Inter national J our nal of Electrical and Computer Engineering (IJECE) V ol. 8, No. 5, October 2018, pp. 3666 3677 ISSN: 2088-8708 3666       I ns t it u t e  o f  A d v a nce d  Eng ine e r i ng  a nd  S cie nce   w     w     w       i                       l       c       m     Nonlinear Contr ol of an Acti v e Magnetic Bearing with Output Constraint Danh Huy Nguy en , T ung Lam Nguy en ;  , Manh Linh Nguy en , and Huy Phuong Nguy en School of Electrical Engineering, Hanoi Uni v ersity of Science and T echnology  Institute for Control Engineering and Automation, Hanoi Uni v ersity of Science and T echnology Article Inf o Article history: Recei v ed December 12, 2017 Re vised July 20, 2018 Accepted Aug 9, 2018 K eyw ord: Acti v e magnetic bearing Backstepping control Barrier L yapuno v function Speed observ er ABSTRA CT In this paper , an appropriate control strate gy is proposed to handle the nonlinear dy- namics of an acti v e magnetic bearing (AMB). The goal of the control design is to dri v e the AMB rotor to the origin with impro v ed transient response. In order to achie v e this task, back stepping control technique with a barrier L yapuno v function are emplo yed to k eep the tracking err or trajectory inside a predefined zone to a v oid possible mechanical contact between rotor and stator . Besides, a speed observ er is also used since informa- tion about rotor speed is not al w ays a v ailable. The stability of the closed-loop system is pro v en. The ef fecti v eness of the proposed control strate gy is v erified by numerical simulations. Copyright c 2018 Institute of Advanced Engineering and Science . All rights r eserved. Corresponding A uthor: T ung Lam Nguyen School of Electrical Engineering, Hanoi Uni v ersity of Science and T echnology No 1, Dai Co V iet street, Hanoi, V ietnam +84989998384 lam.nguyentung@hust.edu.vn 1. INTR ODUCTION Acti v e magnetic bearing (AMB) has been de v eloped in recent fe w decades to replace the con v entional mechanical bearings. The main concept of an AMB is to use an electromagnetic force to support a body with- out an y mechanical contact. In comparison with the traditional mechanical bearings with m an y dra wbacks [1] and [2], AMB e xhibits man y adv antages such as: frictionless, lubricant-free operat ion, acti v e vibration control and unbalance compens ation ability . Hence, AMB is a promising solution for high-speed applications such as high-speed motor [3], flywheel ener gy storage systems (FESS), .etc. F or the proper operation of an AMB, rotor positioning control is a challenging task due to the inherent instability and nonlinearity of the system as men- tioned in [4], [5], and [6]. In order to deal with system nonlinearity , authors of [7] propose a linearized AMB model and emplo yed con v entional PID control, similar approach can be found in [8]. In addition to linear control trend, an of f-line tunning technique for centralized and decentralized linear AMB controllers are demonstrated in [9]. When the v ariation of the rotor position is small, linearizing the model in a small re gion around an equi- librium point in order to use the linear control technique is a n appropriate approach to stabilize the AMB system. Ho we v er , the performance of the positioning system may de grade significantly when the operating point is f ar from the desired equilibrium point. Hence, nonlinear control technique has been studied to further impro v e the systems performance. In order to stabilize rotor when f acing with a vibrating base, [10] design a sliding mode scheme for an AMB system. The closed-loop system e xhibits rob ustness to uncertainties and e xternal disturbances. Based on L yapuno v’ s direct method and the singular perturbation order -reduction technique, the state feedback control is deri v ed in [11]. Through a set of numerical simulations, the state feedback control sho ws its strength in compar - ison with output feedback and deadbeat controls. Inspired by loop shaping properties of H 1 optimization and disturbances rejection ability of the disturbance observ er -based controller , [12] successfully de v elop a h ybrid controller whose ef fecti v eness are v erified via simulations and real-time e xperiments. By in v estig ating strong nonlinearity of 3-poles AMB, a feedback linearization la w is composed in [13]. Rotor responses also are re- stricted to a safe distance from stator boundary , ho we v er , the requirement of initial conditions might be dif ficult J ournal Homepage: http://iaesjournal.com/online/inde x.php/IJECE       I ns t it u t e  o f  A d v a nce d  Eng ine e r i ng  a nd  S cie nce   w     w     w       i                       l       c       m     DOI:  10.11591/ijece.v8i5.pp3666-3677 Evaluation Warning : The document was created with Spire.PDF for Python.
IJECE I SSN: 2088-8708 3667 to archi v e in practice. In a circumstance of the AMB dri ving rotor with unkno wn mass imbalance, [14] proposes an adapti v e m echanism to tackle this situation. T aking v oltage input saturation into account, a rob ust fuzzy con- trol that is able to stabilize the AMB rotor is de v eloped in [15]. Apart from con v entional AMB, [16] suggests a design and decoupling control for axial AMB by combining optimal algorithm and feed-forw ard control. The paper proposes an nonlinear control approach to stabilizing AMB rotor problem based on backst ep- ping control. A speed observ er is utilized in this research to acquire rotor speed information. Barrier L yapuno v candidate function is embedded into design process to dri v e the AMB rotor traj ectory inside a defined range. The stability of the closed-loop system is pro v en. A set of numerical simulations are gi v en to v erify the control ef fecti v eness. 2. MA THEMA TICAL MODEL OF AMB Figure 1 sho ws the one de gree of freedom (1DOF) AMB system considered in this paper . In this figure, Figure 1. An AMB system. F k , u k and i k are the electromagnetic force, applied input v oltage and current of the corresponding coil k , respecti v ely , with k = 1 ; 2 ; F d is the disturbance force caused by rotor mass and other uncertainties. Suppose that the displacement x of the rotor from the nominal posi tion x 0 caused by the initial v oltage and current ( u 0 ; i 0 ) . Denote x 1 and x 2 are the air g aps between the rotor and the left and right side stators, it yields x 1 = x 0 x; x 2 = x 0 + x (1) i 1 = i 0 i; i 2 = i 0 + i (2) u 1 = u 0 u; u 2 = u 0 + u (3) By using (1)-(3), a fundamental operation gi v es 8 > > > > > < > > > > > : dx dt = d dt = K 4 m i 1 x 0 x 2 K 4 m i 2 x 0 + x 2 + F d m di 1 dt = A h R i 1 K 2( x 0 x ) i 1 + u 1 i di 2 dt = B h R i 2 + K 2( x 0 + x ) i 2 + u 2 i (4) where, A = 2( x 0 x ) 2 L s ( x 0 x ) + K ; B = 2( x 0 + x ) 2 L s ( x 0 + x ) + K ; K = g N 2 A g (5) In (5), g is the permeability of air , N is the number of turns in each coil and A g is the cross-section area of the electromagnet. Due to the f act that only the position of the rotor and the tw o current i 1 and i 2 of the corresponding coils are a v ailable for measurement, a speed observ er is needed to estimate the rotor speed. In this research, the nonlinear observ er proposed in [17] is utilized for speed estimation as follo ws ^ = + + k x + F d m 1 k (6) Nonlinear Contr ol of an Active Ma gnetic Bearing ... (Danh Huy Nguyen) Evaluation Warning : The document was created with Spire.PDF for Python.
3668 ISSN: 2088-8708 where, _ = k k 2 x (7) _ = k + i 2 1 ( x 0 x ) 2 i 2 2 ( x 0 + x ) 2 (8) = K 4 m (9) where k is a positi v e constant. Suppose that the initial conditions of the system is (0) = 0 ; (0) = 0 (10) Denote the speed estimation error as = ^ (11) Then, by dif ferentiating both side of (11) and based on (4) and (6)-(9), it gi v es _ = _ _ ^ = i 1 x 0 x 2 i 2 x 0 + x 2 + F d m _ _ k _ x = k + + k x + 1 k F d m = k (12) Relation (12) means that the speed estimation error e xponentially con v er ges to zero and the con v er gence speed depends on k . By using the speed observ er (6), the state space model of the AMB system can be re written as 8 > > > > > > > > > < > > > > > > > > > : ^ _ x = ^ + ^ = + + k x + 1 k F d m _ = k + i 1 x 0 x 2 i 2 x 0 + x 2 _ = k k x 2 _ i 1 = A h R i 1 K 2( x 0 x ) ( ^ + ) i 1 + u 1 i _ i 2 = B h R i 2 + K 2( x 0 + x ) ( ^ + ) i 2 + u 2 i (13) 3. CONTR OL DESIGN It can be observ ed from (13) that the model of the AMB has the form of a strict-feedback system. Hence, backstepping control technique is chosen to stabilize the system. The control design is implemented step-by-step as follo ws. Step 1 : The goal of the control design in this step is to dri v e the rotor to a desired position y r with minimized tracking error . Denote z 1 as the position error z 1 = x y r (14) Dif ferentiating both side of (14) gi v es _ z 1 = _ x _ y r = + + k ( z 1 + y r ) + g k + _ y r (15) Consider the follo wing barrier L yapuno v candidate function V 1 = 1 2 log k 2 b k 2 b z 2 1 + 1 2 k d 1 (16) with d 1 > 0 . By using the proposed barrier L yapuno v function, the position error z 1 is k ept in a re gion restricted by ( k b ; k b ) where k b is a positi v e number . This means the o v ershoot of the AMB in transient state can be IJECE V ol. 8, No. 5, October 2018: 3666 3677 Evaluation Warning : The document was created with Spire.PDF for Python.
IJECE I SSN: 2088-8708 3669 handle by k b . This property is meaningful in practice since the mechanical contact between the rotor and stator can be a v oided. Dif ferentiating both sides of (16), it gi v es _ V 1 = z 1 _ z 1 k 2 b z 2 1 2 d 1 = z 1 ( + + k z 1 + _ Y r + k Y r ) k 2 b z 2 1 2 d 1 (17) in order to render _ V 1 0 , the virtual control action v is selected as v = 1 c 1 z 1 ( k 2 b z 2 1 ) d 1 z 1 k 2 b z 2 1 k ( z 1 + Y r ) + _ Y r (18) with c 1 > 0 . Step 2 : As seen in the pre vious step, if = v , then _ V 1 0 which results in = 0 and z 1 = 0 . Hence, the goal of this step is to guarantee that approaches to v . Denote z 2 as z 2 = v (19) The candidate L yapuno v function is chosen as V 2 = 1 2 log k 2 b k 2 b z 2 1 + 1 2 z 2 2 + 2 2 k 1 d 1 + 1 d 2 (20) with d 2 > 0 . Dif ferentiating both sides of (20) gi v es _ V 2 = z 1 _ z 1 k 2 b z 2 1 + z 2 _ z 2 2 1 d 1 + 1 d 2 (21) T aking from (17) and substitute it into (15) gi v e _ z 1 = z 2 c 1 z 1 ( k 2 b z 2 1 ) d 1 z 1 k 2 b z 2 1 + (22) Dif ferentiating both sides of (19) gi v es _ z 2 = _ @ v @ z 1 ^ v + _ Y r @ v @ Y r + k _ Y r (23) in which, @ v @ z 1 = 1 k c 1 k 2 b + 3 c 1 z 2 2 d 1 ( k 2 b + z 2 1 ) ( k 2 b z 2 1 ) 2 (24) @ v = 1 (25) Then, by substituting (22) and (23) into (21), it results in _ V 2 = c 1 z 1 d 1 z 1 k 2 b z 2 1 2 d 1 2 2 1 d 2 + 3 4 d 1 + z 2   z 1 k 2 b z 2 1 + _ @ v @ _ @ v @ z 1 _ z 1 Y r + k _ Y r ! (26) From (26), to guarantee that _ V 2 < 0 , the virtual control signal _ v is chosen as _ v = c 2 z 2 + @ v @ _ + @ v @ z 1 ( ^ _ Y r ) d 2 z 2 @ v @ z 1 2 z 1 k 2 b z 2 1 + Y r k _ Y r with c 2 > 0 . If _ = _ v (27) Nonlinear Contr ol of an Active Ma gnetic Bearing ... (Danh Huy Nguyen) Evaluation Warning : The document was created with Spire.PDF for Python.
3670 ISSN: 2088-8708 then _ V 2 < 0 . Ho we v er , the solution for (27) can not be obtained since it depends on tw o v ariables i 1 and i 2 . In oder to solv e the abo v ementioned problem and to sa v e the ener gy , a switching control strate gy is emplo yed in which either i 1 or i 2 is used. If x < 0 , coil 2 is turned of f which means i 2 = 0 , then _ = k + i 2 1 ( x 0 x ) 2 (28) Substitute (28) into (27), the desired current of coil 1 is i 1 = i 1 ;d = x 0 x q _ d + k (29) Denote = _ d + k , it gi v es i 1 = i 1 ;d = x 0 x p (30) Similarly , if x 0 , then i 1 = 0 and hence _ = k i 2 2 ( x 0 + x ) 2 (31) The reference current for coil 2 is i 2 = i 2 ;d = x 0 + x p (32) Remark : Equations (28), (31), and the definition of indicate that square root operations in (29) and (32) al w ays hold. Step 3: The goal of this final step is to obtain u 1 and u 2 which dri v es i 1 and i 2 to their desired v alues i 1 ;d and i 2 ;d , respecti v ely . Case 1: x < 0 . Denote z 3 as z 3 = i 1 i 1 ;d (33) Dif ferentiate (33), it gi v es _ z 3 = _ i 1 _ i 1 ;d = AR i 1 + AK i 1 2( x 0 x ) 2 @ i 1 ;d @ z 1 ( ^ v + ) _ Y r + Au 1 @ i 1 ;d @ _ @ i 1 ;d @ _ Y r Y r @ i 1 ;d @ Y r ... Y r AK i 1 2( x 0 x ) 2 + @ i 1 ;d @ Y r _ Y r (34) in which @ i 1 ;d @ z 1 = p + ( x 0 x ) 1 2 p @ @ z 1 (35) @ i 1 ;d @ = ( x 0 x ) 1 2 p @ @ (36) @ i 1 ;d @ = ( x 0 x ) 1 2 p @ @ (37) @ i 1 ;d @ Y r = p + ( x 0 x ) 1 2 p @ @ Y r (38) @ i 1 ;d @ _ Y r = ( x 0 x ) 1 2 p @ @ _ Y r (39) @ i 1 ;d @ Y r = ( x 0 x ) 1 2 p @ @ Y r (40) IJECE V ol. 8, No. 5, October 2018: 3666 3677 Evaluation Warning : The document was created with Spire.PDF for Python.
IJECE I SSN: 2088-8708 3671 with, @ @ = @ v @ z 1 2 d 2 c 2 + @ v @ z 1 + k (41) @ @ = @ v @ z 1 2 d 2 c 2 + k + @ v @ z 1 (42) @ @ Y r = c 2 k + d 2 @ v @ z 1 2 k + @ v @ z 1 k (43) @ @ _ Y r = c 2 1 + d 2 @ v @ z 1 2 1 @ v @ z 1 k (44) @ @ Y r = 1 (45) @ @ z 1 = @ 2 v @ z 2 1 ^ + c 2 @ v @ z 1 + d 2 @ v @ z 3 3 + k @ v @ z 1 + k 2 + 2 d 2 ( v ) @ v @ z 1 @ 2 v @ z 2 ( k 2 b + z 2 1 ) ( k 2 b z 2 1 ) 2 (46) @ 2 v @ z 2 1 = 1 6 c 1 z 1 2 z 1 d 1 (3 k 2 b + z 2 1 ) ( k 2 b z 2 1 ) 3 (47) The third L yapuno v candidate function is chosen as V 3 = V 2 + 1 2 z 2 3 + 2 2 k d 3 (48) with d 3 > 0 . The dif ferentiation of (48) is _ V 3 = c 1 z 2 1 c 2 z 2 2 + z 3 _ z 3 2 3 4 d 1 + 3 4 d 2 + 1 d 3 d 1 z 1 k 2 b z 2 1 2 d 1 2 d 2 z 2 @ v @ z 1 2 d 2 2 (49) T o mak e _ V 3 0 , _ z 3 is chose as _ z 3 = c 3 z 3 d 3 z 3 F 2 + F (50) where F = AK i 1 2( x 0 x ) 2 @ i 1 ;d @ z 1 (51) and c 3 > 0 . The control signal u 1 which stabiliz es the AMB system around the equilibrium point x 0 in Case 1 can be obtained from (34) and (50) as follo ws u 1 = 1 A AR i 1 F ^ _ Y r c 3 z 3 d 3 z 3 F 2 + @ i 1 ;d @ _ + @ i 1 ;d @ _ + 1 A  AK i 1 2( x 0 x ) 2 + @ i 1 ;d @ Y r _ Y r + @ i 1 ;d _ Y r Y r + @ i 1 ;d @ Y r ... Y r (52) Case 2: In this case, x > 0 ; i 2 1 = 0 ; u 1 = 0 ; 2 ;d = x 0 + x p (53) Similar to Case 1 , the control action u 2 in this case is u 2 = 1 B B R i 2 G ^ _ Y r c 4 z 4 d 4 z 4 F 2 + @ i 2 ;d @ _ + 1 B @ i 2 ;d @ _ B K i 2 2( x 0 + x ) 2 @ i 2 ;d @ Y r _ Y r + @ i 2 ;d @ _ Y r Y r + 1 B @ i 2 ;d @ Y r ... Y r (54) with c 4 > 0 and G = B K i 2 2( x 0 + x ) 2 @ i 2 ;d @ z 1 (55) Nonlinear Contr ol of an Active Ma gnetic Bearing ... (Danh Huy Nguyen) Evaluation Warning : The document was created with Spire.PDF for Python.
3672 ISSN: 2088-8708 4. ST ABILITY AN AL YSIS Similar to the control design section, the stability analysis of the AMB control system is also di vided into tw o cases. Case 1 : x < 0 and i 2 = 0 . The closed-loop system can be described by 8 > < > : _ z 1 = z 2 c 1 z 1 ( k 2 b z 2 1 ) d 1 z 1 k 2 b z 2 1 + _ z 2 = c 2 z 2 d 2 z 2 @ v @ z 1 2 z 1 k 2 b z 2 1 @ v @ z 1 _ z 3 = _ i 1 _ i 1 ;d (56) where, _ i 1 = A R i 1 K ( ^ + ) i 1 2( x 0 x ) 2 + u 1 (57) _ i 1 ;d = @ i 1 ;d @ z 1 _ z 1 + @ i 1 ;d @ _ + @ i 1 ;d @ _ + @ i 1 ;d @ Y r _ Y r + @ i 1 ;d @ _ Y r Y r + @ i 1 ;d @ Y r ... Y r (58) x = z 1 + Y r (59) The v elocity observ er in this case is ^ = + + k x + g k (60) with, ( _ = k k 2 x _ = k i 2 2 ( x 0 + x ) 2 (61) and @ @ = @ v @ z 1 2 d 2 c 2 + @ v @ z 1 + k (62) @ @ = @ v @ z 1 2 d 2 c 2 + k + @ v @ z 1 (63) @ @ Y r = c 2 k d 2 @ v @ z 1 2 k + @ v @ z 1 k (64) @ @ _ Y r = c 2 + d 2 @ v @ z 1 2 1 @ v @ z 1 k (65) @ Y r = 1 (66) @ @ z 1 = @ 2 v @ z 2 1 ^ + c 2 @ v @ z 1 + d 2 @ v @ z 1 3 + k @ v @ z 1 + k 2 + 2 d 2 ( v ) @ v @ z 1 @ 2 v @ z 2 ( k 2 b + z 2 1 ) ( k 2 b z 2 1 ) 2 (67) @ 2 v @ z 2 1 = 1 6 c 1 z 1 2 z 1 d 1 3 k 2 b + z 2 1 ( k 2 b z 2 1 ) 3 (68) @ i 1 ;d @ z 1 = p + ( x 0 x ) 1 2 p @ @ z 1 (69) @ i 1 ;d @ = ( x 0 x ) 1 2 p @ @ (70) @ i 1 ;d @ = ( x 0 x ) 1 2 p @ @ (71) @ i 1 ;d @ Y r = p + ( x 0 x ) 1 2 p @ @ Y r (72) IJECE V ol. 8, No. 5, October 2018: 3666 3677 Evaluation Warning : The document was created with Spire.PDF for Python.
IJECE I SSN: 2088-8708 3673 @ i 1 ;d @ _ Y r = ( x 0 x ) 1 2 p @ @ _ Y r (73) @ i 1 ;d @ Y r = ( x 0 x ) 1 2 p @ @ Y r (74) T o sho w the stability of the AMB system in this case, consider the follo wing candidate L yapuno v function V C 1 = 1 2 log k 2 b k 2 b z 2 1 + 1 2 z 2 2 + 1 2 z 2 3 + 2 2 k 1 d 1 + 1 d 2 + 1 d 3 (75) Dif ferentiate both sides of (75) yields _ V C 1 = c 1 z 2 1 c 2 z 2 2 c 3 z 2 3 d 1 z 1 k 2 b z 2 1 2 d 1 2 d 2 @ v @ z 1 2 d 2 2 d 3 z 3 F 2 d 3 2 3 4 d 1 + 3 4 d 2 3 4 d 3 2 (76) Since c 1 , c 2 , c 3 , d 1 , d 2 , d 3 are positi v e constants, then _ V C 1 < 0 which means system (56) is stable. Proof of bounded output is gi v en in [18]. Case 2 : x > 0 and i 1 = 0 . In this case, the stability analysis can be treated in the same manner as in Case 1 . 5. SIMULA TION RESUL TS The closed-loop system is numerically tested to v erify the ability of the proposed control design. The AMB used in the numerical simulation has the follo wing parameters: P arameters V alue Nominal air -g ap 0.001m Number of turns 400 Coil resistance 1 Cross-section area 0 : 001 m 2 Rotor mass 2.6Kg Rotor initial position 0.0004Kg Air -g ap permeability 1 : 256 : 10 6 Assume, initially the rotor is at a distance of 0.0004m a w ay from the x 0 ie. attached to coil 1. The simulation is carried out in tw o scenarios. First, a con v entional direct L yapuno v is applied and then L yapuno v function with appended term to limit system output response is embedded into the system. Figure 2. Displacement. Nonlinear Contr ol of an Active Ma gnetic Bearing ... (Danh Huy Nguyen) Evaluation Warning : The document was created with Spire.PDF for Python.
3674 ISSN: 2088-8708 Figure 3. Actual and estimated v elocity . Figure 4. Current. Figure 5. Control v oltage. It can be seen from the system responses that the rotor displacement has a maximum v alue of 0.001m. The phenomenon implies the rotor hits coil 2 during transient period , this is undesirable in practice. IJECE V ol. 8, No. 5, October 2018: 3666 3677 Evaluation Warning : The document was created with Spire.PDF for Python.
IJECE I SSN: 2088-8708 3675 Figure 6. Displacement. Figure 7. Actual and estimated v elocity . Figure 8. Current. Nonlinear Contr ol of an Active Ma gnetic Bearing ... (Danh Huy Nguyen) Evaluation Warning : The document was created with Spire.PDF for Python.