Tóm tắt Luận án Research and develop the control algorithms using artifical neural network to estimate motor parameters and control AC motors

VIETNAM ACADEMY OF SCIENCE AND TECHNOLOGY GRADUATE UNIVERSITY SCIENCE AND TECHNOLOGY ...*** LE HUNG LINH RESEARCH AND DEVELOP THE CONTROL ALGORITHMS USING ARTIFICAL NEURAL NETWORK TO ESTIMATE MOTOR PARAMETERS AND CONTROL AC MOTORS Major: Control Engineering and Automation Code: 62 52 02 16 SUMMARY OF ENGINEERING DOCTORAL THESIS Hanoi - 2016 This thesis is accomplished at: Graduate University of Science and Technology, Vietnam Academy of Science and Technology Supervisors 1: Assoc. Prof. DSc Pham Thuong Cat Supervisors 2: Dr. Pham Minh Tuan Examiner 1:...................................................................... Examiner 2:...................................................................... Examiner 3:...................................................................... The thesis is to be presented to the Defense Committee of the Graduate University of Science and Technology - Vietnam Academy of Science and Technology At Date Month Year 2016 The complete thesis is availabe at the library: - Graduate University of Science and Technology - Vietnam National Library 1 INTRODUCTION 1. A thesis statement necessary Nowadays, AC motor is widely used both in industrial applications and in domestics ones because of perfective technique specifications such as impact, high power, economic, convinient design, control and maintenance. AC motor is used in pumps, compressors, oil and gas industry, industrial or domestic fan, elevator, crane in construction industry, robotic etc Therefore, the three last decades, AC motor is used instead of DC motor because of eleminating the disadvantages of dc motor such as high maintenance cost for brush – commutator system, vibration environments, iginite flammable environments. Consequently AC motor is widely applied. However, there are still some control problems of AC motor when it can be more applied. Many researches want to improve the effective operation, reduce the production price but the results are still drawbacks. For example, the effect of control methods using Kalman filter, nonlinear filters or observers using sliding mode control to estimate rotor speed and flux depends on control algorithm, estimation of some parameters and the accuracy of the motor model. The mathmetic model of motor is quite difficult to obtain as desired because of uncertain parameters similaryly friction coeffection, inertia, resistance. The uncertain parameters change when the system is operating. In addition, the speed and flux estimation insteading of sensor with the high requirement of accuracy is quite difficult and it is necessary to research. Recently the development of artifical neural network is very helpful to solve the control problem, specially controlling nonlinear subjects with uncertain parameters. Artifical neural network can solve the nonlinearity effectively with self-tuning parameters when the system operates. In this thesis, we concentrate on research and develop some control and estimation algorithm for ac motor with uncertain parameters. 2. The objectives of the thesis - Propose algorithms for controlling speed and flux of AC motors - Propose rotor speed and flux estimation algorithms for speed sensorless controlller of AC motors 3. The main contents of the thesis Two control algorithms and two estimation algorithms of motor parameters are proposed. a) The speed control algorithm for AC motor with uncertain parameters and changing loads on rotating coordinate (d,q) using artifical neural network. b) The speed and flux control algorithm for AC motor with uncertain parameters and changing loads on stationay coordinate (α,β) using the decoupling method. c) The speed estiamtion algorithm for AC motor using artifical neural network and self- adaptation. d) The speed estiamtion algorithm for AC motor using self-adaptation. Lyapunov stability theory and Barbalats’s lemma are used to prove the system asympotic stability of the algorithms. Simulations will be implemented on Matlab. Outline: Chapter 1, Presenting some problems of motor control Chapter 2, Developing control algorithm of asynchrounous motors Chapter 3, Developing estimation algorithms of speed and flux of asynchronous motors Conclusion. 2 CHAPTER 1 OVERVIEW 1.1 Problem statement 1 - Obtaining accurately economically rotor flux and speed estimator algorithm, 2 - Developing AC motor control algorithm with uncertain parameters 3 - Designing intelligent motor controller based on the advanced production technology of electronics 1.2 AC control method AC motor control methods are classified as following diagram Figure 1.1 Classification of IM variable frequency control Nowadays motion control in industrial aplications is required accurately. Motor control methods are used as scalar control voltage/frequency (V/F), direct torque control and filed oriented control. In this thesis, field oreinted control method is ued to research and apply for three-phase AC motor with speed and moment control high performance requirement. Recent researches are focus on identifying the effection of rotor resistance without considering uncertain parameters such as friction coefficient, inertia or changing load. Therefore, this thesis proposes control algorithm and speed estimation of AC motor with uncertain parameters. 1.3 Research problems - Developing rotor speed and flux estimation of AC motor - Developing AC motor control algorithm with uncertain parameters - Using Lyapunov stability theory and Barbalat’s lemma to prove global asympotic stability of system and then using Matlab to simulate and check the validity of proposed control algorithm and estimator. Scalar control U/f = const is=f(ωr) stator current Vector control Field oriented control Direct torque control DTC Circular flux trajectory Hexagonal flux trajectory Rotor flux Oriented Stator flux oriented Direct RFO Indirect IRFO Natural Field Orientation NFO AC motor control 3 CHAPTER 2 DEVELOPING FLUX AND SPEED CONTROL ALGORITHM OF AC MOTOR WITH UNCERTAIN PARAMETERS This chapter will present two flux and speed control algorithm - Speed and flux control algorithm of AC motor uses artifical neural network with online learning rules to compensate uncertain on rotating coordiante (d,q). - Speed and flux control algorithm of AC motor does not decouple and then using artifical neural network to compensate uncertain on static coordiante (α,β). 2.1 AC motor control The model of AC motor is written on static coordinate (,): 1 1 s s r r m s r r s s r r s s s r r m s r r s s r r s r r r r r m s r r r r r r r m s r r di R R R L i u dt L L L L di R R R L i u dt L L L L d R R L i dt L L d R R L i dt L L                                                                            (2.13)   3 2 p m M r s r s L r z L d m i i J B m L dt             (2.14) The model of AC motor is written on ratating coordinate (d,q):     1 1 sd s r r m sd s sq rd rq sd s r r s sq s r r s sd m sq rd rq sq s r r s rd r r rd s rq m sd r r rq r r s rd rq m sq r r di R R R L i i u dt L L L L di R R R i L i u dt L L L L d R R L i dt L L d R R L i dt L L                                                                     (2.15)   3 2 p m M rd sq rq sd L r z L d m i i J B m L dt         (2.16) The mathmethic model of AC motor on rotating coordinate (d,q) when flux rq on axis q is eliminated. From the equation (2.15) results 4 1 1 sd s r r m sd s sq rd sd s r r s sq s r s sd m sq rd sq s r s rd r r rd m sd r r di R R R L i i u dt L L L L di R R i L i u dt L L L d R R L i dt L L                                             (2.17) 3 2 p m M rd sq L r z L d m i J B m L dt       (2.18) 2.2 Build speed control algorithm for three-phase asynchronous as motor with uncertain parameters on rotating coordinate (d,q) 2.2.1 Build a controller model From the equation (2.16), results in ( ) L d Ku t J B m dt     (2.22) where ( ) ( )rd sq rq sdu t i i   is control voltage. When rq is eliminated, yields * *( ) ( )rd sq rq sd rd squ t i i i     From equation (2.22), we rewrite: k k k( )u t J B m    (2.23) where: k k k J J J J K      ; k k k B B B B K      ; k Lmm K  ; k k,J B   are known; k k,J B  are unknown. set k k kf m J B    (2.24) k k( )u t J B f       (2.26) In summary, the motor control problem becomes determining the control signal u(t) that regulates motor speed  reaching reference speed ref when there some uncertain parameters. Figure 2.2 Motor control model Speed controller ref * sdi dq   uvw sqi sdi sdu squ si  si  su  su  sui svi Current controller * sqi 3~ dq  vt wt ut Vector modulation s 1 mL refr M3~ Lm u v w Flux model  -  sdi sqi 5 2.2.2 Build a speed control algorithm of motor We choose: 0 1( )u t u u  (2.27) where 0u is feedback signal written in PD form and u1 a signal compemsating unkown parameters f. And then: 0 k ref ref k( ( ))Du J K B          (2.28) Speed error : ref    , We set ' 1 k u u J   , k f f J    , ' k D D K K J   . ' ' DK u f     (2.31) Finally, the motor control problem becomes determining the control signal 'u to guarantee the system (2.31) asympotic stability when 'f is unknown. 'f is aproximated by a neural network with output fˆ . Theorem 1 [1][2]: Speed of induction motor ω (2.16), (2.22) aproaches the disired speed ωref while friction coefficicent B, inertia moment J and load moment mL are unkonwn if control rule u(t) and study rule w of neural network are defined as below k ref ref k k( ) ( ( )) 'Du t J K B J u            (2.34) ' ˆ(1 )u n f            (2.35) w n  (2.36) where optional parameters DK , , 0n   . Proof: We choose a positive definite function V such as :  2 2 1 2 V w  (2.37) 2 2( ) .D DV K K                     (2.38) 2 0DV K        (2.40) Based on the equation (2.40), Obviously, 0V  and 0V   with ∀ 0 ;  0V  while 0  , therefore ,  are always finite. 0V  , semi negative definite does not guarrantee the sysstem asymtopic stability. The system is non-autonomous because neural system is varied by time. Hence, it is nescessary to use Barbalats’s lemma. From (2.38), we obtain: 2 2 ( ) DV K sign             (2.41) where ,  are finite, so V  is always finite =>V is continuous by time. In addition, from Basbalat’s lemma V is continuous then 0 , 0V      . From the equation (2.31), 1f u and ref  meaning motor speed ω aproaches the disired speed ωref with error is equal to 0. 6 Rotor speed regulator as shown on Figure 2.3. 2.2.3 Current regulator Rewrite the equation (2.17) in vector form sdq sdq sdq rd rd r r rd m sd r r d dt d R R L i dt L L               i Ai Bu h (2.42) where: s m s s s s m s R L L R L L A                                ; h         ; 1 0 1 0 s s L L B               We find the stator voltage:  1 *sdq sdq sdq rd    u B Ai i Gξ h (2.43) where G is positive diagonal matrix and sdq sdq  ξ i i is error vector between the disired cunrrenr and regulated current. * * ( )sdq sdq sdq sdq sdq rd     ξ i i i Ai Bu h     (2.44) Subtituting the equation (2.43) into (2.42) results: ξ Gξ  => 0ξ Gξ  (2.45) Hence the error vector ξ 0 meaning sdq sdq i i . Building the current regulator as shown on Figure 2.4: Figure 2.3 Rotor speed regulator of the motor 1 k ˆ(1 )u J n f              fˆ w w n  ref k ref ref k( ( ))DJ K B          -  0u 1u ( )u t * 1 rd * sqi 7 2.2.4 Simulation results Motor control system model with uncertain parameters and speed feedback signal as shown on Figure 2.2. Simulation was conducted using a four-pole squirrel-cage induction motor from LEROY SOMER with the parameters shown in Table 1. The reference angular velocity varies in a trapezoid shape as seen in Figure 2.5 with the maximum ref 100  Rad/s (956 prm) and reference flux * refr =1.5 (Wb). Motor is mounted on the driller system. Table 1 Motor parameters Rated Power 1.5 KW Stator inductance (Ls) 0.253 H Rated stator voltage 220/380 V Rotor inductance (Lr) 0.253 H Rated stator current 6.1/3.4 A Mutual inducatnce (Lm) 0.213 H Stator resistance(Rs) 4.58 Ω Motor inertia (J) 0.023 Nms 2 /rad Rotor resistance (Rr) 4.468 Ω Viscous coefficient friction (B) 0.0026 Nms/rad Figure 2.5 is rotor desired speed and is started in time t=0,1(s). Figure 2.5 Desired speed ref The motor speed control system was simulated with these assumed uncertain parameters: ; 0.05B B B B B     và ; 0.20 sin(100 )J J J J J t      Load mL varies in a shape as seen in Figure 2.6c 1 2L L L Lm m m m   (Nm) where : mL1 is steady load of system, 3 (Nm), mL2 is unknown load while drill on the material as shown on Figure 2.6a. Lm is unknown load depended on the structure of material as shown on Figure 2.6b. 0 5 10 15 20 25 30 35 40 45 50 20 40 60 80 100 Time (s) R a d /s Omega.ref h 1 B  ξ sdqu A r m r r R L L s R G * sdqi sdqi - + - + + - Figure 2.4 Current regulator model rd + d dt sdi 8 Figure 2.6a mL2 unknown load while drill on the material Figure 2.6b ΔmL unknown load depended on the structure of material Figure 2.6c mL load of the system Figure 2.8 Error between desired rotor speed and real rotor speed using neural network 0 5 10 15 20 25 30 35 40 45 50 0 1 2 3 4 Time (s) N m 0 5 10 15 20 25 30 35 40 45 50 -1 -0.5 0 0.5 1 Time (s) N m 0 5 10 15 20 25 30 35 40 45 50 2 4 6 8 Time (s) N m 0 5 10 15 20 25 30 35 40 45 50 -4 -3 -2 -1 0 1 Time (s) R a d /s 9 Figure 2.9 Setting time of speed with the load mL - When the system starts, the error of speed is about 3,5%. When the load is changed suddenly, the error of speed is about 1,5%. - The rotor speed is reached the steady state after the short time about 1s by using the neural network, the speed is approached the desired speed. 2.3 Build speed and flux control algorithm for three-phase asynchronous as motor with uncertain parameters on stationary coordinate (,) 2.3.1 Control model We set 1x  , 2 2 2 r rx     , From equation (2.13) and (2.14), we obtain:           1 1 1 1 1 2 1 1 1 s sr r m m s r s r r s r s s r L L m s r r s r s s R RB R B R x L x L x J L L J L L Kx i i RK x x R m m L J J L L J J K u u J L                                                        (2.49) 0 0.5 1 1.5 2 2.5 3 -4 -3 -2 -1 0 1 Time (s) R a d /s Speed and flux Controller ref  2 refr si  si  su  su  Flux Model 1e + - 2e + - 2ˆ r ˆ r ˆ r  uvw sui svi 3~ vt wt ut Vector modulation M3~ Lm u v w Hình 2.12 Motor control model 10           2 2 2 2 1 2 2 2 2 2 2 1 2 2 2 r r m r r sr r m m r r r r r s r r m r s r s r r mr m s s r s r s r r s R R x x L x L L RR R L L i i L L L R L x i i L R LR L i i u u L L L                                                        (2.50) Rewriting the equation (2.49), (2.50) as formula below: 1 s   x Mx + Nx Q D u  (2.51) where B, J, Rr are unknown parameters: B B B    J J J   r r rR R R    , , rB J R    are known parameters. , , rJ B R   are unknown parts. From the known parameters, r  và r  can be found r r r r r m s r r r r r r r m s r r d R R L i dt L L d R R L i dt L L                                    (2.52) Hence the equation (2.51) can be reprented as below: N = N +ΔN  ; M = M +ΔM  ; Q = Q +ΔQ  ; D = D +ΔD  . (2.53) where , , ,Q D M N     are known matrices and , , ,Q D M N    are unknown matrices. We choose:  s u D v - Q  (2.54) where T v vv      is auxiliary control signal.   v x Mx + Nx f     (2.56) with 1 1     f = ΔMx+ΔNx D Dv D DQ Q   are unknown parts that determine after. In summary, the motor control problem becomes determining the control signal v that regulates motor speed and flux reaching desired values ref  ,  2 2 2 2refr r r r       while , , rJ B R are uncertain parameters and changing load is unknown and is determined after. 11 2.3.2 Speed and flux control method We denote: s = e + Ce (2.57) where C is the positive definite diagonal matrix; refe x - x is the error between the actual value 1 2 2 r x x x               and the desired value 1 2 2 ˆ ref ref ref ref r ref x x x                . Therefore, when s 0 , then e 0 . Figure 2.13 The neural network structure The form of the neural network: ˆf f η Wθ η    (2.58) where 11 12 21 22 w w w w W        is a weighted matrix; 1 2 θ          output function vector of input neuron i; τ bounded approximation error: 0η  . Therefore, to make s 0 and error ref( )e x - x 0  we need to choose v and the learning rule for the weighted W to make the system (2.56) asymptotically stable. Theorem 2 [4][6]: Speed andflux of the AC motor in equation (2.14) approach the desired values ref  ,  2 2 2 2refr r r r       while ,J ,B rR and changeable load LT are unknown if the control signal v and weighted W are defined as below: ref ˆ ˆ 1 v = Hs Mx + Nx + x - Ce + v    (2.59)  1 1 s v Wθ s     (2.60) i iw s   (2.61) where H is a positive definite diagonal matrix, iw is the i th column of the weighted matrix W and 0  , 0    with 0  . Proof: Applying Lyapunov’s stability theory, we chose a positive definite function V suchas: T T1 1 2 2 i i i V s s w w   (2.62)  T T 1 1V s Hs s v - Wθ - η       (2.65) T 0V s Hs s    (2.66) From equation (2.66), it is clearly that 0V  and 0V   with s 0  ; 0V  when s 0 and from equation (2.58), it is obviously that ,η η are always finite. Because of 0V     w11 w22 w12 w21  2s 1s 2 2 2 1 i i i f w    2 1 1 1 i i i f w    12 negative definite, the system is not guaranteed to be asympotic stability. Therefore, we need use Barbalat’s lemma to stabilize the non-autonoumous system asympotical stability. From the equation (2.65), we obtain: T T T2 T V s s s Hs s η s η s          (2.67) where ,s s and ,η η are always finite, then V  is finite, V  is continuous by