Research On Heavy-Duty Trucks Rollover Warning Based On AR-PNN

The present situation of road traffic safety in our country is serious,especially on the mountain road.Heavy-duty trucks are prone to rollover on it because of their large size,large cargo capacity and high center of mass.In addition,it is hard to take effective measure to avoid the risk of rollover for drivers as a consequence of extremely short rollover time.This means huge casualties and property damage.Currently the most effective way to reduce vehicle rollover accidents is early rollover warning of heavy-duty trucks.Combined with the typical form of road traffic accidents in our country,of which the "rollover through curve movement" was taken into investigated and the heavy-duty truck was taken as the research object,the early warning program was proposed on the foundation AR-PNN model and verified by the joint simulation of Trucksim and Simulink.The main research contents are as follows:One kind of vehicle was taken as research object,a three-degree-of-freedom simulation model of heavy-duty trucks was built.Analyze the lateral movement of the truck,the yaw motion,and the force of the roll motion during the curve motion to explore the potential factors affecting the rollover of heavy-duty trucks.The Step_signal and Fishhook were selected for simulation verification to ensure the accuracy of the model.The longitudinal state,lateral state,swing state and steering state of the heavy-duty truck in the movement shows continuity in time dimension while shows strong independence character in special dimension.Based on the pattern recognition theory,a prediction method of the movement state and early warning algorithm of the rollover of heavy trucks was proposed.The AR model adopted to predict the vehicle motion state parameters and the two-layer PNN model applied to identify the vehicle motion state.Using the Trucksim software can get the experiment data to trained the AR-PNN rollover model.Four typical dangerous steering conditions,such as Step_signal and Fishhook,were selected to train the AR-PNN model as experimental conditions.Considering the complexity of the actual driving environment,the reliability of the AR-PNN model was verified offline in the step steering conditions and compound conditions.In this paper,the mode of rollover prediction and warning was built in the way combining the AR model and the PNN model,in which the former model could display the data of vehicle motion state parameters visually so the defects of other models in terms of intuitiveness could be made up for,while the advantages of the latter one reduce the time of identifying vehicle motion status for the model and had good real-time performance on the premise of satisfying accuracy.The experimental results showed that the movement state of heavy-duty trucks could be identified accurately and potential rollover hazard could be warned early by the built AR-PNN model,resulting in that the rollover accidents of heavy-duty truck could be avoided to a certain extent.The modeling theory and method used in this paper have certain theoretical reference significance for the study of the rollover of other vehicles.