A Novel Statistical Method for Spectral Analysis of A Short-Duration Signal and Its Application to Current Data for Stator Fault Diagnosis

In this paper, a novel approach for fault detection in the stator windings of induction motors is presented. The procedure is based on spectral analysis of the current signal. However, due to the specific target application, short duration signals (0.2 s) are utilized, which results in poor spectral resolution. To address this issue, a statistical methodology is developed to minimize uncertainty in decision-making. To construct a health indicator (HI), a statistical analysis is performed to identify spectral components that are both informative and robust. For the selected fault-related frequencies, the HI was created. Using confidence intervals and statistical testing, a fault detection scheme was proposed. The method was validated on an experimental dataset, including both healthy and faulty conditions. The method has been tested on current signals with five levels of fault severity and seven load conditions. Experimental studies on a dedicated test rig demonstrated the high efficiency of the proposed approach for such specific constraints.