| With the advent of the era of automobile electrification,driverless driving has become the direction of the current automobile industry reform.In unmanned driving,it is the mainstream development direction to collect signals through sensors installed on vehicles and match them with signal processing algorithms.Unmanned control strategies need to vehicle traffic and pedestrians near information,traffic information and other driving signal for decision-making,the road surface adhesion estimation is also an important part of the,road surface adhesion refers to the road surface to provide vehicle traction,quickly identify factors of adhesion of the pavement can help the vehicle electronic control unit to respond to the road changes in time,avoid the happening of the risk,So the rapid recognition of pavement adhesion changes has its research value.In this paper,the road adhesion identification technology of three-axle tractors was studied in depth.Aiming at the error of slip rate calculation caused by neglecting the rolling radius of wheels in the current research,a wheel rolling radius error elimination method based on the weighted least square algorithm was proposed to eliminate the error in slip rate calculation.In view of the problem that the influence of vehicle mass on vehicle adhesion is usually not considered in current studies,this paper proposes to use Generalized Regression Neural Network(GRNN)to fit the corresponding relationship between vehicle traction force and mass and speed signals on high adhesion coefficient road surface,and normalize the mass and speed signals.To solve the problem of low feature dimension and small sample size of data set,this paper uses non-uniform coding to enhance the feature of data set,and uses pre-trained Bidirectional Encoder Representation from Transformers model to identify road adhesion coefficient.The main research work of this paper is as follows:(1)Research on error elimination method of original signal of three-axle tractor.For noise existing in the original signal and error signal characteristic is not obvious,study of the existing triaxial dynamic model of tractor,referring to the common triaxial tractor driving conditions,establish the road adhesion recognition model,is proposed based on sliding window algorithm and weighted median filter method to eliminate noise error least squares algorithm,The noise in the original signal is suppressed and the slip rate calculation error caused by different wheel rolling radii is eliminated.(2)Research on normalization method of vehicle mass and Speed.In view of the problem that the change of road adhesion caused by different vehicle mass and speed is usually ignored in current studies,this paper proposes a vehicle mass and speed normalization algorithm based on Back Propagation Neural Network(BPNN)model and GRNN model.After comparing the advantages and disadvantages of the two models,The GRNN model was selected to normalize the mass and speed,and the traction force of the vehicle was modified to eliminate the interference caused by the mass and speed on the recognition of road adhesion.(3)Research on identification method of pavement adhesion coefficient.Dimensions based on the existing data set is low and less quantity of the data is difficult to obtain high recognition accuracy problem,introduces the time series and non-uniform code to the existing data sets feature enhancements,recognition algorithm is proposed based on BERT road adhesion coefficient and improve the recognition accuracy of triaxial tractor road adhesion coefficient. |
