Professional Documents
Culture Documents
U(s)
+
+
Y(s)
P(s)
Ym(s)
Gn(s)
e-Lms
I. I NTRODUCTION
Over the last decades, the study of dead-time systems has
been widely considered, regarding its importance for both
academia and industry. Within this context the Smith Predictor (SP) [1], described as the first dead-time compensator
(DTC) proposed in literature, plays an important role.
Many modifications regarding the SP structure have been
proposed over the last decades as [2] to deal with integrating
processes and [3]. The Filtered Smith Predictor (FSP) includes a robustness filter which makes the proposed structure
able to deal with stable, integrating and unstable cases as
it improves closed loop robustness properties [4], [5]. High
frequencies noise attenuation is dealt in [6] and the study
of SISO systems with multiple dead-time can be found in
[7] and [8]. A FSP structure with simplified tuning rules
is proposed in [9] for stable, unstable and integrating first
order plus dead-time (FOPDT) systems, which considers the
controller and the reference filter as simple gains, easing the
tuning procedure by making it more intuitive.
Following this idea, the work developed in this paper
investigates simple tuning rules, when the automatic tuning
methods proposed by [10][14] and considered by [15] are
applied.
This paper is organized as follows: Section II describes the
Simplified FSP for the continuous-time case, Section III
present the identification method of the automatic tuning,
Section IV describes the simple tuning rules and the performance assessment used by the proposed method, Section V
show simulations results for each case of processes and in
Section VI conclusions are commented.
C(s)
Fig. 1.
Fr(s)
FSP structure.
C(s)Fr (s)
1 + Gn (s)C(s)(1 Fr (s)eLm s )
and
Feq =
F (s)
.
Fr (s)
(3)
(4)
Q(s)
R(s)
Feq(s)
Ceq(s)
U(s)
+
+
Y(s)
P(s)
Fig. 2.
s0
Note that Eqs. (11) and (12) or Eqs. (13) may be arranged
as a linear system of size two whose variables of interest
are b1 and b2 , leading to obtain b2 = kr and, for the stable
case,
(Tm )2
+ b2 Tm ,
(15)
b1 =
Tm eLm /Tm
for the unstable case,
b1 =
s0
1 + kc Gn (s)
.
kc Gn (s)
(6)
Tm
,
Km (Tcl + 1)
(8)
1
,
Km Tcl
(9)
Ceq (s) =
Gn (s)
1+C(s)Gn (s)
C(s)Gn (s)
Fr (s)eLm s
.
b1 = 2 + Lm +
0 (13)
0. (14)
1
.
kc Km
(17)
As
.
(19)
t+
s0
K
From the structure of the SFSP, the control signal expression
can be obtained as:
lim u(t) = lim sU (s) =
(16)
F (s) As
Y (s).
Fr (s) s
(20)
(21)
Substituting Eqs. (20) and (21) in Eq. (19), the static gain
results
As K m
K=
,
(22)
IEf
where
ef (t)dt
IEf =
(23)
and
= 2 b + Lm + Tcl ,
(24)
(25)
IEsu
L.
As
Ad =
IY
.
Km
(29)
(30)
Ad Km
.
IJ
(32)
(33)
KeLs
(36)
Ts 1
1) Set-point step change: Applying the FVT to the
control signal u(t) after a set-point step change, it can be
obtained for processes like Eq. (36) that
P (s) =
t+
s0
As
.
K
K=
As K m
.
IEf
(35)
(38)
(39)
IY = Ad Km .
(42)
IY
.
Km
(43)
(34)
(37)
Substituting the control variable U (s) for Eqs. (20) and (21)
in Eq. (37), the static gain is obtained as the same relation
for stable processes:
Ad Km
.
IJ
Defining the following variable:
B. Unstable processes
(27)
(45)
(46)
= K(T L)IJ .
(47)
C. Integrative processes
L=
DIEij
.
DIJ
(63)
Ke
(49)
s
1) Set-point step change: Applying the FVT to the output
variable of a process described by Eq. (49) after a set-point
step change, it is obtained
P (s) =
t+
s0
where
U (s)
= KIU = As ,
s
A. Tuning
(50)
IU =
u(t)dt.
(51)
(52)
(53)
(54)
IEiu
.
As
Once the new model (for the uncertain system) has been
identified, we proceed to apply simple tuning rules to the
controller. Once the controller is reset, then the performance
assessment is performed based on the new parameters and
the actual response of the system, in order to compute
performance indexes.
(55)
(56)
= 2 + 2Tcl L2m /2 + b1 Lm .
(57)
where
The tuning rules that are defined for the SFSP were
devised for the FSP in [15]. Given the similarities of the two
controllers, that tuning rules can apply to the corresponding
tuning parameters of the SFSP, namely, and Tcl .
For disturbance rejection, the tuning rules were devised
through the nominal sensitivity peak Ms using an interval
of Lm /Tm (0, 10], that is, covering a considerable
range of lag-dominant and dead-time dominant models. The
tuning can be done with = Lm /2 (for Ms = 1.5) for
stable processes and with = Lm (for Ms = 2.2) for
integrative processes. For unstable processes, it can also be
done = Lm (for Ms = 3), since Lm /Tm < 0.6.
For set-point tracking, considering lag-dominant processes,
the closed-loop time constant Tcl can be chosen between the
interval [Lm /4,Lm /2], where there is a trade-off between
robustness and performance. Hence, lower values result
in higher performance and higher values result in robust
responses. For dead-time dominant processes, a conservative
rule is to choose Tcl = Tm [17].
B. Performance assessment
In general, the performance assessment of a control
system is made by comparing the actual performance with
a target performance. As the performance index used to the
assessment it is chosen the integrated absolute error IAE ,
defined as
Z
IAE =
|e(t)|dt.
(64)
IY
.
Km
(58)
Taking the double integral of the variable j(t) (Eq. 30) and
applying the FVT, it is obtained
Z Z t
Ad Km
J(s)
. (59)
DIJ =
j()ddt = lim s 2 =
s0
s
K
0
0
Therefore, the static gain is
Ad Km
.
DIJ
Defining the variable below:
Z t
eij (t) = K
j()d y(t),
K=
(60)
This choice is based on the characteristic that its minimization results generally in low overshoot and reduced settling
time [18].
The actual IAE value is computed after a set-point step
change or a step load disturbance. After the identification,
the target IAE can then be computed assuming that the
identified model perfectly represents the plant.
1) Set-point step change: For set-point tracking, the
closed-loop transfer function of the SFSP for FOPDT, IPDT
and UFOPDT models is given by
Hr (s) =
(61)
eLm s
Y (s)
=
.
R(s)
Tcl s + 1
(65)
For the SFSP, as for the FSP, considering the above transfer
function and a monotonic response (no overshoot) of the
(69)
IAEld = |Ad Km |,
Set-point
Initial tuning
SP Autotuning
LD Autotuning
0.5
(70)
3) Performance indexes: The Set-Point Performance Index (SPPI) can then be defined below as the ratio between
the target and the actual IAE values:
IAEsp
.
(71)
IAE
Similarly, the Load Disturbance Performance Index (LDPI)
is defined as
IAEld
LDPI =
.
(72)
IAE
Theoretically, the SFSP has satisfactory performance when
the performance index value is equal to 1. Nevertheless,
in a pratical context, a FSP can be considered well tuned
when that value is greater than 0.8 [15]. Certainly, if the
system has tighter performance requirements, that value can
be higher.
SPPI =
50
100
150
time
control variable u
IAEsp = As (Tcl + Lm ).
Initial tuning
SP Autotuning
LD Autotuning
60
40
20
0
0
50
100
150
time
Fig. 3. Set-point and load disturbance step responses for process P1 (s)
with the initial tuning and with the autotunings.
B. Unstable process
The second example is a model of concentration control
in an unstable reactor [5].
3.433e20s
.
(74)
103.1s 1
The uncertainties were +10% for the static gain, 5% for
the time constant and +10% for the dead-time. The resulting
parameters are Km = 3.776, Tm = 97.945 and Lm = 22.
The chosen tuning rules are = Lm and Tcl = Lm /4.
After a set-point unit step change is applied, actual IAE1
= 39.69 and the estimated parameters are Km = 3.457, Tm
= 100.318 and Lm = 20.3, resulting SPPI = 0.639. After
a step load disturbance of magnitude 0.2, the actual IAE2
= 20.388 and Ad = 0.2, Km = 3.489, Tm = 106.505 and
Lm = 22.2, resulting LDPI = 0.786. The output and control
signal responses are shown in Fig. 4.
With the autotuning based on the set-point step response it
was obtained actual IAE1 = 27.826, SPPI = 0.912, actual
IAE2 = 14.08 and LDPI = 1. While with the autotuning
based on the load disturbance step response it was obtained
actual IAE1 = 29.897, SPPI = 0.849, actual IAE2 = 16.089
and LDPI = 0.996.
P2 (s) =
V. S IMULATION RESULTS
The chosen examples are stable, unstable and integrative
processes presented in [5] and [19]. For each case, it was
added model uncertainties considering that the initial model
was incorrectly estimated or that the parameters of the plant
changed. In the same simulation are performed both the
autotunings based on the set-point (SP) step change as
on the step load disturbance (LD). Its responses are then
plotted in the same figure with the initial tuning response
for comparison. Furthermore, the tuning rules of Section
IV-A were used for each controllers.
A. Stable process
The model for temperature control in a heat exchanger
[5] is presented below.
P1 (s) =
0.12e3s
.
6s + 1
(73)
C. Integrative process
The integrative process below is presented in [19].
P3 (s) =
0.1e8s
.
s(s + 1)(0.5s + 1)(0.1s + 1)
(75)
output variable y
1.5
1
Set-point
Initial tuning
SP Autotuning
LD Autotuning
0.5
0
0
100
200
300
400
500
600
700
800
900
time
control variable u
6
Initial tuning
SP Autotuning
LD Autotuning
4
2
0
-2
0
100
200
300
400
500
600
700
800
900
time
Fig. 4. Set-point and load disturbance step responses for process P2 (s)
with the initial tuning and with the autotunings.
ACKNOWLEDGMENT
The authors would like to thank the financial support
for scientific research provided by the Brazilian institutions
CNPq and FUNCAP.
R EFERENCES
output variable y
Set-point
Initial tuning
SP Autotuning
LD Autotuning
0.5
0
0
100
200
300
400
500
600
700
800
control variable u
time
Initial tuning
SP Autotuning
LD Autotuning
0
0
100
200
300
400
500
600
700
800
time
Fig. 5. Set-point and load disturbance step responses for process P3 (s)
with the initial tuning and with the autotunings.
VI. C ONCLUSION
It was presented an automatic tuning methodology applied
to a SFSP dead-time compensator that can cope with stable,