Robust Smith Predictor Design for Time-Delayed Systems
with H∞ Performance
Vinicius de Oliveira and Alireza Karimi
Abstract—A new method for robust fixed-order H∞ con-
troller design for uncertain time-delayed systems is presented.
It is shown that the H∞ robust performance condition can be
represented by a set of convex constraints with respect to the
parameters of a linearly parameterized primary controller in
the Smith predictor structure. Systems with uncertain dead-time
and with multimodel and frequency-domain uncertainty can be
considered straightforwardly in the proposed approach. Further-
more, it is shown that the design method can also be extended
to design of robust gain-scheduled dead-time compensators for
Linear Parameter Varying (LPV) systems. The performance of
the method is illustrated by simulation examples.
I. I NTRODUCTION
Most industrial processes present dead time in their dynam-
ics. Generally, dead times are caused by the time needed to
transport energy, mass or information, but they also can be
caused by processing time or by accumulation of time lags
in a sequence of simple dynamic systems interconnected in
series [1].
The presence of dead times in the control loops has two
main consequences: it greatly complicates the analysis and
the design of feedback controllers and it makes satisfactory
control performance more difficult to achieve [2]. Dead-
time compensators can be used to improve the closed-loop
performance of classical controllers (PI or PID controllers)
for processes with delay. The Smith Predictor (SP) (See Fig.1),
proposed in the late 1950s by Smith [3], was the first dead-
time compensation structure used to improve the performance
of the classical controllers and became the most known and
used algorithm to compensate dead time in the industry.
Although the SP offers potential improvement of the closed-
loop performance of process with large dead-time, it requires
a good model since small modeling errors can lead to very
poor performance. In fact, even if the SP is nominally stable,
it is possible to become unstable for infinitesimal change in
the process dynamics [4].
For this reason, research efforts have been focused on
robustness issues of the SP. A tuning method for models with
one uncertain parameter is proposed by [5]. In [6], tuning
guidelines are presented for set-point tracking considering
model mismatches in the dead-time. Easy tuning rules for SP
in the presence of dead-time uncertainty is addressed in [7] and
a guideline for selection the closed-loop bandwidth based on
the dead-time uncertainty bound is proposed. In [8], a robust
tuning rule is developed which considers the modeling error
in the dead time. Robust PID tuning for SP considering model
The authors are with Automatic Control Laboratory, Ecole Polytechnique
Fédérale de Lausanne (EPFL), Switzerland.
Corresponding author: alireza.karimi@epfl.ch
uncertainty is proposed by [9]. In particular, first and second
order plus dead-time systems which may contain uncertainty
in multiple parameters of the model are considered.
Recently, many researchers are interested in the optimal
control of dead-time systems, especially H ∞ control, i.e.,
to find a controller to internally stabilize the system and to
minimize the H∞ -norm of an associated transfer function.
Many relevant results have been presented in this framework
using modified versions of the SP. See, for instance, [10],[6]
and [11]. Nonetheless, as pointed out by [12], these controllers
are too involved and the predictor always includes additional
unstable hidden modes even for stable plants. As the author
argues, these hidden modes are not safe as they tend to desta-
bilize the system when implemented. Therefore, the practical
significance of these results are limited.
This paper presents a new method to design fixed-order
SP controllers that considers uncertainty simultaneously in the
dead-time and in the rational part of the model. The perfor-
mance specification, like the standard H ∞ control problem, is
a constraint on the infinity norm of the weighted sensitivity
function and is represented by a set of convex constraints in
the Nyquist diagram. A line search on the upper bound γ of the
infinity norm of the weighted sensitivity function can be used
in association to a convex feasibility problem to determine the
primary controller parameters. The proposed method can be
used for PID controllers, which are of great practical interest,
as well as for higher order linearly parametrized controllers
in discrete or continuous time. The main idea is introduced
for LTI-SISO systems and further extended to cope with LPV
systems.
It is important to point out that a method to design fixed-
order controllers for standard control loops based on convex
constraints in the Nyquist diagram has been firstly proposed in
[13]. This paper uses the same framework to design dead-time
compensators for time-delayed systems.
This paper is organized as follows: In Section II the class
of models, controllers and the control objectives are defined.
Section III introduces the control design methodology based
on convex constraints in the Nyquist diagram. In Section IV
the results are extended to unstable time-delayed systems.
Gain-scheduled SP is designed for LPV time-delayed systems
in Section V. Each section includes a simulation example to
show the effectiveness of the proposed approach. Finally the
concluding remarks are given.
 II. P ROBLEM FORMULATION r + u y
Ceq (s) P (s)
A. Class of models −

Consider the class of stable time-delayed LTI-SISO systems

with bounded infinity norm. It is assumed that the plant model
can be represented by: Fig. 2. Equivalent control diagram

P (s) = G(s)e−τ s (1)

C. Design specifications
where the time delay τ belongs to a known interval, i.e., τ ∈
[τmin τmax ] and the dead-time free part of the model has The SP can be represented by a classical feedback loop (see
unstructured multiplicative uncertainty described as: Fig. 2) where Ceq (s) is the equivalent controller given by

Ceq (s) = C(s)[1 + C(s)H(s)]−1 (6)

G(s) = Gn (s)[1 + ∆(s)W2 (s)] (2)
with H(s) = Gn (s) − Pn (s) = Gn (s)(1 − e−τn s ).
where W2 (s) is a known stable uncertainty filter, G n (s) the
Let the sensitivity function of the closed-loop system with
nominal dead-time free model and ∆(s) an unknown stable
the i-th plant model P i (s) be defined as
transfer function with ∆ ∞ < 1. The dead-time interval can
be converted to a finite set {τ 1 , τ2 , . . . , τq } using a fine grid. 1 1 + C(s)H(s)
Then we can assume that P (s) belongs to a set P of q models Si (s) = = (7)
1 + Pi (s)Ceq (s) 1 + C(s)[H(s) + Pi (s)]
given by:
and the complementary sensitivity function as
 −τi s
P = {Pi (s) = Gn (s)e ; i = 1, . . . , q} (3) Pi (s)Ceq (s) C(s)Pi (s)
Ti (s) = = (8)
1 + Pi (s)Ceq (s) 1 + C(s)[H(s) + Pi (s)]
B. Class of controllers
A standard robust control problem is to design a controller
The SP control structure shown in Fig.1 is considered. The
that satisfies W1 Si ∞ < 1 for a set of models where W 1 (s)
nominal model P n (s) = Gn (s)e−τn s with a fixed τn chosen
is the performance weighting filter. If the model is described
inside the interval [τmin τmax ] is used for the implementation
by unstructured multiplicative uncertainty, the necessary and
of the controller.
sufficient conditions for robust performance is given by [14]:
r + u y |W1 Si | + |W2 Ti |∞ < 1 for i = 1, . . . , q (9)
C(s) P (s)
−
The goal of the proposed approach is to design the primary
+ controller C(s) in the SP structure to guarantee robust perfor-
−
Gn (s) e−τn s mance of the closed loop.

+ III. P ROPOSED METHOD

yp
+ The robust performance condition (9) can be written as:

Fig. 1. Smith Predictor |W1 (jω)Si (jω)| + |W2 (jω)Ti (jω)| < 1, ∀ω (10)

and for i = 1, . . . , q. The dependency on frequency ω and s

The primary controller C(s) is linearly parametrized by will be omitted for brevity but the dependency on the controller
parameter ρ will be highlighted.
C(s) = ρT φ(s) (4)
A. Basic Idea
where ρT = [ρ1 , ρ2 , . . . , ρnc ] is an nc dimensional vector of
the controller parameters and Let Li (ρ) be defined as

φT (s) = [φ1 (s), φ2 (s), . . . , φnc (s)] (5) Li (ρ) = C(ρ)(H + Pi ) (11)

is a vector of basis functions with φ i (s) transfer functions with It should be mentioned that this transfer function is not equal
no RHP poles. For instance, a PID controller could be linearly to the open loop transfer function L eqi (ρ) = Pi Ceq (ρ) of the
parametrized by equivalent control loop in Fig. 2. The main properties of L i (ρ)
are:
1 s 1) it is linear with respect to the controller parameter ρ;
ρT = [Kp , Ki , Kd ] , φT (s) = [1, , ]
s 1 + Tf s 2) the closed loop poles are the roots of 1 + L i (ρ) = 0.
The first property is clear because C(ρ) is linearly parameter- that leads directly to (15). To prove that all closed-loop transfer
N
ized. To show the second property, let C(ρ) = D Nc
c
, Gn = Dgg functions are stable, consider (16) which gives:
be defined. The equivalent controller can be written as
Re{[1 + L∗d (jω)][1 + Li (jω, ρ)]} > 0 ∀ω (20)
Nc Dg
Ceq (ρ) = (12) or, alternatively,
Dc Dg + Nc Ng (1 − e−τn s )
Then, we have: wno{[1 + L∗d (jω)][1 + Li (jω, ρ)]} = 0 (21)

Dc Dg + Nc Ng (1 − e−τn s + e−τi s )
1 + Leqi (s) =
Dc Dg + Nc Ng (1 − e−τn s )
It is important to point out that the numerator of 1 + L eqi (s)
is a quasi-polynomial and its zeros are the closed-loop poles
of the system. On the other hand, we have
where wno stands for winding number around the origin. Since
(13) both L∗d (jω) and Li (jω, ρ) are constant or zero for the semi-
circle with infinity radius of the Nyquist contour the wno
depends only on the variation of s in the imaginary axis. Thus,
wno{[1 + Ld (jω)]} = wno{[1 + Li (jω, ρ)]} (22)

Dc Dg + Nc Ng (1 − e−τn s + e−τi s ) Since Ld (jω) satisfies the Nyquist criterion L i (jω, ρ) will do
1 + Li (ρ) = (14) so and all zeros of 1 + L i (jω, ρ) will be in the left-hand side
Dc Dg
of the complex plan. As the zeros of 1 + L i (jω, ρ) are the
It is clear that 1+Li (ρ) shares the same zeros with 1+L eqi (s). closed-loop poles, the system will be internally stable.
This property will be used to prove the closed loop stability. Remark I: The constraint in (16) is an inner convex
approximation of the non convex constraint in (15) or (10).
B. Main Result The quality of this approximation depends on the choice of
The main result of this section is given in the following Ld . It can be shown that better approximation is achieved if
theorem. Ld is close to Li (ρ) [13]. Note that L i (ρ) can be written as:
Theorem 1: Consider the set of models P in (3) with
Li (ρ) = C(ρ)Gn + C(ρ)(Pi − Pn ) (23)
multiplicative uncertainty filter W 2 (jω), then the linearly
parametrized controller in (4) in the SP structure guarantees Therefore, L d can be seen as the desired open loop transfer
closed-loop stability and satisfy the following robust perfor- function of the system without dead-time, i.e., C(ρ)G n . This
mance condition: will be a good approximation if τ i is close to τn or the model
mismatch Pi − Pn is small. For systems with large uncertainty
|W1 Si | + |W2 Ti |∞ < 1 for i = 1, . . . , q (15) in the time delay, for each model P i different L d can be
if considered. In this case, L di can be computed as follows:
Ld
|{W1 (jω)[1 + C(jω, ρ)H(jω)]+ Ldi = Ld + (Pi − Pn ) (24)
Gn
W2 (jω)C(jω, ρ)Pi (jω)}[1 + Ld (jω)]|− Note that Ldi should not encircle the critical point. An iterative
Re{[1 + L∗d (jω)][1 + Li (jω, ρ)]} < 0 algorithm can be used to improve the results. In this algorithm
∀ω for i = 1, . . . , q (16) Ld in the (k + 1)-th iteration is computed using the controller
of the last iteration Ck (ρ) and (23):
where Ld (jω) is a strictly proper transfer function which
does not encircle the critical point and L ∗d (jω) is its complex Ldi = Ck (ρ)Gn + Ck (ρ)(Pi − Pn ) (25)
conjugate. C. Primary controller design
Proof: Since the real part of a complex number is less
The problem of minimizing the upper bound γ of the in-
than or equal to its magnitude, we have
finity norm of the weighted sensitivity function is considered.
Re{[1 + L∗d ][1 + Li (ρ)]} ≤ |[1 + L∗d ][1 + Li (ρ)]| (17) Therefore, the primary controller should be obtained from the
following optimization problem:
Then, using (16) and the fact that |1 + L d | = |1 + L∗d |, one
obtains min γ
ρ
Subject to: (26)
|W1 (1 + C(ρ)H)| + |W2 C(ρ)Pi | − |1 + Li (ρ)| < 0
|W1 Si |+|W2 Ti |∞ < γ for i = 1, . . . , m
∀ω for i = 1, . . . , q (18)
This optimization can be convexified using Theorem 1 and
Using Li (ρ) = C(ρ)(H + Pi ) we have solved by an iterative bisection algorithm. At each iteration


j, γj is fixed and W1 and W2 are replaced by W 1 /γj and

W1 (1 + C(ρ)H) + W2 C(ρ)Pi



<1 W2 /γj . Then, a feasibility problem is solved under the convex

1 + C(ρ)(H + Pi )

constraints (16). If the problem is feasible, γ j+1 is chosen
∀ω for i = 1, . . . , q (19) smaller than γj . Otherwise γj+1 is increased.
 Notice that the condition (16) is defined for every frequency
ω leading to an infinity number of constraints. In practice,
it is possible to define a frequency range and grid it with a
sufficiently large number of frequency points N . The effect of
gridding on the stability and performance of the closed loop
system has been studied in [15].
Remark II: It is also possible to handle multimodel un-
certainty in the proposed approach. Consider that the plant
belongs to a set of stable models P where:
P = {G1 e−τ1 s , G2 e−τ2 s , . . . , Gm e−τm s }
and leads to γ = 0.313. This controller is compared to that
proposed in [16]. Kaya’s controllers performs better than other
controllers presented in the literature ([8], [17] and [18]). Fig.
3 depicts the performance of both controller on unitary step
setpoint change considering the time-delay τ = 4.5s, τ = 5.0s
and τ = 5.5s. As it can be seen, both controller performed
well, however, the proposed controller achieves faster response
even for the uncertain plant.
(27) 1

−τn s
It is assumed that nominal model P n = Gn e

Outputs
representing 0.5
τ = 4.5s

the set is available. In this case, the robust performance

condition is 0
0 5 10 15 20 25 30 35 40

W1i Si ∞ < γ for i = 1, . . . , m (28) 1

τ= 5.0s

Outputs
which can be approximated by the following convex con- 0.5

straints:
0
0 5 10 15 20 25 30 35 40
|W1i (jωk )[1 + C(jωk , ρ)H(jωk )][1 + Ldi (jωk )]|−
Re{[1 + L∗di (jωk )][1 + Li (jωk , ρ)]} < 0
1

Outputs
τ= 5.5s
k = 1, . . . , N ; m = 1, . . . , m (29) 0.5

Note that it is possible to define different performance speci- 0

0 5 10 15 20 25 30 35 40

fication for each model in the set (by different W 1i and Ldi ).
Time [s]

D. Example 1
Fig. 3. Example 1: Blue solid line: proposed; black dot-dashed line: ref [16]
Consider the process described by (1) with multiplicative
uncertainty as in (2) with
IV. E XTENSION TO U NSTABLE S YSTEMS
1
Gn (s) = (30) The SP in the scheme shown in Fig. 1 cannot be used for
(5s + 1)(10s + 1)
unstable plants since the controller will contain zeros in right-
and hand side of the s-plan which cancel the unstable poles in
−s2 − 2s
W2 (s) = (31) the plant and leads to instability. To avoid this unstable zero-
s2 + 2s + 1
pole cancellation, the control structure shown in Fig.1 should
The unknown time delay τ belongs to the interval [4.5, 5.5]. be changed. Several alternatives are available in literature to
The nominal model used in the SP structure is chosen as cope with unstable processes with dead-time (see, for example,
Pn (s) = Gn (s)e−5s . By griding the range of allowable dead- [19], [1], [20], [21], [22]). Consider, for instance, the SP with
times a set P = {Pi (s), i = 1, . . . , q} of models is generated modified dead-time free model depicted in Fig. 4 which is
where: discussed in [1]. In this case, the dead-time free model is
Pi (s) = Gn (s)e−τi s , i = 1, . . . , q (32) defined as Gm = N m
Dn and the equivalent controller becomes:
In this example q = 3 is chosen. The performance specification C(ρ)
is defined by W1 (s) which is given by Ceqm (ρ) = (35)
1 + C(ρ)H
2
W1 (s) = (33)
(30s + 1)2
r + u y
The goal is to compute a PID primary controller that mini- C(s) P (s)
mizes |W1 Si | + |W2 Ti |∞ < γ for i = 1, 2, 3. −

Since the controller has an integrator, L d is chosen as +

Ld (s) = ωc /s where ωc = 0.1 rad/s which is 20% higher than Gm (s) Pn (s) −
open loop bandwidth. Then, the optimization problem (26) is
solved considering N = 100 equally spaced frequency points
yp +
between 10−3 and 103 rad/s. The resulting primary controller
is: +
12.3s2 + 3.28s + 0.2201
C(s) = (34) Fig. 4. Smith Predictor with modified dead-time free model
0.01s2 + s
where
1
H = Gm − Pn = (Nm − Nn e−τ s )
Dn
Note that the poles of H are zeros of C eqm (ρ). Therefore,
Nm must be tuned such that the zeros of N m − Nn e−τ s
cancel the unstable poles in D n . Once Nm has been properly
designed, the primary controller can be obtained by solving
the optimization problem in (26) redefining H = G m − Pn
and
Li (ρ) = (Gm − Pn + Pi )C(ρ)
Here, care should be taken in the choice of L d . As it has
been shown, the wno of 1 + L i equals the wno of 1 + L d .
Therefore, L d should be chosen such that the number of
encirclement of the critical point (−1 + 0j) by its Nyquist
plot is equal to the number of unstable poles in P i .
A. Example 2
Consider the model studied in [6] given by:
problem. With this new L d (s) = Gm (s)C0 (s) the optimal
primary controller is:
2.994s + 0.4612
C(s) = (42)
s
and γ = 0.6074. Figure 5 depicts the function
Γi (jω) = |W1 (jω)Si (jω)| + |W2 (jω)Ti (jω)|
where Si and Ti are respectively given by (7) and (8) with
H = Gm − Pn and Pi is obtained by gridding of τ . Note that
the maximum value of the function is 0.6072 which occurs
when τ = τn +0.02 = 0.22 is close to the bound γ. It is worth
to point out that, although the conditions given in Theorem 1
are only sufficient to guarantee |W 1 Si |+|W2 Ti |∞ < γ, with
a proper choice of L d it is possible to obtain a solution with
very low conservatism. Furthermore, the resulting controller is
a standard PI which can be implemented in a straightforward
manner and has great practical significance.

k
P (s) = [1 + ∆(s)W2 (s)]e−τ s (36)
s−a 0.65

where k = 1, a = 1, τ = τn ± 0.02 and τn = 0.2. The 0.6

τ +∆τ
interval of variation of τ is gridded using q = 3 equally spaced 0.55
n
points. The performance and uncertainty filters are respectively
chosen as: 0.5

s+1 s + 1.1 τ −∆τ

Magnitude

0.45 n
W1 (s) = 2 and W2 (s) = 0.2 (37)
10s + 1 s+1 0.4
The objective is to minimize γ such that the robust per-
0.35
formance condition |W 1 Si | + |W2 Ti |∞ < γ holds for
i = 1, 2, 3. 0.3
Here, we use the SP with modified dead-time free model
0.25
(Fig. 4) due to its simplicity. The dead-time free model G m (s)
is chosen as 0.2
Tm s + 1
−2 0 2
10 10 10
Gm (s) = . (38) Frequency, [rad,s]
s−1
Tm is computed in order to obtain H(s) = G m (s) − Pn (s) Fig. 5. Example 2: Blue solid line: Γ for τ = 0.18; green solid line: Γ for
τ = 0.2; black solid line: Blue solid line: Γ for τ = 0.22; red dot-dashed
without a pole in s = 1. Since line: γ.
1
H(s) = [Tm s + 1 − e−0.2s ], (39)
s−1 For the same example, controller designed in [6] leads to
s − 1 is canceled if Tm = e −0.2
− 1. the optimal γ = 0.9407 which is 55% higher than the value
It is desired to compute a PI as the primary controller. The obtained with the proposed method. It should be mentioned
first step is to choose the transfer function L d (s), which must that the true robust performance criterion in (26) is not
encircle the critical point in the Nyquist diagram once and minimized in [6]. Instead, the maximum singular value of
must contain the integrator. Therefore, it is chosen as [W1 (jω)S(jω) W2 (jω)T (jω)] for all ω is minimized by the
H∞ control theory.
s+1
Ld (s) = 10 . (40)
s(s − 1) V. E XTENSION TO LPV S YSTEMS
Optimization problem (26) is solved considering N = 100 Consider an uncertain LPV plant P (s, θ) belonging to the
equally spaced frequency points between ω = 10 −3 rad/s and set:
ω = 103 rad/s and the following controller is obtained: P = {G(s, θ)e−τi (θ)s , i = 1, . . . , q} (43)
3.582s + 0.5838
C0 (s) = (41) where the dead-time free part of the model has unstructured
s
multiplicative uncertainty and is described as:
which yields γ = 0.6854. This result can be further improved
by using a new L d (s) based on C0 (s) in the optimization G(s, θ) = Gn (s, θ)[1 + ∆(s)W2 (s)] (44)
where θ is a vector of scheduling parameters. The dead-time Optimization problem (49) is again solved using an iterative
is also a function of the scheduling parameter and uncertain, bisection algorithm as previously presented. At each iteration,
so for a given value of θ it lies within the interval τ min (θ) ≤ a feasibility problem is solved with the following convex
τi (θ) ≤ τmax (θ). constraints:
We will consider the SP shown in Fig. 6 where both,
|{W1 (jωk )[1 + C(jωk , θl )H(jωk , θl )]+
the nominal model P n (s, θ) = Gn (s, θ)e−τn (θ)s and the
primary controller C(s, θ) are functions of the scheduling W2 (jωk )C(jωk , θl )P (jωk , θl )}[1 + Ld (jωk )]|− (50)
parameter vector θ. The goal is to compute a primary gain- Re{[1 + L∗d (jωk )][1 + Li (jωk , θl )]} < 0
scheduled controller for this scheme that meets the H ∞ robust for k = 1, . . . , N, i = 1, . . . , q, l = 1, . . . , m.
performance specification.
where θl is a vector with particular values of the scheduling
+ y
parameters θ obtained from the gridding.
r u
C( s, θ) P( s, θ) It should be noted that since the design is performed in the
− frequency domain, the designed controller does not guarantee
the closed-loop stability and performance for fast variation of
+
Gn( s, θ) − the scheduling parameter.
e− τn( θ) s
A. Example 3
+
yp The design method is applied on a simulated system having
+ a resonance whose frequency changes as a function of a
scheduling parameter θ. Consider the following LPV plant
Fig. 6. Gain-Scheduled Smith Predictor model
P (s, θ) = G(s, θ)e−τ s (51)
The primary controller C(s, θ) is linearly parametrized by:
where G(s, θ) = Gn (s, θ)[1 + ∆(s)W2 (s)] and
C(s, θ) = ρT (θ)φ(s) (45)
(2 + 0.2θ)2
Gn (s, θ) = (52)
where the basis function vector φ(s) is defined in (5) and s2 + 0.2(2 + 0.2θ)s + (2 + 0.2θ)2
ρT (θ) is given by 1.1337s2 + 6.8857s + 9
W2 (s) = 0.8 (53)
ρT (θ) = [ρ1 (θ), ρ2 (θ), . . . , ρn (θ)] (46) (s + 1)(s + 10)
and θ ∈ [−1, 1]. This model could represent the dynamics of
Every gain is a polynomial function of order δ of the schedul-
a mechatronic system, where the frequency of the resonance
ing parameters and is defined as
is a function of the moving mass. Consider also that the dead-
ρi (θ) = (νi,δ )T θδ + . . . + (νi,1 )T θ + νi,0 (47) time is within the interval τ ∈ [2.7, 3.3] but its exact value is
unknown in runtime. The objective is to design a primary gain-
and θ denotes element-by-element power of k of vector θ.
k
scheduling PID controller for the Smith Predictor structure
The sensitivity and complementary sensitivity functions are considering the performance filter
respectively given by:
2
1 + C(s, θ)H(s, θ) W1 (s) = (54)
Si (s, θ) = (48) (20s + 1)2
1 + C(s, θ)(H(s, θ) + Pi (s, θ))
The parameters ρ of the primary controller will be affine
C(s, θ)Pi (s, θ) functions of the scheduling parameter θ. The filter of the
Ti (s, θ) = , ∀θ
1 + C(s, θ)(H(s, θ) + Pi (s, θ)) derivative action is chosen to have a time constant of T =
where H(s, θ) = Gn (s, θ) − Pn (s, θ). The primary controller 0.01s. The range of the scheduling parameter θ is gridded
should be obtained from the following optimization problem: using m = 5 points. The uncertainty interval of the dead-time
is gridded using q = 3 points. Finally, optimization problem
min γ (49) is solved considering L d = 1/s and N = 100 equally
ρ
Subject to: (49) spaced frequency points between 10 −2 and 102 rad/s. The
resulting gain-scheduled controller is given by:
|W1 Si (s, θ)| + |W2 Ti (s, θ)|∞ < γ
Kp (θ) = −0.0168θ + 0.2152
for i = 1, . . . , q, ∀θ
Ki (θ) = 0.0144θ + 2.4736
As an LPV model is available, there will be an infinity
Kd (θ) = −0.1224θ + 0.6424
number of models corresponding to different values of the
scheduling parameters. This problem can be solved by grid- This controller leads to:
ding θ leading to m different LTI models. Considering the
|W1 Si (s, θl )| + |W2 Ti (s, θl )|∞ < γ = 0.8928 (55)
multimodel uncertainty, the procedure should be repeated for
each one of the q uncertain LPV models. l = 1, . . . , 5, i = 1, . . . , q
The gain-scheduled controller is evaluated considering θ 1 =
−1, θ3 = 0 and θ5 = 1 and considering the dead-time 10%
higher than its nominal value. The performance is compared
to a fixed-gain PID designed for the nominal case (θ 3 = 0
and dead-time 3s). Figure 7 shows the results of a setpoint
step response. As it can be seen, the gain-scheduled controller
performs very well in all conditions (blue, red and green solid
lines) while the fixed PID is highly oscillating (black dashed
line).
1 [10] L. Mirkin, “On the extraction of dead-time controllers and estimators
0.9
0.8
0.7
0.6
Outputs
0.5
0.4
0.3
0.2
0.1
0 USA, 2010.
0 5 10 15 20 25 30
Time [s]
Fig. 7. Example 3: Blue, red and green solid line: gain-scheduled PID Smith
Predictor and G2 using θ1 = −1, θ3 = 0 and θ5 = 1 respectively; black
dashed line: fixed PID Smith Predictor using θ = −1.
VI. C ONCLUSION
This paper presents a new method to design robust Smith
predictor for uncertain time-delayed systems using convex
optimization techniques. The proposed approached allows one
to design PI/PID as well as higher order primary controllers in
the Smith predictor structure which provide robust H ∞ perfor-
mance for systems with uncertain dead-time and multiplicative
or multimodel uncertainty in the dead-time free model of the
system. The method is based on a convex approximation of
the H∞ robust performance criterion in the Nyquist diagram.
This approximation relies on the choice of a desired open-
loop transfer function L d for the dead-time free model of the
plant. The choice of this transfer function can be performed
iteratively using the last designed controller. The simulation
examples show good performance of the designed controllers
with simple choices of L d .
R EFERENCES
[1] J. Normey-Rico and E. Camacho, Control of dead-time processes.
Springer Verlag, 2007.
[2] Z. Palmor, “Time-delay compensation Smith predictor and its modifica-
tions, the control handbook, w. levine, ed,” 1996.
[3] O. Smith, “Closer control of loops with dead time,” Chemical Engineer-
ing Progress, vol. 53, no. 5, pp. 217–219, 1957.
[4] Z. Palmor, “Stability properties of Smith dead-time compensator con-
trollers,” International Journal of Control, vol. 32, pp. 937–49, 1980.
[5] C. Brosilow, “The structure and design of Smith predictors from the
viewpoint of inferential control,” in Proceedings of Joint American
Control Conference, Denver, Colorado, 1979.
[6] G. Meinsma and H. Zwart, “On H∞ control for dead-time systems,”
IEEE Transactions on Automatic Control, vol. 45, no. 2, pp. 272–285,
2000.
[7] C. Santacesaria and R. Scattolini, “Easy tuning of Smith predictor in
presence of delay uncertainty,” Automatica, vol. 29, no. 6, pp. 1595–
1597, 1993.
[8] Z. Palmor and M. Blau, “An auto-tuner for Smith dead time compen-
sator,” International Journal of Control, vol. 60, no. 1, pp. 117–135,
1994.
[9] D. Lee, M. Lee, S. Sung, and I. Lee, “Robust PID tuning for Smith
predictor in the presence of model uncertainty,” Journal of Process
Control, vol. 9, no. 1, pp. 79–85, 1999.
from delay-free parametrizations,” IEEE Transactions on Automatic
Control, vol. 48, no. 4, pp. 543–553, 2003.
[11] Q. Zhong, “On standard H∞ control of processes with a single delay,”
IEEE Transactions on Automatic Control, vol. 48, no. 6, pp. 1097–1103,
2003.
[12] ——, “H∞ control of dead-time systems based on a transformation,”
Automatica, vol. 39, no. 2, pp. 361–366, 2003.
[13] A. Karimi and G. Galdos, “Fixed-order H∞ controller design for
nonparametric models by convex optimization,” Automatica, vol. 46,
no. 8, pp. 1388–1394, 2010.
[14] C. J. Doyle, B. A. Francis, and A. R. Tannenbaum, Feedback Control
Theory. New York: Mc Millan, 1992.
[15] G. Galdos, A. Karimi, and R. Longchamp, “Robust controller design
by convex optimization based on finite frequency samples of spectral
models,” in 49th IEEE Conference on Decision and Control, Atlanta,
[16] I. Kaya, “Tuning Smith predictors using simple formulas derived from
optimal responses,” Industrial & engineering chemistry research, vol. 40,
no. 12, pp. 2654–2659, 2001.
[17] T. Hagglund, “A predictive PI controller for processes with long dead
times,” IEEE Control Systems Magazine, vol. 12, no. 1, pp. 57–60, 1992.
[18] C. Hang, Q. Wang, and L. Cao, “Self-tuning Smith predictors for
processes with long dead time,” International journal of adaptive control
and signal processing, vol. 9, no. 3, pp. 255–270, 1995.
[19] J. Normey-Rico and E. Camacho, “Unified approach for robust dead-
time compensator design,” Journal of Process Control, vol. 19, no. 1,
pp. 38–47, 2009.
[20] T. Liu, Y. Cai, D. Gu, and W. Zhang, “New modified Smith predictor
scheme for integrating and unstable processes with time delay,” IEE
Proceedings-Control Theory and Applications, vol. 152, no. 2, pp. 238–
246, 2005.
[21] S. Majhi and D. P. Atherton, “A new Smith predictor and controller for
unstable and integrating processes with time delay,” in Proceedings of
the 37th IEEE Conference on Decision and Control, vol. 2. IEEE,
1998, pp. 1341–1345.
[22] A. M. De Paor, “A modified Smith predictor and controlled for unstable
processes with time delay,” International Journal of Control, vol. 41,
no. 4, pp. 1025–1036, 1985.