Fault detection and isolation for dynamic non-stationary processes with stationary subspace-based canonical variate analysis

Abstract

As modern science and technology advance, industrial processes have grown more complex, characterized by dynamic and non-stationary features. Existing methods often focus on single features, necessitating the development of approaches capable of addressing multiple characteristics. This study introduces a novel approach based on stationary subspace canonical variate analysis for fault detection and isolation in dynamic non-stationary processes. The proposed model combines the strengths of stationary subspace analysis and canonical variate analysis (CVA) by introducing new detection indices and their corresponding contributions. These new indices, derived from CVA indices, are further transformed into quadratic form to facilitate easy calculation of contributions, which are based on reconstruction. A numerical example and simulation of the continuous stirred tank reactor process are presented to demonstrate the superior sensitivity and accuracy of the proposed approach.