Vibration-Based Detection of Bearing Damages in a Planetary Gearbox Using Convolutional Neural Networks

Tapered roller bearings are used partly in very rough and highly stressful environmental conditions. Therefore, the need for condition monitoring is increasing. This study intended to provide an approach a two-stage planetary gearbox based on vibration analysis. In total, data of six damage phenomena one healthy bearing collected. A convolutional neural network (CNN) trained evaluated by using balanced accuracy. Mainly, it investigated how many severities can be detected. addition, robustness model regarding unknown speeds should proven. The results show good differentiation up all presented phenomena. classifier reaches averaged accuracy 0.96. Also, samples collected at classified well speed values within known range. For phenomena, shows limits so that reliable classification only applicable with binary classifier, which differentiates between damaged. investigations therefore detection possible gear. Furthermore, transferability successfully tested implemented classifier.