Professional Documents
Culture Documents
P 1120701001
P 1120701001
Direct Torque Control of Induction Motor Using Sophisticated Lookup Tables Based on Neural Networks
1
Y.V.Siva Reddy1, M.Vijayakumar2 and T. Brahmananda Reddy1 E.E.E Department, G. Pulla Reddy Engineering College, Kurnool, Andhra Pradesh, India 2 EE Department, JNTU, Anantapur, Andhra Pradesh, India [yvsreddy_123,mvk_2004,tbnr]@rediffmail.com the first sector from -30 up to 30.In order to overcome the draw backs of the classical DTC two different approaches of sophisticated lookup tables are presented [6]. The first approach is to use six sectors, as in classical DTC, but changing the zones. Here the zone of the first sector is 0 to 60 and is named as modified DTC (m_DTC).In the second approach the stator flux locus is divided into twelve sectors instead of six so all six active states will be used in each sector and is named as Twelve sector DTC (12-DTC). The neural network is well known for its learning ability and approximation to any arbitrary continuous function [7].It has been proposed in the literature that neural networks can be applied to parameter identification and state estimation of induction motor control system. A state selector emulator for classical DTC using neural network with different training algorithms has been reported in [8].In the present paper the neural network by using Levenberg-Marquardt training algorithm has been proposed to select the sector of operation for classical DTC, m_DTC and 12-DTC. The remainder of the paper is organized in detail as follows: Section (2) focuses on modelling of the induction motor. In Section (3) the conventional DTC is briefly reviewed. Section (4) deals with sophisticated lookup tables used for DTC. Section (5) emphasizes on the application of neural networks for direct torque control. Section(6) deals with the results and discussions. In section(7) some conclusions are drawn.
Abstract
Induction motor drive based on direct torque control (DTC) allows high dynamic performance to be obtained with very simple hysteresis control scheme. Direct control of the torque and flux is achieved by proper selection of inverter voltage space vector through a lookup table .In this paper apart from six sector look up table used for classical DTC, a modified look up table, which also use six sectors but with different zones and a twelve sector table are presented .This paper also presents the application of neural networks to control induction machines with DTC. Neural network is used to emulate the state selector of the DTC. In this paper Levenberg-Marquardt algorithm is used to train the neural network. Finally DTC is simulated with and without neural networks and results are compared. Keywords: Direct torque control, Neural Networks.
1. Introduction
High performance electric drives require decoupled torque and flux control. This control is commonly provided through field oriented control FOC [1].Although field oriented control schemes are capable of controlling the torque and flux of the induction machine independently, they are highly dependent on the parameters of the induction machine [2], its speed and hence the requirement of a shaft encoder which reduces the ruggedness of the machine. In contrast to FOC, direct torque control (DTC) [3-5] method requires the knowledge of stator resistance only, thereby decreasing the associated sensitivity to parameters variation and the elimination of speed information .DTC as compared to FOC is also advantages in other aspects like absence of co-ordinate transformation and PWM modulator. DTC is also very simple in its implementation because it needs only two hysteresis comparators and a lookup table for both flux and torque control. In classical Direct torque control [3-5] the position of the stator flux is divided into six different sectors ,the zone of
v ds = R s i ds + v qs = R s i qs +
p ds
p qs
0 = R r i dr + r qr + p dr 0 = R r i qr r dr + p qr
(1)
Where p indicates the differential operator (d/dt). The stator and rotor flux linkages are defined using their respective self leakage inductances and mutual inductance as given below
ds = L s i ds + L m i dr qs = L s i qs + L m i qr
dr = L r i dr + L m i ds qr = L r i qr + L m i qs
Te =
3P ( ds i qs qs i ds ) 2 2
(3)
vs =
d s dt
(4)
The above equation shows that the applied voltage space vector produces a stator flux variation which has the same direction of the voltage space vector and amplitude which is proportional to the voltage and time interval for which the voltage vector is applied. Fig. 1 shows the possible dynamic locus of the stator flux and its variation with respect to the VSI states chosen. The possible global locus is divided into six different sectors as signaled by the discontinuous line. Accordingly a six sector lookup table is derived as given in Table I.
V3 (010) II
V2 (110)
V3 (010) II
V2 (110) I V1 (100)
V0 (000) VI V6 (101)
V6 (101)
Fig.1 Stator flux vector locus and different possible switching vectors. Table I: Lookup table for classical DTC
Fig.2 Modified DTC and its switching vectors Table 2: Lookup table for modified DTC
FI
S1 V2 V0 V6 V3 V7 V5
S2 V3 V7 V1 V4 V0 V6
S3 V4 V0 V2 V5 V7 V1
S4 V5 V7 V3 V6 V0 V2
S5 V6 V0 V4 V1 V7 V3
S6 V1 V7 V5 V2 V0 V4
FD
TI T= TD TI T= TD
FI FD
TI T= TD TI T= TD
S1 V2 V0 V1 V4 V7 V5
S2 V3 V7 V2 V5 V0 V6
S3 V4 V0 V3 V6 V7 V1
S4 V5 V7 V4 V1 V0 V2
S5 V6 V0 V5 V2 V7 V3
S6 V1 V7 V6 V3 V0 V4
10
V2 (110) II I V1 (100)
VIII
V0 (000) XII XI IX X
V5 (001)
V6 (101)
Fig.3 Twelve sector modified DTC (12_DTC) and its switching sectors Table 3: Lookup table for twelve sector DTC Fig.5 Schematic diagram of DTC with neural networks
The screenshot of the Network/Data manager(GUI) and the generated neural network are given in Fig.6 and Fig.7 respectively.
Fig.6.Screenshot of GUI
The setup to teach the neural network to emulate the DTC state selector is shown in Fig. 5. The outputs of the torque and stator flux estimator are the torque, the stator flux magnitude, and angle. The torque and flux references are compared with their corresponding outputs from the estimator to determine the errors in torque and stator flux Ete and E, respectively. Furthermore these two errors signals are passed through hysteresis bands. The outputs of these two hysteresis bands plus the stator flux angle are the inputs to the neural network. The desired outputs are the switching states.
11
The Levenberg-Marquardt algorithm which is used to train the neural network is an approximation to the Hessian matrix in the following Newton-like update:
The algorithm used in this paper is ten to hundred times faster than the algorithm used in [6] and also takes less memory.
X k +1 = X k [ J T J + I ]1 J T e
When the scalar is zero, this is just Newton's method, using the approximate Hessian matrix. When is large, this becomes gradient descent with a small step size. Newton's method is faster and more accurate near an error minimum, so the aim is to shift towards Newton's method as quickly as possible. Thus, is decreased after each successful step (reduction in performance function) and is increased only when a tentative step would increase the performance function. In this way, the performance function will always be reduced at each iteration of the algorithm. As such there is no optimal value of . The training results of the MATLAB program for conventional, modified six sectors and twelve sectors are as shown in the Fig 8 to Fig.10.
Fig. 11 Simulation results of classical DTC; a 10 N-m load is applied at 0.7 sec and removed at 0.9 sec.
Fig .10 Mean square error vs Epochs for Twelve sector DTC
Fig.12 Transient responses during speed reversal for classical DTC; Speed is changed from +1300rpm to -1300rpm
12
Fig. 13 Simulation results of m DTC; a 10 N-m load is applied at 0.7 sec and removed at 0.9 sec.
Fig.16 Transient responses during speed reversal for DTC_12; Speed is changed from +1300rpm to -1300rpm
Fig.14 Transient responses during speed reversal for m_ DTC; Speed is changed from +1300rpm to -1300rpm Fig. 17 Simulation results of classical DTC with neural networks; a 10 N-m load is applied at 0.7 sec and removed at 0.9 sec.
Fig. 15 Simulation results of DTC_12; a 10 N-m load is applied at 0.7 sec and removed at 0.9 sec.
Fig.18 Transient responses during speed reversal for classical DTC with neural networks; Speed is changed from +1300rpm to -1300rpm
13
Fig. 19 Simulation results of m_DTC with neural networks; a 10 N-m load is applied at 0.7 sec and removed at 0.9 sec.
Fig.22 Transient responses during speed reversal for DTC_12 with neural networks; Speed is changed from +1300rpm to -1300rpm
For the simulation, reference flux is taken as 1wb and starting torque is limited to 40 N-m. The simulation results are shown in Fig. 11 to Fig. 16 for classical DTC, m_DTC and DTC-12 without neural networks respectively. The simulation results are shown in Fig. 17 toFig. 22 for classical DTC, m_DTC and DTC-12 with neural networks respectively From the simulation results it can be observed that, the results with and without neural networks are almost similar. But the stability of the system, an inherent phenomena of neural networks will be improved in case Neural network based DTC.
7. Conclusions
In this paper, a new training algorithm is proposed for DTC. The proposed algorithm gives very fast response. From the simulation results, it can be observed that the performance of the classical DTC and neural network based DTC schemes are almost similar. But owing to the stability and learning capability of neural networks, the proposed method can be considered as better technique.
Fig.20 Transient responses during speed reversal for m_DTC with neural networks; Speed is changed from +1300rpm to -1300rpm
8. References
[1] F. Blaschke, The principle of field orientation as applied to the new TRANSVECTOR closed loop control system for rotating field machines Siemens Review XXXIX, 1972, Vol No.5, pp. 217-220. [2] K.B. Nordin, D.W. Novotny and D.S.Zinger, the influence of motor parameter deviation in feedforward field orientation in conf. rec.IEEEIAS, 1984, pp 525-531. [3] Isao Takahashi, Toshihiko Noguchi, A new quickresponse and high-efficiency control strategy of an induction motor, IEEE Trans Ind Appl, Vol.IA-22, No.5, Sep/Oct, 1986, pp. 820-827. [4] M.Depenbrock, "Direct-self control (DSC) of inverter-fed induction machine", IEEE Trans. Power Electron., Vol.3, pp. 420-429, Oct.1988. [5] James N. Nash, Direct Torque Control, Induction Motor Vector Control Without an Encoder, IEEE Trans. Ind. Appl., Vol 33, pp. 333-341, March/April 1997.
14
Fig. 21 Simulation results of DTC_12 with neural networks; a 10 N-m load is applied at 0.7 sec and removed at 0.9 sec.
[6] Antoni arias Pujol, improvements in direct torque control of induction motors PhD thesis submitted to Universitat politechnica De catalunya. [7] B. Kosko, Neural Networks and Fuzzy Systems: A Dynamic Systems Approach to Machine Intelligence. Englewood Cliffs, NJ: Prentice-Hall, 1992. [8] Luis A. Cabrera, Malik E. Elbuluk, and Donald S. Zinger, Learning Techniques to Train Neural Networks as a State Selector for Inverter-Fed Induction Machines Using Direct Torque Control IEEE-PE, Vol. 12, No. 5, pp 788-799, Sep, 1997.
Dr. M. Vijaya Kumar graduated from NBKR Institute of Science and Technology, Vidyanagar, A.P, India in 1988. He obtained M.Tech degree from Regional Engineering College, Warangal, India in 1990. He received Doctoral degree from JNT University, Hyderabad, India in 2000. Currently he is working as Associate professor in Electrical and Electronics Engineering Department, JNT University, Anantapur, A.P, India. He is a member of Board of studies of S.V. University, Tirupathi and JNT University, Hyderabad, India. He has published over 40 research papers. He received two research awards from the Institution of Engineers (India). His areas of interests include Electrical Machines, Electrical Drives, Microprocessors and Instrumentation.
9. Bibliographies
Y.V. Siva Reddy received B.Tech and M.Tech degrees from JNT University, Anantapur in the year 1995 and 2000 respectively. He is currently pursuing Ph.D at J.N.T.University, Hyderabad. He is presently Associate Professor in the Electrical and Electronics Engineering Department, G. Pulla Reddy Engineering College, Kurnool. His research areas include Power systems and Electrical Drives.
T. Brahmananda Reddy graduated from Sri Krishna Devaraya University, Anantapur in the year 2001, M.E from Osmania University, Hyderabad in the year 2003. He is currently pursuing Ph.D at J.N.T.University, Hyderabad. He is presently Assistant Professor in the Electrical and Electronics Engineering Department, G. Pulla Reddy Engineering College, Kurnool. His research areas include PWM techniques, and Electrical Drives.
15