Content uploaded by Thorsten Pawletta
Author content
All content in this area was uploaded by Thorsten Pawletta on Sep 13, 2017
Content may be subject to copyright.
S N E TE C H N I C A L NO T E
SNE 27(2) 6/2017 115
A Multimodeling Approach for the Simulation of
Energy Consumption in Manufacturing
Thorsten Pawletta1*, Artur Schmidt1, Peter Junglas2
1Research Group Computational Engineering & Automation, Wismar University of Applied Sciences,
Philipp-Müller-Straße 14, 23966 Wismar, Germany; *
thorsten.pawletta@hs-wismar.de
2PHWT Vechta/Diepholz/Oldenburg, Rombergstraße 40, 49377 Vechta, Germany
Abstract. In the design of manufacturing systems the
consideration of resource usage, especially energy con-
sumption, is getting more attention. However, the inclusion
of all relevant physical processes in a unified modeling
approach is a non-trivial task, if detailed analyses are re-
quired. The commonly used modeling approach for manu-
facturing systems is the discrete event modeling technique.
However, models of physical processes are often continuous
in nature and are modeled using ordinary differential equa-
tions or differential algebraic equations. Indeed, the investi-
gation of such physical processes in manufacturing systems
often demands a more specific consideration of process
control operations, which are favorably modeled using state
machines. To combine those different paradigms a multi-
modeling approach for manufacturing systems is proposed.
The approach is illustrated by the example of a production
line with an industrial furnace facility.
Introduction
The modeling and simulation of manufacturing systems
has been a subject of study for several decades. Accord-
ing to [1], the typical modeling approach is discrete
event modeling in this domain. This fact is reflected in
the popular simulation tools applied in this field today.
However, the situation has recently been changing,
because of new aspects that have been taken into ac-
count and increasing requirements for accuracy.
One of these aspects is the time-dependent energy
consumption of single-process operations, process
chains or an overall production system, which becomes
important in context with the increasing influence by
renewable energy sources and the associated volatile
energy availability and energy prices. Approaches for
single-process operations, such as in [2], are focused on
the energy consumption of single machine operations.
They are often based on differential algebraic equations.
However, approaches for investigating several machines
coupled to a process chain use more abstract models
mostly based on discrete event methods, such as in [3, 4].
Today most approaches related to energy processes in
manufacturing are only focused on the simulation of
energy consumption. In [5] it is emphasized that the
energy consumption of a production line (PL) has to be
considered in context with production planning and
scheduling operations. In fact, the energy consumption
then has to be examined together with all the other pro-
duction performance indicators, such as through-put
time, load factors, utilization etc. Hence, considerations
regarding the model design and permissible model sim-
plifications are important to master model complexity.
For instance, it is necessary to determine how finely
grained approximations for continuous energy con-
sumption processes should be. In [6] a discrete-event
approximation of those continuous behaviors is dis-
cussed, but depending on the research a more accurate
approximation can be required.
In [7] a simulator coupling is proposed to execute
manufacturing models with mixed discrete event and
continuous process behaviour, what we call hybrid
system dynamic. This approach is a customized solution
and it shows well-known problems of simulator cou-
plings.
SNE 27(2), 2017, 115 - 124, DOI: 10.11128/sne.27.tn.10377
Received: May 25, 2017,
Accepted: June 10, 2017 (Special Issue Review)
SNE - Simulation Notes Europe, ARGESIM Publisher Vienna
ISSN Print 2305-9974, Online 2306-0271, www.sne-journal.org
Pawle tta
et al.
A Multimodeling Approach for Simulation of Energy Consumption
116 SNE 27(2) 6/2017
T
N
A hybrid modeling approach based on the Discrete
Event and Differential Equation System Specification
(DEV&DESS) in [8] is discussed in [9]. It uses an in-
line integration method that schedules the integration
time as discrete events. Thus, continuous processes can
be modeled using ordinary differential equations and are
solved within a discrete event-oriented simulation envi-
ronment. Both approaches are limited according to the
modularity and clear separation between model specifi-
cation and simulation execution.
This paper is a refined version of [10] and introduces
a multimodeling approach for manufacturing systems to
overcome those inadequacies. According to the theories
in [11, 12], multimodeling means breaking a system into
a network or hierarchy structure of individual models.
The models may be specified by different dynamical
behavior or are described using different methods [13,
14]. Hence, the overall multimodel is from the dynam-
ical point of view often a hybrid model.
The approach is illustrated by the example of a com-
ponent based PL with an industrial furnace facility that
is refined using multimodeling in different layers. Be-
side the classical production performance indicators, it
predicts the time-dependent energy consumption of its
main consumer, the furnace facility. The prototypical
example is implemented in the MATLAB/Simulink [10]
environment using different modelling methods, such as
entities, events, statecharts, ODEs and DAEs. Some
parts of the implementation, pitfalls and simulation
results will be presented to strengthen important parts of
the approach.
1 Multimodeling Approach for
Manufacturing Systems
In the past, methods of modeling and simulation were
mainly used for the planning and optimization of the
operation of manufacturing systems to determine pro-
duction performance indicators, such as through-put
time, facility utilizations, etc. The usage of discrete
event modeling approaches is typical for those investi-
gations. Figure 1 shows such a model of a simple PL
implemented using SimEvents within MATLAB/ Sim-
ulink. The PL is reduced for simplicity to a minimal
structure composed of: (i) a source component for gen-
erating the parts, called entities; (ii) a queue component
with FIFO policy; (iii) a server component named fur-
nace; (iv) a sink component for handled parts.
Statistical ports and signal scopes are omitted in the
figure. Such a discrete event-based simulation model
allows the prediction of the above mentioned perfor-
mance indicators.
Figure 1: Simple discrete event-based production line
model with a furnace component in
SimEvents
.
However, a precise determination of time-dependent
energy consumptions of specific components or of an
overall system requires a more detailed modeling of
relevant components. In the case of the PL, the abstrac-
tion of the furnace facility as server in the sense of
queuing theory is insufficient. Its abstraction has to be
refined. According to [11, 14], states and events at this
level of abstraction have to be refined to more accurate
events and states at a next lower level. Such refinement
leads to a model or network of models at a next lower
level, which may be subject to refinement in subsequent
steps. The resulting model of such a refined component
is called a multimodel, consisting of interacting sub-
models. The multimodel of a component itself or the
overall model often combines several modeling para-
digms, and operates with different scales, or is a hybrid
system, according to [13], if it includes both continuous-
time and discrete-event behavior.
For a refinement of system components in discrete
event-based manufacturing models we suggest a layer
structure, as illustrated in Figure 2. Each layer repre-
sents a specific aspect of the component with well-
defined interaction relations between the layers. This
approach corresponds to the multilayered architecture, a
common design pattern, used in software development
[16]. Figure 2 suggests three general layers to refine a
manufacturing system component, in which the models
of different layers should specify the following charac-
teristics.
•
The material flow layer describes as highest
abstraction the event-based flow of entities (i.e. parts)
into and out of the component. It is the basic layer for
connecting components to a production line or process
chain model, such as illustrated in Figure 1.
This layer may be refined in further steps using the
entity-based modeling method to map internal
material flows in more detail or to provide an
interface to the other layers.
Pawletta
et al.
A Multimodeling Approach for Simulation of Energy Consumption
SNE 27(2) 6/2017 117
T
N
•
The process control layer maps the local process
control operations of the component. This is
especially important for components with several
internal manufacturing operations, where the parts
are handled in several phases, which may be iterated
according to an internal control program or other
internal conditions.
•
The process physics layer implements details of
internal process operations that are relevant in the
different manufacturing phases, such as energy flows,
chemical reactions etc.
Figure 2: General layer structure to refine a manufacturing
system component.
Depending on the specific characteristics of a compo-
nent, the suggested layers can be arranged hierarchically
or as a network of models to form a multimodel. Of
course, each layer itself can be refined using models at a
next lower level or a layer may be omitted. It is im-
portant to adapt the level of abstraction to the questions
at hand and only include the processes that are needed
to answer them.
Generally, each layer should provide its specific
kind of information, which influences the interaction
relations between the layers. The material flow layer is
mainly concerned with logistic quantities such as wait-
ing times and utilizations; the process control layer
gives the order and timing of the different manufactur-
ing phases, which can be useful in the context of other
information. The kind of information provided by the
physics layer can vary widely depending on the specific
component characteristics and implemented details.
The various layers typically demand diverse model-
ing approaches and, depending on the level of abstrac-
tion, their scales are often different. For instance, in the
considered example of the furnace component energy
flows are accurately modeled based on physical laws,
which are described by ODEs or DAEs.
2 Hybrid System Modeling and
Simulation
The suggested multimodeling approach for manufactur-
ing system modeling combines several modeling meth-
ods to describe different dynamical behaviour, called a
hybrid system. Subsequently, we want to highlight some
modelling methods and related simulation software.
2.1 Modeling methods
We will consider the modeling methods regarding the
suggested layer structure in Figure 2. The material flow
layer is often specified using a discrete-event modeling
method, which defines abstract entities moving between
stationary components and acting on them [8]. The
entities are identified in manufacturing systems with
workpieces or tools; temporary components move be-
tween production facilities.
A convenient method to describe the different manu-
facturing phases in a production facility, which is the
concern of the process control layer, is that of state
graphs [17]. The phases directly correspond to the states
and the transitions describe the internal process logic.
Alternatively, one could again use a process-based ap-
proach, wherein the entities denote abstract control
tokens.
The actual manufacturing operations, here summa-
rized under the term process physics, are often modeled
based on natural or technical laws, e.g. from mechanics,
thermodynamics or chemistry. This results in continu-
ous models based on differential equations (ODEs). If
the description contains algebraic constraints or the
equations are constructed automatically using a physical
modeling approach [18], then the mathematical model is
enlarged to a system of differential algebraic equations
(DAEs).
2.2 Related simulation tools
For the simulation of the logistic and process-oriented
aspects of a production system several discrete event-
based simulation environments exist and are in wide
industrial use, such as Arena [19] and Plant Simulation
[20]. Usually these programs lack algorithms such as
ODE or DAE solvers to cope with continuous system
specifications. This is why different simulators are cou-
pled to solve such problems, such as in [7]. The intro-
duced multimodel structure supports such simulator
couplings, but it cannot rectify its general problems.
Pawle tta
et al.
A Multimodeling Approach for Simul ation of Energy C onsum ption
118 SNE 27(2) 6/2017
T
N
Instead, one should use a software environment that
is capable of hybrid modeling and simulation. Such an
environment, increasingly used for manufacturing sys-
tem simulation, is AnyLogic [21]. It offers system dy-
namics, discrete event and agent-based methods. How-
ever, the mapping of complex ODEs to system dynam-
ics diagrams is quickly confusing and physical model-
ing techniques, according to [18], are not supported. As
a consequence, it may be only a useful choice for mul-
timodeling problems with relatively simple continuous
process physics.
Another widely used software supporting multimod-
eling, although less popular in the manufacturing simu-
lation domain, is the Matlab/Simulink environment.
Originally designed for the simulation of continuous
systems using the signal flow paradigm, it can be ex-
tended using additional toolboxes and blocksets to in-
clude discrete, discrete event or physical modeling fea-
tures: (i) The SimEvents blockset enables discrete event
modeling based on the entity approach; (ii) Stateflow
provides state chart modeling techniques; and (iii) Sim-
scape expands the continuous tool chest with physical
modeling features. This software environment provides
the widest range of features for multimodeling today
and it is already used and accepted in other engineering
domains. Hence, it will be used in the following to illus-
trate and validate the suggested multimodeling approach
by implementing some concrete examples.
An alternative choice could be to use a Modelica
based solution [18] with the additional packages de-
scribed in [22, 23] or Ptolemy [24] with the OpenModel-
ica extension according to [25]. However, both Modeli-
ca and Ptolemy are even more unknown than
MATLAB/Simulink in the manufacturing system com-
munity.
3 Basic Application to a
Manufacturing System
Component
To illustrate the introduced approach the furnace com-
ponent of the PL model in Figure 1 will be analyzed for
multimodeling and a set of models with different levels
of abstraction will be designed.
3.1 Multimodeling of furnace component
Our objective of multimodeling is the refinement of the
furnace component to investigate its time-related energy
consumption. Industrial furnaces are widely used in
metalworking processes and are one of the most exten-
sive energy consumers in manufacturing systems. In
addition, their internal operation is generally rather
complex, making this an ideal example for refinement
using different modeling approaches. The operation of
such a furnace is patterned in the following after the
descriptions in [9, 26].
When parts arrive at the furnace, they are collected
until a given batch size is reached. A complete batch
then enters the furnace, is processed and leaves the
furnace. Then, the batch will probably be resolved for
processing the parts by other facilities. The refinement
of such internal processes of a component is part of the
material flow layer, as demonstrated in Figure 2.
Moreover, the heat treatment itself consists of sever-
al phases that are implemented by a local control; these
are mapped and refined at the process control layer.
Figure 3 shows an example of such a control with six
operation phases.
Figure 3: State graph describing a local control of
furnace operations.
The idle phase spans the times before parts have entered
and after all parts have left the furnace. During the load
phase parts enter the furnace. In the heat-up phase the
furnace is heated until a given temperature is reached;
this is then held constant during the hold phase. Accord-
ing to the requirements, the heat-up and hold phases can
be iterated several times with different temperatures.
Finally, the parts leave the oven in the unload phase. An
additional cool-down phase may be included either to
make sure that the parts leave the furnace with a moder-
ate temperature or to describe a shutdown of the fur-
nace.
Pawletta
et al.
A Multimodeling Approach for Simulation of Energy Consumption
SNE 27(2) 6/2017 119
T
N
Since the focus of our example study lies on the en-
ergy consumption, the heat flows in the furnace have to
be considered in more detail, and must be refined at the
process physics layer (Fig. 2). The energy source is the
power supply of the actual heater. From here the heat
flows mainly through convection and radiation process-
es to the parts and to the internal structures and the
casing of the furnace. During the heat-up and hold phas-
es losses are mainly due to conduction through the cas-
ing into the environment, while in the load or unload
phases additional losses are caused by the open doors.
Because of the complicated geometry, the physical
details, especially of the convection processes, are also
rather complicated. However, for the estimation of the
total heat flows common approximative methods usual-
ly give quite accurate results. Concrete mathematical
models for specifying the process physics will be con-
sidered afterwards in context with their prototypical
implementation.
Based on the previous considerations, Figure 4
shows the multimodel structure with defined interfaces
for the refinement of our furnace component based on
the layers introduced in Figure 2. In this abstraction it
consists of three interacting models: (i) MF for the ma-
terial flow; (ii) PC for the process control; and (iii) PP
for the process physics. The labels (B) and (C) are not
of interest at this point.
The models have to communicate in several ways:
•
MF receives parts from the external input port <1>
and sends to the PC the number of parts that have
entered the MF.
•
When the batch size is reached, the PC starts the PP,
which models the different manufacturing phases
during the operation of the furnace.
•
During the operation of the furnace the PC signals
the current manufacturing phase to the PP, which
adapts the internal heat flows accordingly. This can
mean changing the supplied heat between the heating
and holding phases or increasing the losses due to the
doors being open while loading or unloading.
•
In return the PP sends the current temperature values
of the furnace and the parts to the PC, which uses
them to determine whether the heat-up phase or the
optional final cool-down phase of the furnace is
complete (Fig. 3).
•
When the manufacturing phase unload (Fig. 3) is
finished, the PC sends a leaving signal to the MF,
which accordingly forwards the processed parts to its
external output port <2>.
Figure 4: Basic multimodel structure with defined
interfaces for the refinement of the furnace
component.
Each model has its specific set of parameters and output
quantities: The MF defines the batch size and logistic
properties such as average waiting time and machine
utilization; the PC gets the heating program and outputs
the time in the different manufacturing phases. The PP
needs a lot of physical parameters for the calculation of
the heat flows and provides the power requirements
during the process phases as well as the total energy
consumption.
3.2 Design of a model library for the furnace
component
Based on the basic multimodel structure in Figure 4, a
set of models for each layer has been implemented and
organized in a library (Fig. 5). The several models use
different modeling methods or have varying levels of
detail. They are labeled with two letters denoting the
layer and a third letter in brackets giving the level of
detail in ascending order, with (A) being the simplest
model:
•
The basic material flow model MF(A) uses only a
simple server, while MF(B) explicitly contains an
input tray, where the batch is compiled.
•
The process control subsystem PC(A) uses a simple
entity-based model, implemented in SimEvents, to
describe one pass through the four basic process
phases and ignoring idle and cool-down phases. PC(B)
enables a repetition of heat-up and hold phases
according to its heat program parameter, implemented
as an entity-based model in SimEvents. Additionally,
PC(C) adds a cool-down phase, but because of the
more complex control logic it is implemented using
the state machine approach with Stateflow.
Pawle tta
et al.
A Multimodeling Approach for Simulation of Energy Consumption
120 SNE 27(2) 6/2017
T
N
•
The process physics model PP(A) uses only the
internal oven temperature and a simple formula for the
global losses, while PP(B) adds the temperature of the
parts, the heat transfer between oven and parts and
additional losses during the load and unload phases.
Both use standard Simulink blocks to implement the
corresponding differential equations. Finally, PP(C)
employs physical modeling, which is modeled using
Simscape for the same physical processes as PP(B).
This makes the physical model structure more trans-
parent and easier for engineers to expand.
Figure 5: Library with models for composing several
multimodels with different levels of detail for the
furnace component.
4 Some Modeling &
Implementation Details
In the following, three multimodel variants for the fur-
nace will be described in more detail: (i) a very basic
multimodel named ovenBAA the capital letters stand
for the composition of MF(B), PC(A) and PP(A) mod-
els; (ii) a medium complex multimodel ovenBCA; and
(iii) the most complex multimodel variant ovenBCC.
While this section is devoted to some implementation
details of the single models used in the three multimod-
els, the next one will discuss some simulation results.
The implementation was carried out using the
MATLAB/Simulink environment and related tools.
4.1 Material flow model
All of the examples considered here use the extend-
ed model MF(B) for mapping the internal material flow.
The entity-based model structure of MF(B) (Fig. 6)
consists of two simple servers: the first for the input
tray; and the second (N-Server) for the furnace proper.
Both hold incoming entities (i.e. parts) up to the given
batch size, until their succeeding gates open to transfer
the entities to the next stage. The intermediate Gate
guarantees that a full batch is always delivered to the
furnace; the final Release Gate is triggered by the
model at the process control layer (Figs. 7, 8) at the
end of the unload phase.
Figure 6: Discrete event-based model of MF(B) using
SimEvents
.
4.2 Process control models
The process control model starts when the number of
parts that have entered at the material flow layer is equal
to the batch size. In the process control model PC(A)
(Fig. 7), implemented using SimEvents, this leads to the
creation of a control entity that passes through a line of
servers denoting the different phases. Except for one, all
servers simply have a fixed processing time; only the
heat-up phase is different. Here the entity is held in a
server until the current oven temperature, which is com-
puted in the model at the physics layer (Figs. 9, 10), has
reached the given temperature Tset. After the unload
phase, the control entity is destroyed and the wake-up
signal (trigger) is generated, which opens the Release
Gate in the model at the material flow layer (Fig. 6).
Figure 7: Discrete event-based model of PC(A) using
SimEvents
.
Pawletta
et al.
A Multimodeling Approach for Simulation of Energy Consumption
SNE 27(2) 6/2017 121
T
N
The much more elaborate process control model
PC(C) (Fig. 8) incorporates all six phases, pictured in
Figure 2, as well as possible repetition of the heat-up
and hold phases. According to the theory in [8], it is a
hybrid DEV&DESS model. It is modeled based on the
state machine approach, but it includes continuous state
event handling and has been mainly implemented using
Stateflow. The two auxiliary subsystems (schedTEvent,
checkTemperature) create events when the waiting time
has changed or the heat-up or cool-down temperature
has been reached. The model PC(C) is essentially an
adapted version of a model described in [9].
Figure 8:
Discrete event-based model with continuous
state event handling of PC(C) mainly using
Stateflow
.
4.3 Process physics models
The basic process physics model PP(A) (Fig. 9) uses a
simple power balance to compute the change of the
oven temperature :
(1)
where is the total heat capacity of the oven. The
power loss is computed with Newtons simple law of
cooling:
(2)
where is the temperature of the surroundings and
a constant that subsumes all convective and conductive
processes. The heating power is assumed to be constant
during the heat-up phase and to match the losses during
the hold phase:
(3)
For the computation of the total power demand of
the furnace, a constant power is added, which sub-
sumes all non-heating processes, such as a base load or
the power electronics.
Figure 9: ODE-based model of PP(A) using Simulink.
The model PP(C) (Fig. 10) uses physical modeling
based on Simscape to incorporate much more physical
details, such as convection and radiation from the fur-
nace during load and unload phases to the environ-
ment. The governing differential algebraic equations are
not built up explicitly here; instead, the physical com-
ponents such as heat capacities and various kinds of
heat flow are represented directly, which makes the
physical structure of the model much clearer.
Figure 10: Physical model (DAE) of PP(C) using
Simscape
.
4.4 Implementation pitfalls
The extension of MATLAB/Simulink by several packag-
es (blocksets) is necessary to make a multi-paradigm
model possible, but it also leads to small inconsisten-
cies, which have to be overcome. A minor nuisance here
is the large number of different ways used to express
values and signals in the packages used: Simple scalar
values can be time-based in Simulink, event-based in
SimEvents or physical-valued in Simscape. An event can
be defined by a sample time hit, a rapid value change
(edge), a special trigger signal or a function call.
Pawle tta
et al.
A Multimodeling Approach for Simulation of Energy Consumption
122 SNE 27(2) 6/2017
T
N
To clearly define a common interface for the inter-
acting models, their external values are fixed in the
following way: The temperature and phase values are
time-based; the number of parts is event-based; and the
leaving signal is a function call. If the internal imple-
mentation of a block generates differing types, then one
has to use one of the many converter blocks to get the
proper kind of signal, e.g. the round Event to Timed
Signal blocks that can be seen in Figures 7 and 8.
This problem is especially annoying in the physical
modeling environment. The large number of necessary
converters and reference points clutters the model and
destroys the clear physical structure. Hiding them in
subsystems is an obvious way to regain an ordered visi-
ble representation of the underlying physics model.
5 Exemplary Simulation Results
Using the introduced approach, a large number of simu-
lation studies is possible to calculate multifaceted per-
formance indicators. We will restrict our exemplary
consideration to the energy consumption of the furnace
that is the main consumer in our simple production line.
Moreover, we will investigate the costs of refinement
relating to the simulation run time by comparing differ-
ent multimodels, which use different modeling para-
digms or are based on a different level of detail.
5.1 Results relating to the energy aspect
In the exemplary experiment 24 parts were processed
using a batch size of six. After the heat-up phase their
temperatures were held first to 400 for 40 minutes,
then to 800 for 30 minutes. Figure 11 shows some
simulation results related to the energy consumption,
calculated using our most complex multimodel variant
ovenBCC: the temperatures of the furnace and parts, the
power consumption of the furnace, the accumulated
used energy and the current internal manufacturing
phase, each as functions over time.
The results are similar to those presented in [9], but
ovenBCC incorporates more details, especially of the
physical model layer. The other two example models
ovenBAA and ovenBCA basically reproduce the results
from [9], since they are based on the same physical
model assumptions. For comparison, Table 1 shows the
total energy use E1 at the exit of the last part and E2 at
the end of the simulation; Figure 12 displays the power
consumption for the three models.
Model E
1
[kWh] E
2
[kWh]
ovenBAA 71.3 108.8
ovenBCA 69.0 106.6
ovenBCC 82.6 117.6
Table 1: Comparison of the total energy consumption.
Figure 11: Simulation results of model variant ovenBCC.
The higher energy needs in variant ovenBCC are due to
the heating of the parts, which enter the oven with the
low temperature of the environment an effect that has
not been taken into account in the other models. The
small difference between the two simpler multimodel
variants results from the different timing of the phases,
as can be seen from Figure 12.
It is interesting to note that the total energy results
only differ by 10
% - 15
%. If this level of accuracy is
sufficient for the question at hand, e.g. for a global as-
sessment of a complex production line, then one can use
one of the simple physics models. However, Figure 12
shows that for a detailed examination of the power
needs during the individual phases one has to stick to
the complexities of multimodel variant ovenBCC.
Pawletta
et al.
A Multimodeling A pproach for Simul ation of E nergy C onsum ption
SNE 27(2) 6/2017 123
T
N
Figure 12: Comparison of the power results.
5.2 Comparison of run times
How much do we pay for the additional level of detail
in the complex multimodel? For a comparison of run
times, the number of parts was raised to 600 and three
runs were performed for each multimodel. To get rid of
loading and compile times, the means of only the last
two results were taken, leading to the values in Table 2.
Model Run Time [s]
ovenBAA 25.1
ovenBBA 27.7
ovenBCA 20.7
ovenBCB 39.3
ovenBCC 49.8
Table 2: Comparison of run times.
The results show that the complex physical model using
Simscape in multimodel ovenBCC (line 5) costs a factor
of two in execution time, which can be reduced in this
case to a factor of 1.6 by replacing DAE-based physical
modeling with standard ODE methods using Simulink
(ovenBCB, line 4). However, we see essential differ-
ences in execution time for multimodels with the same
simple process physics model PP(A) (line 1-3), which
are surprising.
The multimodels ovenBBA (line 2) and ovenBCA
(line 3) use process control models of nearly the same
complexity, but in ovenBBA implemented with Si-
mEvents and in ovenBCA with Stateflow. Moreover, the
multimodel ovenBCA (line 3) uses the more complex
process control model PC(C), than ovenBAA (line 1)
with the PC(A) model that was implemented with Si-
mEvents. Apparently, the implementation of SimEvents
has some potential for optimization.
6 Conclusion
The introduced multimodel approach supports the com-
ponent-oriented refinement of manufacturing system
models, using various modeling methods and models
with different levels of abstraction. Generally, it sug-
gests a refinement of components based on the three
logical layers material flow, process control and process
physics.
The first layer maps the internal material flow. It de-
livers an external input and output interface for connect-
ing with other components in a manufacturing system
model and an internal interface for communication with
the second layer. The dynamic behavior at this layer is
generally discrete event-based. In the second layer local
process control operations are modeled. This layer pro-
vides event-based control inputs for the other two lay-
ers, but it could be necessary to handle state events of
continuous values as well. In the third layer specific
process operations should be mapped. The dynamic
behavior at this layer can vary significantly and particu-
larly depends on the level of abstraction. Of course,
depending on the characteristics of a system component
and the intended level of abstraction, layers may them-
selves be omitted or refined using several interacting
models.
The paper illustrated the approach using the example
of a production line with an industrial furnace as the
main component. According to the layers, models with
different levels of abstraction have been implemented
for the furnace, which could be aggregated to a multi-
model. Then, the energy consumption of the furnace
was investigated with three different multimodels and
the simulation run times were measured. Hence, accura-
cy effects and computing costs could be compared.
The approach opens many new ways for investigating
manufacturing systems, but the complexity is increasing.
Hence, in the next step it should be combined with met-
amodeling techniques such as those considered in [27].
Pawle tta
et al.
A Multimodeling Approach for Simul ation of Energy C onsum ption
124 SNE 27(2) 6/2017
T
N
Acknowledgments. Thorsten Pawletta and Artur
Schmidt thank the German Research Foundation for
funding this research (no. PA 631/2-2). Peter Junglas is
grateful for the hospitality extended to him by Tom
Schramm at the Department of Geomatics, HCU Ham-
burg, during his work on this project.
References
[1] Lefeber E, Rooda JE. Modeling and Analysis of Manu-
facturing Systems. In: Fishwick PA. editor. Handbook of
Dynamic System Modeling, Boca Raton: Cpaman &
Hall/CRC, 2007, p. 34_1-34_20.
[2] Eisele C. Simulation-based optimization of the electrical
consumption of metal-cutting machine tools (German).
[dissertation], TU Darmstadt, Germany, 2014.
[3] Weinert N. An approach for the planning and operation
of energy-efficient production systems (German). [disser-
tation], TU Berlin, Germany, 2010.
[4] Larek R. Resource-efficient design of manufacturing
process chains by simulation and numerical optimization
(German). [dissertation], Univ. Bremen, Germany, 2012.
[5] Schrems S. A method for the model-based integration of
machine-specific energy demand in production planning
(German). [dissertation], TU Darmstadt, Germany, 2014.
[6] Larek R, Brinksmeier E, Meyer D, Pawletta T, Ha-
gendorf O. A discrete-event simulation approach to pre-
dict energy consumption in machining processes. Pro-
duction Engineering Research and Development, Ber-
ling: Springer Pub., 2011, p. 575-579.
[7] Peter T, Wenzel S. Simulation-based Planning and Eval-
uation of Energy Efficiency for Production Systems in
Car Manufacturing (in German). Simulation in Produc-
tion and Logistics, Fraunhofer Pub., Stuttgart, Germany,
2015, p. 535-544.
[8] Zeigler BP, Kim TG, Prähofer H. Theory of Modelling
and Simulation. SanDiego: Acad. Press, 2nd ed., 2000,
510 p.
[9] Schmidt A, Pawletta T. Hybrid modeling of manufactur-
ing process chains and their energy consumption in a
discrete event simulation environment (German). In:
Wittmann, J. editor. Proc. ASIM 2014-22. Symp. Simula-
tionstechnik, Sep 2014: Berlin, Germany, p. 109-116.
[10] Schmidt A, Pawletta T, Junglas P. A layered structure
for modeling manufacturing processes with the inclusion
of energy consumption. In: Proc. ASIM Symp. STS/
GMMS, Mar. 2016: Lippstadt, Germany, p. 155-164.
[11] Fishwick PA. Simulation Model Design and Execution:
Building Digital Worlds. Upper Saddle River: Prentice-
Hall Inc., 1995, 432 p.
[12] Park M, Fishwick PA, Lee J. Multimodeling. In: Fish-
wick P.A. editor. Handbook of Dynamic System Model-
ing, Boca Raton: Cpaman & Hall/CRC, 2007, p. 14_1-
14_28.
[13] Mosterman PJ. Hybrid Dynamic Systems: Modeling and
Execution. In: Fishwick PA. editor. Handbook of Dy-
namic System Modeling, Boca Raton: Cpaman &
Hall/CRC, 2007, p. 15_1-15_26.
[14] Shephard MS, Seol ES, Dale BF. Towards a Multimodel
Hierarchy to Support Multiscale Simulation. In: Fish-
wick PA. editor. Handbook of Dynamic System Model-
ing, Boca Raton: Cpaman & Hall/CRC, 2007, p. 12_1-
12_17.
[15] The MathWorks. Simulink: Simulation and Model-Based
Design. Online: www.mathworks.de/products/simulink/
(called 2015-11-11).
[16] Buschmann F, Meunier R, Rohnert H, Sommerlad P,
Stal M. Pattern-Oriented Software Architecture: A Sys-
tem of Patterns. vol. 1. New York: John Wiley & Sons,
1996, 476 p.
[17] Harel D. Statecharts: A Visual Formalism for Complex
Systems. Science of Computer Programming, 8, 1987, p.
231-274.
[18] Fritzson PA. Principles of Object-Oriented Modeling
and Simulation with Modelica 3.3. New York: Wiley &
Sons, 2015, 1256 p.
[19] Kelton WD, Sadowski R, Zupick N. Simulation with
Arena. 6th ed. New York: McGraw-Hill, 2014, 656 p.
[20] Bangsow S. Tecnomatix Plant Simulation: Modeling and
Programming by Means of Examples. Springer: Berlin,
2016, 713 p.
[21] Grigoryev I. AnyLogic 7 in Three Days. 2nd ed. Noth
Charleston: CreateSpace Independent Pub., 2015, 256 p.
[22] Elmqvist H, Gaucher F, Mattsson SE, Dupont F. State
Machines in Modelica. In: Otter M, Zimmer D, editors.
Proc. 9th Int. Modelica Conf., Sep 2012: Munich, Ger-
many, p. 37-46.
[23] Sanz V, Urquia A, Cellier FE, Dormido S. System mod-
eling using the Parallel DEVS formalism and the Model-
ica language. Simulation Modelling Practice and Theory,
18(7), 2010, p. 9981018.
[24] Ptolemaeus C. System Design, Modeling, and Simulation
using Ptolemy II. Ptolemy.org, 2014, 690 p.
[25] Mirzaei M. Integration of OpenModelica into the Multi-
paradigm Modeling Environment of Ptolemy II. Master
Thesis, Univ. of Linköping, Sweden, 2014.
[26] Heuer V, Löser K. Energy optimization of thermal-
chemical vacuum processes and aggregates in large-scale
productions (German). Elektrowärme International, 69,
2011, p. 261-268.
[27] Pawletta T, Schmidt A, Zeigler BP, Durak U. Extended
Variability Modeling Using System Entity Structure On-
tology within MATLAB/Simulink. In: Proc. SCS Int.
SpringSim/ANSS, Apr. 2016: Pasadena/CA, USA, 2016,
p. 62-69.
