Fault diagnosis of time-varying processes using modified reconstruction-based contributions

Abstract

This paper presents a modified reconstruction-based contributions for sensor fault diagnosis in continuous time-varying processes. The proposed fault diagnosis method is based on recursive updating of the loading subspaces of principal component analysis (PCA) with a low computational cost. The diagnosability of the proposed diagnosis method is proved mathematically for single sensor faults with large magnitudes. The control limits of the reconstruction contributions indices are computed and updated recursively to adapt the time-varying characteristics. Moreover, a complete adaptive algorithm for fault detection and diagnosis phases is provided for adaptive process monitoring. The efficiency of the proposed approach is demonstrated using a simulated time-varying example and a continuous stirred tank reactor (CSTR) process. The results show the ability of the proposed approach to adapt with the time-varying characteristics and still correctly diagnose the sensor faults even in the case of relatively moderate and small faults.