Rolling Element Bearings Multiple Defect Detection and Classification via Vibration Signal Analysis

Ball bearing is one of the most widely used components in rotary machines. Condition monitoring of such elements is counted as pattern recognition problem. Pattern recognition has three main steps: feature extraction, feature reduction and classification. We use features obtained from three different representations of measured signals which are time, frequency, and time-frequency domains. In this study smoothed pseudo wigner ville distribution is used for feature calculation in time-frequency domain. All of the features are extracted from vibration signals. The signals from a piezo-electric transducer are captured for the following conditions: healthy bearing and defective bearings with inner race, outer race and ball faults. In addition, experiments are repeated under various load conditions. After calculation of features, principal component analysis is employed for redundancy reduction. Finally K-NN classifier is built and tested in order to identify the condition of the ball bearing. Experimental results demonstrate that the proposed method is effective.

Keywords:

Ball bearing fault detection , Time- Frequency analysis , Principal component analysis , K-NN classifier.