ARTICLE IN PRESS

Control Engineering Practice 13 (2005) 485–497

Design and tuning of a ratio controller$

Antonio Visioli*
Dipartimento di Elettronica per l’Automazione, University of Brescia, Via Branze 38, 25123 Brescia, Italy
Received 16 July 2003; accepted 12 April 2004
Available online 25 May 2004

Abstract

In this paper a design and tuning procedure for a ratio control architecture is proposed. The overall control scheme is based on the
use of the Blend station proposed in (Control Eng. Pract. 9 (11) (2001) 1215) and standard PI controllers. Since all the control
parameters can be automatically selected based on a simple model of the process under control, the proposed methodology is easy to
implement and therefore suitable to be applied in the industrial context. Simulation and experimental results show the effectiveness
of the methodology for a wide range of processes.
r 2004 Elsevier Ltd. All rights reserved.

Keywords: Ratio control; Auto-tuning; PID controllers; Blending

1. Introduction Recently, this effort has been further motivated by the

Proportional-integral-derivative (PID) controllers are
the controllers most adopted in industry due to the good
cost/benefit ratio they are able to provide for a wide
range of processes. Often, they are employed as basis of
more complex control schemes where couplings between
simple control systems are exploited. An example is
ratio control, which consists of keeping a constant ratio
between two process variables. This is actually required
in many applications, such as chemical dosing, water
treatment, chlorination, mixing vessels, waste incinera-
tors. For example, in combustion systems the air-to-fuel
ratio has to be controlled to obtain an high efficiency,
and in blending processes a selected ratio of different
flows has to be maintained to keep a constant product
composition.
In the last 60 years, a major effort has been provided
by researchers to develop useful techniques for the
implementation of the basic PID algorithm (tuning and
automatic tuning methods) and of additional function-
alities such as anti-windup, gain scheduling, adaptive
control and so on (Astr ( om
. & H.agglund, 1995).
$
This work was supported in part by MIUR scientific research
funds.
*Tel.: +39-030-371-5460; fax: +39-030-380-014.
E-mail address: visioli@ing.unibs.it (A. Visioli).
increase of the computational capability which is
available in modern single-station industrial controllers
and distributed control systems (DCS). Conversely, the
design of methodologies for the implementation of the
above mentioned basic couplings has been much over-
looked. Obviously, to be suitable for industrial settings,
in addition to the achievement of high performances, the
ease of understanding and of use of new techniques is a
major requirement.
A relevant recent work in this context is the one of
H.agglund (2001) in which a new ratio control structure
is proposed. Based on this control scheme, in this
paper a design methodology for a ratio controller is
proposed. A salient feature of the proposed method is
that all the control parameters are selected based on
simple models obtained by accomplishing standard
identification experiment, and therefore the overall
method can be easily performed automatically. The
paper is organised as follows. In Section 2, a short
introduction of ratio control is provided and H.agglund’s
Blend station is briefly reviewed. In Section 3, the new
ratio control architecture is proposed. The tuning
procedure is revealed in Section 4. Simulation results
are presented and discussed in Section 5, whilst
experimental results obtained with a laboratory equip-
ment are shown in Section 6. Finally, conclusions are
drawn in Section 7.

0967-0661/$ - see front matter r 2004 Elsevier Ltd. All rights reserved.
doi:10.1016/j.conengprac.2004.04.010

486 A. Visioli / Control Engineering Practice 13 (2005) 485–497
2. Ratio control and the Blend station
The aim of a ratio control system is to keep the ratio
between the values of two process variables y1 and y2
equal to a constant value a; in order to meet some
higher-level requirements. For this purpose, the control
scheme shown in Fig. 1 is usually implemented. Each
variable is controlled by two separate controllers C1 and
C2 (typically of PI type) and the output y1 of the first
process is multiplied by a and adopted as the set-point of
the closed-loop control system of the second process, i.e.
it is r2 ðtÞ ¼ ay1 ðtÞ (Shinskey, 1996). In this way, at the
steady state, provided that the gain of the second loop is
equal to unity (note that this condition is normally
verified by the presence of the integral part in the
controller) the requirement
y2 ðtÞ
¼a
y1 ðtÞ
is satisfied.
The main disadvantage of this scheme is related to the
transient response to a change in the set-point r1 ; as the
output y2 is necessarily delayed with respect to y1 ; due to
the closed-loop dynamics of the second loop. To
overcome this drawback, H.agglund proposed an alter-
native architecture, named the Blend station (H.agglund,
2001). This is shown in Fig. 2. The main feature of the
scheme is that the value of the set-point r2 depends both
on the value of the process output y1 and on the value of
the set-point r1 ; according to the expression
r2 ðtÞ ¼ aðgr1 ðtÞ þ ð1  gÞy1 ðtÞÞ:
Note that g is a constant parameter that weights the
relative influence of the set-point r1 on r2 with respect to
y1 (for g ¼ 0 the classical scheme of Fig. 1 is obtained).
r1 u1
C1 P1
a
r2 u2
C2 P2
Fig. 1. The typical ratio control scheme.
r1
C1 P1
γ model of a plant and adopting it in the design of a
BS
r2
C2 P2
Fig. 2. The ratio control scheme using the Blend station.
ARTICLE IN PRESS
The value of g can be selected as the ratio of the time
constants of the two closed-loop systems (or, if they are
not available, as the ratio of the integral time constants
of the two controllers) or, alternatively, by applying a
suitable adaptive procedure, i.e. by applying the
following formula (H.agglund, 2001):
dg S
¼ ðay1  y2 Þ; ð2Þ
dt Ta
where SAf1; 0; 1g is a sign parameter that takes into
account if the set-point step is positive or negative. In
(H.agglund, 2001) it is suggested to select the value of the
adaptation rate Ta as a factor times the longest integral
time of the two loops. Note that, for the two PI
controllers, explicit tuning rules to be adopted in this
context are not given.
3. The new ratio control architecture
The ratio control architecture proposed in this paper
is based on the Blend station but aims at achieving
better transient responses by adopting a time-varying
parameter gðtÞ: Assume that a transition from the initial
value yi1 to the final value yf1 is required to be performed
at time t ¼ t0 from the process variable y1 (i.e. a step set-
point signal of amplitude yf1  yi1 is applied to the set-
point signal r1 ðtÞ at time t ¼ t0 ). Without loss of
generality, in the following it will be assumed that a
positive step signal is applied, i.e. yf1 > yi1 : First, the
ð1Þ second loop, has to be selected as the one with the fastest
dynamics, i.e. the dynamics of process P2 is faster than
the one of P1 : This is actually the obvious choice in the
typical ratio control scheme of Fig. 1, as the output of
process P2 (appropriately scaled) can follow easier the
one of process P1 as requested.
Processes P1 and P2 are modelled with first order plus
dead time (FOPDT) transfer functions:
y
1
K1
P1 ðsÞ ¼ eL1 s ; ð3Þ
T1 s þ 1
K2
y P2 ðsÞ ¼ eL2 s : ð4Þ
2 T2 s þ 1
This choice is motivated by the fact that the dynamics of
many industrial processes can be well-captured by a
FOPDT model and that in any case, the knowledge of a
higher-order model cannot be significantly exploited in
y1 the synthesis of a simple PI controller (Astr ( om. &
H.agglund, 1995). Actually, obtaining a high-order
controller might prevent the good cost/benefit ratio that
is the main reason of the extensive use of PI(D)
controllers in industrial settings. However, it should be
y2
stressed that whereas a second-order plus dead time
(SOPDT) model is available, the controller can be
selected of PID type, where a zero of the controller is
 ARTICLE IN PRESS
A. Visioli / Control Engineering Practice 13 (2005) 485–497
employed to cancel the fastest pole of the process (see
e.g. Skogestad, 2003). This yields again to the tuning of
a PI controller based on a FOPDT model of the plant.
Thus, based on FOPDT models (3)–(4) of the two
processes, the two single-loop controllers C1 and C2 are
therefore selected as PI controllers with set-point
weighting, i.e. the manipulated variables u1 and u2 (see
Fig. 2) can be expressed as
 Z t  overall performance.
1
u1 ðtÞ ¼ Kp1 b1 r1 ðtÞ  y1 ðtÞ þ ðr1 ðtÞ  y1 ðtÞÞ dt ;
Ti1 0
ð5Þ
 Z t  (3)–(4)) has to be estimated. This can be easily
1
u2 ðtÞ ¼ Kp2 b2 r2 ðtÞ  y2 ðtÞ þ ðr2 ðtÞ  y2 ðtÞÞ dt :
Ti2 0
ð6Þ
Then, the value of g is chosen as the output of a PI
controller as well, whose input is the current ratio error,
summed to a constant value g : An additional condition
has to be set to account for the case in which L1 > L2 ; in
order to avoid that at the beginning of the transient
response the condition y2 ðtÞ > ay1 ðtÞ holds, i.e. the
output y2 starts its transient before that of y1 : Formally,
it is
8 the following formula:
<0
> if L1 > L2 and
gðtÞ ¼ tot0 þ L1  L2 ;
: g þ K ðe ðtÞ þ 1 R t e ðtÞ dtÞ elsewhere;
> T2 L1
p r Ti 0 r
ð7Þ
where
er ðtÞ ¼ y2 ðtÞ  ay1 ðtÞ: ð8Þ
In this way, the two process outputs are forced to start
their transient response at the same time instant.
It appears that, in fact, an additional feedback of the
current ratio value is implemented and therefore an
improvement in the performances with respect to the
Blend station has to be expected. From another point of
view, the adoption of a time-varying parameter g aims
actually at ‘‘shaping’’ the reference function r2 ðtÞ in such
a way that the response of the second closed-loop system
is as equal as possible to that of the first one, despite
their possible different dynamics (see the results
presented in Section 5). The adoption of this control
strategy obviously implies that an extra tuning effort is
required with respect to the Blend station. Hence, a
tuning procedure is necessary to ensure the ease of
implementation of the proposed scheme and therefore
its suitability to be adopted in industrial settings.
Table 1
Overall tuning rule of the proposed ratio controller
Kp1 Ti1 b1 Kp2
0:9T1 =K1 L1 3L1 0 0:9T2 =K2 L2
487
4. Tuning
The classical approach adopted for the tuning of a
ratio controller (see Fig. 1) is to try to obtain a value of
the dominant time constant of the second closed-loop
system much smaller than that of the first closed-loop
system. This might also lead to a detuning of the
controller C1 ; which is obviously detrimental for the
The tuning procedure for the new ratio controller
proposed in Section 3 is as follows. First, a FOPDT
transfer function for the two processes P1 and P2 (see
accomplished with standard methodologies based on
the open-loop step response, such as the well-known
( om
area method (Astr . & H.agglund, 1995). Then, the two
PI controllers C1 and C2 (see (5)–(6)) are tuned
according to the Ziegler–Nichols formula (Astr ( om. &
H.agglund, 1995) and the set-point weights b1 and b2 are
set to zero in order to avoid significant overshoots, as it
is implemented in many industrial controllers (Astr ( om.

& H.agglund, 1995, p. 110). Finally, g is chosen as
Ti2 =Ti1 and the gains of the PI controller that provides
the current value of g (see (7)) are selected according to
L2 T1 T1
Kp ¼ 0:5 ; Ti ¼ : ð9Þ
L1
The overall tuning rule is summarized in Table 1.
It appears that, being based on a simple identification
experiment and on the direct application of simple
formulas, the tuning procedure can be easily performed
automatically.
Remark 1. It is well-known that the original Ziegler–
Nichols tuning rules might yield to large overshoots in
the set-point step response. However, the use of a set-
point weight equal to zero prevents this fact and extends
the range of processes for which they provide satisfac-
tory results (see Section 5). Obviously, this implies also
that the rise time increases, but this can be accepted in a
ratio control framework, where keeping the desired ratio
is of major concern, rather than obtaining a high-
performance step response. In this context, the Ziegler–
Nichols rules have been selected for their simplicity.
However, in case the dynamics of a process is not
suitable for the Ziegler–Nichols formulas, i.e. the dead
time is greater than the dominant time constant, then a
more appropriate (though more complex) tuning rule
such as the Kappa–Tau (Astr ( om. & H.agglund, 1995)
Ti2 b2 g Kp Ti
3L2 0 Ti2 =Ti1 0:5L2 =T2 T1 =L1 T1 =L1
 ARTICLE IN PRESS
488 A. Visioli / Control Engineering Practice 13 (2005) 485–497
should be used for the two PI controllers C1 and C2 ;
whilst formula (9) is maintained (see Example 4 in
Section 5.4).
Remark 2. The value of g has been selected according
to the considerations done in (H.agglund, 2001).
Actually, this appears to be a sensible choice by taking
into account that g is the initial value of g; as the
integral part of the PI controller in (7) is capable to
maintain the most appropriate value of g at the end of
each set-point transient response.
Remark 3. Expression (9) has been obtained by
performing many simulations in which many FOPDT
processes with different values of the time constant and
of the time delay have been considered. It is therefore
empirical. However, it has a clear physical meaning. For
example, the value of the proportional gain Kp is as high
as the value of the ratio L2 =T2 between the dominant
time constant and the apparent dead time of process P2
is greater than L1 =T1 ; i.e. the one of process P1 : Taking
into account that this ratio is somehow a measure of the
ease of a plant to be controlled, and therefore of the
bandwidth of the system that can be obtained, this
means that if the dynamics of the second closed-loop
system is much faster than that of the first one, then the
value of Kp is very low, according to the intuition, as the
performance of a standard ratio controller (see Fig. 1) is
high and there is therefore no great need of using an
additional feedback of the ratio measure.
Remark 4. In case process P1 has no apparent dead
time, i.e. L1 ¼ 0; the tuning rule presented in Table 1
cannot be applied as it yields to an infinite value of the
proportional gain Kp and also of Kp1 : In this case the
value of Kp1 can be chosen by considering a fictitious
dead time equal to a small fraction of the process time
constant (for example L1 ¼ T1 =50). Then, it is no more
worthy to applying a time-varying coefficient g; as the
original Blend station performs very well. Thus, Kp ¼ 0
has to be chosen.
Remark 5. It is also worth stressing that, in any case,
being based on a standard PI controller (7), the user can
easily modify the performances of the ratio controller by
increasing or decreasing the value of Kp and Ti
according to its typical know-how, which is therefore
conveniently fully retained.
Remark 6. The proposed tuning rule is not applicable
for the ratio control of processes with underdamped
open-loop response (this class of processes has not been
considered also for the Blend station). It should be
noted that this class of processes are rather unusual in
practice and the knowledge of the type of plant is in
general sufficient to determine if an oscillatory dynamics
is present and therefore if the proposed technique
cannot be applied.
5. Simulation results
In the following examples at time t0 ¼ 0 s a unit step
is applied to the set-point signal r1 ðtÞ (i.e. yi1 ¼ 0 and
yf1 ¼ 1). Further, the value a ¼ 1 has been fixed.
5.1. Example 1
As a first example, the following two FOPDT
processes have been considered:
1 1
P1 ðsÞ ¼ e2s ; P2 ðsÞ ¼ e2s : ð10Þ
6s þ 1 2s þ 1
By applying the tuning procedure, it results in Kp1 ¼ 2:7;
Ti1 ¼ 6; Kp2 ¼ 0:9; Ti2 ¼ 6; g ¼ 1; Kp ¼ 1:5; Ti ¼ 3:
The process outputs y1 ðtÞ (thick solid line) and y2 ðtÞ
both in the case where the new methodology has been
applied (dashed line) and in the case where a fixed
value of g ¼ g is adopted following (H.agglund, 2001)
(dash-dot line) are shown in Fig. 3. For the purpose of
comparison, it has been also reported (thin solid line)
the result obtained by applying the standard ratio
control scheme of Fig. 1 (i.e. by fixing g ¼ 0). In
addition, a comparison has been made with the
adaptive Blend station. First, the value of Ta ¼ 10
maxfTi1 ; Ti2 g ¼ 60 has been fixed (see (2)), as suggested
in (H.agglund, 2001). Then, a large number of set-point
steps has been applied to the ratio control architecture,
until the value of g converges around its optimal value.
The resulting process output y2 ðtÞ is again reported in
Fig. 3 (dotted line).
The reference input r2 for the second loop with the
new method is plotted in Fig. 4 (solid line), as well as the
variation of the value of g during the transient for the
new method (dashed line) and for the adaptive blend
station (dash-dot line), respectively. It can be seen that
the variation of g for the case of the adaptive Blend
station is hardly visible (i.e. gðtÞ is nearly constant during
the transient response), and this indicates that the best
result that can be obtained with this scheme has been
achieved.
To better evaluate the results, the following
performance index has been calculated in the considered
cases:
Z N
J¼ jay1 ðtÞ  y2 ðtÞj dt: ð11Þ
0
It results in J ¼ 2:07 for the proposed method, J ¼ 4:44
for the original Blend station, J ¼ 0:35 for the adaptive
Blend station and J ¼ 12:65 for the standard ratio
controller. It has to be stressed that the achievement of
 ARTICLE IN PRESS
A. Visioli / Control Engineering Practice 13 (2005) 485–497 489

y
1

0.8

y2 standard ratio controller

process outputs

0.6
y Blend station
2

y2 proposed ratio controller

0.4

y2 adaptive Blend station

0.2

0
0 10 20 30 40 50 60 70 80
time [s]

Fig. 3. Process outputs for Example 1.

γ (t) proposed ratio controller

1.8

1.6

γ (t) adaptive Blend station

1.4

1.2
r (t) proposed ratio controller
2

0 10 20 30 40 50 60 70 80
time [s]

Fig. 4. Different signals obtained for Example 1.

the best performance by means of the adaptive Blend
station is paid by the need of performing many
experiments (i.e. of evaluating many set-point step
responses) before obtaining that result, conversely to
the other considered methodologies. Actually, if no
adaptation is employed, a significant improvement in
the performances obtained by applying the new control
scheme emerges.
For a thorough analysis of the achieved perfor-
mances, the resulting manipulated variables for the
 ARTICLE IN PRESS
490 A. Visioli / Control Engineering Practice 13 (2005) 485–497

1.5

u1

1
control variables

u2 standard ratio controller

u2 Blend station

0.5
u2 proposed ratio controller

u2 adaptive Blend station

0
0 10 20 30 40 50 60 70 80
time [s]

Fig. 5. Control variables obtained for Example 1.

addressed schemes have been plotted as well in Fig. 5. It 5.2. Example 2

can be seen that the adoption of the new methodology
does not imply a significant increase of the control effort
(although, obviously, it appears that the control effort
decreases when the ratio control is more sluggish). Note
that, since the same considerations can be done for all
the performed experiments, the resulting control vari-
able signals are not shown anymore in the following for
the sake of brevity.
In order to evaluate the performances when a series of
set-point step changes occurs, three unit steps have been
applied to the set-point signal r1 at time t ¼ 0; 8 and
20 s: Results are shown in Fig. 6 where the thick solid
line is the output of the first process. It can be seen that
with the new method (dashed line) results are still
satisfactory ðJ ¼ 2:11Þ: For comparison, the response of
the original Blend station is plotted in Fig. 6 (dash-dot
line) as well as the responses of the adaptive Blend
station (dotted line) and of the standard controller (thin
solid line). In these cases the values of the resulting
performance index are J ¼ 4:80; 0.95 and 13.27,
respectively. In Fig. 7 function gðtÞ has been plotted
both for the new scheme (dashed line) and for the
adaptive Blend station (dash-dot line) with the reference
signal r2 obtained with the new control architecture
(solid line).
It turns out that similar considerations to the case of a
single set-point step can be done when a series of set-
point changes occurs in the plant.
As a second example, the following two FOPDT
processes have been considered:
1 1
P1 ðsÞ ¼ e3s ; P2 ðsÞ ¼ e2s : ð12Þ
4s þ 1 8s þ 1
By applying the tuning procedure, it results in Kp1 ¼
1:2; Ti1 ¼ 9; Kp2 ¼ 3:6; Ti2 ¼ 6; g ¼ 0:67; Kp ¼ 0:17;
Ti ¼ 1:33: Results are reported in Fig. 8, where again the
new approach has been compared with the standard one
and with the Blend station. In this case, by applying the
adaptive procedure (with Ta ¼ 90 as suggested in
(H.agglund, 2001)) for the Blend station on a sequence
of set-point steps, the value of g converges around a
value of about 0.49. Thus, the process output y2 ðtÞ has
been reported in the case the adaptive procedure is
employed, by starting with a value of g equal to 0.49. In
Fig. 9 the value of g for the case of the Blend station
with the adaptive procedure has been plotted together
with gðtÞ and r2 ðtÞ for the new method. The resulting
values of the performance index are J ¼ 1:37 for the new
ratio controller, J ¼ 3:33 for the original Blend station,
J ¼ 1:044 for the adaptive Blend station and J ¼ 7:67
for the standard ratio controller. Indeed, the same
considerations done for Example 1 can be done also for
Example 2.
 ARTICLE IN PRESS
A. Visioli / Control Engineering Practice 13 (2005) 485–497 491

y1

2.5

2 y proposed ratio controller

2
process outputs

1.5

y Blend station
2

y adaptive Blend station

2

0.5
y standard ratio controller
2

0
0 10 20 30 40 50 60 70 80
time [s]

Fig. 6. Process outputs for Example 1 with a series of set-point step changes.

r (t) proposed ratio controller

2
3.5

2.5

γ(t) adaptive Blend station

1.5

γ(t) proposed ratio controller

1
0 10 20 30 40 50 60 70 80
time [s]

Fig. 7. Different signals obtained for Example 1 with a series of set-point step changes.

5.3. Example 3 scheme, i.e.

1 1
To demonstrate the effectiveness of the proposed P1 ðsÞ ¼ e2s ; P2 ðsÞ ¼ e3s : ð13Þ
8s þ 1 4s þ 1
methodology in different contexts, we consider as a
third example, the same processes of Example 2 but In this case we have, evidently, Kp1 ¼ 3:6;
their position has been swapped in the overall control Ti1 ¼ 6; Kp2 ¼ 1:2; Ti2 ¼ 9; g ¼ 1:5; Kp ¼ 1:5; Ti ¼ 4:
 ARTICLE IN PRESS
492 A. Visioli / Control Engineering Practice 13 (2005) 485–497

1
y2 proposed ratio controller
0.9

0.8
y2 Blend station
0.7
process outputs

0.6
y1
0.5
y2 standard ratio controller
0.4

0.3
y adaptive Blend station
2
0.2

0.1

0
0 10 20 30 40 50 60 70 80
time [s]

Fig. 8. Process outputs for Example 2.

0.9

r2(t) proposed ratio controller

0.8

0.7
γ (t) proposed ratio controller
0.6

0.5

0.4

γ (t) adaptive Blend station

0.3

0.2

0.1

0
0 10 20 30 40 50 60 70 80
time [s]

Fig. 9. Different signals obtained for Example 2.

Results obtained with the different control architectures been adopted as initial condition for the reported
are reported in Fig. 10. In this case the adaptive process output y2 ðtÞ obtained with the adaptive Blend
procedure (again with Ta ¼ 90) for the Blend station, station. As for the previous examples, in Fig. 11 the
applied when a series of set-point steps occurs, value of g for the case of the Blend station with the
converges around a value of g ¼ 2:12: This value has adaptive procedure has been plotted together with gðtÞ
 ARTICLE IN PRESS
A. Visioli / Control Engineering Practice 13 (2005) 485–497 493

y adaptive Blend station

2
1

y1

0.8
y proposed ratio controller
2
process outputs

y2 Blend station
0.6

y standard ratio controller

2
0.4

0.2

0
0 10 20 30 40 50 60 70 80
time [s]

Fig. 10. Process outputs for Example 3.

2.5

γ (t) proposed ratio controller

γ (t) adaptive Blend station

1.5

r (t) proposed ratio controller

2

0 10 20 30 40 50 60 70 80
time [s]

Fig. 11. Different signals obtained for Example 3.

and r2 ðtÞ for the new method. The resulting values of the standard ratio controller. By comparing the results
performance index are J ¼ 2:37 for the new ratio obtained for this example with those obtained for
controller, J ¼ 5 for the original Blend station, J ¼ Example 2, it appears that, as already mentioned in
2:07 for the adaptive Blend station and J ¼ 16:5 for the Section 3, it is more sensible to choose as process P1 that

494
with the fastest dynamics but in any case the perfor-
mances obtained with the proposed ratio controller are
still satisfactory (indeed, the same conclusion of
Examples 1 and 2 can be drawn also for this example).
5.4. Example 4
As a fourth example, the following two high-order
processes have been considered:
1 1
P1 ðsÞ ¼ ; P2 ðsÞ ¼
ðs þ 1Þ8 ð0:25s þ 1Þ8
By applying the identification procedure, it results
in K1 ¼ 1; T1 ¼ 2:99; L1 ¼ 5:55; K2 ¼ 1; T2 ¼ 0:71;
L2 ¼ 1:82: Being the dead time of the two processes
significantly greater than the corresponding dominant
time constant, the Kappa–Tau tuning rules have
been adopted instead of the Ziegler–Nichols ones (see
Remark 1). Thus, it results: Kp1 ¼ 0:13; Ti1 ¼ 2:62;
b1 ¼ 2:91; Kp2 ¼ 0:11; Ti2 ¼ 0:71; b2 ¼ 3:67; g ¼ 0:27;
Kp ¼ 0:69; Ti ¼ 0:54: Results are reported in Fig. 12,
where the process output y2 ðtÞ for the adaptive Blend
station has been obtained by starting with a value of g
equal to 0.32, which results after the application of a
sequence of set-point steps with Ta ¼ 26:2: In Fig. 13
the value of g for the case of the Blend station with
the adaptive procedure has been plotted together with
gðtÞ and r2 ðtÞ for the new method. The resulting values
of the performance index are J ¼ 0:52 for the new
ratio controller, J ¼ 1:81 for the original Blend station,
1
y2 proposed ratio controller
0.9
y adaptive Blend station
0.8 2
0.7
process outputs
0.6
0.5
0.4
0.3
0.2
0.1
0
0 10
ARTICLE IN PRESS
A. Visioli / Control Engineering Practice 13 (2005) 485–497
J ¼ 1:70 for the adaptive Blend station and J ¼ 4:56 for
the standard ratio controller. It appears that in this case
the proposed method provides better performances even
than the adaptive Blend station. This is possibly
explained by the fact that it is gðtÞ ¼ 0 when to3:73
(see (7)), thus allowing the two process outputs to start
their transient almost at the same time so that a very
satisfactory result is achieved.
: ð14Þ 6. Experimental results
In order to prove the effectiveness of the devised
technique in practical applications, a laboratory experi-
mental setup (made by KentRidge Instruments) has
been employed (see Fig. 14). Specifically, the apparatus
consists of two small perspex tower-type tanks (whose
area is 40 cm2 ) in which a level control is implemented
by means of a PC-based controller. Each tank is filled
with water by means of a pump whose speed is set by a
DC voltage (the manipulated variable), in the range 0–
5 V; through a PWM circuit and it is fitted with an
outlet at the base in order for the water to return to a
reservoir. The measure of the level of the water is given
by a capacitive-type probe that provides an output
signal between 0 (empty tank) and 5 V (full tank). Note
that the two processes actually have a nonlinear
dynamics, since the flow rate out of a tank depends on
the square root of its level. The task to be accomplished
is to perform an output transition from 2 to 3 V for the
y1
y2 standard ratio controller
y2 Blend station
20 30 40 50 60 70
time [s]
Fig. 12. Process outputs for Example 4.
 ARTICLE IN PRESS
A. Visioli / Control Engineering Practice 13 (2005) 485–497 495

0.9

0.8
r2(t) proposed ratio controller
0.7

0.6

γ (t) proposed ratio controller

0.5

0.4

0.3

0.2
γ (t) adaptive Blend station
0.1

0
0 10 20 30 40 50 60 70
time [s]

Fig. 13. Different signals obtained for Example 4.

In order to diversify the dynamics of the two level

Fig. 14. The experimental setup.
first tank as well as for the second one, maintaining a
desired ratio value a ¼ 1 during the whole transient
response.
control loops, as a first experiment, a software time
delay of 10 and 5 s has been added to the measure of the
level of the first and of the second tank, respectively. A
FOPDT model of each process has been estimated
separately by employing the well-known area method to
the open-loop step response. The transfer functions
obtained are
1:98 2:27
P1 ðsÞ ¼ e11s ; P2 ðsÞ ¼ e6s : ð15Þ
ð25s þ 1Þ ð25s þ 1Þ
By applying the formulas of Table 1 it results in
Kp1 ¼ 1:03; Ti1 ¼ 33; Kp2 ¼ 1:65; Ti2 ¼ 18; g ¼ 0:55;
Kp ¼ 0:27; Ti ¼ 2:27:
Results presented in Fig. 15 show again the effective-
ness of the proposed methodology that provides better
performances than both the standard ratio controller
and the Blend station. The resulting performance index
(11) (calculated over the time interval from t ¼ 0 to
t ¼ 195 s) is J ¼ 5:32 for the new approach, J ¼ 7:41 for
the Blend station and J ¼ 23:59 for the standard ratio
controller.
A second experiment has been performed by modify-
ing the added dead time of the second process,
decreasing it to 4 s: In this case the control parameters
are Kp1 ¼ 1:03; Ti1 ¼ 33; Kp2 ¼ 1:98; Ti2 ¼ 15; g ¼
0:45; Kp ¼ 0:23; Ti ¼ 2:27: Results are shown in
Fig. 16 and the calculated performance indexes are
J ¼ 12:18 for the new approach, J ¼ 22:60 for the Blend
station and J ¼ 20:35 for the standard ratio controller.
 ARTICLE IN PRESS
496 A. Visioli / Control Engineering Practice 13 (2005) 485–497

y1

2.8

y2 Blend station
process outputs

2.6

y standard ratio controller

2
2.4

y proposed ratio controller

2

2.2

0 20 40 60 80 100 120 140 160 180

time [s]

Fig. 15. Process outputs for the first experiment.

3
y Blend station
2

2.8
process outputs

2.6

y
1

2.4
y proposed ratio controller
2

2.2

y standard ratio controller

2

0 20 40 60 80 100 120 140 160 180

time [s]

Fig. 16. Process outputs for the second experiment.

By comparing the obtained results, it appears that making 7. Conclusions

the second loop dynamics faster yields, as expected, to a
better performance obtained by the standard controller. In In this paper a new ratio control structure has been
any case, the new ratio controller provides again the best proposed. An automatic tuning procedure has been
performances and the difference with the original Blend devised so that no tuning effort from the user is needed.
station is more evident in this case. The methodology is easy to implement (note that no
 ARTICLE IN PRESS
A. Visioli / Control Engineering Practice 13 (2005) 485–497
extra measurements are required with respect to
the standard ratio controllers) and it is based on the
use of classical PI controllers, so that it can be
easily understood by operators, who retain their
know-how. Results show its effectiveness for a wide
range of processes. Thus, the overall methodology
appears to be suitable to be implemented in DCS as
well as in single-station controllers for use in the
industrial context.
497
References
( om,
Astr . K., & H.agglund, T. (1995). PID controllers: Theory, design
and tuning. Research Triangle Park: ISA Press.
H.agglund, T. (2001). The Blend station—a new ratio control structure.
Control Engineering Practice, 9(11), 1215–1220.
Shinskey, F. G. (1996). Process control systems—application, design,
and tuning. USA: McGraw-Hill.
Skogestad, S. (2003). Simple analytic rules for model reduction and
PID controller tuning. Journal of Process Control, 13, 291–309.