Quality-Related Fault Detection and Diagnosis Based on Total Principal Component Regression Model

This paper investigates the issue of quality-related fault detection and diagnosis. A total principal component regression (TPCR) model is build, based on which process variables space is divided into two orthogonal subspaces. Subsequently, two statistical indices with different correlations with output space are designed in each subspace, respectively. An appropriate decision logic is used to determine whether a fault is quality-related or not. Once a fault is detected, it is necessary to explore the cause of the failure. Due to traditional contribution plots often provide inaccurate diagnostic result, this paper introduces an improved method without smearing effect, which is integrated into TPCR model for accurate fault diagnosis. Simulation results demonstrate the effectiveness of the proposed method.