Theoretical and practical issues in system identification for process control applications

This dissertation addresses theoretical and practical issues related to efficient system identification. System identification experiments can be time-consuming and disruptive to normal operation of a process. Therefore, it is important to find the most expedient means of performing the identification, in order to minimize the disruption to the process. Yet in order to obtain the maximum amount of information from an experiment, certain conditions on the inputs must be met. Input sequence design and plant test procedures are crucial to minimize the impact on normal operations.; There are many reasons why a process may change over time. Unusual disturbances, instrumentation problems, contamination, or even external environmental fluctuations may cause the process to change significantly. In these cases, the process model no longer matches the actual process and thus any model-based control scheme or controller tuning may result in poor performance. Consequently, re-identification or updating the old model is an option. However, poor agreement between model predictions and the corresponding output data does not necessarily imply that re-identification is required. Relatively large residuals could be the result of unusual disturbances, instrumentation problems, etc. In order to understand the true behavior of the process, it would be helpful to know if the process dynamics have changed significantly. A new metric based on the Akaike Information Criterion (AIC) for detecting changes in the process has been developed. Rigorous statistical tests of hypothesis for model change can be posed. By combining the information from this new metric and data analysis based on Principal Component Analysis (PCA), a methodology has been developed to help decide whether a significant process change has actually occurred.; Theoretical analysis, simulation studies, and an experimental application have been completed to address these issues. The theory will be discussed, followed by description of simulated systems used in the research. Results from the simulation studies are presented. The UCSB pH neutralization process was used for experimental validation of the simulation results.