Canonical correlation analysis–based fault diagnosis method for dynamic processes

Abstract This chapter focuses on the application of the canonical correlation analysis (CCA) technique in dynamic process fault diagnosis. CCA is a typical multivariate analysis tool that is now widely used in process monitoring and fault diagnosis. Conventional CCA assumes that variables are independent; however, in practice, autocorrelation presents in data due to the dynamics of process. When data contains dynamic information, applying CCA on data will not reveal the exact relations between the measurement vectors but rather a linear static approximation. Hence, the fault diagnosis performance of the conventional CCA method will degrade. To this end, we present two variants of the CCA-based method—the dynamical CCA method and the gated recurrent unit–aided CCA method—to deal with the fault diagnosis of the dynamic process. The effectiveness of these methods are favorable. Validation on two industrial benchmarks shows the superior detecting effects of the presented methods.