Distributed dynamic process monitoring based on dynamic slow feature analysis with minimal redundancy maximal relevance

Since modern industrial processes contain a lot of variables and the relationships and dynamic characters among them are also complex. Hence, it is difficult to implement process monitoring by conventional methods. Aiming at the problem, a distributed dynamic slow feature analysis method with minimal redundancy maximal relevance (mRMR-DDSFA) is proposed. Firstly, the minimal redundancy maximal relevance (mRMR) is utilized to divide the most related variables into same block, which not only reduce the redundancy, but also retain the maximal correlation among variables. Then, a monitoring model based on dynamic slow feature analysis (DSFA) is established in each sub-block. In addition, Bayesian inference is carried out to integrate the detection results in each sub-block to achieve comprehensive statistical indicators, after that, the fault detection is realized for the plant-wide process. Finally, the effectiveness and superiority of the new method are verified by the simulations on the real-world diesel working process, the Tennessee Eastman (TE) platform and the continuous stirred tank reactor (CSTR) process. •The mRMR can reduce the redundancy between the selected variables, which is more conducive to the establishment of reasonable sub-blocks.•The DSFA model established in each sub-block is more consistent with the truth of real industrial processes. And has rarely been studied in large-scale industrial process.•The global statistic is obtained by using Bayesian inference. And the advantageously method framework provides a possible solution for fault detection in large-scale plant-wide industrial process.