음성인식 및 머신러닝 기술을 적용한 모터 소음 진단법에 대한 연구

Abstract

Motor noise is one of the important quality factors for motor performance. Generally, in a motor production line, motor noise is inspected by a skilled worker. Because motor noise can be caused by a variety of sources or combinations thereof, it is difficult to isolate each specific noise source. An objective/automatic noise-source detection method would be helpful for motor manufacturers. This study introduces a noise-diagnosis method using a sound recognition technique and machine learning methods. First, the raw noise data are filtered through a spectral noise-gate algorithm to reduce the background noise. Then, mel-frequency cepstral coefficient features, which are widely used in the speech-recognition technique, are extracted from the noise data. Finally, a noise-classification model is developed using a support vector machine technique for motor data whose noise sources are known. This model can be used to identify malfunctional motor noises and their sources. Finally, this method was validated by comparing its results with those of frequency analysis from the motor production line.