Machine learning based SRM control using FPGAs for torque ripple minimization

Switched Reluctance Motor (SRM) is an attractive candidate for electric vehicles and much research is ongoing on motor and its control particularly in the last decade due to advancement in power electronics. Non-linear characteristics of SRM have been a challenge for control engineers for SRM control implementation. In such cases, learning models such as Artificial Neural Network (ANN), fuzzy logic, genetic algorithms etc., are helpful which are easier in terms of the programming and computation. One of the challenge for SRM application is its inherent torque ripple, which can affect drivability, vehicle dynamics and acoustics. Existing approaches for minimizing the ripple are based on standard normal Torque Sharing Function (TSF). In this paper, a novel approach is proposed to bring further reduction in the torque ripple in base, medium and high speed of the motor and also the implementation of this control algorithm is discussed. Two pre-trained ANN models are implemented one for providing run-time torque feedback and another one for calculating instantaneous reference currents for each phase. The novelty of this proposed approach is in providing a dynamic limiter for instantaneous current, based on the speed and torque error, in real time. The machine-learning model is implemented as a sequential logic on Xilinx Zynq-Ultrascale SoC Field Programmable Control Arrays (FPGA). The complete motor control is realized on the Programmable Logic (PL) part of the FPGA. FPGA utilization, along with timing diagram is evaluated for the ANN implementation, which could be used as a basis for selection of FPGAs.