Performance assessment of multivariable control systems using PCA control charts

In the MIMO system, the benchmark of the multivariable system in the past is computed from the traces of the multivariable covariance matrices. This may cause false detection since the correlation among variables is not considered. In this paper, a new method, called ARMA-PCA based MV, is proposed. It is an integrated framework of multivariable principal component analysis (PCA) and the autoregressive moving average (ARMA) filter for building up the minimum variance (MV) performance bound of the MIMO feedback control system. The benchmark of the multivariable system is achievable by the minimum score of the variances using the resulting decomposition structure of the PCA model without the risk of the overdetermined or underdetermined problems in the MIMO system. After extracting the essential features of the benchmark information, subsequently, improved statistics is selected as the performance benchmark index to monitor the progress of the current operation. The case study indicates that the proposed method yields better monitoring results than the conventional approaches in terms of sensitivity and time to the degraded system.