Data-driven plant-wide control performance monitoring

Abstract

In this work a new data-driven plant-wide control performance monitoring methodology is proposed. The main constitutive parts of the suggested method are based on three well-known research areas from process systems engineering (PSE): (1) the sum of squared deviations (SSD) concepts from the plant-wide control design topic, (2) the partial least-squares (PLS) modeling technique from the multivariate statistics area, and (3) the covariance-based performance index and diagnosis (CID) from the control performance monitoring field. All these approaches are integrated and reformulated in the current work to perform a MIMO control structure performance/feasibility assessment, an open-loop steady-state model identification by using closed-loop normal data, and a covariance-based procedure for diagnosis purposes. This strategy requires minimum interference with the industrial process operation and generates valuable information (off-line as well as on-line) to evaluate the already installed control policy and suggest potential control structure modifications and/or potential controller retuning. Two typical case studies are proposed to analyze the scope of the suggested approach.