Canonical residual based incipient fault detection method for industrial process

Abstract

Incipient faults with low amplitudes can be easily submerged by noise and process disturbance. As these faults can slowly evolve into the serious failure of processes, early detection is increasingly becoming important. To monitor incipient faults more efficiently, canonical variate analysis with two statistics indices based on canonical residuals are proposed for monitoring dynamic processes. First, canonical variate can be obtained by performing singular value decomposition on the Hankel matrix, which assembles with covariance and cross-covariance matrices of the past and future observations. Next, the canonical residual can be generated by calculating the difference between projected past data and projected future data. Then, two weighted average statistics based on canonical residual and its control limit can be determined. Finally, with the application in the continuous stirred tank reactor (CSTR) process, the performance of the proposed detection method is verified by comparing with fault detection rates. Simulation results indicate those two weighted average statistics based on canonical residual can improve the fault detection rates compared with the conventional statistics T2 and Q.