| The motivation for system identification can be manifold. In this work, the provocation to identify unknown system characteristics is derived from the control engineering point of view. That is, one intends to design a control strategy based on the identified system properties. The used system identification methods are the Open-Loop Kalman filter System Identification method (OKID) and the Closed-Loop System Identification method (CLID). It is shown that the quantitative largest error of the system identification is given by its model representation, that is the attempt to describe a system with model parameters which poses a linear relationship with the input/output data. Parameter identifiability is reduced to the problem of consistent estimation. The identifiability is largely determined by the way the system is excited, and in addition by the output of the system for the indirect system identification. A quantitative comparison between the indirect and direct system identification method is given, where indirect system identification showed to be slightly superior in accuracy if a suitable controller is selected. The example models used in the comparison are a heat-mass transfer model, a macro economical model, a structural model, NASA's Large-Angle Magnetic Suspension Test Facility (LAMSTF), and a human respiratory system. The problem of defining the input data such that accuracy and identifiability are increased is addressed and controller design criteria can be developed from it. The excitation input is calculated from input/output data and substituted into the current input. Simulations indicate that only a few substitutions are necessary to successfully identify the system. The new input design results in very accurate identification with reduced noise influence and data length requirement. Controller design criteria can be formed based on the input design, such that identification leads to more accurate and more reliable results. |
