Bearing remaining useful life prediction under starved lubricating condition using time domain acoustic emission signal processing

In this paper, the estimation of the remaining useful life (RUL) of angular contact ball bearing using time-domain signal processing method is discussed. An experimental setup based on acoustic emission (AE) signal is used to extract and collect the desired data. The residual life test is performed on the SKF 7202 BEP angular contact ball bearing. Sixty-time domain features have been introduced and used for fault detection. Improved Distance Evaluation (IDE) method has been used for feature dimensionality reduction and the best 10 features have been selected. K-Nearest Neighbors (KNN) algorithm has been used to investigate the classification accuracy of IDE based on selected features for classifying healthy and faulty bearings. The results show that the IDE method enables natural fault detection in bearings with high precision. To validate the performance of the KNN classifier, performance indices such as accuracy, precision, and specificity are applied. The results show that kurtosis, FM4, k factor, energy, and peak are the best features and kurtosis has the highest KNN rank with accuracy, precision, and specificity of 97%, 93%, and 94%, respectively.

Automated summary

This paper discusses the estimation of the remaining useful life of angular contact ball bearings using time-domain signal processing methods, introducing 60 time-domain features for fault detection and utilizing the IDE method for feature dimensionality reduction. The KNN algorithm is used for bearing classification based on selected features, with results showing high precision in fault detection. The study validates the performance of the KNN classifier with performance indices, highlighting the importance of features such as kurtosis in achieving high accuracy, precision, and specificity in bearing classification.