Fault Detection Based On MPCA Methods For Batch Process

Batch processes have become more and more important in modern industrial processes. In ensuring the safety and stability of batch processes and high quality product, batch processes monitoring is of great significance.This paper mainly studies fault detection method in batch processes, on the base of Multi-way Principal Component Analysis (MPCA) method, when the MPCA method is used on on-line fault detection of batch processes, we must complete the whole batch data while the reaction is not finished. So the systematic errors are introduced. A new MPCA method based on variable-wise unfolding is studied to reduce the systematic errors. The simulation results on penicillin fermentation process show that the new MPCA method can construct more exactly statistical monitoring model and thereby effectively reduces the false alarms generated.Considering the nonlinearity existing in batch process data, a nonlinear technique is studied in this paper, which is called Multi-Kernel Principal Component Analysis (MKPCA) method. Since the fault detection sensitivity of MKPCA method is not good, two improved MKPCA method are given in this paper: Exponential Weighted Moving Average Multi-Kernel Principal Component Analysis (EWMA-MKPCA) method and Cumulative Sum Multi-Kernel Principal Component Analysis (CUSUM-MKPCA) method. EWMA-MKPCA method is able to capture the dynamic characteristics of the process data in the time series better, there is a high sensitivity when a small process diversification appears by using EWMA-MKPCA method. CUSUM-MKPCA method accumulates a greater amount of information on the process, so the fault detection of this method is very sensitive. The simulation results on penicillin fermentation process show that the proposed approaches can detect fault earlier than MKPCA method and MPCA method.