Novel Cross-validity Criteria and Statistical Index in Non-Gaussian Space

Abstract

Quality-relevant fault detection is crucial to ensure the safe and stable operation of industrial processes, but the existing cross-validation methods based on the sum of squares have shortcomings and may not be able to determine the appropriate number of features in the non-Gaussian case, so this study proposes a novel cross-validation method incorporating entropy, which uses entropy instead of the sum of squares to deal with errors and prediction errors. In addition, the conventional statistics $T^{2}$ and Q are low-order statistics that can hardly summarize the non-Gaussian information in the data appropriately, so this paper proposes an entropy-based statistic for process monitoring, using the higher-order statistic (entropy) to analyze the non-Gaussian information. Experimental results from synthetic numerical simulations and the Tennessee-Eastman process benchmark test verified the effectiveness of the proposed methods.