汽车工程 ›› 2023, Vol. 45 ›› Issue (9): 1765-1771.doi: 10.19562/j.chinasae.qcgc.2023.09.024

所属专题: 底盘&动力学&整车性能专题2023年

• • 上一篇    下一篇

基于卷积神经网络的汽车操纵稳定性试验类型识别方法

管欣1,仲昭辉1,詹军1(),奚腾龙1,叶昊1,高深圳1,成健2,3,廖世辉2,3,蔡均2,3   

  1. 1.吉林大学,汽车仿真与控制国家重点实验室,长春  130025
    2.汽车智能仿真重庆市重点实验室,重庆  401100
    3.重庆长安汽车股份有限公司,重庆  401100
  • 收稿日期:2022-11-29 修回日期:2023-02-28 出版日期:2023-09-25 发布日期:2023-09-23
  • 通讯作者: 詹军 E-mail:zhanj@jlu.edu.cn
  • 基金资助:
    国家重点研发计划项目(2018YFB1502700)

Vehicle Handling and Stability Test Type Recognition Method Based on Convolutional Neural Network

Xin Guan1,Zhaohui Zhong1,Jun Zhan1(),Tenglong Xi1,Hao Ye1,Shenzhen Gao1,Jian Cheng2,3,Shihui Liao2,3,Jun Cai2,3   

  1. 1.Jinlin University,State Key Laboratory of Automotive Simulation and Control,Changchun  130025
    2.Chongqing Key Laboratory of Automobile Intelligent Simulation,Chongqing  401100
    3.Chongqing Changan Automobile Co. ,Ltd. ,Chongqing  401100
  • Received:2022-11-29 Revised:2023-02-28 Online:2023-09-25 Published:2023-09-23
  • Contact: Jun Zhan E-mail:zhanj@jlu.edu.cn

摘要:

针对汽车操纵稳定性试验评价指标自动化处理需要自动识别试验类型的需求,提出一种基于卷积神经网络的汽车操纵稳定性试验类型自动分类方法。在分析汽车操纵稳定性试验类型数据图像特征的基础上,建立了由1个输入层、3个卷积层、3个批归一化层、2个最大池化(Max-pooling)层、5个线性整流函数(ReLU)层、3个全连接层、2个活化(Dropout)层、1个激活函数(Softmax)层和1个分类输出层组成的汽车操纵稳定性试验类型分类卷积神经网络模型。利用2 250组试验采集的数据对模型进行了训练和验证。经验证,类型分类准确率为99.33%,平均识别时间为0.05 s。结果表明,本文提出的基于卷积神经网络的汽车操纵稳定性试验类型自动识别方法可有效区分不同试验类型,可用于汽车操纵稳定性试验结果的自动处理,显著提升汽车操纵稳定性试验自动化处理水平。

关键词: 汽车试验, 类型识别, 卷积神经网络, 汽车操纵稳定性

Abstract:

To meet the need of automatic identification of test types, which is aimed at automatic processing of vehicle handling and stability test evaluation indicators, this paper proposes a vehicle handling and stability test type recognition method based on convolutional neural network. On the basis of analyzing the image characteristics of the test type data, a vehicle handling and stability test type recognition model based on convolution neural network is established, which consists of 1 input layer, 3 convolution layers, 3 batch normalization layers, 2 Max-pooling layers, 5 linear rectification function (ReLU) layers, 3 full connection layers, 2 Dropout layers, 1 Softmax layer and 1 classification layer. The model is trained and verified using 2 250 groups of data collected from the tests. The accuracy of type recognition is 99.33%, and the average recognition time is 0.05 s. The results show that the vehicle handling and stability test type recognition method based on convolutional neural network proposed in this paper can effectively distinguish different test types, which can be used for automatic processing of vehicle handling and stability test results, and can significantly improve the automatic processing level of vehicle handling and stability test.

Key words: automobile test, type recognition, convolution neural network, vehicle handling and stability