Neural network-based model reference adaptive systems for high performance motor drives and motion controls

Recently, a number of estimation techniques have been developed to achieve speed or position sensorless motor drives. However, most of these suffer from the variation of motor parameters such as the stator resistance, stator inductance or torque constant. It is known that conventional linear estimators are not adaptive to variations of the operating point. Also, model reference adaptive systems (MRAS) have shown to give better solutions for on-line adaptation and estimation problems, but the adapting mechanism is mostly linear. Neural networks (NN) have shown better results when estimating or controlling nonlinear systems. This paper presents model reference adaptive systems with neural network-based adaptation mechanism, to achieve more robust control systems. The technique can be generalized to many motor drives and motion control systems. It is applied in this paper to a Permanent Magnet Synchronous Motor (PMSM) drive. The effects of torque constant and stator resistance variations on the position and/or speed estimations over a wide speed range have been studied. In particular, the rotor speed and/or position neural estimators with on-line adaptation of torque constant and stator resistance are studied. The neural network estimators are able to track the varying parameters, speed and position at different speeds with consistent performance. Compared to other methods, they are adaptive to operating conditions and are easy in design. Simulation results wit Experimental implementation and results that justify the claims are presented.