Modeling And Position Sensorless Control Of Switched Reluctance Machine Based On Neural Networks

The Switched Reluctance Drive (SRD) is favored in a lot of industrial applications due to its low cost in mass production, reduced maintenance requirements, rugged behavior and large torque output over very wide speed range. Because of the double salient structure of the machine, the flux linkage is a nonlinear function of both the rotor position angle and phase current, which brings difficulties in accurate modeling of the machine. The requirement of mechanical position sensor in traditional control methods not only increases the cost and complexity of the structure, but also weakens the reliability of the system. As a result, accurate modeling of the machine and position sensorless control of the system is drawing more and more attention at present.This paper presents a flux linkage characteristic measurement system for switched reluctance machine using DSP TMS320F2812. Phase currents and voltages are sampled and converted using the internal ADC module, then the data is transferred to PC, which calculates the flux linkage using digital method and draws the flux linkage/current curves at different rotor positions. Based on the measurement results, a nonlinear model from phase current, rotor position to flux linkage is constructed using the wavelet neural network (WNN). The comparison shows that there is tiny error between the estimated and measured results and the WNN flux model has a good adaptability.This paper presents a new approach to the position sensorless control of switched reluctance machine based on wavelet neural networks. Two wavelet neural networks with different parameters are constructed to switch on and turn off each phase respectively. The WNNs form a very efficient nonlinear mapping structure from phase currents, flux linkages to commutation signals with currents, flux linkages as inputs and switching signals as outputs, therefore the commutation signals can be obtained by manipulation of the WNNs' outputs. After trained by the data acquired from the system with position sensor, the WNNs replace the position sensor and make SRM switch to position sensorless operation. The simulation and experiment results show that there is tiny error of commutation signals between estimation and reality. SRM can operate with little torque fluctuation and slight speed vibration. This paper introduces an experiment control system of switched reluctance machine based on DSP TMS320F2812. All feasible control methods of the machine are tested with the high-powered processor which is known by excellent digital signal processing and real-time control abilities. The hardware system not only provides training samples for offline learning of wavelet neural networks, but also supplies good research foundation for position sensorless control of switched reluctance machine.