Port-based System Identification

In this thesis, an attempt is made to synthesize a mapping from the observed behaviors of a physical system to its geometric and material parameters. Two different approaches are adopted: one is the standard "black-box" approach; the other is a physical-structure based approach. A DC torque motor is used as an example system. In this thesis, it is desired to identify four unknown motor parameters: rotor inertia, friction coefficient, torque sensitivity, and rotor winding resistance. The two identification methods are first used on a set of simulated noise-free data and later verified with an experimental data set. It is found that the conventional linear black-box identification method is inadequate to synthesize a sought-for mapping. Though, generally the approach is powerful in capturing the observed behavior, it does not provide a guide to the choice of the observed input and output which would possibly lead to a unique identification of the physical parameters. So the result from this approach is a non-unique mapping from observed behaviors to motor parameters. Furthermore, the approach could lead to a model order higher than necessary to identify the physical parameters. The excess order in the form of a free integrator could jeopardize the derived model by making it unstable. The physical-structure based approach, on the other hand, provides an appropriate model structure for a particular pair of observe