Semi-Active Suspension Control Using the RNN Inverse System of MR Dampers

Abstract

To suppress the vibration of a semi-active suspension with four magneto-rheological dampers, a new two-layer control strategy is put forward. The upper layer provides the desired forces of four MR dampers according to the optimal control of the full-vehicle suspension, while the lower layer applies the input voltages to MR dampers to make their forces to approximate the desired forces, through the use of the recurrent neural network (RNN) inverse system of MR damper. Based on the neural network (NN) direct system of MR damper, a RNN with back-propagation can be used to establish the inverse system of MR damper and to control the semi-active suspension. The numerical simulation demonstrates that the trained direct NN system and the RNN inverse system of MR damper can accurately describe the nonlinear relationship between the inputs and the outputs. When adapting such inverse RNN system on line, together with the optimal control of the full-vehicle suspension, in the control of a semi-active suspension of the full-vehicle model, the RNN inverse system of MR damper can greatly suppress the suspension vibration and improve the handling stability in time-domain because of the reduction of the vertical acceleration, pitch angular acceleration and the roll angular acceleration respectively. The PSD of the above three accelerations in frequency domain also show the control effect for the vertical acceleration, pitch angular acceleration and roll angular acceleration.