Quality-Relevant Fault Monitoring Based on Locally Linear Embedding Enhanced Partial Least Squares Statistical Models

The multivariate statistical process monitoring methods represented by the partial least squares method are widely used in quality control and fault diagnosis. However, the existing partial least squares method has certain shortcomings in application. In order to enhance the nonlinear processing ability of system monitoring, a locally linear embedding enhanced partial least squares (LLEEPLS) model is proposed to enhance the local retention ability. By integrating the advantages of partial least squares and local linear embedding, the LLEEPLS model not only has the capability of PLS to extract the maximum correlation information between process variables and quality variables, but also enhance the local retention ability of PLS to keep the local structural information of the sampling data. The simulation of S-curve three-dimensional data shows that the LLEEPLS model can keep the local and global features of the original data better. And the results of TEP simulation verify the performance of LLEEPLS method for the quality-related fault diagnosis is better than that of the existing PLS model.