Multimode Process Monitoring Based on Local Neighborhood Standardization and Bayesian Inference

Practical industrial processes are often characterized by different operation modes, in which the sampling data no longer follow the same distribution. Multi-distribution characteristics and distribution uncertainty of the sampled data have made traditional fault diagnosis methods difficult. In this paper, a novel process monitoring method, based on local neighborhood standardization and Bayesian inference, is proposed for a multimode process. First, the data is preprocessed through a local neighborhood standardization algorithm. Second, an independent component analysis and principal component analysis (ICA-PCA) method is used to extract the Gaussian and non-Gaussian characteristics of the process dataset. A monitoring statistic can then be obtained to realize online monitoring, by combining multiple statistics with the Bayesian inference approach. Finally, the reliability and effectiveness of the proposed method are verified through simulation results of a numerical example and the Tennessee Eastman (TE) process.