Artificial neural networks and inductance vector based sensorless torque estimation in switched reluctance motor drive

Abstract

Switched Reluctance (SR) motors have taken remarkable role in the industry due to their rugged behavior, simple structure and low cost in mass production. An SR motor can produce large torque in a wide speed range. However, it has some drawbacks such as high torque ripple and noise. Therefore, torque ripple reduction strategy is required. Direct Torque Control (DTC) allows users to control the torque and the torque ripple. Feedback of motor torque is essential in the DTC as in most of the SR motor control techniques. The DTC conventionally employs position sensor and estimates motor torque from rotor position and current via look-up table. This paper proposes a new sensorless method for torque estimation. Principle of the proposed method is based on the determination of instantaneous phase torque from phase torque magnitude and torque sign. The phase torque magnitude is estimated by Artificial Neural Networks (ANN) using flux linkage and current while the torque sign is form the inductance vector angle. It is realized that waveforms of the torque magnitude and the inductance vector angle match well at zero levels. The inductance vector angle is obtained by applying ? - s transformation to the phase inductances. A switch connects one of the two constant sources (?1, 1) to the output according to the inductance vector angle and defines the torque sign. Multiplication of the phase torque magnitude and the torque sign results the phase torque. Thus, the phase torque is directly estimated without position sensor.