Position Sensorless Control For BLDCM Based On Modified Neural Network

Brushless DC motor is a kind of electric energy/kinetic energy conversion device with simple principles,compact structure,high efficiency,stability and reliability,which is widely used in industrial control,aerospace,agricultural technology,communication,medical and other fields.The brushless DC motor removes the commutator in the brushed DC motor and uses the sensor to detect the rotor position for commutation.However,due to the existence of the sensor,the volume of the motor and the development cost are increased to a certain extent,which limits its application field.Therefore,studying brushless DC motor control technology without position sensor is a very challenging and practical work.When there is no position sensor,the rotor position signal cannot be directly measured.Only the indirect detection method can be used.The rotor position signal is obtained by signal mapping such as voltage and current closely related to the position signal.This is a typical nonlinear system identification process.Among many system identification methods,artificial neural networks have become a very important system identification tool due to good self-learning,self-organizing ability,and the ability to implement nonlinear mapping at arbitrary precision.It is also a focus to study how to use artificial neural network to realize sensorless control of brushless DC motor.In this paper,we first refer to a large number of domestic and foreign literatures related to the position sensorless control of brushless DC motor,and then analyze the system structure and working principle of the brushless DC motor in detail,and then establish the differential equation and transfer function of the motor.On this basis,the principle of using the artificial neural network to realize the position sensorless control of the brushless DC motor is clarified.Secondly,based on the detailed analysis of the traditional position-free sensor control methods,such as,back EMF and flux linkage estimation,a nonlinear identification model between motor phase voltage,phase current and power device commutation signal is established by using RBF(Radial Basis Function)neural network.At the same time,the network structure and parameter optimization method based on K cross-validation principle is proposed,and the adaptive adjustment strategy of learning rate in network training algorithm is proposed.The simulation experiments under MATLAB/Simulink platform show that the sensorless control method of brushless DC motor based on improved RBF neural network has better performance with accurate commutation signal,no hysteresis and good speed regulation.Finally,based on the structural principle of the brushless DC motor system and the RBF neural network identification model,the hardware system with TMS320F2812 as the control core is designed.The design of device selection,drive circuit,detection circuit and anti-interference measures are introduced in detail.At the same time,this paper briefly introduces the design idea of embedded software program and gives the flow chart.The experimental results show that the hardware system designed in this paper can get the commutation signal more accurately,and the software program runs smoothly and meets the design requirements.