Assessment of multivariable controller performance

The need to achieve reliable and effective automatic control systems has directed numerous research efforts towards the development of assessing and monitoring methods of closed-loop performance. The primary contributions of this thesis are to (a) develop new informative and judicious assessment indices for the performance of univariate and multivariate controllers, which will offer additional and more reliable information than the existing indices; and (b) employ an accurate closed-loop identification methodology, which would give accurate system models, which are necessary for the estimation of the proposed assessment indices.; The Relative Variance Index or RVI is the basis of the overall controller performance assessment methodology. The main premise of the RVI is the comparison of closed-loop variance with the best theoretical control action (minimum variance control) and no control. The analysis which is presented in this thesis shows that the RVI is a judicious and easy-to-interpret controller assessment index, which is consistent with the classical assessment methods based on time response analysis. Moreover, it offers a comparison of overall controller performance on each output to detect which outputs are not controlled adequately and need immediate attention.; The estimation methodology of the proposed controller performance index from closed-loop data is the other key contribution highlighted throughout this thesis. The system models are identified via subspace identification algorithms which are particularly efficient for multivariable systems. Particularly, the accuracy of the plant estimation is significantly enhanced by the use of process unit delays for the input-output data shifting, which enable the estimation of the less biased delay-free plant. Subsequently, the estimation of the interactor matrix, the most important parameter of designing the Minimum Variance Controller (upper performance benchmark), is accurate as calculated from the estimated process. The application to several univariate and multivariable systems shows the increased accuracy and effectiveness of the overall controller assessment methodology.; The RVI controller performance assessment methodology extends to other applications. It enables the evaluation of the effect of improving the current control system by either retuning or redesigning it, based on the estimated process model. It can be also utilized for the evaluation of the robustness of the controller, based on expected plant mismatch. The RVI indices can further monitor the controller performance and diagnose the potential sources of controller performance degradation.