You are on page 1of 6

Robust Tuning of the Simplest Dead Time

Compensators

Mikuláš Huba
Slovak University of Technology in Bratislava, Slovakia
Email: mikulas.huba@stuba.sk

Abstract—This paper rediscovers and modifies the oldest and with a gain estimation error in the range 10-20% of the nominal
the simplest dead-time compensator (DTC) introduced originally value is shown. So, in total, the DTC performance is used to be
by [1] and shows new possibility for its robust tuning by the demonstrated in 3-6 different working points. This is, however,
performance portrait method. This method comprising computer still far from being enough for fully characterizing such a
based analysis, compressed characterization and visualization complex closed loop performance, as the DTC have, from a
of the closed loop properties achieved by evaluating set of
performance measures over chosen grid of loop parameters with
potential user view. In this paper the loop performance will
an optimization based on sweeping all possible stored data may be characterized by information extracted from hundreds, or
be successfully used for an optimal nominal or robust tuning thausends of experiments each one carried out for a different
[2]–[4] based either on the known plant model with parameters working point. The interesting shape and time related closed
taking values upon uncertainty intervals, or directly on measured loop properties will be mapped, stored, visualized and used
experimental data [5]. In order to enable an easy comparing with for the robust control design based on a compressed form
other possible solutions new robustness characteristics will be of information representation called here as the closed-loop
derived showing how a model uncertainty influences decrease of performance portrait (PP). The PP method (PPM) is then used
the loop performance. for comparing and optimizing properties of different DTCs.

I. I NTRODUCTION When considering tuning rules appropriate for education


& practice, according to [9] they should be 1. well motivated,
Tuning of a closed loop systems involving dead-time still 2. preferably model-based, 3. analytically derived, 4. simple
represents a challenging domain of control research. Thereby, and easy to memorize and 5. work well on a wide range
the importance of dead-time systems that are being used to of processes. Thereby, a controller tuning is usually based
describe transport of mass, energy and information and to on trade-off between speed of response, good disturbance
approximate accumulation of time lags in a chain of low order rejection, stability, robustness, less input usage, or less sen-
systems is permanently increasing, to mention just different sitivity to measurement noise [9]. Despite that one may agree
new applications arising in the field of remote control via with all these requirements, it is interesting to note that the
computer networks and telecommunication links. As it was up now most frequently used and cited tuning method by
shown in many contributions (see e.g. [6]), in the loops Ziegler and Nichols [10] does not fulfill at least the 3rd
with PID controllers an increase of the dead-time values requirement in previous list. Of course, their method is not
with respect to the dominant plant time constant leads to a universally applicable, but the main reason for mentioning it
rapid performance deterioration. In robust design controllers is that we may learn from it. It was proposed to generalize
proposed to eliminate dead-time (denoted frequently as the results of numerous experiments on a sample of representative
dead-time compensators, DTCs) are usually proposed for a processes under chosen shape-related performance measure
nominal plant representing whole family of plants defined by (quarter amplitude damping). Such an approach can today
a norm-bounded multiplicative uncertainty error. The required be easily performed by using computers. Thereby, one can
performance is then tuned in the frequency domain with the easily extend spectrum of different qualitative & quantitative
aim to keep the modeling error within prescribed limits also for measures chosen on demand and in different combinations
the plant with the largest deviation from the nominal operating by engineer carrying out a control design requiring particular
point. In carrying out such a control design one may e.g. specifications.
use the phase, or the gain margin, the maximum sensitivity
[7], or the dead-time margin [8]. However, when aiming to This new computer analysis of control loops based on
design control loops and having higher requirements on the generating the closed loop PP [2], [3], [11]) is briefly described
resulting control quality that should be respected by all plants to be used in analyzing, designing and comparing DTCs and
of a given family, the frequency domain approach shows to be in visualizing their properties. The key question is, how to
not sufficiently effective [2], [3]. Instead of this, the shape and robustly design control loops with stable dead-time dominated
time related performance measures based upon closed loop plants to keep their setpoint and disturbance step responses
step responses will now be used. As it is well known, it is nearly monotonic at the the plant input and output also in
not enough to demonstrate the quality of DTCs just by the the case of possible plant-model mismatch. To cope with this
responses measured in nominal case. Their robustness [6] is problem, it is firstly necessary to introduce measures for evalu-
then usually shown for the dead-time uncertainty in the range ating deviations of real responses from ideal monotonic (MO)
of 10-20% of its nominal value. In some papers, also the case shapes and then to propose algorithms processing achieved
1P One-Pulse, with one extreme point
2D Two Dimensional or in many different ways by achieved IAE, or TV values,
DO Disturbance Observer by the maximal overshooting, damping ratio, etc. When these
FSP Filtered Smith Predictor measures increase over some value chosen e.g. as a multiple
I Integral
IAE Integral of Absolute Error of optimal value, transients may be denoted as unstable.
MO Monotonic, Monotonicity
PDO Predictive Disturbance Observer %labelsec:DisturbanceResponse In evaluating disturbance
PI Proportional Integrative step responses one can test the same properties as for the
PP Performance Portrait
PPM Performance Portrait Method setpoint response, but to note that:
PrI Predictive Integral Controller
SP Smith Predictor 1) Immediately after a disturbance step change the plant
TV Total Variance output starts to rise (fall) and controller needs some time to
TV0 Integral deviation from MO
TV1 Integral deviation from 1P
reconstruct a new disturbance value, to balance its effect and
to reverse output to move back to the reference value. So, the
TABLE I. BASIC ACRONYMS USED IN THE PAPER evaluation of MO output increase (decrease) may start just after
its turnover characterized by an extreme point ym . Deviations
from such a response will be quantified by means of
information. X
T V1 (y) = |yi+1 − yi | − |2ym − y∞ − y0 | (4)
The remaining paper is structured as follows. Chapter 2 i
recalls basic performance measures for evaluation of nearly Thereby, y0 and y∞ represent initial and final input values and
monotonic setpoint step responses. In Chapter 3 the evaluation ym ∈ / (y0 , y∞ ) is an extreme output value. Though, to keep
of disturbance step responses is briefly mentioned. Chapter 4 the discussion simple, we will speak about MO response, even
deals with predictive integral controller design for the simplest it actually is a one-pulse (1P) response.
stable dead time dominant plant. In Chapter 5 history of the
predictive I controllers is briefly discussed. Chapter 6 shows 2) By a prefilter in the reference signal it is possible to
an application example with a higher-order plant and Chapter slow down dynamics of the setpoint step responses and thus to
7 brings finally short conclusions to the paper. avoid an overshooting. In this way it is possible to get tighter
disturbance rejection by simultaneously having MO setpoint
II. P ERFORMANCE MEASURES FOR EVALUATING NEARLY step response.
MONOTONIC RESPONSES In general, MO areas for disturbance response are different
A. Stable (ST) control and indication of instability from those corresponding to setpoint response and the con-
troller design has to compromise these differences.
In the time domain the property of stability (ST) and
monotonicity (MO) may be easily tested numerically by eval- III. PREDICTIVE (DEAD TIME) I0 -CONTROLLER
uating simulated or experimentally measured transients, in our (P rI0 )
case the setpoint step responses. The Bounded-Input-Bounded-
Output (BIBO) and Internal Model Control (IMC) stability Predictive I0 controller may be considered as a gener-
require for a bounded plant input u(t) a bounded plant output alization of the first known predictive (dead-time) controller
y(t) that for a given limits Ymax , Umax introduced by Reswick [1] yet before the well known Smith
predictor [12]. Due to its high sensitivity it has been never
|y(t)| < Ymax < ∞; |u(t)| < Umax < ∞; ∀t > 0 (1) broadly used. However, after its extension by low-pass filters
can be tested numerically. Although in this paper we are known from the much later introduced disturbance observer
dealing with constrained continuous signals, we will consider [13], one gets a robust and universal controller that may be
their values sampled with a sampling frequency sufficiently considered as a counterpart to the well known and broadly used
high enabling to preserve all their important features. PI controller. Thus, despite the seemingly simple structure, a
broader exploitation of this new solution has been enabled just
A stable setpoint step responses running to the reference by proposing a new performance portrait method (PPM) for
value r may further be characterized by performance indices its reliable tuning [14].
such as an IAE (Integral of Absolute Error)
Z ∞ Consider a stepwise constant reference signal r(t) and a
stable uncertain plant with dominant dead-time Td approxi-
IAE = |e(t) − e∞ | dt ; e = r − y (2)
0 mated by a model
or a relative Total Variance measure proposed by modification S(s) = Ke−Td s
(5)
of Total Variance [9] as T V0 (y) ( [2]) K ∈ hKmin , Kmax i ; Td ∈ hTdmin , Tdmax i
X
T V0 (y) = |yi+1 − yi | − |y∞ − y0 | (3) with parameters defined e.g. as
i Kmin = 1/2 ; Kmax = 1 ; Tdmin = 1/2 ; Tdmax = 1 (6)
Here, y0 and y∞ represent the initial and the final output
values. The task is to design a robust controller according to Fig-
ure 1 by choosing an operating point K0 , Td0 , a DO filter
Stability, or more precisely instability degrees can be characterized by n and Tn and a prefilter given by m and Tp
indicated by values of the parameters Ymax and Umax in (1), that would offer the fastest possible dynamics of the setpoint
and the input disturbance-to-output y1 closed loop transfer
function is
KKm e−Td s (Tn s + 1)n − e−Tm s

Fi1 (s) = (12)
Km ((Tn s + 1)n − e−Tm s ) + Ke−Td s
Since Fw1 (0) = 1 and Fdi1 (0) = 0, this scheme guarantees
rejection of piece-wise constant disturbances also for K0 6= K
and Tm 6= Td . These properties hold also for the output y0 ,
where
Fs0 (s) = Fw1 (s) eTd s ; Fi0 (s) = Fi1 (s) eTd s (13)
Figure 1. Predictive I controller of the dynamical class 0 (P rI0 ) designed as For the setpoint and disturbance step responses that do not
a filtered static feedforward control with an input disturbance reconstruction
and compensation; δ-measurement noise
change its sign the IAE values may be analytically calculated
as
IAEs1 = Km (Tm + nTn )/K ; IAEi1 = Km (Tm + nTn )
and disturbance responses and thereby guaranteeing the shape (14)
related output constraints Thus, for a given plant (5), to keep the IAE values as small
T V0 (ys ) ≤ ys ; T V1 (yd ) ≤ yd (7) as possible, one usually has to choose as small as possible
parameters Km , Tm , n and Tn . Thereby, it is to remember that
i.e. MO setpoint step responses of the output variable with a such an approach will be limited by:
tolerable output deviation from MO chosen ys , i.e. as ys ∗
100% of the reference value step r = 1 and with a tolerable • validity of the used plant model and
output deviation of the output disturbance responses from a 1P
• the limits imposed by the dynamics of the correspond-
response specified with yd , i.e. as yd ∗ 100% of the output
ing closed loop.
disturbance step di = 1. In this way, one may add also shape
related input constraints The first type of limitation requires that the filter time
T V0 (us ) ≤ us ; T V1 (ud ) ≤ ud (8) constants Tn influencing not only the measurement noise
attenuation, but also the closed loop robustness, should not
with some us and ud . Even when wishing to get responses be arbitrarily low, especially in situations with no considered
with ideal shapes, in order to respect the always limited dead time uncertainty. This is caused by the fact that several
precision of control and also that of computer simulations, possible loop time constants may be included (for the sake of
some ”sufficiently” small model simplicity) into the model dead time Td , which may
 = ys = yd = us = ud = 0.001 (9) become apparent especially in situations with Tm = Td .

will be chosen Such a task has been firstly treated for n = 1 A. Performance portrait method (PPM)
and Tp = Tn in [14]. Experience gained in dealing with DO
The second type of above mentioned limitations will be
based controllers for the first order plants gives motivation to
analyzed by the performance portrait method (PPM). To be
come back to this controller tuning with a question, how its
able to use the generated PP for any plant (5), the setpoint
closed loop robustness and noise attenuation depend on the
step responses will be mapped by using normed coordinates
order of the filter
1 K Tn Td
Qn (s) = (10) κ= ; τf = ; τd = ; p = Tm s (15)
(1 + Tn s)n Km Tm Tm
that may be used both in the disturbance reconstruction and which yields
as a prefilter in the setpoint filtration that prevents possible κe−τd p
Fs1 (p) = ((τn p+1)n −e−p )+κe−τd p
overshotings. (16)
Km κe−τd p ((τn p+1)n −e−p )
Fi1 (p) = κ((τn p+1)n −e−p )+κe−τd p
Definition 1 (Predictive I0 controller - P rI0 ): Under the
P rI0 controller we will understand a static feedforward control
with a gain 1/K0 extended by the input or output disturbance The robustness analysis by the PPM will be based on theo-
reconstruction by means of a ”predictive disturbance observer” rems similar to the Theorem 2 proven in [2] stating that when
(PDO) and its compensation at the static feedforward output considering PP including items IAE(ȳs ), IAE(ȳd ), T V 0 (ȳs ),
compensation (Figure 1) with the PDO filter time constant Tn T V 1 (ȳd ), T V 0 (ūs ) and T V 0 (ūd ) generated for the considered
and by the prefilter with the time constant Tp (in the simplest plant (5) with K = 1, Td = 1 over chosen grid of points
case with Tp = Tn ). Tm , n, Tn from the setpoint responses (r = 1, di = 0) with
output and input ȳs (τ ), ūs (τ ) and from the input disturbance
For Tp = Tn , the setpoint-to-output closed loop transfer responses (r = 0, di = 1) with ȳd (τ ), ūd (τ ) stored and
function of the PrI0 controller corresponding to the output y1 expressed over grid of normalized variables
and a plant-model parameter mismatch may be derived as
κ = K/Km ∈ [κP P min , κP P max ]
Ke−Td s τd = Td /Tm ∈ [τdP P min , τdP P max ] (17)
Fs1 (s) = (11)
Km ((Tn s + 1)n − e−Tm s ) + Ke−Td s τn = Tn /Tm ∈ [τnP P min , τnP P max ]
3 6
n=1
n=2 5.5
2.5 n=3
n=4
5
n=5
2
4.5
∆IAEΣ

IAEΣ
1.5 4

3.5
1
n=1
3
n=2
0.5 n=3
2.5 n=4
n=5
0 2
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
∆ =(T −T )/T ∆ =(T −T )/T
T dmax dmin dmin T dmax dmin dmin

Figure 2. Relative dead time robustness characteristics for  = 0.001 and Figure 3. Absolute dead time robustness characteristics for  = 0.001 and
different filter orders different filter orders

then, the loop performance measures corresponding to any trace out a line segment denoted as an Uncertainty Line
plant parameters K, Td and controller parameters Km , Tm , Tn Segment (ULS) that represents a special case of an uncertainty
belonging to the range (17) with the corresponding responses curve segment (UCS) [2]. It has been shown there that the aim
ys (t), us (t), yd (t) and ud (t) may be calculated according to of an optimal robust controller tuning satisfying given shape
IAE(ys ) = Td IAE(ȳs ) ; IAE(yd ) = KTd IAE(ȳd ) constraints may be interpreted as to optimally locate an UCS
(21) to lie in an area with admissible shape parameter values
T V0 (ys ) = T V 0 (ȳs ) ; T V1 (yd ) = K T V 1 (ȳd )
1 and, according to the Theorem 3 in [2], by locating an UCS
T V0 (us ) = K T V 0 (ūs ) ; T V1 (ud ) = T V 1 (ūd ) into areas with zero deviations from ideal shapes one gets
(18)
full performance invariance against dead time perturbations.
The aim is to keep for all possible values of an uncertain
B. Robustness characteristics for a dead time uncertainty parameter (19) and a chosen controller tuning the tolerable
In [14] it has been shown, how this controller may be tuned shape related deviations (3),(4) (9) yielding a minimal value
for n = 1 by the PPM by considering uncertainty of both of the cost function (2).
model parameters (6) in a 3D coordinate space. However, in
To simplify the evaluation of the performance changes, it
order to be able to evaluate impact of the filter order n on
will be characterized by the sum of the IAE values for the
the overall closed loop robustness, it is more appropriate to
setpoint and disturbance steps of the output y1 corresponding
derive new robustness characteristics expressing influence of a
to the dead time in the feedforward path
particular uncertain parameter (approximated by a fixed value
Km , or Tm ) on the loop performance. Thereby, we will require
to keep the shape related constraints (3) and (4) and to evaluate IAEΣ = IAEs + IAEd (22)
the corresponding increase of the IAE (2) values of the setpoint
and disturbance step responses. For such situations it is enough For an increasing ∆T (20), one gets by the PPM the deadtime
to generate 2D performance portraits corresponding either to robustness characteristics showing for particular filter orders n
κ = 1, or τd = const ≈ 1. the corresponding increase of the IAE (22) on ∆T (Figure 2).
It is to note that the minimal relative performance decrease
Thus, to characterize robustness aspects of the closed loop corresponds to n = 2. However, at the same time, the
control, the corresponding properties will now be checked by absolutely lowest performance decrease due to the dead time
the PPM [2], [3] for a single uncertain parameter. Firslty, the uncertainty corresponds to n = 1 (Figure 3). Thus, when
closed loop dead-time wishing to discuss about impact of the filter order on the
loop robustness, one has strictly define its meaning. Since
Td ∈ hTdmin , Tdmax i (19) the discussion of robustness have been mostly oriented on a
will be considered. Thereby, the dead time uncertainty may be robust stability [15], this aspect has been neglected up to now.
characterized by a relative uncertainty value In interpreting these characteristics by the PP (Figure 4) one
has to note that by increasing the dead time uncertainty (19)
∆T = (Tdmax − Tdmin )/Tdmin (20) the length of an uncertainty line segment (corresponding to
Then, in a normed 2D parameter subspace τd , τn , all possible all possible working points with the first coordinate Td /Tm
working points increases and thus, to be located in regions with the shape
related deviations below the chosen value (9) it must be shifted
κ = K/Km = 1 ; τd = Td /Tm ; τn = Tn /Tm (21) to higher Tn
TV0(us) TV1(ud) TV0(us) TV1(ud)
1 1
0.001
τn=Tn/Tm

τn=Tn/Tm
0.001 4 4
0.1 0.1
0. 1
0.00.1

1
0.5 1 0.5 1 0 1
0.00

0.00
0.1 0.101 2 0 2
10 10
1 0.1
1 1 1
0.6 0.8 1 0.6 0.8 1 0.5 1 1.5 2 0.5 1 1.5 2
TV0(ys) TV1(yd) TV0(ys) TV1(yd)
1 1
τn=Tn/Tm

τn=Tn/Tm
0.001 0.001 4 4
0.1 0.1
1 1
0.00.1 0.00.1
1

1
0.5 1 0.5 1
0.00

0.00
0.1 0.1 2 0 2 0
10 10
1 1 1 1
0.6 0.8 1 0.6 0.8 1 0.5 1 1.5 2 0.5 1 1.5 2
IAE(ys) IAE(yd) IAE(ys) IAE(yd)
1 2 1 2 3 3
τn=Tn/Tm

τn=Tn/Tm
4 4
4

4
3 3

8
0.5 0.5 6 2 6 2
4 1.5 4 1.5 2 4 3
2 4 3
65
18
30 65
18
30
23 23 2 1.5 234 2 1.5 234
0.6 0.8 1 0.6 0.8 1 0.5 1 1.5 2 0.5 1 1.5 2
τ =T /T τ =T /T κ =K/K κ =K/K
d d m d d m d m d m

Figure 4. The closed loop performance portrait for n = 1 and an uncertainty Figure 5. The closed loop performance portrait for n = 1 and an uncertainty
line segment (ULS) corresponding to an uncertain dead time (19), κ = 1 line segment (ULS) corresponding to an uncertain gain (23), τd = 1

τn≥τn,min=2
C. Robustness characteristics for a gain uncertainty 1.4
n=1
Let us now consider a plant gain uncertainty n=2
1.2
n=3
K ∈ hKmin , Kmax i (23) n=4
1 n=5
that may also be characterized by a relative value
∆K = (Ksmax − Ksmin )/Ksmin (24) 0.8
∆IAEΣ

Then, for a P rI0 controller with fixed parameters Km , Tm


and Tn acting on an plant with a fixed Td and an uncertain 0.6

gain K, in a normed 2D parameter subspace κ, τn , all possible


working points 0.4

κ = K/Km ; κ ∈ hκmin , κmax i


(25) 0.2
Td = Tm = const, ; Tn = const
trace out an Uncertainty Line Segment (ULS). The aim of an 0
0 0.2 0.4 0.6 0.8 1
optimal robust controller tuning satisfying the shape related ∆K=(Kmax−Kmin)/Kmin
constraints and minimal IAE values may be interpreted as to
optimally locate an ULS (25) to lie in an area with admissible
Figure 6. Relative gain robustness characteristics for  = 0.001 and
shape parameter values. Thereby, one has to remember that τnmin = 2
a model dead time Tm represents always a sum of smaller
delays (among them also time constants) that are dominated
by a dead time value. Due to this, an assumption Tm = Td , i.e. which should guarantee a constant decrease of the dynamics
τd = 1 leads to an unrealistic results that may be characterized corresponding to a constant increase of the term nTn in the
by Figure 5. In such a case, the ULS will always be located IAE values (14). Such a constraint frequently follows from the
at the lowest available τn P P min. During such a position at fact that such controllers are mostly implemented as quasi-
the lower PP border, up to the moment, when the whole ULS continuous, i.e. with a sampling period Ts that should be
fits into the PP area with required shape related deviations, by sufficiently short [16]. That means that one may only choose
increasing the uncertainty κ the IAE values do not change. filter orders n fulfilling the condition
However, since an ULS placement left from κ = 1 τnmin /n >> Ts (28)
may only be achieved by Km > 1, which leads to slower
responses (increased IAE values (14)), this trivial case of gain-
Already in the case of the relative gain robustness charac-
invariant, but slow transient responses may be eliminated by
teristics that are starting from the optimal nominal performance
an additional optimization constraint
values corresponding to ∆K = 0 one may see that the lowest
κmax > 1 (26) sensitivity to the gain uncertainty corresponds to n = 1
(Figure 6). From the absolute gain characteristics in Figure 7
A similar effect may be achieved by adding an additional
it is obvious that the lowest absolute IAEΣ values correspond
constraint on the filter time constants
again to n = 1. In general, the differences between particular
τn > τn,min /n (27) filter orders increase by increasing τn,min .
τ ≥τ =2
n n,min R EFERENCES
7.5
[1] J. B. Reswick, “Disturbance-response feedback - a new control con-
7 cept,” Trans. ASME, vol. 1, pp. 153–162, 1956.
[2] M. Huba, “Performance measures, performance limits and optimal PI
6.5 control for the IPDT plant,” Journal of Process Control, vol. 23, 4, pp.
500–515, 2013.
6 n=1
[3] M. Huba, “Comparing 2DOF PI and Predictive Disturbance Observer
n=2 Based Filtered PI Control,” Journal of Process Control, vol. 23, 10, pp.
IAEΣ

5.5 n=3 1379–1400, 2013.


n=4 [4] M. Huba and D. Soós, “Performance portrait - a 3D approach,” in IFAC
5 n=5 World Congress, Cape Town, South Africa, 2014.
[5] D. Sóos and M. Huba, “Experimental performance portrait based
4.5 optimal controller tuning,” in 15th Int. Carpathian Control Conference
- ICCC, Velké Karlovice, Czech Republic, 2014.
4 [6] J. Normey-Rico and E. Camacho, Control of dead-time processes.
London: Springer, 2007.
3.5 [7] D. E. Rivera, M. Morari, and S. Skogestad, “Internal model control.
0 0.2 0.4 0.6 0.8 1
∆ =(K −K )/K 4. pid controller design,” Ind Eng. Chem. Res., vol. 25,1, pp. 252–265,
K max min min
1986.
[8] A. Ingimundarson and T. Hägglund, “Performance comparison between
Figure 7. Absolute gain robustness characteristics for  = 0.001 and PID and dead-time compensating controllers,” Journal of Process
τnmin = 2 Control, vol. 12, no. 8, pp. 887 – 895, 2002.
[9] S. Skogestad, “Simple analytic rules for model reduction and PID
controller tuning,” Journal of Process Control, vol. 13, pp. 291–309,
IV. C ONCLUSIONS 2003.
[10] J. G. Ziegler and N. B. Nichols, “Optimum settings for automatic
The analyzed predictive I-controller represents the simplest controllers,” Trans. ASME, pp. 759–768, 1942.
dead time compensator for control of dead time systems that [11] M. Huba, “Designing robust controller tuning for dead time systems,”
may be interpreted as a static feedforward extended by a pre- in Int. Conf. System Structure and Control. Ancona, Italy: IFAC, 2010.
dictive (dead time) disturbance observer. It is well known that [12] O. Smith, “Closser control of loops with dead time,” Chemical Engi-
the model based controllers may hardly be tuned by traditional neering Progress, vol. 53, pp. 217–219, 1957.
intuitive approaches known from the PID control. Similarly as [13] K. Ohishi, “A new servo method in mechantronics,” Trans. Jpn. Soc.
a computer tomography, the performance portrait method gave Elect. Eng., vol. 107-D, pp. 83–86, 1987.
birth to a new tool enabling an easy analysis and design of such [14] M. Huba, “Basic Fundamental Controllers of DC0,” in Textbook:
”tricky” systems. To compare such solutions, it is now possible Selected topics on constrained and nonlinear control, M. Huba, S. Sko-
to derive their robustness characteristics showing impact of an gestad, M. Fikar, M. Hovd, T. Johansen, and B. Roháľ-Ilkiv, Eds. STU
Bratislava - NTNU Trondheim, 2011, pp. 139–186.
uncertain model parameter on the resulting loop performance
[15] E. Sariyildiz and K. Ohnishi, “Performance and Robustness Trade-off in
evaluated by the shape and time related performance measures Disturbance Observer Design,” in 39th Annual Conference of the IEEE
in the time domain. Although the carried out analysis of the Industrial Electronics Society (IECON). Vienna, Austria: IEEE, 2013,
influence of higher order DO filters essentially confirms earlier pp. 3679–3684.
results achieved in the frequency domain [15], [17]–[19], they [16] V. R. Segovia, T. Hägglund, and K. Aström, “Measurement noise
are fully focused on the robust performance, not only on the filtering for PID controllers,” Journal of Process Control, vol. 24, no. 4,
robust stability. Thereby, differences between the relative and pp. 299 – 313, 2014.
absolute robustness characteristics have been firstly noticed. [17] E. Sariyildiz and K. Ohnishi, “Design constraints of disturbance ob-
server in the presence of time delay,” in Mechatronics (ICM), 2013
The analysis also confirmed another known fact that a dead IEEE International Conference on, Feb 2013, pp. 69–74.
time uncertainty has a worse impact on the loop performance [18] E. Sariyildiz and K. Ohnishi, “Analysis the robustness of control sys-
than a gain uncertainty. Furthermore, the experience gained in tems based on disturbance observer,” International Journal of Control,
solving this formally simple problem may further be utilized vol. 86, no. 10, pp. 1733–1743, 2013.
in many not yet solved practical problems, as e.g. in [20], [21]. [19] E. Sariyildiz and K. Ohnishi, “A guide to design disturbance observer,”
J. Dyn. Sys., Meas., Control, vol. 136, no. 2, 2013.
ACKNOWLEDGMENT [20] M. Huba, “Robustness analysis of a disturbance-observer based PI
control,” in 8th Int. Conf. and Exposition on Electrical and Power Eng.
This work has been partially supported by the grants - EPE, Iasi, Romania, 2014.
APVV-0343-12 Computer aided robust nonlinear control de- [21] I. Bélai, P. Bisták, and M. Huba, “Optimal Tuning for a Double
sign and VEGA 1/0937/14 Advanced methods for nonlinear Integrator Plus Dead Time,” in 8th Int. Conf. and Exposition on
Electrical and Power Eng. - EPE, Iasi, Romania, 2014.
modeling and control of mechatronic systems.

You might also like