Linear Subspace Principal Component Regression Model for Quality Estimation of Nonlinear and Multimode Industrial Processes

Abstract

Principal component regression (PCR) has been widely used for quality estimation in industrial processes. However, the traditional PCR method is restricted in linear processes. Although several nonlinear forms of PCR have been proposed, most of them have high algorithm complexities, which make them difficult to use in practice. In this paper, a new linear subspace PCR model is proposed for quality estimation of nonlinear processes. Through monitoring key statistics and Bayesian inference, the quality estimation results in different linear subspaces can be effectively integrated. By introducing an additional information combination direction, the basic linear subspace PCR model is further extended to a two-dimensional form for quality estimation of multimode processes. Two industrial examples are provided for performance evaluation of the proposed methods.