You are on page 1of 12

Control Engineering Practice 21 (2013) 1519–1530

Contents lists available at ScienceDirect

Control Engineering Practice
journal homepage: www.elsevier.com/locate/conengprac

Design and application of a sliding mode controller for accurate motion
synchronization of dual servo systems
Burak Sencer a,n, Tatsuya Mori b, Eiji Shamoto c
a
Department of Mechanical Science and Engineering, Graduate School of Engineering, Nagoya University, Furo-cho, Chikusa-ku, Building 2,
2F Room 205, Nagoya 464-8603, Japan
b
Daido Amistar Co., Ltd., 3-152 Hino, Daito-city, Osaka 574-0061, Japan
c
Nagoya University, Furo-cho, Chikusa-ku, Building 2, 2F, Room 202, Nagoya 464-8603, Japan

art ic l e i nf o a b s t r a c t

Article history: This paper presents a continuous time sliding mode controller (SMC) design to deal with the problem of
Received 19 October 2012 motion synchronization in dual spindle servo systems. Synchronization error is defined as the differential
Accepted 7 July 2013 position error between the two servo drives that follow identical reference motion trajectory. Proposed
Available online 17 August 2013
SMC controller penalizes three error states; namely individual axis tracking errors and the synchroniza-
Keywords: tion error for accurate synchronization. The control law is derived from Lyapunov energy function
Servo system without switching condition. The controller shows robust motion synchronization against disturbances
Synchronization and parameter variations. Proposed SMC control is implemented in conventional double-sided machin-
Tracking error ing operation.
Sliding mode control
& 2013 Elsevier Ltd. All rights reserved.
Double-sided milling

1. Introduction The machine tool is presented in Fig. 1a. Dual spindles equipped
with cutting tools rotate at high speed and engage with work-piece
In order to attain higher productivity and precision at a lower at identical time instant at each side to remove material. Non-
cost, there is a strong demand for developing advanced manufac- linear cutting forces are generated acting on the rotational motion
turing techniques. Various precision motion systems, such as of the spindles. Compared to conventional single sided material
robots and multi-axis computer numerical control (CNC) machine removal process, this strategy provides higher efficiency and finish
tools require high-speed coordinated motion. As the system traces accuracy. However, double-sided milling method relies on cutting
complex trajectory at high-speed, motion accuracy depends not both sides of the work-piece with precision cutter position syn-
only on the capability of individual axes that are in motion, but chronization to cancel out process forces in cutting. As a matter of
also on their synchronization. A part from the drive based fact, thin work-piece has low structural stiffness in the normal
controller design, more robust and coupled control strategies are direction due to clamping. A force unbalance will occur if the
required to improve the accuracy of complex motion systems. In cutters do not engage with the surface at the same instant.
these applications “synchronization” between the drives is favored Resultant unbalanced force can trigger forced, or even chatter
in parallel to individual set point tracking objective. vibrations (Altintas, 2000) causing thin work-piece to vibrate
A practical example showing the importance of motion syn- heavily in normal direction and destroy the surface finish (Mori
chronization in manufacturing can be observed in the double-sided et al., 2010). This phenomenon is explained in Fig. 2. Therefore,
milling (Mori, Hiramatsu, & Shamoto, 2010; Shamoto, Mori, accurate synchronization of teeth positioning, i.e. position syn-
Nishimura, Hiramatsu, & Kurata, 2010) operation. Double-sided chronization between the right and left spindles, is crucial for
milling has been proposed for finishing thin steel plates to attain achieving high manufacturing quality in this material removal
improved flatness and surface quality. This process is shown in process.
Fig. 1. In this machining process, thin steel plates are clamped Several approaches have been developed to achieve “synchro-
vertically and both sides of the plate are cut simultaneously nization” or “coordination” in general multi-axis motion systems.
employing multiple face cutters attached on left and right spindles. A major strategy to introduce coordination between drives is
through ‘cross coupling controller’. This method is developed
essentially for contouring control of CNC machine tools where
n
Corresponding author. Tel.: +81 52 789 4491; fax: +81 52 789 3107.
contour error or so called the ‘path following error’ is calculated in
E-mail addresses: burak@upr.mech.nagoya-u.ac.jp, real-time, and a cross coupling controller (CCC) is utilized to inject
burak0307@gmail.com (B. Sencer). extra control action to pull the end effector closer to reference

0967-0661/$ - see front matter & 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.conengprac.2013.07.001

2. and later Chen. path. Resembling a relay controller. Importance of motion synchronization for vibration avoidance: (a) With time difference. & Yu.e ux - Command x X Axis Master Controller Motor(X) Position(Y) ey Y Axis uy Slave Slave . Jinbang. PI and PID are based on the complex path following applications (Koren & Lo. 2006). which estimates and Nijmeijer. ‘master–slave’ control approach has been law. Sabanovic. Master x X Axis Controller Motor(X) Reference Command Motion Position(Y) ey Y Axis uy Slave Slave . Anwen. 2003. disturbance sensitive applications. 2006) can be implemented in drive level for precision existing drive level controller causes controllers to work against each positioning with vibration suppression (Horowitz. The original contouring controller for 5-axis CNC machine tools. Slotine and Li (1988) proposed motion synchronization in dual servo drives (Rodriguez-Angeles & adaptive sliding mode controller (ASMC). & Matthews. For instance. 3a. Ren. and guidelines fundamentally introduces an unavoidable delay between the servos are given in review papers to tackle the control signal chattering making it practically difficult to achieve perfect synchronization in problem in SMC design (De Carlo. The first servo application of sliding mode in polar coordinates. shown in Fig. & Huang. (b) Without time difference. Koren (1991) founded the CCC and extended it for more Traditional algorithms such as P. sliding mode controller had required high frequency switching In practice. control. which layer or disturbance observers have been proposed. 1991). & Emami-Naeini. Zak. Chiu and Tomizuka ities. Su the idea of utilizing a contouring controller in conjunction with the et al. Control algorithms utilized for motion synchronization: (a) ‘Tandem’ and forced vibration. and Ting (2007) re-formulated it in drive dynamics. and servo closed loop dynamics are matched for accurate (2000) tacked this problem by developing a coordinate transforma. 1. Oldham. on the other hand. It caused control signal chattering and defeated its use in widely employed and has become a practical “gold standard” for precision servoing. 3b. Furthermore. Several countermeasures. synchronization (Yeh & Hsu. position of the brium point. In this scheme set point tracking sliding mode control law. Position(X) Master Feedback Master dx . Controller . 2009. uSMC ¼ ue þ sgnðsÞ ð1Þ . other. To overcome this.e ux . However. Sun. sliding errors of drives are minimized using high bandwidth controllers mode control laws mostly include a term such as and synchronization in-between is achieved passively through dynamics matching and the reference trajectory generator. 1988.1520 B. 3. Su. 2013) capabil- the coupled stability may become cumbersome. The master–slave cancels uncertainties that do not necessarily vanish at the equili- control scheme is shown in Fig. Sencer et al. (a) Double-sided milling machine and process and (b) Work-piece machined by double-sided milling.. and it may jeopardize overall robustness. Sencer and Altintas (2009) designed a robust controller (SMC) was introduced by Utkin (1977). Wenyu. In this scheme. & Mills. 2004). (b) ‘Master–slave’ control. / Control Engineering Practice 21 (2013) 1519–1530 Clamping plate Rotation Exit side Clamping plate Entry side Accelerometer Right facemill Left face mill Left face Right face mill mill C B A Feed direction of workpiece Feed direction of workpiece Right side of Left side of workpiece workpiece Fig. Li. 2005. Recent efforts are directed tion approach to directly design the controller on contour error towards improving bandwidth and disturbance rejection of the dynamics. proving Kon. Powell. Controller . (c) thrust cutting forces and (d) excitation force Fig. One of the main issue lies on the use of Another strategy is the indirect or so called the ‘tandem control discontinuous switching (saturation) function to derive stable approach’. 2011). much simpler methods have been around the sliding surface resulting in a discontinuous control favored. feedback principle (Franklin. Peng and Chen (2007) designed a back-stepping sliding drives using sliding mode controllers that are also robust to changes mode controller. Liu. such as the use of boundary master motor is used as reference command to the slave. Motor(Y) dy Slave Feedback Reference Master Feedback Position(X) Motion dx Master .. Motor(Y) dy Slave Feedback Fig.

y€ T presents drive's actual posi- nated motion. x€ ¼ ½x€ . x x Ka Kt 1 1 (2000) proposed an adaptive sliding mode controller without the rg ms+b s switching function and applied it successfully on precision control DAC Voltage Current Motor Inertia and Gear Backlash of ball-screw driven servo systems (Kamalzadeh & Erkorkmaz. Okwudire & Altintas. x ¼ ½x. Khan. Neglecting the non-linearities and |fflfflfflfflfflfflffl{zfflfflfflfflfflfflffl} e assuming that there is no physical coupling between the two F . Levant (1998) proposed an exact finite time conver. rg. point tracking and at the same time actively minimize the tion of the SMC in practice. in real industrial applications effect of those A continuous-time synchronization sliding mode controller is parameters becomes significant in precision motion synchroniza- designed to achieve both set-point tracking and robust motion tion. 5. 1998. These approaches deliver a chattering-free system. in dual servo drives is addressed. i. A part from the common literature. coordination that may be detrimental to the process accuracy. 4 and 5. Talole & Phadke. motion synchronization. x_ ¼ ½x_ . s. Synchronization errors The initial step in designing a robust synchronization sliding (ε) between two drives can be calculated from individual tracking mode control is to determine the plant model. Rigid body model of an industrial feed drive system. Bottleneck with the axis based approach lies in its tion. To address this problem.. Thus. . diagonal matrices that contain drives' equivalent inertia Mx ¼m/ However. 1998) of the servo 6e 7 6 7 ex 4 y 5¼40 1 5 ð5Þ system is considered. & Altintas. which can successfully estimate the derivative The model considers that amplifiers run in the torque control of the sliding variable but required a priori bounded trajectory. Ferrara. which in practice may not be measured and requires uy dy 0 My y€ 0 By y_ |fflffl{zfflffl} |fflffl{zfflffl} |fflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflffl}|ffl{zffl} |fflfflfflfflfflfflffl{zfflfflfflfflfflfflffl}|ffl{zffl} estimation.e. Spurgeon. A synchroniza- industrial application is utilized to prove the effectiveness of the tion error state is defined as the differential position error between controller. 1993) has been proposed where an integrator is used to eliminate the chattering drives. respectively. Particularly. e€ x parameter variations as well as un-modeled dynamics within the e€ ¼ € ¼ x€ ref M1 ðudBxÞ _ ð3Þ ey sliding mode control design scheme. & Nikkhah. been designed (Won & Hedrick. By Þ are output derivatives to be measured or observed. Nersesov. How. Introducing the reference motion trajectory. several non-linearities such as the friction. 2008). 2008). Sencer et al. 2008). the two servo drives. y_ T . Altintas. 2 3 2 3 ex 1 0 " # fied rigid body dynamics based model (Levant. amplifier saturation as well as cutting disturbances affect the In this research the problem of position synchronization. mode where u [V] is the control equivalent torque command Later. Dynamic sliding mode control Fig. Erkorkmaz. / Control Engineering Practice 21 (2013) 1519–1530 1521 which forces the control signal polarity to change depending on the Non-Linear Friction location of states around the sliding manifold. synchronization in dual servo systems. and Zhu u . the use of Disturbances due to SMC controllers has been limited in precision motion control Cutting Forces & Process practice. B. & Usai. and d [V] is the external disturbances reflected at the algorithm to stabilize non-linear systems that does not require control signal input. M ¼ diagðM x . Block diagram of the proposed sliding mode controller. Sira-Ramirez. such as the X and Y as ε ¼ ex ey ð4Þ 2. An to cutting disturbances and un-modeled dynamics. M y Þ and B ¼ diagðBx . xref ¼ ½xref . controller performance (Mori et al. Quantization Saturation Amplifier Gain Viscous Damping Ratio 2007). but a finite steady-state error may exist due to imperfect disturbance rejection. yT . velocity and acceleration kinematics. The rigid body model for a single servo drive ey ε 1 1 |fflffl{zfflffl} is presented in Figs. widening the tracking bandwidth or just improving KaKtrg and viscous friction Bx ¼b/KaKtrg terms normalized by the the disturbance rejection of a single drive does not necessarily amplifier Ka and motor torque constants Kt as well as the gear improve synchronization between the drives involved in a coordi- ratio. Ashrafiuon. u d M x B x_ gence differentiator. proposed controller generates continuous control signal The objective of the proposed controller is to satisfy accurate set- defeating the chattering issue and allows convenient implementa. In parallel. yref . and Levant (2003) proposed a 2-sliding delivered. In this work higher errors using the following transformation: order structural dynamics of the drives are neglected and simpli. 4. (Bartolini. backlash. robust discrete time sliding mode controls have Fig. Recently higher order as well as model reference sliding mode controller has been introduced to eliminate control signal chattering problem and achieve optimal tuning of the sliding mode gains (Muske. 2012. the un-coupled dual servo dynamics can be written in and at the same time achieve robust disturbance recovery. Controller's asymptotic convergence and synchronization errors between two servo drives that occur due disturbance adaptation are proven by the Lyapunov function. Okwidure (Okwudire & Altintas. 2000). the drawback of dynamic sliding mode control relies in the " # " # " #" # " #" # ux dx Mx 0 x€ Bx 0 x_ fact that it requires higher order derivatives of the sliding surface to  ¼ þ ð2Þ be known. It should be noted that passive motion synchronization where disturbances such as although the model is based on linear dynamics of the servo un-modeled friction or the cutting forces deteriorate motion system. 2010. Dual servo drive model where ex ¼ xref x is set-point tracking error of X-axis and ey ¼yref  y is for the Y-axis. The strategy of the proposed tracking error dynamics can be written from Eq. He also compared and proved robustness of the sliding mode controller against the industry standard P-PI cascade position control in precision linear motor system (Okwudire. (2) as controller is to introduce robust coupling between two drives and " # penalize synchronization errors that arise due to disturbances. matrix form as ever. 2009) designed a discrete time sliding mode control with mode compensation for vibration suppression and widening the tracking bandwidth of servo systems.

Synchronization sliding mode controller design convergent trajectory to the sliding manifold. (6) and (7) as d^ ¼ ^ ¼ ρκΓ S_ . In order to generate a stable sliding Sz mode control law. As shown in a 3 dimensional 1st order sliding surface (S) is defined as Eq. it eliminates the illusion that synchronization errors may converge terms in the Lyapunov derivative (Eq. Thus. due to Lyapunov energy function is computed from Eq. i. This allows the con. By imposing the negative definite. ^ troller to introduce higher static stiffness on the drive level and and due to the disturbance observer bounds imposed in Eqs. (10) ^ ≤0 can be observed.d^ ¼ ^ ¼ ρκΓ ðλFeþFe_ Þdτ: ð8Þ dy dy E_ ¼ ST ½λFM eþFM _ x€ ref FðudBxÞðd _ ^ T ðρÞ1 d_^ dÞ _^ ^ T ðρÞ1 d: As shown in Eq. E penalizes the total energy of the system. if remaining and (11). _ ¼ ST Sðd ^ T ðMρÞ1 d_^ dÞ ð13Þ bance observer d^ is defined. the tracking and synchronization errors. First step of designing a SMC is to select a stable sliding surface. and adaptation law will convergence speed of the errors on the sliding surface. errors states λ¼ diag(λx. where the final terms can be grouped as ST Fd^ þ ST ðFΓ T κÞðddÞ. (8). (5). integrations. which may generate  þ  þ within the predetermined bounds d^ x ≤ d^ x ≤ d^ x and d^ y ≤ d^ y ≤ d^ y . Eq. stable control law can be derived.λy. This ensures that the energy will decrease. the disturbance adaptation errors by S31 ¼ 4 Sy 5 ¼ λ33 F32 e21 þF32 e_ 21 : ð7Þ the ðddÞ ^ T ðMρÞ1 ðddÞ ^ term. κ y Þ is defined to impose bounds on the observed control where sgn or sat function is simply employed for forcing disturbances so that the adaptations. discontinuous control law (Slotine & Li. which penalizes deviation of errors It should be noted that since only slowly changing disturbances from the sliding manifold by integrating the sliding surface based are adapted for. Derivative of the inhered to cancel the open loop dynamics. However. which defines the will be pulled onto the sliding manifold.λε) is the sliding surface parameter. (12) as modeling errors of the servo system. κ x is defined for d^ x is written as _ þ ST Fd^ ¼ ST KS: E_ ¼ ST ½λFMe_ þ FMx€ ref FðuBxÞ ð17Þ 8 ^  9 < 0 if d x ≤dx and Sx ≤0 > > = where K33 ¼ diagðK x . K y . and substituting the disturbance observer expression from Eq. Owing the converge for slowly varying disturbances. More importantly. ρy Þ is the diagonal positive definite observer gain leads to matrix. ST ðFΓ T κÞðddÞ provide robustness against disturbances. Diagonal sliding mode control law. derivative of the Lyapunov function ðEÞ _ must be negative. and external disturbances such as the process forces.e. Sε . (8) ρ ¼ diagðρx . (15) is re-written by adding and subtracting S F d. namely the individual drive The following energy-like Lyapunov function: h i tracking errors and the synchronization errors in the dual servo E ¼ 12 ST S þ ðddÞ ^ T ðMρÞ1 ðddÞ ^ ð12Þ system. Unlike the conventional sliding mode κ ¼ diagðκx . In order to ensure zero steady state errors 2 and introduce robustness to the controller. (7) as to be zero. first depending 2 3 on the discrepancy of the errors from the sliding surface dynamics Sx 6 7 by the ST S term. it becomes a tuning variable for the sliding mode controller. error states are pulled away from the 1 h_T _^ T _ dÞ ^ þ ðddÞ ^ T ðMρÞ1 ðd _^ i _ dÞ E_ ¼ S S þ ST S_ þ ðd ðMρÞ1 ðddÞ sliding manifold. Sencer et al. As a result. 1 otherwise λFMe_ þ FMx€ ref þ FBx_ þ Fd^ þ KS ¼ Fu: ð18Þ . drive system synchronized by the proposed strategy. The key is to disturbance observer only in the drive level controller makes sure select an appropriate forcing function to generate a continuous that steady state tracking errors are strictly diminished. are kept derivative of the Lyapunov to be negative. and the control command is extracted from Eq. is postulated to derive a stable SMC control law for the dual feed In order to penalize both. the following condi- _ o0Þ: tion is postulated for achieving asymptotic stability ðEðtÞ This resembles a conventional integral anti-wind-up function. (3) and (5) as < 0 if d y ≤dy and Sy ≤0 > > = 2 3 2 3 κy ¼ 0 if d^ y ≥d and Sy ≥0 : ð11Þ e_ x e€ x > > y > > : . asymptotic nature of sliding mode controllers. 3 error states presented in Eq. For instance. In the implementation. transformation matrix:   E_ ¼ ST ½λFM eþFM _ x€ ref FðuBxÞ _ 1 0 0 ^ T ^ ^ Γ¼ ð9Þ þST Fd þ ST FdS FdST Γ T κðddÞ ð16Þ 0 1 0 |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} T^ S FdþS T ^ ðFΓ T κÞðddÞ so that the sliding surface is integrated in the Sx and Sy directions excluding the synchronization dynamics. / Control Engineering Practice 21 (2013) 1519–1530 where F3  2 is a transformation matrix. Augmented with the and similar for d^ y it becomes synchronization error state.1522 B. (13) is further expanded by substituting " # " # Z _ d^ x d^ x derivative of the sliding surface from Eqs. postulated observer integrates the sliding ¼ ST ½λFM e_ þFM x€ FðuBxÞ _ þ ST FdðddÞ ref ð14Þ surface to penalize deviation of error dynamics from the surface. The observer bounds ^ d^ þ . it functions as a switching operator that only becomes active in case if the adaptation for external disturbances exceed the predetermined bounds and the error states are not on a 3. (12). (17) as > : x > . derivative of the actual disturbances is considered on Eq. there is a plant inversion stability of the overall system could be achieved. 6 e_ 7 6 7 1 1 otherwise 4 y 5 ¼ Fe_ and 4 e€ y 5 ¼ Fx€ ref FM ðudBxÞ: _ ð6Þ ε_ ε€ It should be noted that although κ22 could be interpreted as time variant. (16)) could be forced to be to zero even if the tracking errors are non-zero. d_ ¼ 0. d^   can be selected as the saturation limits of servo d∈½ The overall objective of the controller is to minimize all the amplifiers to ensure robust operation. error dynamics for the dual servo 8 9 > ^  > system can be computed from Eqs. It should be noted that integration is E_ ¼ ST ½λFM eþFM _ x€ ref FðuBxÞ ^ _ þ ST FdST Γ T κðddÞ ð15Þ performed only on the drive level by the use of the following ^ T T Eq. the following distur. and second. K ε Þcontains positive definitive feedback κ x ¼ 0 if d^ x ≥d and Sx ≥0 : ð10Þ gains. 1988).

(8) and the sliding where Kp. (24) can be simplified to calculate sliding mode controller and gives guidelines for proper tuning for the transfer function governing synchronization error dynamics as control engineer's practice. It determines how mainly to improve the synchronization performance of a dual much control action is produced to penalize deviation of error motor servo system against disturbances and un-modeled states from sliding surface. Kd matrices are computed from Eq. and the tracking dynamics of vidual X and Y servo drive dynamics do not differ dramatically. derivative and the integral action matrices mapped it reveals that proposed controller achieves both set-point tracking from the sliding mode controller's gains. proposed controller is implemented on gains. The 1st term presents feed-forward mass compensation to system.y . proposed synchronization SMC control introduces nega- widen the tracking bandwidth and additional damping. Further proof on the xref þ dx Gx C xy Gx ey yref þ dy Gy C yx Gy ex error convergence is presented in Appendix section.. Depending on the application. ρ ¼ diagðρx . Grouping the terms |fflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflffl} leads to the final control law: GðsÞ Z   _ þ ðρΓ þ F† KλÞFe þ ρΓλF edτ þ ðF† λFM þ F† KFÞe_ ¼ u. It can be noted that in this case cross terms do not ▪ λ¼ diag(λx. affect the controller performance. Contrarily. K y . 2010) while adding stiffness to the system. (23) results in ð20Þ the coupled error dynamics: which does not contain any switching function. tracking bandwidths λx and λy can be set identical to achieve the same error decay. the servo system is not altered. are driving carts simulating a synchronized assembly line. e. At last. 6 Kpyx þ Kdyx s Kpyy þ Kdyy s þ Kiyy 7 Linear motors are driven separately by identical amplifiers that 6 |fflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflffl} s5 7 4 |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} run in torque. Experimental results of control bandwidth. K ε ¼ 1–3K x. Further expansion of Eq. (18) leads to the following control law: the closed loop dynamics using Eqs. Ki. ½Mx€ ref þ Bx where G(s) contains plant dynamics such as Gx ðsÞ ¼ 1= M x s2 þ Bx s and Gy ðsÞ ¼ 1=ðM y s2 þ By sÞ.e. drive's saturation limits. and the Dspace 1103 DSP Cyx Cyy (Digital Signal Processing) system is used for real time control. Increasing Kε will essentially intro. Re-writing surface S ¼ λFe þ Fe_ . K ε Þis SMC feedback gains.g.1. Guidelines for tuning the and at the same time motion synchronization. Synchronization bandwidth term λε determines the coupling action in low and mid frequency range 5. objective of the controller is ▪ K ¼ diagðK x . ρy Þ to eliminate steady state tracking errors a commercial double-sided milling machine tool (Mori et al. proportional. (26) reveals that synchronization errors are governed by the unmatched disturbances dx dy in a dynamically matched servo The continuous control law resembles a PID structure with 4 main system. B. i.g. Eq. Finally. By inspecting the terms in the pro. the reference trajectory com- _ þ ðρΓ þ F† KλÞF e þ ρΓλF edτ ðMx€ ref þ BxÞ manded to the drives is also identical xref ¼ yref.y Þ large enough e. Controller analysis Assuming that the plant dynamics are identical as well as the controller gains are matched for a servo system Gp ¼ Gx ≅Gy . ðxref yref Þ þ Gp ðdx dy Þ posed synchronization SMC law (Eq. λε ¼ 5–50λx.  ¼ : ð23Þ 0 Gy ðsÞ dy yref ey allows omitting the adaptation bound matrixðκÞ. into Eq.λy. The last tive cross term C xy to robustly introduce synchronization and 3 terms can be interpreted as multivariable PID gains that present coupling between the dual servos against disturbances. In contrast to a conventional tandem PID controlled servo terms. In this setup two identical 6 |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffls} |fflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflffl} 7 6 7 blushless Permanent Magnet (PM) linear motors. which results in |fflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflffl} ProportionalGain IntegralGain ðdx dy Þ ε¼ : ð26Þ † † C c þ 1=Gp C xy þ ðF λFM þ F KFÞ e_ ¼ u: ð21Þ |fflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflffl} DerivativeGain Eq. / Control Engineering Practice 21 (2013) 1519–1530 1523 Substituting disturbance observer from Eq. Motor X and 6 Cxx Cxy 7 CðsÞ ¼ 6 6 1 7: 7 ð22Þ Motor Y.λε) dictate desired bandwidth for tracking. Experimental implementation and MIMO analysis matrix: 2 3 The experimental dual motor test platform for controller 1 Kp 6 xx þ Kd xx s þ Ki xx Kp xy þ Kd xy s 7 implementation is shown in Fig. 6. ex ¼ . As a guideline λε should be selected higher than the axial parameters. SMC control law can also be implemented as a multi-input multi- output (MIMO) PID controller with the following compensator 5. synchronization errors would be ε ¼0. . |fflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflffl} GðsÞ CðsÞ assuming that the system works in the linear region or setting the " #" # " # 7 Gx ðsÞ 0 dx xref ex observer bounds ðd^ x. where controller performance is validated in real manufacturing operation against conventional as well as robust servo control Neglecting the feed-forward terms. Controller performance is first experimentally tested duce synchronization effect at higher frequencies and should on a dual linear motor slider setup where uneven external not be set much higher than the axis level. ey ¼ : ð24Þ 1 þ C xx Gx 1 þ C yy Gy 4. Therefore. (21). when proposed controller's gains are summarized as the disturbances are identical dx dy ¼ 0. and a MIMO analysis is presented to give an insight on the ▪ The final parameter set contains the disturbance observer parameter tuning. Sencer et al.y to disturbances are injected to the system to disturb synchronization. current control mode. This section presents insight on the proposed synchronization C c ¼ C xx ≅C yy and C xy ≅C yx . (20)). (3) and (22) results in Z " #" #" # Gx ðsÞ 0 C xx ðsÞ C xy ðsÞ ex Mx€ ref þ Bx_ þ F† λFMe_ þ F† KλFe þ F† KFe_ þ ρκΓ Sdτ ¼ u: ð19Þ 0 Gy ðsÞ C yx ðsÞ C yy ðsÞ ey It should be noted that F† is the pseudo-inverse of F. particular gains can be ε¼ : ð25Þ C c Gp þ 1C xy Gp grouped as follows: Z For synchronized motion tracking. eliminate excess noise entering the system. If indi. dynamics. proposed synchronization schemes.

7. This can also be seen by computing the interaction value of the Encoder feedback is taken from the motor side and the torque transmission matrix. solid lines present maximum singular value s ¼ maxfΣ LðjωÞ g B ¼ b=K a K t r g and dashed lines show the minimum singular value s ¼ minfΣ LðjωÞ g of the loop transfer matrix as controller gains are altered. As Motor X 0. identical except some high frequency fluctuations due to amplifier determinant of the multivariable loop transmission matrix noise.76467  10  3 4000 Tpos ¼0. −600 100 101 102 Freq[Hz] synchronization is not affected directly. Singular values of disturbance transfer function DðjωÞ ¼ SðjωÞGðjωÞ are shown in Fig. there is no coupling between the drives. 8c. which roughly doubles its inertia ðM x ¼ 0:000482 V=mm=s2 Þ . As around a higher frequency range. Difference between max- −100 imum and minimum singular value of the disturbance transfer Phase[deg] −200 function can be observed easily. when K ε is increased. Fig.y is delivered to amplifiers directly.15725 shown. open loop frequency responses of the drives are almost simply be tested from MIMO Nyquist criteria. The controller is diagonal terms: simply discretized using backward differentiation. which in return reduces the stability margins. Sencer et al. In fact. thus maximum Motor Y 0. overall system shows a more compliant response. indicates that if disturbances act in reverse directions. λx ¼ λy ¼ λε ¼ 1. λε is increased. for the initial parameter set. The real time control system is set to run at 1 kHz closed loop sampling. It can be observed that increase in the sensitivity is caused mainly by K ε . 7 10 V. Physically. As the synchronization gain. Identified rigid detðI22 þ LðjωÞÞ does not show any counter-clockwise encircle- body model parameters match the frequency response measure. Open loop frequency contrast. Observed from previous Motor masses M and viscous friction B parameters are identi.26006  10  3 0. Results are summarized in Fig. increasing λε does not affect the sensitivity unfavorably Motor Y(Rigid Body Identified) by just causing a slight increase and still improving synchroniza- −100 tion robustly. The observer satura- 7 Table 1 tion bounds are matched with the amplifiers' torque limits.18512 and minimum singular values are identical. 2009). In −50 Motor X(Rigid Body Identified) contrast. amplitude and (b) open loop phase. parameters. zation of the motors. MIMO 50 Motor X(Measured) sensitivity function is defined as SðjωÞ ¼ ðI22 þ LðjωÞÞ1 . controller parameters are given in Table 2 where both synchroni- zation gains λε and K ε are increased gradually. on the other Fig. This indicates coupling action −300 imposed by the controller.23367 Tneg ¼ 0. and the system becomes less compliant showing stronger resistance against external forces. 8d shows the interaction value. and presented in Table 1. In (Volts/mm/s2) friction resolution friction M ¼ m=K a K t r g (Volts/mm/s) (counts/mm) (Volts) Fig. Open loop frequency response of the dual linear motor setup: (a) open loop hand. ρy ¼ 1 7 1 6 7 4 4 7 5 1 3 ^d ¼ 7 10 5 x.y 1 Fig. are given in Fig. From maximum singular value. increasing K ε introduces stronger synchronization mostly response for both drives. 7. more exact Interaction valueðjωÞ ¼ crossðLðjωÞÞ⋅diagðLðjωÞÞ: ð27Þ methods can also be employed (Franklin et al. As shown in Fig. range.e. λy ¼ 1 1 1 2 6 K ¼ 0:02.25536  10  3 0. maximum amplification is increased gradually around all frequencies. Gx ðjωÞ and Gy ðjωÞ. 6. 8a.89919  10  3 4000 Tpos ¼0. 8b. i. / Control Engineering Practice 21 (2013) 1519–1530 Table 2 Parameter set of the controller. and amplifier delay is dominant at high frequency region. ment around the origin. parameter set given in Table 2. its effect can be observed especially around higher frequencies. 8a presents singular values of the multivariable loop transfer matrix LðjωÞ ¼ GðjωÞCðjωÞ Motor Equivalent mass Equivalent Encoder Coulomb indicating combined amplification of controller and the plant. K ¼ 0:02 7 25 1 6 x y 7 6 7 50 3 6 ρx ¼ 1. and K x ¼ K y ¼ Tneg ¼ 0. torque ripples and resolution of the resolver. 2001). Permanent magnet dual linear motor setup. it can be noted that if the disturbances acting on the motors are in the same direction. Stability of the controller can observed. Minimum singular value. and for validating the performance of the proposed controller for the they are ought to be compensated by the controller. and the Magnitude [dB] Motor Y(Measured) maximum singular value of sensitivity max sfΣ SðjωÞ g is presented in 0 ω Fig. increase in the synchronization gain λε introduces fied from time domain identification method (Erkorkmaz & coupling at lower frequency spectrum and at a wider range. In order to disturb the synchroni- MIMO analysis is performed to give an insight on the perfor. Tracking experiments are performed on the linear motor setup Those discrepancies are considered to be internal disturbance.. In contrast. which considers the ratio between cross and command ux. 8e. Controller Base gains Synchronization Synchronization parameter sliding surface gain feedback gain set ðλε Þ ðK ε Þ 2 3 1 λx ¼ 1. synchronization is deteriorated severely. a 10 kg payload mass is placed on the Motor mance of proposed controller for 5 different parameter sets.y ¼ Rigid body parameters of the dual linear motor system. or it does not cross the origin indicating ments apart from the effect of coulomb friction in low frequency that the control system is stable for all the given parameter sets.1524 B. 8. d^ x.14677 K ε ¼ 1. Tested X. Fig. In Altintas. the difference between −400 maximum and minimum singular values can be interpreted as −500 follows.

Furthermore. dominant frequency con- disturbance against motion synchronization during acceleration tent of a jerk-limited trajectory is generally below 20–30 Hz and sections. (d) interaction value of the loop transfer matrix. disturbances induced by the added mass would affect Fig. (c) singular values of disturbance transfer matrix. there is also presented in Fig.15 0. thus. MIMO analysis of the control system: (a) singular values of the loop transfer matrix. 9 are commanded to both motors. and for Motor Y. Gradient of this change is higher as compared to motion trajectories with jerk-limited trapezoidal acceleration increasing only K ε . increasing λε . 10c.5 Real −1 −50 −50 0 50 −1 −0. tracking errors trajectory. 8). and (e) multivariable Nyquist plot. It is assist the heavier axis while still tracking the reference motion also noticeable that as λε is increased from 1 to 50. As noted. Summarized in Fig. 10a and b presents tracking errors of the drives for the synchronization errors accordingly in low frequency range. Results of tracking tests are summarized in Fig. 11. B.25 Parameter Set 5 0. 8.5 Imag 0 0 −0. (b) maximum singular value of sensitivity matrix. Experimentally recorded synchronization errors are of Motor X decrease gradually. / Control Engineering Practice 21 (2013) 1519–1530 1525 30 Parameter Set 1 Parameter Set 2 20 Parameter Set 3 [dB] 10 Parameter Set 4 0 Minimum singular values of all cases coincide Parameter Set 5 −10 −20 0 1 2 10 10 10 freq [Hz] 2 1.5 [Mag] 1 0. in contrast. and identical back-and-forth a correlated rise. Motor X shows tracking errors Y tracking errors show some trajectory un-correlated and transient mostly correlated with the acceleration profile of the reference components since the controller commands the lighter Motor Y to trajectory (see Fig. from the original value (see Table 1). 9) and roughly 2 times that of the Motor Y.5 0 0.5 1 Fig.05 Parameter Set 3 Parameter Set 2 0 0 1 2 10 10 10 freq [Hz] 50 1 w Zoomed Vie 0. This can be attributed to the fact that λε has profile shown in Fig. 10.5 0 0 1 2 10 10 10 freq [Hz] 10 0 Maximum singular values [dB] −10 −20 Minimum singular values −30 −40 0 1 2 10 10 10 freq [Hz] 0. Sencer et al.1 0. more significant effect at lower frequency on the synchronization due to the added mass. reference trajectory provides a natural performance (see Fig.2 Parameter Set 4 [Mag] 0. Motor given set of controller parameters.

K x.y ¼ 7 10. Thereafter. fine-tuning is performed in the sided milling process was introduced in Fig. Proposed sliding controller for achieving robust synchronization performance. and maximum as material. the MIMO analysis. as the . J y ¼ 3:4301  104 V=rad=s2 showing close agreement in the inertia The proposed control algorithm is implemented and tested in a and slight difference in viscous friction. synchronization errors and torque signals could be logged least amplified direction of the controller and determined by the from the amplifier’s analog monitor channels. First. d^ x. controller gains could not be determined accurately. (b) reference velocity profile and (c) reference acceleration profile. The master–slave controller is tuned by the commer- 2 2 cial manufacturer of the control system. 12a. The bench- same instant. 2009) eCOG ¼ xr  ¼ . 10d. average of the tracking errors of Motor X and Y compared against the industrial standard master–slave approach xþy ex þ ey where both drives are equipped with P-PI (Franklin et al. 2007) is tuned with 7 the left and right cutters as they engage and disengage the identical base gains.y ¼ 7 10. In order to implement the Especially. λx. In order to ensure moderate synchronization. Reference motion trajectory: (a) reference position profile. λε ¼ 20. it is At last. 9. mode controller is compared against 2 controllers. How- interpreted that the center of gravity response governed by the ever.y ¼ 5.y ¼ 3. 2007) is implemented to benchmark by the increase agreeing with the MIMO analysis presented above. and the dSpace system was used for closed loop real-time jeopardize overall tracking performance. During the uous time adaptive sliding mode controller (ASMC) given in tracking experiments. COG errors do not show significant change (Kamalzadeh & Erkorkmaz. improves the synchronization more directly. Spindle X and well as the noise.y ¼ 1. The proposed controller's gains are set to 7 Spindle Y. where they are desired to enter the material at the λx. It can be turer. 1. Sencer et al. a contin- minimum singular value of the loop transfer matrix. conventional master–slave control is canceled by utilized heavily. ρx. Reflected rigid body plant dynamics are identified for the X and Y spindles to be. Bx ¼ 3:9813  104 V=rad=s. J x ¼ 5. K ε ¼ 1. Cutting forces disturb the synchronization between mark ASMC (Kamalzadeh & Erkorkmaz.y ¼ 5. Proposed controller's gains conventional manufacturing system. Second. control at 1 kHz. It should be noted that or so called the ‘center of motion errors’ (known as the ‘center of due to the black box structure of the commercial servo manufac- gravity’ (COG) errors in practice) is plotted in Fig. maximum COG errors are identical for all the switching amplifiers of the CNC into torque mode..y ¼ 1.2. which demands not only set are simply determined based on the above rigid body model and point tracking but also synchronization performance. ρx. This process is implementation on the double-sided milling CNC machine tool to based on cutting both sides of a thin work-piece with cutters accommodate for the un-modeled drive and process dynamics as mounted on identical left and right spindles. master– well as the RMS of synchronization errors is reduced around ∼50% slave control approach has already been utilized in this conven- in this setup. Industrial implementation and validation 3:7337  104 V=rad=s2 and By ¼ 4:3592  104 V=rad=s.y ¼ 2. / Control Engineering Practice 21 (2013) 1519–1530 400 300 Pos[mm] 200 100 0 0 1 2 3 4 time[sec] 7 8 9 10 600 vel [mm/sec] 400 200 0 −200 −400 −600 0 1 2 3 4 time[sec] 7 8 9 10 5000 acc [mm/sec2] 0 −5000 0 1 2 3 4 time[sec] 7 8 9 10 Fig. K x. Double. modifying the parameter sets validating that the proposed controller does not PLC. ð28Þ controllers. during acceleration segments where driving force is controllers. the proposed sliding mode controller. d^ x. This shows the effectiveness of the sliding mode tional milling machine as depicted in Fig.1526 B.

B.2 4 21 7 55 error[mm] 91 1 1. Maximum Synchronization Errors RMS Synchronization Errors 2. Double-sided milling control system and cutting experiment: (a) double- down to 25 counts after disengagement of the left and right tooth. 1b).5 55 0.7 0.1 mm of thick material on each side of the hardened steel (JIS SS400) plate. Since cutting is repetitive. In Rotary encoder 8192 p/rev the first experiment. 13a–c.5 −2 −1 0 2 4 6 time[sec] 10 0 2 4 6 time[sec] 10 2 1 eps [mm] 0 Parameter Set 1 Parameter Set 2 −1 Parameter Set 3 Parameter Set 4 −2 Parameter Set 5 0 1 2 3 4 5 6 7 8 9 time[sec] 1 0. Each spindle removes 0. Cutter 0 Parameter Set 1 2 3 4 5 Left Right Fig. 27 1. The synchronization Spindle errors rise up to 30 counts with the engagement and fluctuate Fig. 01 26 0. The cutter-work-piece engagement is detected by placing an accelerometer on the side of the work. (b) tracking errors of Motor Y and (c) synchronization errors. Evaluation of the synchronization performance.6 1. P-PI master– Left slave control presents the worst performance. Spindle Spindle proposed controller to achieve a fair and moderate comparison. sided milling machine master–slave control system and (b) cutting experiment. and a steel plate in the size of Cutter Spindle 400 mm  200 mm  13:2 mm is fed at a speed of 800 mm/min.5 1. which at max generates 300– 500 N of cutting force.4 0.5 −1 0 1 2 3 4 5 6 7 8 9 time[sec] Fig.5 1 ex[mm] ey[mm] 0 0 −1 −0. a section of the Workpiece experiments is presented in Fig. Synchronization and synchronization errors under conventional master–slave con- trol and tandem ASMC are compared against the proposed con- troller.4 Motor 1.5 eCOG [mm] 0 −0. 12. Experimental tracking and synchronization response: (a) tracking errors of Motor X. Feed Direction piece (see Fig. / Control Engineering Practice 21 (2013) 1519–1530 1527 1 2 0. 11.5 1 69 2 Slave unit Master unit 71 2. . Synchronization performance of the controllers is compared in Fig. 10.1 Rotary 6 Gear box 76 1 04 encoder 38 88 1 56 2 70 0. Cutting conditions are selected to be conventional where Cutter Toolpath Right spindles rotate at 200 rev/min.7 0. single tooth cutting is performed (see Fig. 12b). As shown. and (d) center of gravity motion errors. Sencer et al. 13.

5 11 11. Single tooth cutting experimental results. process quality is confirmed by measuring the vibration marks left 25–30% of the maximum motor load capacity is utilized during on the finished work-piece. 2007): synchronization error (b). acceleration pick  up (d). The cutter's path can be seen on .5 11 11. P-PI master–slave control: synchronization error (a).5 11 11. since synchronization errors are kept small and the master–slave design structure. 13c. robust disturbance controller introduces significant robustness and flexibility to the rejection characteristics of the ASMC can keep synchronization double-sided machining process. Some control signal Additionally. Proposed control: synchronization error (c). even if process parameters nization errors under the ASMC control scheme. As shown. From the manufacturing requirement of 7 4 counts at all times.5 11 11.5 11 11.5 Motor Y 0 0 0 Voltage[V] Voltage[V] Voltage[V] −0. acceleration pick  up (e).5 −1. such as the cutting speed. in the range of 7 8 counts.5 12 1 1 1 Slave Motor Motor X Motor X 0. motor X–Y control signal (i).5 Motor Y 0.5 −1 −1 −1 −1. it can also be seen that the slave results in a more favorable surface finish quality. Surface topography of this small controller delivers the most accurate synchronization and keeps section is significant as it is dynamically compliant due to clamp- errors steady by robustly coupling the drives for synchronization ing and thus easy to vibrate if the teeth do not enter with precision objective.5 12 10 10. or depth is altered. The performance of the proposed synchronizing sliding 30 mm section from the upper right corner of cut work-piece is mode control is presented in Fig.5 11 11. 13. the synchronization chronization errors occur as soon as the teeth engage/disengage performance would not deteriorate.5 11 11. Tandem ASMC (Kamalzadeh & Erkorkmaz. which in return deteriorates the manufacturing quality. steady under the proposed control. However. proposed taken and presented in Fig.5 −0.5 −2. Thus.5 −2. proposed sliding mode drives are not perfectly identical. which motor torque signal (Fig.5 12 20 20 20 Engage Engage Engage Engage Engage Engage Engage Engage Engage 10 10 10 [m/sec2] [m/sec2] [m/sec2] 0 0 0 −10 −10 −10 −2 0 −20 −20 10 10. synchronization errors are kept within the synchronization into the material. Fig.5 12 10 10. As a result.5 −1. acceleration pick  up (f).5 −2 −2 −2 −2.5 12 time[sec] time[sec] time[sec] Fig.5 12 10 10. motor X–Y control signal (h). which could not be satisfied the work-piece due to the fact that disturbances acting on the nor by the ASMC neither by the P-PI. It should also motor follows the master with a lag of approximately 20 ms due to be noted that. syn.5 11 11. 14.5 12 10 10. 13g). 13b presents the synchro. Sencer et al. motor X–Y Control Signal (g).5 10 10.5 11 11.5 12 10 10. Surface topography of a 20 mm  cutting. / Control Engineering Practice 21 (2013) 1519–1530 30 Error Bounds for Process Quality 30 30 20 20 20 eps [counts] eps [counts] eps [counts] 10 10 10 0 0 0 −10 −10 −10 −20 −20 −20 10 10.5 Master Motor 0. In both cases.1528 B. Similarly.5 −0. errors smaller.5 12 10 10. performance of the proposed controller on the chattering and effect of noise can be observed.

the error states Fe. Particularly. the sliding manifold S. 14b shows the vibration amplitude along the tooth's path. _ In conclusion. cutting with the master–slave or the As a result. i. Error [counts] 5 Proposed Control A continuous time sliding mode controller for motion synchroni- zation between dual servo systems is proposed in this research. A part from this particular experimental condition. First. 2007) Tooth 14 2 Vibrati on Valley 0 10 Marks −2 6 Peak −4 Proposed Control 2 −6 0 5 10 15 20 25 30 0 2 4 6 8 10 12 14 16 18 X[mm] (horizontal direction) Y [mm] Fig. (19): roughness and desired flatness in the finished product. which proves that E is a . 14. Even S_ in this condition. Lyapunov function (E) amplitude around 50% from 4 μm peek-to-peek down to 2 μm for is postulated in Eq. 15. proposed contro\ller's performance would in Eq. (10) since the settling time will be less after the cutter work-piece and (11). which is normal to the surface. This dynamics condition essentially favors the master–slave control since physi- ^ Fe€ þ λFe_ þ M1 KS ¼ FM1 ðddÞ. This results in a better surface forced to decrease by the control law in Eq.Fe_ and the tandem ASMC control would present even higher vibration marks estimated disturbances d^ are all bounded considering Eqs. −6 −8 9. E_ E€ ¼ 2ST KS. (6) reveals not decay. the second derivative of encoder counts.e. removal are visible from Fig.2) leads to the following sliding manifold between the left and the right spindle are presented in Fig. The controller OM presents practical implementation of the heavily researched Sliding ED 4 Master-Slave Mode Controller theory using conventional PID blocks for the benefit Synch. Synchronization Errors in 11-Teeth Cutting 6. and the cutting marks due to material providing equipment and experimental facilities. In this Fe€ ¼ Fx€ ref λFeF _ d condition due to the width of the plate several cutters are 1 1 FM KS þ FM d þ FM Bx 1 _ ðA:2Þ always in cut generating a physical coupling between X and Y spindles through the work-piece. (A. Measured synchronization errors Re-arranging (A. is negative definitive and E€ is bounded.1 Time[sec] 9. The authors would like to express their gratitude to Murata Machinery and Daido Amistar for their support in surface topography. Surface quality in single tooth cutting: (a) topography of the work-piece (sample section) and (b) relative vibration measured along cutter path. proposed controller outperforms the master– Eq. However._ SþM 1 ^ KS ¼ FM1 ðddÞ ðA:3Þ cal coupling is provided by the work-piece and subsequent cutter |fflfflfflfflfflffl{zfflfflfflfflfflffl} engagement does not induce severe disturbance fluctuation. synchronization sliding mode controller in a practical manufactur. Thus.8 Acknowledgment Fig. 15. Sencer et al. and its derivative is the given experimental condition. when the spindle rotation λFMe_ þ FMx€ ref þ FBx_ þ Fd^ þ KS ¼ Fu: ðA:1Þ speed is further increased. (12) is lower bounded. 11 teeth cutting is performed. / Control Engineering Practice 21 (2013) 1519–1530 1529 Path 6 Z [micron] (depth direction) Commercial P-PI Path ASMC Y[mm](vertical direction) 18 (Master-Slave) (Kamalzadeh & Peak 4 Tooth Erkorkmaz. This proves the effectiveness of the proposed the Lyapunov function becomes also bounded by Eq. ing application. Its performance is validated experimentally on a dual Time[sec] linear motor setup and also through implementation in a commercial −10 ZO 0 5 10 15 20 25 30 35 40 manufacturing system by machining experiments. Appendix The surface height is measured along the path of the tooth. Substituting the control law (A. 14a in the depth direction. B. This research is sponsored by the Japan Society for Promotion of Science (JSPS). Synchronization errors in 11-teeth double-sided cutting. It The convergence of the errors by the proposed sliding mode can be noted that the proposed controller reduces the vibration control is expanded in this appendix. (17). Vibration occurs in the z (depth) direction. _ x€ ref FM1 BxFM 1 ^ In the second experiment. and Fig.1) into the plant dynamics engagement. The 0 proposed controller shows robust coupling and motion synchroniza- tion in dual servo systems against disturbances and un-modeled −5 dynamics. Error [counts] 2 Proposed Control of control engineer's practice. E is positive definite lower bounded. Conclusions and final remarks 10 Master-Slave Synch. the structured derivation of 0 the control law allows variety of other motion systems such as dual −2 motor driven feed drives or gantry type precision motion systems to −4 benefit from the developed design scheme.3) shows that derivative of the sliding surface S_ is slave by keeping synchronization errors only within several bounded since ðddÞ ^ is bounded.

Part II – Precision contour controller design. C. Feedback control systems ((6th systems. A. (2008). (1977). 1393–1400. J. & Huang. Okwudire. H.. Barbalat's lemma (Slotine & Li. K. Proceedings of the IEEE. 265–272. (2008). At last. ASME Journal of Dynamics systems. E. Measurement. & Ting. & Altintas. Ferrara. & Erkorkmaz.. Measurement and Control. IEEE Transactions Khan. & Usai. Dual-stage servo Slotine.. IEEE/ASME Transactions on Mechatronics. flexible plates with high accuracy and high efficiency. C. Manufacturing automation: Metal cutting mechanics. H-L. Modeling and control of contouring errors for five- De Carlo. & Li. & Li. S. High speed CNC system design: Part II – of regenerative chatter vibration in simultaneous double-sided milling of Modeling and identification of feed drives. M. International Journal of Control. D. ASME Journal of Manufacturing Science and Engineering. E. 212–232. S. 47(10).. Measurement and Control. 49(1). J. 1–9. Erkorkmaz. T. H.. M. Sun. Y. J. synchronization errors ε ¼ ex ey both converge to origin proving Muske. Cross-coupled biaxial computer control for manufactur. controller for systems of relative degree two. W. Ashrafiuon. using a contour index approach. Kon. Powell.. Biaxial contouring control with friction dynamics high speed feed drives. (2000). X. which ensures that all the matica. 162–172. E. Adaptive manipulator control: A cast study. & Altintas. 9(1). 22(1). Sira-Ramirez. Measurement. Perfectly matched feedback control and its integrated ing systems. (1998). Robust exact differentiation via sliding mode technique. G. 15(3). W... A. Control Engineering on Robotics. (1991). Applied nonlinear control. stability of the overall controller. Accurate tracking controller design for uncertainty and disturbance estimator. Disturbance adaptive discrete-time sliding control control conference (ECC ’03). Oldham. (2013). (2007). and Control. Slotine.. and CNC design. formation Fe. 58(1).. 329–345. 202–207. P. 387–390. Hiramatsu. Transactions of ASME. International Journal of Machine Tools flexible plates by speed difference. 1–5. J. & Emami-Naeini. & Phadke. Okwudire. Y. & Lo. (1993). E. Transactions of nonlinear multivariable systems: A tutorial. Simultaneous double-sided milling of errors slide/converge to the origin. L. (2007). sliding mode control. Transactions of ASME. ment and Control. Yeh. Y. K. 126(1).. Variable Structure systems with sliding modes in motion systems: A task coordinate frame approach. ed. Journal of Dynamic Systems. 335–338. Y... (1988). and Control. (2003). Journal of Dynamic Systems. Levant. Y. (2011). 212–222. B.. & Nijmeijer. 59(2). (2000). 47. 35. Shamoto. Integration of saturated PI synchronous Jinbang. T. R. (2009). 899–907. (2000). (2001). UK. Contouring control of biaxial systems based flexible joint robots via estimated state feedback. B.. IEEE systems and vibration compensation in computer hard disk drives. Horowitz. Journal of Dynamic Systems. Cambridge. (2004). 134(2).. 41(10).-L. 130–139. S-0 as t-1. Mori.. M. high speed drives. A. 34(3). A. C.. Koren. Transactions of ASME. (2006). (2000). 130(3).-S. 689–698. 265–270. Dynamic stiffness enhancement of direct- Altintas. 126. A. Variable structure systems with sliding modes. & Hedrick. Liu.. Rodriguez-Angeles. Precision Engineering. Utkin. S. 76(3). T. Altintas. Contouring control of machine tool feed drive Sabanovic. J. / Control Engineering Practice 21 (2013) 1519–1530 uniformly continuous function. Optimal sliding mode cascade control for stabilization of underactuated nonlinear systems. Simple output feedback 2-sliding on Automatic Control. Nersesov. Su. vibrations. T. M. Auto- 1991) dictates that E-0. ey and the 416–423. Transactions of the ASME.. (2003). S. W. 291–305. 1–11. Y. & Mills.1530 B. 102. 1039–1061. (2010). In Proceedings of the European Won. X. considering the trans. (2007).. H. on polar coordinates. ASME Journal of Dynamic Systems. Detection and reduction of control and PD feedback for control of parallel manipulators. & Matthews. and Manufacture. & Chen. K. Chattering avoidance by second-order Manufacture. Measurement and Control. Nishimura. P. 995–1003. Model following sliding mode control based on Kamalzadeh. E. Synchronizing tracking control for Chen. C. Annals of CIRP. 123. C. 57. 379–384. Cambridge University Press. Mori. 22(2). 131(5). 1487–1509. 031007.). W. (1988). NJ: Prentice Hall. Y. International Journal of Machine Tools and Bartolini. S. E. Advanced controllers for feed drives. E. J. 241–246. 41(2). & Altintas. Machine tool driven machine tools using sliding mode control with disturbance recovery. Annals of CIRP.. 212–223. & Zhu. 1542–1555. R. F. Talole... Ren. Sencer et al. Systems Technology. C. Levant. CIRP Annals. R. & Nikkhah.. Wenyu. (2005)... 7(3). C. Zak. & Kurata. 7(2). Variable structure control of axis machine tools. IEEE Transactions on Automatic Control. (2012). & Shamoto. L.. J. ASME Journal of Dynamics systems... IEEE Transactions on Control control – A survey. (1991). Anwen. . with application to engine speed control. K. Y. K. (1991). S. Control Transactions on Automatic Control. Measure- Koren. A. Journal of Dynamics Systems Measurement and Control. & Yu. Chiu. S. Suppression Erkorkmaz. Hiramatsu. (2009). IEEE Transactions on Industrial Informatics.. & Altintas. D.. 43(2). Li. Z.. Y. Tracking error control of flexible ball-screw References drives using a discrete-time sliding mode controller.. (2009). International Journal of Machine Tools and Manufacture. H. S.. K. design for multiaxis motion systems. 131(3).. it can also be concluded that tracking ex . A. 547–557. IEEE Transactions middle frequency resonance for an industrial servo. (2010).. 21(7).. C.. G. Spurgeon. Practice. Prentice Hall. Y. C. Sliding mode controller design for Peng.. & Tomizuka. Sencer. S.. & Hsu... H. CIRP Annals. Y. On the dynamical sliding mode control of nonlinear Franklin. (1998). Engineering Practice. I. V. S-L... A. K. 33(11).. T. K.