0% found this document useful (0 votes)
103 views9 pages

Glass Sagging Simulation With Improved Calculation of Radiative Heat Transfer by The Optimized R

This document summarizes a study that models glass sagging using finite element analysis. Glass sagging is a process used to shape glass through controlled heating and cooling. The study models glass sagging in 2D and compares different approaches to modeling heat transfer, including neglecting radiation, using Rosseland approximation, and using an optimized reciprocity Monte Carlo method. The results show that neglecting radiation leads to large errors in predicted glass temperature distribution and final shape, while Rosseland approximation and predefined temperature provide acceptable but less accurate results than Monte Carlo simulation, especially for fine prediction of temperature over time. Accounting properly for radiative heat transfer is important for accurate modeling of the glass sagging process.
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
103 views9 pages

Glass Sagging Simulation With Improved Calculation of Radiative Heat Transfer by The Optimized R

This document summarizes a study that models glass sagging using finite element analysis. Glass sagging is a process used to shape glass through controlled heating and cooling. The study models glass sagging in 2D and compares different approaches to modeling heat transfer, including neglecting radiation, using Rosseland approximation, and using an optimized reciprocity Monte Carlo method. The results show that neglecting radiation leads to large errors in predicted glass temperature distribution and final shape, while Rosseland approximation and predefined temperature provide acceptable but less accurate results than Monte Carlo simulation, especially for fine prediction of temperature over time. Accounting properly for radiative heat transfer is important for accurate modeling of the glass sagging process.
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 9

International Journal of Heat and Mass Transfer 70 (2014) 215–223

Contents lists available at ScienceDirect

International Journal of Heat and Mass Transfer


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

Glass sagging simulation with improved calculation of radiative heat


transfer by the optimized reciprocity Monte Carlo method
B. Le Corre a,b,⇑, Anthony Collin a, Laëtitia Soudre-Bau c, Yves Meshaka b, Gérard Jeandel a
a
LEMTA, Université de Lorraine, Vandoeuvre-lès-Nancy 54500, France
b
IJL, Université de Lorraine, Nancy 54000, France
c
EEIGM, Université de Lorraine, Nancy 54000, France

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

Article history: Glass sagging is used to process glass industrial products such as windscreens, mirrors or lenses. A 2D
Received 22 July 2013 glass sagging process, simulated with the Finite Element Method (FEM), is presented in this work. Differ-
Received in revised form 18 October 2013 ent thermal cases are reviewed with special care brought to radiative transfer model, with an optimized
Accepted 28 October 2013
reciprocity Monte Carlo method used as the reference. Results show that ignoring radiative transfer is a
too rough hypothesis. This leads to large errors on the glass temperature distribution, on the forming pro-
cess and on the final shape in case of glass sagging without mold. However, predefining glass temperature
Keywords:
or using Rosseland approximation give acceptable results, less accurate than Monte Carlo simulations
Glass forming
Finite elements
especially for a fine prediction of the transfer as a function of time, but with smaller CPU times.
Monte Carlo method Ó 2013 Elsevier Ltd. All rights reserved.
Radiative heat transfer
Rosseland approximation

1. Introduction – optical surface roughness [9],


– important residual mechanical stress [10,11].
Glass sagging operation consists in forming a sheet or a plate of
glass by heating it in a furnace. Glass temperature rises and reaches These problems lead to material waste and expensive produc-
a work temperature at which viscosity is low enough to allow glass tion cost.
to sag under its own weight due to gravity. The main complexity of the process lies on the thermomechanical
In the case of windscreen processing, glass sagging is the most coupling. Glass viscosity is very dependent on temperature: at the
often used forming process [1,2]. Automotive windscreen produc- working stage, a temperature variation of a few dozens degrees
tion requires glass sheet bending in order to obtain complex can change the viscosity by several decades [12]. Conversely, when
shapes and a good optical surface. Glass sheets are placed on a ring glass shape changes, it modifies the thermal flux inside. Some
frame and the forming phase ends when windscreen reaches the authors conducted glass sagging experiments in order to better
desired shape [3]. Glass sagging is also used to produce spherical, understand and improve the process [4]. Experimental results can
aspherical and even freeform mirrors [4] or lenses [5]. These mir- also be used to validate numerical modeling of the forming operation
rors can be used in many industrial applications such as solar pan- [2,8]. Computer simulations are generally less expensive and time
els or scientific instruments, e.g., astronomical telescopes [6]. For consuming than experiments. They can also give many information
these products, the process ends when the object totally fits the on the influence of process parameters on the final product [13].
shape of a mold placed below as shown in Fig. 1. Numerically, mechanical equations are usually solved using a
In theory, glass sagging looks like a simple operation process, Finite Element Method (FEM) as in [14,15]. Concerning the thermal
but in practice, parameters need to be tuned accurately. If not, part, different methods have been used to determine temperature
some flaws can happen, for example: fields during glass forming processes. Two main hypotheses are
usually found in the literature: glass temperature is directly set
– sticking problems between glass and mold [7], without calculations [16] or temperature is evaluated thanks to a
– heterogeneity of plate thickness [8], heat transfer model but neglecting radiative transfer [17].
However, this last hypothesis could be too rough to give accurate
temperature distributions [18].
⇑ Corresponding author at: LEMTA, Université de Lorraine, Vandoeuvre-lès-Nancy
In this work, a two-dimensional glass sagging case is modeled
54500, France. Tel.: +33 3 83 68 46 87.
and different hypotheses on temperature distribution are
E-mail address: [email protected] (B.L. Corre).

0017-9310/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.ijheatmasstransfer.2013.10.068
216 B.L. Corre et al. / International Journal of Heat and Mass Transfer 70 (2014) 215–223

Glass plate expansion, strain tensor  can be written as the sum of elastic, vis-
cous and thermal strains:
 ¼ e þ v þ t : ð3Þ
e is the elastic strain tensor:
 
mold 1 1 1 1
e ¼ rþ  trðrÞI: ð4Þ
2G 3 3B 2G
(a) Initial geometry (b) Final geometry
tr is the matrix trace and I is the identity matrix. G is the shear mod-
Fig. 1. Glass sagging process. ulus and B is the bulk modulus. They are expressed as a function of
the Young’s Modulus E and the Poisson’s ratio mp :
reviewed. The setup involves a glass plate processed on a concave
E
mold. Mechanical modeling is achieved by FEM and glass behavior G¼ ; ð5Þ
is assumed to obey a Maxwell law at working temperature. 2ð1 þ mp Þ
Concerning the thermal part of the problem, four cases are E
considered: B¼ : ð6Þ
3ð1  2mp Þ
(i) Glass temperature is set uniform inside the plate; The viscous strain tensor v is linked to the stress tensor r using the
(ii) During the sagging process, conduction inside the medium relation:
and convection at the boundaries are considered but radia-  
1 1 1 1
tive transfer is neglected; _v ¼ rþ  trðrÞI: ð7Þ
2gs 3 3gb 2gs
(iii) Radiation is taken into account in addition to the other heat
transfer modes using Rosseland approximation; By analogy with the elastic part, gs and gb can be assimilated to vis-
(iv) Radiation calculations are conducted with an optimized cous shear modulus and viscous bulk modulus, respectively:
reciprocity Monte Carlo method (ORM). g
gs ¼ ; ð8Þ
2ð1 þ mg Þ
The last case (iv) is considered as the reference in this present
work as Monte Carlo methods are known to give accurate results, g
in spite of heavier computational (CPU) times [19]. Comparison gb ¼ : ð9Þ
3ð1  2mg Þ
of cases (i)–(ii) with case (iv) will highlight the relevance of taking
radiative transfer into account during the simulation of the glass mg is the viscous Poisson’s ratio and is assumed to be equal to the
sagging process. Then, comparing (iii) and (iv) simulation results Poisson’s ratio. Viscosity g is the key parameter during glass sagging
will give interesting information on the accuracy of Rosseland process. Its value is directly linked to the glass speed forming and is
approximation, knowing that the concept of radiative conductivity supposed to be controlled by a Williams–Landel–Ferry (WLF) law
derived from Rosseland model is sometimes used for glass forming such as:
process simulation [20].  
gðTÞ C 1 ðT  T 0 Þ
The first part of the present paper describes the physical model lg ¼ ; ð10Þ
of sagging process. Then, the second part deals with thermal
gðT 0 Þ C 2 þ ðT  T 0 Þ
numerical schemes and the validation of the Monte Carlo code. where gðT 0 Þ, C 1 and C 2 are empirical coefficients.
The next section is dedicated to the numerical setup and finally Finally, t is the thermal expansion term:
simulation results are presented and discussed. a
t ¼ DTI: ð11Þ
3
2. Physical model for sagging process
a is the thermal expansion coefficient and DT is the temperature
Ò variation.
Abaqus is used in the present work in order to address the
thermomechanical problem by the FEM. The model presentation
2.2. Heat balance equation and boundary conditions
is only focused on the main parameters involved in the mechanical
behavior laws and the specific model of radiative transfer which
In its classical form, the heat balance equation can be written:
has been developed and combined to the FEM code.
q cp T_ ¼ divðqc þ qr Þ; ð12Þ
2.1. Mechanical behavior laws
where T_ is the temperature time derivative, cp is the heat capacity
The continuity equation gives the displacement vector u as a per mass unit at constant pressure, qc and qr are the conductive
function of the stress tensor r: and radiative heat flux vectors, respectively. Convection is ne-
glected inside glass as deformation is very slow and piece deforma-
qu€ ¼ div r þ f; ð1Þ tion has a very little influence on temperature distribution. At the
where q stands for the density and f is the body forces vector. Then, glass/air interface, boundary conditions can be written:
 
in case of small deformation, displacements are linked with the KðT a  T g Þ þ qa!g  n ¼ qg!a þ kc gradðTÞ  n; ð13Þ
r r
strain tensor  according to the relation:
where n is the normal vector oriented outward from glass. T a is the
1
 ¼ ðgradðuÞþt gradðuÞÞ: ð2Þ air temperature and T g is the glass temperatures. K is a global heat
2 exchange coefficient [22] and kc is the thermal conductivity. qi!j r
Making the assumption that glass behaves like a Maxwell rheolog- corresponds to the radiative flux density crossing the interface
ical material [21] (from a rheological point of view, it corresponds between two media i and j. Radiative boundary conditions depend
to the connection in series of a dashpot behaving like a Newtonian on the glass properties, namely in the semitransparent and the
fluid with a spring obeying a Hooke’s law) which undergoes thermal opaque ranges. qi!j
r can be written:
B.L. Corre et al. / International Journal of Heat and Mass Transfer 70 (2014) 215–223 217

 
N Z
X mmax N Z
X mmax
qi!j  n ¼ qi!j i!j
rðSTÞ þ qrðopaqueÞ  n: ð14Þ
r div qFM
rðiÞ ¼ ðPea ea e
ij;m  P ji;m Þdm ¼ P i  P ea
ji;m dm; ð20Þ
j¼1 mmin j¼1 mmin
The subscripts ST (semitransparent) and opaque correspond to the
frequency ranges. The radiative heat flux in the semitransparent where Pei is the total radiative power emitted by the element i. N is
range is related with the transmissivity coefficient s, the spectral the total number of elements. Assuming that the element i of vol-
blackbody intensity I0m , the polar angle hi and the azimuthal angle ume V i is at uniform temperature T i and that the absorption coeffi-
ui : cient ji does not vary in the considered range, then Pei is equal to:
Z mmax Z 2p Z p=2 Z mmax
ST
qi!j
rðSTÞ  n ¼ sðhi ; nm ÞI0m ðT i ; nm;i Þ Pei ¼ 4pji V i I0m ðT i ; nm;i Þ dm: ð21Þ
mmin
ST
0 0 mmin

 cos hi sin hi dhi dui dm: ð15Þ The forward method has been widely used for the estimation of
min max radiative power exchanges between cells [31,32]. However, a reci-
m is the frequency. m and m are the boundaries of the semi-
ST ST
procal Monte Carlo method has been presented quite recently
transparent range. nm;i is the optical index of the medium i and
[33] in order to decrease the CPU time without impacting the accu-
nm ¼ nm;i =nm;j is the relative optical index. I0m is given by the Planck’s
racy. The method is based on the fact that the ratio between the Peaij;m
law [23]:
and P ea
ji;m powers is equal to the ratio of intensities emitted by the i
2hm3 n2m and j elements:
I0m ðT; nm Þ ¼ : ð16Þ
c20 ðehm=kT
 1Þ Pea I0m ðT i ; ni Þ
ij;m
¼ : ð22Þ
c0 is the light speed in the vacuum, h and k are the Planck’s and Pea
ji;m I0m ðT j ; nj Þ
Boltzmann’s constants, respectively. The transmissivity coefficient Tesse et al. [33] presented two methods called Emission Reciprocity
s is given by the Fresnel equation: Method (ERM) and Absorption Reciprocity Method (ARM) to calcu-
 2 late the source term div qrðiÞ of an element i. In the ERM, div qrðiÞ is
1 cos hi  n1
m Pðhi ; nm Þ expressed as a function of Pea
sðhi ; nm Þ ¼ 1  ij;m :
2 cos hi þ n1
m Pðhi ; nm Þ
 2 N Z
X mmax
1 cos hi  Pðhi ; nm Þ div qERM
rðiÞ ¼ Pexch
ij;m dm
 ð17Þ mmin
2 cos hi þ Pðhi ; nm Þ j¼1
!
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi N Z
X mmax
2 I0m ðT j ; nj Þ
with Pðhi ; nm Þ ¼ 1  n2 m sin hi . In the opaque range, the radiative ¼ Pea
ij;m 1 dm: ð23Þ
heat flux is: j¼1 mmin I0m ðT i ; ni Þ
Z mmax
opaque
In the ARM, div qrðiÞ is expressed as a function of Pea
ji;m :
qi!j
rðopaqueÞ  n ¼ pem I0m ðT i ; nm;i Þdm; ð18Þ
mmin
opaque N Z
X mmax
div qARM
rðiÞ ¼ Pexch
ji;m dm
em is the emissivity of the body. j¼1 mmin
!
N Z
X mmax
2.3. Radiative transfer I0m ðT i ; ni Þ
¼ Pea
ji;m  1 dm: ð24Þ
j¼1 mmin I0m ðT j ; nj Þ
1
For frequencies above 2000 cm , glass is usually assumed to
be semitransparent. Below 2000 cm1, glass is opaque and its Then, Dupoirieux et al. [34] proved that the power exchanged be-
emissivity is considered equal to 0.9 [24] as a first approxima- tween elements i and j should be calculated with ERM when
tion. Many methods have been developed in order to calculate T i > T j and with ARM if T i < T j in order to limit the standard devi-
the radiative transfer in semitransparent media, i.e., approxima- ation of div qrðiÞ .
tion methods (Rosseland and two-flux approximations, among This method which selects ARM or ERM depending on the
others) [25], spherical harmonics method [26], zonal method temperatures of elements i and j is called Optimized Reciprocity
[27], discrete ordinates method [28], finite element method Method (ORM). The ORM is particularly suitable for determining
[29], finite volume method [30] and Monte Carlo method [31]. temperature distribution in a low temperature gradient domain.
Rosseland and Monte Carlo methods will be used here. The In the case of glass sagging, the plate thickness is quite thin so tem-
Monte Carlo method is particularly suitable for complex media perature distribution is relatively homogeneous. This last point
and gives accurate results. It consists in simulating radiative clearly justifies the use of ORM as a reference method to calculate
transfer by tracking quanta of energy in all the domain. These radiative heat transfer.
quanta undergo a set of random events (absorption, transmis- In an easier way than ORM, Rosseland approximation can be
sion, reflection) that satisfy, on a statistical basis, the general used for optically thick materials. Rosseland approximation can
principles of thermal radiation. The main drawback of this give acceptable results with small CPU time even if the radiative
method is a large CPU time that is progressively offset by the heat flux near the boundaries may be still underestimated [35]. It
increasing power of computers and method improvements [19]. consists in considering radiative transfer as a diffusive flux inside
Assuming that the studied domain is discretized in boundary glass, assuming that radiation is absorbed at boundaries:
and volume elements, the idea is to calculate the radiative power qr ¼ kr gradðTÞ: ð25Þ
P exch
ij exchanged between two elements i and j. The spectral
radiative power P exch
ij;m can be expressed as: kr can be called radiative conductivity and is equal to [25]:
Z
Pexch
ij;m ¼ Pea
ij;m  Pea
ji;m : ð19Þ þ1
4p @I0m ðs; nm Þ
kr ¼ dm: ð26Þ
0 3bm @T
P ea
ij;m is the spectral radiative power emitted by the cell i and ab-
sorbed by the cell j. In the Forward Monte Carlo Method (FM), the bm is the spectral extinction coefficient. As there is no scattering in
radiative source term div qrðiÞ of an element i is calculated as: glass, bm can be replaced by the absorption coefficient jm . After some
218 B.L. Corre et al. / International Journal of Heat and Mass Transfer 70 (2014) 215–223

x = 0.5 L 1 1
Rousse (2000)

Dimensionless temperature
T3 + Present model
0.9 0.8

Relative error (%)


(N = 0.05)

0.8 error 0.6


(N = 1) (N = 0.05)
L κ n=1
T2 T4 0.7 0.4
error
(N = 1)
0.6 0.2
T1
y
0.5 0
0 0.2 0.4 0.6 0.8 1
L
x y (x=L/2)

Fig. 2. Numerical benchmark. Fig. 3. Results of the radiative code.

derivations and considering that the optical index nm does not 62,5mm
depend on temperature, kr finally becomes: 6mm

Z 2 glass
þ1
4p 2h m4 n2m ð1=T 2 Þðehm=kT Þ
kr ¼ dm: ð27Þ
0 3jm c20 k ½ehm=kT  1
2
R 65mm

The validity of this assumption has to be checked considering the y axis of


radiative properties of the medium of interest. In particular, the symmetry
mold
x
mean free path has to be shorter than the characteristics length of
the medium [25]. This will be discussed in the next sections based Fig. 4. Geometry of the studied case.
on the actual properties of the studied glass.

axis of
symmetry
3. Validation of the ORM radiative code in a pure thermal case C(xC  y C )

6mm
Glass temperature field is determined using the FEM combined 12 elem.
y
with the ORM radiative code which calculates the radiative source 62.5mm
term. This numerical setup is validated on a numerical benchmark x 125 elem.

consisting in a square enclosure of side length L where the medium


Fig. 5. Mesh.
is absorbing, non-scattering and the optical index is equal to 1
(Fig. 2). L is set to 1 m and the absorption coefficient is 1 m1 in or-
der to keep an optical thickness equal to 1, allowing comparisons 4.1. Geometry/mesh
with available data in [29]. The bottom wall is at temperature T 1
and the other walls are at temperature T 2 ¼ T 3 ¼ T 4 ¼ 0:5 T 1 . Emis- The geometry is given in Fig. 4. A 6 mm  125 mm glass plate is
sivity is equal to 1. Thermal conductivity kc varies depending on placed on a 65 mm radius concave mold. The plane defined by
the value of the Stark number N ¼ j kc =4rT 3 , which characterizes x ¼ 0 is used as a symmetry plane in order to decrease the calcula-
the ratio between conductive and radiative transfer. For the pres- tion cost. For all the simulations, a mesh with 12  125 elements is
ent simulations, N is set to 0.05 and 1. used (Fig. 5) with the C point corresponding to the center of the
In the Fig. 3, the dimensionless temperature (T 1 taken as the ref- plate. The finite elements are linear quadrilateral.
erence) is plotted versus y at x ¼ 0:5 L when the steady state is
reached. The results given by our model are compared white those
obtained with a Control Volume Finite Element Method on a 4.2. Glass properties
41  41 nodes grid [29]. There are very few differences between
the results given by the two codes. For the two values of N, the Glass characteristics [8,37] presented in Table 1 are assumed to
relative error is below 0.8%, which provides a verification of the not depend on temperature, except for viscosity (see Eq. (10)) and
present model. radiative conductivity (see Eq. (27)) when the Rosseland approxi-
As the Monte Carlo method is stochastic, results are sensitive to
the number of emitted quanta. It implies that results variance is in-
versely proportional to the sample size. It also explains why the Table 1
relative error values vary along the y-axis. In this model, 10 million Physical properties of glass.

quanta were emitted inside the medium and 1 million from each Property Unit Value
wall. C1 – 40
C2 K 710
g (T0 = 400 °C) Pa s 1019
4. Numerical model q kg m3 2483
kc W m1 K1 0:98
cp J K1 kg1 1140
The present numerical model is now adapted from an existing
E Pa 68:5  109
experiment presented in [36]. A glass plate is placed on a concave
mp – 0:22
mold in a furnace and has to fit the mold shape after heating. a K1 9:36  106
Details of the simulation are given below.
B.L. Corre et al. / International Journal of Heat and Mass Transfer 70 (2014) 215–223 219

Table 2 expected, it strongly increases with temperature, reaching values


Optical indices, absorption coefficient and mean free path in the [0.25 lm; 5.25 lm] more than six times higher than the standard heat conductivity
range.
indicated in Table 1.
k range (lm) n (–) k (–) j (m1) MFP (mm)
[0.25; 0.75] 1.528 1:492  10 7 3.8 263 4.3. Loading
[0.75; 1.25] 1.514 4:591  106 57.7 17.3
[1.25; 1.75] 1.508 4:234  106 35.5 28.2 During glass sagging experiment, glass plate is heated by infra-
[1.75; 2.25] 1.501 4:423  106 27.8 36.0 red heaters. The air temperature was measured in the furnace with
[2.25; 2.75] 1.494 6:324  106 31.8 31.4 thermocouples [40] and is represented as a function of time in
[2.75; 3.25] 1.485 7:432  105 311 3.22 Fig. 7. The radiation coming from the heaters is not exactly simu-
[3.25; 3.75] 1.474 1:065  104 382 2.62 lated in a straightforward manner. It is considered that radiation
[3.75; 4.25] 1.463 1:060  104 333 3.00 is coming from the surroundings, emitting like a high temperature
[4.25; 4.75] 1.449 2:978  104 832 1.20 blackbody source and providing the exact spectral and total
[4.75; 5.25] 1.397 3  103 7540 0.133
emissive power. Actually, radiation is mainly coming from the sur-
rounding walls, with temperature very close to the one presented
in Fig. 7 and with emissivity close to one. This is of course a first
mation is used. Radiative calculations are done considering a band approximation, which makes the calculations simpler however
model for the spectral dependence. and which will be discussed in a dedicated sensitivity study. From
The optical properties are listed in Table 2. These values are ta- a mechanical point of view, glass plate is only submitted to gravity.
ken from [38] and correspond to the properties at the center of Sliding contact between glass and mold is considered as it is ob-
each wavelength band, between 0.25 lm and 5.25 lm with a step served in practice for temperature under 680 °C [41]. Exchanges
of 0.5 lm. The corresponding mean free path of radiation (MFP) is between air and plate occurs according to a coefficient of
given in the last column in the Table 2. As it can be seen in the K ¼ 10 W m2 K1 [22].
range between 0.25 lm and 2.75 lm, the MFP is larger than the
thickness of the glass plate (6 mm). Consequently, the use of Rosse- 4.4. Numerical implementation
land approximation which requires an assumption of optically
thick medium (thickness equal to 3 to 5 times the MFP) may be Glass sagging is a process where thermal and mechanical mod-
questionnable. However, one should take into account that the els are closely coupled. Looking at Eqs. (10) and (11), temperature
highest temperature of the process (during working stage) is has a strong influence on strain tensor and viscosity. Then, when
672 °C, meaning that the proportion of radiative power below glass deforms, external geometry changes and so heat transfer
2.75 lm is about 18% [23] in the worst case. Moreover, the thick- equations (12) and (13) are affected. The model implies a strong
ness is taken here as a reference length, but for some of the radia- thermomechanical coupling. It is taken into account by the FEM
tion directions, total reflection will occur at the interface, which of by using an implicit solution. It means that reaching the next time
course leads to a larger optical thickness, whatever the wave- step can require several iterations before convergence between
length. Similarly, reflection phenomena will contribute to a higher heat transfer and mechanical models.
effective optical thickness even for directions only concerned by Concerning the heat balance, a numerical study has been car-
partial reflection. Hence, the assumption of an optically thick med- ried out showing that, when ORM is used, the radiative source term
ium holds. One supplementary required criterion has been pro- can be calculated only at the beginning of a time step to get the
posed in [39] in terms of applied temperature gradient, as follows: temperature field. However, this model has to be performed using
a 5 s time step and a density of 1000 quanta/mm2 for the medium
1 @T and 1000 quanta/mm for the boundaries in the Monte Carlo code.
< 0:033 jeff ð28Þ
T @n
involving the temperature gradient and with jeff being the Planck’s 5. Results and discussion
average of the absorption coefficient. Actually, jeff is equal to
535 m1 according to the values extracted from Table 2. Assuming Different models involving or neglecting radiative transfer are
that the temperature gradient is smaller along the x-axis and that used and discussed in this section.
the temperature variation is linear along the y-axis, the criterion re-
quires that DT < 0:033 jeff T Dy. At the working stage temperature 5.1. Comparison of different simulation results
(T ¼ 672 °C), this means that the temperature variation DT between
the center of the plate and the boundary (Dy ¼ 3 mm) must be less Fig. 8a shows the temperature of the glass plate center (point C
than 50 °C. This will be checked in the results section. in Fig. 5) versus process time for the four simulations. (PT) stands
Radiative conductivity kr is ploted in Fig. 6 as a function of glass for Predefined Temperature (glass temperature is assumed to be
temperature using the coefficients of Table 2 and Eq. (27). As equal to the surrounding temperature presented in Fig. 7), (RN)

5
Air temperature (°C)

600
4
kr (W.K−1.m )
−1

3 400
2
200
1

0 0
0 200 400 600 0 2000 4000 6000 8000 10000 12000
Temperature (°C) Time (s)

Fig. 6. Radiative conductivity as a function of temperature. Fig. 7. Air temperature during the process.
220 B.L. Corre et al. / International Journal of Heat and Mass Transfer 70 (2014) 215–223

(a) Plate center temperature Table 3


800 Process times.

Methods Beginning of the forming phase (s) Glass/mold contact (s)


600
Temperature ( oC)

(PT) 2934 3925


(RN) 3980 5449
(RA) 3030 4040
400 (ORM) 2980 3980
PT
ORM
200
RA
500
RN

Temperature ( C)
0

o
450
0 2000 4000 6000 8000 10000 12000 PT
Time (s) ORM
400 RA
(b) Plate center position along the y−axis RN
350
10
300
0 0 20 40 60
X (mm)
Position (mm)

−10
Fig. 9. Temperature along the x-axis at t = 2000 s.
−20
PT
ORM
−30 Fig. 9 shows the temperature on a central line at the middle
RA
−40
RN height of the plate (line initially defined by yC ¼ 0, which is pro-
gressively curved) at t = 2000 s. The difference in temperature be-
−50 tween (PT), (ORM) and (RA) results is below 30  C but reaches
2000 3000 4000 5000 6000
150 °C when comparing (RN) results with the other methods.
Time (s)
There is, of course, no variation of temperature along the x-axis
Fig. 8. Temperature (a) and displacement (b) of the plate center. for (PT). The variation between the center and the end of the plate
is about 5  C for (ORM), 10  C for (RA) and 25 °C for (RN). The dif-
for Radiation Neglected (the heat balance equation (12) is solved ference is larger for (RN) because radiation cannot contribute to
omitting the radiative flux term), (RA) for Rosseland Approxima- distribute the energy flux inside the medium in this case. Note that
tion and (ORM) for Optimized Reciprocity (Monte Carlo) Method. the maximum temperature discrepancy in the y-direction (not
It appears that the temperature of the glass plate center strongly shown here) is always below 3 °C, which is in accordance with
depends on the method used. the criterion presented in the previous section for the use of the
During the increase in temperature, (PT) shows the highest Rosseland approximation and explains why results obtained with
temperature. It is, of course, due to the fact that glass temperature the Rosseland approximation are satisfactory. Such a small tem-
is uniformly set to surrounding temperature, as if a perfect heat ex- perature gradient inside the medium even justifies the method of
change occured. Considering (ORM) and (RA), temperature shifts intensity linearization itself (which is used in the derivation of
from (PT) results during the first 2000 s. However, when the work- the Rosseland approximation) whatever the wavelength and the
ing stage temperature is reached (t = 4000 s), the three curves of optical thickness. This is another justification for the application
(PT), (ORM) and (RA) are overlapping. of a Fourier’s type law for the radiative model.
(RN) shows a significantly different behavior. During the Predicted plate shapes are shown in Fig. 10b at t = 12,000 s. The
increasing step, the discrepancy in temperature is very large as shapes provided with (PT), (ORM) and (RA) models are overlap-
compared to the other methods. The maximum difference is larger ping, whereas the shape obtained with (RN) model does not fit well
than 150  C at t = 2000 s. For (PT), (ORM) and (RA), the working to the mold.
stage temperature lasts from 4000 s to 5600 s. For (RN), the work- As a whole, all these results show that the hypothesis of
ing stage temperature is reached lately at t = 5600 s, then temper- neglecting radiation gives too rough predictions: (i) working stage
ature begins to decrease immediately following the setpoint in air temperature is reached too late, (ii) glass temperature is clearly
temperature. underestimated and (iii) the plate shape is not well simulated.
Vertical positions for the plate center as a function of time can When radiative transfer is taken into account hovewer, ORM,
be seen in Fig. 8b, in the time interval between 2000 s and 6000 s. Rosseland approximation and predefined temperature models
The initial position of the plate center is at 0 mm. At the beginning show weak discrepancies in the predictions, except during the time
of the process, plate lifts up slightly because of thermal expansion period of strong deformation which may appear with a small time
due to the increase in temperature. It explains why the positions gap when using the less accurate methods.
are above 0 mm in Fig. 8b at t = 2000 s. Calculations were performed on a HP Z820 workstation with an
For (PT), (ORM) and (RA) the forming phase begins at around Intel Xeon processor. Considering the CPU times required by each
t = 3000 s when viscosity becomes low enough to allow glass to method (Table 4), it seems that Rosseland approximation and to
sag under its own weight. This is observed numerically through some extent, predefined temperature model, provide acceptable
sign change in the vertical displacements. The vertical position of results with CPU times around 7 times inferior to those of ORM.
the plate center decreases and at around t = 4000 s, the bottom sur- This conclusion holds for glass sagging with mold, but other
face of the plate comes into contact with the mold. For (RN), this forming conditions may be found. In the case of windscreen pro-
phase begins later (starting at t = 4000 s) and plate comes into con- duction, there is no mold under the plate, but rather a skeleton
tact with the mold at 5450 s. frame which does not support the glass shape as a mold does. It
All the characteristic times, corresponding to the beginning of means that even for relatively similar time process durations, pre-
forming step and the time when the bottom surface of the plate dicted final shapes can be significantly different. This can be illus-
and the mold come into contact, are summed up in Table 3. trated considering results predicted in the previous case at an
B.L. Corre et al. / International Journal of Heat and Mass Transfer 70 (2014) 215–223 221

(a) t = 3855 s capacity per mass unit cp and the boundary conditions are investi-
20 gated, focusing on the ORM simulation.
Initial shape Results for simulations with K multiplied by 2 or divided by 2
10
are compared with the original simulation in Table 5. It appears
0 RN that, at t = 3000 s (time at which glass forming is beginning), taking
different values for K has no actual impact on the temperature of
−10 the glass plate center, the maximum difference being 0.5 °C. It does
Y (mm)

RA not change either the time at which the bottom surface of the plate
−20 ORM enters into contact with the mold.
Such low influence of K can be understood because radiative
−30
PT transfer is the dominant heat transfer mechanism. Moreover, the
−40
plate temperature has to follow the increasing set point, but with
a slow heating rate. For example at t = 3000 s, the temperature rate
−50 is 0.1 K s1, which means that an increase about 1 K occurs in a
time interval of t = 10 s. K has a weak influence with such a slow
0 10 20 30 40 50 60 70
process. Coupling between heat transfer and mechanics is another
X (mm)
possible explanation, as deformation occurs after a long stage of
heating and was even little affected by the choice of the radiation
(b) t = 12000 s
20 model as seen in Fig. 8.
Regarding the influence of cp in Table 6, the temperature of the
10 Initial shape glass plate center at t = 3000 s is 1.5 °C higher with
cp ¼ 912 J kg1 K1 (20%) than with the original value
0 cp ¼ 1140 J kg1 K1. As cp is smaller, glass temperature is rising
faster. Hence, the time required before contact between the middle
−10
of the plate and the mold is 16 s (0.4%) shorter than in the origi-
Y (mm)

RA
−20
nal simulation. With cp ¼ 1368 J kg1 K1 (+20%), the temperature
ORM
PT of the glass plate center decreases by 1.6 °C and the time before
−30 RN contact increases by 12 s (+0.3%).
This is again a weak influence and the explanation also comes
−40 from the slow heating rate. Some discrepancy may appear in this
first step in the temperature evolution but the global time before
−50
contact which is also affected by the coupling with the mechanics
0 10 20 30 40 50 60 70 shows little variations. Complementary simulations have been per-
X (mm) formed checking that the actual temperature modification in rela-
tion with cp was low. A maximum deviation of 15 K has been found
Fig. 10. Plate shape at 3855 s (a) and 12000 s (b).
between simulations carried out with average cp value and either
the minimum or the maximum cp value. This discrepancy occurs
at the early stage of the heating when the variation of set point
Table 4
CPU times.
with time is faster. However, the temperature rate decreases with
time and the results of the different simulations get very close (as
Methods CPU times (min) stated in Table 6). As a whole the total time before contact is little
Predefined temperature 24 affected.
Radiation neglected 25 As a conclusion, change in the K value does not affect the pro-
Rosseland approximation 27
Monte Carlo 185
cess and modifying cp has little influence.
The radiative boundary conditions are now modified in such a
way that the radiative intensity received by the top surface is dif-
ferent from the one of the other surfaces (Fig. 11).
intermediate instant during deformation and before contact of the The top surface will receive a radiative intensity from a black-
whole plate with the mold, as presented in Fig. 10a at t = 3855 s. body at the temperature T heaters which corresponds to the temper-
The plate center position predicted using a predefined temper- ature of the furnace heaters, while the other surfaces are still
ature model is about 34 mm below the initial position, whereas the submitted to radiation from black walls at temperature T air . In this
predictions with ORM and Rosseland approximation are 20 mm simulation, heaters of the furnace are considered at their maxi-
and 14 mm below the initial position, respectively. This shows mum power (maximum temperature T heaters ¼ 675 °C) from the
the possible advantage of involving the most accurate radiation beginning of the process until the working stage. Then, they follow
model in this case. Neglecting the radiative transfer (RN model), the same temperature evolution as the air and the walls (Fig. 12a).
the forming temperature has not been reached yet at this time. Obviously, the temperature of the plate center rises much faster
The plate shape is totally wrong only illustrating the above dis- with the modified boundary conditions (Fig. 12b). When forming
cussed effect of an upward movement due to thermal expansion begins at around t = 3000 s, the temperature difference with the
since the plate position is still above its initial position.
Table 5
Influence of K variations.
5.2. Sensitivity analysis
K (W K1 m2) Temperature at t = 3000 s (°C) Glass/mold contact (s)

In this section, a sensitivity analysis is carried out in order to 5 (50%) 606.8 (0.3) 3998 (+0.08%)
10 (Original model) 607.1 3995
check the dependence of the results to some questionable param-
20 (+100%) 607.6 (+0.5) 3991 (0.11%)
eters. The influence of the global transfer coefficient K, the heat
222 B.L. Corre et al. / International Journal of Heat and Mass Transfer 70 (2014) 215–223

Table 6 temperature value, displacements predicted by the different mod-


Influence of cp variations. els may be significantly different, the plate quickly reaching the
cp (J kg1 K1) Temperature at t = 3000 s Glass/mold contact same final shape however, due to the mold contact. Hence, this
(°C) (s) observation only holds for glass sagging with mold. During the
912 (20%) 608.6 (+1.5) 3979 (0.40%) forming stage, the shapes can be different and the model could
1140 (Original 607.1 3995 be more influent for forming cases without mold.
model) The sensitivity analysis has shown that the K global exchange
1368 (+20%) 605.5 (1.6) 4007 (+0.30%)
coefficient and the heat capacity per mass unit cp have a little influ-
ence on the results.
On the contrary, modification in the boundary conditions
changes the glass forming prediction. As a first approximation,
Iν0 (Theaters ) the hypothesis was used that radiation comes from a surrounding
axis of
symmetry Iν0 (Tair ) blackbody source. This is a strong assumption, which allowed the
Iν0 (Tair ) evaluation of the radiative influence and various model compari-
sons, however. The simulation should be now improved introduc-
Fig. 11. Modification of the boundary conditions. ing a more realistic radiative flux coming from the infrared heaters.
This would definitely allow checking the validity of Rosseland
approximation in comparison to ORM.
(a) Temperature
Temperature ( C)

600 Acknowledgments
o

T
400 heaters
ORM (modified BCs) The authors gratefully acknowledge the financial and technical
200 supports provided by the CERFAV (Centre Européen de Recherches
T
air et de Formation aux Arts Verriers), the FEDER (Fonds Européen de
ORM (original model)
0 Développement Régional) and the ANR (Agence Nationale de la
0 2000 4000 6000 8000 10000 12000
Recherche).
Time (s)

(b) Plate center position along the y−axis References


0 ORM (original model)
[1] P. Groombridge, A. Oloyede, P. Doherty-Bigara, Development and
Position (mm)

ORM (modified BCs)


implementation of visual feedback technology in automotive windscreen
−20 manufacture, Journal of Materials Processing Technology 139 (1) (2003) 357–
361.
[2] G. Buonanno, M. Dell’Isola, A. Frattolillo, G. Giovinco, Thermal analysis of a
−40 glass bending process, Applied Thermal Engineering 25 (2005) 2108–2121.
[3] R. Hunt, Numerical solution of the flow of thin viscous sheets under gravity
and the inverse windscreen sagging problem, International Journal for
2000 3000 4000 5000 6000
Numerical Methods in Fluids 38 (6) (2002) 533–553.
Time (s)
[4] Y. Chen, A.Y. Yi, Design and fabrication of freeform glass concentrating mirrors
using a high volume thermal slumping process, Solar Energy Materials & Solar
Fig. 12. Temperature (a) and displacement (b) evolution after modification of the Cells 95 (7) (2011) 1654–1664.
boundary conditions. [5] Y.M. Stokes, Numerical design tools for thermal replication of optical-quality
surfaces, Computers & Fluids 29 (4) (2000) 401–414.
[6] W.W. Zhang, K.W. Chang, D.A. Content, S.M. Owens, R. Petre, P.J. Serlemitsos,
original model is 16 °C (623 °C versus 607 °C). There is therefore an T.T. Saha, Y. Soong, Development of mirror segments for the constellation-x
observatory, SPIE Proceedings 4851 (2003) 503–518.
effect on the displacements. With the modified boundary condi- [7] D. Rieser, G. Spieß, P. Manns, Investigations on glass-to-mold sticking in the
tions, the contact between the bottom surface of the plate and hot forming process, Journal of Non-Crystalline Solids 354 (12-13) (2008)
the mold occurs at t = 3869 s, which is 126 s faster than the original 1393–1397.
[8] M.H. Parsa, M. Rad, M.R. Shahhosseini, M.H. Shahhosseini, Simulation of
prediction (contact at t = 3995 s). windscreen bending using viscoplastic formulation, Journal of Materials
For this model, the boundary conditions have consequently a Processing Technology 170 (1-2) (2005) 298–303.
stronger influence on the temperature field of the plate and the [9] A. Jain, A.Y. Yi, X. Xie, R. Sooryakumar, Finite element modelling of stress
relaxation in glass lens moulding using measured temperature-dependent
speed of forming than the K and cp parameters. elastic modulus and viscosity data of glass, Modelling and Simulation in
Materials Science and Engineering 14 (3) (2006) 465–477.
[10] W. Potze, F.H. Veld, J.P.V. den Brink, Structural and stress relaxation in a glass
6. Conclusion slab, International Journal of Forming Processes 2 (1999) 63–80.
[11] T. Soules, R.F. Busbey, S.M. Rekhson, A. Markovsky, M.A. Burke, Finite-element
Four heat transfer models have been investigated numerically, calculation of stresses in glass parts undergoing viscous relaxation, Journal of
the American Ceramic Society 70 (2) (2005) 90–95.
focusing on radiative transfer on a 2D sagging glass process. It ap- [12] J. Zarzycki, Glasses and the Vitreous State, Cambridge University Press, 1991.
pears that neglecting radiative heat transfer leads to large errors on [13] J.M.A. Cesar de sa, Numerical modelling of glass forming processes,
temperature distribution in the glass plate, which results in a sig- Engineering Computations 3 (4) (1986) 266–275.
[14] S. Manservisi, An optimal control approach to an inverse nonlinear elastic shell
nificant influence on the final shape of the piece. problem applied to car windscreen design, Computer Methods in Applied
On the contrary, direct setting of glass temperature or use of Mechanics and Engineering 189 (2) (2000) 463–480.
Rosseland approximation give results relatively close to those pre- [15] Y. Agnon, Y.M. Stokes, An inverse modelling technique for glass forming by
gravity sagging, European Journal of Mechanics B/Fluids 24 (3) (2005) 275–
dicted with ORM, which is considered as the reference method.
287.
Moreover, noticing that ORM requires large CPU times, Rosseland [16] E. Feulvarch, N. Moulin, P. Saillard, T. Lornage, J.-M. Bergheau, 3D simulation of
approximation seems to be a good alternative to get an acceptable glass forming process, Journal of Materials Processing Technology 164–165
accuracy with reasonable CPU times. (2005) 1197–1203.
[17] S. Grégoire, J.M.A. Cesar de sa, P. Moreau, D. Lochegnies, Modelling of heat
Note that some inaccuracies may occur at definite instants. transfer at glass/mould interface in press and blow forming processes,
When the plate is fastly changing after reaching a threshold Computers and Structures 85 (2007) 1194–1205.
B.L. Corre et al. / International Journal of Heat and Mass Transfer 70 (2014) 215–223 223

[18] A. Klar, J. Lang, M. Seaïd, Adaptive solutions of SPN-approximations to [31] T. Wong, M.P. Menguc, Comparison of Monte Carlo techniques to predict the
radiative heat transfer in glass, International Journal of Thermal Sciences 44 propagation of a collimated beam in participating media, Numerical Heat
(2005) 1013–1023. Transfer, Part B 42 (2) (2002) 119–140.
[19] J.R. Howell, The Monte Carlo method in radiative heat transfer, ASME Journal [32] A. Collin, P. Boulet, G. Parent, D. Lacroix, Numerical simulation of a water spray
of Heat Transfer 120 (3) (1998) 547–560. – Radiation attenuation related to spray dynamics, International Journal of
[20] J. Yan, T. Zhou, J. Masuda, T. Kuriyagawa, Modeling high-temperature glass Thermal Sciences 46 (9) (2007) 856–868.
molding process by coupling heat transfer and viscous deformation analysis, [33] L. Tessé, F. Dupoirieux, B. Zamuner, J. Taine, Radiative transfer in real gases
Precision Engineering 33 (2009) 150–159. using reciprocal and forward Monte Carlo methods and a correlated-k
[21] F. Richter, Upsetting and viscoelasticity of vitreous SiO2: experiments, approach, International Journal of Heat and Mass Transfer 45 (13) (2002)
interpretation and simulation, Ph.D. thesis, Technischen Universität Berlin, 2797–2814.
2006. [34] F. Dupoirieux, L. Tessé, S. Avila, J. Taine, An optimized reciprocity Monte Carlo
[22] N. Berour, D. Lacroix, G. Jeandel, Radiative and conductive heat exchanges in method for the calculation of radiative transfer in media of various optical
high-temperature glass melt with the finite-volume method approach. thicknesses, International Journal of Heat and Mass Transfer 49 (7) (2006)
Influence of several spatial differencing schemes on RTE solution, Numerical 1310–1319.
Heat Transfer, Part A 49 (6) (2006) 567–588. [35] K.H. Lee, R. Viskanta, Comparison of the diffusion approximation and the
[23] M.F. Modest, Radiative heat transfer, McGraw-Hill Series in Mechanical discrete ordinates method for the investigation of heat transfer in glass,
Engineering (1993). Glastechnische Berichte-Glass Science and Technology 72 (8) (1999) 254–265.
[24] K.H. Lee, R. Viskanta, Two-dimensional combined conduction and radiation [36] L. Soudre-Bau, Y. Meshaka, G. Parent, P. Boulet, B.L. Corre, G. Jeandel, Combined
heat transfer: comparison of the discrete ordinates method and the diffusion temperature and deformation measurement during glass forming in a real
approximation methods, Numerical Heat Transfer, Part A 39 (3) (2001) 205– scale setup Experimental Mechanics 53 (2013), http://dx.doi.org/10.1007/
225. s11340-013-9764-z.
[25] R. Siegel, J.R. Howell, Thermal Radiation Heat Transfer, Taylor & Francis, 1992. [37] L. Su, P. He, A.Y. Yi, Investigation of glass thickness effect on thermal slumping
[26] M. Modest, S. Lei, The simplified spherical harmonics method for radiative by experimental and numerical methods, Journal of Materials Processing
heat transfer, Journal of Physics: Conference Series 369 (2012). Technology 211 (12) (2011) 1995–2003.
[27] R. Méchi, H. Farhat, K. Guedri, K. Halouani, R. Said, Extension of the zonal [38] M. Rubin, Optical properties of soda lime silica glasses, Solar Energy Materials
method to inhomogeneous non-grey semi-transparent medium, Energy 35 (1) 12 (1985) 275–288.
(2010) 1–15. [39] H. Gomart, J. Taine, Validity criterion of the radiative fourier’s law for an
[28] H.M. Koo, R. Vaillon, V. Goutiere, V.L. Dez, H. Cha, T.H. Song, Comparison of absorbing and scattering medium, Physical Review E 83 (2011). paper 021202.
three discrete ordinates methods applied to two-dimensional curved [40] L. Soudre, Numerical and experimental study of glass plate forming, Ph.D.
geometries, International Journal of Thermal Sciences 42 (4) (2003) 343–359. thesis, Université Henri Poincaré, 2008.
[29] D.R. Rousse, G. Gautier, J.F. Sacadura, Numerical predictions of two- [41] D. Lochegnies, T. Vicart, J. Oudin, Finite element elastovicoplastic models for
dimensional conduction, convection, and radiation heat transfer. II. optimisation of horizontal and vertical press bending of sheets of glass, Glass
Validation, International Journal of Thermal Sciences 39 (3) (2000) 332–353. Technology 36 (1995) 166–170.
[30] S.V. Patankar, Numerical Heat Transfer and Fluid Flow, Hemispher Publishing
Corporation, 1980.

You might also like