Improving Bearing Diagnostic Performance by Using New Discriminatory Fault-Feature Evaluation

Locating different defect types in bearings using the information of the characteristic frequencies in the envelope power spectrum of analyzed acoustic emission (AE) signals has been widely utilized. However, this approach only shows effectiveness as the rotational speed of bearing elements is constant. In contrast, if the bearing speed frequently alters during operation, the value of these characteristic frequencies is not stable, therefore, it is useless for diagnostic purposes. In order to resolve this issue, this study proposes an approach that (a) adopts heterogeneous feature modes to extract as many statistical features as possible in transformed domains (i.e., the time domain, the frequency domain, and the wavelet domain); (b) explores the most discriminatory features using new feature selection scheme. The scheme is the combination of the genetic algorithm (GA)-based feature analysis and the k-nearest neighbors (k-NN); (c) the defect types of a typical bearing are categorized by the k-NN-based classifier. The performance of the proposed method is validated by two datasets of AE samples measured from our bearing testbed.