You are on page 1of 6

Robustness Analysis of a Disturbance-Observer

Based PI control
Mikuláš Huba
Slovak University of Technology in Bratislava
Ilkovičova 3, 812 19 Bratislava, Slovakia
Email: mikulas.huba

Abstract—This paper deals with a robustness analysis of a


disturbance observer (DO) based filtered PI control (DO-FPI)
for integral first order plants based on the performance portrait
method. For the loop robustification and a noise attenuation the
loop is augmented by n-th order filters used in the inner DO
loop, as well as in the outer loop with the stabilizing P controller.
Although the derived results may be used in a broad range of
tasks originating in a control of plants with a dominant first
order dynamics, for the sake of simplicity they focus on control
of a single integrator.

I. I NTRODUCTION
Figure 1. DO-FPI control for an integral first-order plant (2) with an n-th
In the disturbance observer (DO) based PID control concept order filter (9) and input and output disturbance di , δ-measurement noise
proposed already more than two decades ago [1], [2], a
central role is played by the DO filter required to achieve
a proper plant inversion, to avoid an algebraic loop, but also
to influence the loop robustness and the noise attenuation. In DO realization based on inversion of the plant dynamics and
a traditional DO based control, choice of an optimal filter aiming just to get a proper DO transfer function.
order and its influence on the resulting loop robustness and The possibility of a significantly improved nominal loop
performance has been recently treated for example by [3], performance against the traditional PI control has been pointed
[4], [5], [6]. Thereby, the robustness has been analyzed in out also for the first order plants with dead time [14], but
the frequency domain under a two-degree-of-freedom (2DOF) without giving a clear picture of a robustness of the new
design paradigm considering DO as a tool for reducing uncer- approach.
tainties and enabling a nominal performance motivated design II. P ROBLEM FORMULATION
of the main controller. A new less conservative tool for a
robust analysis has been introduced [3] and showed that by To keep the analysis as simple as possible, this paper
increasing the filter order the robustness decreases, although considers design of a disturbance observer base filtered PI
the performance may increase [7]. However, without a clear (DO-FPI) controllers (Fig. 1) for a single integrator with an
recommendation regarding an optimal filter degree choice. input disturbance di , with y being the plant output
The presented closed loop analysis of the robust perfor- ẏ = Ks (ur + di ) (1)
mance issues in a DO based control of the first order integrator
with an uncertain plant gain, acting disturbances and a possible Thereby, the control constraints on the plant input ur and
measurement noise has been inspired by previous experimental possbile output disturbances will be neglected here. Where
works devoted to the disturbance observer based filtered PI appropriate, the plant will also be described by a “pole-zero
and PID control. These are based on a fully new two stage form“ transfer function
 
modular design paradigm dealing firstly with a basic filtered Y (s) Ks
F (s) = = (2)
controller tuning that may be (in a nominal case) directly Ur (s) di =0 s
adapted also by the loop extended by the DO. In [8], [9],
[10], [11], [12], [13] it has been shown that by such a For the loop robustification, as well as a noise attenuation,
noise attenuation motivated modification of the DO based PI this dominant loop dynamical term will further be extended
control it is possible significantly decrease a torque ripple of a by intentionally introduced filters (Fig. 1)
speed DC motor control expressed in terms of a relative total For quantitative evaluation of the speed of responses the
variance of the control signal. The achieved improvements IAE (Integral of Absolute Error) will be used defined as
Z ∞
against the traditional PI control go up to two digit figures
IAE = |e(t)| dt ; e = w − y ; w = setpoint (3)
and occur for a filter order that is higher than required for the 0
Although the DO based filtered PI control [14] has been shown TV (u ), n=1 TV (u )
1 s 1 d
to yield the same optimal tuning both for the setpoint as well 0.8 0.1 0.8 1
0.00
as for the disturbance steps, design of an optimal DO-FPI 0.6 1
00.0.001 0.6

Tn
0.4 0.1 0.4

1
1

0.
control will be based on minimizing the cost function 0.2 .011 0.2 0.0 1
0.00
00.0
0.5 1 1.5 2 2.5 0.5 1 1.5 2 2.5
IAEΣ = IAEs + IAEd (4) TV0(ys) TV1(yd)

under constraints put on the tolerable integral deviations from 0.8 0.8 1
0.1 0.0 .001
0

1
0.6 1 0.6
00.0.001

1
ideal shapes of the setpoint and disturbance step responses at

Tn
0.4 0.4 1
00.11 0.
0.2 0.2 0.001.001
the plant input and output [15], [14] that may be formulated 00.0.

in form of inequalities 0.5 1 1.5 2 2.5 0.5 1 1.5 2 2.5


IAE(ys) IAE(y )
d
T V0 (ys ) ≤ ys ; T V1 (yd ) ≤ yd 0.8 0.6 0.8
0.4
(5)
T V1 (us ) ≤ us ; T V1 (ud ) ≤ ud 0.6 0.6

7
0.5

0.
Tn
0.2

6
2
1. 1.5
0.4 0.6 0.4 8

1.
01
0.

8
0.7

0.
0.2 0. 0.8 0.2 0.4 .2 0.1
specified by a vector of chosen tollerable shape-related de- 9 0.9 01
0.
0.05
0.05
0.01
0.5 1 1.5 2 2.5 0.5 1 1.5 2 2.5
viations ys , yd , us , ud with the index s corresponding to κ=Ks/K
m
κ=Ks/K
m
the setpoint and d corresponding to the disturbance response.
Thereby, we will denote as MO disturbance responses those Figure 2. The closed loop performance portrait for n = 1 and an uncertainty
characterized by T V1 (yd ) → 0. line segment (ULS) corresponding to an uncertain gain ratio (14), K0 = 1,
When wishing to define an optimal control as something Ksmax /Km = Ksmax KP ≤ 2.5
unique, it should be characterized in the simples possible way,
as for example
In situations, when the setpoint and disturbance responses
 = ys = yd = us = ud → 0 (6) do not change their sign, their IAE values are given as
But, in order to respect the always limited precision of control 1 − KP Km nTn nTn
IAEs = ; IAEd = (12)
and also that of computer simulations, some ”sufficiently” KP Km KP
small Let us introduce a required loop gain K0 that is based on
 = ys = yd = us = ud = 0.001 (7) the known model gain Km and determining the speed of the
setpoint responses in a loop without filters as
will be chosen instead. As an optimization result, optimal
parameter KP of the P control algorithm with a disturbance K0 = KP Km (13)
feedforward
ur = KP (r − y) − di (8) The ratio of the plant and model gains (that describes the only
model uncertainty) let us denote as
and an appropriate feedback filter time constant Tn (Fig. 1) in
κ = Ks /Km ∈ K = [κmin , κmax ] (14)
1
Qn (s) = (9) Then, the loop transfer functions become
(1 + Tn s)n
will be determined yielding for an uncertainty in the plant gain K0 κ(1 + Tn s)n
Fs (s) = (15)
Ks and for a chosen filter order n transients with required s[(1 + Tn s)n − 1] + κ(s + K0 )
IAE values satisfying the shape related constraints (5). and
Ks [(1 + Tn s)n − 1]
III. ROBUSTNESS ANALYSIS BY THE PPM Fd (s) = (16)
s[(1 + Tn s)n − 1] + κ(s + K0 )
When a model gain Km does not precisely match the plant
When choosing a controller gain KP and a model gain Km
gain Ks , one has to tune accordingly the controller gain KP
(i.e. the required speed of transients specified by (12)), then
and also the filter Qn determined by its order n and the time
it is possible to look for some optimal n and Tn guaranteeing
constant Tn . The closed loop transfer functions are
for a given model uncertainty (14) the fastest disturbance
Y (s) KP Ks Km (1 + Tn s)n responses with tolerable shape deviations. Next, we will try
Fs (s) = =
R(s) Km s[(1 + Tn s)n − 1] + Ks (s + KP Km ) to show how to solve this step by the performance portrait
(10) method.
and
Y (s) Ks Km [(1 + Tn s)n − 1] A. Basic design limitations
Fd (s) = = With respect to the always present nonmodelled dynamics,
Di (s) Km s[(1 + Tn s)n − 1] + Ks (s + KP Km )
(11) the value (13), as well as the product KP Ks ≤ KP Ksmax
From Fd (0) = 0 and Fs (0) = 1 it follows that the loop must be kept limited. For example, by approximating the
guarantees zero steady state error also in the case of admissible nonmodelled dynamics by a dead time Td [14], for a nom-
constant input disturbances di = const. inal tuning the gain recommended for output monotonic and
TV (u ), n=2 TV (u ) TV (u ), n=4 TV (u )
1 s 1 d 1 s 1 d
0.01
0.001

1
0.3 00..000.1 0.3 0.15 0.15

1
011

0.1
Tn

Tn
1 0.1 0.1
0.2 0.111 0.2 0. .011

0.1
1
0. 1 1

1
0.001

0.001
0.1 00
00..0 0.1 00
0.0 0.05 0.10011 0.05 .0 0

1
00.0

0.0

0.0
0.00
1 0.
0.1
0.5 1 1.5 2 2.5 0.5 1 1.5 2 2.5 0.5 1 1.5 2 2.5 0.5 1 1.5 2 2.5
TV0(ys) TV1(yd) TV0(ys) TV1(yd)

0.1
0.0
0.0 1 0.001.001 1
0.3 01 0.3 0.15 0.15 0.1
1
00.00.0.1

1
0

1
1

0.1 .01 0.0

1
Tn

Tn
0.2 0.2 0.1 0.1
00.111
1
0. 01 0..01011 011 0

0.0
0.1 .0
00.0 0.1 10
0.00.
0.05 00.0 0.05

1
0.001
0.5 1 1.5 2 2.5 0.5 1 1.5 2 2.5 0.5 1 1.5 2 2.5 0.5 1 1.5 2 2.5
IAE(ys) IAE(y ) IAE(ys) IAE(y )
d d
0.5 0.5
0.
6 0.4 0.4

6
0.3 0.3 0.15 0.15

0.
0.7

0.5

0.7
0.2 0.2
Tn

Tn
1.6

6
1. 1.25

1. 1.25
0.2 0.6 0.2 8 0.1 0.6 0.1 8

1.
0.81
0. 0.4
0.81

0.7 0. 0.7

0
0

0.1 0.8 0.1 0.1 0.05 0.8 0.05 0.1


0.9 0.9 01.2 0.05 0.9 0.9 01.2 0.05
0.
0.05 0.01 0.
0.05
0.01
0.5 1 1.5 2 2.5 0.5 1 1.5 2 2.5 0.5 1 1.5 2 2.5 0.5 1 1.5 2 2.5
κ=Ks/K κ=Ks/K κ=Ks/K κ=Ks/K
m m m m

Figure 3. The closed loop performance portrait for n = 2 and an uncertainty Figure 5. The closed loop performance portrait for n = 4 and an uncertainty
line segment (ULS) corresponding to an uncertain gain ratio (14), K0 = 1, line segment (ULS) corresponding to an uncertain gain ratio (14), K0 = 1,
Ksmax /Km = Ksmax KP ≤ 2.5 Ksmax /Km = Ksmax KP ≤ 2.5

TV1(us), n=3 TV1(ud)


TV1(us), n=5 TV1(ud)
00.. 0.0
0001 01 1
0.01

1
0.1

0.15 1 0.15 0.1


1

00..1

00.0
01

0.

1
0.1 0.1
Tn

0.1 0.1

0101
00
Tn

1
1

0. ..00101 11
00..000
0.

1
0.05 0 0.05 0.05
0.1 11
0.05 0.11
0.00 00..0001
0.5 1 1.5 2 0.5 1 1.5 2
0.5 1 1.5 0.5 1 1.5
TV0(ys) TV1(yd)
TV0(ys) TV1(yd)
0.001
0.15 0.15 0.1
0.15 0.15
00.00.00.111 00.0.01
0.1 0.1 01
Tn

0.1 0.1
1
1

Tn

1 1

1
11
1
1

00.0.0.010 00..000 0.11


0.

0.05 0.05 0.05 00.111 0.05 .0 1


00.0. 0 00.00
0.5 1 1.5 2 0.5 1 1.5 2
0.5 1 1.5 0.5 1 1.5
IAE(ys) IAE(yd)
IAE(ys) IAE(yd)
0.5 0.15 0.15
0.7

0.15 0.15 7 0.4


0.6 0.2 0.
0.8

0.1 0.5 0.1


Tn

0.1 0.1
01. .6

0.6
Tn
1.1.5

8
2

0.8
01

0.4.2 0.1 0.2


0.

6
0.7
0.9

0.9 .5

0.05 0.05 01 0.08.4


12

0.9 0.05 0.05 0.05 1.


00.1
0.0 5 0.01 1. 0.8
0.9 0.1
0.2
0.
0.051 0.05
0.01
0.5 1 1.5 2 0.5 1 1.5 2
κ=Ks/Km κ=Ks/Km 0.5 1 1.5 0.5 1 1.5
κ=Ks/Km κ=Ks/Km

Figure 4. The closed loop performance portrait for n = 3 and an uncertainty


Figure 6. The closed loop performance portrait for n = 5 and an uncertainty
line segment (ULS) corresponding to an uncertain gain ratio (14), K0 = 1,
line segment (ULS) corresponding to an uncertain gain ratio (14), K0 = 1,
Ksmax /Km = Ksmax KP ≤ 2.5
Ksmax /Km = Ksmax KP ≤ 2.5

input 1P responses should be chosen much less than K0 =


KP Km < exp(−1)/Td . For the sake of the simplicity, a items as IAE(ȳs ), IAE(ȳd ), T V 0 (ȳs ), T V 1 (ȳd ), T V 0 (ūs )
precise analysis of the nonmodelled dynamics impact will be and T V 0 (ūd ) that are generated for the considered plant (2)
omitted here. But, in applying the presented tuning method with Ks = 1 over chosen grid of points Km , n, Tn from
one has to keep Td << Tn . The second important limitation the setpoint responses (r = 1, di = 0) with output and
follows from the usually discrete time implementation of input ȳs (τ ), ūs (τ ) and from the input disturbance responses
the considered loop, when a sampling period Ts should be (r = 0, di = 1) with ȳd (τ ), ūd (τ ). These are stored and
negligible with respect to the filter time constants [16], i.e. it expressed for different n and a chosen KP over grid of points
should hold
Tn ≥ Tnmin >> Ts (17) κ = Ks /Km ∈ KP P = [κP P min , κP P max ]
(18)
Tn ∈ TP P = [TnP P min , TnP P max ]
B. Performance portrait of the considered loop
Similarly as in the Theorem 2 in [15] one has firstly to It may be shown that the loop performance measures corre-
establish a relation of the normed PP items corresponding to sponding to any plant parameters Ks and controller parameters
(14)-(16) with variables of a real system. The PP includes Km , Tn belonging to the range (18) with the corresponding
0.16 0.14
n=1 n=1
0.14 n=2 0.12 n=2
n=3 n=3
n=4 n=4
0.12 0.1
n=5 n=5

0.1 0.08
∆IAEΣ

∆IAEΣ
0.08 0.06

0.06 0.04

0.04 0.02

0.02 0

0 −0.02
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
∆ =(K −K )/K ∆ =(K −K )/K
K smax smin smin K smax smin smin

Figure 7. Relative robustness characteristics for K0 = 1 and an additional Figure 9. Relative robustness characteristics for K0 = 1 and an additional
constraint Ksmax /Km = Ksmax KP ≤ 2.5 constraint Ksmax /Km = Ksmax KP ≤ 2.0

1 0.95
n=1 n=1
0.95 n=2 n=2
n=3 n=3
n=4 0.9
n=4
0.9
n=5 n=5

0.85
0.85
IAEΣ

IAEΣ

0.8

0.8
0.75

0.7
0.75
0.65

0.7
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
∆K=(Ksmax−Ksmin)/Ksmin ∆K=(Ksmax−Ksmin)/Ksmin

Figure 8. Absolute robustness characteristics for K0 = 1 and an additional Figure 10. Abolute robustness characteristics for K0 = 1 and an additional
constraint Ksmax /Km = Ksmax KP ≤ 2.5 constraint Ksmax /Km = Ksmax KP ≤ 2.0

responses ys (t), us (t), yd (t) and ud (t) may be calculated as Ksmax KP ≤ 2.5. For n > 1 the tuning found are already
limited only by the shape related constraints: for n = 2 and
IAE(ys ) = IAE(ȳs ) ; IAE(yd ) = Ks IAE(ȳd )
n = 3 by the T V0 (us ), for n = 4 and n = 5 by T V1 (ud ).
T V0 (ys ) = T V 0 (ȳs ) ; T V1 (yd ) = Ks T V 1 (ȳd ) (19)
The gain uncertainty degree in the considered robustness
T V0 (us ) = K1s T V 0 (ūs ) ; T V1 (ud ) = T V 1 (ūd )
characteristics will be characterized by
The limits for calculations should be chosen in such a way to
guarantee K ⊂ KP P and the condition (17). ∆K = (Ksmax − Ksmin )/Ksmin ≥ 0 (20)

C. Optimal robust controller design The nominal performance will be characterized by the cost
function (12) coresponding to he nominal case with ∆K = 0
Examples of a closed loop performance portrait correspond- denoted as IAEΣ (0).
ing to K0 = 1 and different n are in Figs 2-6. Sets of all pos- For a considered n, an absolute robustness characteristic
sible working points corresponding to different uncertainties represents dependance of a cost function (4) on a given
(14) and a found Tn form uncertainty line segments (ULS) uncertainty ∆K , whereas a relative robustness characteristics
with a length increasing with the increasing uncertainty (14). express dependance of a cost function IAEΣ (∆K )
For n = 1 the tuning found respect the shape related
requirements on T V1 (us )) and T V0 (ys ) and the requirement ∆IAEΣ (∆K ) = IAEΣ (∆K ) − IAEΣ (0) (21)
TV (u ), n=1 TV (u ) 0.12
1 s 1 d

0.8 0.1 0.8 1


1 0.00
0.6 00.0.001 0.6
0.1
Tn

0.4 0.1 0.4

1
1

0.
.011 0.0 1
0.2 00.0 0.2 0.00
0.5 1 1.5 2 2.5 0.5 1 1.5 2 2.5 0.08
TV0(ys) TV1(yd)

0.8 0.8 1 0.06


0.1 0.0 .001

∆IAEΣ
0

1
0.6 1 0.6
00.0.001
1
Tn

0.4 0.4 1
00.11 0. 0.04
0.2 0.2 0.001.001
00.0.
0.5 1 1.5 2 2.5 0.5 1 1.5 2 2.5
IAE(ys) IAE(y ) 0.02 n=1
d
n=2
0.8 0.6 0.8
0.4 n=3
0.6 0.6 0
7

0.5
0.
Tn

0.2 n=4
6
2
1. 1.5

0.4 0.6 0.4 8


1.
01

0.
8

0.7
0.

0.2 0. 0.8 0.2 0.4 .2 0.1 n=5


9 0.9 01
0. 0.05
0.05 0.01 −0.02
0.5 1 1.5 2 2.5 0.5 1 1.5 2 2.5 0 0.2 0.4 0.6 0.8 1
κ=Ks/K κ=Ks/K ∆ =(K −K )/K
m m K smax smin smin

Figure 11. The closed loop performance portrait for n = 1 and an uncertainty Figure 13. Relative robustness characteristics for K0 = 1 and an additional
line segment (ULS) corresponding to an uncertain gain ratio (14), K0 = 1, constraint Ksmax /Km = Ksmax KP ≤ 1.5
Ksmax /Km = Ksmax KP ≤ 2

0.94
TV1(us), n=2 TV1(ud)

0.01
0.001
0.3 00..000.1 0.3 0.92
011
Tn

1
0.2 0.111 0.2 0. .011
00 00
0.1 00..0 0.1 0.0 1 0.9
0.1 0.00
0.5 1 1.5 2 2.5 0.5 1 1.5 2 2.5
TV0(ys) TV1(yd) n=1
0.88 n=2
IAEΣ

0.1
0.0
0.0 1 0.001.001 n=3
0.3 01 0.3
1
1

n=4
Tn

0.2 0.2 1 0.86


00.111 0. 01 n=5
0.1 .0 0.1 10
00.0 0.00.
0.5 1 1.5 2 2.5 0.5 1 1.5 2 2.5 0.84
IAE(ys) IAE(yd)

0.
6 0.4 0.82
0.3 0.3
0.7

0.5 0.2
Tn

1.6
1. 1.25

0.2 0.6 0.2 8


0.81

0.7 0.
0

0.1 0.8 0.1 0.1 0.8


0.9 0.9 01.2 0.05 0 0.2 0.4 0.6 0.8 1
0.
0.05 0.01
0.5 1 1.5 2 2.5 0.5 1 1.5 2 2.5 ∆K=(Ksmax−Ksmin)/Ksmin
κ=Ks/Km κ=Ks/Km

Figure 14. Abolute robustness characteristics for K0 = 1 and an additional


Figure 12. The closed loop performance portrait for n = 1 and an uncertainty constraint Ksmax /Km = Ksmax KP ≤ 1.5
line segment (ULS) corresponding to an uncertain gain ratio (14), K0 = 1,
Ksmax /Km = Ksmax KP ≤ 2

Ksmax KP ≤ 1.5 would yet more stress advantages of higher


Fig. 7 shows that for Ksmax KP ≤ 2.5 the strongest order filters expressed in terms of the absolute robustness
performance decrease (IAE increase) corresponds to n = 1. (Fig. 14), whereas the relative robustness characteristics show
Advantage of using a higher order disturbance observer filter the lowest increase for n = 1 (Figs 13).
than required by implementation of the plant model inversion Examples of the output step responses in Fig. 16 show that
is documented also by the absolute characteristics course in for an idealized plant they keep the shape related require-
Fig. 8 (for n = 2 and ∆K > 0.45). ments both for Ksmax , as well as Ksmin . However, for a
This fact may become more evident by narrowing limits relatively significant nonmodelled dynamics characterized by
imposed on the controller gain, when e.g. for Ksmax KP ≤ 2 Td = Tn /5 the corresponding transients may already show
one gets the absolute robustness characteristics in Figs 9-10. shapes that already do not fulfill the requirements.
Now, a use of higher order filter may give absolutely better
IV. C ONCLUSIONS
performance not only for n = 2, but also for n = 3, ∆K >
0.3, which may be easily explained by checking the PPs, where The carried out robustness analysis has brought several
the constraint Ksmax KP ≤ 2 has an impact just for n = 1 new results. Firstly, it demonstrated that the notion of ro-
and n = 2. bustness needs to be rigorously defined and used carefully.
A further decrease of the limit product Ksmax KP to In order to avoid misinterpretations, one has to distinguish
0.7
order filters.
n=1 It is once more to be stressed that all possible advantages
0.6
n=2 of the higher order filters are conditioned by a possibility
n=3
n=4 to work with sufficiently short sampling periods, since by
0.5 n=5 increasing the filter order an equivalent loop behavior may
only be guaranteed by decreasing the filter time constants Tn
0.4 (Fig. 15), which increases demands on the control implemen-
Tn

tation. The design may also be extended by consideration of


0.3
a nonmodelled dynamics. It is also interesting to note that the
carried out analysis shows for this DO based controller for the
0.2
first order integrator much more complex results than in the
0.1
case of the simpless dead time compensator [17].
ACKNOWLEDGMENT
0
0 0.2 0.4 0.6 0.8 1 This work has been supported by grants APVV-0343-12 and
∆ =(K −K )/K
K smax smin smin
VEGA 1/0937/14.
Figure 15. Equivalent filter time constants for K0 = 1 and an additional R EFERENCES
constraint Ksmax /Km = Ksmax KP ≤ 2.5
[1] K. Ohishi, “A new servo method in mechantronics,” Trans. Jpn. Soc.
Elect. Eng., vol. 107-D, pp. 83–86, 1987.
[2] K. Ohishi, M. Nakao, K. Ohnishi, and K. Miyachi, “Microprocessor-
1 controlled dc motor for load-insensitive position servo system,” IEEE
0.9
Trans. Industrial Electronics,, vol. IE-34, no. 1, pp. 44 –49, feb. 1987.
[3] E. Sariyildiz and K. Ohnishi, “Bandwidth constraints of disturbance
0.8 observer in the presence of real parametric uncertainties,” European
Journal of Control, vol. 19, no. 3, pp. 199 – 205, 2013.
0.7 [4] E. Sariyildiz and K. Ohnishi, “Design constraints of disturbance observer
SR: Ks=Ksmax in the presence of time delay,” in Mechatronics (ICM), 2013 IEEE
0.6
International Conference on, Feb 2013, pp. 69–74.
−−−> y

SR: Ks=Ksmin
0.5 [5] E. Sariyildiz and K. Ohnishi, “Analysis the robustness of control systems
DR: Ks=Ksmax based on disturbance observer,” International Journal of Control, vol. 86,
0.4 DR: Ks=Ksmin no. 10, pp. 1733–1743, 2013.
[6] E. Sariyildiz and K. Ohnishi, “A guide to design disturbance observer,”
0.3 J. Dyn. Sys., Meas., Control, vol. 136, no. 2, 2013.
0.2
[7] E. Sariyildiz and K. Ohnishi, “Performance and Robustness Trade-off in
Disturbance Observer Design,” in 39th Annual Conference of the IEEE
0.1 Industrial Electronics Society (IECON). Vienna, Austria: IEEE, 2013,
pp. 3679–3684.
0 [8] M. Huba, “Open and flexible P-controller design,” in Int. Conf.
0 0.5 1 1.5 2 2.5 3
−−−> t Advances in Motion Control. Sarajevo, BIH: IEEE, 2012.
[9] M. Huba, “Modular disturbance observer based constrained PI ontroller
design,” in Int. Conf. Advances in Motion Control. Sarajevo, BIH:
Figure 16. Setpoint (SR) and disturbance step responses (DR) for K0 = 1, IEEE, 2012.
Ksmax = 1, Ksmin = Ksmax /3 and n = 2 designed for  = 0.001 for [10] M. Huba and I. Bélai, “Noise attenuation motivated controller design.
an idealized loop (full curves) that are applied for a loop with an additional Part I: Speed control,” in Speedam Symposium, Ischia, Italy, 2014, pp.
dead time Td = Tn /5 (dotted) 1325–1330.
[11] M. Huba, I. Bélai, and P. Bisták, “Noise attenuation motivated controller
design. Part II: Position control,” in Speedam Symposium, Ischia, Italy,
2014, pp. 1331–1336.
between relative and absolute robustness characteristics and [12] M. Huba and I. Bélai, “Experimental evaluation of a DO-FPID controller
to take ito account that an increase of a relative robustness with different filtering properties,” in IFAC World Congress, Cape Town,
South Africa, 2014.
may be accompanied by an absolute robustness decrease, [13] M. Huba and I. Bélai, “Comparison of Two Approaches to a Positional
and vice/versa. Thereby, the relative robustness characteristics Servo Control,” in 15th Int. Carpathian Control Conference - ICCC,
express a performance decrease (in our case ∆IAEΣ (∆K ) Velké Karlovice, Czech Republic, 2014.
[14] M. Huba, “Comparing 2DOF PI and Predictive Disturbance Observer
increase) achieved under the shape-constraints and related to Based Filtered PI Control,” Journal of Process Control, vol. 23, 10, pp.
the nominal case on the considered uncertainty measure ∆K , 1379–1400, 2013.
whereas the absolute robustness characteristics corespond to [15] M. Huba, “Performance measures, performance limits and optimal PI
control for the IPDT plant,” Journal of Process Control, vol. 23, 4, pp.
an IAEΣ (∆K ) course in dependance on ∆K ≥ 0. 500–515, 2013.
Secondly, there are not possible simple straightforward [16] V. R. Segovia, T. Hägglund, and K. Aström, “Measurement noise
recommendations regarding an optimal filter order choise. By filtering for PID controllers,” Journal of Process Control, vol. 24, no. 4,
pp. 299 – 313, 2014.
restricting the admissible loop gains, use of higher order filters [17] M. Huba, “Robust Tuning of the Simplest Dead Time Compensators,”
may guarantee not just a relatively, but also an absolutely in 8th Int. Conf. and Exposition on Electrical and Power Eng. - EPE,
superior loop robustness. This fact is then yet more stressed Iasi, Romania, 2014.
by a more effective noise filtration offered by use of higher

You might also like