Accurate Low-Order Dynamic Model of A Compact Plate Heat Exchanger

This document contains a post-print version of the paper
Accurate low-order dynamic model of a compact plate heat exchanger

authored by A. Michel and A. Kugi
and published in International Journal of Heat and Mass Transfer.
The content of this post-print version is identical to the published paper but without the publisher’s final layout or
copy editing. Please, scroll down for the article.
Cite this article as:

A. Michel and A. Kugi, “Accurate low-order dynamic model of a compact plate heat exchanger”, International Journal
of Heat and Mass Transfer, vol. 61, no. 33, pp. 323–331, 2013. doi: 10.1016/j.ijheatmasstransfer.2013.01.072
BibTex entry:
@ARTICLE{MichelIJHAMT2013,
author = {Michel, A. and Kugi, A.},
title = {Accurate low-order dynamic model of a compact plate heat exchanger},
journal = {International Journal of Heat and Mass Transfer},
year = {2013},
volume = {61},
number = {33},
pages = {323-331},
doi = {10.1016/j.ijheatmasstransfer.2013.01.072},
url = {http://www.sciencedirect.com/science/article/pii/S0017931013001038}
}
Link to original paper:

http://dx.doi.org/10.1016/j.ijheatmasstransfer.2013.01.072
http://www.sciencedirect.com/science/article/pii/S0017931013001038
Read more ACIN papers or get this document:

http://www.acin.tuwien.ac.at/literature
Contact:
Automation and Control Institute (ACIN) Internet: www.acin.tuwien.ac.at
Vienna University of Technology E-mail: [email protected]
Gusshausstrasse 27-29/E376 Phone: +43 1 58801 37601
1040 Vienna, Austria Fax: +43 1 58801 37699
Copyright notice:
This is the authors’ version of a work that was accepted for publication in International Journal of Heat and Mass Transfer. Changes
resulting from the publishing process, such as peer review, editing, corrections, structural formatting, and other quality control mechanisms
may not be reflected in this document. Changes may have been made to this work since it was submitted for publication. A definitive
version was subsequently published in A. Michel and A. Kugi, “Accurate low-order dynamic model of a compact plate heat exchanger”,
International Journal of Heat and Mass Transfer, vol. 61, no. 33, pp. 323–331, 2013. doi: 10.1016/j.ijheatmasstransfer.2013.01.072
Accurate low-order dynamic model of a compact plate heat exchanger
Alexander Michel∗, Andreas Kugi

Automation and Control Institute (ACIN), Vienna University of Technology, Gußhausstraße 27-29 / E376, 1040 Vienna, Austria
Abstract
This paper deals with the derivation of a low-dimensional
mathematical model of a compact plate heat exchanger
capturing the significant nonlinearities and the essential
dynamic behavior in an accurate way. Thereby, the model
is based on the basic laws of thermodynamics and the
similitude theory of Nußelt. It is shown that reasonable
simplifications according to the specific design and the typ-
ical operating conditions of compact plate heat exchangers
together with a semi-discretization of the spatial domain
by means of the finite volume method provides a compact
finite-dimensional approximation of the underlying partial
differential equations (pdes). In this context, two differ-
ent interpolation schemes of the finite volume method are
compared, i.e. a classical upwind scheme and a new con-
cept based on an approximate stationary solution of the
underlying pdes. The latter ensures high accuracy even for
very low-order discretizations, which is shown by means of
simulation and measurement results.
Keywords: low-order dynamic model, compact plate heat exchanger, finite volume method, global power balance
1. Introduction
Compact plate heat exchangers like small sized brazed

plate heat exchangers are increasingly used in the field of
district heating, heat recovery, industrial process cooling
and heating, hydraulic oil cooling or cooling of machine
tools. Their advantages are high heat transfer rates, a
Figure 1: Design of a 1/1 pass countercurrent plate heat exchanger.
small overall size, a high resistance against fouling, high
working pressures, a simple design and thus low costs in
production. The working principle of all plate heat ex-
changers is the same, namely heat is exchanged between a desired outlet temperature under dynamically changing
two fluid circuits divided by plates. Thereby a 1/1 pass operating conditions, a model-based control design, which
flow arrangement, where the fluid paths of the two circuits exploits the nonlinear structure of the heat exchanger
are arranged in an alternating manner, is commonly used model, has the potential to outperform the traditionally
in industry [1], see Figure 1. In order to accurately track used linear control strategies. A prerequisite for the appli-
cation of model-based control strategies is the derivation
of an accurate low-dimensional mathematical model as a
∗ Corresponding author: Tel.: +43 1 58801 376 21, fax: +43 1
design model. Otherwise the resulting control strategies
58801 376 99
Email addresses: [email protected] (Alexander
may get computationally too expensive to be implemented
Michel), [email protected] (Andreas Kugi) in real-time on a standard industrial process control unit.
URL: http://www.acin.tuwien.ac.at (Alexander Michel) Therefore, we aim at developing a mathematical model of
Postprint submitted to International Journal of Heat and Mass Transfer
Post-print version of the article: A. Michel and A. Kugi, “Accurate low-order dynamic model of a compact plate heat exchanger”, Inter-
national Journal of Heat and Mass Transfer, vol. 61, no. 33, pp. 323–331, 2013. doi: 10.1016/j.ijheatmasstransfer.2013.01.072
The content of this post-print version is identical to the published paper but without the publisher’s final layout or copy editing.
Nomenclature
Greek Symbols
Ac cross section Ac = BH [m2 ] α abbreviation (see Eq. (21)) [1/s]
A, B, C, D system matrices of system Eq. (29) β abbreviation (see Eq. (21)) [1/s]
B channel width [m] Γ abbreviation (see Eq. (35))
CNu correlation parameter γ abbreviation (see Eqs. (33) and (34))
cp specific heat capacity [J/(kg K)] δϑ thickness of the thermal boundary layer [m]
dh hydraulic diameter dh = 2H [m] µ dynamic viscosity [kg/(m s)]
H channel heigth [m] ϑ temperature [K]
h heat transfer coefficient [W/(m2 K)] θ temperature mean value of a volume element [K]
Hp plate heigth [m] θ state vector of system Eq. (29)
k thermal conductivity [W/(m K)] ϑin input vector of system Eq. (29)
L channel length [m] ϑout output vector of system Eq. (29)
M number of volume elements ξ linear constant parameter (see Eq. (39))
m mass [kg] % mass density [kg/m3 ]
ṁ mass flow rate [kg/s]
Nc number of channels Subscripts
Np number of plates 1, 2, 3 direction x1 , x2 or x3
nPr correlation parameter I side I of the heat exchanger
nRe correlation parameter II side II of the heat exchanger
Nu Nußelt number Nu = hdh /k f fluid
Pr Prandtl number Pr = µcp /k p plate
q̇ heat flux density [W/m] s stationary solution
Re Reynolds number Re = |u|%dh /µ
t time [s] Superscripts
U overall heat transfer coefficient [W/(m2 K)] c calculated value
u flow velocity [m/s] in inlet
w weighting factor m measured value
x coordinate [m] out outlet
– (overbar) average value
∞ core flow
a plate heat exchanger, which is low-dimensional, easy to ear dynamic behavior. In this context, the specific design
parameterize and accurate enough to capture the essential and the typical operating conditions of compact plate heat
dynamics and nonlinearities. exchangers remain mostly disregarded.
Several mathematical models have been reported in the In this paper, an accurate low-order dynamic model
literature. Usually, these models consider one spatial di- of a 1/1 pass compact plate heat exchanger is derived
mension and they differ in the level of detail, especially if in a systematic way. Thereby, the mathematical model
maldistribution of the fluid flow or the heat capacity of the is systematically simplified by exploiting the specific de-
plates are taken into account. Moreover, different meth- sign and the typical operating conditions of compact plate
ods for solving the underlying partial differential equations heat exchangers. Starting with the basic equations of heat
(pdes) are proposed. In [2] a mathematical model for arbi- transfer all assumptions and simplifications being made
trary flow patterns is given, where the pdes are solved by are thoroughly explained in order to clarify the validity
means of the frequency response in the Laplace domain. range of the resulting low-order dynamic model. The pa-
Further models utilizing the Laplace transform are pre- per is organized as follows: In Section 2, the physical
sented in [3, 4, 5]. An alternative approach is suggested in fundamentals of heat exchanger design are summarized.
[6], where Galerkin’s method is employed to solve the pdes. Based on the special design and the typical operating con-
In addition, [7, 8, 9] provide mathematical models based ditions of compact plate heat exchangers, in Section 3,
on the finite volume method which is commonly used for specific simplifications are made to deduce a distributed-
numerical heat transfer problems. parameter model consisting of three pdes. Furthermore,
Although several mathematical models are available in a semi-discretization of the spatial domain is performed
the literature, most of them are either high-dimensional by means of the finite volume method which gives rise
or give a rather inaccurate approximation of the nonlin- to a compact low-order model of only three ordinary dif-
ii
with the velocity field ui (x1 , x2 , x3 , t), i = 1, . . . , 3 and
ferential equations. Thereby, two different interpolation suitable boundary and initial conditions. The left hand
schemes are used, a classical upwind scheme and a new side of Eq. (1) describes the convective heat flux density.
concept based on the approximate stationary global power Herein % denotes the mass density and cp is the specific
balance. Finally, simulation and measurement results for heat capacity of the fluid. Moreover, the right hand side of
different industrial compact plate heat exchangers demon- Eq. (1) characterizes the heat transport over the boundary
strate the accuracy of the proposed model. The last sec- of V .
tion, Section 5, contains some conclusions.
2.1. Temperature field of a channel
2. Physics of heat exchanger modeling
In order to calculate the temperature field
Figure 1 shows the basic design of the considered 1/1 ϑf (x1 , x2 , x3 , t) inside the volume V , the velocity
pass compact plate heat exchanger. The fluid flow pas- field ui (x1 , x2 , x3 , t) has to be known. An accurate model
V of the velocity field yields a nonlinear system of pdes de-
pending on the exact geometry of the plates. This entails
x3 high-order models which cannot be used for controller
ϑin , u1 q̇ H design. However, the assumption of a turbulent fluid flow
x1 justifies the use of a plug flow model which divides the flow
x2
in a core flow with a constant flow velocity in x1 -direction
1 (t) and a velocity boundary layer [14]. This plug
L u∞
Figure 2: Channel with a simplified geometry.
flow directly influences the evolving temperature field ϑf ,
which again is assumed to be composed of a thermal core
sages are formed by gaps between two adjacent plates and a boundary layer, see Figure 3, whereby the thickness
which have typically a chevron or herringbone corruga- of the thermal boundary layer δϑ is negligible compared
tion pattern [10]. In the following, typical assumptions in to the thickness of the thermal core. Inside the thermal
heat exchanger analysis are supposed [7, 11, 12]: core, due to the high turbulences of the fluid flow, the
temperature can be considered homogeneous, that is
1. The increase of temperature due to friction is negligi-
ble.
ϑf (x1 , x2 , x3 , t) = ϑ∞
f (x1 , t) (2)
2. The fluid is assumed to be incompressible.
3. There are neither heat sources nor heat sinks inside a
and
channel.
4. The heat exchange with the environment is negligible. ∂ϑf ∂ϑf
5. The herringbone corrugation patterns induce high = =0. (3)
∂x2 ∂x3
vorticities and turbulences in the fluids even for low
Reynolds numbers. Thus, integrating Eq. (1) over the cross section Ac = BH
6. The fluid velocity in the main flow direction is high we get
enough to justify that heat conduction along the flow !
Z BZ H 3
direction can be neglected compared to heat convec- ∂ϑf X ∂ϑf
tion. %cp + ui dx2 dx3
0 0 ∂t ∂xi
7. No phase changes of the fluid occur inside the heat Z BZ HX
i=1
(4)
3
exchanger. ∂ q̇i
=− dx2 dx3 .
In a first step, the impact of the velocity field on the tem- 0 0 i=1 ∂xi
perature is discussed before the temperature field of one
channel of the plate heat exchanger is derived. In a sec- Because the thickness of the boundary layer δϑ is consid-
ond step, the thermal coupling between the channels will erably small compared to the height of the channel H, the
be determined. For this, a simplified geometry of a chan- energy stored in the boundary layer will be neglected. Fur-
nel is considered, see Figure 2, where L, B and H refer thermore, the heat fluxes q̇1 , q̇2 can also be considered zero
to the effective values of the channel length, the channel due to the assumptions that the fluid is ideally mixed (cf.
width and the channel height. Eqs. (2) and (3)), no heat exchange takes place with the
Consider a small volume V inside the channel. If environment and that there is no heat conduction along
no phase changes occur and the material parameters x1 -direction. Thus, the mathematical model of the tem-
are constant inside the volume V , the temperature field perature field in a channel of a plate heat exchanger ac-
ϑf (x1 , x2 , x3 , t) inside V can be written as [13] cording to Eq. (4) simplifies to
3
! 3 x3 =H
∂ϑf X ∂ϑf X ∂ q̇i ∂ϑ ∂ϑ
%cp + ui =− , (1) Ac %cp +u = −B q̇3 (5)
∂t i=1
∂xi i=1
∂xi ∂t ∂x1 x3 =0
iii
relies on a suitable approximation of the functional depen-
with the boundary conditions dence of the Nußelt number Nu, on the Prandtl number
( Pr and the Reynolds number Re [14], i.e.,
ϑ(0, t) = ϑin for u ≥ 0
(6) Nu = fNu (Pr, Re) (9)
ϑ(L, t) = ϑin for u < 0
with
and appropriate initial conditions. Note that for the sake
of readability in Eq. (5) and henceforth ϑ∞
f is replaced by hdh µcp |u|%dh
in Nu = , Pr = , Re = , (10)
ϑ and u∞ 1 by u, respectively. Moreover, ϑ denotes the k k µ
inlet temperature in the channel.
where dh = 2H denotes the hydraulic diameter, k is the
fluid thermal conductivity and µ the dynamic viscosity.
2.2. Thermal coupling of neighboring channels
In recent years, several relations for the Nußelt correlation
In order to model the thermal coupling between neigh- Eq. (9) of plate heat exchangers have been reported in the
boring channels, a plate of height Hp exposed to two fluids literature, see, e.g., [15, 16, 17, 18]. It is common practice
of different core temperatures ϑI > ϑII and fluid veloci- to assume that the Nußelt number Nu is constant along
ties uI , uII is considered, see Figure 3. As it is shown in the flow direction where the material parameters k, µ, cp
x3 and % are evaluated at the average temperature of inlet
ϑII ϑin and outlet ϑout . According to [17] the Nußelt number
ϑII for compact plate heat exchangers can be expressed in the
1 form
q̇II uII δϑ,II
hII
ϑp,II Nu = CNu PrnPr RenRe , (11)
1 Hp
2 kp ϑp,m Hp with the constant empirical parameters CNu , nPr and nRe
1 Hp which have to be determined by means of suitable experi-
2 k
p ments. By combining Eq. (10) and Eq. (11) the heat trans-
ϑp,I
1 fer coefficients hI and hII can be calculated which solely
δϑ,I
uI depend on ϑin out in out
I , ϑI , uI and ϑII , ϑII , uII , respectively.
q̇I hI
Analogous to Eq. (1) the temperature field
ϑI
ϑI ϑ ϑp (x1 , x2 , x3 , t) of the plate satisfies the pde
X3
Figure 3: Schematics of the heat transfer between two adjacent chan- ∂ϑp ∂ q̇i
nels. %p cp,p =− (12)
∂t i=1
∂x i
Figure 3, heat is transferred from the core flow of fluid I with appropriate boundary and initial conditions. Hence-
through the boundary layer to the plate and then to fluid forth, in x2 -direction the plate temperature is assumed to
II. A common way to describe this kind of convective heat be homogenous and boundary effects are neglected, i.e.
transfer is to introduce a so-called heat transfer coefficient q̇2 = 0. Moreover, since the height Hp of a typical plate is
h, which allows to calculate the transferred heat flux be- rather small and the thermal conductivity kp of metal is
tween fluid I and the plate by means of Newton’s law of large, the temperature profile in x3 -direction can be con-
cooling in the form sidered in a quasi-stationary manner. Utilizing Fourier’s
law of heat conduction in x1 -direction, see, e.g., [15]
q̇I = hI (ϑI − ϑp,I ) , (7)
∂ϑp
where ϑp,I denotes the temperature of the wall in contact q̇1 = kp (13)
∂x1
with fluid I. In an analogous way the heat flux q̇II reads
as and integrating Eq. (12) over the cross section Ac,p = BHp
of the plate yields
q̇II = hII (ϑp,II − ϑII ) . (8) Z Hp Z Hp 2
∂ϑp ∂ ϑp x3 =Hp
%p cp,p B dx3 = −kp B 2 dx3 − B q̇3 |x3 =0 .
It is well known that an analytical expression for the heat 0 ∂t 0 ∂x 1
transfer coefficient h is only available for very simplistic (14)
scenarios. In the general case the calculation of h requires
an accurate knowledge of the fluid velocity field which in The stationarity assumption in x3 -direction implies (see
turn leads to high-order models with high computational Figure 3)
effort. Therefore, semi-empirical methods like the simili- Z Hp
tude theory according to Nußelt are often employed in the 1 1
ϑp dx3 = (ϑp,I + ϑp,II ) := ϑp,m (15)
design of heat exchangers. Basically, the latter approach Hp 0 2
iv
and thus Eq. (14) can be written in the form been shown for typical compact heat exchangers in [17].
In addition, due to the design with equal plates, (v) the
∂ϑp,m ∂ 2 ϑp,m cross section Ac and (vi) the heat transfer coefficients hp of
%p cp,p Ac,p = − Ac,p kp + B (q̇I − q̇II ) , (16)
∂t ∂x21 every channel passed through by the same fluid are equal,
see Eq. (19). Furthermore, the small overall volumes of
with the boundary conditions (no heat exchange with the both sides (I and II) together with the high flow rates of
environment) the fluids lead to a (vii) low residence time of the fluids in
each channel.
∂ϑp,m ∂ϑp,m
= =0 (17) Due to the same inlet temperatures, fluid flows and geo-
∂x1 x1 =0 ∂x1 x1 =L
metric parameters of each channel of fluid I and II, locally
and suitable initial conditions. Note that ϑp,m is exactly the same temperature distribution evolves in flow direc-
the temperature in the middle of the plate, i.e. for x3 = tion. This suggests to describe the temperature distribu-
Hp /2, and the heat fluxes q̇I and q̇II correspond to Eqs. (7) tion of each channel of the same fluid by only one aver-
and (8). age temperature distribution ϑ̄ and thus reduce the model
In view of the stationary temperature profile in x3 - from 2Np − 1 pdes to only 3 pdes. Henceforth, the index
direction, the surface plate temperatures ϑp,I and ϑp,II can I (II) always refers to parameters of a channel with fluid
be expressed as I (II) and the index p indicates parameters of a plate. In
order to determine the average model, we introduce the ef-
1 Hp fective heat transfer area BL (Np − 2) for both fluids and
ϑp,I = ϑp,m + q̇I , (18a)
2 kp the effective cross section of fluid I and fluid II, given by
1 Hp Ac,I = Nc,I Ac and Ac,II = Nc,II Ac , respectively. Thereby,
ϑp,II = ϑp,m − q̇II . (18b) Nc,I (Nc,II ) denotes the total number of parallel channels
2 kp
with fluid I (II), with Nc,I + Nc,II = Np − 1. Thus, the
Substituting Eq. (18) into Eqs. (7) and (8) and solving mathematical model of the compact plate heat exchanger
with respect to the heat fluxes results in with the local average temperature distribution ϑ̄I and ϑ̄II
of fluid I and II and the local average plate temperature
1 1 1 Hp distribution ϑ̄p results from Eqs. (5), (16) and (19) in the
q̇I = hp,I (ϑI − ϑp,m ) , = + , (19a)
hp,I hI 2 kp form
1 1 1 Hp
q̇II = hp,II (ϑp,m − ϑII ) , = + . (19b) ∂ ϑ̄I ∂ ϑ̄I
hp,II hII 2 kp = − uI − αI ϑ̄I − ϑ̄p (20a)
∂t ∂x1
The thermal equivalent network of Eq. (19) is depicted on ∂ ϑ̄p kp ∂ 2 ϑ̄p
the left hand side of Figure 3. = − 2 − βI ϑ̄p − ϑ̄I − βII ϑ̄p − ϑ̄II
∂t cp,p %p ∂x1
(20b)
3. Compact plate heat exchanger model ∂ ϑ̄II ∂ ϑ̄II
= − uII − αII ϑ̄II − ϑ̄p (20c)
The temperature field of every fluid channel can be mod- ∂t ∂x1
eled by Eq. (5) and the temperature field of every plate with the abbreviations
by Eq. (16). Furthermore, the heat flux q̇3 characterizing
the thermal coupling of the fluid with a plate is given by hp,i Np − 2 2hp,i
αi = , βi = , i ∈ {I, II} ,
Eq. (19). The resulting model of a plate heat exchanger H%i cp,i Nc,i Hp %p cp,p
with Np plates thus consists of 2Np − 1 nonlinear pdes. (21)
In particular for large number of plates this model is not
suitable for model-based control design. However, in the the boundary conditions
derivation of the mathematical model the specific design (
and the typical operating conditions of compact plate heat ϑ̄I (0, t) = ϑin
I for uI ≥ 0
(22a)
exchangers have been disregarded so far. As it will be ϑ̄I (L, t) = ϑin
I for uI < 0
shown in the sequel, the consideration of these conditions (
ϑ̄II (0, t) = ϑin for uII ≥ 0
not only reduces the number of pdes, but also allows for an II
(22b)
accurate low-order approximation of the spatial domain. ϑ̄II (L, t) = ϑinII for uII < 0

In this context the following simplifications are made: ∂ ϑ̄p ∂ ϑ̄p
Due to the small volume of the gatherer, (i) the trans- = =0 (22c)
∂x1 x1 =0 ∂x1 x1 =L
port time as well as the (ii) pressure drop inside the gath-
erer is negligible. This in turn suggest to assume that (iii) and the channel velocities
the inlet temperature and (iv) the fluid velocity of every
channel with the same fluid are equal. The latter implies ṁI ṁII
uI = and uII = . (23)
that no maldistribution of flow occurs which has already %I Ac,I %II Ac,II
Here ṁI and ṁII denote the overall mass flow rates of with i ∈ {I, II} and Eq. (24) takes the form
fluid I and II passing through the heat exchanger. In the
next step the distributed-parameter system of Eq. (20) is dθI |ṁI | out
=− ϑI − ϑin I − αI (θI − θp ) (28a)
spatially discretized in order to obtain a system of ordi- dt mI
nary differential equations (odes). It is common practice dθp
= −βI (θp − θI ) − βII (θp − θII ) (28b)
to approximate hyperbolic pdes as Eq. (20) by means of dt
the finite volume method, see, e.g., [19, 20, 21]. For this, dθII |ṁII | out
=− ϑII − ϑin II − αII (θII − θp ) , (28c)
the particular control volume is partitioned in M equally dt mII
spaced volume elements of length M L
. Then Eq. (20) is
integrated over the whole volume, where the integral is di- where mI = %I Ac,I L and mII = %II Ac,II L denote the to-
vided into a sum of integrals over the partitioned elements, tal mass inside the heat exchanger of fluid I and fluid II,
which results in respectively. The next step is to determine a relationship
between the outlet temperatures ϑout I and ϑout
II and the
L dθj,I
XM
ṁI x1 =j M
L
state variables θI , θII , and θp . It will be shown that the
+ ϑ̄I x1 =(j−1)
dt overall model can be written in the compact form
L
M % I A
(24a)
c,I M
j=1
+
L
αI (θj,I − θj,p ) = 0 d
θ = A(ṁ)θ + B(ṁ)ϑin (29a)
M dt
x1 =j M
ϑout = C(ṁ)θ + D(ṁ)ϑin (29b)
L
L dθj,p
XM
kp ∂ ϑ̄p
+
M dt cp,p %p ∂x1 x1 =(j−1) L
j=1 M (24b) with the state vector θ = [θI , θp , θII ] , the input vectors
T
L L
ṁ = [ṁI , ṁII ] and ϑin = ϑin in T
, the output vector
T
+ βI (θj,p − θj,I ) + βII (θj,p − θj,II ) = 0 I , ϑII
M M
ϑout = [ϑout out
] and appropriate initial conditions.
T
I , ϑ II
X L dθj,II
M
ṁII x1 =j M
L
+ ϑ̄II x1 =(j−1)
dt 3.1. Upwind scheme
L
M % A
II c,II
(24c)
M
j=1
L One of the simplest approximation schemes in the finite
+ αII (θj,II − θj,p ) = 0 . volume method is the so-called upwind scheme, where the
M
outlet temperatures are interpolated in the form [20]
Thereby, the temperature mean values of the different vol-
ume elements ϑout = θi , i ∈ {I, II} . (30)
i
L
jM
Z
θj,i =
M
ϑ̄i dx1 i ∈ {I, II, p} (25) Applying the upwind scheme to Eq. (28) the system ma-
L trices of Eq. (29) read as
(j−1) M
L
 
are introduced as new state variables of the mathematical − |m ṁI |
I
− α I α I 0
 
model. Since not only the three sums in Eqs. (24a), (24b) A= βI − (βI + βII ) βII  ,
and (24c) must vanish, but also its summands, this results 0 αII |ṁII |
− mII − αII
in a finite-dimensional model of 3M ordinary differential  
equations (odes). In general, the choice of the number of
|ṁI |
0
(31)
 mI
 1 0 0
volume elements M has a great impact on the dynamic be- B= 0 0  , C= ,
0 0 1
havior of the resulting finite-dimensional model, especially 0 |ṁII |
m
on the approximation of the transport phenomena. How- II
0 0
ever, as will be seen in Section 4, due to typical operating D= .
0 0
conditions of compact plate heat exchangers with low resi-
dence time and limited dynamics of the inlet temperatures Note that αi and βi , i ∈ {I, II}, depend on the mass flow
the transport phenomena are less dominant. This allows rates in a nonlinear manner, cf. Eqs. (10), (11), (21) and
to model the heat exchanger by means of only three odes. (23). The upwind scheme is quite simple and leads to
For M = 1 the diffusion term in Eq. (24b) a compact mathematical model but has the drawback of
x1 =j M
L rather high approximation errors, as will be seen in Sec-
∂ ϑ̄p tion 4. Of course, the accuracy of the model could be im-
=0 (26)
∂x1 x1 =(j−1) L proved by increasing the number of volume elements M .
M
vanishes due to the boundary conditions of Eq. (22c), the However, this contradicts the desire to derive low-order
average temperatures at the element border in Eqs. (24a) models which serve as a basis for optimization and real-
and (24c) simplify to time nonlinear control. Therefore, an alternative approach
for the interpolation of the outlet temperatures based on
ṁi x1 =L 1
considerations of the stationary behavior of the heat ex-
ϑ̄i x1 =0 = |ṁi | ϑout − ϑin (27)
%i Ac,i %i Ac,i i i
changer will be applied in the following.
vi
3.2. Power balance scheme
For the subsequent derivation of the stationary tem- Remark: For cocurrent flow with ṁI > 0 and ṁII > 0,
perature profiles the heat conduction of the plates in x1 - ϑ̄I,s (0) = ϑin in
I,s and ϑ̄II,s (0) = ϑII,s similar expressions can
direction will be neglected, i.e. kp = 0 in Eq. (20). In be derived for the stationary solution
fact numerical calculations have shown that this term only
1 − exp −Γ xL1
marginally influences the results and only with this sim- ϑ̄I,s (x1 ) = ϑin in in
,
I,s − γI ϑI,s − ϑII,s
plification an analytical solution is made possible. Γ
Assuming countercurrent flow with ṁI > 0 and ṁII < 0 (41a)

the boundary conditions are given by ϑ̄I,s (0) = ϑin I,s and
1 − exp −Γ x1
ϑ̄II,s (x1 ) = ϑin
II,s + γII
in in
ϑI,s − ϑII,s L
ϑ̄II,s (L) = ϑin II,s and the stationary solution of Eq. (20), Γ
subsequently referred to with the index s, for kp = 0 reads (41b)
as
and for
in in in
1 − exp −Γ xL1
ϑ̄I,s (x1 ) = ϑI,s − ϑI,s − ϑII,s (32a) exp(Γ ) (Γ − 1) + 1
1 + γγIII exp(−Γ ) ξI = , ξII = ξI . (42)
Γ (exp(Γ ) − 1)
1 + γIII exp −Γ xL1
γ
ϑ̄II,s (x1 ) = ϑin in in
I,s − ϑI,s − ϑII,s ,
1 + γγIII exp(−Γ ) The idea of the power balance scheme is now to approx-
(32b) imate the dynamic outlet temperatures ϑout I and ϑout
II by
in in
θI , θII , ϑI and ϑII according to the stationary relationship
with the abbreviations of Eq. (39), i.e.
U (Np − 2) BL
γI = , (33) 1 ξI − 1 in
cp,I ṁI ϑout
I = θI + ϑI (43a)
U (Np − 2) BL ξI ξI
γII = , (34) 1 ξII − 1 in
cp,II ṁII ϑout
II = θII + ϑII . (43b)
ξII ξII
Γ = γI + γII (35)
Utilizing Eq. (43) the system matrices of Eq. (29) read as
and the overall heat transfer coefficient
1 1 1  |ṁ | 
= + . (36) − mI ξI I − αI αI 0
U hp,I hp,II  
A= βI − (βI + βII ) βII  ,
Thus, the stationary outlet temperatures can be directly 0 αII |ṁII |
− mII ξII − αII
inferred from Eq. (32) in the form ϑout I,s = ϑ̄I,s (L) and
 |ṁ | 
I
0 1
ϑout
II,s = ϑ̄II,s (0). It can be easily seen that the stationary  mI ξI
 0 0 (44)
B= 0 0  , C = ξI 1 ,
temperature mean values (cf. Eq. (25) for M = 1) 0 0
0 |ṁII |
mII ξII
ξII
ZL 1
1 1 − ξI 0
θi,s = ϑ̄i,s dx1 , i ∈ {I, II} (37) D= .
L 0 1 − ξ1II
0
are bounded from below and above, for instance for ϑin Contrary to the upwind scheme, the power balance scheme
I,s >
ensures that the outlet temperatures stationary match the
ϑin
II,s we have
solution of the pde of Eq. (20) for kp = 0.
ϑin out
I,s ≥ θI,s ≥ ϑI,s , (38a) Remark: Note that the power balance model Eq. (29),
Eq. (44) is a finite-dimensional approximation of the
ϑin out
II,s ≤ θII,s ≤ ϑII,s . (38b)
distributed-parameter model Eqs. (20) – (22). The ap-
The latter guarantees the existence of constant parameters proximation scheme was designed in such a way that the
ξI , ξII ∈ [0, 1] such that the following relations resulting mathematical model exhibits a high stationary
accuracy even for only M = 1 volume element. However,
θI,s = ϑin out in
I,s + ξI ϑI,s − ϑI,s , (39a) in case of very fast changes of the inlet temperatures the
transient accuracy is limited due to the insufficient approx-
θII,s = ϑin
II,s + ξ II ϑ out
II,s − ϑ in
II,s (39b)
imation of the transport phenomenon. In order to counter-
are satisfied. Inserting Eq. (32) into Eq. (37) and then into act this shortcoming, the number M of volume elements
Eq. (39), ξI and ξII can be calculated in the form and thus the dimension of the power balance model has
to be increased. As a rule of thumb, the residence time of
exp(Γ ) (Γ − 1) + 1 the fluid inside a volume element should be smaller than
ξI = , ξII = 1 − ξI . (40)
Γ (exp(Γ ) − 1) the minimum rise time of the inlet temperatures.
vii
pump
ϑout
I , ṁI
ϑin
B
II
A
turbine
ϑout
II , ṁII
tank
Figure 6: Examined plate heat exchangers.
RTD Description Symbol A B Unit

plate heat
exchanger channel length L 46 25 cm
channel width B 10.6 10.6 cm
channel height H 2 1.8 mm
plate thickness Hp 0.5 0.5 mm
Figure 4: Photo of the test bed.
number of plates Np 20 40
correlation parameter CNu 0.5 0.38
correlation parameter nPr 0.26 0.26
correlation parameter nRe 0.67 0.68
density (plate) ̺p 8000 8000 kg/m3
specific heat capacity (plate) cp,p 500 500 J/kg/K
4. Validation of the proposed model
thermal conductivity (plate) kp 15 15 W/m/K
4.1. Experimental setup
Table 1: Parameter of the analyzed plate heat exchangers.
In order to validate the presented mathematical models,
a test bed was set up wherein two fluid circuits, one with
a heater and one with a cooling system, are thermally
coupled by a 1/1 pass countercurrent brazed plate heat with their temperature dependencies can be found in the
exchanger in U-form. A picture of the test bed is shown in relevant literature, see, e.g., [10]. Each flow is driven by
Figure 4 and the schematics are depicted in Figure 5. For a pump where the mass flow rates can be controlled by
proportional valves. The inlet and outlet temperatures
ϑin
I , ṁI
are measured by resistance thermometers (RTDs) and the
plate heat
coolant two volume flow rates by means of turbine meters. The
exchanger
supply empirical parameters CNu , nPr and nRe of Eq. (11) were
heater ϑin identified by means of manufacturer’s steady state data
II , ṁII
with the procedure explained in the following section.
pump ϑout
II
ϑout
I 4.2. Identification of the thermal coupling
M
tank As mentioned in Section 2, the parameters CNu , nPr and
nRe of Eq. (11) have to be determined by means of suit-
able experiments. Manufacturer often provide measure-
ments of stationary outlet temperatures ϑout,m
I,s , ϑout,m
II,s for
different but constant mass flow rates ṁm
I,s , ṁII,s and inlet
m
Figure 5: Schematic design of the test bed.
temperatures ϑin,m in,m
I,s , ϑII,s . These steady state data can
the experiments, two different designs of industrial plate be used in order to adapt the calculated stationary out-
heat exchangers have been used, see Figure 6, with the let temperatures ϑout,c out,c
I,s , ϑII,s to the measurements. Note
parameters listed in Table 1. The fluid I is a water- that henceforth the added superscript m and c refer to
glycol mixture with 44% concentration. As coolant, water measured and calculated variables, respectively.
is used in experiment A and a water-glycol mixture with Assuming countercurrent flow with ṁI > 0 and ṁII < 0,
40% concentration in experiment B. All fluid parameters cf. Eq. (32), the stationary outlet temperatures can be
viii
even less than 1% for experiment A. Larger deviations only
occur for fast changing mass flow rates. At this point it
calculated as, is worth mentioning that already small errors in the pa-
1 − exp(−Γ c ) rameters of the Nußelt correlation (given in Eq. (11)) or
ϑout,c
I,s = ϑin,m in,m in,m
I,s − ϑI,s − ϑII,s γc (45a) in the mass flow rates result in relative errors of the outlet
1 + γIIc exp(−Γ c ) temperatures in the percentage range. Thus, the quality
I
1+
c
γII of the low-order mathematical model with the power bal-
ϑout,c = ϑin,m − ϑin,m − ϑin,m
γIc
, ance scheme can be rated very high, in particular if we
II,s I,s I,s II,s c
1+ γIc exp(−Γ )
γII c take into account that the dynamics of the sensors was
(45b) disregarded in the evaluation. The nearly perfect match-
ing between the upwind scheme model with M = 100 and
where the overall heat transfer coefficient U c in γIc , γII
c
and the power balance scheme model with M = 1 confirms
Γ c according to Eqs. (36), (19), (10) and (11) reads as the assumption that for compact plate heat exchangers
the residence time of the fluid can be neglected. The only
1 2H 2H Hp need for higher approximations within the upwind scheme
= + + .
Uc kI CNu PrnI Pr RenI Re kII CNu PrnIIPr RenIIRe kp is up to the nonlinear quasi-stationary behavior, which is
(46) inherently accounted for by the power balance scheme due
to its specific construction.
For all stationary measurements the error between the
measured and the calculated outlet temperatures ϑout,m
I,s −
5. Conclusion
ϑout,c
I,s and ϑ out,m
II,s − ϑ out,c
II,s are combined in the error vec-
tors eI and eII , respectively. Then the parameters CNu , Starting with a detailed modeling based on the funda-
nPr and nRe are determined by solving the nonlinear opti- mentals of heat transfer and the similitude theory accord-
mization problem ing to Nußelt, a temperature mean value model consisting
of three partial differential equations (pdes) was derived
min wI eTI eI + wII eTII eII (47) for a class of compact plate heat exchangers. A semi-
CNu ,nPr ,nRe
discretization by means of the finite volume method was
for suitable weighting factors wI , wII > 0. In our case an performed in order to obtain a model of ordinary differen-
interior-point method was employed utilizing the Matlab tial equations. Due to the small fluid residence time and
function fmincon. the limited dynamics of the inlet temperatures a discretiza-
tion by only one single volume element was possible. The
4.3. Measurement Results relation between the outlet temperatures and the states
For the experiments, the inputs, i.e. the two inlet tem- of the reduced model was established by the classical up-
peratures ϑin in
I and ϑII and the two mass flow rates ṁI and wind and a newly developed power balance scheme. The
ṁII , are varied over the operating range and are depicted latter was derived utilizing an approximate stationary so-
in the left pictures of Figure 7 and Figure 8 for the two lution of the underlying pdes. Finally, the models were
experiments A and B, respectively. These measured time validated by means of experimental results. It was shown
evolutions also serve as inputs for the different mathemat- that the mathematical model based on the power balance
ical models. The model accuracy is then assessed by com- scheme with only one volume element exhibits similar high
paring the resulting measured outlet temperatures ϑout I accuracy as the upwind scheme model with 100 volume el-
and ϑoutII with the simulated ones. Thereby, two low-order ements. Since this low-order model succeeds in capturing
mathematical models, each only with M = 1 volume ele- the essential nonlinearities and the dynamic behavior in
ment, based on the upwind and the power balance scheme the considered operating range it serves as a suitable ba-
according to Section 3.1 and Section 3.2, and two further sis for model-based (nonlinear) control and optimization.
higher order models employing the upwind scheme with Apart from the advantages of low complexity and high ac-
M = 10 and M = 100 volume elements are used for com- curacy the model has the nice feature that the states are di-
parison purposes. The time evolutions of the simulated rectly measurable and thus nonlinear state controllers may
and measured outlet temperatures ϑout I and ϑout
II as well be directly applied without the need for a state observer.
as the relative errors eI and eII of the simulation results Note that this kind of low-order models can be applied to
are given in the right pictures of Figure 7 and Figure 8. any kind of heat exchanger fulfilling the assumption of low
As expected, the errors of the upwind scheme mod- fluid residence time compared to the dynamics of the inlet
els decrease for an increasing number of volume elements temperatures.
M . The mathematical model based on the power balance The stationary accuracy of the power balance scheme
scheme already shows very good results for only one vol- directly depends on the quality of the heat transfer model
ume element (M = 1) and is comparable to the upwind and thus on the identified correlation parameters. Because
scheme model with M = 100. Here the resulting station- these parameters also appear as exponents in the Nußelt
ary simulation error is less than 3% for experiment B and correlation, already small differences lead to high errors
ix
25
ṁI in kg/min
30
in °C
20
20
ϑout
10 15
I
35 8
I in °C
eI in %
30
0
25
ϑin
−4
20
−8
20
ṁII in kg/min
30
II in °C
15
25
10
20
ϑout
5
0 15
14 4
II in °C
eII in % 0
10
ϑin
−4
6
0 5 10 15 20 0 5 10 15 20
time in min time in min
Upwind scheme with M = 1 Upwind scheme with M = 10

Upwind scheme with M = 100 Power balance scheme with M = 1
Measurement
Figure 7: Measurement and simulation results for experiment A.
of the overall heat transfer coefficient and consequently [2] M. Masubuchi, A. Ito, Dynamic Analysis of a Plate Heat Ex-
to errors in the outlet temperatures. But this is not a changer System, Bulletin of the Japan Society of Mechanical
Engineering 20 (1977) 434–441.
drawback of the power balance scheme but will appear in [3] S. Das, W. Roetzel, Dynamic Analysis of Plate Heat Exchangers
every finite volume scheme based on the proposed Nußelt with Dispersion in both Fluids, International Journal of Heat
correlation. Moreover, the proposed power balance model and Mass Transfer 38 (1995) 1127–1140.
with only one discretization element has its limitation. If [4] S. Das, B. Spang, W. Roetzel, Dynamic Behavouir of Plate
Heat Exchangers - Experiments and Modeling, ASME Journal
the dynamics of the inlet temperatures are higher than of Heat Transfer 117 (1995) 859–864.
the corresponding residence time of the fluid inside a dis- [5] X. Luo, X. Guan, M. Li, W. Roetzel, Dynamic Behavouir of
cretization element, the transport phenomenon gets more one-dimensional Flow Multistream Heat Exchangers and their
dominant and a higher order finite volume model is indis- Networks, International Journal of Heat and Mass Transfer 46
(2002) 705–715.
pensable. However, the power balance scheme is also ap- [6] J. Yin, M. Jensen, Analytical Model for Transient Heat Ex-
plicable in this case, but the number of volume elements changer Response, International Journal of Heat and Mass
and hence the dimension of the resulting model has to be Transfer 46 (2003) 3255–3264.
increased. [7] F. Bechmann, Das stationäre und instationäre Verhalten von
Platten-Wärmeübertragern im industiellen Maßstab, VDI Ver-
lag, Düsseldorf, 1996.
[8] A. Dwivedi, S. Das, Dynamics of Plate Heat Exchangers Subject
Acknowledgements to Flow Variations, International Journal of Heat and Mass
Transfer 50 (2007) 2733–2743.
We highly appreciate the technical and financial support [9] G. Jonsson, O. Palsson, Use of Emperical Relations in the Pa-
provided by Hydac Cooling GmbH, Germany. rameters of Heat-Exchanger Models, Industrial & Engineering
Chemistry Research 30 (1991) 1193–1199.
[10] VDI-Gesellschaft Verfahrenstechnik und Chemieingenieurwe-
sen, VDI Heat Atlas, Springer, Berlin/Heidelberg, 2. edition,
References
2010.
[11] M. Bassiouny, Expermientelle und theoretische Untersuchungen
[1] R. Shah, D. Sekulić, Fundamentals of Heat Exchanger Design,
über Mengenstromverteilung, Druckverlust und Wärmeüber-
John Wiley & Sons, New Jersey, 2003.
35
ṁI in kg/min
30 30
in °C
20 25
20
ϑout
10
I
15
35 3
I in °C
eI in %
30
0
25
ϑin
−3
20
20
ṁII in kg/min
30
II in °C
15 25
10 20
ϑout
5 15
18
3
II in °C
14 eII in % 0
ϑin
−3
10
0 5 10 15 20 0 5 10 15 20
time in min time in min
Upwind scheme with M = 1 Upwind scheme with M = 10

Upwind scheme with M = 100 Power balance scheme with M = 1
Measurement
Figure 8: Measurement and simulation results for experiment B.
gang in Plattenwärmeaustauschern, VDI Verlag, Karlsruhe,

1985.
[12] W. Roetzel, Y. Xuan, Dynamic Behaviour of Heat Exchangers,
WIT Press, Boston/Southampton, 1999.
[13] H. Baehr, K. Stephan, Heat and Mass Transfer, Springer, Berlin
/ Heidelberg, 2. edition, 2006.
[14] F. Incorpera, D. DeWitt, T. Bergman, A. Lavine, Fundamentals
of Heat and Mass Transfer, John Wiley & Sons, US, 2007.
[15] J. Holman, Heat Transfer, McGraham-Hill, New York, 10. edi-
tion, 2010.
[16] R. Heavner, H. Kumar, A. Wanniarachchi, Performance of an
Industrial Plate Heat Exchanger: Effect of Chevron Angle, in:
Proceedings of the American Institute of Chemical Engineers
Symposium Series, volume 89, Atlanta, GA, USA, pp. 262–267.
[17] S. Jain, A. Joshi, P. Bansal, A New Approach to Numerical
Simulation of Small Sized Plate Heat Exchangers With Chevron
Plates, ASME Journal of Heat Transfer 129 (2007) 291–297.
[18] H. Martin, A Theoretical Approach to Predict the Performance
of Chevron-Type Plate Heat Exchangers, Chemical Engineering
and Processing 35 (1996) 301–310.
[19] J. Ferziger, M. Perić, Computational Methods for Fluid Dy-
namics, Springer-Verlag, Berlin / Heidelberg, 3rd edition, 2002.
[20] R. LeVeque, Finite-Volume Methods for Hyperbolic Problems,
Cambridge Texts in Applied Mathematics, Cambridge Univer-
sity Press, Cambridge, 2002.
[21] S.V. Patankar, Numerical Heat Transfer and Fluid Flow, New
Hemisphere Publishing Corporation, USA, 1980.
xi

Accurate Low-Order Dynamic Model of A Compact Plate Heat Exchanger

Uploaded by

Copyright:

Available Formats

Accurate Low-Order Dynamic Model of A Compact Plate Heat Exchanger

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Accurate Low-Order Dynamic Model of A Compact Plate Heat Exchanger

Uploaded by

Copyright:

Available Formats

This document contains a post-print version of the paper

Accurate low-order dynamic model of a compact plate heat exchanger

Cite this article as:

Link to original paper:

Read more ACIN papers or get this document:

Alexander Michel∗, Andreas Kugi

Compact plate heat exchangers like small sized brazed

Figure 6: Examined plate heat exchangers.

RTD Description Symbol A B Unit

Upwind scheme with M = 1 Upwind scheme with M = 10

Figure 7: Measurement and simulation results for experiment A.

Upwind scheme with M = 1 Upwind scheme with M = 10

Figure 8: Measurement and simulation results for experiment B.

gang in Plattenwärmeaustauschern, VDI Verlag, Karlsruhe,

You might also like