(375g) Fast, Accurate Process Monitoring Based on Multi-Block Mutual Information and Nonparametric Statistical Process Control

Early fault detection is essential for smooth, safe, and efficient process operation in manufacturing plants (Arunthavanathan et al., 2020). Multivariate statistical process control (MSPC) has been widely applied for real-time process monitoring (Nawaz et al., 2021). Unfortunately, chemical process data are often nonparametric and heterogeneous, which is ignored by traditional MSPCs. To address the issue, a quantile-based nonparametric cumulative sum method is proposed and has obtained favorable results in the application of chemical processes (Ye and Liu, 2022; Jiang, 2023). However, there are two major drawbacks associated with the current state-of-the-art quantile-based MSPC algorithm. First, it monitors the entire process based on a single global statistic, in which the valuable process knowledge is often lost. Second, this method does not explicitly account for correlation between data streams being monitored, which often provides important information about the causality of faults. To address these drawbacks, we integrate quartile-based MSPC with an innovative idea of mutual information (MI). Mutual information is a measure of the interdependence between variables, which can be employed to quantify variation in variable correlation (Ji et al., 2022). The idea is that when a fault occurs, correlations between variables will also change. By monitoring the MI streams generated from original data streams using MSPC, early fault detection could be achieved.

Nevertheless, directly monitoring all MI streams for every data stream pair is computationally intensive, as modern manufacturing plants are equipped with numerous sensors and the resulting data stream combinations scales computational challenge, we propose a multi-block MI monitoring framework, in which the collected data streams are divided into several distributed blocks according to the process topology information and process prior knowledge. On this basis, the MI values between the variables within each block are calculated separately. Then, the quantile-based nonparametric cumulative sum method is applied to identify the MI shift, thereby achieving early detection of faults. A case study on the Tennessee Eastman Process is investigated to validate the proposed method.

Reference:

Arunthavanathan, R., Khan, F., Ahmed, S., Imtiaz, S., & Rusli, R. (2020). Fault detection and diagnosis in process system using artificial intelligence-based cognitive technique. Computers & Chemical Engineering, 134, 106697.

Nawaz, M., Maulud, A. S., Zabiri, H., Taqvi, S. A. A., & Idris, A. (2021). Improved process monitoring using the CUSUM and EWMA-based multiscale PCA fault detection framework. Chinese Journal of Chemical Engineering, 29, 253-265.

Ye, H., & Liu, K. (2022). A generic online nonparametric monitoring and sampling strategy for high-dimensional heterogeneous processes. IEEE Transactions on Automation Science and Engineering, 19(3), 1503-1516.

Ji, C., Ma, F., Wang, J., Sun, W., & Zhu, X. (2022). Statistical method based on dissimilarity of variable correlations for multimode chemical process monitoring with transitions. Process Safety and Environmental Protection, 162, 649-662.

Jiang, Z. (2023). Online Monitoring and Robust, Reliable Fault Detection of Chemical Process Systems. In Computer Aided Chemical Engineering (Vol. 52, pp. 1623-1628). Elsevier.