关键词: 智能轮胎, 加速度, 轮胎磨损, 分类, 台架试验

Abstract: In light of the issue that the tread depth of passenger car tires was shallow and the conventional identification of wear characteristics were not obvious based on acceleration time-domain signals. This paper aimed to explore the classification and estimation of tire wear through the analysis of frequency-domain features of internal tire acceleration. Firstly, an intelligent tire test system was built by self-developed special rim assembly and data collector, and a three-axis accelerometer was arranged in the tire inner liner. The acceleration values were obtained by wired method, and the sampling frequency was 50 kHz. Secondly, based on the built test system, the typical tire pure rolling test was carried out on the Flat Trac bench, and the data was analyzed to clarify the parameters of the classification algorithm and construct the data set. The test tires included new tire, semi-grinding tire and full-grinding tire. The data analyses show that the circumferential acceleration Ax and radial acceleration Az of tires with different wear degrees are significantly different in the frequency domain of 5 kHz. Therefore, the vibration amplitude of Ax and Az in the frequency domain of 5 kHz was extracted at an interval of 10 Hz as the feature point, and the frequency domain data sets FDAx and FDAz were established respectively with vertical load, speed and tire pressure. Finally, the random forest algorithm was used to train and predict the two data sets respectively. When the number of decision trees and the minimum number of leaves are 20 and 20 respectively, the model classification effectivenes is the best. The results show that the average classification accuracy of the frequency domain data set FDAz is 95.1543%, which is higher than that of the data set FDAx. Compared with the time domain data sets TDAx and TDAz constructed by extracting Ax and Az time domain features from the same experimental data, the classification accuracy is increased by 16.18% and 10.08% respectively. At the same time, the feature values of the FDAz data set are optimized to obtain the optimal model classification accuracy and real-time performance when the feature frequency band and the number of feature points are within 1000 Hz and 100, respectively. The research shows that it is feasible to identify the degree of tire wear based on the frequency domain signals of the acceleration in the tire, which provides an effective means for the identification of the degree of tire wear of passenger cars.

Key words: intelligent tire, acceleration, tire wear, classification, bench test