Sensorless Control of Induction Motor Drives at Very Low and Zero Speeds Using Neural Network Flux Observers

A new method is described which considerably improves the performance of rotor flux model reference adaptive system (MRAS)-based sensorless drives in the critical low and zero speed regions of operation. It is applied to a vector-controlled induction motor drive and is experimentally verified. The new technique uses an artificial neural network (NN) as a rotor flux observer to replace the conventional voltage model. This makes the reference model free of pure integration and less sensitive to stator resistance variations. This is a radically different way of applying NNs to MRAS schemes. The data for training the NN are obtained from experimental measurements based on the current model avoiding voltage and flux sensors. This has the advantage of considering all drive nonlinearities. Both open-and closed-loop sensorless operations for the new scheme are investigated and compared with the conventional MRAS speed observer. The experimental results show great improvement in the speed estimation performance for open-and closed-loop operations, including zero speed