Industrial process fault detection based on locally linear embedded latent mapping

The aim of this study is fault detection in industrial processing with nonlinear or high dimensions. We attempt to integrate the LM algorithm into the local linear embedding algorithm, and propose a novel local linear embedding latent mapping (LLELM) algorithm for fault detection. The LLELM is designed to improve the performance of process monitoring. Our objective is to preserve the global characteristics of the original sample space in the projected lowdimensional space, while preserving the local neighborhood structure of the sample. This research comprises three aspects. First, LLELM overcomes the shortcomings of LM and can effffectively deal with the nonlinear problem. Second, LLELM makes up for the shortcomings of LLE, and can retain the main information from the data after dimensional reduction. Third, LLELM can improve the fault detection rate and can reduce the false alarm rate compared with LM, which is proven by simulation of the TE process.