基于轮胎垂向加速度的路面不平度等级识别研究  

Research on Pavement Unevenness Grade Identification Based on Tire Vertical Acceleration

作　　者：卢俊辰 崔凯特 钟妤馨 胡梦宜 董林玺[1,2] LU Junchen;CUI Kaite;ZHONG Yuxin;HU Mengyi;DONG Linxi(Smart Microsensors and Microsystems Engineering Research Center of Ministry of Education,Hangzhou Dianzi University,Hangzhou Zhejiang 310018,China;Zhejiang Hongzhen Smart Chip Co.,Ltd.,Huzhou Zhejiang 313200,China)

机构地区：[1]杭州电子科技大学电子信息学院智能微传感器与微系统教育部工程中心,浙江杭州310018 [2]浙江宏振智能芯片有限公司,浙江湖州313200

出　　处：《传感技术学报》2025年第3期458-467,共10页Chinese Journal of Sensors and Actuators

基　　金：浙江省级人才项目(2021R52009)。

摘　　要：针对路面不平度等级识别的问题,提出了一种基于轮胎垂向加速度最优特征的BP神经网络识别方法。首先搭建了路面不平度模型和二自由度车辆平顺性模型,通过仿真获得轮胎垂向加速度和路面不平度等级。然后,基于轮胎垂向加速度先后构造40种特征提取方案,引入随机森林模型评估各个特征的重要性,初步确定最优特征。最后,将40种特征分别作为BP神经网络的输入,路面不平度等级为输出,构造3层BP神经网络,引入准确率等评价标准验证最优特征。研究表明在基于轮胎垂向加速度特征的路面不平度等级识别任务中,最优特征为信号的二阶差分绝对值之和,以该特征为输入构造的BP神经网络不仅识别准确率可达到99%,并且具备较低复杂度以及较好的速度鲁棒性。Targeting at the problem of road surface unevenness grade recognition,a BP neural network recognition method based on the optimal characteristics of tire vertical acceleration is proposed.Firstly,the road surface unevenness model and the two-degree-of-freedom vehicle harshness model are constructed,and the tire vertical acceleration and road surface unevenness level are obtained through simu-lation.Then,based on the vertical acceleration of tires,40 feature extraction schemes are constructed successively,and a random forest model is introduced to evaluate the importance of each feature and preliminarily determine the optimal feature.Finally,40 features are used as input to the BP neural network,and the road surface unevenness level is used as the output,a three-layer BP neural network is constructed,and evaluation criteria such as accuracy are introduced to verify the optimal features.The results show that in the road sur-face unevenness grade recognition task based on the vertical acceleration feature of tires,the optimal feature is the sum of the absolute values of the second-order difference of the signal,and the BP neural network constructed with this feature as input not only has a recog-nition accuracy of 99%,but also has low complexity and good speed robustness.

关 键 词：路面不平度识别 BP神经网络 特征提取 随机森林 轮胎垂向加速度 

分 类 号：U463[机械工程—车辆工程] U461.4[交通运输工程—载运工具运用工程] TP391[交通运输工程—道路与铁道工程]