Semi-Active Suspensions Using Neural Networks

This paper presents a design method of neural controllers for semi-active suspensions with nonlinear devices such as electrotheological or magnetotheological dampers. It is shown that the use of two neural networks corresponding to each signs of the relative velocity is effective to realize the desired controllers, i. e., semi active controllers inherently include discontinuous mappings due to the passivity constraint and it is generally very difficult to realize them with single neural network. A model-following type controller with these neural networks is proposed, and two learning methods, the feedforward learning and the recurrent learning, for the controller are investigated from a viewpoint of system identification theories : the equation error method and the output error method