Fault detection for multimode process based on local neighborhood-density standardization and ensemble serial global-local preserving projections processes

Tao Li, Yongming Han, Xiaoyan Duan, Bo Ma, Zhiqiang Geng

Abstract: Highlights•The LNDS method is proposed to eliminate the multimode character of the process data.•A serial hybrid model of the KPCA and the KLPP is designed.•Multiple SGLPP submodels are constructed by choosing Gaussian kernel functions with different width parameters.•A weighted fusion strategy based on Bayesian inference is designed.•The proposed method has better fault detection performance in a numerical example and a penicillin fermentation process.