Neural network learning control of automotive active suspension systems

The paper contains theoretical treatments of neural network controllers allied to various simplified automotive active suspensions. In each case considered, the controlled system has a defined objective, the minimisation of a cost function, and the neural network is set up in a learning structure such that it systematically improves the system performance via repeated trials and modifications of parameters. The learning process is shown to be effective and, in selected cases, to give system performance in agreement with prior results, for both linear and non-linear plants. The final objective of the work is effective neuro-control of a variable geometry active suspension system, the action of which is essentially non-linear. This problem is introduced. The learning structures established are important milestones on the journey to the final objective. Also, they are generic and are likely to be more widely applicable.