Welcome to the Modeling & Simulation (M&S) Research Group

The Modeling & Simulation Research Area forms part of the ICOS unit of the Computer Science Department at the ETH in Zurich.

Research Areas

Object Oriented Modeling of Physical Systems Using Bond Graphs

Bond graphs describe graphically the power flows through a physical system. Energy conservation and power flows are fundamental to all physical systems. Bond graphs therefore are well suited for describing dynamical models of physical systems that extend into multiple energy domains.

Bond graphs are object oriented, i.e., the model topology is preserved under coupling. Bond graph models can also be structured hierarchically. Consequently, bond graphs can be used to describe large scale systems.

Our research group has recently developed three different bond graph libraries for use in Modelica, a general-purpose object oriented physical systems modeling and simulation language. All three libraries are in the public domain. Two of them received first prices for free libraries by the Modelica Association in 2005 and 2006.

Our primary application areas include electronics, 3D mechanical systems, and thermodynamics.

Modeling of Ill-defined Systems Through Fuzzy Inductive Reasoning

The fuzzy inductive reasoning (FIR) methodology offers an alternative to artificial neural networks (ANNs) for inductively modeling unknown systems using observations of their input/output behavior.

FIR models belong to the class of non-parametric qualitative models. They are qualitative as they operate on coarsely discretized (quantified) versions of observed signals. They are non-parametric, as they refer back to the training data sets throughout the testing period. FIR models are synthesized rather than trained, i.e., setting up a FIR model consumes considerably less time than setting up an ANN model.

Our research has developed two different MATLAB toolboxes implementing the FIR methodology. These toolboxes are in the public domain.

Our primary application areas include biomedical systems, water distribution networks, and macroeconomic models.

Mixed Symbolic and Numerical Solution of Differential and Algebraic Equation Systems

Object oriented models of physical systems lead to large sets of implicitly formulated differential and algebraic equation (DAE) systems. Efficient solvers for large higher-index DAE systems are not available. Hence these equation systems need to be preconditioned symbolically to make them amenable to a numerical simulation.

In the process of converting the DAE system to a form that can be simulated, there are algorithms needed for symbolic index reduction, for symbolically converting implicitly formulated DAEs to explicit ordinary differential equations (ODEs), for determining an optimal set of state variables, and for determining a minimal set of iteration variables.

Our research group is involved in the development of symbolic algorithms for the preconditioning of DAE models, as well as in the development of numeric algorithms for the simulation of the resulting ODE models.

Many of these algorithms have meanwhile been implemented in the Dymola, the most advanced physical systems modeling and simulation environment currently on the market. Dymola models are internally coded in Modelica.