Neural Network Semi-active Control For Vibration Of Vehicle Suspensions

Traveling and ride comfort is the basic evaluation criterion for a ground vehicle. This dissertation presents an effective method based on the new-type smart material for vibration isolation of vehicle suspensions, and further to improve the performance of suspensions. The paper reports a systematical investigation on semi-active vehicle suspension, including dynamic characteristic of vehicle suspension, the properties and functions of a magnetoreological (MR) damper, the adaptive neural network control and the experiment on a test rig of quarter car model. The four topics are studied from Chapters 2 to 5, respectively.The ride comfort of a vehicle mainly depends on the vertical vibration of vehicle body. Chapter 2 presents the study on a two-degree-of-freedom model of quarter car. The vibration characteristic and the effect of the suspension parameters on vertical vibration are mainly analyzed. The amplitudes of vehicle body acceleration of half car model of four degrees of freedom verify the efficacy of quarter car model. The passive suspension via the specified parameters can optimize the suspension sometimes. However, the global optimization is difficult to achieve under disturbance. This paper presents an adaptive control for the semi-active suspension using the combination of MR damper and feedback neural networks.Analyzing the MR damper, a smart actuator, Chapter 3 gives the design of pre-amplifier of current, which provides the external magnetic field. And it discusses the additional nonlinear stiffness resulted from the transition of MR fluid from liquid to semi-solid or solid. According to the study of the vibration of car body, it can be concluded that this additional stiffness, a weakly nonlinear term, would not influence the purpose of vibration isolation. At the same time, the nonlinear stiffness hardly affects the rolling stiffness and roll angle. Consequently, the suspension with MR damper not only increases ride comfort, but also guarantees the controlling stability of vehicle.As the MR damper features nonlinearity, the vehicle suspension equipped with MR damper is a nonlinear system. Nonlinear neural network (NN) control strategy, which was certified the high capacity of approximation, is adopted to control this typical nonlinear system. Chapter 4 presents an error back propagation algorithm with quadratic momentum of the multilayer forward neural networks that will speed up the error convergence velocity. And it proposes an indirect adaptive neural network law. This law optimizes the control structure and improves the quality of control. The numerical simulation results show that the semi-active suspension with MR damper using NN strategy is superior to those with traditional control or without any control. To narrow the hysteretic loops and improve saturation, the author suggests partially modifying the parameters of the RD1005 MR damper hi terms of the exclusive utilization, such as the vehicle suspension.Chapter 5 presents an experiment on a test rig of quarter car model equipped with MR damper. The low frequency of road-induced vibration of a vehicle and quick response of MR damper make NN real-time control of possible. In the experiment the timer and counter are adopted to realize computer multi-work environment, and the Turbo-C computer language is used to call on the hardware. The effects of various neural networkparameters on the vibration of car body are compared and the results demonstrate that only single implicit layer structure can considerably improve the ride comfort. However, excessive implicit nodes can not increase the performance but lengthen the time consuming. The real-time control of neural network is adopted successively for the semi-active suspension with MR damper. The experiments indicate that the suspension with MR damper and NN control is superior to the passive one in the frequency band of concern. The isolation performance is particularly pronounced at the resonance of car body. The existence of additional nonlinear stiffness and its weak affecti...