You are on page 1of 6

2010 American Control Conference WeB21.

1
Marriott Waterfront, Baltimore, MD, USA
June 30-July 02, 2010

An Active Disturbance Rejection Based Approach to Vibration


Suppression in Two-Inertia Systems
Shen Zhao1 and Zhiqiang Gao1,2

Abstract— This study concerns with the resonance problems are equivalent mathematically to the mechanical methods
in motion control, typically described in a two-inertia system mentioned above. Active resonance damping control [4]
model as compliance between the motor and load. We refor- actually increases the effective physical damping by adding
mulate the problem in the framework of active disturbance
rejection control (ADRC), where the resonance is assumed a torque that is proportional to the speed difference between
unknown and treated as disturbance, estimated and mitigated. the motor and load. Acceleration feedback control [4]-[5],
This allows the closed-loop bandwidth to go well beyond the however, increases the motor inertia equivalently. There are
resonant frequency, which is quite difficult with the existing still other control methods available, such as center of mass
methods. In addition, such level of performance is achieved control [2], [4] and resonance ratio control [6]-[7].
with minimum complexity in the controller design and tuning:
no parameter estimation or adaptive algorithm is needed, and All of the above control methods predicate on the detailed
the controller is tuned by adjusting one parameter, namely,
the bandwidth of the control loop. It is also shown that the mathematical model of the physical process that may or may
proposed solution applies to both the velocity and position not be readily available. Even if such a model is obtained
control problems, and the performance is monitored at both at considerable cost, the parameters of the model often
motor side and load side, with the latter case much more change during operation, which may lead to variations in
indicative of true quality of the control system, and the fact the resonant frequency, leaving the notch filter approach, for
that ADRC offers an effective and practical motion control
solution, in the presence of unknown resonant frequency within example, vulnerable. The attempt to address this flaw leads
the bandwidth of the control system. to solutions such as the adaptive notch filter [8], which is
designed to tune the filter parameters on the fly based on
I. I NTRODUCTION adaptive control theory, adding complexity and cost to the
Vibration suppression is important in motion control ap- design, implementation, and tuning of the control system. It
plications because vibration causes dynamic stresses, en- is in this background that an alternative solution is proposed
ergy wastes and performance degradations [1]. By law of in this paper.
physics, mechanical resonance is unavoidable in every sys- To deal with the resonance problem in motion control, as
tem involving motion, but the natural frequencies of such described above, we resort to a rather novel control method
systems are usually quite high and not excited during most that requires very little system model information and makes
common motion maneuvers, where a simple proportional- the control system tolerant of unknown changes in system
integral-derivative (PID) controller is often sufficient to meet dynamics. This method is known as active disturbance re-
the design requirements. Control design becomes an issue, jection control (ADRC) [9]-[14], based on the key concept
however, when the performance improvements push the loop of treating the unknown dynamics and disturbances in a
bandwidth to its limit where the resonant modes come into physical process as the total disturbance, building a state
play. The most common resonance seen in industry can be observer, known as the extended state observer (ESO), to
attributed to the compliant couplings, such as gear boxes, estimate it in real time, and then canceling its effect using a
long shafts and belts, which can be treated as springs [2]. part of the control signal. In the context of the motion control,
To deal with resonance, there are mechanical and electrical the resonant mode is not canceled out using a notch filter,
means. Since the resonance is caused by compliance, a stiffer but its effect to motion, the ripples in torque, is estimated
transmission, i.e. a direct coupling in place of a belt, will be and canceled in real time using the motor torque, after which
an obvious solution. Adding more mechanical damping will the motion dynamic behaves largely like a rigid body.
surely be helpful. In addition, increasing the motor inertia
is found to be an effective way to alleviate the resonance This paper is organized as follows. The problem descrip-
[2]. These mechanical methods are costly, which leads us tion based on two-inertia system model is given in Section II,
to electrical options, consist of low-pass filter, notch filter followed by the main result in Section III, where the motion
[3] and bi-quad filter [4], all for the objective of attenuating control problem is reformulated in the context of ADRC.
the loop gain amplitude at the resonant frequency so that Simulation results and comparison to existing methods are
the resonance is suppressed. Some of the electrical methods shown in Section IV. Going beyond the existing methods, we
show in Section V that the proposed method yields a more
1 Center for Advanced Control Technologies, Fenn College of Engineer-
practical solution when the feedback is taken at the load side,
ing, Cleveland State University, Cleveland, OH 44115, USA
2 The Corresponding author. E-mail: z.gao@ieee.org. Tel:1-216-687- as oppose to the motor side. Finally concluding remarks are
3528, Fax:1-216-687-5405 included in Section VI.

978-1-4244-7425-7/10/$26.00 ©2010 AACC 1520


II. P ROBLEM D ESCRIPTION AND E XISTING S OLUTION Bode Plots of Velocity Transfer Functions
50
The compliant resonance problem can be simplified and Resonant Frequency

represented by the two-inertia system model [2], [7] as shown 0

Magnitude (dB)
in Figure 1. −50 Anti−resonant
Frequency
−100

−150

−200
90
Rigid Body Model
Motor − Compliant Model
0 Load − Compliant Model

Phase (deg)
Fig. 1. Two-inertia system model. −90

−180
Motor inertia (JM ) is connected to load inertia (JL ) by a
spring (KS ) with damping (bS ). A torque (TE ) is applied on −270
1 2 3 4 5 6
10 10 10 10 10 10
the motor side to drive the system. Angular acceleration, Frequency (rad/sec)

angular velocity and angular position of the motor and


the load are denoted as αM , ωM , θM , and αL , ωl , θL , Fig. 2. Bode plots of velocity transfer functions - Rigid vs. Compliant.
respectively. Through simple analysis, we can derive the
transfer functions from input (TE ) to different outputs (ωM ,
ωL , θM , and θL ). The transfer function from TE to ωM is Several existing methods are described in [4] that deal
ωM 1 JL s 2 + b S s + K S with the resonance. A notch filter in the form of
= (1)
TE (JM + JL )s JP s2 + bS s + KS s 2 + ωR
2
TN (s) = 2 (7)
where JP = JM JL /(JM + JL ) . Similarly, we can get the s2 + 2ζωR + ωR
other three transfer functions.
is often used to attenuate the open loop gain at the resonant
θM 1 JL s 2 + b S s + K S frequency. The bi-quad filter
= (2)
TE (JM + JL )s2 JP s2 + bS s + KS
s2 + 2ζR ωR + ωR
2

ωL 1 bS s + K S TBQ (s) = 2 (8)


= (3) s2 + 2ζAR ωAR + ωAR
TE (JM + JL )s JP s2 + bS s + KS
as another solution, not only attenuates the open loop gain at
θL 1 bS s + KS the resonant frequency but also increases the open loop gain
= (4)
TE (JM + JL )s2 JP s2 + bS s + KS at the anti-resonant frequency making it more like a rigid
body system. The acceleration feedback method employs
The first term of each transfer function is exactly the
a rigid-body Luenberger observer to estimate the motor
same as the transfer function for the rigid body model; the
acceleration and uses it as a feedback for the purpose of
second term which contains resonance is introduced by the
increasing the motor inertia.
compliance. In both motor and load transfer functions, the
In a typical configuration of two-inertia system, the sensor
denominators of the resonance term will produce a resonant
is normally mounted at the motor end, where only the motion
frequency (ωR ), and the numerator of the resonance term
of the motor is measured and fed back. We denote this
in motor transfer functions will produce an anti-resonant
set up as motor feedback and this is the common practice
frequency (ωAR ) [2]. They can be calculated by following
in industry. In most cases seen in industry, however, the
equations. p objective is to control the motion of the load. Consequently,
ωR = KS /JP (5)
we will also investigate the alternative where we mount the
p
ωAR = KS /JL (6) sensor at the load end and use the measurement of the load as
feedback, which is denoted as the load feedback. Although
The Bode plots of velocity transfer functions of rigid body the load feedback provides the direct information on how the
model and compliant model (two-inertia system model) are load behaves, there is a considerable amount of phase lag,
shown in Figure 2 for comparison. At low frequency (below comparing to the motor feedback, which makes the control
the anti-resonant frequency) the two models behave the same. design more challenging. One may suspect that this might
The motor and load are connected as a whole just like be a main reason why the motor feedback configuration is
the rigid body. As frequency goes higher, the motor and widely used in industry.
load become disconnected and behave differently. Around Different applications may have different design objec-
resonant frequency there is a 180 degree phase difference tives. Some regulate velocity, others position. To show the
between the motor and load, which to some extent represents generality of the proposed method, both velocity control and
the resonance as well. position control are addressed in this paper.

1521
III. T HE P ROPOSED S OLUTION Here we can see clearly that the total disturbance is “can-
As mentioned in Section I, active disturbance rejection celled” and the plant becomes a pure integrator which can
control (ADRC) provides an alternative design paradigm for be easily controlled using a proportional controller given in
the resonance problem in motion control. The main idea of Equation (13).
ADRC is to treat any unknown dynamics of the system B. Velocity Control with Load Feedback
together with external disturbance as a total disturbance,
use an extended state observer (ESO) to estimate this total Considering an external disturbance w, Equation (3) can
disturbance in real time, and then cancel it in the control be rewritten as
...
law [9]. In this manner we do not have to know the exact y + a1 ÿ + a2 ẏ = b1 u̇ + b2 u + w (15)
system model in order to control it, and particularly in this
application we can treat the resonance, no matter what the where y is the load velocity, and u is torque applied to
frequency is, as part of the total disturbance. the motor. Integrating Equation (15) once on both sides, the
For completeness, we consider two types of motion con- third-order system with a relative degree of two becomes a
trol, velocity control and position control, and two feedback second-order system
w w
options, motor feedback and load feedback. Since the only ÿ = b1 u + (−a1 ẏ − a2 y + b2 u + w)
difference between velocity control and position control is w w
that the plant has one more integrator in position control, = b1 u + f (ẏ, y, u, w) (16)
we will only present the problem reformulation for velocity
control in the ADRC structure with both feedback options. Similarly, the ADRC control solution for this plant is
constructed as [15]
A. Velocity Control with Motor Feedback 
 eo = y − z1
With b0 = 1/JM , b1 = bS /(JM JL ), b2 = KS /(JM JL ), 
 ż = z + β e
1 2 1 o
a1 = bS /JP , a2 = KS /JP , and considering an external (17)
 ż2 = z 3 + b̂ 1 u + β2 e o
disturbance w, Equation (1) can be rewritten as 

ż3 = β3 eo
...
y + a1 ÿ + a2 ẏ = b0 ü + b1 u̇ + b2 u + w (9)
u0 − fˆ u 0 − z3
u= = (18)
where y is the motor velocity, and u is torque applied to b̂1 b̂1
the motor. Integrating Equation (9) twice on both sides, the
u0 = kp (r − y) − kd ẏ (19)
third-order system with a relative degree of one becomes a
first-order system [11] as below where r is the reference input, y is the output, u0 is the virtual
w w x x control signal, u is the actual control signal, e is tracking
ẏ = b0 u + (−a1 y − a2 y + b1 u + b2 u+ w)
w w x x error, eo is observer error, β1 , β2 and β3 are observer gains,
= b0 u + f (y, y, u, u, w) (10) kp and kd are controller gains, b̂1 is the estimated value of
b1 , z1 , z2 and z3 are observer states tracking y, ẏ and f
Here f (· ), including both external disturbance and internal respectively. In this case z3 is the extended state and a PD
dynamics — the resonance, represents the “total disturbance” controller is designed for the double integrator plant.
to be estimated and mitigated. For this purpose, the ADRC For the detailed derivation of the ADRC control law and
control law is constructed as [15] recent mathematical analysis of this design approach, the
readers are referred to [12]-[14]. The focus of this paper is

 eo = y − z1
ż = z2 + b̂0 u + β1 eo (11) on its possible application in motion control in the presence
 1 of resonant mode.
ż2 = β2 eo
u0 − fˆ u 0 − z2 IV. S IMULATION R ESULTS AND C OMPARISON
u= = (12) In this section, the proposed method is tested in simulation
b̂0 b̂0
and compared to the three existing methods described in [4],
u0 = kp e = kp (r − y) (13) using the motor feedback configuration as in [4]. The more
where r is the reference input, y is the output, u0 is the virtual practical load feedback configuration will be explored in the
control signal, u is the actual control signal, e is tracking next section.
error, eo is observer error, β1 and β2 are observer gains, kp A. Parameters and Profile Selection
is controller gain, b̂0 is the estimated value of b0 , z1 and z2
are observer states tracking y and f respectively. Here z2 is The proposed method is tested in simulations using the
what we called “extended state”, which is the estimation of same system parameters as those in [4], with KS = 372
f in real time. N·m/rad, bS = 0.008 N·m·s/rad, JM = 1.88×10−3 kg·m2 ,
Substituting Equation (12) into Equation (10), JL = 3.13×10−3 kg·m2 . In this case, the anti-resonant fre-
quency ωAR is 345 rad/s (or 55 Hz), and resonant frequency
u0 − fˆ ωR is 563 rad/s (or 90 Hz). We also compare our method with
ẏ = b0 + f ≈ u0 (14)
b̂0 those discussed in [4] applying their fine tuned parameters

1522
(a) Tracking Performance
in velocity control with motor feedback. The comparison is 1.1

Velocity (rad/s)
not done for other cases because [4] only considers velocity
control with motor feedback. 1
Using the parameterization technique proposed in [15], the
observer gains and controller gains are selected as follows: 0.9
0.57 0.58 0.59 0.6 0.61 0.62 0.63 0.64 0.65 0.66 0.67
in second-order ADRC, β1 = 2ωo , β2 = ωo2 , kp = ωc ; in Time (s)
third-order ADRC, β1 = 3ωo , β2 = 3ωo2 , β3 = ωo3 , kp = ωc2 , Control Signals ADRC (ω =400 Hz)
o
0.1
kd = 2ωc ; and in fourth-order ADRC, β1 = 4ωo , β2 = 6ωo2 , Notch Filter

Torque (N*m)
Bi−quad Filter
β3 = 4ωo3 , β4 = ωo4 , kp = ωc3 , kd = 3ωc2 , kdd = 3ωc . 0.05 Acceleration Feedback

Above, ωo is the observer bandwidth, and ωc is the controller 0


bandwidth which is set to ωo /2. After parameterization, only −0.05
0.57 0.58 0.59 0.6 0.61 0.62 0.63 0.64 0.65 0.66 0.67
one tuning parameter ωo is left making the tuning process Time (s)
very easy.
(b) Disturbance Rejection Performance
Step reference is a commonly used profile in simulations 2.5

Velocity (rad/s)
and real tests, but it is too aggressive and contains com- 2
1.5
ponents with very broad bandwidth, which will excite the
1
resonant mode of the system. So in industry the trapezoidal 0.5
0.98 0.99 1 1.01 1.02 1.03 1.04 1.05 1.06 1.07 1.08
profile, which is less aggressive and also energy saving, is Time (s)
widely used instead of step reference. Control Signals ADRC (ω =400 Hz)
o
Even if a trapezoidal profile is used, the rising time of the 0.5

Torque (N*m)
Notch Filter
0 Bi−quad Filter
profile is still crucial to the system performance. The faster −0.5 Acceleration Feedback
the rising time is, more possible the system is going to have −1
resonance. In order to avoid the resonance, we choose our −1.5
0.98 0.99 1 1.01 1.02 1.03 1.04 1.05 1.06 1.07 1.08
Time (s)
rising time between 0.05 s and 0.1 s in our simulations.
B. Comparison
Fig. 3. Motor Response Comparison. (a) tracking response; (b) disturbance
The proposed method is simulated and compared to the response.
notch filter, bi-quad filter, acceleration feedback methods,
with the rising time set to 100 ms (0.1 s), the profile starting
time set to 0.5 s and a disturbance of 1 N·m applied to the decreases as the bandwidth increases. The disturbance rejec-
motor at 1 s. The results are shown in Table I and II, as well tion ability of acceleration feedback is better than both notch
as in Figure 3. filter and bi-quad filter, which have big errors and oscillate.
But ADRC has the best disturbance rejection ability which
TABLE I
increases as the bandwidth increases.
M OTOR RESPONSES : TRACKING PERFORMANCE
Note that the bandwidth of ADRC can go well beyond the
Overshoot 5h Settling Time resonant frequency, which is quite difficult to achieve with
(%) (ms) other methods. As shown in [4] the closed-loop bandwidths
Notch Filter 4.2 133
Bi-quad Filter 1.3 115
associated with the notch filter, the bi-quad filter and ac-
Acceleration Feedback 4.8 137 celeration feedback design are 32 Hz, 47 Hz and 37 Hz,
ADRC with 100 0.6 108 respectively, well below the resonant frequency (90 Hz).
different ωo 200 0.2 97
(Hz) 400 0.1 96
With ωo set to 400 Hz, however, the closed-loop bandwidth
of ADRC is found to be 192 Hz, which is well beyond the
resonant frequency, unlike the existing methods.
TABLE II The robustness of each controller is also tested by varying
M OTOR RESPONSES : DISTURBANCE REJECTION PERFORMANCE the load inertia without changing the controller parameters.
The tests are performed with the load changing to 0.9, 1.1, 2
Maximum Error 5h Settling Time
(%) (ms)
and 5 times of its original value. The bi-quad filter is found
Notch Filter 135 >1000 to be the most fragile, because the system becomes unstable
Bi-quad Filter 70 >1000 for all four load changes. With the notch filter, the system is
Acceleration Feedback 72 72 stable for the first two changes but becomes unstable for last
ADRC with 100 58 66
different ωo 200 34 86 two in the presence of external disturbances. Acceleration
(Hz) 400 18 94 feedback and ADRC are stable for all four cases, but the
former results in a bigger overshoot of 15%. The motor
It is observed that acceleration feedback has the biggest overshoot in ADRC remains mostly unchanged, but the load
overshoot. Bi-quad filter has less overshoot because it cancels oscillation becomes more pronounced with the increasing
out both resonant and anti-resonant terms in the transfer load.
function. ADRC has even less overshoot and the overshoot We have so far concerned with only the response of the

1523
motor in control design, as in the existing methods mentioned from 100 to 400 Hz. Note that the best load response
above. In reality, however, load response is equally impor- performance is obtained at ωo = 200 Hz, proving that the
tant, if not more so. The good response from the motor in optimal performance is corresponding not to the highest
Figure 3 may be a bit misleading, considering that the load bandwidth. The reason is rather intuitive: higher bandwidth
may experience significant oscillations, as shown in Figure in the motor loop leads to faster movement of the motor
4. How to better regulate the load response will be explored shaft, which in turn leads to more stimulation of the resonant
in the next section. mode.
TABLE III
(a) Tracking Performance
1.1 L OAD RESPONSES : TRACKING PERFORMANCE
ADRC (ω =400 Hz)
o
Notch Filter
1.05
Overshoot 5h Settling Time
Bi−quad Filter
Velocity (rad/s)

Acceleration Feedback (%) (ms)


Notch Filter 4.6 134
1
Bi-quad Filter 2.1 253
Acceleration Feedback 5.3 136
0.95
ADRC with 100 0.9 109
different ωo 200 0.6 104
0.9
0.57 0.58 0.59 0.6 0.61 0.62 0.63 0.64 0.65 0.66 0.67
(Hz) 400 0.8 157
Time (s)

(b) Disturbance Rejection Performance


2.5
ADRC (ω =400 Hz)
o
TABLE IV
Notch Filter L OAD RESPONSES : DISTURBANCE REJECTION PERFORMANCE
2 Bi−quad Filter
Velocity (rad/s)

Acceleration Feedback
Maximum Error 5h Settling Time
1.5 (%) (ms)
Notch Filter 130 >1000
1 Bi-quad Filter 60 >1000
Acceleration Feedback 87 68
0.5 ADRC with 100 74 66
0.98 0.99 1 1.01 1.02 1.03 1.04 1.05 1.06 1.07 1.08
Time (s) different ωo 200 32 116
(Hz) 400 10 343

Fig. 4. Load responses of motor feedback velocity control with different For position control, the results are similar as shown in
controllers. (a) tracking response; (b) disturbance response.
Table V. The best performance is obtained at the medium
bandwidth of 80 Hz; when bandwidth goes beyond 150 Hz
V. M ORE R ESULTS ON L OAD R ESPONSE R EGULATION the system becomes unstable. The load still has some oscil-
lation but the amplitude is relative small. The performance
The ultimate objective in control design is to regulate the
of disturbance rejection is also quite good. To further reduce
load response, which seems to suggest that the load feedback
load oscillation, we investigate the alternative load feedback
configuration is the preferred choice. With motor feedback,
configuration below.
since it is open loop from the motor to the load, all we
can do is to control the motor. From the observation in the TABLE V
previous section, the better the motor is controlled, the more L OAD RESPONSES OF POSITION CONTROL WITH MOTOR FEEDBACK
oscillation is found in load response. In this situation, we are
left with the only option of either manipulating the motion ωo Overshoot 5h Settling Time
(Hz) (%) (ms)
profile to influence the load response indirectly, or to degrade 40 0.4 158
the motor response performance to avoid load oscillation. 60 0.3 133
A better solution to this dilemma, perhaps, is to use the 80 0.1 114
load feedback configuration. But as mention above, the load 100 0.4 123
120 0.7 154
feedback comes with it additional phase lag, which could
complicate the control design. In this section, we revisit
this issue in the context of the proposed method, in hope
B. Load Control with Load Feedback
of finding a viable load feedback solution.
In this section, both approaches for better load responses In this configuration, the output measurement, i.e. position
are considered, starting with the motor feedback and a or velocity, is taken at the load side, leading to significantly
detuned ADRC controller. more phase lag, as shown in Equation (3) and (4). Note
that comparing to the motor feedback, there is one less zero
A. Load Control with Motor Feedback in the transfer function and the remaining zero moves to
Still considering velocity control with motor feedback, the the frequency of 7 KHz. This means that at the resonant
load responses of all four methods are summarized in Table frequency (90 Hz), there is additional 180 degree phase lag,
III and IV, with ADRC tuned at three different bandwidths, making the control design more challenging. But, as it turns

1524
out, the benefits of measuring the output variable, i.e. the For practicality, in addition to the standard motor feedback
load, outweigh disadvantage in the phase lag. configuration, the proposed method is also applied to the
Applying the proposed ADRC solution to this config- scenarios where the measurement is taken at the load, adding
uration at the sampling frequency of 50 KHz, excellent further phase lag and design difficulty to the task. Results
performance is obtained. With ωo set to 200 Hz and rising show that the proposed solution handles both configurations
time set to 50 ms, the overshoot is only 0.1%, and the settling quite well, making it a rather robust and practical solution
time is 52 ms (shown in Figure 5). With load feedback the for motion control.
load response gets better and the oscillation migrates to the
R EFERENCES
motor response. It is an evidence that ADRC has the ability
to remove the oscillation from whichever response that is the [1] C. Beards, “Vibration analysis and control system dynamics,” Halsted
Press, 1981.
main concern. [2] G. Ellis, “Control system design guide, 2nd edition,” Academic Press,
2000.
(a) Tracking Performance
[3] P. Schmidt and T. Rehm, “Notch filter tuning for resonant frequency
1.1 reduction in dual inertia systems,” Proceedings of IEEE Industry
Motor Applications Conference, vol. 3, pp. 1730–1734, Oct. 1999.
Load
1.05
[4] G. Ellis and R. D. Lorenz, “Resonant load control methods for
Velocity (rad/s)

industrial servo drives,” Proceedings of IEEE Industry Applications


Conference, vol. 3, pp. 1438–1445, Oct. 2000.
1
[5] G. Ellis, “Cures for mechanical resonance in industrial servo systems,”
PCIM Conference, pp. 187–192, 2001.
0.95 [6] Y. Hori, “2-mass system control based on resonance ratio control and
manabe polynomials,” Asia Control Conference, pp. 741–744, 1994.
0.9
0.54 0.545 0.55 0.555 0.56 0.565 0.57 0.575 0.58 0.585 0.59
[7] Y. Hori, H. Sawada, and Y. Chun, “Slow resonance ratio control for
Time (s) vibration suppression and disturbance rejection in torsional system,”
IEEE Transactions on Industrial Electronics, vol. 46, no. 1, pp. 162–
(b) Disturbance Rejection Performance
1.1 168, Feb. 1999.
Motor [8] Y. Zhou, F. Peng, and B. Li, “Adaptive notch filter control for the
Load torsion vibration in lead-screw feed drive system based on neural
1.05
Velocity (rad/s)

network,” Intelligent Robotics and Applications, vol. 5315, pp. 803–


812, Oct. 2008.
1
[9] Z. Gao, Y. Huang, and J. Han, “An alternative paradigm for control
system design,” Proceedings of the 40th IEEE Conference on Decision
0.95 and Control, vol. 5, pp. 4578–4585, 2001.
[10] Z. Gao, “Active disturbance rejection control: a paradigm shift in
0.9 feedback control system design,” American Control Conference, vol. 6,
0.99 0.995 1 1.005 1.01 1.015 1.02 1.025 1.03 1.035 1.04
Time (s) pp. 2399–2405, Jun. 2006.
[11] J. Han, “From pid to active disturbance rejection control,” IEEE
Transactions on Industrial Electronics, vol. 56, no. 3, pp. 900–906,
Mar. 2009.
Fig. 5. Motor and load responses of velocity control with load feedback. [12] Q. Zheng, L. Dong, D. Lee, and Z. Gao, “Active disturbance rejection
(a) tracking response; (b) disturbance response. control and implementation for mems gyroscopes,” to appear in IEEE
Transactions on Control System Technology, 2008.
[13] Q. Zheng, Z. Chen, and Z. Gao, “A practical dynamic decoupling
The simulation results for position control with load control approach,” Control Engineering Practice, vol. 17, pp. 1016–
feedback are listed in Table VI. Note that the overshoot is 1025, Jul. 2009.
basically nonexistent. Excellent disturbance rejection is also [14] W. Zhou, S. Shao, and Z. Gao, “A stability study of the active distur-
bance rejection control problem by a singular perturbation approach,”
observed. Applied Mathematical Sciences, vol. 3, no. 10, pp. 491–508, 2009.
[15] Z. Gao, “Scaling and bandwidth-parameterization based controller
TABLE VI
tuning,” American Control Conference, vol. 6, pp. 4989– 4996, Jun.
L OAD RESPONSES OF POSITION CONTROL WITH LOAD FEEDBACK 2003.

ωo Overshoot 5h Settling Time


(Hz) (%) (ms)
50 1.2 104
100 0.3 60
200 0 55

VI. C ONCLUDING R EMARKS


A novel solution for resonance suppression in motion
control is proposed. By reformulating the problem in the
framework of active disturbance rejection control, solutions
for both velocity control and position control are presented
and compared with the existing methods favorably. It is
shown that, with the proposed method, vibration can be
eliminated even when the control bandwidth is pushed well
beyond the resonant frequency, which is assumed unknown.

1525

You might also like