Study On The Estimation Methodology Of Maximum Tire-Road Friction Coefficient Based On Multi-Vehicle Fusion Under Pulse Braking Condition

In the intelligent and connected transportation,real-time road condition monitoring by vehicle-self-perception is an effective way to improve road safety and traffic efficiency.The large number of samples provided by the multi-vehicle data and the rapid processing method of the central cloud computation platform combining together help timely and accurately respond to changes on the road surface,such as the icing emergency,which would reduce the hazards of traffic accidents if monitored.When the tire is driving on a road surface at a stable tire-road friction coefficient,the slip rate of wheels will perform a definite corresponding relationship with the friction coefficient received by tire on the moment.Therefore,the peak of friction coefficient can be deduced based on the characteristics of the slip rate and the synchronous friction coefficient,and then the origin type and changes of road surface can be matched through typically defined correspondence.On account of the application of estimating the maximum tire-road friction coefficient in a real traffic environment,supported by Burckhardt model and vehicle self-sensing data,this study took the pulse braking conditions as the studying object,which commonly occur in non-intersection environments.It explored the estimation mechanism,which guaranteed the accuracy by multi-vehicle data fusion and the real-time effect of communication in the networked environment.The main research contents are as follows:First of all,according to the characteristics of the target conditions,this research conducted pretests on the data and time series and optimized the method of estimating the maximum friction coefficient under one pulse of braking operation.Burckhardt tire-road model is widely used in related research on road surface condition analysis based on driving data.Because the Burckhardt model is an empirical model,it will introduce estimation bias when tire replacement or tire wear and other problems occur.In addition,the model features in the lower slip rate interval are different from the higher slip rate interval.Therefore,the Burckhardt model is rarely used in the past studies to estimate the friction coefficient in the low slip rate interval,and some of them choose to ignore the data in the low slip rate interval or separate the road model with abnormal characteristics.This research focused on the image comparison using Burckhardt classic road-tire model to estimate the tire-road friction coefficient,and proposed the following improvements in comparison method:（1）Separated definition mechanism was raised to solve the problem of the failure of the comparison result due to the intersection line in the low slip rate area;（2）For the dynamic data characteristics of the pulse braking,a solution based on the braking time window and the screening and cleaning strategy for the shortest and effective data interception after fitting was examined;（3）on account of the characteristics of pulse braking,an optimization mechanism was introduced which ignored the turning point of the linear and non-linear sections of the experimental curve,and adopted the fusion of the two estimation methods based on the landing point and the slope of curve.There are obvious differences between the estimation methods based on the location of the landing point before and after optimization,and the estimation results based on the characteristic of curve slope and on the point have significantly different growth feature.The final result showed that the optimization improved the estimation accuracy of the tire-road friction coefficient calculation from 56.9%to 75.6%.After completing the working condition analysis and data interception rules,the Car Sim-Simulink vehicle dynamics simulation platform was designed to collect vehicle self-perception original data under diversified working conditions and road working conditions.The experiment design 6 kinds of brake torque,8 kinds of braking initial speed,5 kinds of vehicle,3 kinds of road types（the peak friction coefficient of the road are respectively 0.35,0.69 and 0.91）full factor experiment,a total of 720 sets of experimental data（710 valid results）.The research analyzed the experimental results from various aspects,and then summarized the influence mechanism of braking torque,initial vehicle speed and vehicle type on the experimental results.and conducted principal component analysis of vehicle static parameters.The key 14 feature parameters introduced based on the analysis results including:maximum slip rates_max,maximum tire-road friction coefficientμ_{ma x 1}（based on the landing region of point）,μ_{max 2}（based on slope of curve）,and the result of principal component analysis of vehicle static parameters,etc.Through various classification and regression neural network experiments,the research explored the scheme based on multi-vehicle data fusion estimation,and selected the optimal feature parameter combination.After cleaning the data with similarity limitation of 99.9%,the following conclusions were obtained:（1）Based on Gaussian process regression,the RMSE could be stably optimized at about 0.1,and the end speed v_t dropped less effect on the regression;（2）The quadratic support vector machine was the optimal classification model,and the classification accuracy rate reached 83%,of which the first,second,and third principal components of vehicle static parameters had no significant impact on the classification effect.The result of feature learning based on neural network showed a little bit over-fitting performance.Due to the limitation of the number of experimental samples,data cleaning was conducted only in similarity of feature parameters and abnormal values of data.If the number of experimental samples could be increased,the similarity threshold can be further reduced,improving the classification and regression effect.