Non-Gaussian Quality Relevant Process Monitoring Based on Higher-Order Statistics Projection to the Latent Structure and Independent Signal Correction

Abstract

A novel statistical model based on the higher-order statistics projection to the latent structure (HPLS) is proposed, which uses a combination of higher-order statistics (mutual information and differential entropy) instead of covariance to extract latent variables. This model not only has an explicit representation similar to the projection to latent structure model but can also capture the non-Gaussian process features. Furthermore, in order to reduce the redundant components contained in the latent variables independent of the quality variables, a novel strategy called independent signal correction is proposed. Finally, the novel independent signal correction HPLS (ISC-HPLS) model and the corresponding process monitoring strategy are developed. This model considers higher-order statistics that may involve certain key features of non-Gaussian processes and eliminates redundant components. Experimental results from synthetic numerical simulations and the Tennessee-Eastman process benchmark test verified the effectiveness of the proposed method.