Professional Documents
Culture Documents
3, JUNE 2013
while preserving the linear characteristics seen for small driv- where A1 , A2 , and A3 are three unknown weights with A1 >
ing currents. An ARC algorithm with effective compensation 0. Such an approximation model is used due to the following
of nonlinear electromagnetic field effect is subsequently de- several considerations.
veloped. The unknown weights and other system parameters R1: The nonlinear phenomenon is effectively captured by the
are tuned online via certain adaptation law to obtain a better quadratic and cubic term of the polynomial. And the fitting
model compensation. Comparative experiments of ARC with error can be very small for all driving currents.
and without compensation of electromagnetic nonlinearity have R2: Since the coefficients A2 and A3 are relatively very small
been carried out on both axes of a linear-motor-driven indus- when compared to A1 , the effect of the quadratic and cubic
trial gantry. The results show that the proposed ARC algorithm term can be neglected at normal driving currents which are
achieves better tracking performance than existing ones, vali- relatively small, leading to the usual linear characteristics
dating the effectiveness of the proposed approach in practical of (2) between thrust force and current with A1 = K.
applications. R3: Since the nonlinear electromagnetic effect has symmetry
with respect to the positive and negative driving currents,
II. PROBLEM FORMULATION |u| is used in the quadratic term to guarantee the odd
function of Fm (u). Note that the resulting function Fm (u)
When neglecting the fast electrical dynamics and structural
in (3) is still continuous and differentiable.
flexible modes [9], the dynamic model of an iron-core linear
R4: By restricting Ai ’s to some known ranges, the function
motor [30] can be described by
Fm (u) can be made strictly monotonic in the entire oper-
M ÿ = Fm (u) − B ẏ − Af Sf (ẏ) + Fdis (1) ating range of the physical system where the thrust force
Fm is below the peak force and the control input u is less
where y, ẏ, and ÿ represent the displacement, velocity, and ac-
than its limit. Such restrictions are necessary as the actual
celeration of the inertia load, respectively. u is the control input
electromagnetic nonlinearity is strictly monotonic. With
voltage representing the driving coil current, M is the inertia, B
this conditioning, the resulting function (3) is a one-to-one
is the viscous friction coefficient, and Af is the Coulomb fric-
mapping. Thus, for any required thrust force Fm , a unique
tion coefficient. Sf (ẏ) represents a known continuous or smooth
input u exists and can be obtained by letting (3) be zero
function used to approximate the traditional discontinuous sign
and solving for the real root of the resulting third-order
function sgn(ẏ) for effective Coulomb friction compensation
equation within the operating range.
in implementation. Fm (u) is the electromagnetic force gener-
With (3), the state-space form of the system (1) is
ated by the linear motor, and Fdis represents the lumped effect
of external disturbances and various types of modeling errors, ẋ1 = x2
such as cogging force and unmodeled friction. For small driv-
ing coil currents, the nonlinear electromagnetic field effect can θ1 ẋ2 = u − θ2 x2 − θ3 Sf (x2 ) + θ4 + θ5 u |u| + θ6 u3 + d˜
be ignored, and the electromagnetic force is considered to be (4)
proportional to driving coil current, which has been assumed in
previous study [14]. With this assumption, Fm is related to u by where x1 and x2 are the position and velocity, respec-
tively. The unknown parameter set is defined as θ =
Fm (u) = Ku (2) [θ1 , θ2 , θ3 , θ4 , θ5 , θ6 ]T ∈ R6 , with θ1 = AM1 , θ2 = AB1 , θ3 =
Af
where K = Km Ki . Km and Ki represent the average force A1 , θ5 = A A3 Fd i s
A 1 , θ6 = A 1 , and θ4 being the nominal value of A 1 .
2
constant and the current amplifier gain, respectively. d˜ = Ad 1i s − θ4 represents the time-varying portion of the nor-
F
For large driving current, the nonlinear electromagnetic field malized lumped uncertainties. The following assumptions are
effect is obvious and the resulting force constant decreases sig- made.1
nificantly as the driving current increases. As such, when driving Assumption 1: The unknown parameters and uncertain non-
heavy load and/or during high acceleration movements, using linearities are bounded by known bounds, i.e.,
the linear assumption as in the previous studies may lead to
a rather inaccurate modeling of the resulting electromagnetic θ ∈ Ωθ = {θ : θm in ≤ θ ≤ θm ax } (5)
force, which may result in instability of the resulting closed- d˜ ∈ Ωd = {d˜ : |d|
˜ ≤ δd } (6)
loop system. In order to make full use of the available driving
T
capability of the motor hardware, it is necessary to explicitly where θm in = [θ1m in , . . ., θ6m in ] , θm ax = [θ1m ax , . . .,
take into account the effect of nonlinear electromagnetic field. θ6m ax ]T , and δd are known.
Furthermore, to avoid the potentially very expensive accurate Assumption 2: For any θ̂5 and θ̂6 where θ5m in ≤ θ̂5 ≤ θ5m ax
modeling and identification of the nonlinear electromagnetic ef- and θ6m in ≤ θ̂6 ≤ θ6m ax , the function fu (u) = u + θ̂5 u |u| +
fect, the nonlinearity will be treated unknown and learned online θ̂6 u3 is strictly monotonic in the operating range of |u| ≤ uM ,
through proper parameter adaptation. Specifically, a third-order where uM represents the physical limit of the control input
polynomial function of the input u with unknown weights is voltage.
used to approximate the unknown electromagnetic nonlinearity
between the thrust force and the input voltage as 1 The following notations are used throughout this paper: ˆ • denotes the esti-
Fm (u) = A1 u + A2 u |u| + A3 u3 (3) mate of •, ˜• denotes the parameter estimation error of •, e.g., θ̃ = θ̂ − θ.
1124 IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 18, NO. 3, JUNE 2013
ary of Ωθ , respectively, and nθ̂ represents the outward unit Substituting (13) into (12) leads to
normal vector at θ̂ ∈ ∂Ωθ . Properties of this projection-type θ1 ṗ = −ks1 p + (θ1 k1 − θ2 − θ3 g)ė + vs2 + va2 − ϕTd θ̃ + d˜
adaptation law are detailed in [9]. (16)
where ϕTd = [−ẍd , −ẋd , −Sf (ẋd ), 1, u |u| , u3 ].
B. ARC Law Define d0 and d˜∗ (t) to be the nominal value and the time-
˜ respectively, i.e.,
varying portion of −ϕTd θ̃ + d,
With the use of projection-type adaptation law (7), the pa-
rameter estimates are bounded with known bounds, regardless d0 + d˜∗ (t) = −ϕTd θ̃ + d.
˜ (17)
of how the actual estimation function τ and adaptation-rate ma-
trix Γ are designed. In the following, this property will be used to d0 is compensated through the fast adaptation of direct ARC
synthesize a desired compensation DIARC control law (DCDI- design [7], [9] by
ARC) for the system (4), which achieves a guaranteed transient va2 = −dˆ0 (18)
and steady-state output tracking accuracy in general.
Define a virtual control input v as where dˆ0 represents the estimate of d0 updated by
˙
v = u + θ̂5 u |u| + θ̂6 u3 . (9) dˆ0 = Projdˆ0 (γd p), |dˆ0 (0)| ≤ dˆm ax (19)
From Assumption 2, for any bounded v, there exists a unique in which γd > 0 and dˆm ax is a preset bound for dˆ0 (t). The
root u of |u| ≤ uM for the cubic equation (9); the root can be projection mapping in (19) guarantees that |dˆ0 (t)| ≤ dˆm ax , ∀t.
obtained analytically by the standard formulas of solving cubic Substituting (18) and (17) into (16), we have
equations. Substitute (9) into (4):
θ1 ṗ = −ks1 p + (θ1 k1 − θ2 − θ3 g)ė + vs2 − d˜0 + d˜∗ (t).
ẋ1 = x2 (20)
˜ The robust function vs2 is then chosen to satisfy the following
θ1 ẋ2 = v − θ2 x2 − θ3 Sf (x2 ) + θ4 − θ̃5 u |u| − θ̃6 u3 + d.
robust performance conditions:
(10) 1) p{vs2 − d˜0 + d˜∗ (t)} ≤ η
Define the tracking error e = x1 − xd (t) and a switching- 2) pvs2 ≤ 0 (21)
function-like quantity p as
where η is a design parameter that can be arbitrarily small. One
p = ė + k1 e = x2 − x2eq , x2eq = ẋd − k1 e (11) smooth example of vs2 satisfying (21) is
e(s) 1 1 ˆ
where k1 > 0 is a positive gain. Since Gp (s) = p(s) = s+k 1
is a vs2 = −ks2 p, ks2 = (dm ax + θM ϕd + δd )2 (22)
stable transfer function, the rest of the control design is to make 4η
p as small as possible. Differentiating (11) and noting (10), one in which θM = θm ax − θm in .
CHEN et al.: ACCURATE MOTION CONTROL OF LINEAR MOTORS WITH ADAPTIVE ROBUST COMPENSATION 1125
Theorem 1: Consider the parameter estimates θ̂ updated by the Theorem 2: Consider the situation where only parametric
projection-type adaptation law (7) and the dynamic compensa- uncertainties exist after a finite time t0 , i.e., d˜ = 0, ∀t ≥ t0 .
tion dˆ0 updated by (19). Regardless the adaptation function τ to Using the same DCDIARC law as in Theorem 1 but with the
be chosen, the DCDIARC law given by (13) and (9) guarantees least-squares-type estimation function (28) and the adaptation
that all signals are bounded. Furthermore, the positive-definite rate matrix (27), if the following persistent excitation condition
function Vs defined by is satisfied:
t+T
1 1
Vs = θ1 p2 + θ1 k12 e2 (23) ϕf ϕTf dτ ≥ βIp , ∀t > t0 for some T > 0 and β > 0
2 2 t
(29)
is bounded above by
then θ̂ converges to its true value θ. Consequently, in addition to
η the robust performance results stated in Theorem 1, asymptotic
Vs (t) ≤ exp(−λt)Vs (0) + [1 − exp(−λt)] (24)
λ output tracking is also achieved, i.e., e → 0 as t → ∞.
where λ = min{2k2 /θ1m ax , k1 }
All the theorems in this paper can be proved using the same IV. EXPERIMENTAL RESULTS
techniques as in [8] and [9] and the details are omitted. A. Experimental Setup
The two-axes commercial Anorad Gantry by Rockwell
C. Parameter Estimation Automation setup at Zhejiang University and described in [17] is
used as a testbed. Both axes of the gantry are powered by Anorad
Regardless the estimation function τ to be used, the proposed LC-50-200 iron core linear motors. The maximum thrust force
DCDIARC law achieves a guaranteed transient and final track- for each axis is 632 N and the continuous force is 279 N. The
ing performance even in the presence of uncertain nonlinearities, position sensors of the gantry are two linear encoders with a
as shown in the previous sections. Thus, this section focuses on resolution of 0.5 μm. The velocity signal is obtained by the dif-
the construction of suitable estimation functions τ so that good ference of two consecutive position measurements. The exper-
online parameter estimates can be achieved as well for an im- iments have been conducted on both axes with a 5-kg payload
proved steady-state tracking accuracy. As such, in this section, mounted on the gantry and the control algorithms are imple-
uncertain nonlinearities are assumed to disappear after a finite mented in real time through a dSPACE CLP1103 controller
time t0 , i.e., assuming d˜ = 0, ∀t ≥ t0 in (4). board as detailed in [17].
To reduce the effect of measurement noises and avoid the need
of acceleration feedback, a stable filter Hf (s) with a relative
degree no less than 1 is used. Namely, applying the filter to both B. System Identification
sides of (4) and noting the assumption of d˜ = 0, one obtains First, offline parameter identification is carried out without
considering nonlinear electromagnetic field effect. For small
uf = −ϕTf θ (25) driving coil current, this linear assumption is reasonable and the
where •f represents the filtered value of •, and the regressor is nominal model for identification can be rewritten as
ϕTf = [−ẋ2f , −x2f , −Sf f , 1f , u |u|f , u3f ].
θ1 ẋ2 = u − θ2 x2 − θ3 Sf (x2 ) + θ4 . (30)
Defining the prediction output and the prediction error as
Position(m)
800 0.4
0.3
Asterisk: experimental data 0.2
700 Solid: cubic polynomial fitting 0.1
Dashed: linear approximation of the fitting 0
600 0 5 10 15 20 25 30 35 40
Thrust Force (N)
Velocity(m/s)
2
500
1
0
400 −1
−2
0 5 10 15 20 25 30 35 40
300
Acceleration(m/s )
2
40
200 20
0
100 −20
−40
0 5 10 15 20 25 30 35 40
0 Time(s)
0 2 4 6 8 10
Input Voltage(V) Fig. 3. Point-to-point motion trajectory.
800
ks = 500 × 0.19, k1 = 250. For the fast compensation term va2 ,
Asterisk: experimental data
700 the two parameters are set as γd = 500 and dˆm ax = 1. The con-
Solid: cubic polynomial fitting
Dashed: linear approximation of the fitting
tinuous function Sf (ẋd ) is chosen as π2 arctan(900ẋd ). Namely,
600 the lower and upper bounds of the parameter variations for θ are
chosen as θm in = [0.05, 0.08, 0.05, −0.5, −0.015, −0.002]T
Thrust Force (N)
TABLE I
Compensantion(V) Compensantion(V)
5
Algorithms eM (µ m) eF (µ m) L2 [e] (µ m) uM (V )
0 C1 65.5 44.4 4.54 9.55
−5 C2 37.6 28.1 3.40 9.85
−10
0 5 10 15 20 25 30 35 40
Estimate of θ1
10 0.2
Input with
5 0.18
0 0.16
−5 0.14
−10 0 20 40 60 80 100 120
0 5 10 15 20 25 30 35 40
5
Estimate of θ
Time(s) 0
−0.005
Fig. 4. Control input of two control algorithms (X-axis).
−0.01
−0.015
Compensantion(μm) Compensantion(μm)
0 20 40 60 80 100 120
40 Time(s)
Error without
−3
6
20 x 10
Estimate of θ
0 −0.5
−20 −1
−40 −1.5
−60
0 5 10 15 20 25 30 35 40
−2
0 20 40 60 80 100 120
Time(s)
40
Fig. 7. Estimation of parameters of θ1 , θ5 , and θ6 (X-axis).
Error with
20
0
−20
−40
2
Estimate of θ
−60 0.25
0 5 10 15 20 25 30 35 40
Time(s) 0.2
Fig. 5. Tracking errors of two control algorithms (X-axis).
0 20 40 60 80 100 120
Estimate of θ3
Compensantion(μm) Compensantion(μm)
0.4
40
Error without
20 0.2
0
−20 0
0 20 40 60 80 100 120
−40
Time(s)
Estimate of θ4
40 0
Error with
20
−0.1
0 0 20 40 60 80 100 120
−20 Time(s)
−40
22 22.5 23 23.5 24 24.5 25 25.5 Fig. 8. Estimation of parameters of θ2 , θ3 , and θ4 (X-axis).
Time(s)
Fig. 6. Magnification of tracking errors over one running period (X-axis). TABLE II
TRACKING PERFORMANCE OF Y-AXIS BY QUANTITATIVE MEASURES
cycle. The reduced peak transient tracking errors over the cycles Algorithms eM (µ m) eF (µ m) L2 [e] (µ m) uM (V )
also demonstrate the usefulness of online parameter adaptation. C1 75.4 66.1 9.14 9.32
C2 39.1 35.0 5.90 9.60
Figs. 7 and 8 show the parameter estimation history of C2. It
is clear that the online parameter estimates do converge to their
offline identified values over cycles even with significant initial
parameter estimation errors. All these results confirm the excel- C1 and 0.014 ms for C2, which is relatively very short when
lent tracking performance and accurate parameter estimation of compared to the sampling period 0.2 ms. The results show that
the proposed algorithm. The computational loads of two control the complexities of both algorithms are acceptable, validating
algorithms are also tested. The turnaround time is 0.010 ms for their use in practical applications.
1128 IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 18, NO. 3, JUNE 2013
2
Compensantion(μm) Compensantion(μm)
Estimate of θ
0.4
Error without
50 0.2
0 0
0 20 40 60 80 100 120
−50
3
Estimate of θ
0.4
0 5 10 15 20 25 30 35 40
0.2
0
50 0 20 40 60 80 100 120
Error with
Time(s)
4
Estimate of θ
0 0.05
−50 0
0 5 10 15 20 25 30 35 40 −0.05
Time(s) 0 20 40 60 80 100 120
Time(s)
Fig. 9. Tracking errors of two control algorithms (Y-axis).
Fig. 12. Estimation of parameters of θ2 , θ3 , and θ4 , Y-axis.
Compensantion(μm) Compensantion(μm)
50
[0.55, 0.2, 0.1, 0, −0.0186, −0.0003] , and the initial adapta-
T
0
tion rates are Γ(0) = diag{2, 50, 100, 500, 0.05, 0.005}. Other
−50 control sets are the same as the X-axis control experiments. A
21.5 22 22.5 23 23.5 24 24.5 25 25.5 desired trajectory with the maximum velocity of 1 m/s and the
maximum acceleration of 10 m/s2 is used here.
Similarly, the experimental results in terms of the quantita-
50
tive indexes are given in Table II, and the tracking error and
Error with
Fig. 10. Magnification of tracking errors over one running period (Y-axis). V. CONCLUSION
In this paper, approximation of electromagnetic nonlinearity
1
Estimate of θ
−0.015
oped for accurate motion control of linear motors. The unknown
−0.02 weights and other system parameters are tuned online via certain
−0.025 adaptation law to obtain a better model compensation. Theoret-
0 20 40 60 80 100 120
Time(s) ically, the proposed ARC algorithm achieves a guaranteed tran-
−3
sient and steady-state performance for position tracking, as well
Estimate of θ6
x 10
0
as zero steady-state tracking error when the system is subjected
−0.5
to parametric uncertainties only. Comparative experiments have
−1 been conducted on a biaxial Anorad gantry system. The results
0 20 40 60 80 100 120
Time(s) show that the proposed ARC algorithm achieves better tracking
performance than existing ones, validating the effectiveness of
Fig. 11. Estimation of parameters of θ1 , θ5 , and θ6 (Y-axis).
the proposed approach in practical applications.
[3] B. Yao, C. Hu, and Q. Wang, “An orthogonal global task coordinate frame [26] S. Zhao and K. Tan, “Adaptive feedforward compensation of force ripples
for contouring control of biaxial systems,” IEEE/ASME Trans. Mecha- in linear motors,” Control Eng. Practice, vol. 13, no. 9, pp. 1081–1092,
tronics, vol. 17, no. 4, pp. 622–634, Aug. 2012. Sep. 2005.
[4] D. M. Alter and T. C. Tsao, “Control of linear motors for machine tool [27] L. Bascetta, P. Rocco, and G. Magnani, “Force ripple compensation in
feed drives: Design and implementation of H ∞ optimal feedback control,” linear motors based on closed-loop position-dependent identification,”
ASME J. Dyn. Syst., Meas., Control, vol. 118, pp. 649–656, 1996. IEEE/ASME Trans. Mechatronics, vol. 15, no. 3, pp. 349–359, Jun. 2010.
[5] S. Komada, M. Ishida, K. Ohnishi, and T. Hori, “Disturbance observer [28] J. G. Slootweg, M. J. Hoeijmakers, and J. C. Compter, “Modeling of a
based motion control of direct drive motors,” IEEE Trans. Energy Con- linear PM machine including magnetic saturation and end effects: Max-
vers., vol. 6, no. 3, pp. 553–559, Sep. 1991. imum force-to-current ratio,” IEEE Trans. Ind. Appl., vol. 39, no. 6,
[6] G. Otten, T. Vries, J. Amerongen, A. Rankers, and E. Gaal, “Linear motor pp. 1681–1688, Nov./Dec. 2003.
motion control using a learning feedforward controller,” IEEE/ASME [29] Y. Kano, T. Kosaka, and N. Matsui, “Simple nonlinear magnetic analysis
Trans. Mechatronics, vol. 2, no. 3, pp. 179–187, Sep. 1997. for permanent-magnet motors,” IEEE Trans. Ind. Appl., vol. 41, no. 5,
[7] B. Yao and M. Tomizuka, “Adaptive robust control of SISO nonlinear sys- pp. 1205–1214, Sep./Oct. 2005.
tems in a semistrict feedback form,” Automatica, vol. 33, no. 5, pp. 893– [30] Anorad Linear Motors, Hauppauge, NY: Anorad Corporation. [Online].
900, 1997. Available: http://www.rockwellautomation.com/anorad
[8] B. Yao, “Desired compensation adaptive robust control,” ASME J. Dyn.
Syst., Meas., Control. vol. 131, no. 6, pp. 1–7, Nov. 2009.
[9] B. Yao, “Advanced motion control: From classical PID to nonlinear
adaptive robust control,” in Proc. 11th IEEE Int. Workshop Adv. Motion
Control, Nagaoka, Japan, Mar. 2010, pp. 815–829.
[10] S. Gayaka and B. Yao, “Accommodation of unknown actuator faults using
output feedback based adaptive robust control,” Int. J. Adaptive Control
Signal Process., vol. 25, no. 11, pp. 965–982, Nov. 2011. Zheng Chen (S’11) received the B.Eng degree from
[11] A. Mohanty and B. Yao, “Integrated direct/indirect adaptive robust con- Zhejiang University, Hangzhou, China, in 2007,
trol hydraulic manipulators with valve deadband,” IEEE/ASME Trans. where he is currently working toward the Ph.D. de-
Mechatronics, vol. 16, no. 4, pp. 707–715, Aug. 2011. gree in mechatronic control engineering.
[12] C. Hu, B. Yao, and Q. Wang, “Performance oriented adaptive robust con- He was a Visiting Scholar in the School of
trol of a class of nonlinear systems preceded by unknown dead-zone with Mechanical Engineering, Purdue University, West
comparative experimental results,” IEEE/ASME Trans. Mechatronics, Lafayette, IN, from 2008 to 2010.
[Online]. Available: http://ieeexplore.ieee.org, DOI: 10.1109/TMECH.
2011.2162633
[13] T. A. Davis, Y. C. Shin, and B. Yao, “Observer-based adaptive robust con-
trol of friction stir welding axial force,” IEEE/ASME Trans. Mechatronics,
vol. 16, no. 6, pp. 1032–1039, Dec. 2011.
[14] L. Xu and B. Yao, “Adaptive robust precision motion control of linear
motors with negligible electrical dynamics: Theory and experiments,”
IEEE/ASME Trans. Mechatronics, vol. 6, no. 4, pp. 444–452, Dec. 2001.
[15] Y. Hong and B. Yao, “A globally stable high performance adaptive robust
control algorithm with input saturation for precision motion control of
linear motor drive system,” IEEE/ASME Trans. Mechatronics, vol. 12,
no. 2, pp. 198–207, Apr. 2007. Bin Yao (S’92–M’96–SM’09) received the B.Eng.
[16] B. Yao, C. Hu, L. Lu, and Q. Wang, “Adaptive robust precision motion degree in applied mechanics from the Beijing Univer-
control of a high-speed industrial gantry with cogging force compensa- sity of Aeronautics and Astronautics, Beijing, China
tions,” IEEE Trans. Control Syst. Technol., vol. 19, no. 5, pp. 1149–1159, in 1987, the M.Eng. degree in electrical engineering
Sep. 2011. from Nanyang Technological University, Singapore,
[17] L. Lu, Z. Chen, B. Yao, and Q. Wang, “Desired compensation adaptive in 1992, and the Ph.D. degree in mechanical engi-
robust control of a linear motor driven precision industrial gantry with im- neering from the University of California at Berkeley
proved cogging force compensation,” IEEE/ASME Trans. Mechatronics, in 1996.
vol. 13, no. 6, pp. 617–624, Dec. 2008. He has been with the School of Mechanical Engi-
[18] C. Hu, B. Yao, and Q. Wang, “Integrated direct/indirect adaptive robust neering, Purdue University, West Lafayette, IN, since
contouring control of a biaxial gantry with accurate parameter estima- 1996, and was promoted to the rank of Professor in
tions,” Automatica, vol. 46, no. 4, pp. 701–707, Apr. 2010. 2007. He was honored as a Kuang-Piu Professor in 2005 and a Changjiang
[19] C. Hu, B. Yao, and Q. Wang, “Adaptive robust repetitive control of an Chair Professor at Zhejiang University, Hangzhou, China, by the Ministry of
industrial biaxial precision gantry for contouring tasks,” IEEE Trans. Education of China in 2010 as well. More detailed information can be found at
Control Syst. Technol., vol. 19, no. 6, pp. 1559–1568, Nov. 2011. https://engineering.purdue.edu/∼byao.
[20] C. Hu, B. Yao, and Q. Wang, “Global task coordinate frame based contour-
ing control of linear-motor-driven biaxial systems with accurate parameter
estimations,” IEEE Trans. Ind. Electron., vol. 58, no. 11, pp. 5195–5205,
Nov. 2011.
[21] C. Hu, B. Yao, Q. Wang, Z. Chen, and C. Li, “Experimental investigation
on high-performance coordinated motion control of high-speed biaxial
systems for contouring tasks,” Int. J. Machine Tools Manufacture, vol. 51,
no. 9, pp. 677–686, Sep. 2011. Qingfeng Wang (M’11) received the M.Eng. and
[22] L. Lu, B. Yao, Q. Wang, and Z. Chen, “Adaptive robust control of lin- Ph.D. degrees in mechanical engineering from Zhe-
ear motors with dynamic friction compensation using modified LuGre jiang University, Hangzhou, China, in 1988 and 1994,
model,” Automatica, vol. 45, no. 12, pp. 2890–2896, Dec. 2009. respectively.
[23] B. Bona, M. Indri, and N. Smaldone, “Rapid prototyping of a model-based He then became a faculty member at Zhejiang
control with friction compensation for a direct-drive robot,” IEEE/ASME University where he was promoted to the rank of Pro-
Trans. Mechatronics, vol. 11, no. 5, pp. 576–584, Oct. 2006. fessor in 1999. He was the Director of the State Key
[24] T. Lee, K. Tan, and S. Huang, “Adaptive friction compensation with a Laboratory of Fluid Power Transmission and Con-
dynamical friction model,” IEEE/ASME Trans. Mechatronics, vol. 16, trol, Zhejiang University, from 2001 to 2005, where
no. 1, pp. 133–140, Feb. 2011. he currently serves as the Head of the Institute of
[25] S.-L. Chen, K. Tan, S. Huang, and C. Teo, “Modeling and compensation of Mechatronic Control Engineering. His research in-
ripples and friction in permanent-magnet linear motor using a hysteretic terests include electrohydraulic control components and systems, hybrid power
relay,” IEEE/ASME Trans. Mechatronics, vol. 15, no. 4, pp. 586–594, systems and energy saving techniques for construction machinery, and system
Aug. 2010. synthesis for mechatronic equipment.