Process Monitoring Using Kernel PCA and Kernel Density Estimation-Based SSGLR Method for Nonlinear Fault Detection

Fault monitoring is often employed for the secure functioning of industrial systems. To assess performance and enhance product quality, statistical process control (SPC) charts such as Shewhart, CUSUM, EWMA statistics have historically been utilized. When implemented to multivariate procedures, unfortunately, univariate demonstrate low fault sensing ability. Due some limitations charts, numerous techniques dependent on approaches principal component analysis (PCA) partial least squares (PLS) designed. Yet, in challenging scenarios chemical biological processes with notably nonlinear properties, PCA works poorly, according its presumption that dataset generally be linear. However, Kernel Principal Component Analysis (KPCA) a reliable precise methodology, but interaction mainly through upper limits (UCLs) Gaussian distribution may weaken output. This article introduces time-varying error tracking based Generalized Likelihood Ratio (GLR) using sequential sampling scheme named KPCA-SSGLR detection. The main issue employing just T2 Q statistic KPCA they cannot correctly give practitioners change point system fault, preventing from diagnosing issue. Based this perspective, study attempts incorporate (SSGLR) utilized dimension reduction, while SSGLR statistic. kernel density estimation (KDE) was approximate UCLs variational operation relying KPCA. testing efficiency corresponding KPCA-KDE-SSGLR technique then analyzed competed locality preserving projection (KLPP), which were focused distribution. purpose development accomplish future enhancements advance practical use established model by implementing GLR approach. demonstrated different simulation scenarios, one utilizing synthetic data, other Tennessee Eastman technique, lastly hot strip mill. findings indicate applicability KPCA-KDE-based over KLPP KPCA-KDE methods two recognize faults.