A neural network approach to position sensorless control of brushless DC motors

To control brushless de motors without using position sensors has been a challenging task for some time. This paper presents a new approach to the problem based on neural network methods. Instead of using position sensors, neural networks are used to identify the rotating angles of the rotor. Neural networks are trained to associate between the measured phase voltages and currents and the rotor positions. Once this association is established, the networks perform independently to identify the rotor positions based on the measured voltages and currents. The background, theoretical analysis and the results obtained are described in this paper.