Non-Stationary Vibratory Signatures Bearing Fault Detection Using Alternative Novel Kurtosis-based Statistical Analysis

Abstract

Vibration signature-based analysis to detect and diagnose is the commonly used technique in the monitoring of rotating machinery. Reliable features will determine the efficacy of diagnosis and prognosis results in the field of machine condition monitoring. This study intends to produce a reliable set of signal features through an alternative statistical characteristic before available relevant prediction methods. Given the above advantage of Kurtosis, a newly formed feature extraction analysis is adapted to extract a single coefficient out of EMD-based pre-processing vibration signal data for bearing fault detection monitoring. Each set of IMFs data is analyzed using the Z-rotation method to extract the data coefficient. Afterwards, the Z-rot coefficients, RZ are presented on the base of the specification of the defect vibratory signal to observe which IMF data set has the highest correlation over the specification given. Throughout the analysis studies, the RZ shows some significant non-linearity in the measured impact. For that reason, the Z-rotation method has effectively determined the strong correlation that existed in some of the IMFs components of the bearing fault. It corresponds to the first IMF for the inner race and the rolling ball specified a strong RZ coefficient with the highest correlation coefficient of R2 = 0.9653 (1750 rpm) and R2 = 0.9518 (1772 rpm), respectively. Whereas, the 4th IMF decomposition for the outer race bearing fault scored is R2 = 0.8865 (1772 rpm). Meanwhile, the average R-squared score in the correlation between RZ coefficient and bearing fault throughout the study is R2 = 0.8915. Thus, it can be utilized to be the alternative feature extraction findings for monitoring bearing conditions.