Generalized Semisupervised Self-Optimizing Kernel Model for Quality-Related Industrial Process Monitoring

Abstract

The process monitoring performance of traditional kernel-based modeling methods strongly depends on the appropriate selection of the kernel function, which solely aims at solving the nonlinearity and totally ignores the relationships between the process variables and the quality variables. In this article, a supervised self-optimizing kernel model is first proposed. The unique special kernel space with data-dependent kernel function is designed and learned from the modeling data, so that the authentic relationships between the process variables and quality variables can be discovered and extracted combined with dimension reduction on the process variables. To be more general than the commonly used semisupervised assumptions, the generalized semisupervised self-optimizing kernel model is successively proposed to make full use of the collected information in the modeling data with irregular measurements of quality variables. Both of the proposed models only contain one compact step of optimization with scalable parameters. The feasibility and efficiency of the proposed algorithms are demonstrated and evaluated through the industrial case studies.