A new fault detection method for multi-mode dynamic process

Abstract

To deal with multi-mode, dynamic and stochastic characteristics in industrial process data, a new fault detection method based on double local neighborhood standardization and dynamic probabilistic principal component analysis (DLNS-DPPCA) is proposed in this paper. Firstly, a double Local neighborhood standardization method is used to transform the multi-mode data into single mode, which avoids the influence of cross-mode neighbor on mode transformation. Then, a dynamic probabilistic principal component analysis is applied to single mode process data to extract the dynamic and stochastic characteristics. In this way, multi-mode, dynamic and stochastic characteristics are considered and extracted so that the performance of fault detection is improved. Finally, the proposed DLNS-DPPCA method is applied to the TE process for fault detection. The results of simulation demonstrate the effectiveness and superiority of the proposed method.