Distributed Dynamic Process Monitoring Based on Maximum Correlation and Maximum Difference

In order to solve the problem of dynamic characteristics caused by the autocorrelation of different measurement variables and the cross-correlation of variables reflected at different sampling times in industrial processes, a dynamic monitoring method of distributed latent variable model based on maximum correlation maximum difference (MCMD) is proposed. Firstly, delay measurements are introduced into each measurement variable to form an augmented matrix, which is used to establish sub-blocks for each measurement variable, and MCMD is used to describe the influence of delay variables on the current measurement value. This operation can not only eliminate redundant fault information, but also describe the dynamic characteristics of industrial processes more comprehensively. Secondly, dynamic principal component analysis (DPCA) is used to establish a monitoring model for each sub-block. Then, the Bayesian inference mechanism is used to convert the statistical indicators of different sub-blocks into posterior probabilities, and the fusion strategy is used to obtain a more comprehensive monitoring indicator, so that the indicator can indicate the process state more comprehensively. Finally, the effectiveness and feasibility of the proposed method are verified by the Tennessee-Eastman (TE) process.