) found that larval-stage receptors were remarkably responsive to one substance in
particular medetomidine. This bioactive substance prevented barnacle larvae attempting to settle on a
hard substrate.

Medetomidine was also distinguished by its reversible effects. Larvae that came into contact with the
substance could still later metamorphose into juvenile barnacles with no apparent ill effect.

In collaboration with two Finnish universities, Swedish researchers discovered that medetomidine
could bind to a specific group of receptors, the octopamine receptors. The receptors were cloned and
the causality between the receptor and medetomidine was established. Further study led the researchers
to link the binding to octopamine receptors to changes in the larval behaviour at a surface. This
explained the high efficacy in preventing and deterring barnacle larvae in an antifouling paint without
its being toxic to the barnacles.

301
 Fig.1: Medetomidine identity

In a counter-intuitive discovery, given its sedative effect on vertebrates, medetomidine was found to
induce hyperactivity in the barnacle larvae to disrupt the settling process; similar to the effects to a
small dose of adrenaline in humans.

Fig.2: Cyprid larvae stage of the barnacle Fig.3: Adult stage of the barnacle

During initial panel testing, a further discovery was made. Remarkably, a polymer film containing
medetomidine in a concentration equivalent to 0.02% by weight volume rejected 97% of the aggressive
Barnacle improvus after two weeks, and 96% after four weeks. No other macro-fouling organisms were
present at all. A further
the full extent of a surface.

approach to biofouling prevention. I-Tech AB commercialised the use of medetomidine in marine

coatings, owning all IP and regulatory rights to the antifouling agent under the brand name Selektope®.
I-Tech also controls the largest and most efficient source of medetomidine production.

In 2009, buoyed by the further confirmation of earlier research, I-Tech entered a new stage in the
development of Selektope®, by initiating the registration of the marine antifouling agent for regulatory
approvals. evaluation
process; the dossier BPR (EU528/2012) consists of more than 20.000 pages 528 files and refers to 90
investigations regarding human and environmental safety. Even so, Selektope® was granted full
approval in 2016 for use as a marine biocide under EU Biocide Product Regulation (EU) (528/2012).

Today, Selektope® has received approvals in all leading markets for new builds and dry-docking
including China, South Korea and Japan. In the EU, Selektope® has been approved for all relevant use-
types l. For Africa, South America and the rest of Asia, no registration is needed for the use of
Selektope®. Following 15 years of development time and numerous regulatory hurdles, the first
commercial antifouling coating product containing Selektope® was applied to the vertical sides of a
ship in 2015 with the first commercial product containing Selektope® being officially launched into
the market in 2016.

302
2. Increasing demand for advanced antifouling coatings

Ocean going vessels are increasingly at risk from negative commercial impacts associated with
biofouling accumulation on the hull. Marine fouling is the biological process of single celled organisms,
algae and hard-shelled organisms, predominantly barnacles, attaching to submerged surfaces and
colonising at a rapid rate.

Any organisms anchored on the hull create increased drag (commonly referred to as added resistance)
which significantly decreases hull performance. A biofouled vessel must burn more fuel to attain the
same speed through water when in active service, resulting in higher fuel costs for the ship operator. A
hull suffering from heavy fouling is also extremely impactful on maintenance costs. Costs associated
with hull cleaning services are fac
global biofouling risk increases, hull cleaning is likely to be required more frequently, in-creasing
maintenance costs. Repeated cleaning of the hull can also remove layers of the antifouling coating,
reducing its service life.

Different types of biofouling require different hull cleaning practices. Soft fouling can be removed by
diver cleaning with brushes or by Remote Operated Vehicle (ROV) cleaning technology. However,
hard fouling presents a greater cost and risk for the ship operator. Encrusted colonies of barnacles or
other hard-shelled organisms must be removed with methods such as scraping or blasting which are
much more damaging to the antifouling coating.

Biofouling accumulation on vessels experiencing extended periods of static activity is also becoming
an increasingly dominant issue on the agenda for many shipyards. Newly launched vessels remaining
stationary for three or four months at shipyards located in biofouling hotspots, or longer in the case of
LNG carriers, during the fitting out process are becoming so fouled that they perform badly during sea
trials.

In addition, growing regulatory focus on the transportation of invasive aquatic species (IAS) by the
international shipping fleet creates an increased risk of potential commercial impact. Some regional
regulations are already in force that allow ports to refuse entry of heavily bio-fouled ships, resulting in
greater financial costs for the operator. On an international level, IMO has recently shifted its focus on
tackling IAS transfer via ballast water onto hull biofouling.

The afore-mentioned issues are driving the need for high performance, advanced antifouling technology
in the maritime industry. Ship operators are increasingly demanding antifouling paints that are both
well-suited to specific ship trading patterns, and varying activity levels in addition to protecting against
both soft and hard fouling. When looking at the future trading potential, ship operators need to ensure
that their ship is protected whether it be in constant active service, idle for long periods of time, or is at
risk of fluctuating between the two.

This future-proofing approach to antifouling coating selection, without any certainty of future trade, is
exerting great pressure on the coating suppliers, prospering great innovation and new approaches of
fouling prevention technology using the active substance Selektope®. This is supported by increasing
demand for antifouling coatings that contain the anti-barnacle active agent from ship owners and
operators.

3. The use of Selektope® in self polishing coatings

siloxane
elastomers and fluoropolymers, yet. Selektope® is a biocide that is currently suited for use only in
- (SPCs) (CDPs) types. SPCs rely
n through water causing tiny quantities of the base polymer
paint to hydrolyse and to leach at a predetermined rate, while the active antifouling maintains its

303
Selektope® is a biocide that has highly favourable antifouling properties at low concentrations (nano
Molar). To obtain full protection against barnacle fouling, 0.1 - 0.3% w/w of Selektope® should be
used in a wet paint formulation. Just 2 grams Selektope® is used per litre of paint, comparable to 500-
700 grams of copper oxide used per litre of paint for barnacle prevention.

Selektope® binds to pigment and other particles in the paint system and is therefore continuously
released in the same way as other active substances and components. This contributes to long-term
performance as long as the paint remains on the hull. The paint formulation, which mainly comprises
binding agents, biocides, pigment and filler material, is applied to the hull using a traditional spraying
method. The compatibility between Selektope® and the paint matrix in the marine coatings industry,
ensures as slow and steady release secures the antifouling effect over time.

However, how and when Selektope® is added during the formulation process is key to controlling the
release rate of Selektope® from an antifouling paint. To prevent premature depletion of Selektope®
the molecule should be able to interact with a carrier in the paint mixture. A carrier could be an inorganic
particle such as zinc- or cuprous oxide. It could also be a metal ion such as Zn2+ or Cu2+, or an acid
group on a binder, for example the carboxylic acid on rosin.

I-Tech advises that Selektope® should be added early in the process, rather than adding it post
formulation. I-Tech also advises that Selektope® should be added as a solution in a suitable solvent.
Preferably the Selektope® solution and the carrier should be mixed first and then the rest of the
components can be added. Selektope® will adhere to metal ions and metal oxide such as zinc oxide
and cuprous oxide. This has been shown to be an effective way to control the release of Selektope®
and prevent premature depletion.

Inorganic materials such as Al2O3, SiO2, CuO, ZnO, TiO2 and MgO can be used. Cuprous oxide, zinc
oxide and iron oxide are commonly used. ZnO is advised as the best pigment particle for maximum
Selektope® adsorption in xylene.

If Selektope® is added later in the paint formulation process the surface of the metal oxide pigment
may already be occupied and Selektope® adsorption will not take place in an adequate or linear way
with uneven distribution and weaker adhesions. This may cause Selektope® to leach out of the paint
too quickly and result in premature depletion of Selektope® from the treated surface.

Fig.3. Selektope® added before the binder and other components - good adhesion will occur versus if
Selektope® added at the end of the formulation process weaker or no adhesion occurs
potentially leading to premature depletion of Selektope®

Although most Selektope®-containing antifouling paint products on the market are combinations of
copper oxides and Selektope®, Chugoku Marine Paints have launched a paint that is copper free.
Therefore, the concentration of biocides in the paint has been reduced, while other qualities, such as
prevention of soft fouling (e.g. slime and seaweed) have been notably improved.

304
4. Current market uptake and in-service performance data

Application data, as provided by paint manufacturers, suggest that to-date the total of ships coated
with a Selektope®-containing antifouling totals over 300. There are 10 antifouling coating products
available to the market that contain Selektope® (correct as of April 2019).

In this paper, a selection of case studies that demonstrate the hard fouling prevention effectiveness of
Selektope®-containing antifouling coatings are presented.

4.1. CASE STUDY: 36-month medium range tanker trial

The vertical sides of the IMO II 2009-built chemical and products tanker Team Calypso were coated
with a copper free Selektope®-based antifouling product with a service life of 60 months. At the 36-
month position in its drydock interval Team Calypso had a hull completely free of barnacle fouling
after spending more than 50% of its operating time spent in areas of high biofouling with up to 32 oC
water temperatures. The tanker had also encountered several extended idling periods of 25 days or
more.

Month 27: Vertical Side Month 27: Uncoated Letters Month 27: Uncoated Ballast
Suction
Fig.4: Team Calypso dive inspection report, Month 27

Month 32: Flat Bottom With Month 32: Uncoated Ballast Month 32: Flat Bottom
Patch Of Uncoated Surface Suction
Fig.5: Team Calypso dive inspection report, Month 32

Month 36: Vertical Side Month 36: Vertical Side Month 36: Vertical Side
Fig.6: Team Calypso dive inspection report, Month 36

305
resistance also reinforced the underwater hull inspection findings at month 36. Data analysis confirms
that the added
compared to that expected for a reference ship of similar age, size and trading patterns.

The independent data for Team Calypso at Month 36 is described in Fig.7.

Total added resistance 12 %

Hull added resistance 8%
Propeller added resistance 4%
Development rate of added resistance (normally 0.4 %
0.5% to 1.5%)
Excessive fuel consumption since last drydock 2.7 t/24h
at loaded condition
Months since latest dry-docking 36
Months since latest hull cleaning 36
Months since latest propeller polish 6
Fig.7: Team Calypso added resistance data: Month 36

The performance of the Selektope®-containing SPC applied to Team Calypso was compared to two
other sister ships in the fleet (ship A and ship B). Ship A coated with a foul release coating type and
ship B coated with an SPC type. The added resistance data for each ship is presented in Fig.8.

Fig.8: Team Calypso development of hull/propeller added resistance versus sister ships

provide yet more convincing long-term performance results for Selektope®-based coatings. These

306
results hold great importance as they confirm the superior efficacy of Selektope® for a vessel engaged
in active service that encounters significant exposure to severe fouling conditions and undertakes
periodic idling activity. This is proof that I-
using Selektope®-based antifouling coatings superior hard fouling prevention for any vessel regardless
of its activity and trading patterns.

4.2. CASE STUDY: US Navy static panel tests

The Office of Naval Research Biofouling Program supports field testing at five sites located in Florida,
California, Hawaii and Singapore. These sites are characterised by different environmental conditions,
and different biofouling communities. Across the sites, however, any coating under evaluation will
experience biofouling by the full suite of organisms likely to recruit on a ship hull.

The antifouling performance of a number of coating formulations have been examined during a test
period of 12 months using coated test panels under static conditions at the five aforementioned test
sites. Each coating formulation is a different product sourced from multiple paint manufacturers.

The first copper based SPC type containing Selektope® supports no macrofouling, with no barnacle
growth, or slimes at most of the sites after 12-months static. Some panels at one of the Florida tests
sites and at the Singapore site were covered with a layer of slime. At the 6-month point, panels at the
California test site also had very little slime growth.

The second coating formulation, a copper-free SPC type containing Selektope®, after 12 months static
at all test sites supported no macrofouling, with no barnacle growth. At one of the Florida tests sites,
macrofouling, not barnacles, was present and some scattered amphipod tubes. Panels at the Singapore
test site supported slightly heavier slime.

For the testing of third coating formulation, one set of panels were coated with a copper-based SPC and
one set of panels were coated with a copper-based SPC type containing Selektope® at each test site
location, after 12 months there was a clear difference in performance exhibited between the two sets of
panels coated, with and without Selektope®. At the Florida test sites the set of panels with Selektope®-
containing paint being free of barnacles, while scattered barnacles were present on the set of panels
coated with paint not containing Selektope® and both sets of panels also supporting light to moderate
slime and scattered amphipod tubes. At the Hawaii tests sites, both sets of panels were essentially clean,
while those at the Singapore test site slime was slightly heavier. After 6 months, panels coated with
paint containing Selektope® at the California test site remained barnacle free, whereas panels coated
with paint not containing Selektope® supported barnacle growth.

This test provided the conclusion that the presence of Selektope®® in antifouling coatings provides
superior hard fouling prevention performance, even under static conditions of 12-month duration. This
test also demonstrated that copper-free antifouling coatings containing Selektope®® also offer superior
colour retention to those copper-containing SPCs tested under the same conditions. This test will
continue in 2019.

5. R&D projects

The challenges with cleaning submerged surfaces and equipment requires the development of new
materials or combinations of materials that withstand cleaning and other kinds of mechanical wear.
There is also submerged equipment that could benefit from having an antifouling coated surface that
currently may only be coated with ordinary marine paint for corrosion protection. Fenders, buoys, nets,
cables, measuring/monitoring devices and energy production devices can suffer from fouling, for
example.

I-Tech, together with industry partners, has starting screening suitable materials either as coating or
construction materials for submerged objects that are:

307
 Durable to withstanding cleaning / mechanical wear
Slippery: to release fouling
Permeable: to release biocides

Materials already available that can be mixed or tweaked to improve fouling resistance are:

Polyurethanes
Silicone epoxies

In this paper, research conducted by I-Tech in the areas of Polyurethanes and Silicone epoxies
containing Selektope® is presented.

5.1. CASE STUDY: Polyurethanes with Selektope®

I-Tech has demonstrated that it is possible to protect submerged surfaces/object that are made out of
polyurethane material. Test results have confirmed the prevention of barnacle settlement under static
conditions over two seasons (summers).

This test provided the conclusion that the presence of Selektope® in polyurethane material successfully
protects against barnacle fouling.

Selektope® concentration
0% 0.1% 0.2% 0.5% 1% 0%

Fig.13: Reference polyurethane test panels

Selektope® concentration
0% 0.1% 0.2% 0.5% 1% 0%

Fig.14: Polyurethane test results: May 2013 - September 2014

308
Fig.15: Three-panel polyurethane test conducted with only one formulation containing 0.37%
Selektope® versus a reference panel with 0% Selektope® - October 2015 - November 2017

5.2. CASE STUDY: Silicon epoxy coatings

Whereas epoxy-silicone systems offer abrasion resistance and easy to clean properties, they still have
limitations when exposed to sea water at static conditions. The addition of Selektope® enables highly
innovative coating systems and, as such American paint specialist Wearlon®, part of Plastic Maritime
Corp., trialled the inclusion of Selektope® in its Wearlon Super F-3M range to develop an abrasion-
resistant coating with boosted antifouling properties. The trials concluded that the addition of
Selektope® to Wearlon® -silicone coating resulted in significantly improved antifouling
properties when exposed to raft tests in sea water. The coating is expected to be particularly effective
in areas where high-wear strength is required such as around propellers, and for use on off shore
equipment.

Wearlon® -3M is a water-based silicone-epoxy coating whose silicone epoxies are blocked
copolymerised and supplied through emulsion chemistry. The mol % of silicone to epoxy is varied and
coupled with surfactant and filler selection to obtain properties that are unique for a variety of
applications. Wearlon® Super F-3M is one of the most hydrophobic of all the Wearlon® coatings,
having a low surface tension when in contact with water, resulting in drag reduction. Because of its
excellent abrasion and corrosion resistance it is being used in a variety of marine applications including
zebra mussel control.

Developing a coating that was inclusive of Selektope® has been key to producing a more impact-
resistant, yet flexible and low abrasive epoxy silicone coating system that shows excellent antifouling
performance.

The trialing of the Wearlon® epoxy-silicone product containing Selektope® was conducted at I-
test facility on the west coast of Sweden over a period of 19 months. Test panels were coated with two
layers of the test product by a roller. For the formulations that underwent panel testing, a concentration
of 0.3% Selektope® w/w of the final wet paint mixture was used to ensure the steady release of the
biocide.

309
A successful formulation was achieved through the addition of a 1-metoxy-2-propanol solution of
Selektope® to component B of the two-component system of Super F-3M. Traditionally, when
formulating rosin gum-based or acrylic-based antifouling paints, Selektope® is always best added to
the paint as a solution. In order to add it to the formulation, it should be dissolved in an appropriate
solvent before adding pigments, metal oxides (e.g. ZnO), fillers, binder and other additives.

Fig.16: Reference silicon epoxy-coated test Fig.17: Month 18 results

panels submerged in April 2016 off the
west coast of Sweden.

At the 2-month inspection, all reference panels had plenty of barnacle fouling. Some of the test panels
had a few barnacles attached except F-3M with 0.3% Selektope®. At 18 months, panels coated with F-
3M with addition of 0.3% Selektope® had no fouling, whereas every other test panel was completely
covered in barnacle fouling.

The trial of the Wearlon® Super F-3M range product with Selektope® included concluded that the
addition of Selektope® to the Wearlon® epoxy-silicone coating resulted in significantly improved
antifouling properties when exposed to raft tests in sea water. In the future, the partners are convinced
that epoxy-silicone coating technology will achieve wider acceptance on ocean going vessels due to
Selektope®-powered antifouling performance.

6. Conclusion

Selektope® is currently being used on over 300 vessels and I-Tech anticipate this number growing
significantly in the near future. Independent data analysis of added resistance on the ship hull, dive
inspections and static panel tests confirmed, and continue to confirm, the barnacle repellent power of
this biotechnology when used in marine antifouling coatings. Further research efforts conducted by I-
Tech confirm that it is possible to add Selektope® to a variety of materials not only to traditional marine
paints. This can broaden the scope of use to other areas that are not coated today, but that would benefit
from protection against barnacles. I-Tech intends to continue looking for new materials where
Selektope® can be incorporated as well as enabling the use of the biotechnology in all currently used
antifouling solutions on the market today and in the future.

310
Acknowledgement

We thank the University of Gothenburg, the United States Navy and Plastic Maritime Corporation.

References

DAHLSTRÖM, M.; MÅRTENSSON, L.; JONSSON, P.; ARNEBRANT, T.; ELWING, H. (2000),
Surface active adrenoceptor compounds prevent the settlement of cyprid larvae of Balanus improvises,
Biofouling 16(2), pp.191-203

311
 Index by Authors

Arango 143 Mathew 254

Ashida 38 Matsubara 21
Austin 301 McLaughlin 220
Becchi 76 Mieno 21
Bellusci 162 Mizutani 38
Bertelsen 114 Montazeri 211
Bertram 9 Montrose 272
Bonello 128 Murrant 272
Bretschneider 65 Mushiake 172
Castagna 254 Nielsen 229
Chabaane 301 Noh 86
Connelly 201 Östman 153
Daniel 254 Ohlauson 301
De Freitas 201 Paereli 261
Dietz 229 Pakkanen 38
Doran 97 Pallard 272
Eliasson 4 Pappas 201
Galeazzi 229 Park 193
Gangeskar 114 Patey 31
Gatin 13 Petersen 211
Giordamlis 56 Petoumenos 179
Gonzalez 143 Prytz 114
Gorski 292 Rahimi 254
Gundermann 220 Savio 153
Hansen 211 Schmode 162
Helsloot 50 Schrader 65
Holst 229 Shan 242
Holm 242 Shimada 21
Hympendahl 162 Shin 193
Igarashi 172 Sia 103
Ikonomakis 229 Silberschmidt 201
Inukai 172 Smith 128
Isaksson 301 Sogihara 21
Ishiguro 21 Spandonidis 56
Jasak 13 Stefanatos 179
Jeon 86 Sugimoto 21
Kauffeldt 179 Teo 103
Kawanami 172 Themelis 56
Kennedy 272 Väisänen 38
Khan 254 Van Ballegooijen 50
Ki 193 Vargas 242
Kluwe 65 Voosen 254
Koushan 153 Vukcevic 13
Lee 86,193 Wakabayashi 172
Levantis 261 Weigenand 301
Lim 103 Yonezawa 21,172
Manderbacka 38
Marquardt 254
Masuyama 21

312