Quality-related Process Monitoring Using Modified Dynamic PLS

In most industrial processes, the partial least squares (PLS) approach has been widely applied to improve product quality and product yield. However, traditional PLS can only extract cross-correlation in data. For process monitoring, both autocorrelation and cross-correlation need to be extracted effectively. In the paper, a novel auto-regressive with exogenous inputs (ARX) dynamic PLS method is proposed to capture the dynamic features in data. The key innovations of the proposed method can be summarized as: 1) a clear dynamic inner model is given for dynamic data modeling, which improves the accuracy of the quality prediction; 2) The proposed algorithm establishes a consistent dynamic model for the inner and outer model; 3) The proposed DPLS algorithm induces an orthogonal projection structure in the input space, which provides strong support for fault detection. The application in fused magnesium production demonstrates the feasibility and effectiveness of the proposed algorithm.