基于智能轮胎系统的实时路面辨识技术

Real-Time Pavement Recognition Technology Based on Intelligent Tire System

在复杂和极限工况下,路面附着系数是进行轮胎受力分析和车辆动力学控制的重要状态参数。相对于模型估计的方法,智能轮胎技术能够将轮胎与路面的交互信息反馈给车辆控制系统。本文提出了一种将智能轮胎系统和机器学习相结合的车辆路面附着系数获取方法。首先,考虑行驶工况环境进行传感器选型,开发基于MEMS三轴加速度传感器的智能轮胎硬件采集系统,并采用简化硬件结构的无线传输模式。其次,通过采集不同路面上的实车实验数据进行车辆实验收集机器学习训练的数据集,并分析轮地关系及信号特征。最后,将CNN与LSTM两者的优势相结合实现了对加速度时序信号的特征学习。通过与其它神经网络模型训练结果的比较,验证了所提CNN-LSTM双通道融合神经网络模型的有效性和准确性。本文提出的路面辨识方案实现了实时道路识别的目标,硬件与软件架构和神经网络模型更适合车辆系统搭载,为车辆运动控制提供了实时准确的路面信息。

Under complex and extreme conditions,road adhesion coefficient is an important state parame‐ter for tire force analysis and vehicle dynamics control.Compared with the method of model estimation,the intelli‐gent tire technology can feed back the interaction information between the tire and the road surface to the vehicle control system.In this paper,a method of obtaining road adhesion coefficient of vehicle by combining intelligent tire system and machine learning is proposed.Firstly,considering the driving conditions,the sensor selection is carried out,and the intelligent tire hardware acquisition system based on MEMS three-axis acceleration sensor is devel‐oped,and the wireless transmission mode with simplified hardware structure is adopted.Secondly,the data set of machine learning training is collected by vehicle experiments by collecting real car test data on different road surfac‐es and the wheel-ground relationship and signal characteristics are analyzed.Finally,the feature learning of acceler‐ation timing signal is realized by combining the advantages of CNN and LSTM.The effectiveness and accuracy of the proposed CNN-LSTM dual channel fusion neural network model are verified by comparing with the training results of other neural network models.The road identification scheme proposed in this paper realizes the goal of real-time road recognition,and the hardware and software architecture and neural network model are more suitable for vehicle system loading,providing real-time and accurate road information for vehicle motion control.