In this article, a novel data-driven fault diagnosis method by combining deep canonical variate analysis and Fisher discriminant analysis (DCVA-FDA) is proposed for complex industrial processes. Inspired by the recently developed deep canonical correlation analysis, a new nonlinear canonical variate analysis (CVA) called DCVA is first developed by incorporating deep neural networks into CVA. Based on DCVA, a residual generator is designed for the fault diagnosis process. FDA is applied in the feature space spanned by residual vectors. Then, a Bayesian inference classifier is performed in the reduced dimensional space of FDA to label the class of process data. A continuous stirred-tank reactor and an industrial benchmark of the Tennessee Eastman process are carried out to test the performance of DCVA-FDA fault diagnosis. The experimental results demonstrate that the proposed DCVA-FDA fault diagnosis is able to significantly improve the fault diagnosis performance when compared to other methods also examined in this article.

Principal component analysis (PCA) and its modified methods have been widely applied in industrial process monitoring. In practice, industrial processes are with disparate characteristics, the process monitoring system should consider as many process characteristics as possible, such as dynamic and nonlinear characteristics. In this paper, a multi-feature extraction technique based on PCA is proposed for nonlinear dynamic process monitoring. The proposed method integrates dynamic inner PCA (DiPCA), PCA and kernel PCA (KPCA) methods through a serial structure to extract the dynamic, linear and nonlinear features among the process data. Along with the proposed method, the original data space is decomposed into several orthogonal subspaces, in which abnormal variations of different features can be monitored. For real-time process monitoring, a combined Hotelling's T² statistic based on the extracted multi-feature and a squared prediction error (SPE or Q) statistic are established. Case studies on a numerical example and the Tennessee Eastman process are carried out to demonstrate the superior process monitoring performance of the proposed method compared with other relevant methods.