Fault Detection and Root Cause Analysis of a Batch Process via Novel Nonlinear Dissimilarity and Comparative Granger Causality Analysis

Data-driven fault detection and root cause analysis methods become attractive in modern industrial production that can guarantee the safety and stability of process operation. If process monitoring technology is implemented for fault detection, and the root cause of faults is analyzed timely, it is beneficial to maintain and improve the quality of coming batches. In this paper, a framework of fault detection and root cause analysis is proposed to address the aforementioned issue, particularly for a batch process. First, a new algorithm, termed kernel entropy component analysis (KECA)-DISSIM that combines KECA and dissimilarity analysis (DISSIM), is proposed for the batch process monitoring purpose. The KECA can extract nonlinear characteristics of the batch process effectively based on nonlinear mapping with the Renyi quadratic entropy. Then, dissimilarity indices between normal reference datasets and testing datasets can be calculated. If the testing dataset is detected as the non-normal batch by KECA-DISSIM, a novel root cause analysis named comparative Granger causality analysis is introduced for root cause analysis. The testing dataset is decomposed into a series of data slices via the moving window along the time domain. A series of causality values for each pair of variables are obtained by performing Granger causality analysis on these time slices. Lastly, the case studies based on a typical seven-variable nonlinear numerical process and a benchmark fed-batch penicillin fermentation process are studied to illustrate the practicality and effectiveness of the proposed framework.