System identification is the science of mathematical model construction. Processes originated from the fields of engineering are typical examples of systems. It is often necessary to describe the system behavior in time (also known as system dynamics) for purposes, such as simulation and prediction. Therefore, models consisting of mathematical equations are built, using data from the system, which mimic the system behavior as close as possible. The quality of the model depends on factors such as the quality and amount of data, the mathematical structure used to model the system dynamics, etc. The more complex the dynamics is, the more complicated the model should be, in order to capture sufficiently the system behavior, and the more data should be recorded. Even if a complex structure is available (usually not), it is often the case that either long measurements are not available or the data size together with the large number of model parameters constitute modeling a very difficult and often impractical problem to solve. In this thesis, systems are modeled with the Volterra series, a mathematical series able to capture very complex dynamics. A method is proposed on how intuition on the underlying system can be combined together with the (not necessary long) recorded data in order to obtain a sufficiently accurate model. The proposed methodology is tested on several simulated as well as real systems in the fields of electronics, mechanical and biomechanical engineering. The results indicate that this method constitutes a very significant step towards modeling of complex dynamic systems. The importance of this thesis is also emphasized by the fact that this work has already given rise to more scientific contributions, a strong evidence of the work being accepted by the scientific community.
|Award date||22 Jan 2018|
|Publication status||Published - 2018|