Data-driven Based Fault Diagnosis And Abnormal Monitoring Of Intelligent Vehicle Steering System

With the development of automotive industry,the degree of automobile automation is increasing.However,while intelligent automobile bring convenience to travel,they also lead to an increasingly complex internal structure,which increases the potential risk of fault.The steering system directly determines the lateral movement of the vehicle,which plays an important role in the safe operation of the vehicle.If it is not detected and repaired in time at the early stage of fault,serious consequences may occur when the fault expands.This may endanger the lives and properties of drivers and passengers as well as the surrounding traffic participants.Therefore,it is important to conduct fault diagnosis and anomaly monitoring research on the steering system of intelligent vehicles.Firstly,the simulation of intelligent vehicle steering system was carried out based on Simscape physical modeling platform.By analyzing the structure of the steering system,the physical model of the intelligent vehicle steering system was established.Then a joint simulation with Car Sim was conducted to replace the steering system structure in the Car Sim vehicle model with the constructed steering system model,and the accuracy of the constructed model was verified.In order to simulate the real vehicle operation conditions,a variety of common driving conditions were designed.The parameters such as vehicle speeds and driving trajectories were reasonably combined.In addition,by analyzing the key faults of the steering system,different faults were simulated in the constructed physical model separately to obtain the vehicle operation data under the fault condition.Secondly,a steering system fault diagnosis model based on GWO-RF was proposed.The Random Forest(RF)algorithm was used to learn and classify the collected fault dataset.The key parameters were selected by traversal method.The confusion matrix was analyzed and the effectiveness of the RF algorithm was compared with other algorithms.To address the problem of setting key parameters in the RF algorithm,the Grey Wolf Optimizer(GWO)algorithm was used to search for the key parameters,which improved the efficiency of parameter selection and the classification accuracy.By analyzing the confusion matrix and comparing with other algorithms,the proposed method was proved to have better fault diagnosis accuracy and faster diagnosis speed.In order to monitor the unknown faults or other abnormal conditions of the vehicle steering system,an Attention-LSTM-based steering system anomaly monitoring method was proposed.The vehicle operation dataset under normal conditions was collected for training.The vehicle yaw at each moment was predicted by the anomaly monitoring model,and its normal state threshold was determined based on the residual distribution.A test set with fault states was used for validation.The feasibility of the method was demonstrated when the residuals continuously exceeded the threshold range after a fault occurred.For safety reasons,a steering system test bench was built to collect fault data and to validate the GWO-RF-based fault diagnosis model.Two critical faults of the steering system were set up in the bench,and feature datasets with unbalanced samples were collected.The test results demonstrated that the proposed fault diagnosis model had good diagnostic effect and good generalization for steering system datasets with different sources and characteristics.In order to validate the Attention-LSTM-based anomaly monitoring model,a real-vehicle test platform was built to collect the vehicle normal operation dataset.By learning the steering system operation law,the vehicle operation status was monitored.The fault datasets were obtained by fault injection,and the effectiveness of the proposed anomaly monitoring model was verified.