A Quality-Related Statistical Process Monitoring Method Based on Global plus Local Projection to Latent Structures

Abstract

The partial least-squares (PLS) method is widely used in the quality monitoring of process control systems, but it has poor monitoring capability in some locally strong nonlinear systems. To enhance the monitoring ability of such nonlinear systems, a novel statistical model based on global plus local projection to latent structures (GPLPLS) is proposed. First, the characteristics and nature of global and local partial least-squares (QGLPLS) are carefully analyzed, where its principal components preserve the local structural information in their respective data sheets as much as possible but not the correlation. The GPLPLS model, however, pays more attention to the correlation of extracted principal components. GPLPLS has the ability to extract the maximum linear correlation information; at the same time, the local nonlinear structural correlation information between the process and quality variables is extracted as much as possible. Then, the corresponding quality-relevance monitoring strategy is established. Finally, the validity and effectiveness of the GPLPLS-based statistical model are illustrated through the Tennessee Eastman process simulation platform. The experimental results demonstrate that the proposed model can maintain the local properties of the original data as much as possible and yield monitoring results that are better than those of PLS and QGLPLS.

Locally Linear Embedding Orthogonal Projection to Latent Structure

“…9) integrates the advantages of PLS and LLE methods. The distinctive feature of the GPLPLS model is that the local nonlinear features are enhanced by LLE in the PLS decomposition (Zhou et al 2018). GPLPLS uses the strategy of plus but not embedding, in which the new feature space is divided into linear part (global projection) and nonlinear part (local preserving).…”

New Robust Projection to Latent Structure

“…Experiments showed that they are efficient and robust for the data with inherent uncertainty and outliers. However, the two improved RPCA methods do not obtain any useful information from the output quality variables, so it is difficult to directly apply them to quality-relevant process monitoring and fault diagnosis (Zhou et al 2018). The monitoring system will automatically alarm if a fault is detected whether it affects the product quality or not.…”

“…To improve the efficiency of the LAD algorithm, the idea of partial least squares (PLS) regression is used to extend the conventional LAD regression to partial LAD regression. The PLS-based monitoring method decomposes the process space through the correlation between the quality and the process variables, which can reflect the quality-relevant product changes in the process variables (Wang et al 2017;Zhou et al 2018).…”

A novel dynamic nonlinear partial least squares based on the cascade structure

“…For such applications, the data-driven process monitoring (DPM) methods are preferred due to their data-based characteristics, which do not require the accurate models of complex systems. [5][6][7] The DPM methods mainly implement the fault monitoring, the fault identification and diagnosis, the fault reconstruction and the product quality monitoring and control in the production process by making use of the multiple statistics process control (MSPC) and the machine learning methods, 8,9 which has been successfully applied to many industrial fields such as chemical industry, steel, electric power, and so on. [10][11][12] The partial least squares (PLS) is a significant MSPC method, which can be traced back to the nonlinear iterative partial least squares (NIPALS) path model method proposed by Wold and Lyttkens 13 to solve the problems of the model parameter estimation in econometrics and behavioral sciences.…”