Neural Network Modeling of a Magnetorheological Damper

Abstract

The magnetorheological (MR) damper is a newly developed semiactive control device that possesses unique advantages such as low power requirement and adequately fast response rate. The device has been previously tested in a laboratory to determine its dynamic properties and characterized by a system of nonlinear differential equations. This paper presents an alternative representation of the damper in terms of a multilayer perceptron neural network. A neural network model with 6 input neurons, one output neuron and twelve neurons in the hidden layer is used to simulate the dynamic behavior of the MR damper. Training of the model is done by a Gauss-Newton based Levenberg-Marquardt method using data generated from the numerical simulation of the nonlinear differential equations. An optimal brain surgeon strategy is adopted to prune the weights and optimize the neural networks. An optimal neural network is presented that satisfactorily represents dynamic behavior of the MR damper.