Research On Vibration Control Of Magnetorheological Semi-active Seat Suspension Based On Deep Reinforcement Learning

The driver of the vehicle is exposed to the vibration environment for a long time,and the health of the driver is affected by this.In severe cases,driving fatigue will endanger life and safety.As an important link between the body and the human body,the seat suspension system can attenuate and isolate the vibration and impact caused by the uneven ground during the driving process of the vehicle.The selection of appropriate control algorithms can effectively improve the vibration isolation performance of the seat suspension.Semi-active suspension combines the advantages of passive suspension and active suspension,and has excellent characteristics such as low energy consumption,easy control,and wide dynamic range,and has become a research hotspot.In the current semi-active suspension control algorithm,the control system needs to make judgments based on the signals collected by the sensors,which will cause a hysteresis in the system control.Since the device needs a certain response time,the output force value of the magnetorheological damper also has a certain time delay,and these factors will affect the effect of the control algorithm.For the hysteresis problem in the control process of magnetorheological semi-active seat suspension,a visual control method for road obstacle recognition based on YOLOv3 deep learning network is proposed.A seat suspension control system model is established for the magnetorheological semi-active seat suspension,and a deep reinforcement learning controller is designed to control the magnetorheological damper to improve the ride comfort of the vehicle seat.The damping effect of the proposed control strategy on the magnetorheological seat is verified by means of damper test,system modeling,simulation analysis and experiment.The vibration damping device in the magnetorheological seat vibration damping system adopts the magnetorheological damper independently developed by the laboratory,and is tested by the MTS test platform to discuss the mechanical properties of the damper under different working conditions.The data processing software MATLAB is used to analyze the test results,and the neural network model is used to identify the mechanical characteristics of the magnetorheological damper.And determine the relevant parameters of the seat suspension system.In order to improve the vibration isolation performance of the semi-active seat suspension,a visual control method for obstacle recognition based on the YOLOv3 deep learning network was proposed,and the seat suspension system controller was established using the principle of deep reinforcement learning.By classifying and calibrating different obstacles,a data set is produced,the network training parameters are determined,and the network is trained after loading the pre-training model.The information returned after the network recognizes the target can be further calculated to obtain the obstacle category and the distance between the vehicle and the obstacle,which is convenient for subsequent selection of an appropriate control method;based on the depth deterministic policy gradient algorithm(DDPG)theory,magnetorheological seat The suspension controller is designed,and the relevant parameters of the controller are determined to train the agent.In order to verify the effectiveness of the deep reinforcement learning controller and the visual control method for obstacle recognition,the passive suspension system and the switch canopy damping control are used as comparisons for simulation analysis.By establishing passive suspension system,switch ceiling damping control and deep reinforcement learning control semi-active suspension control system model in Simulink,applying different excitations(sinusoidal excitation and random excitation)to the system,and discussing the seat suspension under different control methods The vibration reduction performance of the system is selected,and ISO 2631 "Evaluation Standard for Whole Body Vibration of Human Body" is selected as the evaluation index.The simulation results show that the deep reinforcement learning control can significantly reduce the impact on the human body caused by the uneven ground,and its vibration reduction effect is better than that of the ceiling damping control;the ideal state of the system(no delay)and the passive suspension system(with delay)are used.Compared with the magnetorheological semi-active seat suspension system controlled by constant current(with time delay),the simulation analysis results show that the visual control method of obstacle recognition can effectively solve the delay existing in the control process of magnetorheological semi-active seat suspension.Sexual issues.A magnetorheological seat vibration reduction test platform was built for experimental research to verify the effectiveness of the deep reinforcement learning control algorithm and the visual control method for obstacle recognition and the correctness of the simulation results.Using the switch ceiling as the control group,the effects of different control algorithms under the sinusoidal displacement signal and random signal were analyzed through the vibration experiment of the semi-active seat suspension.The results show that the comfort of the seat suspension is the best under the deep reinforcement learning control strategy,which is better than that of the switch ceiling control.The response of the magnetorheological semi-active seat suspension system under the passive and visual control methods is compared and analyzed,and the results show that the visual control method for obstacle recognition can effectively improve the vibration isolation performance of the seat suspension.