LAUSR

Abstract This study is motivated by the fact that minimum variance control (MVC) benchmark can be insufficient in case of deterministic process inputs, since stochasticty is assumed. Therefore, it is crucial to know the type of disturbance while assessment. In this paper, a new approach is proposed to extend MVC via characterization of disturbance, which require only routine closed loop data and is easy-to-use online. The method can also detect oscillations properly in noisy process measurements. Suggested method uses a decision tree to combine controller performance index (CPI) with correlation coefficient which identifies the type of disturbance. The algorithm is applied into a refinery unit where the root causes of poorly performing loops are known. The results are discussed in the scope of performance monitoring.