Road Friction Coefficient Estimation Based On Multi-sensor And Data Fusion

With the continuous development of modern automobile technology,how to quickly and accurately estimate the road friction coefficient has become an issue in the fields of vehicle safety and autonomous driving.Obtaining an accurate road friction coefficient can not only optimize the control strategy of the vehicle active safety control system,but also can improve the performance of decision-making and planning in autonomous driving.Due to the cost of sensors,complex vehicle conditions and the coupling characteristics of vehicle dynamics,there is no equipment can measure the road friction coefficient in real time on mass-produced vehicles currently.Therefore,based on vehicular sensor,vehicular camera and roadside 3D camera,this paper proposes a multisensor and data fusion algorithm to estimate the road friction coefficient.The specific research contents are as follows:Firstly,considering that the road friction coefficient estimation method based on traditional vehicle dynamics needs to obtain vehicle state,vehicle state estimation system which uses measurable vehicle states as input to estimate unmeasurable vehicle states is built: 1)Judge the current vehicle state based on vehicular sensor signals,and estimate and compensate the vehicle acceleration offset;2)Estimate the normal force,longitudinal force and lateral force of the tire based on vehicle load transfer model,tire longitudinal dynamics model and PID state observer and three degrees of freedom vehicle model and Kalman filter;3)Two fusion strategies are designed to fuse the longitudinal vehicle speed based on the wheel-speed method and kinematics method and sideslip angle based on model method and kinematics method,and the fusion result is the estimated value of longitudinal vehicle speed and sideslip angle;Secondly,considering that the road friction coefficient as an external input parameter is complex and changeable and there are slip rate deviations and difficult to estimate under small excitation conditions,an estimation system is built to estimate the slip rate offset and the friction coefficient in the linear and non-linear regions of the tire based on the simplified Uni Tire model and the gradient descent method: 1)Retained the core parameters of the traditional Uni Tire model and Obtained a simplified Uni Tire model which is suitable for friction coefficient estimation;2)Due to the offset problem of slip rate estimation,the slip rate offset is introduced into the simplified Uni Tire model;3)The simplified model is tested based on the gradient descent method.The results show that the friction coefficient is difficult to estimate in linear region of the tire,estimating the friction coefficient and slip rate offset together will reduce the reliability of the estimation result of the adhesion coefficient;4)An estimation strategy is designed based on the test results: the slip rate offset and the road friction coefficient are estimated separately in the linear and non-linear regions of the tire.The judgment conditions of the linear and non-linear regions require other sensor information;Thirdly,a road condition recognition model based on camera and Convolutional Neural Network is built: 1)The traditional neural network model and neuron model are studied,and the common optimization methods of neural networks are studied;2)Based on the fully connected Neural Network,the Convolutional Neural Network is introduced and the convolution and pooling layers of the network are related to research;3)Taking into account the collection cost,quality of the data set and Hardware performance.Based on virtual data and data enhancement as a means,the existing data set has been trained and tested based on the Convolutional Neural network Model.Then,the road friction coefficient reflects the friction limit between the rubber tire and the road surface,which size is closely related to the roughness of the road surface.Considering that the roadside 3D camera can accurately measure the rough texture of the road surface,a micro analytical model of rubber friction is developed based on the 3D camera: 1)The physical properties of the road surface and rubber in terms of friction were studied separately,and the road power spectrum and rubber composite modulus models were established;2)The hysteretic friction and adhesion friction were studied based on the laws of energy conservation and molecular surface free energy.3)A general theoretical model of rubber friction is established based on Persson’s contact theory,which integrates road power spectrum,rubber composite modulus,temperature and other clear physical parameters;Lastly,a friction coefficient estimation fusion algorithm is designed based on multi-sensors:1)When there is a 3D camera on the road section,the fusion algorithm takes the result of the rubber friction model as the output and corrects the tire model parameters;2)When there is no 3D camera on the road section,The fusion algorithm uses the classification result of the road surface by the camera(Convolutional Neural Network)as the feedforward information of the friction coefficient estimation system.When the vehicle is in a small excitation state,the fusion algorithm uses the feedforward information as the estimated value of the friction coefficient and estimates the slip rate offset based on the simplified Uni Tire model;when the road surface is sufficiently excited and the tire is in the non-linear region,the fusion algorithm estimates the friction coefficient based on the simplified tire model,and the result is used as the output.