Position control of a flexible joint with friction using neural network feedforward inverse models

This paper presents a proposition of a control strategy based on artificial neural networks for mechanisms with hard nonlinearities. The parallelism, the regularity and the ability to approximate nonlinear functions of neural networks make them good candidates for this control task and for real-time and VLSI implementation. The flexible joint model includes Coulomb and static frictions for both motor and load and the model is used in learning and generalization phases of the neural network inverse model of the mechanism. The control structure includes an inverse model based feedforward neural network controller and a partial state feedback control law that consists of a fuzzy sliding mode control law. Simulation results show the performance of the controller, its robustness with respect to load inertia variations and its fast response to mismatch in load position initial condition