A Practical Method Of Speed Estimation Based On Neural Network

Sensorless vector control can eliminate the speed sensors which makes the variable-speed a.c. drives simple and reliable. For this reason, the sensorless vector control is becoming the popular research project. How to get the rotor speed accurately in sensorless vector control system is the key problem and many specialists and scholars have made some progresses on this point. After comparison of some speed estimation schemes, the thesis combines the advantages of artificial neural network and MRAS (Model Reference Adaptive System). Then the model of an induction motor with indirect rotor-flux-oriented control and MRAS speed estimator based on neural network have been done by using Matlab/Simulink. Four parts of research work are briefly listed as follows:Firstly, some simulation results of indirect rotor-flux-oriented vector control system prove that the model possesses good steady and dynamic performance. Then, this paper analyzes the effect of control parameters'variations of speed regulator on drive performance and gives simulation results which are helpful to the adjustment of vector control system. Secondly, in order to confirm the good operation performance of speed estimation model, the paper has done a whole simulation which includes both vector control model and speed estimation model. The simulation results show that the MRAS speed estimation model based on neural network has good estimation precision and steady and dynamic performance.Thirdly, the motor parameters (stator and rotor resistance, magnetizing inductance) will vary because of the change of environment temperature, iron loss and main flux saturation as the motor operates. The parameter variations will lead to the error of estimation. In order to verify the robust performance, the paper has simulated for the parameters variation situation respectively and the results show that the model is sensitive to the variation of stator resistance, however the influence of rotor resistance and magnetizing inductance is small.Finally, the good performance of speed model is proved once more with experimental data (stator voltage and current). The simulation results show that the estimation speed trace the actual speed very well except the oscillations in low speed area.