Detection Capability For Incipient Faults in a Noisy Environment Using PDF and CDF Based Techniques: a Comparative Study

Abstract

Incipient fault detection plays an essential role in the system’s health monitoring and condition maintenance. However, incipient faults will not cause obvious changes in the system parametric information, they tend to cause slight modifications in their data distributions. In high noise level environment, the slight changes caused in the fault features can be masked by the noise information. Thus, obtaining an efficient incipient fault detection in high noise level environments is more tricky. In the literature, it is proved that the traditional statistics methods such as the four statistical moments, the Hotelling’s T2 and Square Prediction Error (SPE) which focus on detecting the changes in the parameters of the data distribution can’t well address the incipient fault detection problem in a high noise level environment. These last years, the Probability Density Functions (PDF)-based distance measure algorithms and the Cumulative Density Functions (CDF)-based ones with good advantages in measuring slight differences of data distributions have shown their potential in fault detection for incipient fault. In this paper, addressing the incipient fault detection of the multivariate system in a high noise level environment, we evaluate the detection performances of several widely used PDF-based and CDF-based methods in the principal component analysis (PCA) framework. Their advantages and limitations are shown and comparatively discussed.