Automatically Abnormal Detection for Radiator Fans Through Sound Signals Using a Deep Learning Technique

In this study, an abnormal detection model of a fan through sound is developed using a deep learning technique. The fan sound datasets include two classes, OK and NG. First, the sound signals are framed to a consistent duration; then, the log-mel spectrogram features are extracted. A deep learning model is proposed to classify fan sound signals based on the extracted features. The results show the high performance and accuracy of the proposed model and can be used to develop a computer application for the abnormal detection of radiator fans through sound signals.