Statistical Process Monitoring Based on the Kernel Mean Discrepancy Test

A fault detection method based on the kernel mean discrepancy test is proposed that deals with the fault detection problem from the aspect of nonparametric statistical test. The basic idea of kernel mean discrepancy is to project the training data and test data into the reproducing kernel Hilbert space, in which the mean discrepancy test is performed. Based on the kernel mean discrepancy test, a quantitative analysis of the sensitivity of the test statistic is presented, and a fault detection strategy is developed. The confidence limit is obtained by a moving-window approach, and the process faults are then detected sequentially. Simulation and application results show that the method is sensitive to process faults.