Improved confidence limits of T2 statistic for monitoring batch processes

Abstract

Multiway principal component analysis (MPCA) is an effective method for batch processes monitoring and fault detection, but it is shown that the low sensitive of T2 statistic and the high confidence limits of T2 statistic commonly appeared in practical monitoring. In order to overcome these shortcomings, an improved method of determining the T2 confidence limits is proposed. The T2 values of normal history data are organized as a new sample dataset after building MPCA model. By applying PCA to this dataset, the confidence limits of T2 statistic will be attained. The simulation results of penicillin fermentation process platform show that the proposed method is able to detect faults more prompt and accurate than traditional method.