Interoperability in modeling and simulation

Sammanfattning: The design of complex engineering systems involves modeling and simulation as means for early estimation of product properties. The models considered in the thesis are mathematical and are computed numerically in the simulation activity. For systems spanning several engineering domains such as mechanics, hydraulics and electronics, modeling and simulation are collaborative processes where engineers and analysts from the different domains take part, making use of a variety of modeling formalisms and software. In the thesis, tools are presented that improve interoperability between models and simulation environments as well as among the environments. This facilitates straightforward collaborative simulation-based design in which appropriate modeling formalisms and software can be applied in a given situation.To fully exploit the advantages of simulation-based design, the models have to be both accurate and easy to create. For these reasons, the system is often decomposed and modeled in several environments specialized in each domain in terms of user interface etc. In order to simulate the system as a whole, the models are made coupled, either directly by merging them into one system model, or by connecting the simulation environments in which they reside. Due to a number of circumstances, which are explored in the thesis, often the only possible way of performing system simulation is to engage the solvers part of each simulation environment in coupled simulation, so called co-simulation. Unfortunately, due to the problem de-coupling, the error in the simulation results might grow unboundedly through time. Today, methods exist for making the error remain constant or decay, but few simulation environments contain the functionality demanded by these methods, such as redoing former computations. In the thesis, a few methods are presented and evaluated analytically and in practice, which can be applied to most co-simulations in order to have reliable results.One of the practicalities that necessitate co-simulation is the strong dependence between models and the environments in which they are created. If one and the same model can be simulated in any environment, it has a longer life and a larger application area than implementation-dependent models. This is of importance since the modeling work needs to be brought down to a minimum; redundant work is not an option in industry. In the thesis it is shown how equation-based models can be made to fit into most simulation environments and other software. A minimum of information is used about the model and the environment, in which the model is to function properly. A framework for open-ended model compilers is also presented that enables a model compiler to adapt to a destination environment given the minimum information mentioned. For evaluation purposes, a model compiler that can adapt models to fit within two simulation environments with fundamental differences in terms of solvers etc is implemented for a subset of the Modelica equation-based language.