A new bearing fault diagnosis approach based on common spatial pattern features

Authors : Nurhan Gürsel Özmen, Yunus Emre Karabacak

Pages : 1545-1557

Doi:10.28948/ngumuh.1330864

View : 73 | Download : 117

Publication Date : 2023-10-15

Article Type : Research Paper

Abstract :Condition monitoring in machines holds significant importance for early fault detection, optimizing maintenance processes, and ensuring operational continuity. In this study, a novel intelligent detection approach for rolling bearings is introduced, utilizing the Common Spatial Pattern (CSP) method to extract distinctive features related to bearing faults. By maximizing the variance ratio of signal matrices from distinct sources, CSP sets itself apart from conventional frequency-based features. This technique captures characteristic vibration patterns unique to each measurement, enabling differentiation between faulty and healthy bearings. The effectiveness of the proposed method was assessed using Artificial Neural Network (ANN), Support Vector Machine (SVM), and K-Nearest Neighbour (k-NN) algorithms across two diverse datasets. The results indicated an 88.5% accuracy in two-class fault detection and 93.5% in fault classification when employing ANN. Comparison with traditional time domain feature sets highlighted the superior performance of CSP features, exhibiting elevated accuracy rates in both two-class and multiclass scenarios. Thus, CSP features emerge as a promising avenue for effectively monitoring bearing conditions through vibration data.
Keywords : Ortak Uzamsal Örüntü, Rulmanlar, Durum İzleme, Titreşim Sinyalleri