Professional Documents
Culture Documents
Composite Structures
journal homepage: www.elsevier.com/locate/compstruct
a r t i c l e i n f o a b s t r a c t
Article history: The distribution of residual thermo-elastic stresses in encapsulated solar cells arising from lamination is
Available online 28 October 2015 relevant for the characterization of the long term performance of photovoltaic (PV) modules during ser-
vice. Accurate modelling of the structural response of the laminate in the transient regime during cooling
Keywords: after lamination is a challenging task from the computational point of view. In this work we propose a
Photovoltaics semi-analytic model based on the Kirchhoff plate theory and the shear-lag approach for the treatment
Residual stresses of the polymeric encapsulant layers and accounting for their time and temperature dependency according
Laminates
to a rheological model derived from fractional calculus considerations. Spatially uniform and non-uniform
Shear-lag theory
Thermo-visco-elasticity
temperature distributions are compared to accurately assess the amount of the residual compressive
Polymeric encapsulant stresses raised in the Silicon cells after lamination. The use of more realistic non-uniform temperature
distributions leads to lower residual compressive stresses in Silicon as compared to the uniform case.
2015 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.compstruct.2015.10.023
0263-8223/ 2015 Elsevier Ltd. All rights reserved.
482 S.O. Ojo, M. Paggi / Composite Structures 136 (2016) 481492
epoxy vinyl acetate (EVA), which also undergoes microstructural tively. According to Fourier law, the heat flows in the x1 , x2 and z
transformation during lamination. Observation shows that the direction can be related to the temperature gradients as follows:
temperature dependent properties of the EVA at the interface @T
influences the stress distribution in the PV laminate [14,15], which z kz
q 2a
@z
in turn affects the long term reliability of PV modules.
To accurately determine and model the temperature dependent @T
properties of EVA, uniaxial relaxation tests have to be performed x kx1
q 2b
@x1
and then traditionally interpreted by using rheological models
la Maxwell. However, these models are in general not easy to fit @T
to real experimental data, thus requiring a significant number of y kx2
q 2c
@x2
free parameters [16]. Based on the work of Di Paola et al. [17,18],
a model based on fractional calculus with only two free parameters By introducing Eq. (2) into (1) and considering a finite differ-
can effectively fit experimental trends and greatly simplify the pro- ence discretization in space, we obtain [11]:
cedure of parameters identification. As a further step with respect @T i;j;k
to [17,18], in the case of a microstructural transformation of the C i;j;k V i;j;k Q i1;j;k Q i1;j;k Q i;j1;k Q i;j1;k Q i;j;k1 Q i;j;k1
@t 2 2 2 2 2 2
K i;j1;k T i;j1;k T i;j;k K i;j1;k T i;j1;k T i;j;k A constant temperature is imposed at the top z 0 and bot-
2 2
tom z Z sides of the laminate:
K i;j;k1 T i;j;k1 T i;j;k K i;j;k1 T i;j;k1 T i;j;k 7
2 2 Tx1 ; x2 ; 0; t Tx1 ; x2 ; Z; t T p 11
By considering that the materials of the layers composing
where T p is the temperature of the press. Heat is absorbed only from
the module are continuous along the x1 and x2 directions,
the lateral sides of the laminate. Therefore, Robin (mixed) boundary
K i1;j;k K i1;j;k K i and K i1;j;k K i1;j;k K j , thus simplifying
2 2 2 2 conditions are imposed at the laminate sides to absorb heat away
Eq. (7) as: from the laminate, i.e., for t > 0:
Q i;j;k K i1;j;k T i1;j;k K i1;j;k T i1;j;k K i;j1;k T i;j1;k K i;j1;k T i;j1;k @TX; x2 ; z; t
2 2 2 2 k hTX; x2 ; z; t hT p 12a
K i;j;k1 T i;j;k1 K i;j;k1 T i;j;k1 @x1
2 2
Substituting Eq. (8) into the overall heat conduction Eq. (3), we
have, for a single cell (see Fig. 2): @Tx1 ; Y; z; t
k hTx1 ; Y; z; t hT p 12c
@x2
DT i;j;k
C i;j;k V i;j;k Q i;j;k 9
Dt @Tx1 ; 0; z; t
k hTx1 ; 0; z; t hT p 12d
During module lamination, the stack is kept hot at 150 C for @x2
about 15 min during which the EVA partially melts and provides
h is the convection coefficient of the air in the cooling press.
the sealing. Afterwards, it is cooled down to the ambient
To solve the thermal problem, a backward Euler implicit time
temperature in the cooling press and the EVA becomes solid. Con-
integration scheme is employed. In this method, a differential
sidering that the press is large and made of a highly conductive
equation of the form:
material, this system can be modelled as a heat sink providing a
uniform temperature. On this basis, the controlled volume to be dy
f t; y 13
analysed can be restricted to the laminate and the temperature dt
at its top and at bottom sides can be set equal to the press
is integrated as
temperature.
yt k1 yt k hf t k1 ; yt k1 14
where yt k1 and ytk denote the approximate solutions of the dif-
ferential equation at t t k1 and t t k , respectively. A total cooling
period of 30 min is specified. The topmost (glass) and bottom (back-
sheet) layers are maintained at a constant temperature of 25 C
(298 K) imposed by the air. Fig. 3 shows the temperature contour
after 30 min of cooling in the x2 z plane of the module.
The solution of the heat equation shows that the temperature
profile of the module is symmetric along the longitudinal
axis, due to the same boundary conditions specified at the edges
of the laminate. In the thickness direction, the degree of symmetry
in the temperature variation is very high due to the same boundary
constraints imposed at the top and bottom of the PV stack.
-3 -3
x 10 x 10
5 5
330 400
4 325 4 380
320
3 3 360
315
2 2 340
310
1 305 1 320
300 300
0 0.05 0.1 0 0.05 0.1
Longitudinal axis (m) Longitudinal axis (m)
(a) (b)
(a)
z
Fig. 3. Temperature contour (K) in the plane x2 z of the PV laminate along the x1 direction. (a) x1 0 (at the boundary of the laminate); (b) x1 31:3 mm (in the middle of
the portion of the laminate); (c) sketch of the laminate.
rx2 ;i Ei ai DT i Ei e0 Ei !z 16
and the unknowns e0 and ! can be determined by imposing the
conditions of vanishing axial force and bending moment:
n Z
X
2
zi
1
rx2 ;i xdz 0 17a
i1 zi
n Z
X
2
zi
Fig. 4. Cross-section of the PV module used for the stress analysis with perfectly 1
rx2 ;i xzdz 0 17b
i1 zi
bonded interfaces.
1 2
where zi and zi are, respectively, the upper and lower interfaces
of the i th layer, measured from the x2 z plane, with thickness
Table 1 2 1
Material properties of the layers composing the laminate. hi zi zi , while x denotes the out of plane thickness. By intro-
ducing Eq. (16) into (17), we obtain a set of equations in matrix
Glass EVA layer Silicon Backsheet
form:
layer layer
Youngs modulus 73 103 Viscoelastic 130 103 2.83 103 M 11 M 12 e0 V1
18
(MPa) M 21 M 22 ! V2
Coefficient of thermal 8 106 2.7 104 2.49 106 5.04 105
expansion (1/C) where the coefficients M11 , M12 , M 21 , M22 , V 1 and V 2 are given by
Poissons ratio 0.23 0.35 0.28 0.40
n Z
X
2
zi
M11 Ei dz 19a
rx2 ;i i1
1
zi
ex2 ;i ai DT i e0 !z for i 1; 2; . . . ; 5 15
Ei
n Z
X
2
zi
where e0 and ! represent, respectively, the longitudinal strain and M12 M21 Ei zdz 19b
the beam curvature at x2 0. In Eq. (15), Ei , ai and DT i are, respec- i1
1
zi
tively, the Youngs modulus, coefficient of thermal expansion and
change in temperature of a generic layer. n Z
X
2
zi
Eq. (16) allows the computation of the stresses rx2 at an arbi- M22 Ei z2 dz 19c
1
i1 zi
trary position along z:
S.O. Ojo, M. Paggi / Composite Structures 136 (2016) 481492 485
n Z
X
2
zi If the strip has dimensions in x2 much larger than in x1 , all the
V1 Ei ai DT i dz; 19d derivatives w.r.t. x1 can be approximately neglected. Hence, r1 and
1
i1 zi
r2 are respectively reduced to:
" #
n Z
X
2
zi 2
Et d w dU 2
V2 Ei ai DT i zdz 19e r1 z a DT t 23a
i1
1
zi 1 t2 dx2
2 dx2
Having computed e0 and !, stresses rx2 ;i and strains ex2 ;i can be " 2
#
determined at any point within the layers of the module. E d w dU 2
r2 z 2 aDT1 t 23b
1 t2 dx2 dx2
4. Shear-lag model with EVA treated as a shear deformable zero
thickness layer where t 1t t.
The transverse strain ezz is given by:
In this alternative structural model, relative displacements are dw
admitted from one layer to the next, as the EVA layer is modelled ezz aDT 0 24
dz
as an adhesive with zero thickness (see Fig. 5). As compared to the
simplified shear-lag theory proposed in [4], which does not Integrating Eq. (24) over the thickness z leads to:
account for the effect of normal peeling tractions in the shear-lag w Wx2 aDTz 25
formulation and satisfies the rotational equilibrium only in an
approximate way, the present formulation accounts for both peel- where W is the mid-plate deflection.
ing and shearing tractions. Each of the layer is modelled as a plate By considering the differential element, dx1 dx2 dz in Fig. 6, we
based on relaxed Kirchhoffs hypothesis [5] in which the displace- can write equations for this differential element in equilibrium as:
ments in the 1 and 2 directions for a thin plate having axes x1 , x2 X X X
and z are: F x1 0 F x2 0 Fz 0 26
@w By neglecting body forces, the equilibrium equations yield:
u1 z U 1 x1 ; x2 20a
@x1
@ r1 @ r12 @ rz1
0 27a
@w @x1 @x2 @z
u2 z U 2 x1 ; x2 20b
@x2
@ r12 @ r2 @ rz2
where U 1 and U 2 represent the displacements in the middle plane 0 27b
@x1 @x2 @z
and w is the deflection of the plate. Strains e1 and e2 are given by:
Fig. 5. PV module cross-section for the shear-lag model. where m r2z;c r2z;c ; r2z;c r2z x2 ; c; r2z;c r2z x2 ; c.
486 S.O. Ojo, M. Paggi / Composite Structures 136 (2016) 481492
d2 U 2
4 3 Taking n r2z;c r2z;c while noting that Eh
1t2 dx2
dN 2
leads
E d w E d U2 2 dx2
z3 3c2 z 2c3 z c2
2
rzz to
61 t2 dx42 21 t2 dx32
1 dm dN2
z c rzz;c n0 41
2 dx2 dx2
Adding Eqs. (32a) and (32b) and simplifying, we get: By similarly subtracting Eq. (32b) from (32a) for the top and
bottom of the plate and simplifying, we get:
4 3
E d w E d U2 2
rzz z3 3c2 z z c2 4Ec3 d w dm
4
61 t2 dx42 21 t2 dx32 c rzz;c rzz;c 0 42
61 t2 dx42 dx2
1 dm p
z 33
2 dx2 2 3
d4 w
Taking q rzz;c rzz;c while noting that 61
4Ec
t2 dx42
dx
dm
2
c dQ
dx2
2
i i1 si
u2 u2 49a
Kxi
ri
wi1 wi for i 1; 2 49b
Kyi
where Ki i
x and Ky are the longitudinal and transverse compli-
ances at the interface i.
In this work, we assume the same value for Ki i
x and Ky which
are given by
EE
Ki i
x Ky x 50
hE
where EE and hE are the Youngs modulus and thickness of the EVA
adhesive layer, and x is the out of plane thickness.
Altogether there are 10 governing equations consisting of 6
equilibrium and 4 continuity equations. By substituting for the
constitutive terms in these equations and further simplification,
we have a system of 6 higher order ordinary differential equations
which has to be solved to obtain the variables U k , W k and their Fig. 6. A differential element subject to internal stresses.
derivatives (for k 1; 2; 3) while si and ri (for i = 1,2) are calcu-
lated from the continuity equations, i.e., by Eq. (49). This formulation has the advantage over Prony series to
The solution of the problem can be achieved by converting the involve only two parameters (a and a) to model the viscoelastic
system of higher order ordinary differential equations to a system behaviour of the EVA accurately. To obtain the relaxation modu-
of first order ordinary differential equations which can then be lus Et of the EVA during cooling, identified parameters for
solved to determine the stress distribution in the plate. For this experimental uniaxial relaxation tests at different temperatures
case, we can write: have been fitted in [19] to determine a and a as functions of
temperature change. Due to modification in the material
dv x2 microstructure of EVA at DT 84 C, two different correlations
Btx2 F 51
dx2 are used to accurately fit the experimental data. The correlations
for a and a in Fig. 7 are mathematically represented as
t f U k ; U 0k ; W k ; W 0k ; W 00k ; W 000k ; for k 1; 2; 3 (Tref = 20 C):
F f Kxi ; Ki
y ; DT; ak constant; for i 1; 2 (
6:5 107 DT 3 1 104 DT 2 0:0093DT 0:225 DT 84 C
a
where B is an 18 18 sparse coefficient matrix which contains 1:2 107 DT 3 5:4 105 DT 2 0:0083DT 0:474 DT > 84 C
constant coefficients of the variables and their derivatives and 55a
F is 18 1 vector (see Appendix A). The set of boundary conditions
for this system of ODEs are (for i = 1, 2 and k = 1, 2, 3): 733:5exp 0:26DT 81:2exp 0:04DT DT 84 C
a 6
k
U k 0; Q 2 0; si 0; at x2 0 52a 6:5 10 exp 0:17DT 1:7exp 0:004DT DT > 84 C
55b
k k
M 2 0; N 2 0; at x2 L: 52b
We can use the temperature distribution history of the module
computed during cooling after lamination to evaluate the temper-
4.2. Modelling viscoelasticity of EVA and transient thermo-elastic ature and time dependent properties of the EVA adhesive layer.
analysis After cooling of the PV laminate, the temperature of the layers var-
ies along the longitudinal coordinate for each layer and along the
To account for the visco-elastic behaviour of the EVA, the rheo- transverse coordinate for the whole module. Thus, the interfacial
logical model described in [19] based on fractional calculus is compliance to be inserted in the shear lag constitutive equations
herein considered in which the constitutive equation of the sim- now varies with temperature and time, as well as along the longi-
plest fractional element (also known as ScottBlair element) is tudinal axis of the module, which is a novelty with respect to stan-
given as: dard shear lag theories that assume constant compliances. Hence,
a this leads to a non-homogeneous system of ordinary differential
d et equations with time-dependent coefficients:
rt a a ; a 2 0; 1 53
dt
and the relaxation modulus assumes the power law form:
dv x2 ; t
Bx2 ; tv x2 ; t Fx2 ; t 56
t a dx2
Et a 54
C1 a where F f Kx x2 ; t; Ky x2 ; t; Tx2 ; t; a
a has the SI unit of MPa sa and its mechanical meaning varies with For non-uniform temperature simulations, a model with only
from a stiffness (a 0) to a viscosity a 1. Time t is measured in the longitudinal compliance is considered. The compressive stres-
seconds. ses in Silicon are computed at some relevant time intervals.
488 S.O. Ojo, M. Paggi / Composite Structures 136 (2016) 481492
Fig. 7. Correlations for a and a as a function of the change of temperature. Tref corresponds to 20 C.
5. Numerical examples For the shear deformable interface configuration with 3 layers
and 2 interfaces (glass-Silicon interface 1 and Silicon-backsheet
An exact solution of the system of ordinary differential interface 2), 2 models are examined:
equations can be achieved for the special case of a uniform
temperature profile for the entire PV module after cooling. In (1) Shear deformable interfaces with longitudinal compliance
the case of a non-uniform temperature analysis, the coefficients only.
in the B matrix vary with time and space, so a numerical (2) Shear deformable interfaces with longitudinal and trans-
method is required. Specifically, for the non-uniform tempera- verse compliances of equal magnitude.
ture case study, an integration scheme using a trapezium rule
method (an average of forward and backward Euler method) is It can be deduced from the results shown in Fig. 9 that the
used. To verify the accuracy of this numerical scheme for this interfacial stress distribution for model 1 and model 2 are almost
analysis, a comparison was made between the exact solution the same. This result is not unexpected, since we assume an
of the uniform temperature analysis and the numerical solution isotropic adhesive layer at the interfaces for model 1 and model
using trapezium rule method and a good agreement was 2 (i.e., Kx = Ky). The effect of the transverse compliance will obvi-
achieved. The results of this comparison are shown in the next ously be significant when an orthotropic adhesive (i.e. Kx Ky )
section. interface is considered.
The results of the interfacial stress distribution pinpoint that
the magnitude of the shear stresses and the peeling stresses
5.1. Uniform temperature across the PV module
increases towards the edge of the laminate. Therefore, those points
are critical locations for possible delamination. It is observed in
We consider a cooling down of the module from 150 C
Fig. 10 that the mean compressive stress in the Silicon layer is
to the room temperature (25 C) in 30 min. At this final tem-
reduced for the shear deformable interface configuration with
perature, the EVA relaxation modulus E(t) is calculated to be
respect to the perfect interface configuration, due to the finite com-
6 MPa.
pliance of the interfaces, which is a more realistic situation in
In the case of perfectly bonded interfaces and EVA layers
practise.
treated as continuum layers with their thicknesses, the axial
stress and strain are plotted in Fig. 8. The Silicon layer results
5.2. Non-uniform temperature
to be in a compressive stress state above 60 MPa. The thermal
strain and the mechanical strain almost counteract each other,
The stress formulation for this analysis involves a shear deform-
which makes the total strain to be almost zero in the entire
able interface with only longitudinal compliance, since the effect of
module.
the transverse compliance was already found to be negligible. The
(a) (b)
Fig. 8. (a) Stress and (b) strain distributions for a perfect interface configuration of uniform temperature analysis.
S.O. Ojo, M. Paggi / Composite Structures 136 (2016) 481492 489
Fig. 9. Shear stress at (a) interface 2, (b) interface 1 for a uniform temperature analysis.
(a) (b)
Fig. 10. (a) Peeling stress along interfaces 1 and 2; (b) silicon compressive stress for model 2.
0.4 0.2
Shear stress at interface 2(MPa)
Shear stress at interface 1(MPa)
Exact Exact
0.3 Numerical Numerical
0
0.2
-0.2
0.1
-0.4
0
-0.1 -0.6
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
X2 / L X2 / L
(a) (b)
Fig. 11. Exact and numerical solution for uniform temperature analysis at (a) interface 1, (b) interface 2.
solution for this problem is achieved by the trapezium rule method The numerical results for the non-uniform temperature analysis
for time integration. As highlighted in Section 5, the accuracy of the show that the magnitude of the interfacial stresses at the Glass-
numerical scheme is tested against the result from the exact solu- Silicon interface is reduced with respect to the uniform tempera-
tion that can be determined for the uniform temperature boundary ture analysis, as we move away from the core to the edge of the
conditions. A very good agreement was found, see Fig. 11, proving laminate. This may be attributed to the temperature variation from
the accuracy of the method. the core to the edge of the laminate, see Fig. 12(a). On the other
490 S.O. Ojo, M. Paggi / Composite Structures 136 (2016) 481492
0.6 0.2
Shear stress at interface 1(MPa)
0 -0.4
-0.2 -0.6
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
X2 / L X2 / L
(a) (b)
Fig. 12. Shear stresses at: (a) interface 1, and (b) interface 2, after a cooling period of 30 min.
Table 2
Mean axial stress in Silicon for different models and different thickness of Silicon cell.
Table 3
1 Interlaminar stresses at the interface 1 between glass and Silicon for uniform and
Transversal coordinate
non-uniform temperatures.
0.8
Normalised Interfacial stresses, interface 1 (MPa)
length
0.6 Shear stresses Peeling stresses
Uniform Non uniform Uniform Non-uniform
0.4 temperature temperature temperature temperature
0.00 0.00000 0.00000 0.00000 0.00000
0.2 0.10 0.00077 0.00037 0.00285 0.00074
0.20 0.00145 0.00090 0.00011 0.00117
0.30 0.00209 0.00160 0.00024 0.00301
0 0.40 0.00246 0.00302 0.00042 0.00561
-60 -40 -20 0 20 40
0.50 0.00214 0.00571 0.00064 0.01248
Mean axial stress (MPa)
0.60 0.00025 0.01153 0.00085 0.02231
0.70 0.01043 0.02557 0.00094 0.03551
Fig. 14. Mean axial stress distribution in the module for non-uniform temperature
0.80 0.04275 0.05874 0.00048 0.06022
analysis.
0.90 0.16236 0.16024 0.00222 0.09496
1.00 0.52913 0.45058 0.01645 0.13839