Conference PaperPDF Available

Inverse-Based Local Loop Shaping and IIR-filter Design For Precision Motion Control

Authors:
Inverse-Based Local Loop Shaping and
IIR-Filter Design For Precision Motion
Control
Xu Chen Atsushi Oshima ∗∗ and Masayoshi Tomizuka
Department of Mechanical Engineering, University of California,
Berkeley, CA, 94720, USA (e-mails:
{maxchen,tomizuka}@me.berkeley.edu).
∗∗ Mechatronics Development Center, NSK Ltd, Fujisawa, Kanagawa
251-8501, Japan (email:atsuoshi508@gmail.com)
Abstract: In motion-control problems such as vibration rejection, periodical reference tracking,
and harmonic disturbance cancellation, the disturbances/references share a common characteris-
tic of exhibiting concentrated energies at multiple bands of frequencies. In this paper, we discuss
a feedback loop-shaping approach to address such a class of control problem. An integration of
(inverse) system models is proposed to bring enhanced high-gain control at the required local
frequency regions. We show that such servo enhancement can be effectively achieved if good
model information is available at the disturbance frequencies, and that a rich class of design
tools can be integrated for the controller formulation. The proposed algorithm is verified in
simulation and experiments on vibration rejection in hard disk drives and an electrical power
steering system in automotive vehicles.
Keywords: digital control, vibration rejection, digital-filter design, loop shaping, hard disk
drives, electrical power steering
1. INTRODUCTION
Driven by the ever increasing demand for higher accuracy,
faster speed, and more robust performance, customized
control design is becoming more and more essential in
precision-motion-control systems. For example, in the ap-
plication to all-in-one personal computers and smart TVs,
modern hard disk drives (HDDs) are placed close to high-
power audio speakers that generate a significant amount of
vibrations. Due to the nature of the disturbances (Deller
et al., 1999), these vibrations occur in several concentrated
bands of frequencies, near or even above the bandwidth
of the servo system. Fig. 1 demonstrates the impact of
actual audio vibrations on an HDD benchmark problem
(IEEJ, Technical Commitee for Novel Nanoscale Servo
Control, 2007). Despite the fact that a set of baseline
controllers have been designed to meet standard indus-
trial requirements, 1in the presence of strong vibrations,
we observe that common feedback design has difficulty
attenuating the local spectral peaks at around 880 Hz
and 1600 Hz. The servo challenge is additionally amplified
by the increasing demand on servo accuracy, as the HDD
disk density continues growing to meet the requirements
in modern and future data storage applications.
A related type of disturbance, with its spectral peaks
much sharper than those in Fig. 1, is the narrow-band
disturbance consisting of single-frequency vibrations. Such
disturbance is common in motion control that involves
periodic movements. Examples include but are not limited
1A set of notch filters and an PID controller have been applied to
attenuate the resonances and achieve a 1.19kHz bandwidth here.
500 1000 1500 2000 2500 3000
0
0.005
0.01
0.015
0.02
0.025
0.03
Frequency (Hz)
Magnitude
Fig. 1. A typical HDD error spectrum under vibrations
to: (i) engine noise in turboprop aircraft and automobiles
(Shoureshi and Knurek, 1996) (ii) repeatable runout, disk
flutter, and fan noise in HDDs (Ehrlich and Curran,
1999; Guo and Chen, 2001) (iii) vibration in suspension
systems (Landau et al., 2009) and (iv) repetitive trajectory
tracking (Tomizuka, 2008). Customized servo design is
essential for attenuating these disturbances.
From the feedback-control perspective, both reference
tracking and disturbance attenuation are about shaping
the dynamic behavior of the servo loop. Based on the spec-
tral properties, we will regard narrow-band disturbances,
and vibrations that generate residual errors similar to that
in Fig. 1, as to belong to the same class of signals. We
denote such disturbances as band-limited disturbances, and
the corresponding closed-loop control design as local loop
shaping.
The application of inverse system models is proposed to
address the aforementioned problem. Inverse-based control
has long been used in motion control (for example, in feed-
forward designs), but not extensively explored in feedback
vibration rejection, especially for attenuating audio vibra-
tions. Different from narrow-band loop shaping that can be
achieved via e.g., peak filters (Zheng et al., 2006), repet-
itive control (Chew and Tomizuka, 1989; Cuiyan et al.,
2004), Youla parameterization (Landau et al., 2009), and
narrow-band disturbance observers (Chen and Tomizuka,
2012), general band-limited vibrations are much more
challenging to reject, since enhanced local loop shape is
usually accompanied by deteriorated servo performance
at other frequencies. This is the “waterbed effect” from
Bode’s Integral Theorem. To balance between the desired
vibration rejection and the theoretical boundaries, we in-
vestigate methods to place a group of structured poles
and zeros for local loop shaping, and provide the design of
infinite-impulse-response (IIR) digital filters to reduce the
disturbance amplification.
The remainder of the paper is organized as follows. Section
2 discusses the controller structure and the loop-shaping
idea. Section 3 provides the customized filter design tech-
niques. Section 4 shows the simulation and experimental
results on vibration rejection. Section 5 concludes the
paper.
2. CONTROLLER STRUCTURE
Consider a general discrete-time feedback system with
the plant and the stabilizing negative-feedback controller
given by P(z1) and C(z1) respectively. In motion con-
trol, C(z1) is designed such that y(k), the output of
the plant, tracks the reference r(k) while rejecting the
disturbance d(k). Corresponding, at frequencies below the
bandwidth of the closed loop, the complementary sensitiv-
ity function To(z1) = Gry(z1) = P(z1)C(z1)/[1 +
P(z1)C(z1)] approximates unity and the sensitivity
function (a.k.a. the output disturbance-rejection function)
So(z1) = 1To(z1) = 1/[1+P(z1)C(z1)] should have
small magnitude at locations where d(k) presents large
spectral peaks.
Consider adding three elements zmˆ
P1(z1), zm, and
Q(z1) around C(z1) as shown in Fig. 2. Computing the
new sensitivity and complementary sensitivity functions
gives (due to space limit, we drop the index (z1) here)
T=¯
Gry=P C +zmQˆ
P1P
1 + P C +zmQ(ˆ
P1P1) (1)
S= 1 T=1zmQ
1 + P C +zmQ(ˆ
P1P1).(2)
Here ˆ
P1(z1) is defined as the inverse model of P(z1).
We add the delay element zmso that zmˆ
P1(z1) is
realizable if the relative degree of P(z1) is larger than
zero in Fig. 2. ˆ
P1(z1) is assumed to be stable. This
is practically not difficult to satisfy for motion-control
systems (see, e.g., Ohnishi et al. (1996); Tomizuka (1987)).
For the case of unstable P1(z1), a related discussion of
optimal inverse design is provided in Chen et al. (2013).
To see the role of Q(z1), note first that when ˆ
P1P= 1,
(1) and (2) reduce to
T(z1) = P(z1)C(z1) + zmQ(z1)
1 + P(z1)C(z1).(3)
S(z1) = 1zmQ(z1)
1 + P(z1)C(z1)(4)
If in addition zmQ(z1) = 1, then we have S(z1)=0
and T(z1) = 1 in the above equations, i.e., perfect
disturbance rejection and reference tracking. Furthermore,
the right hand side of (4) equals So(z1)(1 zmQ(z1)),
where So(z1) is the baseline sensitivity function com-
puted at the end of the first paragraph in this section.
S(z1) is hence decomposed to have the additional free-
dom to shape the loop via 1 zmQ(z1), while the
original sensitivity function So(z1) remains intact.
+-
r(k
)
y(k
)
C
(z
¡
1
)
(z
¡
1
)
z
¡
m
z
¡
m
^
P
¡
1
(z
¡
1
)
Q
(z
¡
1
)
d
(k
)
+
+
+
+
+ +
Fig. 2. Block diagram of proposed loop shaping
The above is an ideal-case analysis due to the perfect-
model assumption. The condition zmQ(z1) = 1 is also
not practical as it requires an anti-causal Q(z1) = zm.
However, replacing every z1with eand re-evaluating
the equations, we see that (4) and (3) still hold in the
frequency domain provided that ˆ
P1(e )P(e ) = 1.
At this specific ωvalue, we still have S(e)=0
and T(e ) = 1 if emjωQ(ej ω ) = 1. Therefore, en-
hanced local loop shaping remains feasible at the fre-
quency regions where good model information is avail-
able. At the frequencies where there are large model
mismatches, efficient servo control is intrinsically dif-
ficult from robust-control theory. We will thus make
emjω Q(e )0, to keep the influence of the un-
certainty elements ejmω Q(e )ˆ
P1(e )P(e ) and
zmQ(e )[ ˆ
P1(e )P(e )1] small in (2) and (1).
More formally, if the plant is perturbed to ˜
P(e ) =
P(e )(1 + ∆(e )),2standard robust-stability anal-
ysis (see, e.g., Chap. 7.5 of Skogestad and Postlethwaite
(2005)) gives that the closed-loop system is stable if and
only if the following hold:
nominal stability–the closed loop is stable when
∆(e ) = 0 ω, i.e., the nominal Nyquist plot has
the correct number of encirclements around (-1,0) in
the complex plane, and does not touch (-1,0).
robust stability–ω,
∆(e )T(e )
<1, where
Tis given by (1). This additionally guarantees that
2∆ is assumed to be stable and has finite magnitude response.
the perturbed Nyquist plot does not touch the (-1,0)
point.
From (4) and (3) we see that nominal stability is satisfied
as long as Q(z1) is stable. If Q(e ) = 0, the add-on loop
shaping is essentially cut off in Fig. 2, and (1) reduces to
T(e )To(e ). We thus have
∆(e )T(e )
∆(e )To(e )
<1, which is the robust stability
condition for the baseline feedback loop, and is satisfied
by our assumption that C(z1) is a stabilizing controller.
To summarize, in regions where good model information
is available, ejmω Q(e )1 in (4) gives small gain
in S(e ) and we have enhanced servo performance at
this local frequency region; at frequencies where vibrations
do not occur or there are large model mismatches, letting
ejmω Q(e )0 maintains the original loop shape and
system stability. This concept will be the central of our
discussions in the following sections.
Finally we remark that when P(z1) is minimum-phase
then the proposed controller structure forms a special
Youla parametrization (see e.g., Zhou and Doyle (1998)), 3
which indicates that any controller that stabilizes the feed-
back system can be formed by picking some stable Q(z1)
in Fig. 2. The design of this Q filter in Youla parametriza-
tion however does not have a common rule. With the plant
inversion already achieved by ˆ
P1(z1), the next section
discusses how we can incorporate structured designs in
Q(z1) for band-limited local loop shaping.
3. BAND-LIMITED LOCAL LOOP SHAPING
3.1 General Concept
Recall from Fig. 1, that band-limited disturbances show
peaks in the error spectrum. To introduce small gains at
these local frequencies in the sensitivity function (4), we
consider the following design on Q(z1):
1zmQ(z1) = A(z1)
A(αz1)K(z1) (5)
Q(z1) = BQ(z1)
A(αz1)(6)
where 0 < α < 1 and
A(γz1),12γcos ω0z1+γ2z2, γ = 1, α (7)
=(1γej ω0z1) (1γej ω0z1).(8)
From the factorization (8), A(z1)/A(αz1) in (5) is
a special notch filter, with its zeros and poles respec-
tively given by e±0and αe±j ω0.ω0is in the unit
of radians here, and equals 2π0Ts(Ω0is in Hz, Ts
is the sampling time in sec). The zeros e±0provide
small gains in A(z1)/A(αz1) around the center fre-
quency ω0(can check that A(e0) = 0 from (8)). The
poles αe±0balance the magnitude response such that
A(e )/A(αe )1 when ωis far away from ω0. The
structured poles and zeros in (5) will be absorbed to the
sensitivity function due to the construction of (4), and
provides the desired local modification to the loop shape.
Approximately, the -3dB bandwidth for this notch filter
is given by (1 α2)/[(α2+ 1)(πTs)]. The notch shape
3The proof is omitted here due to space limit.
of |A(e )/A(αe )|becomes sharper and sharper as
αgets closer to 1.
We will demonstrate the placing of one notch shape in the
sensitivity function. The design of placing multiple notches
is analogous after replacing (7) with A(γz1) = n
i=1(1
2γcos ωiz1+γ2z2).
In (5), the plant delay mis usually non-zero for practical
control systems. K(z1) functions in this case to satisfy
the causality of Q(z1): without K(z1), equations (5-7)
give
Q(z1) = (α21)z2+m(α1) 2 cos ω0z1+m
12αcos ω0z1+α2z2(9)
which contains the unrealizable term z1+mif m > 1.
Assigning K(z1) = k0+k1z1+· · · +knKznK(an FIR
filter) reduces (5) to
A(z1)K(z1) + zmBQ(z1) = A(αz1) (10)
where A(z1), zm, and A(αz1) are known from the
previous design. Matching the coefficients of z1, we
can solve for BQ(z1) and K(z1). 4The minimum-
order solution satisfies deg(BQ(z1)) = deg(K(z1)) +
deg(A(z1)) mand deg(A(αz1)) deg(BQ(z1)) + m.
The dashed line in Fig. 3 presents an example of the solved
Q(z1) and 1zmQ(z1), with α= 0.993, Ts= 1/26400
sec, Ω0= 3000 Hz, and m= 2. We observe that 1
zmQ(z1) is approximately unity except at the desired
attenuation frequency Ω0= 3000 Hz, where we have 1
emjω0Q(ejω0) = K(e0)A(e0)/A(αe0) = 0 due
to A(e0) = 0 in (8). Thus, from (2), S(e0) = 0
and disturbances at 3000 Hz gets perfectly attenuated.
Meanwhile, the magnitude of Q(z1) reduces from 1
at 3000 Hz quickly down to -35dB (0.0178 in absolute
value) in the low-frequency region, and -50dB (0.0032)
in the high-frequency region. As discussed in the last
section, these small gains reduce the influence of the model
uncertainties, so that (4) and (3) are valid approximations
of (2) and (1).
With the same center-frequency configuration, the solid
line in Fig. 3 shows the solved Q(z1) and 1zmQ(z1)
for α= 0.945. For this Q(z1), the width of the pass band
at -3dB is approximately 475 Hz. We can see that the
desired loop shaping is also effectively achieved, and that
such a Q(z1) suits for rejecting disturbances with wide
spectral peaks such as the one described in Fig. 1.
Using the designs in Fig. 3, if we consider a standard
feedback design of 1/(1 + P(z1)C(z1)) and use (4), we
obtain the magnitude responses of S(z1) in Fig. 4. The
dotted line is the magnitude response of So(z1) = 1/(1 +
P(z1)C(z1)). In this example, we used a loop shape
that is common for motion control. The solid and the
dashed lines are the magnitude responses of S(z1) after
we introduce the proposed designs in Fig. 3. We observe
that the shape of 1 zmQ(z1) is directly reflected to
S(z1) due to the relationship S(z1)So(z1)(1
zmQ(z1)).
4(10) is a Diophantine equation. Some existing solvers are provided
in Landau and Zito (2006).
103104
−20
−15
−10
−5
0
Magnitude (dB)
1−zmQ(z−1)
103104
−40
−30
−20
−10
0
Magnitude (dB)
Frequency (Hz)
Q(z−1)
α=0.945
α=0.993
Fig. 3. Magnitude responses of two example Q(z1)’s and
the corresponding 1 zmQ(z1)’s
101102103
−40
−30
−20
−10
0
10
Frequency (Hz)
Magnitude (dB)
baseline
α = 0.945
α = 0.993
Detail at 3000 Hz
Fig. 4. Magnitude responses of the sensitivity functions
with the designs in Fig. 3
3.2 Additional Customization
It is not practical to have an ideal bandpass filter that
equals either unity or zero at each frequency. In Fig. 3,
as the magnitude response of Q(z1) gets sharper and
sharper, we more and more approximate an ideal filter
that passes only the frequency component at 3000 Hz.
In order to achieve the solid line, i.e., a larger-bandwidth
Q(z1), and hence a wider range of attenuation frequencies
in Fig. 4, the computation of (10) has traded off the small
magnitudes of Q(z1) at frequencies far away from 3000
Hz. As a result,
in the top plot of Fig. 3, the solid-line |1
ejmω Q(e )|becomes larger than unity at low and
high frequencies;
the robustness of the algorithm against plant uncer-
tainty is decreased as the term zmQ(ˆ
P1P1) may
become not negligible in (2).
These concerns will not occur in narrow-band loop shap-
ing, where the bandwidth of Q(z1) is very small, but
should be examined with care in the case of rejecting gen-
eral band-limited disturbances. We discuss next methods
to reduce the above design trade offs.
3.2.1 The effect of fixed zeros in Q(z1)
In Section 3.1, we chose the IIR Q(z1) with a cus-
tomized denominator A(αz1) but have not placed specific
structural designs for BQ(z1). Actually BQ(z1) is the
unknown to be solved in (10) and its nontrivial frequency
response is solely determined by the algebraic equation.
Using the concept in pole placement, we can add a fixed
part B0(z1) such that
BQ(z1) = B0(z1)B
Q(z1).(11)
Designing B0(z1) = 1 + z1for example, will yield
Q(e )
ω=π=BQ(1)/A(α) = 0, i.e., zero magnitude
at Nyquist frequency. More generally, introducing fixed
zero near z=1 and/or z= 1 in the z plane will provide
enhanced small gains for Q(z1) in the high- and/or low-
frequency region. Extending this idea, we can essentially
place magnitude constraints at arbitrary desired frequen-
cies, by letting B0(z1) = 12βcos ωpz1+β2z2in (11),
which places the fixed zeros βe±pto penalize
Q(e )
near ω=ωp.
Table 1 summarizes the effects of different configurations
for the fixed term B0(z1). For band-limited loop shap-
ing, it is natural to place magnitude constraints at both
low and high frequencies. In this case, the modules in
Table 1 can be combined to provide, e.g., B0(z1) =
(1 + ρz1)n1(1z1)n2, where ρ[0.5,1]; and n1,n2
are non-negative integers.
Table 1. Effects of placing fixed zeros to Q(z1)
B0(z1) zeros small |Q(e)|
1 + z11 around Nyquist freq.
1 + ρz1, ρ [0.5,1] ρat high freq.
1ρz1, ρ [0.5,1] ρat low freq.
12βcos ωpz1+β2z2
βe±j ωparound ωp
β(0,1]
1z11 at low freq.
3.2.2 Cascading a bandpass IIR filter to Q(z1)
By (11) we essentially have cascaded the FIR filter
B0(z1) to Q(z1). B0(z1) has been designed to con-
trol the magnitude response of Q(z1) at some specific
frequency regions. It is well known that IIR design has
additional flexibility compared to FIR filters. From the
frequency-response perspective, cascading two bandpass
filters with the same center frequency provides an new
bandpass Q(z1), which can have reduced magnitudes
at all frequencies outside the passband. This suggests
to assign an IIR bandpass B0(z1) and let Q(z1) =
Q0(z1)B0(z1), where Q0(z1) is the fundamental so-
lution from Section 3.1. Note that (5) indicates the
frequency-domain equality
1ejmω Q0(e ) = A(ej ω )
A(αe )K(e ).(12)
At ω=ω0, we have A(e0) = 0 from (8) and hence
Q0(e0) = ej 0at the center frequency ω0. Due to
this result, Q0(z1) is not a conventional bandpass filter,
and B0(z1) needs to satisfy B0(e0) = 1 to preserve
the property Q0(e0)B0(ej ω0) = Q0(e0) = ejmω0.
A standard bandpass filter will suffice this requirement.
Recall that A(z1)/A(αz1) is a notch filter. One candi-
date B0(z1) is 1 ηA(z1)/A(αz1): η(0,1] (unity
minus a notch shape generates a bandpass shape).
Fig. 5 presents the Q(z1) and 1 zmQ(z1) solved
from the discussed algorithms in this section. The solid
lines are the direct solution from Section 3.1; the dashed
lines are from Section 3.2.1; and the dotted lines from
Section 3.2.2. We observe from the magnitude responses
of 1 zmQ(z1), that all three methods create the
required attenuation around 3000 Hz. Also, the additional
magnitude constraints on Q(z1) are effectively reflected
in the bottom plot of Fig. 5: in the dashed-line Q(z1), the
design of B0(z1) = 1+ 0.7z1places a zero z=0.7 near
the Nyquist frequency (z=eπ=1), yielding the small
gain in the high-frequency region compared to the solid-
line Q(z1); in the dotted-line Q(z1), by cascading the
bandpass filter B0(z1) we have reduced the magnitude of
Q(z1) at both low and high frequencies.
Finally we note the presence of the “waterbed effect” in
Fig. 5, a result of the fundamental limitation of feedback
loop shaping. Besides the strong disturbance attenuation
around 3000 Hz, the magnitude of 1 zmQ(z1) holds
values higher than unity at other frequencies. The fun-
damental solution (solid line) from Section 3.1 evenly
spreads the amplification throughout the entire frequency
region; the dashed line has enhanced robustness at high
frequencies and makes 1 zmQ(z1) closer to unity
near Nyquist frequency; lastly the IIR-B0(z1) algorithm
concentrates the amplification near the center frequency
3000 Hz. In general, it is preferred to evenly spread the
amplifications, so the solid or the dashed lines are pre-
ferred from the performance perspective. Yet if large model
uncertainty exists which enforces Q(z1) to have small
magnitudes at high and/or low frequencies, the dotted line
may be considered over the other designs. Nonetheless,
the maximum amplification among all designs is around
1.6dB (1.2023) while the attenuation is more than -20dB
(0.1) in a large frequency region (perfect attenuation at
the center frequency 3000 Hz). One way to “smoothen”
the waterbed effect is to replace Q(z1) with kQ(z1) at
the final stage of design. This will trade off the perfect
disturbance attenuation with a less amplified loop shape
at other frequencies.
4. SIMULATION AND EXPERIMENTAL RESULT
4.1 Audio-vibration rejection in HDDs
In this section we apply the discussed control schemes for
audio-vibration rejection on a HDD benchmark system
(IEEJ, Technical Commitee for Novel Nanoscale Servo
Control, 2007). In this case study, the plant dynamics
involves a single-stage actuator that is powered by a
voice coil motor, and the control aim is to regulate the
plant output in the presence of external disturbances.
The frequency response of the sampled plant (sampling
frequency Fs= 26400 Hz) is shown in Fig. 6. The multiple
resonance modes are attenuated via several notch filters,
and the notched plant is treated as P(z1) in Fig. 2.
The plant delay in this case is m= 3. The baseline
design C(z1) is a PID controller which achieves a servo
103104
−8
−6
−4
−2
0
2
Magnitude (dB)
1−zmQ(z−1)
103104
−50
−40
−30
−20
−10
0
Magnitude (dB)
Frequency (Hz)
Q(z−1)
B0 = 1
B0 = 1 + 0.7z−1
IIR B0
Fig. 5. Comparison of the magnitude responses in three
designs of Q(z1) for wide-band disturbances: in the
dotted line, B0(z1) = 1 αA(z1)/A(αz1)
bandwidth of 1.19 kHz. The disturbance source is from a
scaled version of actual experimental results under audio
vibration.
The top plot of Fig. 7 shows the spectrum of the posi-
tion error signal (PES) without the proposed local loop
shaping. To attenuate the main spectral peaks centered at
880 Hz and 1600 Hz, we apply the algorithm in Section
3.1, with a two-band Q(z1), α= 0.945, and kQ(z1) =
0.8Q(z1). Fig. 8 presents the magnitude responses of
the loop shaping elements, where we observe the deep
attenuation at the desired 880 Hz and 1600 Hz. The
bottom plot of Fig. 7 shows the resulted PES, where the
spectrum has been greatly flattened compared to the case
without compensation, and we can see a direct projection
of the shape of 1 zmQ(z1) to the error spectrum.
The corresponding time traces are provided in Fig. 9.
The three-sigma value (sigma is the standard deviation)
has reduced from 33.54% TP (Track Pitch) to 22.79% TP
(here 1 TP = 254 nm), i.e., a 29.07 percent improvement.
Notice that 1600 Hz is above the bandwidth of the servo
system, where disturbance rejection was not feasible for
the original feedback design.
Additional customization on Q(z1) is carried out using
the discussions in Section 3.2. Table 2 shows the attenua-
tion results on the three-sigma value of the position errors.
For all designs in the last three columns, we have enforced
Q(z1) to have small gains in the high-frequency region.
Actually if the high-frequency gain of Q(z1) is larger
than -10 dB the closed-loop becomes unstable. This can
be projected by plotting the frequency response of S(z1)
and noting that the term zmQ(ˆ
P1P1) is no longer
negligible. We observe that all designs provided large servo
enhancements by at least 21.85% reduction on the three-
sigma values.
Table 2. Comparison of servo enhancement
B0(z1), i.e., the fixed part in Q(z1) 1 1 + z11z2IIR bandpass 1 αA(z1)/A(αz1)
3σw/ compensation (%TP) 23.79 24.33 24.76 26.21
3σreduction (baseline 33.54%TP) 29.07% 27.46% 26.18% 21.85%
102103104
0
50
100
Gain (dB)
102103104
−180
−90
0
90
180
Phase (degree)
Frequency (Hz)
Fig. 6. Frequency response of the plant in the HDD
benchmark
500 1000 1500 2000 2500 3000
0
0.01
0.02
0.03
Magnitude
w/o compensation
w/o compensation 3σ = 33.54 %TP
500 1000 1500 2000 2500 3000
0
0.01
0.02
0.03
Frequency (Hz)
w/ compensation
Magnitude
w/ compensation 3σ = 23.79 %TP
Fig. 7. Spectra (FFT) of the position error signals with
and without compensation
4.2 Harmonic cancellation in an EPS system
In this section we apply the algorithm to an electrical-
power-steering (EPS) system described in Sugita and
Tomizuka (2012). This system features a variable-gear-
ratio (VGR) control module for enhanced vehicle steering.
VGR control provides speed-dependent assistive torque
during steering, to reduce driving effort and to improve
driver comfort and safety. The problem of narrow-band
disturbance rejection occurs as the VGR system generates
unnatural reaction torques and contains imperfections in
the gear and motor rotations.
One level of the motion control involves a velocity feedback
servo loop, where the motor accepts torque input and the
rotational velocity is the output. In this case, the plant
under control is G(s) = 1/(Jms+Bm), where Jmis
100101102103104
−20
−15
−10
−5
0
Magnitude (dB)
1−zmQ(z−1)
100101102103104
−25
−20
−15
−10
−5
0
Magnitude (dB)
Frequency (Hz)
Q(z−1)
Fig. 8. 1 zmQ(z1) and Q(z1) used for vibration
rejection in HDDs
21 21.5 22 22.5 23
−30
−20
−10
0
10
20
30
Revolution
PES (%TP)
w/ compensation
w/o compensation
Fig. 9. Time traces of the position error signals with and
without compensation
the inertia of the motor and Bmis the friction damping
coefficient. This plant is discretized at a sampling time of
1 ms. The system has a 4ms input delay, yielding m= 4
in the local loop shaping design.
The top plot of Fig. 10 demonstrates the motor velocity
during a variable-speed steering test (experimental re-
sults), where we observe large tracking errors between the
reference velocity and the actual motor speed. Further in-
vestigation shows that there are strong narrow-band vibra-
tions due to imperfect motor rotations. The dashed line in
Fig. 11 presents the spectrum of the tracking errors when
we apply a constant-speed steering. The strong spectral
peak at 15 Hz contributes greatly to the tracking errors.
Applying the proposed algorithm in Section 3.1 yields the
solid line in Fig. 11. Visual comparison indicates that the
algorithm has removed the original spectral peak at 15 Hz.
Computing the standard deviations for the tracking errors,
we obtain a 74.77% error reduction on the 3σvalue (from
0.20826 rad/s to 0.052549 rad/s). Besides these constant-
speed steering results, back to comparing the variable-
steering-speed test in Fig. 10, the algorithm also provides
significant performance enhancement, both visually and
quantitatively (3σreduces from 0.19402 to 0.102). Here the
vibration frequency no longer stays at 15 Hz but actually
varies based on the steering speed. An adaptive Q(z1)
is used to obtain the bottom plot of Fig. 10, using the
identified relationship between the steering speed and the
vibration frequency.
0 0.2 0.4 0.6 0.8 1 1.2
−1
−0.5
0
0.5
1
Steering speed (rad/s)
w/o compensation: error 3σ = 0.19402
Reference
Measurement
0 0.2 0.4 0.6 0.8 1 1.2
−1
−0.5
0
0.5
1
Time (sec)
Steering speed (rad/s)
w/ compensation: error 3σ = 0.102
Reference
Measurement
Fig. 10. Time traces of the EPS tracking result during
variable-speed steering
5 10 15 20 25 30 35 40 45 50 55 60
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
Frequency (Hz)
Magnitude (rad/s)
w/o compensation: 3σ = 0.20826
w/ compensation: 3σ = 0.052549
Fig. 11. Spectra of the tracking errors during constant
speed steering
5. CONCLUSION
In this paper we have discussed an algorithm for enhancing
the servo performance at several bands of frequencies. In
the presence of bandwidth limitations in feedback design,
such a local loop shaping approach suits well for at-
tenuating strong vibration-type disturbances. Simulation
and experimental results support the proposed design and
analysis.
6. ACKNOWLEDGMENT
This work was supported in part by the Computer Me-
chanics Laboratory (CML) in the Department of Mechan-
ical Engineering, University of California at Berkeley.
REFERENCES
Chen, X., Castellanos Silva, A., and Tomizuka, M. (2013).
Adaptive model inversion for rejection of time-varying
vibrations on a benchmark problem. In 2013 European
Control Conference (submitted).
Chen, X. and Tomizuka, M. (2012). A minimum parameter
adaptive approach for rejecting multiple narrow-band
disturbances with application to hard disk drives. IEEE
Transactions on Control System Technology, 20(2), 408
–415. doi:10.1109/TCST.2011.2178025.
Chew, K. and Tomizuka, M. (1989). Digital control of
repetitive errors in disk drive systems. In Proceedings of
the 1989 American Control Conference, 540–548. IEEE.
Cuiyan, L., Dongchun, Z., and Xianyi, Z. (2004). A survey
of repetitive control. In Proceedings of 2004 IEEE/RSJ
International Conference on Intelligent Robots and Sys-
tems., volume 2, 1160–1166. IEEE.
Deller, J.R., Hansen, J.H.L., and Proakis, J.G. (1999).
Discrete-time processing of speech signals. Wiley-IEEE
Press.
Ehrlich, R. and Curran, D. (1999). Major HDD TMR
sources and projected scaling with TPI. IEEE Transac-
tions on Magnetics, 35(2), 885–891.
Guo, L. and Chen, Y.J. (2001). Disk flutter and its
impact on hdd servo performance. IEEE Transactions
on Magnetics, 37(2), 866 –870. doi:10.1109/20.917633.
IEEJ, Technical Commitee for Novel Nanoscale Servo
Control (2007). NSS benchmark problem of hard disk
drive systems. http://mizugaki.iis.u-tokyo.ac.jp/nss/.
Landau, I.D., Constantinescu, A., and Alma, M. (2009).
Adaptive regulation-Rejection of unknown multiple nar-
row band disturbances. In 17th Mediterranean Confer-
ence on Control & Automation, 1056–1065. IEEE.
Landau, I.D. and Zito, G. (2006). Digital control systems:
design, identification and implementation. Springer
Verlag.
Ohnishi, K., Shibata, M., and Murakami, T.
(1996). Motion control for advanced mechatronics.
IEEE/ASME Trans. Mechatronics, 1(1), 56 –67.
doi:10.1109/3516.491410.
Shoureshi, R. and Knurek, T. (1996). Automotive appli-
cations of a hybrid active noise and vibration control.
IEEE Control Systems Magazine, 16(6), 72–78.
Skogestad, S. and Postlethwaite, I. (2005). Multivariable
feedback control analysis and design. New York.
Sugita, S. and Tomizuka, M. (2012). Cancellation of un-
natural reaction torque in variable-gear-ratio. Journal of
Dynamic Systems, Measurement, and Control, 134(2),
021019.
Tomizuka, M. (1987). Zero phase error tracking algorithm
for digital control. ASME Journal of Dynamic Systems,
Measurements, and Control, 109(1), 65–68.
Tomizuka, M. (2008). Dealing with periodic dis-
turbances in controls of mechanical systems. An-
nual Reviews in Control, 32(2), 193 – 199. doi:
10.1016/j.arcontrol.2008.07.002.
Zheng, J., Guo, G., Wang, Y., and Wong, W. (2006).
Optimal narrow-band disturbance filter for pzt-actuated
head positioning control on a spinstand. IEEE Trans-
actions on Magnetics, 42(11), 3745–3751.
Zhou, K. and Doyle, J.C. (1998). Essentials of robust
control. Prentice Hall New Jersey.
... To improve the positioning accuracy of linear motors through the compensation scheme for the cogging force has thus become another effective approach. Yao et al (2011) proposed an adaptive robust control method to solve the force disturbance problem, taking the cogging force as unknown modelling uncertainties; Luo et al (2010) adopted the fractional order robust control into the compensation of detent force; Chen et al (2012Chen et al ( , 2013 considered the force fluctuation in the motion control as multiple narrow-band disturbances with unknown frequency, and designed adaptive digital filter to eliminate the adverse effect; Luo et al (2010) noticed the periodicity of the cogging force, and proposed a Abstract: The cogging force of linear motors significantly affects the positioning accuracy in ultraprecision lithography machines. Optimal motor design cannot completely eliminate it, leaving the realtime active compensation as the only means to counteract its adverse effect. ...
... To improve the positioning accuracy of linear motors through the compensation scheme for the cogging force has thus become another effective approach. Yao et al (2011) proposed an adaptive robust control method to solve the force disturbance problem, taking the cogging force as unknown modelling uncertainties; Luo et al (2010) adopted the fractional order robust control into the compensation of detent force; Chen et al (2012Chen et al ( , 2013 considered the force fluctuation in the motion control as multiple narrow-band disturbances with unknown frequency, and designed adaptive digital filter to eliminate the adverse effect; Luo et al (2010) noticed the periodicity of the cogging force, and proposed a ...
... To improve the positioning accuracy of linear motors through the compensation scheme for the cogging force has thus become another effective approach. Yao et al (2011) proposed an adaptive robust control method to solve the force disturbance problem, taking the cogging force as unknown modelling uncertainties; Luo et al (2010) adopted the fractional order robust control into the compensation of detent force; Chen et al (2012Chen et al ( , 2013 considered the force fluctuation in the motion control as multiple narrow-band disturbances with unknown frequency, and designed adaptive digital filter to eliminate the adverse effect; Luo et al (2010) noticed the periodicity of the cogging force, and proposed a learning algorithm to reduce the disturbance. A common disadvantage for the existing methods is that the compensation process is time consuming due to the algorithm complexity, therefore they could hardly meet the requirement for some highly fast real-time applications. ...
Article
The cogging force of linear motors significantly affects the positioning accuracy in ultraprecision lithography machines. Optimal motor design cannot completely eliminate it, leaving the realtime active compensation as the only means to counteract its adverse effect. This requires the real-time identification of the cogging force, which is however a challenging issue. Traditional approaches are either computationally too expensive or practically infeasible. This paper proposes to develop a two-stage trained RBF neural model for the detent force identification of the ultra-precision wafer stage platform in a lithography machine. Experimental results confirm the effectiveness of this data-driven method.
... If HDDs encounter high-frequency widespectrum excitations, the amplified vibrations will greatly degrade the track-following performance. Thus, more considerations must be given at those frequencies to "locally" shape the sensitivity function for enhanced vibration suppression [18]. In this paper, we assume that all excitations enter the HDDs as a lumped disturbance d(k). ...
... Also, the frequency response of N(z −1 ) in Eq. (14) is not symmetric in low and high frequencies, as shown in Fig. 4. Although N(z −1 ) provides enhanced low gain near the notch center and at low frequencies, the high frequency gain should be carefully handled [18] to avoid undesired noise/disturbance amplifications. The wider the notch, the more pronounced the increase of the high-frequency gains, which may become problematic. ...
... (11), (13) and (16), we can solve for the Q filter once N L (z −1 ) is designed and the passband of the Q filter is also determined by α L . For example, if the system has one-step delay, i.e., m = 1, then the corresponding Q filter is given by Figure 6 shows the frequency response of Eq. (18), where the band-pass property is evident. ...
Conference Paper
Full-text available
In the big-data era, requirements for storage capacity and access speed in modern Hard Disk Drive (HDD) systems are becoming more and more stringent. As the track density of HDDs increases, vibration suppression of the recording arm in HDDs is becoming more challenging. Vibrations in modern HDDs are environment/product-dependent with different frequency characteristics. Furthermore, they can occur at very high frequencies with wide spectral peaks. This paper presents an adaptive algorithm to identify and suppress these high-frequency widespectrum vibrations. We design a vibration-compensation controller based on an adaptive disturbance observer (DOB), and devise parameter adaptation algorithms not only for the vibration frequencies but also for the spectral peak widths of the vibration. The peak-width parameters are adaptively tuned online to maximally attenuate the vibration with minimal error amplifications at other frequencies. The proposed algorithm is verified by simulations of HDDs for the problem of suppressing highfrequency wide-spectrum vibrations.
... The mechanical load of the manipulator by the workpiece is indicated by M ς . The structural[1][2][3][4][5][6]8,9] model (Fig. 2) also reflects the mode"without loading the wrist"(Fig.2.a), and operational mode "with max load an wrist"(Fig.2.b). The transmission functions i G (1)-(6), G (7), electromechanical propulsion of the arms taking into account M ς , are presented in the expanded structure (Fig.2.c,d). ...
Conference Paper
Full-text available
The industrial robot manipulator FANUC M-430IA/4FH is considered as a control plant. They are designed classical with PID-controller and robust DRC-Disturbance Rejection Control System. The quality and robust analysis of the designed systems are studied. https://ieeexplore.ieee.org/xpl/conhome/9311274/proceeding
... Step The simulation results Fig.3 y Fig.6 confirm the reaching of the desired quality of the system according to the synthesis criterion const V (boundary aperiodic transition process) (7) for optimization, but also show a significant influence in changing the order of the controlled object as a result of the multiplicative disturbance load impact M 9 in "with max load an wrist" mode. To overcome this operational dependency, in the paper is proposes to design a robust position control system [15][16][17] 3. Robust DRC position control system for 5-DOF FANUC M-430iA / 4FH industrial robotic manipulator In contrast to the discussed (7), Fig.2 classical system with PID-regulators (symbolically reflected in Fig.7, where i R * R { ) and from the known ones till now [1][2][3][4][5][6][7][8][9][10][11][12][13][14], a robust DRC position control system is offered ...
Conference Paper
This paper offers the results of the control capabilities of the FANUC M-430iA / 4FH robotic manipulator. Classical and DRC robotic position control systems have been synthesized. Both systems are modelled. Their quality has been researched and comparatively analysed in unloaded, maximum loaded and in operation vibration working modes. In Proc. of TechSys 2020, IOP Conf. Series: Materials Science and Engineering 878 (2020) 012007, IOP Publishing, doi:10.1088/1757-899X/878/1/012007, pp. 1-10 https://iopscience.iop.org/article/10.1088/1757-899X/878/1/012007/meta
... Step The simulation results Fig.3 y Fig.6 confirm the reaching of the desired quality of the system according to the synthesis criterion const V (boundary aperiodic transition process) (7) for optimization, but also show a significant influence in changing the order of the controlled object as a result of the multiplicative disturbance load impact M 9 in "with max load an wrist" mode. To overcome this operational dependency, in the paper is proposes to design a robust position control system [15][16][17] 3. Robust DRC position control system for 5-DOF FANUC M-430iA / 4FH industrial robotic manipulator In contrast to the discussed (7), Fig.2 classical system with PID-regulators (symbolically reflected in Fig.7, where i R * R { ) and from the known ones till now [1][2][3][4][5][6][7][8][9][10][11][12][13][14], a robust DRC position control system is offered ...
Article
Full-text available
This paper offers the results of the control capabilities of the FANUC M-430iA / 4FH robotic manipulator. Classical and DRC robotic position control systems have been synthesized. Both systems are modelled. Their quality has been researched and comparatively analysed in unloaded, maximum loaded and in operation vibration working modes
Article
Full-text available
A repetitive ML-DRC-system for an industrial robot manipulator has been proposed and analyzed. The performance and filtering properties of the system to effectively compensate for production mechanical vibrations have been studied. Results of robust analysis in conditions of a priori uncertainty are presented.
Conference Paper
Full-text available
Disturbance observer based (DOB) control has been implemented in motion control to reject unknown or time-varying disturbances. In this research, an internal model-based disturbance observer (DOB) design combined with a PID type feedback controller is formulated for wind turbine speed and power regulation. The DOB controller facilitates model-based estimation and cancellation of disturbance using an inner feedback control loop. The disturbance observer combined with a compensator is further designed to deal with the model mismatch. The proposed method is applied to National Renewable Energy laboratory (NREL) offshore 5-MW wind turbine. Our case studies show that the DOB controller can achieve improved speed and power regulation compared to the baseline PID controller, and exhibit excellent robustness under different turbulent wind fields. Copyright © 2016 by ASME Country-Specific Mortality and Growth Failure in Infancy and Yound Children and Association With Material Stature Use interactive graphics and maps to view and sort country-specific infant and early dhildhood mortality and growth failure data and their association with maternal
Article
Full-text available
A controller design is presented for directly shaping the sensitivity function in the feedback control of dual-input single-output (DISO) systems. We provide special forms of discrete-time Youla-Kucera (YK) parameterization to obtain stabilizing controllers for DISO systems, and discuss concepts to achieve a reduced-complexity formulation. The proposed YK parameterization is based on an inverse of the loop transfer function, which gives benefits such as clearer design intuition and reduced complexity in construction. The algorithm is verified by simulations and experiments on hard disk drive systems, with application to wideband vibration and harmonic disturbance rejections.
Article
The reticle stage of lithography is a high precision servo motion platform, which requires using macro movement of linear motor and micro movement of voice coil motor to realize an nm-level positioning precision and tracking. In order to increase the control effect and response speed of macro movement motor of reticle stage of lithography, there we present a control strategy in combination with feed-forward thrust fluctuation compensation based on query table, the PID controller, and robust ILC controller. Firstly, the method separate the tracking error that is caused by the motor thrust fluctuation, then according to the speed build a query table to construct feedforward thrust fluctuation model, which is used for system thrust fluctuation compensation; the PID controller aim to improve the anti-interference ability of the system; through adding the attenuation factor, the robust ILC controller ensure the tracking error would be astringe to zero. The results showed that this method can well restrain the effects of thrust fluctuation from the motor, and the macro movement motor positioning quickly and accurately.
Article
The cogging force disturbance of linear motor is one of the main factors affecting the positioning accuracy of ultraprecision moving platform. And this drawback could not be completely overcome by improving the design of motor body, such as location modification of permanent magnet array, or optimization design of the shape of teeth-slot. So the active compensation algorithms become prevalent in cogging force rejection area. This paper proposed a control structure based on internal mode principle to attenuate the cogging force of linear motor which deteriorated the accuracy of position, and this structure could make tracking and anti-disturbing performance of close-loop designed respectively. In the first place, the cogging force was seen as the intrinsic property of linear motor and its model constituting controlled object with motor ontology model was obtained by data driven recursive identification method. Then, a control structure was designed to accommodate tracking and anti-interference ability separately by using internal model principle. Finally, the proposed method was verified in a long stroke moving platform driven by linear motor. The experiment results show that, by employing this control strategy, the positioning error caused by cogging force was decreased by 70%.
Article
Full-text available
A digital feedforward control algorithm for tracking desired time varying signals is presented. The feedforward controller cancels all the closed-loop poles and cancellable closed-loop zeros. For uncancelled zeros, which include zeros outside the unit circle, the feedforward controller cancels the phase shift induced by them. The phase cancellation assures that the frequency response between the desired output and actual output exhibits zero phase shift for all the frequencies. The algorithm is particularly suited to the general motion control problems including robotic arms and positioning tables. A typical motion control problem is used to show the effectiveness of the proposed feedforward controller.
Article
Full-text available
Many servo systems are subjected to narrow-band disturbances that generate vibrations at multiple frequencies. One example is the track-following control in a hard disk drive (HDD) system, where the airflow-excited disk and actuator vibrations introduce strong and uncertain spectral peaks to the position error signal. Such narrow-band vibrations differ in each product and can appear at frequencies above the bandwidth of the control system. We present a feedback control scheme that adaptively enhances the servo performance at multiple unknown frequencies, while maintaining the baseline servo loop shape. A minimum parameter model of the disturbance is first introduced, followed by the construction of a novel adaptive multiple narrow-band disturbance observer for selective disturbance cancellation. Evaluation of the proposed algorithm is performed on a simulated HDD benchmark problem.
Article
Periodic disturbances are common in control of mechanical systems. Such disturbances may be due to rotational elements such as motors and vibratory elements. When the period of a periodic disturbance is fixed and known in advance, repetitive control can be used for attenuating their effect. The most popular repetitive controller is based on the internal model principle. When the period is not fixed and unknown, adaptation capability must be introduced. This paper presents some fundamental issues and new challenges in the design of controllers to deal with periodic disturbances along with applications to mechanical systems.
Article
Gives a comprehensive presentation of the methodology for design and implementation of computer controlled systems.
Article
Variable-gear-ratio steering, also known as active steering, is an advanced steering technology which enhances the driver's comfort and vehicle operability. However, one big problem, namely the unnatural reaction torque created by the variable actuator, restricts the further practical-use of the variable-gear-ratio steering. This paper proposes a control method to cancel the unnatural torque using a simple concept called friction relocation. Effectiveness of the method is experimentally confirmed using a hardware-in-the-loop simulator. [DOI: 10.1115/1.4004577]
Conference Paper
Tracking errors in disk drive systems have a significant repetitive component that is not explicitly taken into account in conventional servo controllers. Three specialized digital controllers applied as plug-in modules to a winchester disk drive with a pre-existing analog feedback controller to demonstrate their efficacy in the reduction of this periodic component. These controllers include the discrete time repetitive controller which is based on the internal model principle and feedforward cancellation controllers based on an adaptive prediction algorithm and on input and output measurements of the system.
Article
Periodic disturbances are common in control of mechanical systems. Such disturbances may be due to rotational elements such as motors and vibratory elements. When the period of a periodic disturbance is fixed and known in advance, repetitive control can be used for attenuating their effect. The most popular repetitive controller is based on the internal model principle. When the period is not fixed and unknown, adaptation capability must be introduced. This paper presents some fundamental issues and new challenges in the design of controllers to deal with periodic disturbances along with applications to mechanical systems.
Conference Paper
Repetitive control method is well known for its high performances in feedback systems that are subject to periodic exogenous inputs. This paper reviews the theory and applications of repetitive control based on both the internal model and external model, together with some repetitive control methods for nonlinear plants.