Incipient Fault Detection Based on Kolmogorov-Smironv Test

Abstract

Most of the traditional data-driven fault detection methods depend on the change of mean value and the shift of variables. However, incipient faults are characterized by low amplitude and masked by unknown disturbances and noise. Incipient faults are not detected effectively by the traditional data-driven methods. If incipient faults are ignored, it will cause potential problems to the operation of industrial processes. It is of great importance to study the detection of incipient faults. In this study, a Kolmogorov-Smironv (K-S) test based incipient fault detection method is proposed. First, the sample data is divided into training data and test data; second, a principal component analysis (PCA) model is established with the training data to obtain the loading matrix; third, the principal component subspace (PCS) of the training data and the PCS of the test data are calculated with the load matrix; fourth, the empirical distribution function (EDF) of the principal components of the training data and the test data are calculated in each sliding window; finally, K-S statistics are calculated to determine whether a fault occurs. Moreover, the effectiveness of the proposed solution is verified in the continuous stirred tank heater (CSTH) process simulation models.