A Quality-Related Fault Detection Approach Based on Dynamic Least Squares for Process Monitoring

The issue of quality-related fault detection has attracted much attention in recent years. Partial least squares (PLS) is considered as an efficient tool for predicting and monitoring. However, due to the fact that PLS performs an oblique projection to input space, it is not suitable for quality-related fault detection. On the other hand, PLS is a static method which cannot be used in dynamic systems. In this paper, a dynamic least squares approach is developed by using the structure of auto-regressive moving average exogenous (ARMAX) time-series model. Furthermore, augmented input matrix is decomposed into two orthogonal parts according to their correlations with output, such that quality-related fault detection can be utilized by designing appropriate statistics in two subspaces corresponding to the two parts. The proposed approached is simple and effective for systems with dynamic input and static output, which is a common case for most industrial processes. A numerical example and an industrial process simulator are applied to test the performance of the proposed approach.