Multi-directional reconstruction based contributions for root-cause diagnosis of dynamic processes

Abstract

Dynamic principal component analysis (DPCA) is required for the modeling and monitoring of dynamic processes. However, the root cause identification of faulty variables is quite desired after a fault is detected. As DPCA based methods construct detection indices in augmented variable space, it is difficult to use contribution analysis for diagnosis in a common way. In recent literature, reconstruction based contribution (RBC) is proposed, which is more efficient to diagnose sensors responsible for a fault than traditional contribution analysis. However, they both suffer from smearing effect. In this paper, an extended method of RBC based on DPCA is proposed to select multiple faulty variables in the sense of reconstruction, which is called multi-directional RBC. The case study on continuous stirred tank reactor (CSTR) process is used to demonstrate the effectiveness of the proposed approach.