This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. IEEE/ASME TRANSACTIONS ON MECHATRONICS

1

Design and Robust Repetitive Control of a New Parallel-Kinematic XY Piezostage for Micro/Nanomanipulation Yangmin Li, Senior Member, IEEE, and Qingsong Xu, Member, IEEE

Abstract—This paper presents mechanism and controller design procedures of a new piezoactuated flexure XY stage for micro-/ nanomanipulation applications. The uniqueness of the proposed stage lies in that it possesses an integrated parallel, decoupled, and stacked kinematical structure, which owns such properties as identical dynamic behaviors in X and Y axes, decoupled input and output motion, single-input-single-output (SISO) control, high accuracy, and compact size. Finite element analysis (FEA) was conducted to predict static performance of the stage. An XY stage prototype was fabricated by wire electrical discharge machining (EDM) process from the alloy material Al7075. Based on the identified plant transfer function of the micropositioning system, an H∞ robust control combined with a repetitive control (RC) was adopted to compensate for the unmodeled piezoelectric nonlinearity. The necessity of using such a combined control is also investigated. Experimental results demonstrate that the H∞ plus RC scheme improves the tracking response by 67% and 28% compared to the stand-alone H∞ for 1-D and 2-D periodic positioning tasks, respectively. Thus, the results illustrate the effectiveness of the proposed mechanism design and control approach. Index Terms—Flexure mechanisms, micro-/nanopositioning, motion control, parallel manipulators, piezoelectric actuation.

I. INTRODUCTION MICROPOSITIONING system is a crucial component for the robotic micro-/nanomanipulation system dedicated to automated ultrahigh-precision positioning and assembly in micro-/nanometer scales, such as bio-cell manipulation [1], optical fiber alignment [2], and scanning probe microscopy (SPM) [3]. Such kind of applications call for micropositioning systems capable of positioning with high resolution, high repeatability, and high bandwidth. Although micromanipulation is an existing topic in both academia and industry, this technique has gained extensive recent attentions in the research domain. The realization of a high-performance manipulation is a challenging work because it requires an integrated consideration of

A

Manuscript received April 29, 2010; revised August 2, 2010, October 28, 2010, and March 25, 2011; accepted May 19, 2011. Recommended by Technical Editor D. Caldwell. This work was supported in part by the Macao Science and Technology Development Fund under Grant 016/2008/A1 and in part by the Research Committee of the University of Macau under Grant SRG006-FST11XQS. The authors are with the Department of Electromechanical Engineering, Faculty of Science and Technology, University of Macau, Macao, China (e-mail: [email protected]; [email protected]). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/TMECH.2011.2160074

all the technical issues including mechanical joints, actuators, sensors, kinematic schemes, materials, fabrication processes, control strategies and so on. In the literature, flexure-based compliant stages [4], [5] are popularly recognized to achieve ultrahigh-precision positioning. The reason lies in that the compliant mechanisms deliver motions by making use of the elastic deformations of notch hinges instead of conventional mechanical joints, which renders the stage with merits of free of backlash, zero friction, repeatable motion, and vacuum compatibility. Besides, piezoelectric actuators (PZTs) [6]–[8], are frequently used for the actuation since they are capable of positioning with (sub)nanometer level resolution, large blocking force, high stiffness, and rapid response characteristics. Some piezostages have been developed in research laboratories and even commercialized on the markets (e.g., the piezostages manufactured by Physik Instrumente GmbH). This research investigates XY stages due to their promising applications in micro-/nanomanipulation and positioning. According to kinematic schemes, the XY stages can be classified into two categories in terms of serial and parallel ones [9]. In serial stages, two one-degree-of-freedom (1-DOF) linear stages are jointed consecutively as a stacked or nested architecture in majority of the cases [10]–[13], where the moving (output) platform is only supported by the last stage. The serial stage has the advantages of simple structure and control strategy since the two stages can be treated independently. However, it exhibits certain disadvantages including high inertia, asymmetric dynamics in X and Y axes, and cumulative errors. In contrast, parallel stages are based on parallel kinematics, whose moving platforms are supported by all the connecting limbs simultaneously [4], [14], [15]–[18]. By this way, the performances of low inertia, high load-carrying capacity, identical dynamic behaviors in X and Y axes, and high accuracy can be relatively easily achieved. Therefore, parallel stages are more appropriate for micro-/nanopositioning from the above considerations. This research is concerned with macro-scale parallel stages with the size of tens to hundreds of millimeters, although the stages can also be fabricated in micro- and mesoscales in accordance with the pertinent application requirements [19]. Within macro-scale, the existing parallel XY stages [4], [14], [15]–[18], are all designed as planar monolithic structures, which are free of assembly and can be easily manufactured by such process as wire electrical discharge machining (EDM). In consequence, the monolithic stage potentially occupies a large planar dimension since it is fabricated from a piece of material. Under such

1083-4435/$26.00 © 2011 IEEE

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. 2

situations, it may not be suitable for the applications where the micro-/nanomanipulation inside a limited space, e.g., inside a scanning electron microscope [20], is required. One motivation of the current research is to overcome the shortcomings of monolithic XY parallel stages. Specifically, a concept of parallel stage with stacked structure is presented to obtain a compact size. By this way, the inherent advantages of parallel kinematics are maintained. Moreover, under the circumstances where the stage is underactuated or sensory feedback of the output motion is not permitted, a decoupled XY stage with proper calibration is desirable [16]. In order to isolate/protect actuators and to obtain a decoupled output motion, it has been recommended in [17] that a totally decoupled stage is preferred to be designed with both input and output decoupling properties. In this paper, a parallel piezostage with both stacked and totally decoupled structure is proposed for micro-/nanomanipulation applications. The major problem of a piezoactuated system arises from the nonlinearity due to hysteresis and creep effects. In order to fulfill the requirements of ultrahigh-precision positioning, the hysteresis has to be suppressed by an appropriate control scheme. Generally, the existing schemes fall into two categories, i.e., hysteresis model-based and hysteresis model-free methods. In the first category of approaches, a hysteresis model (e.g., Preisach model) is generated and used to construct an inversionbased feedforward compensator [21]–[23]. It has been shown that the inversion-based compensation can achieve an accurate positioning, whereas the result is very sensitive to the model accuracy [3], [24]. Fortunately, the combination of feedforward with feedback control can be adopted to suppress the hysteresis as well as creep effects [25], [26]. Concerning the second category, its main advantage lies in that no hysteresis model is required. The unmodeled hysteresis is considered as an uncertainty or a disturbance [27] to the nominal system, which is tolerated by a robust or adaptive controller. For instance, the applications of sliding mode control [28], H∞ robust control [29], fuzzy logic control [30], and neural network control [31] have been reported. In addition, H∞ combined with iterative learning control is applied to nanopositioning systems [32], and the integration of inversion-based feedforward and H∞ control is realized in PZTs [33]. It is noticeable that another method named charge actuation can be used to reduce hysteretic behavior of PZT. If a PZT actuator is driven by a charge source, the hysteresis becomes almost negligible. This technique has been known for some time, and there has been renewed interest in this subject recently [34]. In this paper, a robust repetitive control is adopted to suppress the unmodeled hysteresis and to alleviate tracking errors for periodic reference input. More specifically, a 2-DOF control framework employing an H∞ robust control combined with a plug-in repetitive controller is used to suppress the nonlinearity in the piezo-driven positioning system without modeling the hysteresis and creep effects. In the literature, although both H∞ control [29], [32], [33] and repetitive control (RC) [35]–[38] have been well studied, the investigation on their combination is still limited [39], [40]–[42]. It remains unclear whether such a combined controller is necessary for piezo-driven systems

IEEE/ASME TRANSACTIONS ON MECHATRONICS

Fig. 1. (a) A 1-DOF linear guiding mechanism using a compound parallelogram flexure, (b) a compound bridge-type displacement amplifier with certain tolerance of lateral load, and (c) a PP limb with decoupled 2-DOF translations.

which suffer from hysteretic and creep nonlinearities. In this paper, the necessity to introduce such a combination for a micropositioning system is presented in detail by revealing: 1) the superior performance of H∞ over traditional proportionalintegral-derivative (PID) control, 2) the better performance of H∞ plus RC over PID plus RC, and 3) the better performance of H∞ combined with RC over stand-alone H∞ and RC approaches. In the rest of the paper, the mechanical design procedures of an XY stage are described in Section II, where the stage performance is tested via finite element analysis (FEA). In Section III, a prototype XY stage is fabricated along with plant model identified. Then, an H∞ combined with repetitive control is presented in Section IV to improve the tracking ability of the positioning system. The performance of the piezostage system is verified by experiments conducted in Section V. Finally, Section VI concludes this paper.

II. MECHANISM DESIGN AND FINITE ELEMENT ANALYSIS A. Mechanism Design and Assembly Process To design a decoupled XY stage with parallel structure, a 2-PP (P stands for prismatic joint) mechanism is adopted due to its simple structure. In what follows, the design procedures of a PP limb with decoupled translations are outlined. It is known that the compound parallelogram flexure [see Fig. 1(a)] provides 1-DOF ideal translation (dx ) if a force Fx is applied on its output stage. At the same time, the compound bridge-type displacement amplifier [see Fig. 1(b)] delivers both an amplified translational output motion in vertical direction (y-axis) and a spring preload for the actuator. Besides, this type of amplifier has much larger input stiffness and larger lateral stiffness than the conventional bridge-type amplifier [43]. The large input stiffness calls for an actuator with large blocking force and stiffness. Hence, PZT is the most suitable actuator for driving the amplifier. On the other hand, the large lateral stiffness indicates that the output end of the amplifier can tolerate a large lateral load. This merit offers protection for the inner PZT which can only tolerate small magnitude of lateral load. Therefore, combining the preceding 1-DOF stage with the amplifier, a PP limb with decoupled translations along the x- and y-axes is selected as shown in Fig. 1(c). This PP limb is employed as a basic module to design an XY parallel stage with decoupled structure.

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. LI AND XU: DESIGN AND ROBUST REPETITIVE CONTROL OF A NEW PARALLEL-KINEMATIC XY PIEZOSTAGE

Fig. 2. Top view of a two-layer XY stage with parallel, decoupled, and stacked structure.

3

Fig. 4. Parameters of (a) a right-circular flexure hinge and (b) one quarter of the displacement amplifier. TABLE I ARCHITECTURAL PARAMETERS OF AN XY STAGE

Fig. 3. (a) Exploded view of CAD model for the XY stage. (b) Assembled model of the XY stage.

With different arrangement schemes of two such limbs, various architectures of an monolithic XY stage can be designed. Furthermore, four PP limbs can be adopted to create a doublesymmetric architecture. For instance, a 4-PP stage is developed in previous works [44] of the authors. As monolithic structures, the XY stages are easy to fabricate. However, they own a relatively large dimension in plane. Hence, they are not suitable for the applications where the micro-/nanomanipulation inside a limited space is required. Although the stage structures can be further enhanced to utilize the planar space efficiently to get a more compact 2-D structure, it still occupies a large area if longstroke PZT actuators are employed to obtain a large workspace. To conquer such drawbacks, an XY stage with a two-layer stacked structure is proposed as illustrated in Fig. 2. Using an orthogonal assembly of the two PP limbs, the stage produces decoupled output motion. While comparing to monolithic XY stages, it has a more compact size. Its limitation lies in that, assembly is required to tie the two limbs and output platform together to construct the whole stage. Considering that the fixing holes can be easily machined with a fine tolerance (e.g., ±5 μm), precise assembly of the components is not impossible nowadays. The induced slight squareness errors can be tolerated by the flexures suffering from elastic deformations. Although four PP limbs can be employed to design a double-symmetric XY stage to reduce parasitic motions and thermal gradient effects, the two-limb version as shown in Fig. 2 is adopted in this paper to demonstrate the conceptual design of a compact XY stage with integrated parallel, decoupled and stacked structure. The exploded view of a computer-aided design (CAD) model for such an XY stage is graphically shown in Fig. 3(a). The assembly is completed through the following processes.

1) First, limb 1 (with PZT 1 embedded) is fixed on the base through two bolts, where spacer 1 is used to support the limb 1. 2) Second, the output platform is fixed at limb 1 through two bolts. 3) Third, limb 2 is placed on the top of limb 1 separated by spacer 2, and then mounted on the base by three bolts. 4) Forth, the output platform is assembled on limb 2 via another two bolts. 5) Afterwards, the top platform is mounted on the top of the output platform in order to support external weights. 6) Finally, two displacement sensors are fixed on the base for the measurement of the output platform positions. The assembled model is illustrated in Fig. 3(b). It is noticeable that although the right-circular shape flexure hinge [see Fig. 4(a)] is adopted in the current research for illustrations, other types of hinges can also be employed. B. Performance Evaluation with FEA Finite element analysis (FEA) is performed with ANSYSto predict the performance of the designed XY stage. The architectural parameters of the stage [see Fig. 1(a) and Fig. 4] are tabulated in Table I, and the physical and mechanical parameters of the alloy Al7075 used in the FEA are: Young’s modulus = 71.7 GPa, yield strength = 503 MPa, Poisson’s ratio = 0.33, and density = 2.81 × 103 kg·m−3 . A 3-D finite element model is established with 20-node element SOLID186. The mesh model is created with a medium mesh size and refined at the flexure hinges to obtain smaller mesh size therein in order to achieve more accurate simulation results. Besides, zero displacements are assigned on the surfaces of the mounting holes to constrain the mechanism. In the static structural FEA simulation, a force is applied on the input ends of the amplifier 2 (in limb 2), which produces

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. 4

IEEE/ASME TRANSACTIONS ON MECHATRONICS

Fig. 5. Top view of the deformed finite element model of the XY stage, where the deformations are exaggerated by 30 times for a better view.

depicted in Fig. 3. The assembled XY stage has an overall dimension within 116 × 116 × 46 mm3 . Regarding to actuation, the actuators should be selected with high enough stiffness to drive the stage which has an input stiffness (6.99 N/μm) as evaluated by FEA in Section II-B. Two 20 μm-stroke PZTs (model PAS020 with stiffness of 14–208 N/μm ± 20%, from Thorlabs, Inc.) are adopted in this research. In order to measure the output displacements of the moving platform, two laser displacement sensors (Microtrak II model LTC-025-02, from MTI Instruments, Inc.) are used. Concerning control apparatus, a dSPACE DS1005 (from dSPACE GmbH) rapid prototyping system equipped with DS2001 A/D and DS2102 D/A modular boards are employed. The D/A board produces an analogy voltage which is then amplified by a two-axis voltage amplifier (BPC002 from the Thorlabs) to drive the PZTs. Besides, the sensor outputs are passed through signal conditioners and then acquired simultaneously by using the A/D board. The overall experimental setup is sketched in Fig. 6. Using a signal of 0–10 V provided by the D/A board, the voltage amplifier generates an output signal of 0–75 V to drive the PZT actuator, which produces a workspace range of 138 × 128 μm2 for the XY stage. B. Plant Model Identification

Fig. 6. Connection scheme of experimental apparatus for the developed XY micropositioning stage.

a translational motion along the y-axis direction as shown in Fig. 5. This figure also illustrates the working principle of the XY stage. Since both limb 1 and limb 2 are connected to the output platform, the output end of limb 2 is also deformed as shown in the figure. This is one of the major properties of the parallelkinematics mechanism compared to the serial-kinematics one. By extracting the input and output displacements, the amplification ratio and input stiffness of the stage can be derived as 6.45 and 6.99 N/μm, respectively. By comparing the maximum parasitic translation along the x-axis with the primary output motion in y-axis, it can be derived that the parasitic translation is less than 0.5% and the maximum in-plane rotation is less than 0.023 mrad, which demonstrates that the stage motion is decoupled. III. PROTOTYPE DEVELOPMENT AND PLANT MODEL IDENTIFICATION In this section, the development of a prototype XY parallel micropositioning stage is described and the plant model is identified for subsequent studies. A. Prototype Fabrication and Experimental Setup One prototype of the XY stage is fabricated as shown in Fig. 6. As far as materials are concerned, Al7075 alloy is more elastic and has a lower density compared to steel. Thus, Al7075 plates are used to fabricate the two monolithic limbs of the XY stage, which are then assembled together as per the processes

The system plant consists of the PZTs, XY stage, and laser sensors. Since it is difficult to obtain an accurate physical model of the whole system, a linear model of the plant is experimentally identified with respect to a specific operating point. The identified model is called a nominal model for the system, and the uncertainty (including hysteresis and creep effects) is expected to be compensated by implementing a robust controller. Swept sine waves with the frequencies covering the bandwidth of interest (1–600 Hz in the current research) are created with “Chirp” block in Simulinkand applied to voltage amplifier to drive the PZTs. When the input signal has a large amplitude (compared to the input voltage range of 10 V), piezoelectric hysteresis effects become significant. In contrast, the plant model can be assumed to be linear over a short operating range [29]. Thus, to obtain a linear model of the system, a low voltage of 0.5 V (peak-to-peak amplitude), which only accounts for 5% of the total input range, is selected to reduce the hysteresis effect. With such a small input signal, the modeling error caused by hysteresis can be neglected. To identify the model, PZT 1 is first actuated with the swept sine-wave voltage. At the same time, the displacements of the output platform in both x- and y-axes are measured and recorded with a sampling rate of 2 kHz. To reduce the noise effect, the data collection procedure is experimentally repeated five times and the five sets of output data are averaged as the sensor output. With the two PZTs actuated individually, the frequency responses (see Fig. 7) in the two axes are obtained by fast Fourier transform (FFT). The frequency response of the i-axis motion induced by driving the PZT k is described by Gik (jω) = di (jω)/uk (jω), where the axis index i = x and y, and actuator index k = 1 and 2, respectively. It is noticed that, with PZT 1 driven, the responses in the y-axis are 22 dB

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. LI AND XU: DESIGN AND ROBUST REPETITIVE CONTROL OF A NEW PARALLEL-KINEMATIC XY PIEZOSTAGE

Fig. 7. Magnitudes of frequency responses of the two-input two-output micropositioning stage. The input is the voltage applied to PZT, and output is the displacement measured by laser sensor. G x 1 represents the x-axis frequency response when the stage is driven by PZT 1.

lower than that in the x-axis in low-frequency range except for the frequencies nearby the resonant mode. With PZT 2 driven individually, similar results are obtained. Thus, the two axial motions of the XY stage are well decoupled, and two singleinput-single-output (SISO) controllers can be designed for each axis. In comparison with the FEA result which shows a crosstalk of 0.5%, the experimental result (7.9%) is more significant. The discrepancy may come from the machining error and the installation error of the stage. In addition, the misalignment of the laser sensors with respect to the target platform and the perpendicular error of the two sensors lead to non-orthogonality of the metrology coordinate frame and also result in crosstalk between the two working axes. The input-output data sets are used to identify the plant transfer functions (Gx1 and Gy 2 ) in the two axes by resorting to the command “spafdr” in Matlab,which estimates the model from the frequency responses. For instance, the frequency response of the x-axis plant model Gx1 identified at the null position (0 V initial input for PZT) is displayed in Fig. 8. It is observed that there are several resonance modes in the plant. The one with the most significant magnitude occurs at 183 Hz. A 16th-order nominal model Gxn = Gx1 with the unit of μm/V is identified in continuous-time form as shown in (1), at the bottom of this page, which matches the system dynamics well in the range of 1 to 600 Hz. The response of the corresponding discrete-time

Gxn (s) =

(s2

5

Fig. 8. The x-axis frequency responses obtained by experiment (solid line), identified 16th-order continuous-time model (dashed), and converted discretetime model (dash-dot) with a sampling rate of 2 kHz, which match one another well in the frequencies below 600 Hz.

Fig. 9. Variation of the plant frequency responses (G x 1 ) at different operating points (0–4 V).

model generated using a sampling rate of 2 kHz is also depicted in Fig. 8. It is observed that the discrete-time model approximates the continuous-time one quite well using the selected sampling rate. Besides, the system frequency responses are generated at different operating points (0 to 4 V) for the PZT. Comparing the frequency responses as plotted in Fig. 9, one can observe that the responses vary as the changing of the operating point.

−675.1937(s − 7787)(s2 + 1.225s + 4.127 × 105 )(s2 + 39.91s + 1.075 × 106 ) + 24.07s + 3.721 × 105 )(s2 + 53.84s + 9.283 × 105 )(s2 + 23.48s + 1.307 × 106 )

×

(s2 − 28.21s + 4.329 × 106 )(s2 − 3175s + 7.989 × 106 )(s2 + 66.58s + 1.002 × 107 ) (s2 + 2318s + 3.646 × 106 )(s2 + 23.07s + 4.831 × 106 )(s2 + 52.09s + 1.003 × 107 )

×

(s2 + 34.5s + 1.207 × 107 )(s2 + 93.46s + 1.317 × 107 ) (s2 + 16.26s + 1.203 × 107 )(s2 + 82.15s + 1.313 × 107 )

(1)

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. 6

IEEE/ASME TRANSACTIONS ON MECHATRONICS

Fig. 10. Block diagram of the mixed-sensitivity H ∞ robust control system, which is equivalent to a standard two-part robust control system. G x is the plant model, G c denotes the H ∞ controller, and P x is the augmented plant model.

In addition to the nonlinear hysteresis effect, the plant model variation also necessitates the design of a robust controller. IV. ROBUST REPETITIVE CONTROLLER DESIGN The uncertainties in terms of model incompleteness and model variation are the essential of piezo-driven system attributed to the hysteresis and other nonlinearities. To compensate for the uncertainty, an H∞ -based robust control scheme is designed in this section. A. H∞ Robust Control Design In comparison with other feedback control approaches, the mixed-sensitivity H∞ robust control exhibits the advantage of combining the performance and robustness requirements in one controller design procedure. The achievement of the performance and robustness for H∞ controller is heavily dependant on the selection of the weighting functions. The block diagram of the closed-loop H∞ robust control is shown in Fig. 10, where W1 , W2 , and W3 represent the weighting functions. z1 , z2 , and z3 are weighted signals. The goal of H∞ controller design is to minimize the transfer functions of these weighted signals. The sensitivity function S and complementary sensitivity function T of the closed-loop system can be obtained as S = 1/(1 + Gx Gc ) and T = 1 − S, where Gx is the plant model and Gc denotes the H∞ controller to be designed. In order to design the controller, the three weighting functions (W1 , W2 , and W3 ) need to be determined to shape the sensitivity function S, transfer function Gc S, and complementary sensitivity function T, respectively. With reference to Fig. 10, we can deduce that S denotes the transfer function between the reference input xd and tracking error e, Gc S relates xd to the control input u, and T relates xd to the sensor output x as well as the sensor noise n to sensor output x. In light of the small-gain theorem, the controller Gc can be synthesized by minimizing the transfer functions of the weighted signals, i.e., ⎤ ⎡ W1 S ⎥ ⎢ N ∞ = sup σ ¯ [N (jω)] ≤ 1; N = ⎣ W2 Gc S ⎦ (2) ω ∈

W3 T

where  • ∞ denotes the H∞ norm to measure the size of a transfer function, which is considered as the maximum singu-

Fig. 11. Singular value plots show that the sensitivity function S and complementary sensitivity function T lie below the inverse of the weighting functions W 1 and W 3 , respectively.

lar value σ ¯ [N (jω)] of the transfer function with the argument ω denoting the frequency. Such a controller guarantees that W1 S∞ ≤ 1, W2 Gc S∞ ≤ 1, and W3 T ∞ ≤ 1. It follows that the inverse of the three weighting functions provide upper bounds on the corresponding transfer functions (see Fig. 11). Therefore, the key technique of designing an H∞ robust controller lies in the selection of the weighting functions which are stated below. The weighting function W1 is desired to exhibit high gains at low frequencies and low gains at high frequencies. This ensures that the sensitivity function is small at low frequencies in order to guarantee a small tracking error within the bandwidth of interest. In contrast, the weighting function W3 is chosen such that it has low gains at low frequencies and high gains at high frequencies. This strategy is taken to ensure that the complementary sensitivity function rolls off at high frequencies to attenuate the sensor noise and hence to produce a better resolution. Besides, the weighting function W2 is chosen as a constant to make the control input of PZT stay below the saturation limit. In this research, the weighting function W1 is designed as a first-order low-pass filter in the form W1 (s) =

0.1667s + 188.5 s + 0.01885

(3)

which is chosen to make the sensitivity function S have a gain less than −80 dB at low frequencies with a −3 dB crossover frequency around 30 Hz and thereby insensitive to low frequency variations in the nominal model. The weighting function W3 is selected as a first-order highpass filter transfer function W3 (s) =

s + 37.7 0.001s + 377

(4)

which is assigned to enable the complementary sensitivity function T to have a gain less than −60 dB at high frequencies with a roll-off slope of −20 dB/decade.

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. LI AND XU: DESIGN AND ROBUST REPETITIVE CONTROL OF A NEW PARALLEL-KINEMATIC XY PIEZOSTAGE

Besides, the weighting function for control input scaling is chosen as a constant W2 = 0.5 in order to restrict the control signal within the upper bound of 10 V. Using the selected weighting functions, an optimal H∞ robust controller is designed based on the γ-iteration method (using the function “hinfopt” in Robust Control Toolbox of Matlab.The obtained controller is a 18th-order model, which is then converted to a discrete-time form using zero-order hold (ZOH) with the sampling time interval of 0.5 ms (using the function “c2d” in Matlab.The −3 dB crossover frequencies of the sensitivity function and complementary sensitivity function give a bandwidth measurement of the positioning system. From the singular value plots of S and T as shown in Fig. 11, it is observed that the corresponding crossover frequencies are 29.8 Hz and 49.0 Hz, respectively. A small peak of 1.96 dB in S∞ indicates a good robustness of the control system versus model uncertainties. Besides, the bode diagram of the open-loop transfer function (Gc Gxn ) exhibits a gain margin of 13.7 dB and a phase margin of 73.2◦ , which indicates that the system is stable enough. Since the H∞ control methods have been well investigated, the theoretical basis is neglected in this paper. The theoretical analysis can be easily found in the existing literature, e.g., [29], [32], [33]. In the current research, the three weighting functions are designed to provide certain performances for the closedloop as indicated in Fig. 11. The performances will be verified by experiments conducted in Section V.

B. H∞ Combined with Repetitive Control For many applications such as SPM, the position tracking of a periodic scanning path is usually required. Repetitive control (RC) is a special case of the internal model principle in control systems with periodic signals. As a simple learning control, the RC scheme is a promising approach to achieve a precise tracking. The major advantage of RC lies in that it generates control input using the error signal information in the previous period. Thus, the control signal can be adjusted repetitively by RC for the tracking of a periodic reference motion. It is observed that the RC scheme is very similar to the iterative learning control (ILC) [45]. In fact, the control input of ILC is updated only once per cycle and the plant is reset at the beginning of each iteration. Compared to that, the control input in RC is updated continuously without the need for plant reset. 1) Plug-in Repetitive Controller Design: It has been shown that it is possible to simply plug the repetitive control into the system while the existing controller is kept unchanged. The advantage of using plug-in control lies in that the H∞ control can be designed independently without considering the repetitiveness of the reference input or disturbance. In the current research, a digital repetitive controller Gr (z) is plugged in the control loop just prior to the H∞ controller Gc (z) as illustrated in Fig. 12, where kr is the control gain, and Q(z) denotes a digital low-pass filter. Additionally, z −N represents a repetitive signal generator with a basic period of N = L/Ts , where L is the period of the signal and Ts is the sampling time interval. In the repetitive control, the filter Q(z)

7

Fig. 12. Block diagram of H ∞ robust control (G c ) combined with a plug-in digital repetitive controller (G r ), where d represents the external disturbance, Q is a low-pass filter, z −N is a repetitive signal generator, k r is the control gain, and G f is a filter assuring the closed-loop stability.

is used to improve the robustness, and the filter Gf (z) ensures the closed-loop stability. Without loss of generality, the complementary sensitivity function T (z) of the closed-loop system (with the nominal plant) excluding the repetitive controller can be expressed as T (z) =

z −n B(z −1 ) Gc (z)Gxn (z) = 1 + Gc (z)Gxn (z) A(z −1 )

(5)

where n is the identified time delay. With the feedback controller Gc (z), it is assumed that the system is asymptotically stable, i.e., the roots of denominator A(z −1 ) = 0 all locate inside the unit circle in z-plane. For a minimum-phase system, the filter Gf (z) is usually chosen as the inverse T −1 (z) of the complementary sensitivity function. Concerning a non-minimum-phase system like the one in this research, the numerator can be written as B(z −1 ) = Bc (z −1 )Bu (z −1 )

(6)

where Bc (z −1 ) and Bu (z −1 ) denote the cancelable and uncancellable parts of the numerator B(z −1 ), respectively. That is, Bc (z −1 ) comprises zeros inside the unit circle, and Bu (z −1 ) consists of zeros that are on or outside the unit circle. Then, the filter Gf (z) is designed as z −n u A(z −1 )Bu (z) Bc (z −1 )b

(7)

b ≥ maxπ |Bu (e−j ω T s )|2

(8)

Gf (z) =

ω ∈[0, T s ]

where Bu (z) is obtained from Bu (z −1 ) by replacing z −1 with z, and nu denotes the order of Bu (z −1 ) which makes the filter Gf (z) realizable. Thus, the filter actually employs a pole-zero cancellation or zero-phase compensation design. Finally, the repetitive controller can be obtained as Gr (z) =

kr Q(z)z −N Gf (z) . 1 − Q(z)z −N

(9)

2) Stability Assessment: With the designed H∞ plus repetitive controller, the error transfer function, i.e., the sensitivity

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. 8

IEEE/ASME TRANSACTIONS ON MECHATRONICS

function, of the whole system can be generated as follows. Te (z) =

1 E(z) = Xd (z) − D(z) 1 + [1 + Gr (z)]Gc (z)Gxn (z)

[1 − z −N Q(z)] 1+G c (z1)G x n (z )  = k G f (z )G c (z )G x n (z ) z −N Q(z) 1 − 1 − r 1+G c (z )G x n (z )

(10)

where E(z), Xd (z), and D(z) denote the z-transform of the tracking error e, desired position xd , and external disturbance d, respectively. The last expression in (10) means that the system can be expressed as three cascaded subsystems [46]. The first one [1 − z −N Q(z)] is a filter with a time delay, which is always stable with Q(z) selected as a finite-impulse response (FIR) filter. The second subsystem 1+G c (z1)G x n (z ) has the same denominator as the complementary sensitivity function (5) of the system, which is also stable with the designed feedback controller Gc (z). Additionally, the third a closed one can be regarded as k G f (z )G c (z )G x n (z ) z −N Q(z) loop system with the term 1 − r 1+G c (z )G x n (z ) in the positive feedback path. In view of the small-gain theory, this subsystem is stable if the following condition is satisfied:







1 − kr Gf (z)Gc (z)Gxn (z) Q(z)

(11)

jωT < 1 1 + Gc (z)Gxn (z) s z =e  with ω ∈ 0, Tπs . It can be deduced that the condition (11) is satisfied if the following two conditions hold: |Q(z)|z =e j ω T s < 1 (12)





1 − kr Gf (z)Gc (z)Gxn (z)

<1 (13)

1 + Gc (z)Gxn (z) z =e j ω T s  for ω ∈ 0, Tπs . The condition (12) imposes a constraint on the design of lowpass filter Q(z), and condition (13) provides a guidance for the selection of control gain kr . Furthermore, inserting (5), (6), and (7) into (13), gives



−n −n u

Bu (z −1 )Bu (z)

1 − kr z (14)

j ω T < 1.

b s z =e Then, taking into account (8) and (14), allows the derivation 0 < kr < 2.

(15)

Therefore, the overall system is asymptotically stable once the above conditions (12) and (15) are satisfied at the same time. It is noticeable that the repetitive controller can also be designed with other selection of Gf (z). For instance, with Gf (z) assigned as a phase lead compensator, a digital repetitive controller is constructed in [47] for SPM applications, where a phase stability condition is derived to design Gf (z). Concerning the implementation of the repetitive controller, the low-pass filter Q(z) is usually selected as a finite impulse response (FIR) filter (e.g., Q(z) = 14 z + 12 + 14 z −1 ), which has zero phase shift. This filter enables the closed-loop system to have unitary gain at the disturbance fundamental frequency, which guarantees the disturbance rejection of the system. Unfortunately, it slightly moves the pole positions of the closed-loop

system in z-plane [46]. Hence, in the current research, Q(z) is designed alternatively as the discrete-time realization of a first-order filter Q(s) = 1+T1 q s with ZOH, i.e., 

−Ts 1 − e−T s s 1 − e Tq Q(z) = Z = −Ts s(1 + Tq s) z − e Tq

(16)

where Z[·] denotes the z-transform operator and Tq is the time constant of the filter Q. It is observed that the designed lowpass filter Q(z) satisfies the stability condition (12). The effects of the two parameters kr and Tq on control performance are investigated by the following simulation studies. C. Simulation Studies 1) H∞ Robust Controller: First, a simulation study is carried out to demonstrate the tracking performance of the designed H∞ controller compared with traditional PID control. A digital PID controller can be expressed by Kd (z − 1) Ki z + . (17) z−1 z The controller parameters Kp , Ki , and Kd are initially tuned using the Ziegler-Nichols method and then finely regulated by trial-and-error to Kp = 0.0045, Ki = 6.8538, and Kp = 1.9924 × 10−6 , which give a more rapid response mainly attributed to the integral control efforts as displayed in Fig. 14(a). For a low-frequency sinusoidal reference input (1 Hz) as shown in Fig. 13(b), the tracking errors of PID and H∞ controllers are plotted in Fig. 13(c). Compared to the PID tracking results, the H∞ controller reduces the maximum tracking error by 68.2%. Thus, a better performance of H∞ over PID control strategy is evident. 2) Robust Repetitive Controller: To demonstrate the effects of the control gain kr [see (16)] selection on the tracking results, simulation studies are carried out below. For a 2-Hz periodic sinusoidal reference input as shown in Fig. 14(a), the tracking errors of H∞ plus RC along with different values of kr are compared in Fig. 14(b). The variation tendency of the control errors reveals that kr = 1 leads to the quickest convergence with a low tracking error. Thus, kr = 1 is chosen in the following experimental studies. It is noticeable that for nonperiodic reference input, kr = 0 can be assigned in the robust repetitive controller. Moreover, the robust repetitive controller with a low-pass zero-phase FIR filter Q(z) = 14 z + 12 + 14 z −1 is also implemented, and the simulation results with a longer time (8 s) are shown in Fig. 14(c). It can be observed that although the FIR Qfilter produces almost zero steady-state tracking error after the first two periods, oscillations appear in the tracking results especially as the time elapses. These oscillations are induced by the high gain introduced by the FIR filter at high frequencies where model uncertainties are relevant [46]. This is the reason why the low-pass filter (16) is employed alternatively in the current research. As the sacrifice, the steady-state control error does not converge to zero exactly due to the phase-delay nature of this low-pass filter. Even so, the control error can be reduced by selecting a smaller time constant (Tq ). With different values of Tq , the tracking results are plotted in Fig. 14(d). It is found that too Gpid (z) = Kp +

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. LI AND XU: DESIGN AND ROBUST REPETITIVE CONTROL OF A NEW PARALLEL-KINEMATIC XY PIEZOSTAGE

9

Fig. 13. Simulation results of (a) step responses of PID controllers with parameters tuned by Ziegler-Nichols and trial-and-error methods, (b) sine-wave tracking results, and (c) tracking errors of PID and H ∞ controllers.

small Tq causes instability of the closed-loop system. Therefore, a tradeoff between tracking accuracy and robustness is required for the low-pass filter design. In this research, Tq = 0.01 s is selected. To demonstrate the need of the H∞ controller design, it is also desirable to compare the control performance of RC alone with the H∞ + RC method. However, without the designed H∞ controller Gc , the closed-loop system is unstable because the complementary sensitivity function T in (5) has two poles locating outside the unit circle in z-plane. Hence, a stand-alone RC is not sufficient for the micropositioning system in the current research. To further discover the necessity of combining the H∞ with RC, another simulation study is performed to track a 5-Hz triangular motion trajectory as described in Fig. 15(a). Similar to the treatment in [47], before using the triangular wave as the input, the original trajectory is filtered to remove high-frequency components by adopting a zero-phase digital filter with 30-Hz cutoff frequency. The filter is designed using the command “filtfilt” in Matlab.The tracking results of PID + RC and H∞ + RC are shown in Fig. 15(a) and (b), respectively. Compared to PID results (the first cycle of tracking results), the PID + RC scheme reduces the control error by 70.2%, whereas the presented H∞ + RC approach suppresses the tracking error more significantly by providing a 84.1% reduction.

Fig. 14. Simulation results of the H ∞ plus repetitive control with different control gains k r for (a) the reference input, (b) tracking errors show that k r = 1 gives the best result, (c) tracking errors reveal that a zero-phase FIR filter leads to instability, and (d) tracking results are obtained with k r = 1 and different time constant T q for the Q-filter.

V. EXPERIMENTAL VERIFICATION AND DISCUSSIONS To verify the designed H∞ plus RC scheme, experimental studies are carried out to test the motion tracking performance of the XY piezostage system. The H∞ controller is constructed in Section IV-A and the repetitive controller parameters are set as kr = 1 and Tq = 0.01. For the purpose of comparisons with simulation results, a sampling frequency of 2 kHz is assigned in the following experiments. Equations (1) to (16) represent the controller design process for the H∞ and repetitive control. They also illustrate in detail how to develop a combined control for a piezoactuated micropositioning system. The control algorithms are implemented with Matlab and Simulinksoftware, and then downloaded to the DS1005 PPC board through ControlDesk

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. 10

Fig. 15. Simulation results of PID + RC and H ∞ + RC for (a) 5-Hz triangular motion tracking and (b) tracking errors.

Fig. 16. Experimental results of the x-axis motion response for a 0.2-μm staircase signal input with H ∞ control, which indicates a positioning resolution better than 0.2 μm.

interface to realize the real-time control. It is observed that the developed controller can be realized conveniently. Easyto-implement property of the controller is very important for practical applications. A. Experimental Results First, an experiment for a staircase input signal with 0.2-μm height is performed with the stand-alone H∞ controller, and the experimental results are shown in Fig. 16. It is clear to identify that the positioning resolution of the micropositioning stage is better than 0.2 μm. Second, the single-axis motion tracking for a sinusoidal trajectory with peak-to-peak (p–p) amplitudes of 30 μm is experimentally tested. With the proposed controller, the tracking results are depicted in Fig. 17(a). From the first cycle (0.5–1.0 s) tracking errors as illustrated in Fig. 17(b), it can be deduced that

IEEE/ASME TRANSACTIONS ON MECHATRONICS

Fig. 17. Experimental results of (a) 2-Hz sinusoidal motion tracking and (b) tracking errors with the H ∞ plus repetitive controller.

the maximum p-p tracking error of the H∞ control is 1.939 μm, i.e., 6.5% of the motion range. Starting from the second cycle (1.0 s), the RC begins to take effect. After two cycles (1.5 s), the H∞ plus RC generates the maximum p–p error of 0.716 μm, which accounts for 2.4% of the motion range. Thus, in comparison with stand-alone H∞ control, the proposed controller substantially mitigates the tracking error by 63.1%. Third, the motion tracking of a 5-Hz triangular-wave trajectory along the x-axis [see Fig. 18(a)] is carried out. The tracking errors are plotted in Fig. 18(b). With respect to the H∞ control results (the first cycle), the maximum p-p error has been reduced by the proposed control scheme from 1.412 μm to 0.884 μm, i.e., from 14.1% to 8.8% of the motion range, revealing an improvement of 37.4%. Comparing the experiment with simulation results for the sinusoidal (Figs. 14 and 17) and triangular (Figs. 15 and 18) motion tracking, we can observe that the experimental results agree well with the simulation output. Thus, the accuracy of the identified plant model (1) is verified. Moreover, to disclose the two-axis cooperative tracking performance of the XY stage for 2-D motion, biaxial contouring tests are implemented. Other than the motion tracking error which reflects the difference between the desired and actual positions, contouring error is defined as the minimum distance between the actual position and desired trajectory along an orthogonal direction to the trajectory. For a circle of 5-μm radius, the tracking results of the X and Y axes are shown in Fig. 19, which are obtained with a contouring speed of 40 μm/s. For two speeds of 40 and 80 μm/s, the circular contouring results with and without the RC action are illustrated in Fig. 20. It is observed that H∞ plus RC generates better contouring results than the stand-alone H∞ control. For comparisons of the tracking results of H∞ and H∞ + RC (after two contouring periods),

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. LI AND XU: DESIGN AND ROBUST REPETITIVE CONTROL OF A NEW PARALLEL-KINEMATIC XY PIEZOSTAGE

11

Fig. 20. Biaxial circular contouring results for different feed speeds with the stand-alone H ∞ control (left) and H ∞ plus repetitive control (right).

Fig. 18. Experimental results of (a) 5-Hz triangular motion tracking and (b) tracking errors with the H ∞ plus repetitive controller.

Fig. 21. Peak-to-peak (p–p) contouring errors relative to the circle diameter of H ∞ control with and without repetitive control (RC) versus feed speeds.

B. Discussions on Stage Performance

Fig. 19. Experimental results of the two axes tracking for a circle of 5-μm radius with a feed speed of 40 μm/s generated by H ∞ plus repetitive controller.

the percentage maximum p–p contouring errors relative to the circle diameter are described in Fig. 21. The bar charts show that, for different feed speeds (40, 60, 80, and 100 μm/s), the H∞ plus RC strategy reduces the contouring errors by 28.1%, 33.7%, 28.6%, and 18.3%, respectively, as compared with the H∞ robust control without RC efforts.

The above experimental results demonstrate that the robust repetitive control can substantially reduce both single-axis tracking and biaxial contouring errors of the XY micropositioning stage under the influences of hysteresis and creep. Relatively, the H∞ plus RC exhibits more significant improvement on 1-D tracking performance than on 2-D contouring for the micropositioning system. In order to further improve the 2-D contouring accuracy of the system, the contouring control approach [48] may be employed in the future research. The assembled XY stage has a dimension of 116 × 116 × 46 mm3 . It is much more compact than a monolithic one fabricated from only a piece of material in [17], which achieves almost the same size of workspace with a much larger profile dimension of 250 × 250 × 15 mm3 . Moreover, the proposed stage can be fabricated with more compact dimension if the PZTs with smaller size and leaf-spring flexures instead of the rightcircular flexure hinges are adopted. Compared to a commercial XY piezostage produced by Physik Instrumente, e.g., model

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. 12

IEEE/ASME TRANSACTIONS ON MECHATRONICS

P-541.2CD1 , which has a dimension of 150 × 150 × 16.5 mm3 , a closed-loop workspace 100 × 100 μm2 with a closed-loop resolution of 0.2 nm, the developed stage has a more compact planar dimension with a larger workspace size while owns a much worse resolution. The limitation of the current piezostage mainly arises from the displacement sensors which have a limited resolution of 38 nm. The sensors are the bottle-neck in ultraprecision positioning stages. Use of high-performance sensors with subnanometer-level resolution is expected to significantly increase the system-resolution of the proposed stage. It is noticeable that, as the disadvantage of assembly of modules, the system dynamics and statics could change every time when the system was decomposed and then reassembled. The current amplitudes of steady-state errors (see Figs. 17 and 18) indicate that the performance of the robust repetitive control may be improved, e.g., by designing a more suitable low-pass filter (Q), to further reduce the tracking error down to the positioning resolution. While this robust repetitive control study is preliminary and there is plenty of room for performance improvement, the enhancement of positioning precision for the XY micropositioning system over the sole H∞ , RC control, and PID plus RC control elaborated by the conducted investigations demonstrates the effectiveness of the mechatronic synthesis and displays great potential for the future research. Considering the performances achieved by the piezostage system with the current hardware, the stage may be more suitable for the manipulation of microscopic objects such as biological cells with the size of tens of micrometers. In the future, experiments will be conducted to demonstrate the capability of the stage for such micromanipulation tasks. VI. CONCLUSION This paper has concentrated on the design and control of a piezo-driven XY parallel micropositioning stage with integrated parallel, decoupled, and stacked kinematics structure and submicron accuracy for micro-/nanomanipulation applications. A series of analytical, simulation, and experimental studies were undertaken to facilitate the system design, plant model identification, and controller verification. It is found that SISO control is sufficient for the decoupled XY stage and the H∞ plus repetitive control strategy enables substantial improvement on the positioning accuracy in both single-axis tracking and biaxial contouring tasks with unmodeled hysteresis and creep effects. Moreover, a submicron accuracy is achieved by the micropositioning system, which verifies the effectiveness of the presented mechanism and control design processes. Future research will focus on a more sophisticated controller design with higher bandwidth. The proposed stage will be employed for nanopositioning using high-performance displacement sensors with subnanometer-level resolution. Moreover, the stage can also be miniaturized into micro- and mesoscales for pertinent micro-/nanomanipulation applications.

1 www.physikinstrumente.com/en/products/prspecs.php?sortnr=201530

REFERENCES [1] A. S. Putra, S. Huang, K. K. Tan, S. K. Panda, and T. H. Lee, “Design, modeling, and control of piezoelectric actuators for intracytoplasmic sperm injection,” IEEE Trans. Contr. Syst. Technol., vol. 15, no. 5, pp. 879–890, Sep. 2007. [2] B. Morgan, J. McGee, and R. Ghodssi, “Automated two-axes optical fiber alignment using grayscale technology,” J. Microelectromech. Syst., vol. 16, no. 1, pp. 102–110, 2007. [3] K. K. Leang, Q. Zou, and S. Devasia, “Feedforward control of piezoactuators in atomic force microscope systems: Inversion-based compensation for dynamics and hysteresis,” IEEE Control Syst. Mag., vol. 29, no. 1, pp. 70–82, Sep. 2009. [4] Y. K. Yong, S. Aphale, and S. O. R. Moheimani, “Design, identification, and control of a flexure-based XY stage for fast nanoscale positioning,” IEEE Trans. Nanotechnol., vol. 8, no. 1, pp. 46–54, Jan. 2009. [5] T. J. Teo, I.-M. Chen, G. Yang, and W. Lin, “A flexure-based electromagnetic linear actuator,” Nanotechnology, vol. 19, no. 31, pp. 315 501–1– 315 501–10, 2008. [6] J. W. Li, X. B. Chen, and W. J. Zhang, “A new approach to modeling system dynamics—In the case of a piezoelectric actuator with a host system,” IEEE/ASME Trans. Mechatronics, vol. 15, no. 3, pp. 371–380, Jun. 2010. [7] A. J. Fleming, “Nanopositioning system with force feedback for highperformance tracking and vibration control,” IEEE/ASME Trans. Mechatronics, vol. 15, no. 3, pp. 433–447, Jun. 2010. [8] U.-X. Tan, W. T. Latt, C. Y. Shee, and W. T. Ang, “A low-cost flexurebased handheld mechanism for micromanipulation,” IEEE/ASME Trans. Mechatronics, vol. 16, no. 4, pp. 773–778, Aug. 2011. [9] E. Pernette, S. Henein, I. Magnani, and R. Clavel, “Design of parallel robots in microrobotics,” Robotica, vol. 15, no. 4, pp. 417–420, 1997. [10] T. Ando, T. Uchihashi, and T. Fukuma, “High-speed atomic force microscopy for nano-visualization of dynamic biomolecular processes,” Progr. Surf. Sci., vol. 83, no. 7–9, pp. 337–437, 2008. [11] K. K. Leang and A. J. Fleming, “High-speed serial-kinematic SPM scanner: Design and drive considerations,” Asian J. Control, vol. 11, no. 2, pp. 144–153, 2009. [12] R.-F. Fung, Y.-L. Hsu, and M.-S. Huang, “System identification of a dual-stage XY precision positioning table,” Precis. Eng., vol. 33, no. 1, pp. 71–80, 2009. [13] D. Kang, K. Kim, D. Kim, J. Shim, D.-G. Gweon, and J. Jeong, “Optimal design of high precision XY-scanner with nanometer-level resolution and millimeter-level working range,” Mechatronics, vol. 19, no. 4, pp. 562– 570, 2009. [14] K.-B. Choi and D.-H. Kim, “Monolithic parallel linear compliant mechanism for two axes ultraprecision linear motion,” Rev. Sci. Instrum., vol. 77, no. 6, pp. 065 106–1–065 106–7, 2006. [15] Y. Li and Q. Xu, “A novel design and analysis of a 2-DOF compliant parallel micromanipulator for nanomanipulation,” IEEE Trans. Automat. Sci. Eng., vol. 3, no. 3, pp. 248–254, Jul. 2006. [16] S. Awtar and A. H. Slocum, “Constraint-based design of parallel kinematic XY flexure mechanisms,” ASME J. Mech. Des., vol. 129, no. 8, pp. 816– 830, 2007. [17] Y. Li and Q. Xu, “Design and analysis of a totally decoupled flexurebased XY parallel micromanipulator,” IEEE Trans. Robot., vol. 25, no. 3, pp. 645–657, Jun. 2009. [18] S. Polit and J. Dong, “Development of a high-bandwidth XY nanopositioning stage for high-rate micro-/nanomanufacturing,” IEEE/ASME Trans. Mechatronics, vol. 16, no. 4, pp. 724–733, Aug. 2011. [19] D. Zhang, C. Chang, T. Ono, and M. Esashi, “A piezodriven XYmicrostage for multiprobe nanorecording,” Sens. Actuator A, Phys., vol. 108, no. 1, pp. 230–233, 2003. [20] S. Fatikow, T. Wich, H. Hulsen, T. Sievers, and M. Jahnisch, “Microrobot system for automatic nanohandling inside a scanning electron microscope,” IEEE/ASME Trans. Mechatronics, vol. 12, no. 3, pp. 244–252, Jun. 2007. [21] W. T. Ang, P. K. Khosla, and C. N. Riviere, “Feedforward controller with inverse rate-dependent model for piezoelectric actuators in trajectorytracking applications,” IEEE/ASME Trans. Mechatronics, vol. 12, no. 2, pp. 134–142, Apr. 2007. [22] M. Al Janaideh, S. Rakheja, and C. Su, “An analytical generalized prandtl– ishlinskii model inversion for hysteresis compensation in micropositioning control,” IEEE/ASME Trans. Mechatronics, vol. 16, no. 4, pp. 734–744, Aug. 2011.

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. LI AND XU: DESIGN AND ROBUST REPETITIVE CONTROL OF A NEW PARALLEL-KINEMATIC XY PIEZOSTAGE

[23] Y. L. Zhang, M. L. Han, M. Y. Yu, C. Y. Shee, and W. T. Ang, “Automatic hysteresis modeling of piezoelectric micromanipulator in vision-guided micromanipulation systems,” IEEE/ASME Trans. Mechatronics, 2011. DOI: 10.1109/TMECH.2011.2106136. [24] S. W. John, G. Alici, and C. D. Cook, “Inversion-based feedforward control of polypyrrole trilayer bender actuators,” IEEE/ASME Trans. Mechatronics, vol. 15, no. 1, pp. 149–156, Feb. 2010. [25] P. Ge and M. Jouaneh, “Generalized Preisach model for hysteresis nonlinearity of piezoceramic actuators,” Precis. Eng., vol. 20, no. 2, pp. 99–111, 1997. [26] J.-C. Shen, W.-Y. Jywe, H.-K. Chiang, and Y.-L. Shu, “Precision tracking control of a piezoelectric-actuated system,” Precis. Eng., vol. 32, no. 2, pp. 71–78, 2008. [27] J. Yi, S. Chang, and Y. Shen, “Disturbance observer-based hysteresis compensation for piezoelectric actuators,” IEEE/ASME Trans. Mechatronics, vol. 14, no. 4, pp. 456–464, Aug. 2009. [28] S. Bashash and N. Jalili, “Robust adaptive control of coupled parallel piezo-flexural nanopositioning stages,” IEEE/ASME Trans. Mechatronics, vol. 14, no. 1, pp. 11–20, Feb. 2009. [29] S. Salapaka, A. Sebastian, J. P. Cleveland, and M. V. Salapaka, “High bandwidth nano-positioner: A robust control approach,” Rev. Sci. Instrum., vol. 73, no. 9, pp. 3232–3241, 2002. [30] C.-L. Hwang, “Microprocessor-based fuzzy decentralized control of 2D piezo-driven systems,” IEEE Trans. Ind. Electron., vol. 55, no. 3, pp. 1411–1420, Mar. 2008. [31] H. C. Liaw and B. Shirinzadeh, “Neural network motion tracking control of piezo-actuated flexure-based mechanisms for micro-/nanomanipulation,” IEEE/ASME Trans. Mechatronics, vol. 15, no. 4, pp. 517–527, Oct. 2009. [32] B. E. Helfrich, C. Lee, D. A. Bristow, X. H. Xiao, J. Dong, A. G. Alleyne, S. M. Salapaka, and P. M. Ferreira, “Combined H ∞ -feedback control and iterative learning control design with application to nanopositioning systems,” IEEE Trans. Control Syst. Technol., vol. 18, no. 2, pp. 336–351, Mar. 2010. [33] Y. Wu and Q. Zou, “Robust-inversion-based 2-DOF control design for output tracking: Piezoelectric actuator example,” IEEE Trans. Control Syst. Technol., vol. 17, no. 5, pp. 1069–1082, Sep. 2009. [34] B. J. G. Vautier and S. O. R. Moheimani, “Charge driven piezoelectric actuators for structural vibration control: Issues and implementation,” Smart Mater. Struct., vol. 14, no. 4, pp. 575–586, 2005. [35] G. S. Choi, Y. A. Lim, and G. H. Choi, “Tracking position control of piezoelectric actuators for periodic reference inputs,” Mechatronics, vol. 12, no. 5, pp. 669–684, 2002. [36] H.-S. Ahn, “Design of a repetitive control system for a piezoelectric actuator based on the inverse hysteresis model,” in Proc. 4th Int. Conf. Control Autom., Montreal, Que., Canada, 2003, pp. 128–132. [37] Y. Shan and K. K. Leang, “Repetitive control with prandtl-ishlinskii hysteresis inverse for piezo-based nanopositioning,” in Proc. Amer. Control Conf., St. Louis, MO, 2009, pp. 301–306. [38] C.-Y. Lin and Y.-C. Liu, “Precision tracking control and constraint handling of mechatronic servo systems using model predictive control,” IEEE/ASME Trans. Mechatronics, 2011. DOI: 10.1109/TMECH.2011.2111376. [39] L. Xu and B. Yao, “Adaptive robust repetitive control of a class of nonlinear systems in normal form with applications to precision motion control of linear motors,” in Proc. IEEE/ASME Int. Conf. Adv. Intell. Mechatronics, Como, Italy, 2001, pp. 527–532. [40] Y. Wang, D. Wang, B. Zhang, K. Zhou, and Y. Ye, “Robust repetitive control with linear phase lead,” in Proc. Amer. Control Conf. , Minneapolis, MN, 2006, pp. 232–237. [41] F. Zhao, X.-M. Yao, Y. Liu, and B.-K. Su, “Robust repetitive control design for a class of linear systems with time-varying uncertainties,” in Proc. Int. Joint Conf. Artif. Intell., Pasadena, CA, 2009, pp. 637–641. [42] P. C.-P. Chao, L.-D. Liao, H.-H. Lin, and M.-H. Chung, “Robust dual-stage and repetitive control designs for an optical pickup with parallel cantilever beams powered by piezo-actuation,” Microsyst. Technol., vol. 16, no. 1–2, pp. 317–331, 2010.

13

[43] N. Lobontiu and E. Garcia, “Analytical model of displacement amplification and stiffness optimization for a class of flexure-based compliant mechanisms,” Comput. Struct., vol. 81, no. 32, pp. 2797–2810, 2003. [44] Y. Li and Q. Xu, “Development and assessment of a novel decoupled XY parallel micropositioning platform,” IEEE/ASME Trans. Mechatronics, vol. 15, no. 1, pp. 125–135, Feb. 2010. [45] D. A. Bristow, M. Tharayil, and A. G. Alleyne, “A survey of iterative learning control: A learning-based method for high-performance tracking control,” IEEE Control Syst. Mag., vol. 26, no. 3, pp. 96–114, Jun. 2006. [46] R. Costa-Castello, J. Nebot, and R. Grino, “Demonstration of the internal model principle by digital repetitive control of an educational laboratory plant,” IEEE Trans. Educ., vol. 48, no. 1, pp. 73–80, Feb. 2005. [47] U. Aridogan, Y. Shan, and K. K. Leang, “Design and analysis of discretetime repetitive control for scanning probe microscopes,” ASME J. Dyn. Sys., Meas., Control, vol. 131, no. 6, pp. 061 103–1–061 103–12, 2009. [48] S.-L. Chen, H.-L. Liu, and S. C. Ting, “Contouring control of biaxial systems based on polar coordinates,” IEEE/ASME Trans. Mechatronics, vol. 7, no. 3, pp. 329–345, Sep. 2002.

Yangmin Li (M’98–SM’04) received the B.S. and M.S. degrees from Jilin University, Changchun, China, in 1985 and 1988, respectively, and the Ph.D. degree from Tianjin University, Tianjin, China, in 1994, all in mechanical engineering. He is currently a Full Professor of electromechanical engineering at the University of Macau, Taipa, Macao, China, where he also directs the Mechatronics Laboratory. He has authored more than 240 scientific papers, and has served on 100 international conference program committees. His research interests include micro/nanomanipulation, nanorobotics, micromanipulators, mobile robots, modular robots, and multibody dynamics and control. He also serves as a Technical Editor of the IEEE/ASME TRANSACTIONS ON MECHATRONICS, Associate Editor of the IEEE TRANSACTIONS ON AUTOMATION SCIENCE ENGINEERING, a Council Member and an Editor of the Chinese Journal of Mechanical Engineering, and a Member of a Editorial Board of the International Journal of Control, Automation, and Systems. Prof. Li is a member of the American Society of Mechanical Engineers (ASME).

Qingsong Xu (M’09) received the B.S. degree in mechatronics engineering (with honors) from Beijing Institute of Technology, Beijing, China, in 2002, and the M.S. and Ph.D. degrees in electromechanical engineering from the University of Macau, Macao, China, in 2004 and 2008, respectively. He is currently an Assistant Professor of electromechanical engineering at the University of Macau, Taipa, Macao, China. His current research interests include parallel robots, micro-/nanorobotics, micro-/nanopositioning, computational intelligence, and smart materials and structures.