Data-Driven Nonlinear VRFT for Dead-Zone Compensation in Servo Systems Control

We propose herein a data-driven dead-zone (DZ) compensation strategy using a model-free Virtual Reference Feedback Tuning (VRFT) approach. The VRFT tuning scheme is accommodated for two controller structures: the first one which explicitly includes a model of the DZ inverse to be identified and the second one which uses a Neural Network (NN) to model the controller to be identified. The main question to be answered here is whether if the inclusion of an explicit model of a static nonlinearity (DZ in this case) can be avoided while preserving the control system performance. Thorough investigation case studies are carried out both in simulation and experiment on a laboratory 3D-crane system as a typical servo system control application.