Mathematics and Computers in Simulation 53 (2000) 333–344

Physically oriented modeling of heterogeneous systems

Peter Schwarz∗
Design Automation Department EAS, Fraunhofer Institute for Integrated Circuits,
Zeunerstrasse 38, D-01069 Dresden, Germany
Accepted 28 July 2000

Abstract
Technical systems can be characterized very often as complex heterogeneous. Many simulation algorithms exist
for the analysis of continuous and discrete systems. However, the first modeling steps in the physical domains
have not been sufficiently assisted by CAD tools. Multiports and generalized KIRCHHOFF’s networks proved to
be powerful concepts in a physically oriented modeling methodology. Modeling languages like VHDL-AMS and
Modelica will support the practical application of these approaches. © 2000 IMACS. Published by Elsevier Science
B.V. All rights reserved.
Keywords: Physically oriented modeling; Modeling languages; Simulation

1. Introduction

Modern technical systems like integrated circuits, micro-electro-mechanical systems (MEMS), mecha-
tronic systems or distributed automation systems can be characterized as complex heterogeneous systems.
Typically they show some of the following features:
• mixed-domain (mechanical, electrical, thermal, fluidic, . . . phenomena),
• partially close coupling between these domains, side effects, cross coupling,
• distributed and lumped (concentrated) elements,
• discrete and continuous signals and systems (in electronics: analog and digital),
• very large and stiff systems of differential equations to describe the continuous subsystems.
Depending on the level of abstraction, partial differential equations (PDE) and ordinary differential
equations (ODE) are the mathematical models (system equations) of continuous subsystems. We will
focus on ODE (and briefly discuss the transition from PDE to ODE in Section 3). A lot of algorithms and
computer programs is available for the numerical solution of the system equations. However, there is a
demand for more and better assistance in finding out these system equations. A powerful interdisciplinary
modeling methodology is necessary to analyze real-world problems.
∗
Tel.:+49-351-4640-730; fax: +49-351-4640-703.
E-mail address: [email protected] (P. Schwarz).

0378-4754/00/$20.00 © 2000 IMACS. Published by Elsevier Science B.V. All rights reserved.
PII: S 0 3 7 8 - 4 7 5 4 ( 0 0 ) 0 0 2 2 0 - 2
334 P. Schwarz / Mathematics and Computers in Simulation 53 (2000) 333–344

Fig. 1. Steps in modeling physical systems.

Basically, the modeling process can be divided into two fundamental steps (Fig. 1). At present, the
second step is mostly well supported by the input processing programs of simulators. So we will put
emphasis on the first step, the “physically-oriented modeling” of complex heterogeneous systems. The
term “physically-oriented” means: it is a goal of the modeling approach that as much as possible modeling
steps are closely related to the design process of technical systems and to the intuitive procedure of the
design engineer. Therefore, different description methods have to be considered. We can distinguish
between
• behavioral description: equation-oriented, similar to the textual notation of an ODE,
• structural description: the system model is hierarchically composed of subsystems and basic elements
(the so-called primitives) available with the simulators.
Mostly, system models with spatially lumped elements appear as implicit nonlinear differential-algebraic
equations (DAE). Only in special cases their formulation as explicit state equations is possible.
In Fig. 2, some well-known descriptions of the same system (a simple time-continuous mechanical
system with mass, spring, and damping) are shown:
• a mathematical description which may be processed by tools like Mathematica or Maple,
• the model formulated in a “hardware description language” (HDL) for the electronic system simulator
Saber,
• a block diagram (or signalflow diagram) to be performed by Matlab/Simulink or MatrixX,
• a mechanical network and its equivalent electrical network (which can be simulated by electronic
network analysis programs like SPICE).
Especially, the use of equivalent electrical networks for modeling non-electrical systems has a long and
successful tradition [10,15,16,18,20,31,39]. However, for a long time it had been restricted to relatively
small systems because of the cumbersome handling of the basic elements of older simulators like SPICE
(resistance R, capacitance C, inductance L, . . . ) in modeling complex functionalities. Also the treatment
of translational and rotational effects in three-dimensional mechanical systems is an awkward process.
This process can be considerably simplified by the more general “modeling by multiports” described
in the next section. But only the latest developments of new modeling languages and simulator features
make this multiport modeling approach practicable and powerful, thus, enabling the design engineer to
exploit the potential of mathematical algorithms.
 P. Schwarz / Mathematics and Computers in Simulation 53 (2000) 333–344 335

Fig. 2. Different descriptions of a mechanical system mathematical description.

Physical systems may be classified into:

• conservative physical systems: their quantities are flow quantities and difference quantities subject to
compatibility constraints,
• non-conservative physical systems: other interconnected systems consisting of elements with inputs
and outputs and only one type of quantities.
A flow quantity (through quantity, 1-quantity) is measured at one point of the physical system (e.g.
an electrical current or a hydraulic flow). A difference quantity (across quantity, 2-quantity) is measured
between two points, e.g. a mechanical displacement or an electrical voltage. In Fig. 3 some of the most
important flow and difference quantities are shown.
The term “conservative physical systems” is motivated by the fact that time-independent compatibility
constraints for flow and difference quantities are valid. In electrical systems, these constraints are the
well-known KIRCHHOFF’s current law (KCL, node law) and voltage law (KVL, mesh law). Similar
conservation laws for flow and difference quantities hold also in many other physical domains. This fact
can be reflected in different classes of system models: conservative system models with conservative
signals (flow and difference signals) and non-conservative systems (with non-conservative signals). We
use the term “signal” as a model of all kinds of physical quantities which are able to exchange energy
or information between the subsystems. We will use the term “network” to describe a system model
consisting of elements, connections between these elements, and conservative signals. To stress the fact

Fig. 3. Flow and difference quantities in different physical domains.

336 P. Schwarz / Mathematics and Computers in Simulation 53 (2000) 333–344

that these networks are models of physical systems in different domains, sometimes the term “generalized
KIRCHHOFFian network” is used [41,42].
Bond graphs [11,43] are closely related to networks [30], but they are beyond the scope of this paper.
From the author’s point of view, the construction of bond graphs as models of real physical systems seems
to be more complicated sometimes especially in the case of strongly coupled conservative elements or
subsystems. But the transformation of bond graphs into DAE systems is similar to the transformation of
networks and can be done automatically.
Non-conservative quantities are typical for control systems or digital signal-processing devices. Block
diagrams (see Fig. 2) or signal-flow graphs are appropriate models of these physical systems. In practice,
control systems are often realized as electronic systems where voltages are the signals. Currents do
not have any influence, and therefore, they are not included into the model. Conservation laws do not
exist on this level of abstraction. From the practical point of view, the distinction between “analog”
(time-continuous) and “digital” (time-discrete, mostly also value-discrete) non-conservative signals is
very important. For the simulation of discrete (or digital) systems powerful specialized algorithms are
available, e.g. the discrete-event simulation algorithms [2].

2. The multiport approach

The multiport modeling approach is illustrated in Fig. 4.

• The complete system is decomposed into subsystems, the so-called multiports. Signals are separated
into external signals between the subsystems, and internal signals within the subsystems. The external
signals of a subsystem form its terminal signals.
• External signals may be conservative (flow signals e5 , e6 , e7 , e8 , e9 , difference signals e1 , e2 in the
example) or non-conservative (e3 , e4 ).
• The behavior of subsystems depends only on their terminal signals (and some internal signals) and
their initial conditions, too.
In electrical network theory, multipoles and multiports may be distinguished as such subsystem models.
But in our more general context this distinction is not substantial, and therefore, we will use only the
term multiport. Each subsystem can be substituted by other subsystems with the same terminal behavior
[29] without any influence on the behavior of the rest of the system. The behavior of a subsystem can be
expressed

Fig. 4. Multiports in a system model.

 P. Schwarz / Mathematics and Computers in Simulation 53 (2000) 333–344 337

Fig. 5. Terminal signals of a multiport.

• implicitly by the connection of some other subsystems (or primitives on the lowest level): hierarchical
structural refinement,
• explicitly by a set of equations (e.g. nonlinear differential-algebraic equations): “behavioral modeling”
in a strict sense,
• or a combination of both if the simulator has language constructs to formulate mixed structural-
behavioral descriptions.
The key problem in setting-up the system equations is the description of the terminal behavior of the
subsystems. All other equations are essentially constraints resulting from the connection of the subsys-
tems, especially the node and mesh law for flow and difference signals on conservative terminals, and
can be constructed automatically.
An external view of a multiport is shown in Fig. 5. Terminal signals of the same type are assembled
into vectors. So these signals are vector-valued functions of time t. The terminal signals are divided into
the following categories:
v 1 , i 2 , a in , d in independently chooseable difference, flow, and non-conservative signals
(analog and digital)
v 2 , i 1 , a out , d out dependent difference, flow, and non-conservative signals (analog and digital)

In many cases it is impossible to state the terminal behavior only based on terminal signals. Subsystems
may have internal states, and therefore, the introduction of additional signals s is necessary. The terminal
behavior is determined also by the values of some parameters described by a parameter vector p . On very
wide assumptions, the terminal behavior of multiports (Fig. 5) can be given by the following equations:
Eqs. (1)–(4) are differential-algebraic equations (DAE). They can be solved by the numerical solution
algorithms of a continuous system simulator [8]. The last equation (5) is mostly handled by discrete-event
simulation algorithms which are very popular in general system simulation and in digital electronic
simulation [1,2,25]. The combined treatment of continuous and discrete equations is known as hybrid
simulation, analog-digital simulation, or mixed-signal simulation.
This multiport modeling approach is closely related to object-oriented modeling. Unfortunately, there is
no generally accepted understanding of object-orientation in modeling and simulation. From the author’s
point of view, the following classification is useful:
• Object-oriented modeling: the construction of strictly hierarchical, modular models. This interpretation
of object-oriented modeling was stressed by Cellier et al. [6,7,27].
• Object-oriented simulation: each subsystem is considered as an object. Each object has its own imple-
mented simulation algorithm (a “method”). All objects communicate via message passing, coordinate
338 P. Schwarz / Mathematics and Computers in Simulation 53 (2000) 333–344

their behavior, and so the simulation of the entire system is carried out [5]. A similar definition is:
object-oriented simulation is the concurrent operation of different simulators without a global con-
troller or a master simulator.
• Object-oriented programming: the application of programming languages like C++ or Smalltalk.
All these approaches exist independently from each other but they can be combined. Therefore, some au-
thors [3,17,33] emphasize that object-oriented programming is not only the application of object-oriented
programming languages but also a general method to design complex software systems. Hence, the method
is very useful in developing complex models too.
We will focus on object-oriented modeling. Modularity is guaranteed by the multiport approach because
the interaction between the subsystems occurs only by the signals on the terminals (the interface). There
are no global variables, and side-effects are excluded. Hierarchy is achieved by the structural refinement
mentioned above and by the hierarchy concepts of programming languages or HDL used for behavioral
modeling. Other aspects of modular modeling are
• the structural similarity between a functional-oriented partitioning of the original physical system and
the decomposition of the system model into subsystem models,
• the user has only to model the behavior of subsystems (in a hierarchical description: on a lower level)
and has to describe topologically the interconnection of the subsystems. Setting-up the equations
describing the whole system is the task of the simulator (or its input language processor),and has not
to be done by the user,
• with the same terminal description (interface) different behavioral descriptions may be used to obtain
appropriate degrees of accuracy and multi-level-simulation.
Object-oriented programming languages [17,33] are characterized by the concepts of data encapsulation
and information hiding, message passing, a class concept and instantiation of classes, inheritance for
building new classes, states and “methods” for changing the states. The application of these concepts
allows to support substantially the correct construction of modular behavioral descriptions in different
physical domains [12–14]. Some modern hardware description languages [34,24] and other modeling
languages [23] are object-oriented.

3. Reduction of distributed systems to lumped systems

System partitioning and the multiport approach are described in Figs. 4 and 5 for systems with con-
centrated (lumped) elements only. However, many physical systems have spatially extended components,
and fields have to be considered. Then, additional modeling steps are required to apply the multiport
concept. In Fig. 6, these steps are illustrated. The physical system is partitioned in such a way that field
regions are transformed onto connections. By integration over these field regions, the field quantities
are transferred to signals carried by connections and terminals, respectively. Conservation laws hold in
electrical, magnetic, thermal, . . . fields and, therefore, conservative signals have to be used.
In practice, discretization methods are used to construct network models [28]. By spatial discretization,
a PDE may be transformed into a set of ODEs which can be interpreted as a connection of network
elements or multiports. In Fig. 7, a well-known example is presented.
Another modeling approach is the combination of the multiport concept with the Finite Element Method
(FEM). Both methods are similar: partitioning the system and introducing “nodes” as carriers of flow and
difference quantities. The basic idea of the FEM [4,22,44]:
 P. Schwarz / Mathematics and Computers in Simulation 53 (2000) 333–344 339

Fig. 6. Distributed and concentrated systems.

Fig. 7. Network interpretation of a discretized partial difference equation (thermal conduction).

• definition of an energy functional of the whole system as the sum of the functionals of the subsystems,
• minimizing this energy functional by appropriate choice of the node signals,
• choice of approximating “shape functions” for the quantities inside of the subsystems
can be applied to the construction of behavioral models of subsystems which can be connected to form
a network [26,32,35–37]. In Fig. 8, the approach is illustrated generally.

Fig. 8. FEM approach and multiports.

340 P. Schwarz / Mathematics and Computers in Simulation 53 (2000) 333–344

Fig. 9. Behavioral model of a beam element.

In Fig. 9, a mechanical beam is choosen as an example. The static behavior of the beam with respect
to global coordinates (x, y, z) is given by the equation in Fig. 9. C is the (3 × 3) transformation matrix
between the local coordinate system (l, m, n) and the global system (x, y, z), E is the modulus of elasticity,
Im and In are the planar moments of inertia [26,22].
The multiport modeling approach could be applied very successfully in modeling complex microme-
chanical sensors [19,26,38]. Fig. 10 shows the microphotograph of an acceleration sensor, Fig. 11, illus-
trates the modeling approach (partitioning into subsystems, modeling the subsystem behavior by FEM
methods as multiports, interconnecting all multiports to model the entire system). In Fig. 12, a rotational
sensor and the corresponding multiport network with conservative as well as non-conservative signals is
presented.

Fig. 10. Translational acceleration sensor (Bosch).

 P. Schwarz / Mathematics and Computers in Simulation 53 (2000) 333–344 341

Fig. 11. Multiport model of the translational acceleration sensor.

4. Modeling languages

If the design engineer is able to formulate directly the model equations then they can be solved with
programs like Mathematica, Maple, Matlab, MatrixX or ACSL. Graphical front ends (Simulink, Sys-
temBuild) are important tools to capture the structure of the system. Another approach is the application
of simulator-independent modeling languages (in electronics: Hardware Description Languages, HDL).
Modern languages like VHDL-AMS [21,40] and Modelica [23] support all these aspects of systems and
signals discussed above:
• multi-domain,
• conservative and non-conservative,

Fig. 12. Rotational acceleration sensor (Bosch) and its multiport model.
342 P. Schwarz / Mathematics and Computers in Simulation 53 (2000) 333–344

Fig. 13. Skeleton of a VHDL-AMS description of the multiport behavioral equations (in relation to Fig. 5).

• continuous and discrete (mixed-signal),

• hierarchical structural and behavioral description.
They have been provided with features for the precise definition of terminals at the subsystem interfaces
and of the physical nature of quantities and signals. Hierarchical and modular modeling is assisted, in
addition Modelica is an object-oriented language. VHDL-AMS is under IEEE standardization and will
be supported by several system simulators in the near future. At the moment, Modelica is accepted only
by a prototype simulator developed on the Dymola simulator [9] basis.
In Fig. 13 the structure of a VHDL-AMS description of the multiport behavioral Eqs. (1)–(5), see also
Fig. 5, is shown. The interface description (ENTITY) is strictly separated from the behavioral description
(ARCHITECTURE). It is allowed to combine different ARCHITECTURE bodies with one ENTITY. In
this way it is possible to use models with different levels of abstraction and of numerical accuracy by
exchanging the ARCHITECTURE part without any influence on the structure of the whole system.
The impressive examples of Figs. 11 and 12 illustrate the power of the generalized network modeling
approach but its practicability depends strongly on the modeling language available in modern system
simulators.

Acknowledgements

I would like to thank my colleagues in the modeling and simulation group at FhG EAS Dresden and
Kurt Reinschke for many years of fruitful cooperation as well as Karl-Heinz Diener, Joachim Haase,
Roland Jancke, and Gunter Kurth for their assistance in preparing this paper. The funding by the German
Ministry of Research (BMBF) in different projects and in DFG-SFB 358 “Automated System Design”
is also gratefully acknowledged.
 P. Schwarz / Mathematics and Computers in Simulation 53 (2000) 333–344 343

References

[1] J.R. Armstrong, Chip-Level Modeling with VHDL, Prentice Hall, Englewood Cliffs, 1989.
[2] J. Banks, Handbook of Simulation, Wiley, New York, 1998.
[3] J.-M. Berge, Oz. Levia, J. Rouillard, Object-Oriented Modeling, Kluwer, Dordrecht, 1993.
[4] D. Braess, Finite Elemente, Springer, Berlin, 1997.
[5] J.T. Buck, S. Ha, E.A. Lee, D.G. Messerschmidt, Ptolemy: a framework for simulating and prototyping heterogeneous
systems, Int. J. Comp. Simul. 4 (4) (1994) 155–182.
[6] F.E. Cellier, Continuous System Modeling, Springer, New York/Berlin, 1991.
[7] F.E. Cellier, H. Elmquist, M. Otter, Modeling from physical principles, in: W.S. Levine (Ed.), The Control Handbook, CRC
Press, Boca Raton, FL, 1996, pp. 99–107.
[8] C. Clauß, J. Haase, G. Kurth, P. Schwarz, Extended admittance description of nonlinear n-poles, Archiv Elektronik und
Übertragungstechnik 40 (2) (1995) 91–97.
[9] H. Elmquist, Dymola Dynamic Modeling Language, Users Manual, Dynasim AB, Lund, 1994. See also
http://www.dynasim.se.
[10] F.A. Firestone, The mobility method for computing the vibration of linear mechanical and acoustical systems:
mechanical-electrical analogies, J. Appl. Phys. 9 (6) (1938) 373–387.
[11] D.C. Karnopp, D.L. Margolis, R.C. Rosenberg, System Dynamics: A Unified Approach, Wiley, New York, 1990.
[12] R. Kasper, W. Koch, Object-oriented behavioral modelling of mechatronic systems, in: Proceedings of the 3rd Conference
Mechatronics and Robotics, Paderborn, 1995, pp. 70–84.
[13] A. Kecskeméthy, Objektorientierte Modellierung der Dynamik von Mehrkörpersystemen mit Hilfe von Übertragungs-
elementen, VDI-Fortschrittsberichte Reihe 20, Nr. 88, Düsseldorf, 1993.
[14] A. Kecskemethy, M. Hiller, An object-oriented approach for an effective formulation of multibody dynamics, in: Proceedings
of the 2nd US National Congress Computational Mechanics, Washington, 1993.
[15] A. Klein, G. Gerlach, System modelling of microsystems containing mechanical bending plates using an advanced network
description method, in: Proceedings of the MICRO SYSTEM Technologies, Potsdam, 1996, pp. 299–304.
[16] H.E. Koenig, W.A. Blackwell, Electromechanical System Theory, McGraw-Hill, New York, 1961.
[17] W.P. Kowalk, System Modell Programm, Spektrum Akademischer Verlag, Heidelberg, 1996.
[18] A. Lenk, Elektromechanische Systeme, Vol. 3, Verlag Technik, Berlin, 1971–1973.
[19] G. Lorenz, R. Neul, Network-type modeling of micromachined sensor systems, in: Proc. MSM’98, 1998, pp. 233–238.
[20] R.H. MacNeal, The solution of elastic plate problems by electrical analogies, J. Appl. Mech. 18 (1) (1951) 59–67.
[21] H.A. Mantooth, M.F. Fiegenbaum, Modeling with an Analog Hardware Description Language, Kluwer, Dordrecht, 1994.
[22] U. Meissner, A. Menzel, Die Methode der Finiten Elemente, Springer, Berlin, 1989.
[23] Modelica, see http://modelica.org many links to Modelica-related publications.
[24] W. Müller, F. Rammig, ODICE: Object-oriented hardware description in CAD environment, in: Proceedings of the 9th
Conference on Computer Hardware Description Languages (CHDL’90), Elsevier, Amsterdam, 1990, pp. 19–34.
[25] Z. Navabi, VHDL Analysis and Modeling of Digital Systems, McGraw-Hill, New York, 1993.
[26] R. Neul, et al. A modeling approach to include mechanical microsystem components into system simulation, in: Proceedings
of the Design, Automation and Test Conference (DATE’98), Paris, 1998, pp. 510–517.
[27] M. Otter, Objektorientierte Modellierung mechatronischer Systeme am Beispiel geregelter Roboter, Dissertation, Bochum,
1994.
[28] G. Pelz, et al. MEXEL: Simulation of microsystems in a circuit simulator using automatic electromechanical modeling, in:
Proceedings of the MICRO SYSTEM Technologies, VDE-Verlag, Berlin, 1994, pp. 651–657.
[29] A. Reibiger, On the terminal behaviour of networks, in: Proceedings of the ECCTD’85, Prague, September 1985,
pp. 224–227.
[30] A. Reibiger, H. Loose, Bond graphs and matroids, in: Proceedings of the 3rd IMACS Symposium MATHMOD, Vienna,
February 2000, pp. 303–308.
[31] K. Reinschke, P. Schwarz, Verfahren zur rechnergestützten Analyse linearer Netzwerke, Akademie-Verlag, Berlin, 1976.
[32] B.F. Romanowicz, Methodology for the Modeling and Simulation of Microsystems, Kluwer, Dordrecht, 1998.
[33] J. Rumbaugh, et al., Object-Oriented Modeling and Design, Prentice Hall, Englewood Cliffs, 1991.
[34] G. Schumacher, W. Nebel, Survey of languages for object-oriented hardware design methodologies, in: J.-M. Berge, et al.
(Eds.), High-Level System Modeling: Specification Languages, Kluwer, Dordrecht, 1995.
344 P. Schwarz / Mathematics and Computers in Simulation 53 (2000) 333–344

[35] P. Schwarz, J. Haase, Behavioral modeling of complex heterogeneous microsystems, in: Proceedings of the 1st International
Forum on Design Languages (FDL’98), Lausanne, September 1998, pp. 53–62.
[36] S.D. Senturia, CAD Challenges for Microsensors, Microactuators, and Microsystems, Proc. IEEE’86 (8) (1998) 1611–1626.
[37] S. Senturia, N.R. Aluru, J. White, Simulating the behavior of MEMS devices: computational methods and needs, in: Proc.
IEEE Trans. Comput. Sci. Eng. 1 (1997) 30–54.
[38] D. Teegarden, G. Lorenz, R. Neul, How to model and simulate microgyroscopic systems, Proc. IEEE Spectrum 35 (7)
(1998) 67–75.
[39] E. Tonti, The reason for analogies between physical theories, Appl. Math. Model. 1 (1976) 37–50.
[40] VHDL-AMS (VHDL-Analog and Mixed Signal Extensions): see http://www.vhdl.org/analog/.
[41] P. Voigt, G. Wachutka, Electro-fluidic microsystem modeling based on Kirchhoffian network theory, Sensor and Actuators
A 66 (1–3) (1998) 6–14.
[42] G. Wachutka, Tailored modeling: a way to the ‘virtual microtransducer fab’? Sensor and Actuators A 46–47 (1995) 603–612.
[43] P.E. Wellstead, Introduction to Physical System Modelling, Academic Press, London, 1979.
[44] O.C. Zienkiewicz, R.L. Taylor, The Finite Element Method, Vol. 2, McGraw-Hill, New York, 1989.