Where Fault Detection and Diagnosis Meets MPC Performance Assessment: Review and Case Study of an Integrated Framework

Abstract

Abstract

Various methodologies have been developed over the years for assessing model predictive controller (MPC) performance. However, few are applied in industry, and they remain inefficient in providing a rapid indication of the causes of deteriorated control behavior. This article aims to review methodologies available in the literature addressing these challenges. Additionally, it proposes a structure in which the MPC performance assessment is carried out within a fault detection and diagnosis (FDD) framework. The integrated approach employs cascaded modules of machine learning (ML) binary classifiers arranged in a sequence that mimics the decision-making logic of an operator. To illustrate the integrated strategy both conceptually and operationally, the van de Vusse reactor, controlled by a nonlinear model predictive controller (NMPC), was adopted as a case study. The ML models used were Random Forest, Multilayer Perceptron, and Recurrent Neural Networks. The results showed that the models can correctly distinguish the cause of abnormality, even in the presence of noise in the measurements. Different ML models performed best for different diagnostic tasks, highlighting the flexibility of arranging models according to their most suitable application. The investigation indicated that the proposed ML-based FDD framework, which embeds control performance assessment, is competitive for control-aware diagnosis of MPC-controlled processes.