Abstract: In order to avoid the difficulty of installing vibration sensors and extracting characteristic frequency vectors for the traditional vibration-based abrasion fault diagnosis on the main bearing of diesel engine, this paper presents a new approach based on wavelet packet images processing of sound signal of diesel engine. Thus, the standard time-frequency distribution images of all fault conditions including the gap abrasion information of the main bearing can be defined. Correspondingly, a gap abrasion fault diagnosis model of the main bearing with images matching is set up. Through comparing the Euclid distance between standard fault image and the test image, the model can recognize the gap abrasion condition. The results show that this method makes the best use of fault information and is simple and effective.