Competitive Adaptive Reweighted Sampling Method for Fault Detection

According to Darwin's theory of evolution, a variable combination with the strongest correlation between selection process variables and quality indicators is developed. This is called competitive adaptive reweighted sampling (CARS). In this article, the absolute value of the partial least squares regression coefficient is used to assess the importance of each variable. Next, we select variables based on the regression coefficients of variables and quality indicators, including forced variable selection based on exponential decreasing function (EDF) and competitive variable selection based on adaptive reweighted sampling (ARS). Finally, we use cross-validation (CV) to select the subset with the lowest root mean square error of CV (RMSECV). The Tennessee-Eastman (TE) process is used to evaluate the performance of the proposed fault detection method. The results show that CARS can find the best combination of certain key variables, improving the fault detection capability of quality indicators.