Zdravko Kravanja, Miloš Bogataj (Editors), Proceedings of the 26th European Symposium on

Computer Aided Process Engineering – ESCAPE 26

June 12th -15th, 2016, Portorož, Slovenia © 2016 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/B978-0-444-63428-3.50147-8

Hybrid plant-wide Simulation and Modelling using

Reference nets and Java
Stefan Huberta,* , Darius Rückerta and Antonio Delgadoa
a Friedrich-Alexander-Universität
Erlangen-Nürnberg (FAU), Lehrstuhl für Strömungsmechanik,
Cauerstraße 4, 91058 Erlangen, Germany
*[email protected]

Abstract
Holistic plant-wide modelling and simulation can be used as an advanced tool for decision mak-
ing. Control and operational strategies and even different sets of equipment might be evaluated
virtually. But plant-wide modelling is very demanding, as many models are based on a variety of
state variables and different model types need to be included, which requires hybrid simulation
modelling techniques. This work shows a new approach for hybrid plant-wide modelling, which
is based on Reference nets (a special type of Petri nets), Java programming language, embedded
SQL databases, object-oriented modelling and a model coupling concept called Balance flows. It
was developed and tested within a research project evaluating water and energy management in
breweries.

Keywords: plant-wide, simulation and modelling, hybrid, reference nets, petri nets, java

1. Introduction
Sustainability has evolved as a major goal over the last decades within process systems engineer-
ing (El-Halwagi, 2012). This is also a consequence of environmental treaties and regulations,
such as the Kyoto Protocol, Eco-Management and Audit Scheme and ISO 14001 (United Na-
tions, 1997; European Union, 2009; ISO, 2004; Simate et al., 2011). One key activity in order
to increase sustainability might be a sophisticated energy and water management, which needs a
strong foundation of process knowledge in terms of knowing how much and where energy and
water is demanded. One tool helping to build such a deep understanding is holistic plant-wide
modelling and simulation.
Plant-wide modelling and simulation also offers possibilities like testing plant-wide control strate-
gies and evaluation of different process equipment and operational setups. According to the IEC
62264 standard, process manufacturing systems can be divided into level 1 to 4 (IEC, 2013). Level
1, 2 and 3 are part of the manufacturing control and operation domain, whereas level 4 is the en-
terprise domain (IEC, 2013). A promising approach is to include those levels, or at least a set con-
taining more than one level, into one integrated approach (Chu and You, 2015; Grossmann, 2005).
Some papers can be found about plant-wide modelling and control within the field of chemical
engineering, like the Tennessee Eastman process (TEP) (Lyman and Georgakis, 1995) and within
the field of wastewater treatment, like the benchmark simulation model (BSM2) (Jeppsson et al.,
2006). Literature discussing plant-wide concerns in food and beverage processes, like the produc-
tion of beer, have not been found by the authors. Many papers have been published concerning
single process stages, such as the mashing process in breweries (Koljonen et al., 1995; Kettunen
et al., 1996; Li et al., 2004) and fermentation of wort (Gee and Ramirez, 1994; de Andres-Toro
854 S. Hubert et al.

et al., 1998). Some are dealing with the complete process of brewing but are discrete event based,
missing physical and biological first principle models of the complex processes being involved
(Gläßer, 2012; Scharbrodt, 2000). Tools like gPROMS (Process Systems Enterprise Limited),
Aspen HYSYS (Aspen Technology, Inc) or Plant Simulation (Siemens AG) can be used for plant-
wide modelling. Unfortunately more than 95 % (Söllner, 2011) of the companies in the brewery
sector in Germany are small to medium-sized enterprises (SMEs), which do not have any budget
for such sophisticated software solutions. Hence there is a huge cap between SMEs and bigger
companies with respect to applying simulation based process improvements.
Therefore the presented paper deals with a novel approach for hybrid plant-wide modelling based
on a non-commercial software solution, which is a combination of Reference nets (Kummer, 2002)
and Java programming language (Nagel et al., 2007; Hubert et al., 2016; Pettigrew et al., 2015).
Due to the unlimited opportunities of Java programming the shown approach is capable to deal
with many different model types, like discrete/continuous (hybrid) and deterministic/stochastic
(Hubert et al., 2015). A concept called ’Balance flows’ for coupling different process model out-
puts is being presented. Processes and equipments of breweries are used as an example application
for this new approach.

2. Problem description
2.1. Processes within a brewery
In a brewery four different types of processes can be separated (see Figure 1). The production
of wort aims to dissolve the ingredients of malts, like polysaccharides and proteins, into water.
Different unit operations are utilised to do so. Size reduction (malt milling) is done by different
types of mills. This increases the surface area and releases the grain kernels. Biochemical reactions
(mashing), like the enzymatic hydrolysis of starch molecules, are necessary to prepare the wort
for the metabolisation by yeast in the fermentation step. Solid-liquid separation (lautering and
casting the wort) are required to extract further ingredients and to separate brewers’ spent grain.
Finally, there are thermal operations like evaporation (wort boiling) and cooling. Up to 8 % of the
wort is being evaporated (Kunze, 2010). Second major process within a brewery is the production
of beer. The wort is utilised by yeasts, which are metabolizing the dissolved sugar into alcohol
and carbon dioxide. This process stage is called fermentation and usually takes between 4 to 14
days, depending on the type of beer and yeast organism. Once the sugar molecules are almost
completely digested the beer is cooled down to around freezing temperatures and remains at this
temperature, usually until it is bottled. This stage is called maturation. (Kunze, 2010).
Additional processes involved in a brewery are the bottling and cleaning & disinfection of the
equipment. The processes wort production, beer production and cleaning and disinfection (at least
clean-in-place systems) can be considered as process manufacturing (batch processes), whereas
the process of bottling can be thought of as a type of discrete manufacturing.

2.2. Real life brewery example

The presented work is based on a study about (waste)water and energy management of two SME
breweries, ranging in production volumes of 50,000 to 200,000 hectolitre beer per year. A simpli-
fied P&ID of one of the brewhouses can be seen in Figure 2. Each process equipment receives one
unique identification number which is then used within the simulation environment.

3. A novel modelling approach using Reference nets and Java

The modelling approach is based on Reference nets (RN) (Kummer, 2002) and Java programming
language. RN are an extended version of the well studied Petri nets (Petri, 1962). They feature a
Hybrid plant-wide Simulation and Modelling using Reference nets and Java 855

Malt milling Mashing Lautering Wort boiling

Casting the
Cooling
wort

Fermentation Maturation
Cleaning and
disinfection
Bottling

Figure 1: Processes involved in a typical brewery (Kunze, 2010).

14 15 16
Water Water Hops 17
7 cold warm Vap.

1 12
Malt 10
9 11

2 3 4 5 6 13
18 19
BSG Trub Beer
2 Production

Figure 2: Simplified P&ID of the brewhouse. Devices involved in the production of wort are
shown with their identification number, which is used within the simulation environment.

timing mechanism, which is used to model the discrete dynamic events within a brewery. Those
discrete events are for example the initial start of a new wort production, activation of a stirrer or
start of a bottling process. RN further support complex token types, like tuples and lists or even
arbitrary Java objects. A nets-within-nets formalism, in which tokens can represent other nets, is
used for nested net models. This feature is used for object-oriented modelling of the processes
and equipment. The RN are able to be extended by using additionally Java classes. Within the
shown approach those classes contain models, physical properties of substances (like water, mash
and wort) and also the simulation environment, which is based on a SQL database system. For
example stochastic models, used for the process of bottling, can be integrated into a RN as seen in
(Hubert et al., 2015, 2016). Models based on ordinary differential equations, used for example to
describe the dynamic process of mashing (Koljonen et al., 1995), are also implemented and solved
numerically within the model classes. RN itself do not contain any representation of physical
properties. Therefore a concept called Balance flows (BF) has been developed. Basically BF store
information about mass, mass component and energy balances to an embedded SQL database.
The simulation is initialised in the RN environment where it follows a discrete path, similar to
a process flow chart like the production of wort shown in Figure 1. Depending on the process
stage, classes are loaded and models are being simulated. The results of the continuous models
are then stored as a BF in the embedded database and can be used for further simulation or result
analysis (see Figure 4). A second database is used to store inputs and parameters of recipes,
processes and equipment in a object-oriented way. This data is then to be used within the model
classes for object initialisation. Figure 3 shows for example how a centrifugal pump is stored in
the database. The parameters para cp hq x are results of a polynomial regression of the specific
pump performance curve (power consumption as a function of flow rate). This allows to easily
implement real devices of different manufactures, which then can be tested virtually for decision
making (e.g. for investment decisions). Similar to this other devices and also processes are being
856 S. Hubert et al.

implemented.

Figure 3: Screenshot showing the inputs and parameters of a centrifugal pump. On the upper left
the main type and subtypes are shown.

The simulation outputs BFs, which can be seen as 7-tuple, defined as BFi = (SE, T D, SD, Mt , Mc ,
P, E). SE is a set of information about the simulation environment SE := {id, simid , batchid },
where id is a unique identification number of the BF, simid is a unique number of the simulation run
and batchid is unique number of the batch being simulated (e.g. a special type of wort production).
T D includes information about time, like the start and the end time of BFi . SD includes spatial
data like the equipment numbers (see Figure 2). Mt := {mt , φmt } stores the total mass balance of
BFi , with mt representing the total mass change and φmt a function of time, which describes the
mass change. φ itself is described by a set of parameters B and a function number κ ∈ K. K
is a set of implemented functions, which are either math or Java based. Mc represents the set of
component balances. All of them are defined as (for component j) Mc, j := {mc, j , φc, j }. P is a set
of physical properties, involving temperature, density and specific heat capacity. And E stores the
energy balances defined as E := {Qt , φQ ,Wt , φW , Eel,t , φel,t }, with Qt and φQ being the conductive
heat transport, Wt and φW describing the added work and Eel,t in combination with φel,t is the
electrical energy demand.

4. Simulation examples
The novel modelling approach was used to model processes involved in typical breweries. As
an example the results of a simulation of a wort production, especially at the stage of mashing
is shown. First principle models found in literature (Koljonen et al., 1995; Kettunen et al., 1996;
Durand et al., 2009; Li et al., 2004) have been implemented as Java classes and coupled using the
BF concept. Analysis of the results is done by choosing a single device, such as the mash tun
(ID = 9, see Figure 1) from a list of all devices being implemented. Results can then either be
stored into a comma separated file or analysed using a graphical user interface (see Figure 5).
The state variables for each device are mass, volume, energy content, temperature, total theoretical
chemical oxygen demand (COD), total COD ratio, dissolved COD and ratio, electrical energy
consumption and extract, which is a common measure within breweries for the concentration of
fermentable sugars.
Hybrid plant-wide Simulation and Modelling using Reference nets and Java 857

Start End

Mashing Lautering Boiling Casting

Milling Cooling

BF2

BF4

BF5

BF6
BF3
BF1
Analysis

Density
Mass

Mass

Density
…
Time Time Time Time

Volume
Volume

COD
COD

Time Time Time Time

Device 1 Device 2 Device N

Figure 4: Concept of the simulation environment using Reference nets, Java, embedded Databases
and model coupling via Balance flows.

Simulation runs can also include schedules of arbitrary length, like a complete month of produc-
tion. Multiple types of processes, for example wort production and beer bottling, are simulated
together. The result, in form of a databases filled with Balance Flows, is a huge set of trajectories
of the mentioned state variables with respect to the set of process devices. This database can be
used for additional optimization frameworks. E.g. to design an optimised waste water treatment
strategy based on the simulation result of a complete year of production.

Figure 5: Screenshots of the graphical analysis tool. The total mass as a function of time (left) and
the temperature of a specific device (mash tun) is shown.

5. Conclusion
A novel hybrid plant-wide modelling and simulation approach, based on a non-commercial soft-
ware solution using Reference nets, Java programming language, embedded databases and a model
coupling concept called Balance flows, has been shown. It is capable to deal with various types of
models, like discrete and continuous as well as deterministic and stochastic ones. The vast amount
of process equipment can be included by object-oriented modelling concepts. Future publications
will show how this is done in detail. The novel approach was successfully used within a research
858 S. Hubert et al.

project, dealing with breweries and a focus on wastewater levels. State variables, like the chemical
oxygen demand for wastewater or extract as a specific measure for fermentable sugar, have been
implemented specifically for a brewery environment. Other state variables can be chosen freely by
using the concept of Balance flows. Compared to available software solutions, the non-commercial
nature and the adaptability of the shown approach allows for greater access to simulation-based
process improvements for SMEs.
References
Y. Chu, F. You, 2015. Model-based integration of control and operations: Overview, challenges, advances, and opportu-
nities. Computers & Chemical Engineering.
B. de Andres-Toro, J. M. Giron-Sierra, J. A. Lopez-Orozco, C. Fernandez-Conde, J. M. Peinado, F. Garcıá-Ochoa,
1998. A kinetic model for beer production under industrial operational conditions. Mathematics and Computers in
Simulation 48 (1), 65–74.
G. A. Durand, M. L. Corazza, A. M. Blanco, F. C. Corazza, Dec. 2009. Dynamic optimization of the mashing process.
Food Control 20 (12), 1127–1140.
M. M. El-Halwagi (Ed.), 2012. Sustainable Design Through Process Integration. Butterworth-Heinemann, Oxford.
European Union, 2009. Eco- Management and Audit Scheme. REGULATIONS 1221/2009.
D. A. Gee, W. F. Ramirez, Sep. 1994. A Flavour Model for Beer Fermentation. Journal of the Institute of Brewing
100 (5), 321–329.
J. F. Gläßer, 2012. Ereignisorientierte Simulation der Bierherstellung. Doktor-ingenieurs, Technische Universität München,
München.
I. Grossmann, Jul. 2005. Enterprise-wide optimization: A new frontier in process systems engineering. AIChE Journal
51 (7), 1846–1857.
S. Hubert, T. Helmers, F. Baur, A. Delgado, N. Räbiger, Dec. 2015. Simulation Modeling of Bottling Line Water
Demand Levels using Reference Nets and Stochastic Models. In: Simulation Conference (WSC), 2015 Winter.
pp. 2272–2281.
S. Hubert, T. Helmers, F. Groß, A. Delgado, May 2016. Data driven stochastic modelling and simulation of cooling
demand within breweries. Journal of Food Engineering 176, 97–109.
IEC, 2013. Enterprise-control system integration – Part 1: Models and terminology. Tech. Rep. IEC 62264-1:2013,
Geneva.
ISO, 2004. Environmental management systems – Requirements with guidance for use. Standard ISO 14001:2004,
International Organization for Standardization, Geneva, CH.
U. Jeppsson, C. Rosen, J. Alex, J. Copp, K. V. Gernaey, M. N. Pons, P. A. Vanrolleghem, 2006. Towards a benchmark
simulation model for plant-wide control strategy performance evaluation of wwtps. Water Science and Technology:
A Journal of the International Association on Water Pollution Research 53 (1), 287–295.
A. Kettunen, J. J. Hämäläinen, K. Stenholm, K. Pietilä, Aug. 1996. A model for the prediction of β -glucanase activity
and β -glucan concentration during mashing. Journal of Food Engineering 29 (2), 185–200.
T. Koljonen, J. J. Hämäläinen, K. Sjöholm, K. Pietilä, 1995. A model for the prediction of fermentable sugar concentra-
tions during mashing. Journal of Food Engineering 26 (3), 329–350.
O. Kummer, 2002. Referenznetze. Logos Verlag, Berlin.
W. Kunze, 2010. Technology Brewing and Malting. VLB Berlin, Berlin.
Y. Li, J. Lu, G. Gu, Z. Shi, Z. Mao, May 2004. Mathematical modeling for prediction of endo-xylanase activity and
arabinoxylans concentration during mashing of barley malts for brewing. Biotechnology Letters 26 (10), 779–785.
P. R. Lyman, C. Georgakis, Mar. 1995. Plant-wide control of the Tennessee Eastman problem. Computers & Chemical
Engineering 19 (3), 321–331.
M. Nagel, H. Petermeier, A. Delgado, 2007. Computer aided production planning and scheduling. In: Proceedings of
the European Brewery Convention Congress 2007. Venice, pp. 1427–1434.
C. A. Petri, 1962. Kommunikation mit Automaten. Rhein.-Westfäl. Inst. f. Instrumentelle Mathematik an der Univ.
Bonn.
L. Pettigrew, V. Blomenhofer, S. Hubert, F. Groß, A. Delgado, Jan. 2015. Optimisation of water usage in a brewery
clean-in-place system using reference nets. Journal of Cleaner Production 87, 583–593.
M. Scharbrodt, 2000. Produktionsplanung in der Prozessindustrie: Modelle, effiziente Algorithmen und Umsetzung.
Doktors der naturwissenschaften, Technische Universität München, München.
G. S. Simate, J. Cluett, S. E. Iyuke, E. T. Musapatika, S. Ndlovu, L. F. Walubita, A. E. Alvarez, Jun. 2011. The treatment
of brewery wastewater for reuse: State of the art. Desalination 273 (2–3), 235–247.
R. Söllner, 2011. Ausgewählte Ergebnisse für kleine und mittlere Unternehmen in Deutschland 2009. Wirtschaft und
Statistik, 1086–1097.
United Nations, 1997. Kyoto Protocol. Tech. rep., United Nations Framework Convention on Climate Change.