A novel pattern matching-based fault diagnosis using canonical variate analysis for industrial process

Original data with high dimension and noise are usually directly applied to pattern matching, which will affect the accuracy of models to some extent. To solve this issue, a novel pattern matching method integrating canonical variable analysis with adaptive rank-order morphological f lter (CVA-AROMF) is proposed for fault diagnosis in this article. First, canonical variable analysis (CVA) is used to extract the features of training data with sequence correlation and process dynamics, and then the features are used as the template signal of adaptive rank-order morphological f lter (AROMF). Second, the noise-bearing test signal is used to match the template morphology waveform under the supervision of different fault template signals. Third, the fault mode is classifieds by finding the minimal distance between the filter output signal and the raw test signal of each fault mode. Simulations based on Tennessee Eastman(TE) process data is performed and the result verifies the accuracy and superiority of this proposed method.