Hysteresis Dispersion Compensation with Neural Network based Controller

Abstract

Hysteresis is a commonly encountered physical phenomenon in many systems. It results in a dependence of the state of a system to its history. This non-linearity makes it particularly difficult to control accurately. There are many ways for compensating hysteresis, one of them consists of building an inverse model of the hysteresis and using it as a feedforward controller. Coupled to a feedback mechanism, the hysteresis impact can thus be minimized. However, the performance of these controllers decreases when exposed to dispersion in the hysteresis quantity or shape. The capacity of neural networks to model non-linear phenomena is not to be proven and will be put at use. In this paper, an artificial neural network model was trained to replace the conventional hysteresis inverse model. The controller performance was evaluated on a limited-angle torque motor, which exhibits hysteresis due to the magnetization saturation of the ferromagnetic materials. The experimental results pointed out the superior robustness to system dispersion of the Neural Network based controller for time and frequency response.