Sensorless position detection using neural networks for the control of switched reluctance motors

For high performance position or torque control, or for many of the different possible approaches to torque ripple and acoustic noise reduction in a switched reluctance motor (SRM), position feedback is essential. However, optical position encoders add to the complexity and cost of SRMs, compromising some of their main advantages. The paper describes a novel method of sensorless position detection requiring no special converter or sensor circuitry, and which does not rely on accurate prior knowledge of the magnetic characteristics of the motor. The approach described is novel in two respects. Firstly, it does not rely on accurate prior knowledge of phase winding inductance but merely makes the assumption that it varies substantially as sin(N_rθ), where N_r is the number of rotor poles and θ is rotor angle. Secondly, the approach learns from previous good estimates of position and, once it has done so, makes use of this knowledge where performance of the basic estimation algorithm degrades (principally at low speeds of rotation). The technique has been investigated in simulation and a hardware implementation is under development