Sie sind auf Seite 1von 11

International Journal of Heat and Mass Transfer 96 (2016) 256–266

Contents lists available at ScienceDirect

International Journal of Heat and Mass Transfer


journal homepage: www.elsevier.com/locate/ijhmt

Heat transfer and thermal stresses in a circular tube with a non-uniform


heat flux
C. Marugán-Cruz a,⇑, O. Flores b, D. Santana a, M. García-Villalba b
a
Ing. Térmica y Fluidos, Universidad Carlos III de Madrid, Spain
b
Bioing. e Ing. Aeroespacial, Universidad Carlos III de Madrid, Spain

a r t i c l e i n f o a b s t r a c t

Article history: This paper presents a thermal analysis of thin-wall pipes under non-uniform heat flux in the circumfer-
Received 9 October 2015 ential direction, with a turbulent flow in statistically stationary state inside. The temperature distribution
Received in revised form 9 December 2015 in the solid and in the fluid is obtained using an spectral method that solves the conjugate heat transfer
Accepted 12 January 2016
problem. Special attention is paid to the inner wall fluid temperatures and the thermal stresses on the
Available online 29 January 2016
solid, that are compared to predictions based on 1D models in which the circumferential heat transfer
is neglected. The comparison shows that, even if at sufficiently large Biot numbers (Bi J 0:3) the 1D
Keywords:
model gives a reasonable prediction of the inner wall fluid temperatures (less than 5% of error), the 1D
Temperature distribution
Thermal stresses
model for the thermal stresses is only appropriate for very large Biot numbers (Bi J 10), giving qualita-
Non-uniform heat flux tively wrong results for Biot numbers below 0.3.
Biot number Ó 2016 Elsevier Ltd. All rights reserved.

1. Introduction tube, resulting as well in a circumferentially-varying heat flux on


the inner wall of the pipe. This asymmetry precludes, in principle,
Cumulative fatigue damage plays a key role in life prediction of the use of one-dimensional solutions to obtain the peak thermal
tubular heat exchangers subjected to time varying heat fluxes. To stress.
estimate the tube thermal fatigue, the maximum thermal stress Similar working conditions can be found in other industrial
must be obtained for different operating cycles [1]. Therefore, the applications, like parabolic trough collectors, space-vehicle heat-
effect that the operating conditions and fluid properties have on exchangers or in the cooling system of nuclear reactors. In all these
the tube temperature profiles is of paramount importance from cases, the non-uniform heat fluxes produce radial and circumfer-
an engineering point of view. ential temperature gradients in the tube walls, which result in
There are many circumstances under which the tube thermal thermal stresses on the solid and circumferential variations of
stress determination is straightforward. Such is the case of pipes the fluid temperature at the wall. Both, thermal stresses and inner
under uniform heat flux, where the mean temperatures of the fluid wall fluid temperature, are extremely important from an engineer-
and pipe wall vary linearly along the axial direction [2], while the ing point of view, since the former can produce mechanical failure
temperature in a section of the pipe is a function of only the radial of the pipe and the latter controls the chemical degradation of the
direction. In such situations the thermal stress problem becomes a heat transfer fluid [4].
one-dimensional problem, and the effective stress can be obtained The prediction of heat transfer for non-uniform configurations
knowing the heat flux and the thermo-mechanical properties of has been the subject of many research studies in the past, using
the pipe [3]. analytical methods, experiments and numerical simulations. First
On the other hand, there are applications where uniformity is Reynolds [5] and later Rapier [6] studied the case of a thermally
not found. Concentrating solar power is a good example of such developed flow with circumferential varying heat flux from an ana-
applications. In solar central receivers, for example, the front- lytical point of view. They assumed a fully developed turbulent
tube section must withstand high heat flux (around 1 MW/m2) flow in the pipe, and prescribed a heat flux that was expressed as
whereas the rear-tube section is almost adiabatic, resulting in a a sine or cosine series with zero mean value. The resulting temper-
circumferentially-varying heat flux on the outer wall of the pipe. ature profiles in the fluid were obtained as a series of sines or cosi-
That heat flux is transferred to the molten salt flowing inside the nes, whose coefficients vary with the Reynolds and Prandtl number
of the flow. Comparing theinner wall fluid temperatures predicted
by Reynolds [5] for a sinusoidal heating with the inner wall fluid
⇑ Corresponding author.

http://dx.doi.org/10.1016/j.ijheatmasstransfer.2016.01.035
0017-9310/Ó 2016 Elsevier Ltd. All rights reserved.
C. Marugán-Cruz et al. / International Journal of Heat and Mass Transfer 96 (2016) 256–266 257

temperatures obtained from applying the Dittus and Boetter formulas derived for a circular tube with a radial temperature
correlation yields a difference on the maximuminner wall fluid distribution (i.a, uniform heat flux in the circumferential direc-
temperatures of about 50% for the typical working conditions tion),which can be found in specialized books [24–26].
of concentrating solar power receivers (Reynolds number In principle, the parameter that controls whether a local
Re = 30,000, Prandtl number Pr ¼ 10). approach is valid is the Biot number: the ratio between the charac-
Although Reynolds [5] and Rapier [6] did not take into account teristic heat flux in the radial direction and the characteristic heat
the fluid-wall interaction, the thermal interaction inside ducts has flux in the circumferential direction. If we consider the conjugate
been the interest of many other investigations. Several aspects of heat transfer problem (solid + fluid), the heat fluxes in the radial
this problem have been studied in the context of laminar flows, like direction (r) are controlled by the heat transfer in the solid–fluid
the effect of unsteadiness in the heat flux [7–9], or the effect of spa- boundary, and the heat fluxes in the circumferential direction (h)
tial inhomogeneity in the axial direction [10,11]). by the conductivity of the solid. The definition of the Biot number
Painstaking experimental studies have also been carried out in is then
the past. Black and Sparrow [12] investigated the local and average
he
heat transfer coefficients of an air flow through a tube with Bi ¼ ; ð1Þ
circumferential varying wall heat flux. More recently, Yang and
js
co-workers [13] measured the heat transfer performance of spiral where h is the convective heat transfer coefficient between the fluid
tubes with molten salts as heat transfer fluid. The results from and the solid, e is the thickness of the tube wall and js is the con-
these investigations also show differences between the experimen- ductivity of the solid. For large values of Bi, the radial heat flux is
tal values of the Nusselt number and the correlations generally dominant, and one can expect the correlations and methods based
used in literature. on uniform (radial) heat fluxes to be applicable locally. However,
Other researchers have tried to tackle the problem of non- when the Bi is small these methods should be expected to fail.
uniform heating in turbulent flows from a numerical point of view, The objective of the present work is to establish the range of
in more or less idealized configurations (fully developed turbulent Biot numbers where the local approach is applicable in pipes with
flow, relatively low Reynolds and Prandtl numbers, constant fluid non-uniform heating in the circumferential direction. When the
properties, etc.). Weingand and Gassner [14] analyzed the conju- local approach is applicable, the solution of the conjugate heat
gate heat transfer problem showing that axial heat conduction transfer problem can be avoided, thereby simplifying the stress
effects in the fluid are important for low Peclet numbers and that analysis. To achieve this, the conjugate heat transfer problem for
these effects might drastically influence the heat transfer behavior a circular pipe with a circumferentially varying heat flux and a fully
for short heating sections even at higher Peclet numbers. Some developed turbulent flow inside is solved using a simple eddy dif-
studies focus on the evaluation of the eddy-diffusivity associated fusivity model and constant fluid properties.
to the non-uniformity of the heat flux [15,16]. Other studies are Different fluid velocities, tube diameter and thickness have
concerned with the evaluation of thermal field and thermal stres- been studied, in order to analyze a range of Biot numbers spanning
ses along the solid wall [17]. two orders of magnitude. Particular attention is paid to the effect of
There are also numerical studies of more applied cases, typically the Biot number on the inner wall fluidtemperature and thermal
using Reynolds-Averaged Navier Stokes simulations to solve the stresses, and to our ability to estimate them using simple 1D
heat transfer in the fluid. For instance, Flores and co-workers models.
[18] report an analysis of thermal stresses on the collector of a To some extent, the present numerical work can be considered
solar thermal plant for different configurations, and Rodríguez- as a continuation of previous works that only considered the heat
Sánchez et al. [19] compares two simplified 2D models using a transfer in the fluid part [5,6,27]. Here we add also the heat trans-
RANS calculation with a commercial software for the evaluation fer problem on the solid part.
of inner wall fluid temperatures and thermal stresses on a solar The paper is structured as follows. First, Section 2 describes the
external receiver. In the same vein, Kim et al. [20] proposed a formulation of the conjugate heat transfer problem, and the details
correlation for the convection losses in tower solar receivers using of the numerical solver. Section 3 presents the results obtained in
CFD simulations. Wang and co-workers [21] analyzed the the calculations, discussing the inner wall fluid temperature
conjugate heat transfer problem in a parabolic trough collector (Section 3.2) and the thermal stresses on the solid (Section 3.3)
showing that asymmetric heat flux has a significant influence on in terms of the Biot number. Finally, conclusions are offered in
the local heat transfer process. Section 4.
In view of the above results, it is clear that the prediction of the
thermal stress history for a tubular heat exchanger heated 2. Physical, mathematical and numerical model
non-uniformly is a difficult problem. While simulations solving
the conjugate heat transfer problem provide accurate results, they 2.1. Physical model
typically require additional simplifications regarding flow proper-
ties, geometries, etc. On the other hand, experimental measure- We consider the conjugate heat transfer problem of a circular
ment can be extremely expensive and time consuming. However, pipe subject to homogeneous heating along the axial/streamwise
the engineering approaches used to determine theinner wall fluid direction (x) and non-homogeneous heating in the circumferential
temperatures and thermal stresses are usually based on empirical direction (h). The inner radius (diameter) of the pipe is denoted
correlations derived from uniform heat flux conditions. The ratio- Ri (Di ), while the outer radius (diameter) is denoted Ro (Do ). The
nale for that simplification is that locally the conditions are uni- wall-thickness is e ¼ Ro  Ri , and r is the radial coordinate.
form, with stronger temperature gradients in the radial direction A sinusoidal heat flux is imposed on one half of the wall of the
than in the axial or circumferential directions. For instance,inner pipe, while the other half is considered adiabatic, resulting in a
wall fluid temperatures are usually estimated using Nusselt num- heat flux on the outer surface
ber empirical correlations (e.g., Kolb [22] uses the Gnielinski corre- 
lation and Jianfeng and co-workers [23] use the Dittus and Boelter
qmax sin h if 06h6p
qo ðhÞ ¼ ð2Þ
correlation). Thermal stresses are customarily obtained using 0 if h>p
258 C. Marugán-Cruz et al. / International Journal of Heat and Mass Transfer 96 (2016) 256–266

It is also convenient to define the total heat flux (per unit If we model the turbulent heat flux using the concept of turbu-
length) applied to the outer surface, lent eddy conductivity, we can write the Reynolds-Averaged equa-
Z p tion for the cross-plane temperature fluctuation
Q¼ qo ðhÞRo dh ¼ 2qmax Ro : ð3Þ dT b
0 qC p UðrÞ ¼ r  jT rH
dx
As discussed in Section 2.5, the total heat flux (Q) is kept con-    
1 @ @H 1 @ 1 @H
stant for all the cases presented here, so that the maximum heat ¼ r jT ðrÞ þ jT ðrÞ ; ð7Þ
r @r @r r @h r @h
flux (qmax ) is inversely proportional to the outer radius.
The fluid inside the tube is assumed to be flowing in a fully where jT is the turbulent eddy conductivity. In principle, jT is a
developed turbulent regime, both in terms of velocity and temper- function of r and h . However, it has been shown that the effect of
ature. Moreover, it is assumed that the fluid has constant density the circumferential variation of jT is small compared to its radial
(q), viscosity (l), specific heat (C p ) and thermal conductivity (j). variation [33,27,16]. Hence, in the present paper we assume
Under these conditions, the hydrodynamic problem (fluid) is jT ¼ jT ðrÞ. Moreover, we assume that the turbulent Prandtl number
decoupled from the thermal problem (fluid + solid). PrT ¼ qC p mT =jT is equal to 1. Schwertfirm and Manhart [34] show
that this approximation is appropriate when the Prandtl number
Pr ¼ C p l=j is in the range 0.1–10, which is the case for the database
2.2. Mathematical formulation
presented here (see Table 1). With these approximations,the turbu-
lent eddy conductivity becomes
In this section we briefly describe the mathematical formula-
tion of the problem, which is similar to that used by Reynolds [5] jT l
¼ Pr T ; ð8Þ
and Rapier [6], except that we consider the conjugate heat transfer j l
problem to analyze the flux and temperature distributions in both where lT =l is given in Eq. (4).
the fluid and the tube.
The temperature field in the solid T ðsÞ ðx; r; h; tÞ can also be
With the assumptions detailed in the previous section (fully ðsÞ
developed flow, constant fluid properties), the hydrodynamic decomposed in bulk (T b ¼ T b ), cross-averaged temperature (HðsÞ )
problem in the fluid is decoupled from the thermal problem. In and fluctuation (HðsÞ0 ). Again, the energy conservation equation
particular the mean velocity and the Reynolds stresses are func- yields a linear variation of the bulk temperature with the axial
tions of r, only, and depend on the Reynolds number, coordinate, and the energy conservation in the cross plane yields
Re ¼ qU b Ri =l, where U b is the bulk velocity (i.e. mean velocity the following equation for the cross-averaged temperature
averaged in time and in the cross-section of the pipe). For conve- fluctuations,
nience, we use the analytic expressions given by Cess [28] for the
mean velocity profile (UðrÞ) and for the turbulent eddy viscosity
(mT ), which is given by
2 3
  2 !1=2 Table 1
lT 1 4 2
k Re2s ðg  1ÞRes
¼ 1þ 2 2 2 2
ð1  g Þ ð1 þ 2g Þ 1  exp þ 15 Summary of the parameters of the simulations. Re ¼ qU b Ri =l is the Reynolds
l 2 9 A number, Res ¼ qus Ri =ls is the turbulent Reynolds number, Pr ¼ C p l=j is the Prandtl
number, e=Ri is the wall thickness non-dimensionalized with the inner radius and Bi is
ð4Þ the average Biot number, calculated using the correlation proposed by Gnielinski [40].

where g ¼ r=Ri , and the friction Reynolds number Res is a known Case Re Res Pr e=Ri Bi
function of the Reynolds number (Re). In the present calculations, i 2532 180 0.7 0.085 0.019
the tunable parameters of the model have been chosen to be ii 0.183 0.041
k ¼ 0:42 and A ¼ 27. This model is robust and has been used by sev- iii 3 0.085 0.034
iv 0.183 0.073
eral authors over the last 50 years in various applications [5,29–31], v 10 0.085 0.053
spanning a range of Reynolds number 2:5  103  5  105 . It should vi 0.183 0.114
be noted that similar results are obtained when profiles from DNS vii 5815 360 0.7 0.085 0.038
[32] are used instead of the Cess profiles. viii 0.183 0.082
For a fully developed thermal problem, it is possible to show ix 3 0.085 0.074
x 0.183 0.160
that the temperature of the fluid Tðx; r; h; tÞ can be decomposed in
xi 10 0.085 0.118
0 xii 0.183 0.255
Tðx; r; h; tÞ ¼ T b ðxÞ þ Hðr; hÞ þ H ðx; r; h; tÞ
xiii 8528 500 0.7 0.085 0.051
dT b
¼ T0 þ x þ Hðr; hÞ þ H0 ðx; r; h; tÞ ð5Þ xiv 0.183 0.111
dx xv 3 0.085 0.103
xvi 0.183 0.223
where the bulk temperature, T b ðxÞ, is the time and cross-plane xvii 10 0.085 0.167
ðr  hÞ averaged temperature, Hðr; hÞ is the time averaged tempera- xviii 0.183 0.361

ture fluctuation in the cross-plane, and H0 ðx; r; h; tÞ is the remaining xix 18,863 1000 0.7 0.085 0.095
xx 0.183 0.205
fluctuation. For the case considered here, the bulk temperature
xxi 3 0.085 0.202
increases linearly with x, and the bulk temperature gradient xxii 0.183 0.437
dT b =dx is univocally determined by the total heat flux received xxiii 10 0.085 0.337
through the fluid–solid interface, xxiv 0.183 0.729

Z 2p xxv 41,184 2000 0.7 0.085 0.174


dT b 2
qC p U b p Ri ¼ j@ r HðRi ; hÞRi dh ¼ Q; ð6Þ
xxvi
xxvii 3
0.183
0.085
0.375
0.389
dx 0
xxviii 0.183 0.841
xxix 10 0.085 0.666
which has to be equal to the total heat flux received in the outer
xxx 0.183 1.440
surface of the pipe, Q (defined in Eq. (3)).
C. Marugán-Cruz et al. / International Journal of Heat and Mass Transfer 96 (2016) 256–266 259

! !
1 @ @ HðsÞ 1 @ js @ HðsÞ are found to be in good agreement with the DNS data. The relative
0 ¼ r  js rHðsÞ ¼ r js þ ð9Þ error is smaller than 4%.
r @r @r r @h r @h
A second set of validation tests consists of the calculation of the
where js is the thermal conductivity of the tube material. conjugate heat transfer problem with homogeneous heating, for a
In order to solve the conjugate heat transfer problem, we solve broad range of Re and Pr. Despite the strong assumptions in our
method (constant fluid properties, simple model for the turbulent
Eq. (7) for Hðr; hÞ in 0 < r < Ri and Eq. (9) for HðsÞ ðr; hÞ in
eddy diffusivity), the values obtained for the Nusselt number
Ri < r < Ro . At the outer boundary of the solid pipe, the non-
(Nu ¼ hDi =j, defined in terms of the heat transfer coefficient h),
homogeneous heat flux conditions yields
are in good agreement with the usual correlations found in the lit-
@ HðsÞ erature [38,39]: Gnielinski and Ditus-Boetter. This is shown in
js ðRo ; hÞ ¼ qo ðhÞ ð10Þ Fig. 1(b). In addition, some of the cases computed by Reynolds
@r
for sinusoidal heating in the circumferential direction [5] have
At the solid–fluid interface, the following coupling conditions been recomputed using the present method. Our results agree with
apply Reynolds data with an error smaller than 2% for ReDi ¼ 10; 000 and
Pr ¼ 0:7  10. The simulations are run using 60 collocation points
@ HðsÞ @H
js ðRi ; hÞ ¼ j ðRi ; hÞ ð11Þ in the circumferential direction, and 150 collocation points in the
@r @r
radial direction. Further refinement of the grid did not result in
HðsÞ ðRi ; hÞ ¼ HðRi ; hÞ ð12Þ any appreciable variation of the results.

2.3. Numerical method 2.4. Simplified 1D model

The coupled system of Eqs. (7)–(12) is solved using a spectral As noted in the introduction, we are interested in comparing the
method. We use a collocation method in which a Fourier expansion results obtained from solving the complete 3D conjugate heat
is used in the circumferential direction, and a Chebychev expan- transfer problem with the results obtained from a simplified model
sion in the radial direction. The method is implemented in based on a local approach, that accounts only for conduction on the
MATLAB, using the formulation described by Trefethen [35] (chap- radial direction. For that model we will use the Gnielinski (exper-
ter 11). We employ the Chebychev differentiation matrices from imental) correlation [40] for the convective heat transfer coeffi-
Weideman and Reddy [36], that are more accurate than those from cient, h:
Trefethen [35]. The resulting linear systems of equations are solved
h Di f =8 ðReDi  1000Þ Pr
using the MATLAB backslash operator. Nu ¼ ¼ pffiffiffiffiffiffiffiffi 2=3 ; ð13Þ
We have performed extensive validations of the numerical
j 1 þ 12:7 f =8 ðPr  1Þ
code, and three of them are discussed in the following lines. First,
where Nu is the Nusselt number and ReDi is the Reynolds number
we have used data of a DNS of forced convection in a pipe with uni-
based on the bulk velocity and the inner diameter of the pipe. The
form heating at Re ¼ 5300 and Pr ¼ 0:7 [37]. This simulation did
Darcy friction factor, f, can be obtained using the Petukhov correla-
not consider the wall-fluid interaction, so that only the fluid part
tion [41]:
was considered in the validation test. Using 20 modes in the cir-
cumferential direction and 51 modes in the radial direction, we  2
f ¼ 0:790 logðReDi Þ  1:64 : ð14Þ
performed two calculations. In the first one the tunable parameters
of the model we set as discussed above, A ¼ 0:27 and k ¼ 0:42. For The determination of the model temperature H1D is performed
the second one, we have selected the values employed by Reynolds as follows. Once Nu is known from Re and Pr, it is possible to relate
[5], namely A ¼ 0:26 and k ¼ 0:4. The mean temperature distribu- the inner wall fluid temperature in the fluid (HðRi ; hÞ) to the heat
tion is shown in Fig. 1(a) and the results of the present calculations flux in the outer wall (qo ðhÞ). To that end, we use the convection

Fig. 1. (a) Temperature distribution Hþ þ þ


w  H as a function of the distance to the wall Ri  r (the superscript + indicates standard wall scaling) for a case with homogeneous
þ

heating, Re ¼ 5300, Pr ¼ 0:7. Symbols, DNS data of Piller [37]. Blue, present calculation. Red, present calculation with tunable parameters in Eq. (4), A ¼ 26; k ¼ 0:4 as selected
by [5]. (b) Nusselt number, Nu, as a function of the Reynolds number, ReDi , computed with homogeneous heating. Squares, e=Ri ¼ 0:085. Circles, e=Ri ¼ 0:183. Black, Pr ¼ 0:7.
Red, Pr ¼ 3. Blue, Pr ¼ 10. Solid lines, Dittus–Boetter correlation. Dashed lines, Gnielinski correlation. (For interpretation of the references to colour in this figure legend, the
reader is referred to the web version of this article.)
260 C. Marugán-Cruz et al. / International Journal of Heat and Mass Transfer 96 (2016) 256–266

coefficient h ¼ qi ðhÞ=HðRi ; hÞ appearing in the definition of Nu,


where qi ðhÞ is the heat flux through the fluid–solid interface. Note
that the energy conservation in the solid implies that
Z 2p Z 2p
qo ðhÞRo dh ¼ qi ðhÞRi dh: ð15Þ
0 0

Then, the assumption of conductive heat transfer in the radial


direction only implies that qo ðhÞRo ¼ qi ðhÞRi , so that the inner wall
fluid temperature of the 1D model is
qi ðhÞ qo ðhÞRo =Ri
H1D ðRi ; hÞ ¼ ¼ ; ð16Þ
h h
with h given by Eq. (13) and qo ðhÞ by Eq. (2).
Note that in Eq. (16) the variation of H1D with the circumferen-
tial coordinate is given by the circumferential variation of qo ðhÞ,
while the convective heat transfer coefficient h is constant. In that
sense, the present 1D (o local) model can also be understood as a
model based on the averaged heat transfer of the pipe.

2.5. Cases studied


Fig. 2. Ratio between the local Biot number, Bimax , and the average Biot number, Bi.
Black symbols, Pr ¼ 0:7. Red symbols, Pr ¼ 3. Blue symbols, Pr ¼ 10. Circles, thick
We consider the heat transfer and absorption characteristics for
wall, e=Ri ¼ 0:183. Squares, thin wall, e=Ri ¼ 0:085. Closed symbols, D1 > 0. Open
different fluids (Pr numbers range from 0.7 to 10), different flow symbols, D1 < 0. See Section 3.3 for the definition of D1 . Reynolds numbers increase
regimes (Re numbers vary from 2500 to 42,000) and different wall from left to right for each serie. (For interpretation of the references to colour in this
thickness (e=Ri ¼ 0:085 and 0:183). In all cases we consider a ratio figure legend, the reader is referred to the web version of this article.)
of thermal conductivity between fluid and solid equal to
ks =k ¼ 37:5, typical of concentrating solar power applications.1It
should be noted that the results presented in Section 3 are qualita-
tively similar when smaller conductivity ratios are considered, i.e.
ks =k ¼ 18:75, suggesting that the conclusions of the present work
are not strongly dependent on the conductivity ratio. As mentioned
in the introduction, the total heat rate imposed on the outer wall has
been kept constant in order to maintain the same bulk temperature
gradient dT b =dx in all cases (see Eq. (6)).
Table 1 shows the main parameters of all the cases computed.
The table also includes the values of the average Bi number, where
the h in Eq. (1) has been estimated using the 1D model (Gnielinski
correlation for Nu [40]). Note that, for the range of Re; Pr and e=Ri
considered here, the average Biot number Bi varies by roughly
two orders of magnitude (between 0.019 and 1.44).
It is also possible to define a local Biot number, using the value
of h obtained from the present calculations at the position of max-
imum incident heat flux,
 
ehjh¼p=2 e j @H
Bimax ¼ ¼ ð17Þ
js js Hðr; hÞ @r r¼Ri ;h¼p=2
Fig. 3. Circumferential variation of heat flux on the outer and inner walls of the
The comparison between these two numbers is shown in Fig. 2, tube. The heat flux imposed on the outer surface, qo =qmax , is the red solid line. The
where it can be observed that the difference between them is rel- filled symbols correspond to the thicker tube (case xxx) and the empty symbols to
atively small (less than 30%) and decreases as Pr increases. Since the thinner tube (case i). The squares are the heat flux distribution in the inner wall
obtained with the spectral method, qi ðRi ; hÞ=qmax , and the circles correspond to the
from an engineering point of view the average Bi number is more inner distribution of the heat flux assuming no conduction in the circumferential
accessible (i.e., easier to compute), we will mainly use that Biot direction in the tube wall, qi;r =qmax . (For interpretation of the references to colour in
number in the rest of the paper. In any case, the use of Bimax does this figure legend, the reader is referred to the web version of this article.)
not qualitatively change the results presented, nor the conclusions.

3. Results and discussion with the maximum heat flux qmax . Note that this maximum heat
flux depends on the case: for two cases with the same Reynolds
3.1. Heat flux and Prandtl number and different wall thickness, the magnitude
qmax will be affected by a factor of Ri =Ro .
As was previously explained, a sinusoidal heat flux was The heat flux distribution on the inner wall, qi ðRi ; hÞ, obtained
imposed on one half of the wall of the pipe (front side), while the from the present calculations is also shown in Fig. 3 for two cases
other half was considered adiabatic (rear side) (see Eq. (6)). That wih extreme Biot numbers: the blue empty-squares for case i
heat flux is shown in Fig. 3 (red solid line), non-dimensionalized (ReDi ’ 5000; Pr ¼ 0:7, thin wall e=Ri ¼ 0:085, low Bi ¼ 0:019),
while the blue filled-squares correspond to the case xxx
1
ks =k ¼ 37:5 corresponds to using HITEC (molten nitrate salt, 7% NaNO3, 53% KNO3, (ReDi ’ 82; 000; Pr ¼ 10, thick wall e=Ri ¼ 0:183, large Bi ¼ 1:44).
40% NaNO2 as heat transfer fluid in a stainless steel pipe. It is worth noting that for the lowest Bi, the maximum heat flux
C. Marugán-Cruz et al. / International Journal of Heat and Mass Transfer 96 (2016) 256–266 261

through the inner wall is actually smaller than qmax , due to the both cases, this yields a larger temperature jump in the fluid for
redistribution of heat towards the adiabatic side of the tube. In the case with the thicker thermal boundary layer.
the case of high Bi most of the heat transfer on the tube occurs Fig. 5 compares the inner wall fluid temperature from the pre-
in the radial direction, and the heat flux in the inner wall is larger sent calculations (dashed lines) with the simplified model (solid
by a factor close to Ro =Ri ¼ 1 þ e=Ri  1:2. lines), Eq. (16). The difference between the inner wall fluid temper-
Fig. 3 also includes the heat flux distribution in the inner wall ature, Tðr ¼ Ri Þ, and the bulk temperature, T b , at that section,
using the simplified 1D model (black circles in the figure), to com- Tðr ¼ Ri Þ  T b has been non-dimensionalized with T ref . It can be
pare it with the present results. As expected, assuming that there is noticed that in general, the 1D model underestimates the inner
no heat conduction on the circumferential direction leads to an wall fluid temperature on the hot side, while it overestimates it
overestimation of the maximum heat flux on the inner wall and on the adiabatic side (note that for the 1D model, the inner wall
to an underestimation of the heat flux on the opposite side. This fluid temperature in the adiabatic side has to be equal to the bulk
difference is larger for the low Bi case. temperature). The only exception are the cases with the lowest
Reynolds numbers, Re ¼ 2532, in which the inner wall fluid tem-
3.2. Temperature profile perature is overestimated by the 1D model also in the hot side.
Interestingly, Figs. 5(a–c) show that the difference between the
The importance of the Biot number on the temperature distri- inner wall fluid temperatures predicted by the 1D model and
bution is analyzed in this section, paying special attention to its the spectral method decreases as the Re and Pr increases, even if
effect on the inner wall fluid temperature. the Gnielinski correlation is expected to work for the whole range
Fig. 4 shows the time averaged temperature fluctuation in the of Re and Pr considered here [42]. Note that as Re and Pr increase,
the Bi also increases, suggesting that the improvement of the
cross-plane in the fluid, H, and in the solid, HðsÞ . In both panels,
agreement in the prediction of the inner wall fluid temperature
the temperature fluctuations have been non-dimensionalized
is indeed related to the increasing dominance of the radial heat
using the reference temperature proposed by [5],
flux over the circumferential heat flux. This is confirmed in Fig. 5

@ H

Ri q i (d), where the relative differences between the peak inner wall
T ref ¼ Ri
¼ ; ð18Þ fluid temperatures predicted by the spectral and 1D models are
@r
j
r¼Ri plotted for all the cases in Table 1 as a function of the Biot number.
It can be observed that for Bi J 0:2 the error between the spectral
where H and qi are the circumferentially averaged temperature model and the 1D model is smaller than 5%, which provides a rule
fluctuation and heat flux at the inner wall, respectively. of thumb for the applicability of 1D models to the prediction of
As in Fig. 3, the left panel of Fig. 4 corresponds to the lowest Biot inner wall fluid temperatures.
number (case i, thin wall), whereas the right panel corresponds to
the highest Biot number (case xxx, thick wall). The color scale is 3.3. Thermal stresses
different in each panel, which for the same incident heat flux it
would result in a higher temperature in the left panel (where the In order to evaluate the thermal stresses on the walls it is com-
Re, and hence the flow velocity, is smaller). mon in the recent literature [43–46,18] to use the expressions pro-
In both cases the temperature on the solid is maximum where vided by Faupel and Fisher [24] for a circular pipe with a radial
the incident heat is maximum in the outer wall. Also, there are temperature distribution (i.a, as when the heat flux is uniform in
important temperature gradients in the circumferential direction the circumferential direction). These thermal stresses tensor com-
within the solid. However, for the high Bi case, those gradients ponents are reproduced here for completeness:
are small compared to the temperature gradient in the radial direc- !
Z Z
tion (between the inner and outer wall of the pipe). Temperatures C r 2  R2i Ro r

in the fluid are roughly uniform, except in the thermal boundary


rrr ¼ HðsÞ rdr  HðsÞ rdr ; ð19Þ
r2 R2o  R2i Ri Ri
layers near the hot side of the pipe. It is clear from the figure that Z Z !
C r 2 þ R2i Ro r
the thermal boundary layers are much larger in case i (Fig. 4a) than rhh ¼ HðsÞ rdr þ HðsÞ rdr  HðsÞ r2 ; ð20Þ
in case xxx (Fig. 4b), consistent with the variation in Re and Pr r2 R2o  R2i Ri Ri
between the two cases. Since the heat flux is roughly the same in

Fig. 4. Non-dimensional mean temperature, H=T ref , in the cross plane for (a) case i Re ’ 2500 (Res ¼ 180), Pr ¼ 0:7 and e=Ri ¼ 0:085; Bi ¼ 0:038 (b) case xxx ReR ’ 41; 000
(Res ¼ 2000), Pr ¼ 10 and e=Ri ¼ 0:183; Bi ¼ 2:879. Note that the scale is different in each panel.
262 C. Marugán-Cruz et al. / International Journal of Heat and Mass Transfer 96 (2016) 256–266

where C ¼ aE=ð1  mÞ; E is Young’s modulus, m is Poisson’s ratio, a is


the coefficient of thermal expansion and the cross term rrh is zero.
The longitudinal stress, rxx , is obtained from (19) and (20) by
imposing the condition of plane strain,
1
½rxx  mðrhh þ rrr Þ þ aHðsÞ ¼ 0; ð21Þ
E
yielding2
Z !
2m Ro
rxx ¼ C HðsÞ rdr  HðsÞ ; ð22Þ
R2o  R2i Ri

These expressions were originally developed for a temperature


distribution which is a function of the radius only, that is
HðsÞ ¼ HðsÞ ðrÞ, and they are often used in the literature. The ratio-
nale for the application of these formulas to non-uniform problem
has already been explained in the introduction: if the circumferen-
tial gradients of temperature are small compared to radial gradi-
ents, a 1D model with conduction only in the radial direction
might be applicable. However, it is apparent from Fig. 4 that in
all the cases presented in Table 1 the circumferential gradients
are comparable to the radial gradients, which would prevent the
application of these formulas.
A more general solution for the thermal stresses on a circular
pipe was obtained by Gatewood [47]. First, the temperature distri-
bution on the pipe is expressed as a series of sines and cosines,
X
1
HðsÞ ðr; hÞ ¼ A0 þ E0 log r þ ðAn r n þ C n rn Þ cos nh
n¼1
X
1
þ ðBn r n þ Dn r n Þ sin nh; ð23Þ
n¼1

which corresponds to the most general solution of the Laplace


equation in cylindrical coordinates. Then, the thermal stresses
are obtained from the temperature distribution by solving the
Duhamel–Neumann law, together with the equilibrium condition
for the solid and the boundary conditions of no surface forces.
Gatewood [47] obtained the remarkable result that the thermal
stresses only depend on the coefficients E0 ; C 1 and D1 of the temper-
ature distribution:
"    #
CE 0 Ro R2i R2o  r 2 Ro
rrr ðr; hÞ ¼ log  2 2 log
2 r r Ro  R2i Ri
C ðR2o  r 2 Þðr 2  R2i Þ
 ðC 1 cos h þ D1 sin hÞ; ð24Þ
2r 3 ðR2o þ R2i Þ
"     #
CE 0 Ro R2i R2o þ r 2 Ro
rhh ðr; hÞ ¼ log þ 2 2 log 1
2 r r Ro  R2i Ri
C 4R2i r 2  ð3r2  R2o Þðr 2 þ R2i Þ
 ðC 1 cos h þ D1 sin hÞ;
2r 3 ðR2o þ R2i Þ
ð25Þ

C ðR2o  r 2 Þðr 2  R2i Þ


rrh ðr; hÞ ¼  ðC 1 sin h  D1 cos hÞ: ð26Þ
2r 3 ðR2o þ R2i Þ
The computation of these coefficients (E0 ; C 1 and D1 ) once the
Fig. 5. (a–c) Circumferential variation of the inner wall fluid temperature for thermal field on the solid is known is trivial. Note also that for
different Reynolds number at a fixed thickness (e=Ri ¼ 0:0848). (a) Pr = 0.7. (b)
Pr = 3. (c) Pr = 10. Dashed lines correspond to the spectral method and solid lines to
the present configuration, coefficient C 1 ¼ 0 due to symmetry con-
1D model. (d) Error between the spectral model and the 1D model for the inner wall siderations. As with Faupel and Fisher’s solution, the longitudinal
fluid temperature at the location of maximum temperature (h ¼ p=2; r ¼ Ro ), as a stress rxx can be obtained from the cross-plane stresses by impos-
function of the Bi number. Circles correspond to e=Ri ¼ 0:1833, squares to ing the condition of plane strain, Eq. (21).
1
e=Ri ¼ 0:0848, dashed line correspond to a power law of Bi .

2
Please note that the formula given by Faupel and Fisher [24] for rxx contains a
typo.
C. Marugán-Cruz et al. / International Journal of Heat and Mass Transfer 96 (2016) 256–266 263

Comparing Eqs. (19) and (20) to (24)–(26), we can observe a few normalized with CT ref , and the figure shows from top to bottom
important differences. First, [47] gives a non-zero cross stress term, the components rrr ; rhh and rrh .
Eq. (26), which implies the existence of bending moments on the First, it is important to note that the scale is not the same for all
wall of the pipe. Also, when C 1 and D1 are zero, Gatewood’s solu- the panels: the magnitude of the tangential component, rhh , is con-
tion coincides with Faupel and Fisher’s. That is the case when the siderably larger than the other two. Second, the shape of the ther-
temperature on the solid depends on the radial coordinate only. mal stresses presents clear differences as the Bi number changes.
Note that, if the temperature is n-periodic in the circumferential When the Biot is small, the temperature gradients are larger in h.
direction with n P 2, Faupel and Fisher solution would yield the As a consequence, the thermal expansion of the solid is roughly
appropriate values for the stress if (and only if) the circumferen- uniform in the radial direction. Under this condition, the dilatation
tially averaged temperature is used in Eqs. (19)–(22). of the front part of the pipe results in inward bending stresses in
As an illustration, Fig. 6 shows the components of thermal stres- the front half of the pipe (compression inside, traction outside),
ses tensor calculated using the formulas proposed by Gatewood and outward bending stresses on the rear side of the pipe (traction
[47] for cases ii (low Biot) and xxx (high Biot). The stresses are inside, compression outside).

Fig. 6. Components of the thermal stresses tensor for different Biot numbers. (a) and (b) rrr =CT ref . (c) and (d) rhh =CT ref . (e) and (f) rrh CT ref . (a), (c) and (e) case ii. (b), (d) and
(f) case xxx.
264 C. Marugán-Cruz et al. / International Journal of Heat and Mass Transfer 96 (2016) 256–266

On the other hand, when the Biot is large the radial temperature Note that for the present configuration, the rhh component of
gradients are dominant. As a consequence, there is a large temper- the stress tensor dominates the effective stress. Fig. 7 shows this
ature difference between the outer and inner walls of the pipe, effective stress for three cases of the database, corresponding to
with the former suffering larger thermal expansion than the latter. low, intermediate and large values of the Biot number. The shape
This produces outward bending stresses in the front side of the of the effective stress obtained with the Faupel and Fischer’s solu-
pipe, and smaller thermal stresses on the rear side of the pipe. tion, Fig. 7(b), (d) and (f), is very similar for all Bi numbers,
In order to compare the thermal stresses obtained with the although their magnitude changes considerably. A large stress is
solutions of Faupel and Fischer and Gatewood, it seems convenient predicted where the heating is applied and a negligible stress is
to use the effective stress defined as predicted along the adiabatic part.
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi On the other hand, Gatewood’s solution (Fig. 7(a), (c) and (e))
ref ¼ ðrxx  rrr Þ2 þ ðrrr  rhh Þ2 þ ðrhh  rxx Þ2 þ 6ðr2rh þ r2rx þ r2xh Þ: shows a strong dependence of the thermal stresses on the Biot
ð27Þ number, as already observed in Fig. 6. The maximum thermal
stresses occur on the hot side for high Bi, while they occur on the

Fig. 7. Effective thermal stress, ref =CT ref , (a), (c) and (e) using Gatewood formulas. (b), (d) and (f) using Faupel and Fischer formulas. (a) and (b) case ii. (c) and (d) case xvi. (e)
and (f) case xxx.
C. Marugán-Cruz et al. / International Journal of Heat and Mass Transfer 96 (2016) 256–266 265

Fig. 8. (a) D1 coefficient as a function of the Biot number, for all cases studied. Circles, thick wall, e=Ri ¼ 0:183. Squares, thin wall, e=Ri ¼ 0:085. (b) Maximum effective
thermal stress at the hottest point in the wall (h ¼ p=2 at the outer wall), as a function for the Bi number. Black symbols, Pr ¼ 0:7. Red symbols, Pr ¼ 3. Blue symbols, Pr ¼ 10.
Circles, thick wall, e=Ri ¼ 0:183. Squares, thin wall, e=Ri ¼ 0:085. Close symbols, D1 > 0. Open symbols, D1 < 0. (For interpretation of the references to colour in this figure
legend, the reader is referred to the web version of this article.)

adiabatic side for low Bi. It is interesting to note that for interme-
diate Bi numbers (Bi  0:2), a roughly homogeneous distribution of
effective thermal stress is obtained.
In order to analyze in more detail the effective stresses from
Gatewood’s solution, lets consider the equation for the tangential
stresses (Eq. (25)), the largest contributors to ref , for the particular
case of C 1 ¼ 0:

rhh ¼ E0 f 0 ðrÞ þ D1 f 1 ðrÞ sin h: ð28Þ


In this equation f 0 and f 1 are functions of r only. It is clear that
the shape of the rhh (and hence ref ) is governed by the sign of coef-
ficient D1 . When D1 is close to zero, the thermal stresses are essen-
tially axisymmetric, as it happens for the intermediate Bi shown in
Fig. 7(c). Note that D1 ¼ 0 does not imply that the temperature dis-
tribution is axisymmetric since D2 ; D3 ; . . . ; DN , might all be non-
zero. When D1 is positive the maximum effective stress tends to
be in the adiabatic region, while when D1 is negative the maximum
effective stress occurs in the region where the heating is applied.
Fig. 9. Relative difference between the tangential stresses at the point of maximum
Fig. 8(a) shows the dependence of D1 with the Biot number, for temperature (h ¼ p=2 and r ¼ Ro ), computed using Faupel and Fisher formulas (rhh )
ðFÞ

all the cases considered here. We can observe that as Bi increases, and Gatewood’s formulas (rhh ). The relative error is plotted as a function of the Biot
ðGÞ

D1 decreases and eventually change signs, at Bi  0:1  0:2. The number. Black symbols, Pr ¼ 0:7. Red symbols, Pr ¼ 3. Blue symbols, Pr ¼ 10.
plot suggest that the actual threshold depends on the tube thick- Circles, thick wall, e=Ri ¼ 0:183. Squares, thin wall, e=Ri ¼ 0:085. Close symbols,
D1 > 0. Open symbols, D1 < 0. (For interpretation of the references to colour in this
ness. This is clearer when D1 is plotted against the local Bi (not
figure legend, the reader is referred to the web version of this article.)
shown here), based on the convective coefficient computed from
the spectral simulations at the location of the peak heat flux.
An interesting consequence of the change of sign of D1 is that clear that, even for Bi J 0:2 (which was the limit for obtaining rea-
the effective stress becomes minimal when D1 is zero. This is sonable predictions for the inner wall fluid temperature with the
hinted in Fig. 7, but becomes clear in Fig. 8(b), where the maximum local 1D model), the errors in the thermal stresses are large
ref obtained with Gatewood’s formula is plotted as a function of ( 50%). The plot suggests that, for Faupel and Fisher’s formulas
to provide an error smaller than 5%, the Bi must be at least  10.
the Bi number. In general, the smallest thermal stresses are
obtained for Bi  0:1  0:2 (where D1 crosses zero and the effective
stresses on the pipe are uniform), although there are three cases 4. Conclusions
with thin walls with minimum thermal stresses at Bi  0:05  0:1.
Finally, the results of Fig. 7 shows that for small Bi numbers A conjugate heat transfer numerical study on thin-walled pipes
Faupel and Fisher formulas predict qualitatively wrong thermal subjected to a non-uniform heat flux has been carried out. Differ-
stresses. Moreover, even if at large Bi the thermal stresses seem ent operation conditions have been imposed to study the role of
to be qualitatively right, they are still quantitatively wrong: while the Reynolds, Prandtl and Biot number. The results of the present
ref ;max ¼ 3:9  103 C T ref in Fig. 7(e), it is ref ;max ¼ 5:3  103 C T ref calculations have been compared to the estimations based on
in Fig. 7(f). This is emphasized in Fig. 9, where the relative differ- empirical correlations for the Nusselt number, where the heat
ence between the rhh obtained with Gatewood’s and Faupel and transfer is assumed to occur only in the radial direction (1D
Fisher formulas is plotted as a function of the Biot number. It is model).
266 C. Marugán-Cruz et al. / International Journal of Heat and Mass Transfer 96 (2016) 256–266

As expected, the higher inner wall fluid temperatures are found [13] M. Yang, X. Yang, X. Yang, J. Ding, Heat transfer enhancement and performance
of the molten salt receiver of a solar power tower, Appl. Energy 87 (2010)
near the maximum heat flux. The agreement between the spectral
2808–2811.
method and the 1D model increases as the Bi number increases, [14] B. Weingand, G. Gassner, The effect of wall conduction for the extended Graetz
although there is in general an underestimation of the maximum problem for laminar and turbulent channel flows, Int. J. Heat Mass Transfer 50
inner wall fluid temperature in the 1D model. As a rule of thumb, (2007) 1097–1105.
[15] K. Matsubara, M. Kobayashi, H. Maekawa, Direct numerical simulation of a
for Bi J 0:3 the difference between spectral and 1D models is neg- turbulent channel flow with a linear spanwise mean temperature gradient, Int.
ligible (smaller than 5%). J. Heat Mass Transfer 41 (1998).
From the point of view of the thermal stresses, Gatewood’s for- [16] A. Antoranz, A. Gonzalo, O. Flores, M. Garća-Villalba, Numerical simulation of
heat transfer in a pipe with non-homogenous thermal boundary conditions,
mulas should be used to compute the thermal stresses in non- Int. J. Heat Fluid Flow 55 (2015) 45–51.
uniform configurations, when HðsÞ depends on r and h. The results [17] M.A. Irfan, W. Chapman, Thermal stresses in radiant tubes due to axial,
circumferential and radial temperature distributions, Appl. Thermal Eng. 29
show that the stress distribution on the pipe depends strongly on (2009) 1913–1920.
the Biot number: for Bi < 0:1, the maximum thermal stresses are [18] O. Flores, C. Marugán-Cruz, D. Santana, M. Garća-Villalba, Thermal stresses
concentrated in the adiabatic side of the pipe. For Bi > 0:3, the analysis of a circular tube in a central receiver, Energy Procedia 49 (2014) 354–
362.
maximum thermal stresses appear near the peak heat flux, in the
[19] M. Rodŕguez-Sánchez, C. Marugán-Cruz, A. Acosta-Iborra, D. Santana,
hot side of the pipe. For Bi  0:1  0:3, the distribution of thermal Comparison of simplified heat transfer models and CFD simulations for
stresses is roughly uniform in the circumferential direction. molten salt external receiver, Therm. Eng. 73 (2014) 991–1003.
[20] J. Kim, J. Kim, W. Stein, Simplified heat loss model for central tower solar
Indeed, for that range of Bi the maximum thermal stress seems
receiver, Solar Energy 116 (2015) 314–322.
to be minimum. [21] P. Wang, D.Y. Liu, C. Xu, L. Zhou, L. Xia, Conjugate heat transfer modelling and
Finally, the thermal stresses obtained from Gatewood’s and asymmetric characteristic analysis of the heat collecting element for a
Faupel and Fisher’s formulas are compared. It is concluded that parabolic trough collector, Int. J. Therm. Sci 101 (2007) 68–84.
[22] G. Kolb, An Evaluation of Possible Next-Generation High-Temperature Molten-
Faupel and Fisher’s formulas do not provide quantitatively correct Salt Power Towers, Sandia Report. SAND2011-9320, 2011.
results for Bi K 10, and provide qualitatively wrong results for [23] L. Jianfeng, D. Jing, Y. Jianping, Heat transfer performance of an external
Bi K 0:3. This implies that, while 1D (local) models can provide rea- receiver pipe under unilateral concentrated solar radiation, Solar Energy 84
(2010) 1879–1887.
sonable good estimations of the inner wall fluid temperature in the [24] J. Faupel, F. Fisher, Engineering Design: A Synthesis of Stress Analysis and
fluid, a proper estimation of the thermal stresses on the solid Material Engineering, John Wiley and Sons, 1981.
requires obtaining the temperature distribution on the whole tube [25] W.C. Young, R.G. Budynas, Roark’s Formulas for Stress and Strain, McGraw-
Hill, 2002.
and using Gatewood’s formulas. [26] M. Eslami, R. Hetnarski, J. Ignaczak, N. Noda, Y. Tanigawa Sumi, Theory of
It should be noted that, even if the present results have been Elasticity and Thermal Stresses Explanations, Problems and Solutions,
obtained for a conductivity ratio ks =k ¼ 37:5, qualitatively similar Springer, 2013.
[27] D. Gartner, K. Johannsen, H. Ramm, Turbulent heat transfer in a circular tube
results are obtained for ks =k ¼ 18:5, supporting the present conclu- with circumferentially varying thermal boundary conditions, Int. J. Heat Mass
sions for (at least) lower conductivity ratios. Transfer 17 (1974) 1003–1018.
[28] R. Cess, A survey of the literature on heat transfer in turbulent tube flow.
Report 8-0529-R24, Westinghouse Research, 1958.
Acknowledgments [29] W.C. Reynolds, A.K.M.F. Hussain, The mechanics of an organized wave in
turbulent shear flow. Part 3. Theoretical models and comparisons with
experiments, J. Fluid Mech. 54 (2) (1972) 263–288.
This work was supported by the Spanish Ministry of Economy [30] J.C. del Álamo, J. Jiménez, Linear energy amplification in turbulent channels, J.
and Competitiveness through the Grant ENE2012-34255. O.F. and Fluid Mech. 559 (2006) 205–213.
M.G.-V. were partially supported by Grant TRA2013-41103-P of [31] G. Pujals, M. Garća-Villalba, C. Cossu, S. Depardon, A note on optimal transient
growth in turbulent channel flows, Phys. Fluids 21 (2009) 015109.
the Spanish Ministry of Economy and Competitiveness. The latter [32] X. Wu, P. Moin, A direct numerical simulation study on the mean velocity
grant includes FEDER funding. characteristics in turbulent pipe flow, J. Fluid Mech. 608 (2008) 81–112.
[33] A. Quarmby, R. Quirk, Measurements of the radial and tangential eddy
diffusivities of heat and mass in turbulent flow in a plain tube, Int. J. Heat Mass
References Transfer 15 (1972) 2309–2327.
[34] F. Schwertfirm, M. Manhart, DNS of passive scalar transport in turbulent
[1] A. Fatemi, L. Yang, Cumulative fatigue damage and life prediction theories: a channel flow at high Schmidt numbers, Int. J. Heat Fluid Flow 28 (2007) 1204–
survey of the state of the art for homogeneous materials, Int. J. Fatigue 20 (3) 1214.
(1998) 9–34. [35] L.N. Trefethen, Spectral methods in MATLAB, Siam, 2000.
[2] A. Bejan, 4th ed., Convection Heat Transfer, John Wiley and Sons, 2013. [36] J.A. Weideman, S. Reddy, A MATLAB differentiation matrix suite, ACM Trans.
[3] D. Johns, B. Neal, Thermal Stress Analyses, Pergamon Press, 1965. Math. Software (TOMS) 26 (4) (2000) 465–519.
[4] M.J. Lata, M. Rodŕguez, M.A. de Lara, High flux central receivers of molten salts [37] M. Piller, Direct numerical simulation of turbulent forced convection in a pipe,
for the new generation of commercial stand-alone solar power plants, J. Solar Int. J. Numer. Methods Fluids 49 (2005) 583–602.
Energy Eng. 130 (2008). 021002-1-5. [38] Esdu, 92003: Forced convection heat transfer in straight tubes: Part 1:
[5] W. Reynolds, Turbulent heat transfer in a circular tube with variable turbulent flow, ESDU, 1981.
circumferential heat flux, Int. J. Heat Mass Transfer 6 (1963) 445–454. [39] F.P. Incropera, D.P. Dewitt, T.L. Bergman, A.S. Lavine, 6th ed., Fundamentals of
[6] A. Rapier, Forced convection heat transfer in a circular tube with non-uniform Heat and Mass Transfer, John Wiley and Sons, 2007.
heat flux around the circumference, Int. J. Heat Mass Transfer 15 (1972) 527– [40] V. Gnielinski, New equation for heat and mass transfer in turbulent pipe and
537. channel flow, Int. Chem. Eng. 16 (1976) 359–368.
[7] E.M. Sparrow, F.N.D. Farias, Unsteady heat transfer in ducts with time varying [41] B.S. Petukhov, Heat transfer and friction in turbulent pipe flow with variable
inlet temperature and participating walls, Int. J. Heat Mass Transfer 11 (1968) physical properties, Adv. Heat Transfer 9 (6) (1966) 539–565.
837–853. [42] V. Gnielinski, On heat transfer in tubes, Int. J. Heat Mass Transfer 63 (2013)
[8] R.M. Cotta, D. Mikhailov, M.N. Osizik, Transient conjugated forced convection 134–140.
in ducts with periodically varying in inlet temperature, Int. J. Heat Mass [43] I. Alzahrnah, M.S. Hashmi, B. Yilbas, Thermal stresses in thick-walled pipes to
Transfer 30 (10) (1987) 2073–2082. fully developed laminar flow, J. Mater. Process. Technol. 118 (2001) 50–57.
[9] A. Hadiouche, K. Mansouri, Analysis of the effects of inlet temperature [44] H. Yapc, B. Albayrak, Numerical solutions of conjugate heat transfer and
frequency in fully developed laminar duct, J. Eng. Phys. Thermophys. 83 thermal stresses in a circular pipe externally heated with non-uniform heat
(2010) 380–392. flux, Energy Convers. Manage. 45 (6) (2004) 927–937.
[10] S. Bilir, Transient conjugated heat transfer in pipes involving two-dimensional [45] F. Wang, Y. Shuai, Y. Yuan, G. Yang, H. Tan, Thermal stress analysis of eccentric
wall and axial fluid conduction, Int. J. Heat Mass Transfer 45 (2002) 1781– tube receiver using concentrated solar radiation, Solar Energy 84 (10) (2010)
1788. 1809–1815.
[11] O. Aydin, M. Avci, T. Bali, M.E. Arici, Conjugate heat transfer in a duct with an [46] F. Wang, Y. Shuai, Y. Yuan, G. Yang, L. B., Effects of material selection on the
axially varying heat flux, Int. J. Heat Mass Transfer 76 (2014) 385–392. thermal stresses of tube receiver under concentrated solar irradiation, Mater.
[12] A. Black, E. Sparrow, Experiments on turbulent heat transfer in a tube with Des. 33 (2012) 284–291.
circumferentially varying thermal boundary conditions, J. Heat Transfer 89 (3) [47] B.E. Gatewood, Thermal stresses in long cylindrical bodies, London Edinburgh
(1967) 258–268. Dublin Philos. Mag. J. Sci. 32 (213) (1941) 282–301.

Das könnte Ihnen auch gefallen