Composite Structures 136 (2016) 481492

Contents lists available at ScienceDirect

Composite Structures
journal homepage: www.elsevier.com/locate/compstruct

A thermo-visco-elastic shear-lag model for the prediction of residual

stresses in photovoltaic modules after lamination
S.O. Ojo a,, M. Paggi b
a
Department of Structural, Geotechnical and Building Engineering, Politecnico di Torino, Corso Duca degli Abruzzi 24, 10129 Torino, Italy
b
IMT Institute for Advanced Studies Lucca, Piazza San Francesco 19, 55100 Lucca, Italy

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.

1. Introduction between the electrical response and the elastic deformation in

Degradation and cracking has been reported to affect the per-
formance of Silicon cells embedded in photovoltaic (PV) modules.
Due to mechanical and thermal loads which induce mechanical
stresses in the polymer-encapsulated solar cells, PV modules may
experience electrical power loss during service [1,2]. The presence
of different materials composing the stack of a PV module laminate
further complicates the distribution of stresses which in turn
affects the phenomenon of crack propagation in the Silicon cells.
Due to the non-symmetrical structure of the module in the thick-
ness direction, along with the mismatch in the thermo-elastic
material properties, bending of the PV module occurs. This is the
case when the module is cooled down from a high temperature,
which is the stress-free condition during the production stage, to
the ambient temperature. Depending on the thicknesses and on
the thermo-mechanical properties of the layers composing the
photovoltaic module, residual compressive stresses in Silicon
develop and may also lead to a crack closure state, which is posi-
tively influencing the recovery of electrical conductivity in the
cracked regions, as experimentally shown in [3]. This coupling
Corresponding author. Tel.: +39 0110904868; fax: +39 0110904899.
E-mail addresses: saheed.ojo@polito.it (S.O. Ojo), marco.paggi@imtlucca.it
(M. Paggi).
the presence of cracking is an important concept which has to be
modelled with care.
Several research studies on thermo-elastic lamination theory
within the last few years have been dedicated to the distribution
of stresses inside the laminate with a uniform temperature profile
[49]. Other studies to improve the long term stability of PV mod-
ules focused on modelling the behaviour of a PV module during
thermal cycling and have shown that Silicon cells are under com-
pressive stresses up to 76 MPa [8]. Microcracking can take place
[3,10], and fatigue degradation can also be present in case of
repeated mechanical loading [3]. In view of the interaction
between various fields (elastic, thermal and electric) that charac-
terise a typical PV system, a multiphysics approach has been
proved to be effective [2].
Investigation of the effect of interfacial compliance on thermo-
mechanical stresses of laminate structures used in microelectron-
ics has been presented in [12] where analytical solutions show
that softer adhesive layers reduce significantly the interfacial stres-
ses and that the presence of high peeling stresses at the laminate
free edges can facilitate delamination. This is also confirmed in
[7,13] where a layer-wise theory was used to investigate the non-
linear behaviour of interlaminar peel and shear stresses. The effect
of interfacial stresses in PV modules is however more complicated
due to the presence of a viscoelastic adhesive interface, i.e., the

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

polymer occurring at a given transition temperature during 3

lamination which invalidates the applicability of the classical
timetemperature superposition principle, it is shown in [19] that where Q represents, consistently with energy conservation
the fractional model can still be applied to each of the two temper- principles, the heat flow exchanged at mesh points, which can be
ature regimes below and above the transition temperature, with a further expressed as Q K DT, where DT is the temperature varia-
total of four parameters to be identified, two per each regime. tion between two adjacent mesh points, and K is the corresponding
In the present study, a novel thermo-visco-elastic shear-lag the- thermal conductance. The latter is a function of the equivalent
ory based on the fractional calculus formulation in [19] is herein thermal conductivities of the grid points and on the laminate
proposed for performing coupled thermo-elastic structural composition through its thickness, the width, and the size of the
simulations of laminates with polymeric layers. Specifically, the grid spacings, i.e. Dzk , Dx2j and Dx1i . Since the volume of a cell is
aim is to predict the thermomechanical residual stress distribution V i;j;k Dx1i Dx2j Dzk , the equivalent thermal conductances can be
in the PV module stack after lamination in a semi-analytic form,
computed according to [11]:
thus providing a useful guidance for the design of the laminate. A
careful comparison with other approaches based on a continuum Dx2 j Dz k
K i1;j;k Dx1 D x1
; 4a
description of the encapsulant layer is also provided. 2 i1
2k i
Ri1;j;k
2ki1;j;k i;j;k 2

2. Mathematical formulation of the problem of heat conduction Dx2 j Dz k

K i1;j;k Dx1 D x1
; 4b
2k Ri1;j;k
2 i1 i
A model problem consisting of a Silicon cell embedded in a 2ki1;j;k i;j;k 2
stack composed of glass, EVA and backsheet is considered, see
Fig. 1. For an accurate heat conduction simulation of the Dx1i Dzk
lamination process and its subsequent stages, we consider a 3D K i;j1;k D x2 Dx2
4c
2 j1
2ki;j1;k
2k j
Ri;j1;k
heat conduction equation of the form: i;j;k 2

@T @2T @2T @2T Dx1i Dzk

C kx 2 ky 2 kz 2 H 1 K i;j1;k D x2 Dx2
4d
@t @x1 @x2 @z 2 j1
2k j
Ri;j1;k
2ki;j1;k i;j;k 2

where Tx1 ; x2 ; z; t represents the unknown space- and time-

dependent temperature profile; Cx1 ; x2 ; z is an equivalent volu- Dx1 i Dx2 j
K i;j;k1 Dzk1
4e
metric heat capacity (J/m3 K), which is equal to an equivalent mass 2
2ki;j;k1
2kDzk Ri;j;k1;
i;j;k 2
density multiplied by the equivalent specific heat capacity
(C = q  cp ), taking into account the composite structure of the lam- Dx1 i Dx2 j
inate. The function Hx1 ; x2 ; z; t represents a heat source and, since K i;j;k1 Dzk1
4f
2
2kDzk Ri;j;k1;
there is no heat source in this problem, the term H can be dropped 2ki;j;k1 i;j;k 2

from Eq. (1). The coefficients kx1 x1 ; x2 ; z, kx2 x1 ; x2 ; z and kz x1 ; x2 ; z

where the R terms are the interfacial resistances between adjacent
are the thermal conductivities in the x1 , x2 and z directions, respec- cells. Since the laminate stack can be approximated as periodic in
the x1 and x2 direction, those R terms in the x1 and x2 direction
are vanishing, i.e.:
Ri1;j;k Ri1;j;k 0; Ri;j1;k Ri;j1;k 0 5
2 2 2 2

It is also assumed that heat is absorbed from the upper and

lower free sides of the laminate by natural convection. Therefore,
a parallel thermal resistance configuration applies at the interface
between the layers. If the thermal resistances Rg, RE, Rs, and Rb
represent, respectively, those of the glass, EVA, Silicon and
backsheet resistances, then the thermal resistances at the
interfaces can be computed as follows:
1 1
Rint1 ; 6a
Fig. 1. Module geometry for the thermal analysis. Rg R0E
 S.O. Ojo, M. Paggi / Composite Structures 136 (2016) 481492 483

1 1 In a PV lamination process, homogeneous temperature enables

Rint2 Rint3 6b
Rs R0E a significant faster lamination process, as well as a more homoge-
neous cross-linking and sealing of the encapsulant. Hence it is
1 1 assumed that the PV laminate attains a homogenous temperature
Rint4 6c
Rb RE of 150 C at every point in the module after heating. Therefore,
the initial condition at time t 0 is given as:

2.1. Heat flows Tx1 ; x2 ; z; 0 150  C 0  x1  X; 0 6 x2 6 Y; 0zZ

10
The total heat flow converging to a cell (i; j; k) from the 6 adja-
cent cells is given by (see Fig. 2)
2.2. Boundary conditions for thermal analysis
Q i;j;k K i1;j;k T i1;j;k  T i;j;k K i1;j;k T i1;j;k  T i;j;k
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

 T i;j;k 2K i 2K j K i;j;k1 K i;j;k1 8 @T0; x2 ; z; t

k hT0; x2 ; z; t hT p 12b
@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. Stress analysis in the case of a perfect interface (fully bonded)

configuration

Let us consider the PV laminate whose cross-section is shown in

Fig. 4, made of glass, EVA, Silicon, EVA and backsheet layers with
material properties reported in Table 1. All the interfaces between
the layers are considered here as fully bonded.
Based on the EulerBernoullis hypothesis of conservation of
plane cross-sections, the longitudinal strain at an arbitrary position
Fig. 2. Finite difference discretization, reference system and grid spacing for a unit
cell. x is given by [4]:
484 S.O. Ojo, M. Paggi / Composite Structures 136 (2016) 481492

-3 -3
x 10 x 10
5 5
330 400

Transverse axis (m)

Transverse axis (m)

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:

@ 2 w @U 1 1 @ r1z @ r2z @ rzz

e1 z r1  tr2 aDT 21a 0 27c
@x21 @x1 E @x1 @x2 @z
And reduce to
@ 2 w @U 2 1
e2 z 2 r2  tr1 aDT 21b dr2 drz2
@x2 @x2 E 0 28a
dx2 dz
The product aDT is the thermal strain in the layer due to cooling
of the PV from 150 C to the room temperature. Solving for stres- dr2z drzz
0 28b
ses, we get: dx2 dz
E The boundary conditions required to solve the equilibrium
r1 e1 te2  aDT1 t
1  t2 equation are [5]:
" #
E @ 2 w @U 1 @2w @U 2 h
z  tz t  aD T1 t r2z r2z x2 ; c; rzz rzz x2 ; c; at z c 29a
1  t2 @x21 @x1 @x22 @x2 2
22a
h
r2z r2z x2 ; c; rzz rzz x2 ; c; at z  c 29b
E 2
r2 e2 te1  aDT1 t
1  t2 By integrating the reduced equilibrium equations w.r.t. z and
" #
E @ 2 w @U 2 @2w @U 1 applying the boundary conditions at z = c and z c, we obtain:
z  zv v  aD T1 t " 3 #
1  t2 @x22 @x2 @x21 @x1 2
E d w 2 d U2
22b r2z z  c  2
2
z  c r2z x2 ; c 30a
21  t2 dx32 2
dx2
" #
3 2
E d w 2 d U2
r2z z  c 2
 2 z c r2z x2 ; c
21  t2 dx32 2
dx2
30b
Adding Eqs. (30a) and (30b) and simplifying, we get:
3 2
E d w 2 E d U2 m
r2z z  c2  z 31
21  t2 dx32 1  t2 dx22 2

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

Similarly, by substituting Eq. (31) into (28b) and integrating dM2

mc  Q 2 0 39b
with the boundary conditions in Eq. (29): dx2
4 3 By subtracting Eq. (30b) from (30a) for the top and bottom of
E d w E d U2 2
rzz z3 3c2 z  2c3 z  c2 the plate and simplifying, we obtain:
61  t2 dx42 21  t2 dx32
2
1 dm Eh d U2
 z  c rzz;c 32a r2z x2 ; c  r2z x2 ; c 0 40
2 dx2 1  t2 dx22

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

where p rzz;c rzz;c ; rzz;c rzz x2 ; c; rzz;c rzz x2 ; c.

leads to:
For a generic plate, the axial force, the shear force and the resul-
tant bending moment (N, Q, M) are defined as: dQ 2
Z q0 43
c dx2
Ni ri dz 34a
c
The overall equilibrium equations for the beam type plate are
Z c now expressed as:
Mi ri zdz 34b
c dM2
mc  Q 2 0; 44a
Z dx2
c
Qi riz dz for i 1; 2 34c
c dN2
n 0; 44b
Substituting for ri and riz (for i 1; 2) in the resultant equa- dx2
tions, we have for a generic ply:
  dQ 2
N1 Et dU 2 q 0: 44c


 a DT t 35a dx2
h 1  t dx2
2
Eqs. (44a) and (44c) can be further combined to give
 
N2 E dU 2 2
d M2 dm
 a D T1 t 35b cq0 45
h 1  t2 dx2 2 dx2
dx2
The moment resultants are given as
In the sequel, rzz and r2z will henceforth be denoted respec-
3 2 2
Eth d w EIt d w tively by r and s.
M1   0 36a The rotation variable W in the x2 direction can be obtained from
121  t2 dx22 1  t2 dx22
Eq. (20b)
3 2 2
Eh d w EI d w du2 @w
M2   36b W  46
121  t2 dx22 1  t2 dx22 dz @x2
Differentiating Eq. (25) twice and substituting into Eq. (36) we By combining Eqs. (25) and (46), we get:
get:
@W
W 47
Eth
3
d W
2
EIt d W
2
@x2
M1   37a
121  t2 dx22 1  t2 dx22
4.1. Numerical scheme
3 2 2
Eh d W EI d W
M2   37b For a 3 layer module as in Fig. 5 represented by a beam type
121  t2 dx22 1  t2 dx22
plate, there are six equilibrium equations in total which are stated
Using Eqs. (35) and (37), Eq. (22) for the normal stresses can be as:
simplified as:
k
@ 2 M2 @mk k
Ni Mi c qk 0; 48a
ri z for i 1; 2 38 @x22 @x2
h I
Evaluating Eq. (34c) for Q 2 gives k
@N2
nk 0 for k 1; 2; 3 48b
EI d w
3
@x2
Q2  mc; 39a
1  t2 dx32 where k is the number of layers.
 S.O. Ojo, M. Paggi / Composite Structures 136 (2016) 481492 487

The behaviour of the EVA interfaces modelled as deformable

linear elastic springs is mathematically given as:

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
An exact solution of the system of ordinary differential
equations can be achieved for the special case of a uniform
temperature profile for the entire PV module after cooling. In
the case of a non-uniform temperature analysis, the coefficients
in the B matrix vary with time and space, so a numerical
method is required. Specifically, for the non-uniform tempera-
ture case study, an integration scheme using a trapezium rule
method (an average of forward and backward Euler method) is
used. To verify the accuracy of this numerical scheme for this
analysis, a comparison was made between the exact solution
of the uniform temperature analysis and the numerical solution
using trapezium rule method and a good agreement was
achieved. The results of this comparison are shown in the next
section.
5.1. Uniform temperature across the PV module
We consider a cooling down of the module from 150 C
to the room temperature (25 C) in 30 min. At this final tem-
perature, the EVA relaxation modulus E(t) is calculated to be
6 MPa.
In the case of perfectly bonded interfaces and EVA layers
treated as continuum layers with their thicknesses, the axial
stress and strain are plotted in Fig. 8. The Silicon layer results
to be in a compressive stress state above 60 MPa. The thermal
strain and the mechanical strain almost counteract each other,
which makes the total strain to be almost zero in the entire
module.
(a)
Fig. 8. (a) Stress and (b) strain distributions for a perfect interface configuration of uniform temperature analysis.
For the shear deformable interface configuration with 3 layers
and 2 interfaces (glass-Silicon interface 1 and Silicon-backsheet
interface 2), 2 models are examined:
(1) Shear deformable interfaces with longitudinal compliance
only.
(2) Shear deformable interfaces with longitudinal and trans-
verse compliances of equal magnitude.
It can be deduced from the results shown in Fig. 9 that the
interfacial stress distribution for model 1 and model 2 are almost
the same. This result is not unexpected, since we assume an
isotropic adhesive layer at the interfaces for model 1 and model
2 (i.e., Kx = Ky). The effect of the transverse compliance will obvi-
ously be significant when an orthotropic adhesive (i.e. Kx Ky )
interface is considered.
The results of the interfacial stress distribution pinpoint that
the magnitude of the shear stresses and the peeling stresses
increases towards the edge of the laminate. Therefore, those points
are critical locations for possible delamination. It is observed in
Fig. 10 that the mean compressive stress in the Silicon layer is
reduced for the shear deformable interface configuration with
respect to the perfect interface configuration, due to the finite com-
pliance of the interfaces, which is a more realistic situation in
practise.
5.2. Non-uniform temperature
The stress formulation for this analysis involves a shear deform-
able interface with only longitudinal compliance, since the effect of
the transverse compliance was already found to be negligible. The
(b)
 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
for time integration. As highlighted in Section 5, the accuracy of the
numerical scheme is tested against the result from the exact solu-
tion that can be determined for the uniform temperature boundary
conditions. A very good agreement was found, see Fig. 11, proving
the accuracy of the method.
The numerical results for the non-uniform temperature analysis
show that the magnitude of the interfacial stresses at the Glass-
Silicon interface is reduced with respect to the uniform tempera-
ture analysis, as we move away from the core to the edge of the
laminate. This may be attributed to the temperature variation from
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)

Shear stress at interface 2(MPa)

Uniform temp
Non uniform temp
0.4 0
Uniform temp
0.2 -0.2 Non uniform temp

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.

Normalised thickness Mean axial stress in Silicon (MPa)

Uniform temperature Non-uniform
temperature
Perfect interface Shear-lag Shear-lag
0.01 66.452 56.363 44.005
0.10 66.332 56.184 43.974
0.20 66.212 55.996 43.938
0.30 66.092 55.817 43.901
0.40 65.973 55.629 43.865
0.50 65.853 55.450 43.828
0.60 65.733 55.272 43.792
0.70 65.573 55.084 43.755
0.80 65.453 54.905 43.719
0.90 65.333 54.717 43.682
Fig. 13. Mean residual compressive stresses in the Silicon layer vs. longitudinal 1.00 65.213 54.547 43.648
coordinate, for different time intervals.

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

hand, there is no significant changes in the stress distribution at

the Silicon-backsheet interface, see Fig. 12(b).
Results in Fig. 13 show that the mean compressive stress in the considerably lower by assuming a space-dependent temperature
Silicon cell increases along the longitudinal coordinate in the field instead of a uniform temperature across the whole module,
Silicon layer as the cooling time increases. i.e., about 45 MPa instead of about 60 MPa. A detailed comparison
As shown in Fig. 14, the mean residual axial stress in the of numerical results for the different models is presented in
Silicon layer at the end of the lamination process after cooling is Tables 24.
 S.O. Ojo, M. Paggi / Composite Structures 136 (2016) 481492 491

Table 4 A.2. Coefficients of B matrix

Interlaminar stress distribution at the interface 2 between Silicon and backsheet for
uniform and non-uniform temperatures.
The matrix operator B entering Eqs. (51) and (56) is given by:
Normalised Interfacial stresses, interface 2 (MPa) 2 3
length
0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
Shear stresses Peeling stresses 6j 0 j 0 0 0 0 j 0 0 0 j4 0 0 0 0 0 0 7
6 1 2 3 7
Uniform Non uniform Uniform Non uniform 6 7
6 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 07
temperature temperature temperature temperature 6 7
6 7
6 n1 0 n2 0 n3 0 0 n4 0 0 0 n5 0 0 0 n6 0 0 7
0.00 0.00000 0.00000 0.00000 0.00000 6 7
6 7
0.10 0.02870 0.00832 0.00034 0.00052 6 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 07
6 7
0.20 0.05564 0.01579 0.00130 0.00047 6 0 0 . 0 . 0 0 0 0 0 0 .5 0 0 0 .6 0 0 7
0.00055 6 7
0.30 0.08580 0.02502 0.00290 6 2 3
7
0.40 0.11500 0.03579 0.00480 0.00042 6 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 07
6 7
0.50 0.14866 0.05236 0.00713 0.00067 6 7
6 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 07
0.60 0.18211 0.07454 0.00936 0.00053 6 7
6 7
0.70 0.22163 0.11093 0.01162 0.00019 6 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 07
B66k k k k 0 0 k k k 0
7
k6 k10 k4 0 0 0 0 0 7
0.80 0.26196 0.16174 0.01322 0.00128
6 7 1 8 2 5 9 3 7
0.90 0.31138 0.24590 0.01370 0.00194 6 7
0.00378 6 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 07
1.00 0.36520 0.36719 0.01919 6 7
6 7
6 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 07
6 7
6 7
6. Conclusions 6 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 07
6 7
6c c c c c c c c c 0 c8 c14 c5 0 c9 c15 c6 0 7
6 10 1 11 2 12 3 7 13 4 7
6 7
A detailed investigation of the thermomechanical stress distri- 6 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 07
6 7
bution inside a PV module during lamination has been carried 6 7
6 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 07
out by comparing a spatially uniform temperature distribution 6 7
6 7
4 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 15
dependent on time and a non-uniform temperature distribution
obtained by numerically solving the heat conduction equation in 0 0 d7 d1 d8 d2 0 0 0 0 d5 d9 d3 0 d6 d10 d4 0
space and time. An advanced plate formulation with shear-lag the-
ory is presented and implemented using a semi-analytic approach where the entries of the B matrix are defined as:
accounting for the viscoelastic behaviour of the EVA layer, based on
1  m2i
fractional calculus considerations. The predictions of this shear-lag Ki  ;
Ei hi
model have been compared with another classical structural model
where the EVA layer is considered as a continuum, as in linear elas-
1  m2i
tic finite element analyses of PV laminates. !i i 1; 2; 3
The results of the stress analysis show that Silicon is subjected to Ei Ii
a residual mean compressive stress of about 60 MPa if uniform tem-
j1 K1 Kx1 ; j2 K1 Kx1 ; j3 K1 Kx1 c1 ; j4 K1 Kx1 c2
perature is considered. On the other hand, if a non-uniform temper-
ature distribution is accounted for in the analysis, as it happens
n1 K2 Kx1 ; n2 K2 Kx1 Kx2 ; n3 K2 Kx2 ;
during the simulation of the transient heat conduction regime, the
stress field reduces to about 40 MPa after lamination.
n4 K2 Kx1 c1 ; n5 K2 c2 Kx1  Kx2 ; n6 K2 Kx2 c3 ;
As future developments, a finite element model where a cou-
pled thermo-mechanical formulation is envisaged to further refine .1 K3 Kx2 ; .2 K3 Kx2 ; .3 K3 Kx2 c2 ; .4 K3 Kx2 c3
the present semi-analytical predictions. On the other hand, since
the computation time will drastically increase with respect to the
k1 !1 Kx1 c1 ; k2 !1 Kx1 c1 ; k3 !1 Kx1 c21 ; k4 !1 Kx1 c1 c2
semi-analytical approach proposed in this work, model order
reduction schemes will be probably requested to overcome this
!1 c1 @Kx1 !1 c1 @Kx1
challenge. k5 !1 Ky1 ; k6 !1 Ky1 ; k7 ; k8 
Kx1 @x2 Kx1 @x2
Acknowledgements
!1 c21 @Kx1 !1 c1 c2 @Kx1
k9  ; k10  ;
MP would like to acknowledge funding from the European Kx1 @x2 Kx1 @x2
Research Council under the European Unions Seventh Framework
Programme (FP/20072013)/ERC Grant Agreement No. 306622 c1 !2 Kx1 c2 ; c2 !2 c2 Kx2  Kx1 ; c3 !2 Kx2 c2 ;
(ERC Starting Grant Multi-field and multi-scale Computational c4 !2 Kx1 c1 c2
Approach to Design and Durability of PhotoVoltaic Modules
CA2PVM). SOO acknowledges the support of the Italian Ministry c5 !2 c22 Kx2 Kx1 ; c6 !2 Kx2 c3 c2 c7 !2 Ky1 ;
of Education, University and Research to the Project FIRB 2010
Future in Research Structural mechanics models for renewable !2 c2 @Kx1
energy applications (RBFR107AKG). c8 !2 Ky1 Ky2 ; c9 !2 Ky2 ; c10 ;
Kx1 @x2

Appendix A 1 @Kx2 1 @Kx1 !2 c2 @Kx2
c11 !2 c2  ; c12  ;
Kx2 @x2 Kx1 @x2 Kx2 @x2
A.1. Elements of m vector !2 c1 c2 @Kx1
c13  ;
Kx1 @x2
The vector of unknowns m in Eqs. (51) and (56) is defined as:

1 @Kx2 1 @Kx1 !2 c3 c2 @Kx2
v U1 U01 U2 U02 U 3 U03 W 1 W 01 W 001 W 0001 W 2 W 02 W 002 W 0002 W 3 W 03 W 003 W 0003 T c14 !2 c22
Kx2 @x2

Kx1 @x2
; c15 
Kx2 @x2
;
492 S.O. Ojo, M. Paggi / Composite Structures 136 (2016) 481492

d1 !3 Kx2 c3 ; d2 !3 Kx2 c3 ; d3 !3 Kx2 c3 c2 ; d11 !1 Ky1 DT 2 a2 c2 DT 1 a1 c1

d4 !3 Kx2 c23
References
!1 c1 @Kx1 !1 c1 @Kx1
d5 !3 Ky2 ; d6 !3 Ky2 ; d7 ; d8 
Kx1 @x2 Kx1 @x2 [1] Kntges M, Kunze I, Kajari-Schrder S, Breitenmoser X, Bjrneklett B. The risk of
power loss in crystalline silicon based photovoltaic modules due to
microcracks. Sol Energy Mater Sol Cells 2011;95:11317.
!1 c21 @Kx1 !1 c1 c2 @Kx1 [2] Paggi M, Corrado M, Rodriguez MA. A multi-physics and multiscale numerical
d9  ; d10  ; approach to microcracking and power-loss in photovoltaic modules. Compos
Kx1 @x2 Kx1 @x2 Struct 2013;95:6308.
[3] Paggi M, Berardone I, Infuso A, Corrado M. Fatigue degradation and electric
recovery in Silicon solar cells embedded in photovoltaic modules. Sci Rep
A.3. Coefficients of F vector 2014;4:17. Paper no.: 4506.
[4] Carpinteri A, Paggi M. Thermo-elastic mismatch in non-homogeneous beams. J
Eng Math 2008;61:37184.
[5] Wen Y, Basaran C. An analytical model for thermal stress analysis of multi-
8 9 layered microeloctronic packaging. Mech Mater 2004;36:36985.
> 0 >
>
> >
> [6] Suhir E, Shangguan D, Bechou L. Predicted thermal stresses in a trimaterial
>
> 0 >
>
>
> >
>
assembly with application to silicon based photovoltaic module. J Appl Mech
>
> >
0 >
2013;80:110.
>
> >
>
>
> >
> [7] Tahani M, Nosier A. Free edge stress analysis of general cross-ply composite
>
> 0 >
> laminates under extension and thermal loading. Compos Struct 2003;60:
>
> >
>
>
> >
> 91103.
>
> 0 >
>
>
> >
>
[8] Eitner U, Kajari-Schrder S, Kntges M, Altenbach H. Thermal stress and strain
>
> >
>
> 0 >
>
>
of solar cells in photovoltaic modules. Adv Struct Mater 2011;15:45368.
>
> >
> [9] Wen Y, Basaran C. Thermomechanical stress analysis of multilayered
>
> 0 >
>
>
> >
>
electronic packaging. J Appl Mech 2003;80:1348.
>
> >
0 >
[10] Paggi M, Sapora A. Numerical modelling of microcracking in PV modules
>
> >
>
>
> >
> induced by thermo-mechanical loads. Energy Procedia 2013;38:50615.
< 0 = [11] Ojo SO, Grivet-Talocia S, Paggi M. Model order reduction applied to heat
F conduction in photovoltaic modules. Compos Struct 2015;119:47786.
>
> k11 >
>
>
> >
>
[12] Basaran C, Wen Y. Influence of interfacial compliance on thermomechanical
>
> >
> 0 >
stresses in multilayered microelectronic packaging. IEEE Trans Adv Packag
> >
>
>
> >
> 2006;26:66673.
>
> 0 > > [13] Abdolmajid Yousefsani S, Tahani M. An analytical investigation on
>
> >
>
>
> >
> thermomechanical stress analysis of adhesively bonded joints undergoing
>
> 0 >
>
>
> >
>
heat conduction. Arch Appl Mech 2014;84:6779.
>
> >
c16 >
[14] Schulze S, Pander M, Naumenko K, Altenbach H. Analysis of laminated glass
>
> >
>
>
> > beams for photovoltaic applications. Int J Solids Struct 2012;49:202736.
>
> 0 >>
>
>
> > [15] Paggi M, Kajari-Schroder S, Eitner U. Thermomechanical deformations in
> 0 >
> >
> photovoltaic laminates. J Strain Anal Eng Des 2011;46:77282.
>
> >
>
>
> 0 > > [16] Eitner U. Thermomechanics of photovoltaic modules. Martin-Luther-
>
> >
> Universitt Halle-Wittenberg; 2011. p. 3549.
>
: >
; [17] Di Paola M, Pinnola F, Zingales M. A discrete mechanical model of fractional
d11 hereditary materials. Meccanica 2013;48:157386.
[18] Di Paola M, Pirrotta A, Valenza A. Visco-elastic behavior through fractional
where the entries of the F vector are defined as: calculus: an easier method for best fitting experimental results. Mech Mater
k11 !1 Ky1 DT 2 a2 c2 DT 1 a1 c1 2011;43:799806.
[19] Paggi M, Sapora A. An accurate thermo-viscoelastic rheological model for
ethylene vinyl acetate based on fractional calculus. Int J Photoenergy 2015:17
c16 !2 Ky1 DT 1 a1 c1 DT 2 a2 c2 Ky1  Ky2  Ky2 DT 3 a3 c3  Article ID 252740.