Department of Electrical Engineering Southern Taiwan University Sensorless Control of the Permanent Magnet Synchronous Motor Using Neural Networks Student:

Department of Electrical Engineering Southern Taiwan University

Department of Electrical Engineering Southern Taiwan University

Sensorless Control of the Permanent MagnetSynchronous Motor Using Neural NetworksSensorless Control of the Permanent MagnetSynchronous Motor Using Neural Networks

Student: Chun-Yi Lin Adviser: Ming-Shyan Wang Date : 24th-Jun-2011

100%製作

1,2Department of Electrical and Electronic Engineering, Fırat University 23119 Elazığ, Turkey

2Department of Electrical Engineering Southern Taiwan UniversityDepartment of Electrical Engineering Southern Taiwan University

OutlineOutline

Abstract

I. INTRODUCTION

II. RECURRENT BASED ROTOR POSITION ESTIMATION

A.PMSM MODEL

B. DESCRIPTION OF NEURAL BASED ROTOR ANGLE OBSERVER

III. STRUCTURE AND TRAINING OF NEURAL-NETWORK OBSERVERS

IV. SIMULATIONS RESULTS

V. CONCLUSIONS

References


AbstractAbstract

In this paper, a neural network based rotor position control and speed estimation method for Permanent Magnet Synchronous Motor (PMSM) is proposed.

The proposed method has three recurrent neural networks. They are used for estimating stator current, rotor speed, and rotor position angle.

Each of them is trained in two steps: off-line training for learning dynamic of PMSM and on-line training for realizing parameter adaptation of PMSM. Sensorless control of the permanent magnet synchronous motor using neural networks is simulated in MATLAB/Simulink.


I. INTRODUCTIONI. INTRODUCTION

In permanent magnet synchronous motor (PMSM) rotor positional sensors, which are connected to shaft, and these sensors brings some disadvantages in PMSM applications. These disadvantages are:

• High accuracy sensors are more expensive and system costs increase by using them.• Accuracy of sensors is limited due to environmental factors such as temperature, humidity, and dirt.• Adding friction to shaft reduces ruggedness of drive and forms a fault source.

It is desired to eliminate rotor position sensors and instead of them, new different techniques have been developed for sensorless control.

The estimation method has three recurrent neural networks. One of them is used to estimate rotor speed, and the other is used to estimate stator current.


Fig. 1. Structure of recurrent neural network


II. RECURRENT BASED ROTOR POSITION ESTIMATIONII. RECURRENT BASED ROTOR POSITION ESTIMATION

A.PMSM MODEL:

Mathematical model of PMSM are:



And,



Where, λm, R, Lss, L1 and τ are permanent magnet flux constant phase resistance, self inductance, leakage inductance and electrical time constant of machine respectively.

PMSM model which is transformed to rotating reference frame can be given as in (8).



B. DESCRIPTION OF NEURAL BASED ROTOR ANGLE OBSERVER

Neural network based sensorless control model of PMSM is shown in Fig.2. In the block diagram, it can be seen that control model has two artificial neural network blocks which are used to estimate rotor speed and stator current.





III. STRUCTURE AND TRAINING OF NEURAL-NETWORK OBSERVERSIII. STRUCTURE AND TRAINING OF NEURAL-NETWORK OBSERVERS

In case of using two hidden layer composed of 9 neurons for the neural current observer within the system, better results were obtained in the experiments. Also output layer consists of two neurons.

Speed observer has two hidden layers which consist of 8 and 6 neurons. Position estimate observer includes a single hidden layer consisting of 7 neurons. Tansigmoid transfer function in the hidden layers of the three observers within the system, and linear transfer function in the output layers were used.



IV. SIMULATIONS RESULTSIV. SIMULATIONS RESULTS

Sensorless simulation model of permanent magnet synchronous motor was implemented by MATLAB 6.5 program and shown in the figure…

The model included PI controller, PWM inverter, Permanent magnet synchronous motor, axis transformation blocks, and neural network blocks.








ConclusionsConclusions

In this paper proposed neural network observer with control system for estimating rotor position of PMSM is simulated in MATLAB/Simulink.

By this method, it is possible to eliminate many error calculations and submitting constant values which are necessary in other methods.

In addition, it is possible to estimate speed and position in a very large speed range in high accuracy by the proposed method.


ReferencesReferences[1] D. Yousfi, M. Azizi, A.Saad, “Sensorless Position and Speed Detection for Permanent Magnet Synchronous Motor”, IPEMC 2000, Vol.3, p.1224-1229

[2] T. Senjyu, T. Shimabukuro, K. Uezato, “Position Control of Permanent Magnet Synchronous Motors without Position and Speed Sensors” Industrial Automation and Control: Emerging Technologies, IEEE conference, 1995, p.182-186

[3] S. Ogasawara, H. Akagi, “An Approach Position Sensorless Drive Brushless dc Motors”, IEEE Transactions on Industry Applications, Vol.27, No.5, 1991, p.928-933

[4] M. Schroedl, “Sensorless Control of Permanent Magnet Synchronous Machines. An Overview”, EPE-PEMC, Tagungen, Riga, 2004

[5] J. Hu, D. Zhu, B. Wu, “Permanent Magnet Synchronous Motor Drive without Mechanical Sensors”, CCECE, IEEE, 1996, p.603-606

[6] Y. Li, L. Jiang, “Sensorless control of PMSM with an adaptive observer”, EPE’99, Lausanne, 1999, p.1-6.

[7] G. Qingding, L. Ruifu, W. Limei, “A Shaft Sensorless Control for PMSM Using Direct Neural Network Adaptive Observer”, Industrial Electronics, Control, and Instrumentation, IEEE IECON 22nd International Conference ,1996, p.1729-1734


Thank you for your attention.

Department of Electrical Engineering Southern Taiwan University Sensorless Control of the Permanent Magnet Synchronous Motor Using Neural Networks Student:

Documents