THE IMPLEMENTATION OF PERMANENT MAGNET SYNCHRONOUS MOTOR SPEED TRACKING BASED ON ONLINEARTIFICIAL NEURAL NETWORK

ISSN 1819-668 26-213 AsianResearch PublishingNetwork (ARPN).Allrights reserved. www.arpnjournals.co THE IMPLEMENTATION OF PERMANENT MAGNET SYNCHRONOUS MOTOR SPEED TRACKING BASED ON ONLINEARTIFICIAL NEURAL NETWORK N. M. Zin 1, W. M. Utoo 1,Z. A.Haron 1 1 Electrical Power Departent, Faculty of Electrical and Electronic Engineering, UniversitiTun Hussein Onn Malaysia, 864 Parit Raja, BatuPahat, Johor, Malaysia. adzianie@gail.co, wahyu@uth.edu.y,zainalal@uth.edu.y ABSTRACT This paper deals with the perforance analysis of the field oriented control for a peranent agnet synchronous drive syste with an artificial neural network proportional-integral-derivative for speed control in closed loop operation. Space vector pulse width odulation is used to generate the required stator voltage. The space vector pulse width odulation has the character of wide linear range, little higher haronic and easy digital realization. The field oriented control theory and space vector pulse width odulation technique ake the peranent agnet synchronous otor can achieve the perforance as well as a direct current otor. Therefore an online and offline learning of artificial neural network algorith is derived. The controller is designed to tracks variations of speed references and stabilizes the output for both systes. The effectiveness of the proposed ethod is verified by develop the syste in MATLAB-siulink progra and experiental by using Digital Signal Processing boardand interfacing DAQ with LabView software in order to recorded the result. The results show that the proposed online learning artificial neural network controller produce significant iproveent control perforance for controlling speed reference variations condition copared to offline learning artificial neural network syste. It can conclude that by using proposed controller, the settling tie and speed achieving can be iproved significantly. Key words: Peranent Magnet Synchronous Motor Online Neural Network Field Oriented Control INTRODUCTION The earliest power systes were d.c systes, but by the 198s a.c power syste were clearly winning out over d.c systes. Despite this fact, there were several reasons for the continued popularity of dc otors such as in which wide variations in speed are needed. For dc syste, the flux and torque can be controlled separately by eans of controlling the field and the arature currents respectively. In soe applications today, d.c electric otors are replaced by cobining an a.c electric otor with an electronic speed controllerbecause it is a ore econoical and less expensive solution. Moreover, d.c electric otors have any oving parts that are expensive to replace, and d.c electric otor repair is usually ore expensive than using a new a.c electric otor with an electronic controller. By these reason, Peranent agnet synchronous otors (PMSM) has been selected. PMSM are widely used in low and id power applications such as coputer peripheral equipent, robotics, adjustable speed drives and electric vehicles. Peranent agnet synchronous otor has the characteristics of high power density, free aintenances and high efficiency, which has been widespread application in the various electric drives applications (E. S. Sergaki et al. 28). Since 1988, Pillay, P and Krishnan, R. has been presented about PM otor drives and classified the into two types such as peranent agnet synchronous otor drives (PMSM) and brushless dc otor (BDCM) drives (P. Pragasen& R. Krishnan 1989). The PMSM has a sinusoidal back ef and fed with sinusoidal stator currents while the BDCM has a trapezoidal back ef and fed with direct currents. The PMSM is very siilar to the wound rotor synchronous achine except that the PMSM that is used for servo applications tends not to have any daper windings and excitation is provided by a peranent agnet instead of a field winding. The PM otor faily incorporates two designs: internal rotor and external rotor. Both designs are industrially rated and adopted in critical applications such as elevator winches and wind power generators. However, the ain drawbacks that ake a.c. otor retreats fro industry were the control between flux and torque are inherent coupling but this proble was aended by the exits of electronic control. So, Field Oriented Control (FOC) technique has been chosen for this syste.foc also known as decoupling or vector control, cae into the field of ac drives research in the late 196s and was developed proinently in the 198s to eet the challenges of oscillating flux and torque response in inverter fed induction and synchronous otor drive.in FOC, otor stator currents & voltages are anipulated in the direct-quadrature (d-q) reference frae of the rotor and it s a control procedure for operating the otor that results in fast dynaic response and energy efficient operation at all speeds. There are two ethods to achieve zero steady state error: switching and integration. To eliinate steady state error, a Proportional and Integral (PI) controller should be eployed (L.K. Wong et al. 1998). Other than that by using Proportional-Integral-Derivatives (PID) controller exact dq axis reactance paraeters can be obtained. Moreover, to step change of coand speed, paraeter variations and load disturbances is very sensitive. Since it is slightly siple to ipleent, PI and PID controller becoe ost widely used for PMSM. So, a real tie self-autoated hardware ipleentation of PID controller is desired (MohdMarufuzzaan et al. 21). 1

ISSN 1819-668 26-213 AsianResearch PublishingNetwork (ARPN).Allrights reserved. While, the artificial neural networks (ANN) are best suited for solving the probles that are nonlinear in nature. In ANN we can use parallel processing ethods to solve soe real-world probles where it is difficult to define a conventional algoriths. The ability of ANN to learn large classes of nonlinear functions is well known (Yang Yi et al. 23)( K.S. Narenda&K. Parthasarathy 199). It can be trained to eulate the unknown nonlinear plant dynaics by presenting a suitable set of input/output patterns generated by the plant ( JiangWeidong et al. 29 ). Once syste dynaics has been identified by using an ANN, any conventional control techniques can be applied to achieve the desired objective. In this paper, a odel of ANN closed-loop PMSM control syste that is controlled by SVPWM are develops for speed perforance in FOC PMSM drive.therefore an online and offline training of ANN algorith is derived. The controller is designed to tracks variations of speed references and stabilizes the output for both systes. The effectiveness of the proposed ethod is verified by develop the syste in MATLAB-siulink progra and experiental by using Digital Signal Processing board. FIELD ORIENTED CONTROL Dynaic Modeling of PMSM PMSM is essentially a three phase AC otor with sinusoidal back EMF driven by a DC source, which is converted to three-phase alternating currents supplying to the three stator windings of PMSM. The atheatic odel of PMSM i dq synchronous rotating reference frae can be obtained fro synchronous achine odel. Due to the constant field produced by peranent agnets, the field variation is zero. It is also assued that saturation and losses of core are negligible, the induced ef is sinusoidal and there is no daper winding on rotor. Using these assuptions, the voltage equations can write as follow: d d vd = Rsid+ Ld id Lqω e iq (1) dt dt d d vq= Rsiq+ Lq iq Ldωe id+ ωeλpm (2) dt dt The produced torque of the achine can be presented as follow: 3 Te= P [ λ PMiq+ ( Ld Lq ) idiq ] (3) 2 While, the axiu speed can be identified fro the relationship: d Te= TL+ Kfω + J ω (4) dt The update frequency of the control loops ust be high enough and the SVPWM should be properly configured to ensure sinusoidal currents applied to the stator windings. The paraeters for the PMSM are given as Table 1. Table 1: Paraeters of PMSM Motor Paraeter Value Frequency, f 5 Hz Pole, p 4 Stator Resistance, R s 2.875Ω d-axis Inductances, L d.85h q-axis Inductances, L q.85h Moent of Inertia, J.8kg² PM Flux Linkage,λ PM.175Wb Friction Coefficient, K f.38818 PMSM Drive Syste The operation of PMSM drive syste is based on the easure of two phase currents and of the otor position. The rotor position feedback is necessary to generate the reference speed. In this case, increental encoder (25 pulse per revolution)has been attached. The easured phase currentsi a and i b are transfored into the stator reference frae coponents i alpha and i beta. Then, based on the position inforation, these coponents are transfored into the rotor frae direct and quadrature coponents i d and i q. The speed and current controllers are PID discrete controllers. The inverse coordinates transforation is used for the coputation of the phase voltages references, V a, V b and V c, applied to the inverter, starting fro the values of voltage references coputed in the d and q reference frae (V d, V q ). Thus, the 6 full-copare SVPWM outputs of the DSP controller are directly driven by the progra, based on these reference voltages. The code is developed only in C language, both for the ain structure of the application and for the tie-critical parts (as controllers, coordinates transforation, etc) The direct current coponent reference i d is set to because of the case corresponding to the otion of the otor in the noral speed range, without considering a possible field weakening operation.figure 1 is the diagra of current control loop using FOC technology based on proposed ANN speed controller. Figure 1: PMSM drive syste with ANN controller Proposed ANNController Structure To design the neural network control soe inforation about the plant is required. Basically, the nubers of input and output neuron at each layer are equal to the nuber of input and output signals of the syste respectively. Further the nuber of hidden layers and the total neurons is depended on the coplexity of the syste and the required training accuracy. To ipleent search efficiency optial control of PMSM drive, a ultilayer 2

ISSN 1819-668 26-213 AsianResearch PublishingNetwork (ARPN).Allrights reserved. perceptron neural network control is developed. Based on the type of the task to be perfored, the structure of the proposed ANN speed controller is shown in Figure 2 (NooradzianieMuhd. Zin et al. 213) Figure 2: Block diagra of ANN controller for PMSM drive syste. The controller consists of input layer, hidden layer and output layer. Based on nuber of the neuron in the layers, the ANN is defined as a 1-3-1 network structure. The first neuron of the output layer is used as a torque reference signal (a 2 1= f ). The connections weight paraeter between j th and i th neuron at th layer is given by w ij, while bias paraeter of this layer at i th neuron is given by b i. Transfer function of the network at i th neuron in th layer and output function of neuron at th layer is defined by: S 1 = 1 i wij a j + bi j= 1 n (5) The output function of neuron at th layer is given by: a i = f ( ni ) (6) Where fis activation function of the neuron. In this design the activation function of the output layer is unity and for the hidden layer is a tangent hyperbolic function given by: 2 f ( ni ) = 1(7) 2n 1+ e i Updating of the connection weight and bias paraeters are given by: paraeters which are difficult to define and vary against with environent. The training process iniizes the error output of the network through an optiization ethod. Generally, in learning ode of the neural network controller a sufficient training data input-output apping data of a plant is required. Since the otor paraeters of the PMSM drive vary with teperature and agnetic saturation, the online learning Back propagation algorith is developed. Based on first order optiization schee, updating of the network paraeters are deterined. The perforance index su of square error is given by: 1 2 F = e i (1) 2 i ei = ti ai (11) where:t i is target signal a i output signal on last layer. The gradient descent of the perforance index against to the connection weight is given by: = wij wij The sensitivity paraeter of the network is defined as: si si (12) = (13) ai ai = (14) Gradient the transfer function again to the connection weight paraeter is given by: 1 = a i wij (15) Fro substitution equation (13) and (15) into (8) the updating connection paraeter is given by: 1 i 1 wij ( k + 1) = wi α si ai (16) F ( k ) w ( k + 1) = w ( k ) α ij ij w ij (8) With the sae technique the updating bias paraeter is given by: bi k+ 1) = bi ( k) bi ( α (9) wherek is sapling tie, α is learning rate, and F perforance index function of the network. Online Schee of the Proposed ANN After the neural network architecture is odelled, the next stage defines the learning odel to update network paraeters. By this learning capability, it akes the ANN suitable to be ipleented for the syste with otor 1 i bi ( k+ 1) = bi αsi (17) RESULTS AND DISCUSSION The proposed odel has been developed by Matlab/Siulink. The siulation block diagra for the proposed PMSM drives syste with ANN is shown in Figure 3. The siulation block diagra has been created in order to download all the proposed syste into DSP board. For the experiental set up, all the ain circuit including of 3 inputs 6 outputs gate driver, inverter and current sensor (odel: ACS756KCA-5B-PFF-T) has been built. All the connection between PMSM and all the circuit has been connected as shown in Figure 4. The 3

ISSN 1819-668 26-213 AsianResearch PublishingNetwork (ARPN).Allrights reserved. hardware is interfacing with syste by using TMS32F28335 DSP controller that the progra will be downloaded. While, the PMSM equipped with 5-line quadrature increental encoder (25 pulse per revolution) is used.results for each testing wererecorded by interfacing data acquisition (DAQ)with LabView software. 1 reference speed 2*pi/6 F28335 ezdsp C28x/C28x3x B dac_ac_noyanew1 ADC convert to rad/ec2 NN Controller C28x3x GPIOx GPIO DI Digital Input 2*pi/6 convert to rad/sec1 w ref f pulse_counting1 S-Function2 convert to rp 24 id_ref1 double foc26 Alf* Bet* an bn cn sv pw1 C28x GPIOx GPIO DO GPIO C28x GPIOx double GPIO DO GPIO 2 C28x GPIOx GPIO DO GPIO 4 Figure 3: The siulink block diagra of the proposed PMSM drive syste with ANN. DSP Board Figure 4: Experiental setup for proposed PMSM drive syste with ANN In order to verify the validity of the proposed PMSM drive syste with ANN, both online and offline ANN syste has been testfor a variety of speed. Different operating speed is tested, which is constant speed reference 1rp, step up speed reference is varying fro 4rp to 9rp and step down speed reference is varying fro 9rp to 6rp for both systes. All the results are shown in Figure 5 to Figure 1. 1 Gate Driver Inverter Current Sensor PMSM Speed (rp) 1 8 6 4 2 1 2 3 4 5 6 7 8 9 1 Tie (sec) Figure 6: PMSM drive syste with offline ANN for constant speed reference 1rp Fro the resultsin Figure 5 and Figure 6, it shows that by using an online ANN speed controller produced a better start-up perforance copare to the offline ANN speed controller where the settling tie ore faster than offline speed controller in achieving desired output speed. The settling tie for online ANN speed controller is 1.6sec while 2.74sec of settling tie for offline ANN speed controller. Moreover in speed achieving for both syste, an online ANN speed controller achieved 99rp fro the 1rp speed reference while offline ANN speed controller achieved 98rp fro the 1rp speed reference.based on result fro figure 5 and Figure 6, thedifferencepercentage for settling tie between offline ANN and online ANN is 52.54% iproved. Meanwhile, the rppercentage is 1.2% iproved. Speed (rp) 1 8 6 4 2 1 11 12 13 14 15 16 17 18 19 2 Tie (sec) Figure 7: PMSM drive syste with online ANN for step up response fro 4rp to 9rp Speed (rp) 1 8 6 4 8 2 Speed (rp) 6 4 2 1 2 3 4 5 6 7 8 9 1 Tie (sec) Figure 5: PMSM drive syste with online ANN for constant speed reference 1rp 1 11 12 13 14 15 16 17 18 19 2 Tie (sec) Figure 8: PMSM drive syste with offline ANN for step up response fro 4rp to 9rp Refer to the Figure 7 and Figure 8, it is also shows that by using an online ANN speed controller produced a better step up perforance copared to the offline ANN speed controller where the settling tie ore faster than 4

ISSN 1819-668 26-213 AsianResearch PublishingNetwork (ARPN).Allrights reserved. offline speed controller in achieving desired output speed. Their settling tie is1.32sec and 2.14sec respectively. It is also sae in speed achieving perforance which is an online ANN speed controller achieved 39rp and 89rp fro the speed reference 4rp and 9rp copared to offline ANN speed controller which is reduced 1rp respectively fro online ANN speed controller in speed achieving. Difference percentage between offline ANN and online ANN for the result of settling tie is 47.4% iproved and speed is 2.6% iproved for 4rp reference speedand 1.13% iproved for 9rp reference speed. An online ANN can adapt variety condition because of their syste always update the paraeter even though one of the paraeter during testing such as changing of teperature. In other words the weight and bias is updated together with the testing process. Speed (rp) 1 8 6 4 So, for the result difference percentage for constant speed 1rp, step up response and step down response can be conclude that by using online ANN will be 1.35% iproved in average for speed achieving. While, 58.47% iproved in average for settling tie. CONCLUSION This paper has presented the odelling and hardware ipleentation of the field oriented control for PMSM drive syste using online and offline neural network controller. The effectiveness of the proposed ethod is verified by develop the syste in MATLABsiulink progra and experiental by using Digital Signal Processing board and interfacing DAQ with LabView software in order to recorded the result. The results show that the proposed an online ANN controller produce significant iproveent control perforance for controlling speed reference variations condition copared to offline ANN syste especially for settling tie which is 58.47% iproved in average While 1.35% iproved in average for speed achieving. It can conclude that by using proposed controller, the settling tie and speed achieving can be iproved significantly. Speed (rp) 2 2 22 24 26 28 3 32 34 36 38 4 Tie (sec) Figure 9: PMSM drive syste with online ANN for step down response fro 9rp to 6rp 1 8 6 4 2 2 22 24 26 28 3 32 34 36 38 4 Tie (sec) Figure 1: PMSM drive syste with offline ANN for step down response fro 9rp to 6rp While, by referring to the Figure 9 and Figure 1, it is also shows that by using an online ANN speed controller produced a better step down perforance copared to the offline ANN speed controller where the settling tie ore faster than offline speed controller in achieving desired output speed. Their settling tie is.66sec and 1.46sec respectively. It is also sae in speed achieving perforance which is an online ANN speed controller achieved 89rp and 59rp fro the speed reference 9rp and 6rp copared to offline ANN speed controller which is reduced 1rp and 5rp respectively fro online ANN speed controller in speed achieving.difference percentage between offline ANN and online ANN for the result of settling tie is 75.47% iproved and speed is 1.13% iproved for 9rp reference speed and.85% iproved for 6rp reference speed. REFERENCES Sergaki, E.S. et al., 28. Fuzzy Logic based OnlineElectroagnetic Loss Miniization of Peranent Magnet Synchronous Motor Drives.In ICEM 28.18 th International Conference.pp.1-7, 6--9. Pillay P. & Krishnan R. 1989.Modeling, Siulation, and Analysis of Peranent-Magnet Motor Drives, Part 1: The Peranent-Magnet Synchronous Motor Drive. In Industry Applications, IEEE Transactions, pp. 265--273. Wong L.K. et al. 1998.Cobination of Sliding Mode Controller and PI Controller using Fuzzy Logic Controller. In IEEE International Conference on Fuzzy Syste, vol.1, pp. 296--31. MohdMarufuzzaan et al., 21. FPGA Ipleentation of an Intelligent Current Dq PI Controller for FOC PMSMDrive. In International Conference on CoputerApplications and Industrial Electronics (ICCAIE), pp 62. Yang.Yi et al., 23. Ipleentation of an Artificial neural network based real tie adaptive controller for an Interior PMSM. In IEEE Transaction On Industry Application, vol. 39, pp. 96--13. Narenda K.S. &Parthasarathy K. 199.Identification and Control of Dynaical Systes Using Neural Networks. In IEEE Transaction Neural Network, pp. 4--27. Jiang Weidong et al., 29. SVPWM Strategy for Three Level Inverter based on SVPWM Strategy for Two-Level Inverter. In Transactions of China Electrotechnical Society, vol. 24, No.1, pp. 18--114. 5

ISSN 1819-668 26-213 AsianResearch PublishingNetwork (ARPN).Allrights reserved. NooradzianieMuhdZin&WahyuMulyoUtoo et al.,213. Speed Control of Peranent Magnet Synchronous Motor using FOC Neural Network. In Inforation Technology Convergence, Lectures Notes in ElectricalEngineering, Vol. 253, pp 295-- 33. WahyuMulyoUtoo&NooradzianieMuhdZin et al. 214, Speed Tracking Field Oriented Control of Peranent Magnet Synchronous Motor using Neural Network.In International Journal of Power Electronics and Drive Systes, IJPEDS, pp 29--297. 6