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

The multiple modes of operation of complex chemical production processes lead to problems of center drift in process data and difficulty in mining feature information, which can affect the safety of the production process. Therefore, a novel multimode process fault detection approach based on the local neighborhood-density standardization and ensemble serial global-local preserving projections (LNDS-ESGLPP) is proposed in this paper. Specifically, the set of local neighborhood-density samples of the original data is found to standardize the sample. The local neighborhood-density standardization can shift the center of different modal data to the same point and adjust the dispersion of each modal data. The kernel principal component analysis (KPCA) and the kernel locality preserving projections (KLPP) are used to build a hybrid model for extracting global and local feature information of the process data. Furthermore, the SGLPP sub-model based on different width parameters is developed. A weighted combination of Bayesian inference results from different sub-models use the ensemble learning approach for the monitoring of multimode process data. The proposed method is applied in a numerical example and a penicillin fermentation process, the experimental results verify that the proposed method has better fault detection performance.