基于轮内加速度的乘用车胎面磨损程度分类试验

Classification Test of Tire Tread Wear of Passenger Cars Based on In-wheel Acceleration

围绕乘用车轮胎花纹深度浅、常规的基于加速度时域信号辨识磨损特征不明显的问题,研究了基于胎内加速度频域特征的胎面磨损程度分类估算问题。首先通过自主开发的专用轮辋总成和数据采集器等搭建智能轮胎测试系统,并在轮胎气密层布置1个三轴加速计,采用有线方式获取加速度值,采样频率50 kHz。其次基于搭建的测试系统在Flat Trac台架上进行典型轮胎纯滚动试验,分析用于明确分类算法的参数并构建数据集。试验轮胎包括新胎、半磨胎和全磨胎,数据分析发现不同磨损程度轮胎周向加速度A _(x)和径向加速度A _(z)在频域5 kHz内区别度明显。故以间隔10 Hz提取A _(x)和A _(z)频域5 kHz内振动幅值作为特征点,连同垂直载荷、速度和胎压,分别建立频域数据集FDA _(x)、FDA _(z)。最后利用随机森林算法对这2个数据集分别进行训练与预测,优化参数决策树数目和最小叶子数均为20时模型分类效果最优。结果表明,频域数据集FDA _(z)分类准确度平均值为95.1543%,高于数据集FDA _(x)。与相同试验数据提取A _(x)和A _(z)时域特征构建的时域数据集TDA _(x)、TDA _(z)对比,分类准确度分别提高了16.18%和10.08%。同时优化数据集FDA _(z)的特征取值后发现,特征频段和特征点个数分别为1000 Hz内和100时模型分类准确度和实时性最优。研究结果表明,基于胎内加速度的频域信号进行轮胎磨损程度辨识是可行的,为乘用车胎磨损程度辨识提供了有效手段。

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 A_(x)and radial acceleration A_(z)of tires with different wear degrees are significantly different in the frequency domain of 5 kHz.Therefore,the vibration amplitude of A_(x)and A_(z)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 FDA_(x)and FDA_(z)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 FDA_(z)is 95.1543%,which is higher than that of the data set FDA_(x).Compared with the time domain data sets TDA_(x)and TDA_(z)constructed by extracting A_(x)and A_(z)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 FDA_(z)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.