Counter-current tubular heat exchanger: Modeling and

adaptive predictive functional control

M.A. Arbaoui, L. Vernie`res-Hassimi, D. Seguin, M.A. Abdelghani-Idrissi
*
Centre de De veloppement Durable, Universite de Rouen, IUT, Rue Lavoisier, 76821 Mont Saint Aignan Cedex, France
Received 6 October 2005; accepted 22 December 2006
Available online 26 January 2007
Abstract
The control of the outlet temperature of a counter-current tubular heat exchanger in heater conguration with the predictive func-
tional control is presented in this paper. The outlet temperature of the cold uid is controlled by variation of the ow of the hot uid
while the inlet temperatures corresponding to the principal inputs are maintained constant. An approximated rst order model, corre-
sponding to the response of the heat exchanger to a step change of the ow rate is used to apply the functional predictive control. The
gain and the time-constant of this model depend on the initial and nal steady state temperatures according to the ow rates. This non-
linear dynamic model, obtained from the partial dierential equations (PDE) is taken into account to apply the functional predictive
control, which was validated experimentally in various congurations. The robustness of this controller is also examined when the system
is subjected to the sudden change of the ow rate of the cold uid.
2007 Elsevier Ltd. All rights reserved.
Keywords: Tubular heat exchange; Heater; Temperature control; Model-based predictive control
1. Introduction
In much of industrial thermal applications, the heat
exchanger constitutes important equipment for the control
of the temperatures and the heat exchange. This thermal
system is generally controlled by the ow rate correspond-
ing to a parameter and not by the inlet temperatures corre-
sponding to the principal inputs of this one. The nonlinear
eects must consequently be taken into account to control
this kind of system. A simple controller of type PI (Propor-
tional-Integral), tuned by hand, is frequently adequate to
control the outlet temperature. But if the tubes are rather
long, the tubular counter-current heat exchanger presents
an important lag time corresponding to the residence time
when the inlet temperature changes. In this case, control-
lers PID (Proportional-Integral-Derivative) lose their eec-
tiveness. The parameters of the model involving the various
temperatures change with the ow rates of the uids and
the adjustment of the controller should change with the u-
ids ow rates, which could be a tiresome work. Model-
based Predictive Control MPC gives an alternative for
the eective control of the nonlinear processes such as
the thermal equipment. In the case of the heat exchanger,
if the model of the controller is obtained from the rst prin-
ciples and if its parameters are renewed in real time by real
measurements of the ow rates and the temperatures, a
permanent natural adaptation of the controller with the
environment is feasible. It should be noted that a black
box model type, where the parameters do not have any
physical signicance, could not be used because of the
strong nonlinearities. Simplied models based on enthalpy
and mass balances are used, insofar as they were identi-
ed by real data in order to have a good predictability.
Based on a set of physical equations with a clear physical
1359-4311/$ - see front matter 2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.applthermaleng.2006.12.027
*
Corresponding author. Tel.: +33 235 146 369; fax: +33 235 146 263.
E-mail address: Ahmed.Abdelghani@univ-rouen.fr (M.A. Abdelghani-
Idrissi).
www.elsevier.com/locate/apthermeng
Applied Thermal Engineering 27 (2007) 23322338
signicance, they can be used for on-line adaptation. In this
case, MPC allows the explicit integration of phenomeno-
logical process models in the system control.
Moreover, this kind of systems is governed by the par-
tial derivative equations (PDE) when a local modeling is
employed [1,2]. A model obtained by rst order approxi-
mation describing the response of the temperature along
the heat exchanger to a step change of ow rate is proposed
in Refs. [35]. This model is derived from the system of
PDE describing the transport and the transfer of heat by
convection. This dynamic model can be employed to apply
the control of this system. The good control eectiveness of
these systems makes necessary the use of the advanced con-
trol techniques based on dynamic internal models. Many
researchers have shown the feasibility of the predictive con-
trol in the industrial applications. Richalet et al. [6] have
published the rst paper on the IDCOM and the applica-
tion to the petrochemical area. The rst comprehensive
exposition of generalized predictive control (GPC) was pre-
sented by Clarke et al. [7,8]. Predictive control of chemical
reactors has been the subject of several studies [912]. In
Ref. [13], a model-based predictive controller was coupled
to particle swarm optimisation algorithm to control a
greenhouse air temperature. Richalet [14,15] also devel-
oped the predictive functional control (PFC), which will
be used and presented in this paper. Morari and Lee [16]
gave an overview of the origins of model predictive control
and its popularity. The robustness of the model predictive
control was investigated and applied to the temperature
control of an open-loop unstable continuous stirred tank
reactor [17]. The present paper deals with the predic-
tive functional control applied to a counter-current heat
exchanger. The PFC is based on an internal adaptive model
corresponding to the response to ow rate used in the algo-
rithm command. The principle of predictive functional
control is also recalled.
2. Description and modeling
The system studied in this work corresponds to a water/
water counter-current tubular heat exchanger as described
on Fig. 1. The inner pipe is a copper tube and the outer one
is a stainless steel tube. The hot uid crosses the circular
duct and the cold uid circulates in the annular duct. The
thermocouple probes are placed at the inlet and the outlet
opposite sides of the tubular heat exchanger. The ow rates
of the two uids are controlled using two valves. The
probes and the valves are connected to a data acquisition
system in order to monitor the data and to adjust the
two uids ow rates. In this paper, the heat exchanger is
used in a heater conguration. The controlled variable is
the cold uid outlet temperature and the manipulated var-
iable is the hot uid ow rate. The geometrical and physical
parameters of the heat exchanger are reported in Table 1.
Patankar et al. [18] proposed the mathematical model
used in this study. The assumptions made in our case are:
Fluids are in turbulent ow.
Heat conduction along the ow axis is neglected.
Fluids are incompressible and single phased.
Nomenclature
a wall thickness [m]
A heat transfer area [m
2
]
Bi Biot number
C
p
specic heat [J kg
1
K
1
]
D diameter [m]
h heat transfer coecient [W m
2
K
1
]
H horizon [s]
J quadratic criterion
K coupling coecient [s
1
]
L heat exchanger length [m]
Q mass ow rate [kg s
1
]
t time [s]
T temperature [K]
V mean velocity [m s
1
]
Vol volume [m
3
]
x axial position [m]
Greek symbols
b dimensionless parameter
e error
/ ux [W]
k thermal conductivity [W m
1
K
1
]
l parameter
q density [kg m
3
]
s time constant [s]
v parameter
Subscripts
B base
c cold uid
e outer tube
h hot uid
i inner tube
in input stream
m model
out output stream
p process
r reference
set set point
w separating wall
Superscripts
0 Initial conditions
1 nal conditions
M.A. Arbaoui et al. / Applied Thermal Engineering 27 (2007) 23322338 2333
Thermophysical properties of the uids are assumed to
be constant.
Separating wall is assumed isothermal along the radial
axis. The Biot numbers of both uid sides Bi
h;c

h
h;c
a
kw
corresponding to experimental conditions are lower
than 0.01.
In order to obtain dierential equations of the model, the
heat exchanger is subdivided in several elemental volumes
with length dx as shown in Fig. 2. While owing through
the elemental volume, the hot uid transfers heat to the wall
by convection resulting in reduction in its outlet enthalpy
and internal stored thermal energy. Energy balance applied
to a dierential volume of hot uid leads to the rst equa-
tion of system (1) after simplication and rearrangement.
Similar energy balances on cold uid and separating wall
gives the second and the third equations of system (1).
q
h
Vol
h
C
p
h
oT
h

ot
Q
h
C
p
h
L
oT
h
ox
h
h
A
h
T
w
T
h

q
c
Vol
c
C
p
c
oT
c
ot
Q
c
C
p
c
L
oT
c
ox
h
c
A
c
T
w
T
c

q
w
Vol
w
C
p
w
oT
w
ot
h
h
A
h
T
h
T
w
h
c
A
c
T
c
T
w

_

_
1
The system (1) can be written in the following form accord-
ing to the mean velocities of the two uids:
oT
h
ot
V
h
oT
h
ox
K
1
T
w
T
h

oT
c
ot
V
c
oT
c
ox
K
2
T
w
T
c

oT
w
ot
K
3
T
h
T
w
K
4
T
c
T
w

_

_
2
The parameters K
1
, K
2
, K
3
, K
4
are dened as:
K
1

4h
h
q
h
C
p
h
D
i
; K
2

4h
c
D
i
q
c
C
p
c
D
2
e
D
2
i

;
K
3

h
h
q
w
a
i
C
p
w
; K
4

h
c
q
w
a
i
C
p
w
: 3
The boundary conditions are
T
h
x L T
h;in
and T
c
x 0 T
c;in
The temperatures temporal evolution can be approximated
by an exponential response like a rst order system:
T
h
x; t T
0
h
x T
1
h
x T
0
h
x
_ _
1 exp
t
s
h
x
_ _ _ _
T
c
x; t T
0
c
x T
1
c
x T
0
c
x
_ _
1 exp
t
s
c
x
_ _ _ _
T
w
x; t T
0
w
x T
1
w
x T
0
w
x
_ _
1 exp
t
s
w
x
_ _ _ _
_

_
4
when V
h
(t) = V
h0
+ DV
h
U(t).
acquisition card
multiplexer card
Computer
system
stainless steel
copper
V1
cold fluid
hot fluid
cold fluid
hot fluid
V2
Fig. 1. Counter-current tubular heat exchanger with data acquisition system.
Table 1
Geometrical and physical parameters of the heat exchanger
L (m) D (m) a (m) q (kg/m
3
) C
p
(J/kg K)
Inner 4.5 0.02 0.001 8900 394
Outer 4.5 0.04 0.003 7850 490
d
h
= h
h
(T
w
-T
h
) dA
h
d
c
= h
c
(T
w
-T
h
) dA
c
h
d
c
d
c c c
.
T Cp m
( ) dx
x
T
T Cp m
c
c c c
.

+
wall
h h h
.
T Cp m
( ) dx
x
T
T Cp m
h
h h h
.

+
t
T
Cp
L
x d
Vol
c
c c c

t
T
Cp
L
dx
Vol
w
w w w

t
T
Cp
L
dx
Vol
h
h h h

x x +dx
Fig. 2. Elementary energy balance taking into account convective heat
transfer and counter-current ow.
2334 M.A. Arbaoui et al. / Applied Thermal Engineering 27 (2007) 23322338
U(t) is the Heaviside function. In relations (4) appear the
variations of the time constants s
h
(x), s
c
(x) and s
w
(x) as
function of the space. The analytical expressions of time
constants as function of space are proposed in Refs. [3,5].
The important results of this analysis are used in this paper.
The uid not submitted to step presents two situations. The
rst case corresponds to a uniform time constant along the
heat exchanger. The second case is relative to an increasing
linear function according to the ow axis. Note that the
time constants, as illustrated in Refs. [3,5], are function
of the ow rate magnitude. Therefore, the model is adapt-
able in real time when the exchanger of heat is controlled.
3. Predictive functional control
As discussed before, the predictive control has been
applied successfully in many industrial applications [15].
The principle of this method is depicted in Fig. 3. It is
based on the following elements:
A dynamic model T
m
(n) to predict the system behaviour
in the future.
A reference trajectory T
r
(n), which consists in bringing
the controlled variables up to the future set-point T
set
(n)
within a prediction horizon H
p
more slowly, in order to
avoid overshoot.
A quadratic performance objective with a criterion
J(Q
h
, e) as a function of the future controller error
e(n) between reference trajectory and predicted output
over a coincidence horizon [H
1
, H
c
].
Note that the rst element of this technique is an impor-
tant point, which can aect the performance when the
model poorly represents the process. A large eort has to
be undertaken in this area before implementing the PFC.
The reference trajectory can also be considered as the
desired response of the closed loop. Particularly, the PFC
considers a reference trajectory of rst order. For a given
set-point prole, the objective criterion minimisation leads
to optimal values of the manipulated variable in the future
for the coincidence horizon that gives the closest predicted
target variables by the dynamic model to the reference tra-
jectory values.
In the PFC algorithm, the manipulated variable is con-
stituted of a superposition of a set of basis functions Q
Bk
:
Qn i

k
l
k
Q
Bk
i 5
In this case, the only parameters l
k
have to be evaluated;
thus, the calculation time is reduced. The basis functions
are generally polynomial and are:
Q
Bk
i i
k
Note that there is no limit to select them; they are taken
according to the form of the desired controlled variable.
The increment between the actual model output and the
reference trajectory at the point of coincidence in the future
is:
DH T
r
n H T
p
n 6
with
T
set
n i T
r
n i v
i
T
set
n T
p
n 7
and
v e
DT=TBRF=3
: 8
Taking into account the coincidence of the model output
and the reference trajectory, DH can be written in the fol-
lowing form:
DH T
set
n T
p
n1 k
H
: 9
Considering the forced response of the model and its
homogenous solution, the increment can be transformed
in the following expression:
DH
i
T
p;x
n H
i
T
p
n

k
l
k
Q
Bk
i 10
where T
p,x
represents the forced response.
Note that each output function T
Bk
corresponds to an
input function Q
Bk
. If the set-point prole is characterised
by dierent steps then only the rst function Q
B1
has to be
considered. The response to this function is:
T
B1
K1 b
H
with b e
DT=s
In this case, the law command is given by the following
expression:
T
set
n T
p
n1 k
H
l
1
T
B1
H b
H
T
p
n T
m
n
11
Using this algorithm to control the outlet temperature of
cold uid with the ow rate of hot uid, which can be
extracted from (11) and is given as follow:
Q
h
n
T
set
n T
p
n1 k
H

K1 b
H

T
m
n
K
12
present past
Q(i)
T
p
(i)
T(i) r
T
set
(i)
n-1 n+H
c
n +H
1
n t
reference
trajectory
predicted
target
manipulated
variable
Set point
future
Fig. 3. Predictive control principle.
M.A. Arbaoui et al. / Applied Thermal Engineering 27 (2007) 23322338 2335
This leads to the functional diagram shown on Fig. 4. In
this diagram, the gain K and the time constant s are func-
tion of the hot uid ow rate, which constitutes the system
input. The model parameters are also function of the
amplitude of the ow rate step. These statements were de-
scribed in the previous section. The term a is also a func-
tion of the ow rate and the PFC algorithm used in this
work adapts these parameters according to ow rate in real
time.
4. Results and discussion
The adaptive PFC algorithm is implemented on a com-
puter in order to test experimentally the control of outlet
temperature of product by varying the ow rate of hot uid
circulating through the inner tube. Fig. 5 shows the tempo-
ral evolution of experimental outlet temperature for dier-
ent steps of set point. In this gure, the error is also
reported.
As shown on this gure, the PFC gives good results and
the outlet temperature of the heat exchanger tends towards
the set point for various congurations. This is illustrated
by the curve corresponding to the temporal evolution of
the error of process, which converges towards zero for each
change of the temperature set point. It should be noted
that the transient response to the positive and negative
step changes are dierent. Indeed, the time-constant is
dierent for positive and negative step change of the ow
rate.
Fig. 6 shows the temporal variation of the ow rate,
which represents the manipulated variable, in order to
adjust the outlet temperature of cold uid towards the tem-
perature set point. Fig. 7 illustrates the variation of time
constant and gain as function of time. As discussed previ-
ously, the time-constant and the gain are functions of the
amplitude of the ow rate step change and depend on the
initial value of this one. The curves of this gure, obtained
with the same experimental conditions of Fig. 5, show the
signicant variation of the gain and of the time constant.
The parameters of the model are adapted in real time
according to the ow rate and its step change amplitude
to predict the future in the horizon of coincidence. This g-
ure shows the adaptability of this technique for the control
of the outlet temperature of the heat exchanger by the eval-
uation of the parameters of the model according to the
uid ow rate.
The previous experiment is achieved without perturba-
tions. Indeed, the inlet temperatures and the ow of the
other uid were maintained constant.
The ow rate of the product could be subjected to an
unexpected variation that aects the functioning of the
controller. Fig. 8 illustrates the response of the exchanger
1
1

H
H
K ( )
1
K
K z
z
( ) 1
1
1
1

T
P,out
T
m
Qh
+
+
+
-
Tset
Heat
exchanger
Fig. 4. Functional diagram of temperature control.
10
15
20
25
30
35
0 50 100 150 200 250 300
-10
-5
0
5
10
15
20
t(s)
t
e
m
p
e
r
a
t
u
r
e

(

C
)
e
r
r
o
r

(

C
)
Th,in = 63.8 C Tc,in = 12.6 C Qc = 0.27 kg/s
set point
output
Fig. 5. Closed loop temperature response for dierent set point steps.
0
10
20
30
40
50
60
70
0 50 100 150 200 250 300
t(s)
1
0
0

x

Q
h
(
k
g
/
s
)
Th,in = 63.8 C Tc,in = 12.6 C Qc = 0.27 kg/s
Fig. 6. Temporal evolution of manipulated variable.
8.0
8.5
9.0
9.5
10.0
10.5
11.0
0 50 100 150 200 250 300
40
50
60
70
80
90
100
110
120
G
a
i
n

[

C

/

(
k
g
/
s
)
]

t(s)

(
s
)
Fig. 7. Temporal evolution of time constant and gain.
2336 M.A. Arbaoui et al. / Applied Thermal Engineering 27 (2007) 23322338
of heat in closed loop when a sudden change appears on
the ow rate of the cold uid at t = 350 s. In spite of these
sudden variations, the controller responds and adjusts the
outlet temperature toward the set point in real time. This
reaction is shown on the Fig. 9, which represents the evo-
lution in the time of the ow rates of the hot and cold uid
obtained under the same conditions as Fig. 7. The control-
ler ensures a follow-up of the set point prole by rejecting
this disturbance. By reducing the dierence between the
temperature set point and the measured outlet temperature
by the controller, the error converges towards zero as indi-
cated on Fig. 7. The algorithm adjusts the manipulated var-
iable with respect to the change of the desired value and
rejects perturbations aecting the product ow rate as indi-
cated in Fig. 8.
It is important to note that the change of cold uid ow
rate aects the model parameters. Indeed, Fig. 10 shows
the evolution of the gain and time constant of the heat
exchanger as function of time.
The PFC also gives satisfactory results when the
inlet temperature of the cold uid presents uctuations.
The same results are obtained when both inlet tempera-
tures are subjected to variations. The controller rejects
the inuence of these disturbances in closed loop.
Moreover, the algorithm of PFC takes into account the
inuence of the various parameters to predict correctly
the future by adjusting the model in order to track the
set points.
5. Conclusion
The functional predictive control was applied to a coun-
ter-current heat exchanger. The outlet temperature of the
cold uid circulating in annular duct is controlled by the
ow rate of the hot uid circulating through the inner duct.
The dynamic model associated with the PFC in this study is
an approximation of the transient response to the variation
of ow rate by a rst order system obtained from the PDE
describing the heat exchanger. This model takes into
account the evolution of the gain and the time-constant
according to the ow rate of the hot uid. The PFC algo-
rithm takes account of variations of the model parameters
in real time and control the outlet temperature of the heat
exchanger. The eect of the unexpected variations of the
ow rate of the cold uid is studied. The controller rejects
this disturbance and maintains the outlet temperature of
the cold uid to the desired values.
References
[1] F. Bagui, M.A. Abdelghani-Idrissi, L. Estel, C. Thenard, Estimation
detat repartie dun echangeur de chaleur a` contre courant, Recent
Prog. Genie Proc. 10 (49) (1996) 235240.
[2] L. Estel, F. Bagui, M.A. Abdelghani-Idrissi, C. Thenard, Distributed
state estimation of a counter current heat exchanger under varying
ow rate, Compute. Chem. Eng. 24 (1) (2000) 5360.
[3] M.A. Abdelghani-Idrissi, F. Bagui, L. Estel, Analytical and Exper-
imental response time to ow rate step along a counter current ow
double pipe heat exchanger, Int. J. Heat Mass Transfer 44 (2001)
37213730.
[4] M.A. Abdelghani-Idrissi, F. Bagui, Counter-current double-pipe heat
exchanger subjected to ow-rate step change, Part I: New steady state
formulation, Heat Transfer Eng. 23 (5) (2002) 411.
[5] M.A. Abdelghani-Idrissi, F. Bagui, L. Estel, Counter-current double-
pipe heat exchanger subjected to ow-rate step change, Part II:
Analytical and experimental transient response, Heat Transfer Eng.
23 (5) (2002) 1224.
[6] J. Richalet, A. Rault, J.L. Testud, J. Papon, Model predictive
heuristic control-application to industrial processes, Automatica 14
(1978) 413428.
0
20
40
60
80
100
120
140
0 100 200 300 400 500
7
8
9
10
11
12
G
a
i
n

[

C

/

(
k
g
/
c
)
]

(
s
)

t (s)
Fig. 10. Gain and time constant as function of time with a perturbation
on cold uid ow rate.
10
15
20
25
30
35
0 100 200 300 400 500
-10
-5
0
5
10
15
20
t
e
m
p
e
r
a
t
u
r
e

(

C
)
e
r
r
o
r

(

C
)
t (s)
Th,in=59.0 C
Tc,in=12.0 C
outpu t
se t point
Fig. 8. Reaction to change of cold uid ow rate.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0 100 200 300 400 500
0.2
0.22
0.24
0.26
0.28
0.3
0.32
0.34
Q
h

(
k
g
/
s
)
Q
c

(
k
g
/
s
)
t (s)
Fig. 9. Flow rate of both uids versus time.
M.A. Arbaoui et al. / Applied Thermal Engineering 27 (2007) 23322338 2337
[7] D.W. Clarke, C. Mohtadi, P.S. Tus, Generalized predictive control,
Part I: the basic algorithm, Automatica 23 (1987) 137148.
[8] D.W. Clarke, C. Mohtadi, P.S. Tus, Generalized predictive control,
Part II: extentions and interpretations, Automatica 23 (1987) 149
160.
[9] G. O

zkan, H. Hapoglu, M. Alpbaz, Non-linear generalised predictive

control of a jacketed well mixed tank as applied to a batch process-A
polymerisation reaction, Appl. Therm. Eng. 26 (7) (2006) 720726.
[10] F. Xaumier, M.-V. Le Lann, M. Cabassud, G. Casamatta, Experi-
mental application of nonlinear model predictive control: temperature
control of an industrial semi-batch pilot-plant reactor, J Process
Control 12 (6) (2002) 687693.
[11] K. Preu, M.-V. Le Lann, J. Proth, H. Pingaud, Modelling and
predictive control of fed-batch yeast growth on industrial pilot plant
scale, Chem. Eng. J 78 (1) (2000) 5359.
[12] K. Preu, M.-V. Le Lann, M. Cabassud, G. Anne-Archard, Imple-
mentation procedure of an advanced supervisory and control strategy
in the pharmaceutical industry, Control Eng. Pract. 11 (12) (2003)
14491458.
[13] J.P. Coelho, P.B. de Moura Oliveira, J. Boaventura Cunha, Green-
house air temperature predictive control using the particle swarm
optimisation algorithm, Compute. Elect. Agric. 49 (3) (2005) 330344.
[14] J. Richalet, Pratique de la Commande Predictive, Herme`s, Paris,
1993.
[15] J. Richalet, Industrial applications of model based predictive control,
Automatica 29 (1993) 12511274.
[16] M. Morari, J.H. Lee, Model predictive control: past, present and
future, Compute. Chem. Eng. 23 (1999) 667682.
[17] G. Pannocchia, Robust model predictive control with guaranteed
setpoint tracking, J. Process Control 14 (8) (2004) 927937.
[18] S.V. Patankar, D.B. Spalding, in: N. Afgan, Schlinder (Eds.), Heat
Exchangers, McGraw-Hill, New York, 1974, pp. 155175.
2338 M.A. Arbaoui et al. / Applied Thermal Engineering 27 (2007) 23322338