합성곱 신경망을 이용한 상용차 이상소음 정성 평가 등급 예측

Abstract

This study aims to predict the qualitative evaluation result of abnormal noise generated in commercial vehicles by applying the deep learning method. First, two experts conducted a qualitative test to evaluate the abnormal noise in a commercial vehicle. The correlation between this qualitative evaluation and measured sound pressure was investigated according to the road, driving, and noise source conditions. A quantitative evaluation was then performed by dividing the range of the measured maximum values into the same 7 grades, and the correlation with the qualitative evaluation result was confirmed. Finally, the colormap images obtained from the time-frequency analysis were used to perform a deep learning-based prediction. As a result, a good correlation was observed with the results of the qualitative evaluation. The possibility of using the convolutional neural network (CNN) as an auxiliary means for the qualitative evaluation of abnormal noise was investigated.