Fault detection and diagnosis based on multi-block Just-In-Time-learning slow feature analysis

Abstract

The traditional fault detection method only establishes a global model and does not consider the local information of the process. At the same time, the data in the industrial process has time-varying and non-linear characteristics, limiting the prediction accuracy of fault monitoring. Therefore, fault detection and diagnosis of the method is proposed based on multi-block just-in-time-learning slow feature analysis (JITL-MBSFA). Firstly, mutual information (MI) is segmented into two sub-blocks based on normal observation data sets. Then, through just-in-time-learning (JITL) to screen the optimal data set, and based on slow feature analysis (SFA) to build a sub-model, calculate the corresponding monitoring statistics, the support vector machine (SVDD) to monitor the results of fusion. Finally, the comparative simulation experiment in the Tennessee Eastman (TE) process verified the effectiveness and superiority of the proposed method.