Analysis of Window Size and Statistical Features for SVM-based Fault Diagnosis in Bearings

Bearings are mechanical components used in many rotating devices. They exhibit high failure rates which cause significant maintenance downtime. For this reason, there has been an increase in the efforts for designing techniques that allow early failure detection. Fault diagnostics systems based on machine learning are becoming increasingly prominent in this scenario. These techniques have three fundamental steps: signal acquisition, feature extraction and fault classification. The present work aims to provide a detailed analysis at the second and third steps, in the context of producing efficient statistical attributes for bearing fault recognition using Support Vector Machine (SVM) classifiers. More specifically, the classifier’s performance is studied considering different statistical features, class formulation and different sizes of the window over which features are computed.