Research On Flexible Neural Network Control Approach For Switched Reluctance Motor

The switched reluctance motor drive system (SRD) has been receiving attention for electrical drive applications due to its low cost in mass production, reduced maintenance requirements, rugged behavior and large torque output over very wide speed range. However, because of the double salient pole structure of its stator and rotor and special power supply way, rotor position sensors and torque ripple are often cited disadvantages of the switched reluctance motor (SRM) as the mechanical position sensors add to the cost, complexity and potential unreliability at high speed and obvious torque ripple has existed when the motor is supplied by the conventional square pulse power. So position sensorless operation and torque ripple minimization control become main topics in the research field of SRM.Because of its high nonlinear electromagnetism characteristic, the accurate model of SRM is hard to be accomplished. Artificial neural network (ANN) technology has made a great progress, which gives an efficient method for the modeling of nonlinear system. However, the methods based on conventional ANN can not meet the requirements of real-time control in complex systems because they have some inefficiencies, such as large and complex structure, low learning speed and easily landing into local minimum points. Flexible neural network (FNN) which has simple net structure, quick learning speed and strong generalizing ability compared with the conventional ANN can be used in the intelligent control approach in SRM.At the beginning, with the study on basic theory and nonlinear model of SRM, the paper establishes the dynamic simulation model of the 4-phase (8/6 poles) SRM and its drive system based on the environment of MATLAB/SIMULINK, and then the research on intelligent control approach by using FNN is processed on it. In the research of sensorless control, a FNN is built and trained offline to estimate the rotor position through measurement of the phase flux linkages and phase currents thereby facilitating elimination the rotor position sensor. In the research of torque ripple minimization control, another FNN is built for the estimation of the reference currents with a desired torque and rotor position, then the real currents in the armatures are adjusted according to the reference values, therefore the torque ripple generated by the non-ideal current waveforms is minimized. The simulation results illustrate that the learning speed of FNN is much quicker than the conventional ANN, the sensorless control in SRM is realized by the accurate estimation of the rotor position and the torque ripple is reduced efficiently based on the FNN control approach. The experiment system of SRM based on DSP TMS320F2812 can operate reliably and stably.