Fault Detection of Motor Gear Box Using Two Stage Sound Classification Network

A motor gear box is used to control tilting of the mirror and folding of the side mirror wings in a vehicle. Gear box fault detection in the production line is very important because faulty production returns considerable cost when undertaking vehicle maintenance. Fault detection based on acoustic sounds is widely used because it is simple and efficient. Sound-based method offers the advantage of non-destructive inspection, but it also provides limited classification performance. In this article, we propose a two-stage classification algorithm based on CNN. This method detects anomaly in the first stage, then subsequently classifies the faulty class. Mel-spectrogram is used to extract the features of the acoustic motor sounds. Since the proposed method classified the types of defects after determining faulty components, it is expected not only to improve accuracy, but also to reduce the time required to figure out the cause of failure.