Measured Performance Evaluation of Pid and Neural Net Control of a Hydraulically Driven Inertia Load with Nonlinear Friction

Abstract Hydraulic systems are inherently nonlinear. When used to control an inertial load, which also exhibits nonlinear behaviour due to slip-stick friction at the contact surface, the result is a system which is highly non-linear and poses a difficult control problem. The study described in this paper examines the experimental performance of velocity control of a mass on a sliding contact surface using a servovalve and linear actuator. Conventional PID control is compared to artificial neural net (ANN) based controllers. A modified multi-input PID controller was used to train the ANN controller. The ANN based controller outperformed the PID controller when subjected to a wide variety of input signals. A second ANN co-controller was added to the loop to provide an additional corrective signal in the form of a pulse to give the system an extra surge of input to overcome the stiction friction in the zero velocity cross-over region. Excellent results were achieved with improved accuracy compared to the single ANN controller when subjected to a series of random input signals, indicating the robustness of the ANN controllers.