An Improved Principal Component Regression for Quality-Related Process Monitoring of Industrial Control Systems

Multivariate statistical methods are effective data-driven approaches for complex practical systems. Traditional partial least squares (PLS) serves as a latent projection approach applied to the qualityrelated process monitoring field widely. However, PLS is not suitable for quality-related fault detection which performs an oblique projection to the X variables. In order to address this problem, an improved principal component regression (IPCR) is proposed in this paper. IPCR separates the process measurements into a quality-related part and a quality-unrelated part. Compared with the conventional method, IPCR can represent the relationship between the fault and product quality more clearly. Furthermore, we design the corresponding test statistics to build the logic of fault detection. A numerical experiment and the Tennessee Eastman process simulator are utilized to illustrate the performance of the proposed approach.