Methodology for the on-line design of identification experiments

grantor: University of TorontoIn this thesis, system identification using the Frequency Sampling Filter (FSF) model is evaluated. A new input signal design method is developed specifically for application in industrial environments. Employing both the FSF model and the newly developed input signal design method, a methodology for the on-line design of identification experiments is proposed. The new input signal design method produces a multi-frequency, non-periodic signal with energy concentrated at prespecified frequencies. This method is easy to apply and, in comparison with other popular design methods, is shown to produce more accurate step response model estimates. Four parameters in the identification experiment are shown to act as indicators of data quality and model accuracy. Using these indicators as a basis, a set of guidelines is developed. The guidelines specify changes in the experimental conditions which are necessary to ensure that an accurate process model estimate is obtained. These guidelines are expanded into a methodology for on-line use during the identification experiment.M.A.Sc