Switched Reluctance Motor Torque Ripples Reduction by the Aid Of Adaptive Reference Model

This paper deals with SRM torque ripples reduction. They are one of the major obstacles for SRM wide spread application. Direct torque feedback is the best solution for the case but due to the price it is quite unacceptable. Feedback with neural network torque estimator reduces torque ripples up to 2.5 times comparing them to the opened system but in many cases this reduction is insufficient. That is why very precise parallel running neural network motor model has been suggested. This model copies the behavior of the SRM and can be exploited as an observer of different state parameters including the dynamic torque. Applying this model, it has been discovered that certain overlapping of the phase voltages can reduce 10 times torque ripples comparing them to the regular opened system. In fact, this adaptive reference model is fully implementable by means of parallel running algorithms embedded in middle sized FPGA.