Multi-mode data-based joint fault detection method

Fault detection for multi-mode process are becoming a hotspot. By effective mode division and accurate online identification, a multi-mode hybrid data set can be transformed into multiple single-mode data sets, then the traditional PCA-based approach can still be adopt to process monitoring. However, in the treatment of “fast response” multi-mode procedure, the above idea does not seem to apply, which is mainly due to: 1) the characteristics of the realtime measurements are needed to be similarity or dissimilarity measured with the known characteristics of all modes, which produces large computational cost, further usually appears online data can not be matched with the monitoring model timely, and finally leads to higher false alarm and missing detection; 2) especially when the running cycle of a single mode is short with less data, and the mode is frequency switched, the situation will be much worse. To this end, aiming at the switching frequently multi-mode, the joint multi-mode hybrid data are used to build a unified model for the case that modes unable to be identified effectively. Meanwhile, the generalized eigenvalues and generalized eigenvectors are defined, and a trade-off fault detection approach is established based on the unified model. Simulation results demonstrate the effectiveness of the new method, which can avoid the monitoring model unapplicable. Moreover, it is real-time and practical.