Chemical Engineering Research and Design 1 6 7 ( 2 0 2 1 ) 281–291

Contents lists available at ScienceDirect

Chemical Engineering Research and Design

journal homepage: www.elsevier.com/locate/cherd

Integrating uncertainty quantification in reliability,

availability, and maintainability (RAM) analysis in
the conceptual and preliminary stages of chemical
process design

Ahmad Al-Douri a,∗ , Vasiliki Kazantzi b , Nancy Currie-Gregg c ,

Mahmoud M. El-Halwagi a,d
a Department of Chemical Engineering, Texas A&M University, College Station, TX 77843, USA
b Department of Business Administration University of Thessaly, 41500 Larissa, Greece
c Department of Industrial and Systems Engineering, Texas A&M University, College Station, TX 77843, USA
d Gas and Fuels Research Center, Texas A&M Engineering Experiment Station, College Station, TX 77843, USA

a r t i c l e i n f o a b s t r a c t

Article history: Traditional analysis of a proposed process design uses average input values in the
Received 26 November 2020 performance assessment model, thereby generating single-point estimates. The result-
Received in revised form 13 January ing estimates ignore reliability, availability, and maintainability (RAM) considerations, or
2021 assume a fixed value based on prior experience. As a result, a probabilistic view of the impact
Accepted 18 January 2021 of equipment unavailability on process profitability is not considered. Recent works have
Available online 27 January 2021 proposed a financial framework for incorporating safety and sustainability considerations
in the analysis of proposed designs. Based on this research, we propose a framework to inte-
Keywords: grate RAM aspects during the conceptual design stage in a probabilistic manner using Monte
Reliability, availability, and Carlo simulation. Subsequently, full distribution profiles of key process performance indica-
maintainability (RAM) tors are generated, including system and section availability, annual net profit, and return
Process design on investment (ROI). Probabilistic characterization of equipment availability also facilitates
Uncertainty quantification the prediction of potential safety and sustainability issues, as more frequent process upsets
Monte Carlo simulation may result in increased flaring and other potential negative consequences. A modified avail-
Information modelling ability metric, using restoration instead of repair times, is used in this work to obtain a more
accurate view of expected downtime and thus its effects on profitability. A propane dehy-
drogenation (PDH) process system is used to demonstrate the application and benefits of
the framework. The proposed approach allows designers and decision-makers to compre-
hensively assess the impacts of equipment RAM characteristics on process availability and
economic performance.
© 2021 Institution of Chemical Engineers. Published by Elsevier B.V. All rights reserved.

1. Introduction further narrowed down by a variety of constraining factors. While some

of these constraints are outside the design team’s control (e.g., physi-
The chemical process design methodology involves determining spe- cal laws, government regulations, engineering standards), other critical
cific objectives or customer needs, then developing and evaluating factors, such as the choice of process or process conditions, materi-
possible design solutions to ensure they satisfy those objectives in the als, and equipment, are open in the trade space. Among the major
best possible way. The number of feasible design alternatives is then constraints on any engineering design are economic considerations
since prospective plants must also be profitable. Further, the number
of alternative designs that can be evaluated is often also constrained
by schedule considerations.
∗
Corresponding author.
E-mail address: [email protected] (A. Al-Douri).
https://doi.org/10.1016/j.cherd.2021.01.014
0263-8762/© 2021 Institution of Chemical Engineers. Published by Elsevier B.V. All rights reserved.
282 Chemical Engineering Research and Design 1 6 7 ( 2 0 2 1 ) 281–291

Li and Kraslawski (2004) presented an overview of the development compute the capital cost of each equipment piece. However, it is further
of conceptual process design in the context of three scales: micro- noted that it can be challenging to obtain real-world data on the link
(molecules), meso- (unit operations), and macro-scale (plant). At the between capital cost and inherent availability. Sharda and Bury (2008)
macro-scale, the authors identified the main research issues that will presented a discrete event simulation model to understand the impact
dominate conceptual design to be: (1) combining knowledge of differ- of different critical subsystems on overall plant production capabili-
ent disciplines, (2) dealing with uncertainties at the top decision level, ties using historical failure data. The disadvantage of using a discrete
and (3) developing optimization and simulation techniques for com- events approach is that it does not guarantee optimal solutions. Aguilar
plex systems. In this work, the proposed addition to the traditional et al. (2008) considered redundancy alternatives to address reliability
process design approach is the integration of reliability, availability, issues in utility plants. Recent works have utilized Markov chains to
and maintainability (RAM) considerations in the economic evaluation model utility systems and production sites and determine their opti-
of design concepts for a proposed plant while taking into account the mal designs (Lin and Chang, 2012; Terrazas-Moreno et al., 2010). In
uncertainties involved in the process system. addition, recent studies have utilized optimization methods to deter-
mine optimal allocation of redundancy (Ye et al., 2018; Andiappan
et al., 2019). Al-Douri et al. (2020) introduced an economic framework
1.1. Safety and sustainability considerations in process design
for failure mitigation during the conceptual design stage utilizing a
Bayesian updating procedure to update generic failure rate data with
Within the approach outlined above, the evaluation of process tech-
plant-specific data.
nology is usually based purely on techno-economic analysis. However,
this approach excludes consideration of other critical aspects, such
as safety and sustainability, until after the detailed design has been
1.3. Uncertainty quantification
completed. Alternatively, these considerations can be integrated earlier
into evaluations of the design of the process. For example, inher-
The previous survey shows that significant progress has been made
ent safety principles are applied in the early stages of design (e.g.
in the inclusion of safety, sustainability, and reliability considerations
screening of alternative technologies) to enhance the safety charac-
as part of the evaluation of the design of chemical processes. How-
teristics of the process (Rahman et al., 2005; Kidam et al., 2016; Park
ever, these performance models are based on uncertain input variables
et al., 2019). Sustainability considerations, such as greenhouse gas
and use average values leading to conclusions with significant levels of
emissions, total waste rate, and water footprint, have been used as
uncertainty regarding the overall performance profile of the process
indicators to assess the environmental impact of different process
design. The use of average values is not recommended based on the
designs (Martinez-Gomez et al., 2016; Julián-Durán et al., 2014; Yang
“flaw of averages” concept in probability theory, which underscores
and You, 2017). Other process evaluation techniques include these con-
that evaluating process performance at average conditions does not
siderations, but primarily from an economic perspective. El-Halwagi
necessarily lead to an average process performance. In reality, differ-
(2017a) developed a sustainability-weighted return on investment met-
ent values for an input variable can have different consequences on the
ric (SWROIM) to evaluate the viability of process improvement projects
performance metric being examined. To address this, Monte Carlo (MC)
and their impact on sustainability relative to the requisite capital
simulation methods can be used to develop a model of the system in
investment. Guillen-Cuevas et al. (2018) extended the previous work by
which uncertainties associated with key design variables are included.
including safety considerations to develop a safety-and-sustainability-
Seila et al. (2003) and Law and Kelton (2007) provided comprehensive
weighted return on investment metric (SASWROIM). In addition to
reviews of simulation modelling methods and applications to opera-
these metrics, another approach is the use of loss functions to model
tions research and finance. Monte Carlo techniques have been used to
the economic consequences of process unit deviations from target val-
analyze complex industrial systems in areas such as project portfolio
ues (Hashemi et al., 2014; Khan et al., 2016).
management, finance and accounting, and operational risk evaluation
(Savage, 2003; Kuppens et al., 2018). Kazantzi et al. (2013) presented a
1.2. Reliability, availability, and maintainability (RAM) systematic solvent selection approach for a safety-constrained system,
considerations in process design using MC simulation to evaluate system performance in the presence
of economic and regulatory uncertainties. Kazi et al. (2018) used an
Similar to other engineering systems, the expected function of a chem- optimization framework and MC simulation to explore the effects of
ical process plant can be interrupted due to failure(s) in one or more of uncertainty in flaring sources of an ethylene process on the selection
its units. Some systems or units are less inclined to fail, usually due to of flare reduction alternatives. Specific applications of MC simulation
inherent design characteristics; thus, they are considered more reliable. to systems availability analysis have been demonstrated for a cool-
While reliability engineering is relevant to all aspects of the design and ing tower pump system (Alexander, 2003) and an offshore installation
operation of a process plant, applying reliability analysis and predic- (Zio et al., 2006). Recent work has also combined structural reliability
tion techniques at the conceptual and preliminary stages of the design techniques and MC simulation to formulate stochastic performance
process significantly aids the design team in the technology selection models for optimizing the design and operational reliability of chemical
decision. By analyzing the possible failure modes and mechanisms of processes (Abubakar et al., 2015a, b, c).
process units or subsystems, the consequences of these failures and
their effects on the economic potential or profitability of the design can
be estimated. These consequences include revenue losses as a result 1.4. Problem formulation
of production slowdown or shutdown due to failure; costs of different
maintenance regimes including labor (corrective, preventive, or pre- The constraints on the solutions to a process design problem are both
dictive); environmental damages due to a large quantity of flaring; loss external and internal, but the decisive factor is the economic perfor-
of life or injury to personnel or the public; and costs associated with mance, as plants must be economically viable. Traditional economic
environmental damages. performance assessment and valuation approaches have been proven
The problem of integrating RAM assessment and optimization into inadequate when applied to complex engineering systems, especially
the conceptual design phase has been an active research area for in the early design phase, leading to insufficient outcome character-
many years. Early works examined the sensitivity of system reliabil- ization (de Neufville and Scholtes, 2011; Savage, 2003). This kind of
ity to parallel redundancies (Rudd, 1962; Henley and Gandhi, 1975). economic appraisal framework is usually based on specific economic
More recently, Goel et al. (2003) considered the effects of availability metrics, such as ROI, IRR, NPV etc. at average conditions and does
and maintenance in chemical plant economics and concluded that not correspond to realistic representations of uncertain input variables
revenues and operational costs must be affected by the system inher- (such as equipment RAM characteristics in this case), which could have
ent availability. Their approach used an exponential relationship, first asymmetric impacts on economic performance outcome. Hence, there
reported by Ishii et al. (1997), between investment and availability to is a need to provide an explicit way to embed and quantify uncertain
 Chemical Engineering Research and Design 1 6 7 ( 2 0 2 1 ) 281–291 283

conditions in a process system, especially when RAM considerations where SDN is the failure rate standard deviation reported in
are of particular importance at the early design stage. the new edition, and o and N are the failure rate values from
In this work, the aim is to present an integrated framework which the old and new editions, respectively. In this work, the expo-
combines knowledge from the fields of process synthesis and design nential distribution is used for failure and restoration times,
with reliability engineering into the economic analysis of process
meaning the failure and restoration rates are constant. Thus,
design alternatives in a probabilistic manner. RAM considerations will
the MTBF (mean-time-between-failures) value is the inverse
be included by collecting failure and repair data from two versions of
of the weighted failure rate.
the Offshore and Onshore Reliability Data (OREDA) Handbook—OREDA-
02 and OREDA-15. Furthermore, the proposed approach enables the Conventional analysis of inherent availability utilizes the
simultaneous inclusion of various sources of irreducible uncertainty mean time to repair (MTTR) metric to represent the time
(equipment RAM characteristics) as multiple model input (random) that equipment or a system is not in operation. However, a
variables that becomes feasible (as opposed to the conventional sensi- more encompassing metric is the mean time to restoration
tivity analysis where one model input at a time is considered varying). (MTTRes), which is the time needed to restore an item to full
As a result, full distribution profiles of economic performance out- operational status. It involves delays prior to and after repair
comes are derived in the presence of uncertainty. The derived profiles actions, as well as time needed to ramp down and start up
are amenable to insightful statistical characterization, and risks and
an equipment or system in the case of a failure event. The
opportunities regarding the equipment and process design features can
failure and restoration cycle described here is illustrated in a
be identified and actively managed. According to the methodology fol-
supplementary figure provided with this work.
lowed in this study, all the uncertain model input variables are first
identified and reasonable probabilistic representations through appro-
The MTTRes metric captures important factors involved
priately selected distribution profiles are assigned to them. Applying in the restoration process which can decrease the overall
standard Monte Carlo techniques and performing random sampling availability of a system and are important to consider when
from the above distributions, model input uncertainties are propagated assessing its profitability. These factors include: ability of plant
through the model, and distribution profiles are generated that can personnel to identify and mitigate failure events, responsive-
be probabilistically characterized. The framework will be applied to a ness of management in allocating resources for maintenance
case study of a proposed propane dehydrogenation (PDH) plant using needs, and the skill level of maintenance personnel. For a
a RAM-weighted return on investment metric. system in series, a modified system availability, Atotal , using
MTTRes can be estimated from the following equation:
2. Description of proposed framework

n

n 
m
MTBFk
Atotal = Ai = ( ) (2)
2.1. Classical process design and evaluation MTTResk + MTBFk
i=1 i=1 k=1 i

The framework presented in Fig. 1 begins with the same steps where Ai is the availability of a block, and MTBFk and MTTResk
commonly used in the design process. These steps include are the mean time between failures and mean time to restora-
identification of design objectives and sub-objectives, trans- tion, respectively, of the equipment making up the block.
lation of customer needs into a design basis, development Then, the block with the lowest availability is identified as the
of block flow diagrams (BFDs) for promising alternative tech- critical process step causing the most significant downtime
nologies, performing of stoichiometric targets of feed and and having the most adverse impact on process profitability.
product rates, and determination of profitability estimates for
the alternative technologies. Evaluation of an alternative cul- 2.3. RAM-weighted economic evaluation
minates in an estimate of a return on investment (ROI) value.
The alternatives that do not meet the company’s acceptable Using the estimated system availability range, a modified
threshold ROI value are eliminated, and those that meet the return on investment metric (ROI) range can be calculated as
criterion are considered for further study. follows:
 
2.2. Estimation of system RAM levels for proposed  %  ANP $
year
ROI =   ∗ 100 (3)
design year TCI $

At this stage of the design activity, the integration of reliability, where ANP is the annual net (after-tax) profit and TCI is the
availability, and maintainability (RAM) factors is considered. In total capital investment. This profit can be calculated by the
order to estimate the availability of each process, preliminary following equation, modified from the profit equation in El-
knowledge of the equipment to be used for each major pro- Halwagi (2017b):
cessing step is needed. This can be obtained through process
flow diagrams (PFDs) of previous designs within a company

ANP = Availability ∗ (Annual Revenue − Annual Operating Cost)
or from literature references. Equipment failure rates (lower, 
mean and upper) and restoration manhours data (mean and −Depreciation ∗ (1 − Tax Rate) + Depreciation (4)
maximum) for the proposed PDH plant were derived from
two versions of the OREDA Handbook for Onshore and Off-
shore Reliability Data (OREDA, 2002, 2015). A weighted failure The purpose of this modified profitability metric is to deter-
rate, ˆ , was then calculated using the formula described by mine the inherent RAM characteristics of a process design
Vatn (1993): alternative and their effect on profitability estimates. Differ-
ent design modifications to improve RAM levels of critical
2 equipment are then assessed to determine their feasibility and
2o + 2N (o /N + |o − N | /SDN )
ˆ = (1) the extent to which they enhance process profitability. These
2
o + N (o /N + |o − N | /SDN ) modifications may include the quality of purchased process
284 Chemical Engineering Research and Design 1 6 7 ( 2 0 2 1 ) 281–291

Fig. 1 – Framework for inclusion of reliability, availability, and maintainability (RAM) factors in process design.

equipment, over-design of equipment, addition of redundant equipment RAM characteristics (MTBF, MTTRes), a design
components or subsystems, preventive and predictive mainte- team is able to obtain a range of values for process system
nance plans, and measures to improve human reliability. The availability and economic metrics such as annual profit and
choice of design modification is made based on calculating its return on investment. For each major piece of equipment, both
incremental return on investment (IROI) as follows: failure and restoration data represent uncertain variables and
  were assigned triangular distributions, selected due to limited
 %  ANP $
year
data, as follows:
IROI =   (5)
year TCI $
TriangularFailure (MTBFMinimum , MTBFMean , MTBFMaximum ) (6)
2.4. Uncertainty quantification approach

Fig. 2 below illustrates the approach used to determine the

TriangularRestoration (MTTResMinimum , MTTResMean , MTTResMaximum )
expected ranges for process RAM characteristics and prof-
itability. Using a representative distribution for the constituent (7)
 Chemical Engineering Research and Design 1 6 7 ( 2 0 2 1 ) 281–291 285

Fig. 2 – Uncertainty propagation of process RAM characteristics.

2015, and is expected to increase to 4.5 MMT in the period of

In this manner, the uncertainty inherent in RAM character- 2017−2021. With the decline in propylene production due to
istics is propagated through the economic evaluation, thereby lighter feedstock in crackers, it is projected that approximately
allowing decision-makers in technical and business divisions 30% of propylene production (38 MMT) will be produced by
of an organization to make better-informed decisions about on-purpose technologies in 2023. In North America, propylene
potential projects. demand is expected to reach 20 MMT (about 15% of global sup-
ply), of which 5 MMT is expected to be via on-purpose routes.
Currently, there are three PDH plants in the U.S. producing
3. On-purpose propylene production
about 2 MMT of on-purpose propylene (Galindo, 2019). These
factors indicate that on-purpose propylene technologies will
3.1. Propylene production pathways
be essential to meet North American demand. Thus, this infor-
mation establishes the need and commercial opportunity that
Propylene is the petrochemical substance with the second-
exist for on-purpose propylene. In the PDH process, propane is
largest production volume after ethylene, and an important
converted to propylene over a platinum-alumina-based cata-
raw material for producing polypropylene and propylene
lyst bed at high temperatures (550−650 ◦ C) and low pressures
oxide, both major building blocks in plastics production. Tra-
(2−3 bar). Overall selectivity towards propylene is 90 mol%,
ditionally, propylene has been produced as a byproduct of
and one-pass conversion is 40 mol% (Gregor and Wei, 2004).
ethylene from steam cracking of hydrocarbons, or as a byprod-
The PDH licensing is dominated by Honeywell UOP’s OLEFLEX
uct of gasoline in fluid catalytic cracking (FCC) in refineries.
and CB&I CATOFIN technologies. Of the 28 PDH units operat-
In 2012, steam cracking and FCC accounted for 90% of the
ing worldwide, 18 use the OLEFLEX technology. In this work,
global production of propylene. However, the shale gas revolu-
the OLEFLEX process was selected for examination.
tion has caused a shift in North American cracker feedstocks
from naphtha to ethane, leading to an increase in ethylene 4. Application of proposed framework
yield and a decline in propylene production. For refineries,
gasoline prices dictate the propylene production in those 4.1. Propane dehydrogenation OLEFLEX process
facilities. If prices are high, propylene is produced and used
to make octane-boosting alkylates. If demand is low, gaso- To begin the plant design process, a block diagram of the PDH
line production is lowered causing propylene output to fall. process is constructed showing the major processing steps in
In these two pathways, propylene supply, unlike propylene each route (Fig. 3). The PDH process consists of the follow-
demand, is subject to developments in the gasoline and ethy- ing major processing equipment and operations: depropanizer
lene markets. That demand is expected to grow from 109 column, dehydrogenation reactors and interheaters, reac-
million tons in 2014 to about 165 million tons in 2030, approx- tor effluent compressors and coolers, cryogenic refrigeration,
imately 12–14% greater than the amount of propylene that pressure swing adsorption (PSA), propylene/propane sepa-
can be produced as a by-product of steam crackers and in ration. Fresh and recycled propane enter the depropanizer
refineries (Wood Mackenzie, 2014). The resulting gap between column where hydrocarbon (C4+) material is separated. The
supply and demand can be filled by on-purpose propy- propane-rich stream in the overhead enters the refrigeration
lene (OPP) technologies such as propane dehydrogenation, unit where its autorefrigeration property is utilized to cool
methanol-to-propylene/methanol-to-olefins (MTP/MTO), and the reactor effluent stream. The propane stream leaving the
olefin metathesis. The optimal method of producing propy- cold box is mixed with hydrogen and enters a fired heater
lene is subject to factors such as geographic location, feedstock where the mixture is heated to a temperature to 580−620
prices and availability, market conditions, technology matu- ◦ C before entering the reactor system. The reaction occurs

rity, sustainability and safety (Guillen-Cuevas et al., 2018; Roy over a fluidized catalyst bed in a radial flow reactor to mini-
et al., 2016). mize pressure drop across the beds (Vora, 2012). To burn off
resulting coke that is formed, a continuous catalyst regen-
3.2. Propane dehydrogenation technologies erator (CCR) is used. The reaction is highly endothermic in
nature (H = 124.3 kJ/mol) and a considerable temperature
According to (Eramo, 2017), propylene demand grew at an drop occurs in each reactor. Therefore, interstage heaters are
average rate of 3.5 million metric tons (MMT) from 2011 to needed to raise the outlet stream temperature. The reactor
286 Chemical Engineering Research and Design 1 6 7 ( 2 0 2 1 ) 281–291

Fig. 3 – Block diagram of UOP OLEFLEX propane dehydrogenation (PDH) process.

system effluent is a mixture of propylene; unreacted propane;

light gases, such as methane, ethane and ethylene; diolefins;
and heavier hydrocarbon components formed in the reactor.
The effluent is then cooled and compressed from 96.5 kPa psia
to about 1380 kPa in a multistage compressor with interstage
coolers. The next step is to liquefy the hydrocarbon material
and separate out the hydrogen in the cold box unit. Part of
the hydrogen produced is sent to the dehydrogenation reac-
tors and the rest goes to the selective hydrogenation process
(SHP) reactors. The net hydrogen is then sent to the pressure
swing adsorption (PSA) unit to ensure it meets pipeline qual-
ity specifications. The liquefied hydrocarbons from the cold
box are sent to SHP reactors to convert the diolefins to propy- Fig. 4 – Probability density graph of overall PDH process
lene. Then, a deethanizer column separates C2 components availability.
from the liquefied hydrocarbons and propane-propylene split-
ter (superfractionator) produces the final propylene product.
Unconverted propane is sent from the superfractionator bot-
toms to the depropanizer column as part of the feed.

4.2. High-level analysis of block flow diagrams

A preliminary economic assessment of the proposed plant

is performed using stoichiometric targeting to estimate the
amount of raw materials needed for the process, operating
expenditures, annual revenue and the capital investments.
The capital and operating expenditures for the 450,000 MTA Fig. 5 – Probability distribution profile for the modified ROI
plant are estimated on the basis of the cost data in Agarwal of proposed PDH process.
et al. (2018), which reported on a simulation study with capital
and operating investment estimates for a 600,000 MTA plant major equipment piece, processing section, and the overall
located on the U.S. Gulf coast. Those values and the prices for process are not point estimates, but rather a range of values
raw materials and product used in this study are shown in that can be fit to a probability distribution using the @Risk
Table 1. software.
Fig. 4 shows the probability density distribution for the
4.3. Process design performance evaluation process system’s availability. These results show that the stan-
dard deviation is low, indicating that the upper and lower
4.3.1. Estimation of system RAM levels for proposed confidence limits do not differ significantly from the mean
design value of total inherent availability (72.7%). The result is a range
After the major equipment in each processing step are of estimated system availability that is significantly different
identified, failure and restoration manhours data for these from projections assuming a fixed system availability value
equipment are collected from two versions of the OREDA of 80% or 85%. Greater accuracy with respect to system avail-
database. Table 2 includes this data for the equipment in a ability estimates allow for more accurate outlooks of a plant’s
propane dehydrogenation (PDH) process. The ranges for MTBF projected profitability. Further refinement of the availability
and MTTRes values in this table are representative of different estimate is possible if in-house failure and repair data is avail-
installations as well as varying environmental and operational able to the design team.
conditions under which the data was collected.
The Monte Carlo simulation tool in the Palisade® @Risk 4.3.2. RAM-weighted economic evaluation of proposed
software was used to generate 50,000 random samples for design
each equipment’s MTBF and MTTRes values. Two sampling Uncertainties in equipment RAM characteristics are included
methods are available for use in this simulation feature: Latin in the calculation of the process profitability using the modi-
Hypercube and Monte Carlo. For this work, Latin Hypercube fied return on investment (ROI) metric, as described previously.
sampling was used as the sampling method because it avoids A Monte Carlo simulation was performed to obtain a distri-
the clustering that can occur with Monte Carlo sampling, pro- bution profile for this indicator. The probability distribution
viding a better chance for all values from the input distribution profile for the modified ROI is shown in Fig. 5. The results esti-
to be sampled. Eq. (2) is used to calculate availability for a sub- mate a mean ROI value of 23.9 %/year and a standard deviation
system or system in series. The resulting availability for each value of 7.45%/year. The coefficient of variation, defined as the
 Chemical Engineering Research and Design 1 6 7 ( 2 0 2 1 ) 281–291 287

Table 1 – Feedstock/product prices and economics for PDH process.

Feedstock/Product Minimum Flowrate in Price ($/MT)
Metric Tons/Annum
(MTA)

Propane 524,000 375

Propylene 450,000 1100
Hydrogen 22,000 1600

Process economics results

Fixed capital investment (FCI), $MM 534
Total capital investment (TCI), $MM 628
Annual net (After-Tax) profit (ANP), $MM/year 180
Return on investment (%/year) 28.6

Table 2 – Uncertain inputs into the availability estimation of a proposed propane dehydrogenation (PDH) design.
Processing step Major equipment Mean time between failure (Hours) Mean time to restoration (Hours)

Lower Mean/Most likely Upper Lower Mean/Most likely Upper

Separation of C4+ Depropanizer column 8174 10,146 12,695 10.0 44.5 359
materials in feed and
recycle
Catalytic Fired heater (charge 266 909 41,239 2.0 66.7 706
dehydrogenation heater)
reaction Dehydrogenation 5255 10,632 50,383 1.0 47.0 474
reactor
Reactor feed-effluent 13,325 15,838 20,017 1.0 55.4 419
heat exchanger
Continuous catalyst Continuous catalyst 5255 10,632 50,383 39.0 47.0 474
regeneration (CCR) regenerator
Turbocompressor feed 13,325 15,838 20,017 1.0 55.4 419
Compression and
cooler
cooling of reactor
Steam turbine 6160 10,317 21,529 9.3 16.0 31
effluent
Multi-stage centrifugal 2700 3145 3815 16.9 28.9 1481
compressor
Turbocompressor 13,325 15,838 20,017 1.0 55.4 419
discharge cooler
Heat exchanger 13,325 15,838 20,017 1.0 55.4 419
Cryogenic refrigeration Flash drum 6849 13,522 37,823 5.9 16.9 409
Isentropic expander 1207 3348 181,048 10.0 44.5 728
Selective hydrogenation 5255 10,632 50,383 39.0 47.0 474
Hydrogen Separation
process (SHP) reactor
Pressure swing 8174 10,146 12,695 10.0 44.5 359
adsorption (PSA) unit
De-ethanizer column 8174 10,146 12,695 10.0 44.5 359
Product Recovery
Propylene-propane 8174 10,146 12,695 10.0 44.5 359
splitter
(superfractionator)

ratio of the standard deviation to the mean, can be utilized to in the process performance assessment framework. Identi-
examine and compare the variability from the mean value for fication and characterization of the uncertainty associated
both availability and ROI. These values were determined to be with performance assessments improves the understanding
0.109 for system availability, and 0.312 for ROI. This indicates of risks and design sensitivity, leading to more informed deci-
that there was less variability from the mean value (0.745) for sions regarding proposed designs.
system availability (Fig. 4) than from the mean value (23.7%)
for ROI (Fig. 5).
A table in the supplementary materials of this work shows 4.3.3. Section analysis of process RAM characteristics and
the range of results for the main economic parameters used profitability
to evaluate the proposed PDH process design. The 5% and 95% In addition to the results presented, the availability of individ-
tails represent the lower and upper limits of the 90% confi- ual process blocks can be estimated to identify critical units
dence interval for all the parameters presented. The range of and determine the impact of their downtime in realtion to
possible system availability is reflected in the annual revenue revenue losses. Fig. 6 shows the mean section availability for
losses, annual profit, and return on investment. All of the cat- each of the process blocks, along with the 90% confidence
egories displayed a moderate level of skewness (between −1 bounds. This captures the range of possible availabilities for
and 1), indicating the resulting distributions are moderately each section, depending on the equipment failure and restora-
asymmetric. tion rates. Also, the figure illustrates that the compression and
The results of this work emphasize the importance of cooling of reactor effluent section has a significantly lower
explicitly including uncertainty in all input variables used availability than the other section, even when considering the
predicted upper limit of its availability. The compression and
288 Chemical Engineering Research and Design 1 6 7 ( 2 0 2 1 ) 281–291

The first three modifications can be realized by either

obtaining equipment that are: (1) known to have enhanced
RAM characteristics due to previous operating experience, or
(2) larger vessels that provide a surge capacity in the case of
process upsets. The latter case can be achieved by selecting
a centrifugal compressor with a greater design pressure ratio
and discharge temperature than needed for the process. The
configuration for the base case and modifications 1–3 are illus-
trated in top part of Fig. 9.
For the compression and cooling section of a PDH process,
the discharge temperatures for the two compressor stages are
138 ◦ C and 355 ◦ C, respectively. Accordingly, the compressor
Fig. 6 – Section availability of individual process blocks for
selected should have a higher specified discharge temperature
a proposed PDH process.
than those values. This results in equipment that have higher
pressure ratios and power rating, which typically incur higher
capital costs. The desired margin between the discharge tem-
cooling of reactor effluent section includes a steam turbine-
perature and compressor design temperature limit depends
driven multi-stage centrifugal compressor, which typically
on many factors including: (1) likely operating conditions
experience higher failure rates than non-rotating machinery.
(clean or fouling service), (2) operating environment due to
Using the section availability data, the contribution of each
location, and (3) an organization’s financial resources. Higher
section to the overall process revenue losses (presented in
design margins provide greater protection against failures,
the supplementary table) can be determined. From the results
thus increasing overall equipment availability. The relation-
shown in Fig. 7, it is clear that the compression and cooling of
ships between equipment capacity, availability, and cost is
reactor effluent section is the largest contributor to overall rev-
beyond the scope of this work and have not been estab-
enue losses due its significantly lower availability. In this case
lished for the chemical process industries. Several works have
study, this section accounts for approximately 60% of annual
examined those relationships for electronic equipment using
revenue losses for a PDH process.
correlations and discrete data (Fratta and Montanari, 1976;
Majumder et al., 1976; Govil, 1985).
4.4. Design modifications The fourth modification, a parallel flow configuration,
employs two equal-size centrifugal compressors for the first
Four design modifications to the compression and cooling of and second stage of the compression system with the same
reactor effluent section are evaluated to determine their effect design specifications and RAM characteristics as the base case.
on ROI. The proposed modifications for the multi-stage cen- Each compressor in the parallel arrangement is designed to
trifugal compressor are as follows: take 50% of the flowrate in the base case design. This configu-
ration, shown in Fig. 8, incurs much higher capital costs than
the other modifications due to the loss of economy-of-scale.
(1) Compressor with higher reliability (increased MTBF)
The availability, cost, and revenue losses associated with
(2) Compressor with shorter restoration time (decreased
each modification are shown in Table 3. The parallel flow con-
MTTRes)
figuration has the highest availability and thus achieves the
(3) Compressor with both a lower failure rate and shorter
greatest reduction in revenue losses from equipment unavail-
restoration time
ability. However, it does present a significantly greater increase
(4) Addition of a second, parallel flow compressor train, with
in installed costs over compared to the other modifications.
each train handling 50% of the flowrate.

Fig. 7 – Contribution of processing sections to revenue losses.

 Chemical Engineering Research and Design 1 6 7 ( 2 0 2 1 ) 281–291 289

Fig. 8 – Compression and cooling section configurations for (top) base case design and modifications 1-3, and (bottom)
modification 4 (parallel flow arrangement).

Table 3 – Availability and economic parameters for proposed design modifications.

Alternative MTBF (Hours) MTTRes (Hours) Availabi-lity Installed cost Revenue losses Reduction in
($MM) due to revenue losses
unavailabili-ty (%)
($MM)

Base case 3226 509 0.845 34.1 94.7 –

Lower failure rate 3449 509 0.871 36.7 77.6 18.1
Shorter MTTRes 3226 458 0.876 37.1 75.3 20.5
Lower failure rate and shorter MTTRes 3449 458 0.883 37.7 71.6 24.4
Parallel flow with base case RAM levels 2419 208 0.921 51.8 51.9 45.2

be recommended as the optimal design configuration for the

compression and cooling section of a PDH plant.
The improvement of system availability has a positive
impact on both profitability and sustainability because it
reduces flaring that occurs due to equipment failures. In the
case of a PDH process, valuable propane feedstock, propylene
product, and other chemicals can be flared in a failure event.
This incurs additional costs to replace lost feedstock, reduces
plant revenue, and results in fines due to the greenhouse gas
emissions released during flaring. These fines can have an
adverse effect on a company’s reputation and future growth.
Fig. 9 – Comparison of ROI and IROI for alternative design This approach can be further augmented by considering
scenarios of a PDH process compression system. sustainability aspects along with RAM characteristics. For
example, CO2 emissions reductions can be selected as a sus-
tainability indicator. Two scenarios that can be considered
are implementing: (1) a waste heat recovery (WHR) system to
In this case study, a minimum ROI value of 15% was
recover medium temperature heat from the fired boilers, and
selected as an appropriate threshold for acceptability of a
(2) an offgas recovery system to recover the deethanizer off-
design. Fig. 9 presents the results for both the ROI and IROI
gas stream and use it as fuel, thereby reducing natural gas
metrics for the proposed designs for the cooling and com-
consumption.
pression section. Note that the base case, as well as all
alternatives, exceed the ROI threshold. However, only the par-
allel flow configuration exceeds the 15% threshold for the 5. Conclusions
IROI metric, with a value of approximately 26%/year. This
configuration provides the most significant improvement in This paper presents a framework to integrate uncertainty
equipment availability and increase in annual profit over the quantification of equipment RAM characteristics into eco-
base case. Despite the higher capital cost, the additional rev- nomic analyses of chemical process designs during the
enue generated by greater equipment availability results in conceptual development stage. Instead of the standard prac-
the configuration having the highest IROI. Hence, it would tice of using average or fixed values, this framework employs
290 Chemical Engineering Research and Design 1 6 7 ( 2 0 2 1 ) 281–291

distributions for failure and restoration data of process equip- De Neufville, R., Scholtes, S., 2011. Flexibility in Engineering
ment to estimate an availability range for a base case design Design. MIT Press.
using Monte Carlo simulation. Using restoration instead of El-Halwagi, M.M., 2017a. A return on investment metric for
incorporating sustainability in process integration and
repair times accounts for delays in the maintenance process
improvement projects. Clean Technol. Environ. Policy 19 (2),
which increase downtime and negatively impact availabil- 611–617.
ity and economic performance. By doing so in a probabilistic El-Halwagi, M.M., 2017b. Sustainable Design through Process
manner, full distribution profiles of the key performance indi- Integration: Fundamentals and Applications to Industrial
cators, instead of single-point estimates which is the case Pollution Prevention, Resource Conservation, and Profitability
in a deterministic approach, are generated and associated Enhancement. Butterworth-Heinemann.
to their statistical characterization. These process economic Fratta, L., Montanari, U.G., 1976. Synthesis of available networks.
IEEE Trans. Reliab. 25 (2), 81–87.
performance indicators include: annual net profit, annual rev-
Goel, H., Grievink, J., Herder, P., Weijnen, M., 2003. Optimal
enue losses, and return on investment. This approach also reliability design of process systems at the conceptual stage
allows for identifying which specific section(s) in a process of design. In: Annual Reliability and Maintainability
contribute significantly to revenue losses due to equipment Symposium, 2003, IEEE, pp. 40–45.
unavailability. Proposed design modifications can then be Govil, K.K., 1985. Fitting cost-reliability data in a standard curve.
examined to determine how best to improve availability for Microelectron. Reliab. 25 (5), 913–915.
Gregor, J., Wei, D., 2004. UOP Oleflex Process for light olefin
the section(s) in question in an economically-optimal way
production. In: Handbook of Petrochemicals Production
using an incremental return on investment (IROI) metric.
Processes.
A case study involving a propane dehydrogenation (PDH) Guillen-Cuevas, K., Ortiz-Espinoza, A.P., Ozinan, E.,
plant was used to demonstrate the merits of integrating Jiménez-Gutiérrez, A., Kazantzis, N.K., El-Halwagi, M.M., 2018.
uncertainty quantification in RAM estimates used in system Incorporation of safety and sustainability in conceptual
performance assessments. This approach differs significantly design via a return on investment metric. ACS Sustain. Chem.
from practices which consider system design modifications Eng. 6 (1), 1411–1416.
Hashemi, S.J., Ahmed, S., Khan, F.I., 2014. Risk-based operational
to improve availability based solely on budgetary constraints.
performance analysis using loss functions. Chem. Eng. Sci.
In summary, the proposed framework can be implemented by 116, 99–108.
decision-makers, design engineers, and operations teams to Henley, E.J., Gandhi, S.L., 1975. Process reliability analysis. AICHE
make better-informed decisions on the economic potential of J. 21 (4), 677–686.
a project or design modification. This approach can also be Ishii, N., Fuchino, T., Muraki, M., 1997. Life cycle oriented process
extended to characterize the sustainability and process safety synthesis at conceptual planning phase. Comput. Chem. Eng.
impacts of project or system modifications. 21, S953–S958.
Julián-Durán, L.M., Ortiz-Espinoza, A.P., El-Halwagi, M.M.,
Jiménez-Gutiérrez, A., 2014. Techno-economic assessment
Declaration of Competing Interest and environmental impact of shale gas alternatives to
methanol. ACS Sustain. Chem. Eng. 2 (10), 2338–2344.
Kazantzi, V., El-Halwagi, A.M., Kazantzis, N., El-Halwagi, M.M.,
The authors report no declarations of interest. 2013. Managing uncertainties in a safety-constrained process
system for solvent selection and usage: an optimization
approach with technical, economic, and risk factors. Clean
References Technol. Environ. Policy 15 (2), 213–224.
Kazi, M.K., Eljack, F., Amanullah, M., AlNouss, A., Kazantzi, V.,
Abubakar, U., Sriramula, S., Renton, N.C., 2015a. Stochastic 2018. A process design approach to manage the uncertainty of
techno-economic considerations in biodiesel production. industrial flaring during abnormal operations. Comput.
Sustain. Energy Technol. Assess. 9, 1–11. Chem. Eng. 117, 191–208.
Abubakar, U., Sriramula, S., Renton, N.C., 2015b. A hybrid method Khan, F., Wang, H., Yang, M., 2016. Application of loss functions
for stochastic performance modeling and optimization of in process economic risk assessment. Chem. Eng. Res. Des.
chemical engineering processes. Chem. Eng. Commun. 202 111, 371–386.
(2), 217–231. Kidam, K., Sahak, H.A., Hassim, M.H., Shahlan, S.S., Hurme, M.,
Abubakar, U., Sriramula, S., Renton, N.C., 2015c. Reliability of 2016. Inherently safer design review and their timing during
complex chemical engineering processes. Comput. Chem. chemical process development and design. J. Loss Prev.
Eng. 74, 1–14. Process Ind. 42, 47–58.
Agarwal, A., Sengupta, D., El-Halwagi, M., 2018. Sustainable Kuppens, T., Rafiaani, P., Vanreppelen, K., Yperman, J., Carleer, R.,
process design approach for on-purpose propylene production Schreurs, S., Van Passel, S., 2018. Combining Monte Carlo
and intensification. ACS Sustain. Chem. Eng. 6 (2), 2407–2421. simulations and experimental design for incorporating risk
Aguilar, O., Kim, J.K., Perry, S., Smith, R., 2008. Availability and and uncertainty in investment decisions for cleantech: a fast
reliability considerations in the design and optimisation of pyrolysis case study. Clean Technol. Environ. Policy 20 (6),
flexible utility systems. Chem. Eng. Sci. 63 (14), 3569–3584. 1195–1206.
Al-Douri, A., Kazantzi, V., Eljack, F.T., Mannan, M.S., El-Halwagi, Law, Averill, Kelton, W. David, et al., 2007. Simulation Modeling
M.M., 2020. Mitigation of operational failures via an economic and Analysis. McGraw-Hill, New York.
framework of reliability, availability, and maintainability Li, X., Kraslawski, A., 2004. Conceptual process synthesis: past
(RAM) during conceptual design. J. Loss Prev. Process Ind., and current trends. Chem. Eng. Process. Process. Intensif. 43
104261. (5), 583–594.
Alexander, D.C., 2003. Application of Monte Carlo simulations to Lin, Yi-Kuei, Chang, Ping-Chen, 2012. Reliability evaluation for a
system reliability analysis. In: Proceedings of the 20th manufacturing network with multiple production lines.
International Pump Users Symposium, Texas A&M University. Comput. Ind. Eng. 63 (4), 1209–1219.
Turbomachinery Laboratories. Majumder, D.D., Pal, S.K., Chaudhuri, B.B., 1976. Fast algorithm for
Andiappan, V., Benjamin, M.F.D., Tan, R.R., Ng, D.K., 2019. An reliability and cost of a complex network. IEEE Trans. Reliab.
integrated framework to address criticality in biomass 25 (4), 258-258.
tri-generation systems via redundancy allocation. Process. Martinez-Gomez, Juan, Sánchez-Ramírez, Eduardo,
Integr. Optim. Sustain. 3 (1), 65–73. Quiroz-Ramírez, Juan José, Segovia-Hernandez, Juan Gabriel,
 Chemical Engineering Research and Design 1 6 7 ( 2 0 2 1 ) 281–291 291

Ponce-Ortega, José María, El-Halwagi, Mahmoud, et al., 2016. Vatn, J., 1993. OREDA Data Analysis Guidelines. Report STF75
Involving economic, environmental and safety issues in the F93024. SINTEF Sikkerhet og pålitelighet, Trondheim.
optimal purification of biobutanol. Process Safety and Vora, B.V., 2012. Development of dehydrogenation catalysts and
Environmental Protection 103, 365–376. processes. Top. Catal. 55 (19–20), 1297–1308.
OREDA, 2002. Offshore and Onshore Reliability Data, 4th ed. Wood Mackenzie, 2014. Global Propylene Long-Term Outlook 2H
OREDA, 2015. Offshore and Onshore Reliability Data, 6th ed. 2014.
Park, S., Xu, S., Rogers, W., Pasman, H., El-Halwagi, M.M., 2019. Yang, M., You, F., 2017. Comparative techno-economic and
Incorporating inherent safety during the conceptual process environmental analysis of ethylene and propylene
design stage: a literature review. J. Loss Prev. Process Ind., manufacturing from wet shale gas and naphtha. Ind. Eng.
104040. Chem. Res. 56 (14), 4038–4051.
Rahman, M., Heikkilä, A.M., Hurme, M., 2005. Comparison of Ye, Y., Grossmann, I.E., Pinto, J.M., 2018. Mixed-integer nonlinear
inherent safety indices in process concept evaluation. J. Loss programming models for optimal design of reliable chemical
Prev. Process Ind. 18 (4–6), 327–334. plants. Comput. Chem. Eng. 116, 3–16.
Roy, N., Eljack, F., Jiménez-Gutiérrez, A., Zhang, B., Zio, Enrico, Baraldi, Piero, Patelli, Edoardo, et al., 2006.
Thiruvenkataswamy, P., El-Halwagi, M., Mannan, M.S., 2016. A Assessment of the availability of an offshore installation by
review of safety indices for process design. Curr. Opin. Chem. Monte Carlo simulation. Int. J. Pressure Vessels Piping 83 (4),
Eng. 14, 42–48. 312–320.
Rudd, D.F., 1962. Reliability theory in chemical system design. Galindo, Enrique (2019). Enterprise targets a second PDH unit in
Ind. Eng. Chem. Fundam. 1 (2), 138–143. the US.
Savage, S.L., 2003. Decision Making with Insight (with Insight. xla <https://www.woodmac.com/news/opinion/enterprise-
2.0 and CD-ROM). Duxbury Press. targets-a-second-pdh-unit-in-the-us/>. Accessed: August, 10,
Seila, A.F., Ceric, V., Tadikamalla, P.R., 2003. Applied Simulation 2020.
Modeling. Brooks/Cole. Eramo, Mark, 2017. Global Petrochemical Market Outlook: Will
Sharda, B., Bury, S.J., 2008. A discrete event simulation model for Expansion Activity in North America Continue? Presented to:
reliability modeling of a chemical plant. In: 2008 Winter VMA Market Outlook Workshop, Boston, MA
Simulation Conference, IEEE, pp. 1736–1740. https://cdn.ymaws.com/www.vma.org/resource/resmgr/2017
Terrazas-Moreno, S., Grossmann, I.E., Wassick, J.M., Bury, S.J., MOW Presentations/Eramo Presentation.pdf.
2010. Optimal design of reliable integrated chemical
production sites. Comput. Chem. Eng. 34 (12), 1919–1936.