Selection of optimal methods for intelligent process monitoring

摘要

This work proposes a statistics-based approach to the selection of the best-performing numerical methods for the detection and diagnosis of faults in the process industry. It is assumed that the performance of each method cannot be measured directly for each user-specified fault. It is shown how in those cases one can evaluate the expected performance of each method for fault detection and diagnosis by using the nonparametric statistical tests and kernel density estimation.