Avalable onlne at www.scencedrect.com Systems Engneerng Proceda 3 (22) 259 267 The 2 nd Internatonal Conference on Complexty Scence & Informaton Engneerng The PWM speed regulaton of DC motor based on ntellgent control Wenbn Yan a, Dada Wang b, Pengfe Ja c, Weguo L c, a* a Graduate Workstaton of North Chna Electrc Power Unversty & Yunnan Power Grd Corporaton,Kunmng,6527,Chna b Yunnan Electrc Power Test & Research Group CO,LTD Electrc Power Research Insttute,Kunmng,6527,Chna c North Chna Electrc Power Unversty,Bejng,226,Chna Abstract The PWM speed regulaton of DC motor based on ntellgent control s dscussed. The smulaton s carred out wth the SIMULINK after that the mathematcal model of controlled object s bult. Ths artcle ntroduces the PWM bpolar drve of DC motor, desgns a fuzzy controller and a neutral network controller and then dscusses the applcaton of artfcal ntellgence n the speed regulaton of DC motor. 2 Publshed by Elsever Ltd. Selecton and and peer-revew under under responsblty responsblty of of Desheng Desheng Dash Dash WuWu. Open access under CC BY-NC-ND lcense. Key Words: speed regulaton of DC motor; PWM bpolar speed regulaton; fuzzy PID control; neutral network control. Introducton DC motor s wdely used n metallurgy, machnery manufacturng and lght ndustry because of ts good performance n startng and breakng and ts easly controlled speed regulaton. In recent years, wth the development of the power electronc technology, the thyrstor rectfer s commonly used for the power supply of the DC motor, wtch replaces the AC motor DC generator power supply system. But DC motor speed control system s a complex multvarable nonlnear control system, because the varous parameters nfluence each other, t s ant nterference ablty s weak and t s not sutable for hgh control performance occason. Therefore, n order to enhance DC motor speed control system of ant jammng and robustness, and mprove the response speed and stable precson of the speed regulaton system, ths paper dscuss the PWM DC motor speed control system based on the fuzzy control and neural network control. a * Correspondng author. Tel.: +86 35 88477; fax: +86 87 63457. E-mal address: 3588477@39.com. 22-389 2 Publshed by Elsever Ltd. Selecton and peer-revew under responsblty of Desheng Dash Wu. Open access under CC BY-NC-ND lcense. do:.6/j.sepro.2..28
26 Wenbn Yan et al. / Systems Engneerng Proceda 3 (22) 259 267 2. DC motor speed control system smulaton model In Fg., t shows the smulaton model bult wth MATLAB/SIMULINK, n wtch the ASR s the speed controller, the ACR s the armature current controller, the PWM module provdes requred PWM wave for the dual polarty H brdge[]. A 5-HP DC motor of 24-V ratng 22 rpm s used n the smulaton models. The equvalent crcut parameters of the DC motor used n the smulaton are R = 24Ω, L = 2H, R =.6Ω, L = 2mH [2]. F F A A + g A + - v.4 TL m Gan w - B A+ dc A- Ia node F+ F- node If Out In Te PWM Clock InOut InOut - ASR ACR.83.2s+ Transfer Fcn Gan Contnuous powergu Fg.. SIMULINK modle of the DC motor speed control system In Fg.2, t shows the nternal structure of the PWM model and the ACR model.[3] In Sgnal(s) PWM Generator Pulses In kpn Saturaton Out Out Sgnal(s) Pulses Selector. Constant PWM Generator s G-acr I-acr Fg.2. The nternal structure of the PWM model and the ACR model 3. Ordnary PID controller As s known to all, the tradtonal PID control s a mature and wdely used engneerng control method. On condton that the structure and parameters of the lnear tme-nvarant system are known, t has good control performance, and ts algorthm s smple and t s easy to realze. The adjustment object of the PID
Wenbn Yan et al. / Systems Engneerng Proceda 3 (22) 259 267 26 controller s the system error, t s a knd of scale, ntergral, dfferental adjustment rules, and ts equaton s: t de() t ut () = KPet () + KI etdt () + K D () dt In the equaton, K P, K I, KD are the parameters of the PID controller, et () s the devaton nput sgnal of the controller, ut () s the control sgnal. In Fg. 3, t shows the smulaton model of the ordnary PID controller. In Out -Kkpn G-asr s I-asr Fg.3. The smulaton model of the ordnary PID controller 4. Fuzzy Controller Fuzzy control s a knd of computer ntellgent control based on the fuzzy set theory, the fuzzy language varables and the fuzzy logc. The basc concept s proposed by the famous professor of the unversty L.A.Zadeh. After over 2 years development, t makes a great success n the fuzzy control theory. Fuzzy controller s also called Fuzzy Logc Controller. Because the fuzzy control rules are descrbed by the fuzzy condtonal statement of the fuzzy theory, t s a knd of language controller, so t s also called Fuzzy Language Controller.[4] The composton of the fuzzy controller s showed n Fg. 4: Data Base Rule Base Knowledge Base Input\ Fuzzfcaton Interface Reasonng Defuzzfcaton Interface Output Fg. 4 The composton of the fuzzy controller
262 Wenbn Yan et al. / Systems Engneerng Proceda 3 (22) 259 267 4.. Fuzzfcaton Interface The fuzzfcaton of the nput of the fuzzy controller s mportant so that t can be used for solvng the control output, so t s actually the nput nterface of the fuzzy controller. Its man effect s puttng the true ndeed quanttatve nto a fuzzy vector. In ths case, t s a sngle varable 2D fuzzy controller. The fuzzy set of error E, error rate EC and control quantty u s descrbled as: { } e = NB,,,,,, NM NS Z PS PM PB 3,-2,-,,,2,3 The doman of dscourse of E and EC s: { } The doman of dscourse of u s: { 4.5,-3,-.5,,.5,3,4.5} (2) 4.2. Knowledge Base Knonledge base conssts of Data base and Rule base. The data base conssts of all membershp vector value of all nput and output varables fuzzy subsets. If the doman s a contnuous doman, t s a membershp functon. In solvng the fuzzy relaton equaton of rule reasonng, t provdes data to the reasonng machne. The rule base conssts of all the rules of the fuzzy control. In reasonng, t provdes control rules to the reasonng machne. The number of the rules s concerned wth the fuzzy subsets dvson of the fuzzy varables. The more fuzzy subsets, the more rules, but t does not represent that the accuracy of the rule base s hgher. The accuracy of the rule base s also concerned wth the accuracy of the expert knowledge. 4.3. Reasonng In fuzzy control, reasonng s a part that uses nput fuzzy quantty and the fuzzy control rules to complete fuzzy nference and solve fuzzy relatons equaton, and also get fuzzy control volume. In fuzzy control, consderng the reasonng tme, a smple method of reasonng s commonly used. Foregong fuzzy control rules can be descrbled by the fuzzy rule table (Table. ), there s 49 fuzzy rules, and the relatonshp between the varous fuzzy statement s or. The fuzzy rules that the table above shows can be expressed as follows: R:IF E s NB and EC s NB then U s PB R2:IF E s NB and EC s NS then U s PM Table. The fuzzy rule table ec u e NB PB PB PM PM PS Z Z NM PB PB PM PM PS Z ZS
Wenbn Yan et al. / Systems Engneerng Proceda 3 (22) 259 267 263 ec u e NS PM PM PM PS Z NS NS Z PM PM PS Z NS NM NM PS PS PS Z NS NS NM NM PM PS Z NS NM NM NM NB PB Z Z NM NM NM NB NB The basc structure can be reduced to If A and B then C, among whch A s a fuzzy subsets of doman U, and B s a fuzzy subsets of doman V. Accordng to the control experence, the control decson table R can be organzed offlne.r s a fuzzy subsets of the cartesan product U V. In a moment, ts control volume s gven out by the followng equaton: C = ( A B) R In the equaton, fuzzy drect product opetaton fuzzy synthetc operaton 4.4. Defuzzfcaton After gettng the results, the reasonng of the fuzzy control has been completed. However, at present, the results obtaned s stll a fuzzy vector, whch can t be drectly used as a control volume. Therefore a converson must be done on the results so that t can get a clear output. The process s the defuzzfcaton. Usually the output part that has a converson functon s called defuzzfcaton nterface. To obtan accurate control volume, t requres the fuzzy method to express the calculated output of the membershp functons. In ths paper, the weghted average method s used. For each element on the doman, x ( =, 2,, n), t s used as the weghtng factor of the output fuzzy set membershp degree u (), that s to take the product xu (), then calculate the sum of the product and the membershp, and then calculate as follows: (3) x = n = n = xu() u () (4)
264 Wenbn Yan et al. / Systems Engneerng Proceda 3 (22) 259 267 The average x s the requred output of the fuzzy sets obtaned by the weghted average method. Fnally, the output x s multpled by the quanttatve factor to meet the control requrement. Then the practcal value of control volume s obtaned. In Fg. 5 and Fg. 6, t shows the SIMULINK model of the fuzzy PID controller and the fuzzy control rules, and also the fuzzy membershp functon graph of the error E, error change rate EC and the control volume u. In du/d Dervatve t Mux Fuzzy Logc Controller Out Fg.5. The smulaton model of the fuzzy PID controller.8 e (7) fuzzf (mamdan) 49 rules Degree of membershp.6.4 u (7).2 ec (7) System fuzzf: 2 nputs, outputs, 49 rules -3-2 - 2 3 e.8.8 Degree of membershp.6.4 Degree of membershp.6.4.2.2-3 -2-2 3 ec -4-3 -2-2 3 4 u Fg.6. The fuzzy membershp functon gragh of the error E, error change rate EC and the control volume u
Wenbn Yan et al. / Systems Engneerng Proceda 3 (22) 259 267 265 5. Neural network controller The neural network control s one of the front subject n the automatc control feld whch s developed n the 98 's. It s a new branch of the ntellgent control and opens up new ways to solve the control problem of the complex nonlnear, uncertan and unknown system. The sngle neuron adaptve ntellgent PID controller whch s conssted of the sngle neuron wth selflearnng and adaptve ablty not only has a smple structure, but also can adapt to the changes of the envronment. It also has strong robustness. PID control needs to adjust the three control effects nclude scale, ntegral and dfferental to form the coordnate and nterdependent relatonshp n order to get good control effect. The relatonshp s not a smple lnear combnaton, t can form the best relatonshp from the boundless change combnaton of nonlnear optmal relatonshp. Neural network has arbtrary nonlnear ablty and can acheve the best combnaton of PID control by learnng the performance of the system.[5] The learnng rules of neurons: no supervson Hebb learnng rules, supervson Delta learnng rules and supervsory Hebb learnng rules. The sngle neuron adaptve controller realzes ts functon of selfadapton and self-organzaton through the adjustment of the weghtng coeffcent, The realzaton of weght coeffcent adjustment s accordng to the supervson Hebb learnng rules. Control and learnng algorthm are: 3 uk ( ) = uk ( ) + K w( kx ) ( k) (5) = j j j= 3 w( k) = w ( k)/ w ( k) (6) w( k) = w( k ) + η z( k) u( k) x ( k) (7) I w ( k) = w ( k ) + η z( k) u( k) x ( k) (8) 2 2 P 2 w ( k) = w ( k ) + η z( k) u( k) x ( k) (9) 3 3 D 3 In the equatons, x( k) = ek ( ) ; x ( ) ( ) ( ) 2 k = ek ek ; x k = ek = ek ek + ek ; ( ) 2 ( ) ( ) 2 ( ) ( 2 ) 3 zk ( ) = ek ( )
266 Wenbn Yan et al. / Systems Engneerng Proceda 3 (22) 259 267 η I, η P, η D are respectvely the learnng rate of ntegral, proporton and dfferental. K s the proportonalty coeffcent of neurons, K >. The ntegral I, proporton D and dfferental P respectvely used dfferent learnng rate η I, η P, η D. So as to separately adjust the dfferent weght coeffcent. The choce of K value s very mportant. The greater the K value, the better the speed. But the bg overshoot may even make the system out of stablty. When the controlled object delay ncreases, the k value must be reduced to ensure that the system s stable. If K value selecton s too small, t can also make the system effcency becomes poor. In Fg. 7, t shows the SIMULINK smulaton model of the neural network PID controller. Model Reference Controller MATLAB Functon Reference Fcn In Reference Plant Output Neural Network Controller Control Sgnal Out Fg.7. The smulaton model of the neural network controller 6. Concluson In Fg. 8, t shows the square-wave response curve of the ordnary PID controller, the fuzzy PID controller and the neural network PID controller. We can see from the pcture that the square-wave responses of the fuzzy PID controller and ordnary PID controller are smlar,but the trackng curve of fuzzy PID controller s more smooth; Neural network PID controller has the most smooth trackng curve and the mnmum error wth the nput sgnal. It s obvously that the applcaton of neural networks PID control can be a very good way to mprove the control performance, and fuzzy PID control can realze concse and effectve control requrements n the face of a more complex nonlnear system for ts smple applcaton..8.8.8.6.6.6.4.4.4.2.2.2 rn,yout -.2 rn,yout -.2 rn,yout -.2 -.4 -.4 -.4 -.6 -.6 -.6 -.8 -.8 -.8 -.2.4.6.8.2.4.6.8 2 tme(s) -.2.4.6.8.2.4.6.8 2 Tme(second) -.2.4.6.8.2.4.6.8 2 tme(s) Fg.8. The square-wave response curve of the ordnary PID controller, the fuzzy controller and the neural network PID controller
Wenbn Yan et al. / Systems Engneerng Proceda 3 (22) 259 267 267 References [] Saffet Ayasun, Gültekn Karbeyaz. DC motor speed control methods usng MATLAB/Smulnk and ther ntegraton nto undergraduate electrc machnery courses[c]. Computer Applcatons n Engneerng Educaton, 27:347-354. [2] Ayütel Gelen, Saffet Ayasun. Realzaton of Power Electronc Convertor Based DC Motor Speed Control Methods Usng MATLAB/Smulnk[C]. Int J Engng Ed. Vol. 25, No., pp. 33-4, 29. [3] Seung-K Sul. Control of Electrc Machne Drve Systems[M], 2: 46-47, 57-28. [4] Kevn M. Passno, Stephon Yurkovch. Fuzzy Control[M], 2: 2-72. [5] L.P.J. Veelenturf. Analyss and Applcatons of Artfcal Neural Networks[M], 995: 6-45.