Machine Vision–Based Measurement Approach for Engine Accessory Belt Transverse Vibration Based on Deep Learning Method

In this paper, to address the problem of using displacement sensors in measuring the transverse vibration of engine accessory belt, a novel non-contact method based on machine vision and Mask-RCNN model is proposed. Mask-RCNN model was trained using the videos captured by a high speed camera. The results showed that RCNN model had an accuracy of 93% in detection of the accessory belt during the test. Afterward, the belt curve was obtained by a polynomial regression to obtain its performance parameters. The results showed that normal vibration of the center of the belt was in the range of 2 to 3 mm, but the maximum vibration was 8.7 mm and happened in the engine speed of 4200 rpm. Also, vibration frequency of the belt was obtained 124 Hz. Moreover, the minimum belt oscillation occurred at the beginning point of the belt on the TVD pulley, whereas the maximum oscillation occurred at a point close to the center of the belt at a distance of 16 mm from it. The results show that the proposed method can effectively be used for determination of the transvers vibration of the engine accessory belts, because despite the precise measurement of the belt vibration at any point, can provide the instantaneous position curve of all belt points and the equation of the belt curve at any moment. Useful information such as the belt point having the maximum vibration, belt slope, vibration frequency and scatter band of the belt vibration can be obtained as well.