Improved Weighted PLS for Quality-Relevant Fault Monitoring Based on Inner Matrix Similarity

Monitoring the influence of fault towards the product quality is of great importance to modern manufacturing enterprise. Traditional projection to latent structures (PLS) method as well as its variants still face many problems. In this paper, a new improved weighted PLS (IWPLS) is proposed to utilize the local information of the process data, handle noises and build regression models with better generalization capability. The objective function of IWPLS is weighted through calculating the similarity between the target inner matrix (IM) and the other inner matrices (IMs). Two types of weight matrices are given for different process data set. The IWPLS-based monitoring scheme is developed with additional restrains and decomposition operation. A designed numerical experiments and Tennessee Eastman Process (TEP) are employed to evaluate the validity of the proposed method.