Magnetorheological (MR) dampers are adaptive devices whose properties can be adjusted through the application of a controlled voltage signal. A semi-active suspension system incorporating MR dampers combines the advantages of both active and passive suspensions. For this reason, there has been a continuous effort to develop control algorithms for MR-damped vehicle suspension systems to meet the requirements of the automotive industry. The overall aims of this thesis are twofold: (i) The investigation of non-parametric techniques for the identification of the nonlinear dynamics of an MR damper. (ii) The implementation of these techniques in the investigation of MR damper control of a vehicle suspension system that makes minimal use of sensors, thereby reducing the implementation cost and increasing system reliability. The novel contributions of this thesis can be listed as follows: 1- Nonparametric identification modelling of an MR damper using Chebyshev polynomials to identify the damping force from both simulated and experimental data. 2- The neural network identification of both the direct and inverse dynamics of an MR damper through an experimental procedure. 3- The experimental evaluation of a neural network MR damper controller relative to previously proposed controllers. 4- The application of the neural-based damper controller trained through experimental data to a semi-active vehicle suspension system. 5- The development and evaluation of an improved control strategy for a semi-active car seat suspension system using an MR damper. Simulated and experimental validation data tests show that Chebyshev polynomials can be used to identify the damper force as an approximate function of the displacement, velocity and input voltage. Feed-forward and recurrent neural networks are used to model both the direct and inverse dynamics of MR dampers. It is shown that these neural networks are superior to Chebyshev polynomials and can reliably represent both the direct and inverse dynamic behaviours of MR dampers. The neural network models are shown to be reasonably robust against significant temperature variation. Experimental tests show that an MR damper controller based a recurrent neural network (RNN) model of its inverse dynamics is superior to conventional controllers in achieving a desired damping force, apart from being more cost-effective. This is confirmed by introducing such a controller into a semi-active suspension, in conjunction with an overall system controller based on the sliding mode control algorithm. Control performance criteria are evaluated in the time and frequency domains in order to quantify the suspension effectiveness under bump and random road excitations. A study using the modified Bouc-Wen model for the MR damper, and another study using an actual damper fitted in a hardware-in-the-loop- simulation (HILS), both show that the inverse RNN damper controller potentially gives significantly superior ride comfort and vehicle stability. It is also shown that a similar control strategy is highly effective when used for a semi-active car seat suspension system incorporating an MR damper.
展开▼
