Multi-block kernel probabilistic principal component analysis approach and its application for fault detection

In this paper, a decentralized fault diagnosis approach of complex processes is proposed based on multi-block kernel probabilistic principal component analysis (MBKPPCA). Under the probabilistic modeling framework, this paper introduced MBKPPCA into process monitoring and gave a qualitative analysis on the problems of determining the parameters in MBKPPCA. Efficient Expectation-Maximization algorithms were developed for parameter learning in models analysis and algorithm is proposed and applied to monitor large-scale processes. By mapping nonlinear data into high-dimensional space by kernel function, the method eliminated process nonlinear features. Electro-fused magnesia furnace study was provided to evaluate the modeling and performances of the new method.