POLYMERTESTING

Polymer Testing 25 (2006) 615–622

ARTICLE IN PRESS

0142-9418/$ - see

doi:10.1016/j.po

�Correspondifax:+358 9 4513

E-mail addre

www.elsevier.com/locate/polytest

Test Method

Absorption and desorption of water inglass/ethylene-vinyl-acetate/glass laminates

Thomas Carlssona,�, Petri Konttinena, Ulf Malmb, Peter Lunda

aAdvanced Energy Systems, Helsinki University of Technology, P.O. Box 4100, 02015 HUT, FinlandbAngstrom Solar Center, Uppsala University, P.O. Box 534, SE-751 21, Sweden

Received 15 March 2006; accepted 20 April 2006

Abstract

Ethylene-vinyl-acetate (EVA) is used as the encapsulation polymer in photovoltaic (PV) modules to protect the sensitive

parts of the module from the exterior environment. For many module types, exposure to water is an important lifetime-

limiting factor. An important question is how quickly the EVA absorbs water in wet conditions and how fast water is

desorbed when the module is at operational temperature in sunny conditions. In this work, the water penetration and

escape rates in glass/EVA/glass laminates were compared. A question of particular interest was how sample temperature

affects the escape rate. It was found that water desorption at 50 1C sample temperature is 16 times faster than absorption at

room temperature, which implies that unsealed modules will dry out rapidly on sunny days. The implications of water

absorption/desorption experiments for module design were also discussed.

r 2006 Elsevier Ltd. All rights reserved.

Keywords: Ethylene-vinyl-acetate; Encapsulation; Absorption; Desorption; Water

1. Introduction

Ethylene-vinyl acetate (EVA) is commonly usedas the encapsulation polymer in photovoltaic (PV)modules [1]. A good encapsulation polymer shouldprovide protection from water and impurities, goodadhesion to the other components of the moduleand high transparency. These qualities shouldremain stable for an expected lifetime of at least20 yr in outdoor use. Additionally, cost and proces-sing considerations are very important. Since it isdifficult to find a solution which meets all of these

front matter r 2006 Elsevier Ltd. All rights reserved

lymertesting.2006.04.006

ng author. Tel.: +358 9 4513212;

195.

ss: thomas.carlsson@tkk.fi (T. Carlsson).

requirements perfectly, it is necessary to knowhow different encapsulation polymer alternativesmeet each criterion so that the necessary compro-mises can be made. This paper focuses on thebehaviour of water in EVA, in particular on howabsorption and desorption of water in a PV modulecan be expected to alternate with the weatherconditions.

A standard PV module has an area of 0.5–1m2

and a thickness of 0.5–2 cm. A common modulestructure is glass/EVA/photovoltaic cells/EVA/backsheet. The back sheet is sometimes made of glass,but other materials can also be used since transpar-ency is required only on one side of the module. Ingeneral, the front and back sheets which give themodule rigidity are by far the thickest components

.

ARTICLE IN PRESST. Carlsson et al. / Polymer Testing 25 (2006) 615–622616

of the module, the cells and the encapsulationpolymer are both usually less than a millimetrethick.

The front and back sheets of moisture-sensitivePV modules are normally impermeable to water,which makes water penetration through the poly-mer from the edges of the module the primary watermanagement concern. The edges are often sealed,which reduces water penetration. However, efficientsealing may not always be beneficial. Typicaloperating temperatures for PV modules in fullsunlight are between 50–70 1C, so the water whicha module has absorbed earlier (in rainy conditions,for example) can desorb from the module to somedegree during operation. Even efficient sealingmethods allow some absorption of water into themodule. If the sealant also significantly hinderswater desorption, the total water exposure of asealed module may in some situations be higherthan that of an unsealed module.

In spite of the importance of water managementfor PV modules, a number of questions remainunaddressed to date, especially with regard tokinetics of moisture transfer in the encapsulationpolymer [2]. This is partly because measurementsneed to be conducted on laminates with a structureresembling that of PV modules in order to achieveresults which are relevant for module design.Commercially available humidity sensors are ingeneral too large to be used inside a laminatedstructure, and measurement wires may introduceunwanted penetration paths for the water. Throughthe recent development of a thin moisture sensingelement integrated onto a glass surface with a TiO2

film as the moisture-sensitive component [3], non-invasive measurements are now possible for thisapplication.

This study takes advantage of the new moisturesensor and presents a quantitative comparisonbetween the water absorption and desorption ratesof laminated EVA at different temperatures. Thestudied encapsulation scheme is an unsealed glass/EVA/glass laminate. The diffusion coefficient ofwater at 25 1C in PV-grade EVA (with a vinylacetate content close to 33%) is in the range10�7–10�6 cm2/s [4,5] which is a fairly high valuecompared to other potential PV encapsulant poly-mers [5]. Water movement in EVA is therefore fairlyrapid both in the absorption and desorption phase.The results presented in this paper represent the firstabsorption/desorption rate comparison for lami-nated EVA samples.

The temperatures at which absorption anddesorption experiments were conducted in thispaper correspond to those at which absorption/desorption can be expected to occur in PV modulesduring real operation. The results indicate thatwater desorption at module temperatures of 35 and50 1C is four and 16 times quicker than absorptionat room temperature, respectively. Since PV mod-ules in most locations spend at least 8–10 h attemperatures above 50 1C on sunny days, significantdrying can be expected to occur at the edge ofunsealed EVA-encapsulated modules. Moisture-induced damage is therefore expected to occurprimarily during extended periods of wet exposurewithout sunshine.

2. Experimental methods

2.1. Sensor preparation, measurement and

lamination

Thin moisture sensors were prepared on indiumtin oxide (ITO)-coated glass patterned with photo-lithographic methods. The ITO pattern for eachsensor included an electrode area with two inter-meshed fork patterns and conductor lines from theelectrodes to the edge of the glass where externalconductors could be connected. Each fork patternincluded 14 fingers and each finger was 150 mm wide.Adjacent fingers from opposite sides were spaced150 mm apart from each other. The sensor wasprepared by covering the 3.45� 8mm2 electrodearea with Solaronix HT-L TiO2 paste. Aftersintering, the thickness of the TiO2 layer was inthe range 573 mm. The TiO2 film is porous and cancontain a lot of water relative to its volume. TheAC-resistance of the sensor depends exponentiallyon the water content inside the TiO2 film. Sensorresponse is independent of temperature in the range25–85 1C and it withstands the standard EVAlamination process [3].

Since manual sensor preparation causes someresistance variation between sensors, a scaledresistance parameter r was defined as

r ¼lgðRÞ

lgðR0Þ. (1)

R is the measured AC-resistance value and R0 is anAC-resistance value obtained in a weather chambercalibration measurement with the same sensorbefore lamination. In the calibration measurement,the R of the open sensor was measured at several

ARTICLE IN PRESS

Fig. 1. Laminated glass/EVA/glass sample for absorption/

desorption experiments. Sensors are numbered from I to VI

and ITO contact squares for measurement leads are indicated.

T. Carlsson et al. / Polymer Testing 25 (2006) 615–622 617

relative humidity (RH) values, and R0 was definedas the intersection of the linear R vs. RH curve withthe RH ¼ 0 axis [3]. In this study lgR0 was in therange 5.4–5.7 for all sensors, with R0 expressed in O.Sensor impedance was measured with a Zahner IM6Electrochemical Workstation at frequencies be-tween 1Hz and 1MHz. All measurements wereconducted at zero DC polarization with an ACvoltage amplitude of either 200 or 500mV. The AC-resistance was calculated from the measured spec-trum by fitting an equivalent circuit consisting of R

in parallel with a constant phase element (CPE) tothe measured data. All laminations were carried outin a Panamac L A3-A Automatic PV ModuleLaminator with 0.5-mm-thick EVA (VistaSolar485.00). The vinyl-acetate content of the EVAwas measured to be 33%. Sensor preparation,measurement and calibration have been presentedand discussed in detail elsewhere [3]. Sincethe sensor is prepared on the glass surface, thestructure of the finished samples was glass/sensor/EVA/glass.

2.2. Absorption and desorption experiments

In order to use the moisture sensor in desorptionexperiments, verification of its behaviour duringdesorption was needed to confirm that the waterdoes not remain locked inside the TiO2 film whenthe EVA on top of it dries out. For this purpose,absorption and desorption experiments were firstperformed without a top glass in a glass/sensor/EVA configuration in Arctest ARC-150 and ARC-400 weather chambers. Two samples were tested,one at 25 1C and the other one at 50 1C. The waterconcentration in the EVA was first allowed tosaturate at an initial RH level. Saturation wasjudged by the moisture sensor response, a nearlyconstant r indicating that the sensor, the EVA andthe surrounding environment were near equili-brium. From a given state of sufficiently stableequilibrium, the RH was changed rapidly to a newvalue and sensor response was measured during thetransition. As the external RH changed, the waterconcentration in the EVA also changed throughinward or outward water diffusion depending onwhether RH was increased or decreased. Thischanged the water concentration in the sensor as anew equilibrium was approached. The purpose ofthe experiment was to determine whether the sensorexhibits any time lag in responding to changes inwater concentration in the EVA. If the sensor reacts

equally quickly to a step from high to low RH as itdoes to an RH change in the opposite direction atthe same temperature, then water moves freely outof the sensor when the EVA dries out and sensorresponse gives information on the desorption rate.The RH cycle was 70–90–70% at 25 1C and60–70–80–90–60% at 50 1C.

In the main part of this work, glass/sensor/EVA/glass samples were studied. The top glass had to besmaller than the bottom glass to accommodateelectrical contacting to the edge of the ITO-patterned bottom glass. Fig. 1 illustrates thepositioning of the sensors and the contacts to whichexternal leads were connected. The bottom glasswas 100� 100mm in size with the etched ITOpattern shown in grey colour. TiO2 was applied tothe sensor areas numbered I–VI. The distancebetween the sides of adjacent sensors was 5 or6mm depending on location. The top glass and theEVA layer laminated beneath it were 90� 90mm insize. Both the top and bottom glass sheets wereapproximately 1mm thick. During lamination, thetop glass was kept in place above the bottom glasswith a copper frame. The EVA which had beensqueezed out beneath the edges of the top glass wasremoved with a scalpel after lamination. The edgesmarked B in Fig. 1 were sealed to allow waterpenetration only through edge A. The distancesfrom the sensor edges to the top glass edge A, whichequalled the required water penetration depth toreach the sensors, were approximately 2mm forsensors III and IV, 4mm for II and V, and 6mm forsensors I and VI, respectively. The positioning ofthe top glass and the distance from the sensors tothe glass edge varied approximately 70.5mmbetween the samples.

ARTICLE IN PRESST. Carlsson et al. / Polymer Testing 25 (2006) 615–622618

When water penetrates the encapsulation fromthe sample edge A and reaches the TiO2 film, itprobably fills out the TiO2 film approximatelyuniformly because water moves easily in the porousTiO2 material. Water movement in the EVA isexpected to be significantly slower. The detailedmechanisms of sensor filling are not of criticalimportance for the results in this paper, but someremarks on expected sensor behaviour are in order.The porosity of the TiO2 was about 50%, whichmeans that the 5-mm-thick TiO2 layer can hold onthe order of 2.5 mg/mm2 of water when all pores arefilled. The water saturation concentration of EVAwith 33% vinyl acetate content is 5 mg/mm3 [4],which means that in a state of complete saturationthe amount of water in the TiO2 is comparable tothe amount of water in the 0.5-mm-thick EVAcolumn above it. It is therefore likely that the TiO2

film acts as an effective sink for the water whichpenetrates the EVA. The points made aboveindicate that water enters the sensor mainly at theouter edge of the TiO2 film. During the experimentsdescribed below, sensors I, II, V and VI did not givemeasurable signals. Since the width of all sensors(3.45mm) was larger than the difference in distanceto edge between sensor pairs II/V and III/IV(2mm), it can be concluded that the water front inthe EVA probably did not penetrate past the surfacearea of sensors III and IV. This being the case, the

Fig. 2. Weather chamber measurement to verify sensor response to wat

in desorption than in absorption, but the difference is insignificant for t

to maintain clarity.

response from sensors III and IV followed theinward and outward movement of water at the edge,and gave a valid comparison between the waterabsorption and desorption rates.

During water absorption half of the sample wasimmersed in a bath of distilled water at roomtemperature (22–25 1C) and sensor response waslogged as water absorption proceeded at edge A.The experimental set-up corresponds roughly to thesituation in field conditions when water penetrationis likely, i.e. on cloudy and rainy days when moduletemperature is equal to ambient temperature.Absorption experiments were followed by deso-rption measurements on a hot plate at both 35 and50 1C and on the table at room temperature. Thesurroundings in the desorption experiments werenormal room conditions (22–25 1C, RH 20–50%).On sunny days PV modules in the field regularlyreach temperatures clearly above 50 1C for extendedperiods of time, so the experimental set-up approx-imates the real desorption conditions outdoors.

3. Results

3.1. Verification of sensor behaviour

Fig. 2 shows the measured sensor response at25 1C in a glass/sensor/EVA structure exposed tothe relative humidity changes indicated by the solid

er absorption and desorption at 25 1C. The delay is slightly longer

he purposes of this study. RH curves have been partially modified

ARTICLE IN PRESST. Carlsson et al. / Polymer Testing 25 (2006) 615–622 619

line. The RH in the weather chamber oscillatedaround the set values within the limits 9072% and7071.5% during the experiment. The data on theweather chamber curve during and immediatelyafter each abrupt change in RH is shown asmeasured, but the remaining part of the curve atconstant RH has been straightened manually tomaintain clarity in the figure. The RH oscillationdid not have any effect on sensor response, becauseshort-term variations (with a period of approxi-mately 2min) were averaged out as the waterdiffused through the EVA to the sensor.

Sensor absorption response in Fig. 2 shows thatit takes 15–20 h before the sensor is close to anew equilibrium after the RH step. It can be seenthat the desorption process occurred at approxi-mately the same rate as absorption after a smallinitial delay. In the upward step from 70% to90% RH, the first signs of change in sensorresponse were detected about 20min after the step.In the downward step from 90% to 70% RH,changes were observed after 50min. Fig. 3 showsthe same measurement for a different sensor at50 1C with several upward RH steps. In this case,the diffusion process is quicker and the first changesin sensor response are observed after 1–2min ineach absorption phase and within 7min in deso-rption. The sensor response curve has a slightly

Fig. 3. Weather chamber measurement to verify sensor response to w

slightly more than absorption, but the difference is insignificant for the

longer time constant in the drop from RH 90%to 60% than in the upward RH steps. This isexplained by the slowness of the weather chamberin the downward step (17min was needed tocomplete this step, compared to 2–3min in theupward steps).

As seen in Fig. 3, the step from 80% to 90% at50 1C resulted in a differently shaped curve in ther(t) data compared to the other steps at thistemperature. A possible explanation for this is thatthe accuracy of the fit calculated to the measuredAC-spectrum was reduced by the large amount ofwater present in the TiO2 at RH values above 80.While this brings some uncertainty to this experi-ment, the onset of desorption is still clearly visible inFig. 3. Additionally, it will be observed that sensorsdid not approach equally low values for r in theactual absorption/desorption experiments describedbelow, so a similar loss of accuracy was notexpected there. The time scales of interest in theabsorption/desorption tests for glass/EVA/glasssamples were over 1 h at 50 1C and over 10 h atroom temperature. The small delays in responseevidenced by the data in Figs. 2 and 3 haveinsignificant effects on the results of the experimentsperformed in this work. It can therefore beconcluded that the sensor works as required at bothtemperatures.

ater absorption and desorption at 50 1C. Desorption is delayed

purposes of this study.

ARTICLE IN PRESST. Carlsson et al. / Polymer Testing 25 (2006) 615–622620

3.2. Absorption and desorption from glass/EVA/

glass laminates

The results presented here are all from sensorsclosest to edge A, labelled with III and IV in Fig. 1.Sensors further away from the edge did not registermeasurable signals during the experiments. Fig. 4shows the time series of sensor response r inconsecutive absorption and desorption measure-ments for two laminated glass samples, S1 and S2.In both samples, the measured sensors wereaccompanied by a second sensor at the samedistance from the edge. These reference sensorsmirrored the behaviour of the primary sensorsclosely. The precise distance of the TiO2 film edgefrom the glass edge was 1.6mm in S1 and 2.4mm inS2. The logarithm of R for the sensor in S1 is shownon the right axis to indicate the magnitude of themeasured resistance values. The time scale showsthe number of elapsed hours since the beginning ofthe experiment when the samples were put in thewater bath for the first time. Sample S1 wentthrough the following sequence: (1a) absorption ofwater at room temperature, first sensor responseobtained after �75 h; (1b) start of first desorption at50 1C sample temperature; (1c) start of secondabsorption at room temperature; (1d) start ofsecond desorption at 35 1C sample temperature;(1e) start of third absorption at room temperature;(1f) start of third desorption at 35 1C sample

Fig. 4. Complete series of measurement data from consecutive absor

exemplify the measured resistance values, the lgR axis for the sensor in

temperature; and (1g) end of experiment. Thesecond desorption lacks data points due to ameasurement error. The third desorption wastherefore also conducted at 35 1C to gather sufficientdata. Each desorption was continued until thesensor reached resistances corresponding to ap-proximately r ¼ 1.3–1.4, which is close to themaximum measurable resistance. The samples werenot dried out completely in the desorption phasebecause it would not have been possible todetermine when drying is complete. Stopping thedesorption at a specific r value enabled controlledrepetition of the absorption experiment with a well-defined starting point. Sample S2 underwent: (2a)absorption of water at room temperature, with thefirst signals measured from the sensor after 175 h;(2b) desorption of water at room temperature; and(2c) end of experiment. The time difference betweenthe first absorption curves of S1 and S2, with S2reaching a given r value about 75 h after S1, is dueto the difference in the distance from the sensor tothe edge of the glass.

To gain insight into changes in the absorptionrates during multiple cycles, the different absorptionphases of sample S1 in Fig. 4 have been plottedtogether in Fig. 5. The time axis of the second andthird phase absorption data was chosen so that thecurves passed through the same point, denoted bya in Fig. 5. This enables a direct comparison ofthe absorption rate from that point onward. It is

ption and desorption experiments with samples S1 and S2. To

sample S1 is shown on the right.

ARTICLE IN PRESS

Fig. 5. Comparison of absorption rates in the first, second and third absorptions into sample S1 in Fig. 4. The first absorption proceeds

slower than the following ones.

Fig. 6. Comparison of desorption curves at different sample temperatures for S1 and S2. Surroundings were at room temperature in all

experiments.

T. Carlsson et al. / Polymer Testing 25 (2006) 615–622 621

observed that the second absorption is signifi-cantly faster than the first one, reaching r ¼ 1.00in approximately half the time. The third absorp-tion occurs very nearly at the same rate as thesecond one. The second and third curves can thusbe taken as an initial approximation of the

absorption rate after a few absorption/desorptioncycles.

Two of the desorption phases of S1 and thedesorption phase of S2, in Fig. 4, are showntogether in Fig. 6, where the zero-point timecorresponds to the time of removal from the water

ARTICLE IN PRESS

Table 1

Comparison of absorption and desorption times for first, second

and third absorption cycles in sample S1 and desorption times at

different sample temperatures in samples S1 and S2

# Absorption Tsample Desorption

First 182 h (S1) Room 260 h (S2)

Second 81 h (S1) 35 1C 20 h (S1)

Third 95 h (S1) 50 1C 5h (S1)

The table presents the time needed for the transition between

r ¼ 1.05 and 1.35.

T. Carlsson et al. / Polymer Testing 25 (2006) 615–622622

bath (and, in the case of high temperature tests,placement of the sample on the hot plate). It isobserved that the desorption rate is a very sensitivefunction of sample temperature. At 50 1C thesample returns from r ¼ 1.05 to 1.35 in 5 h, at35 1C the same process takes about 20 h. Thedesorption curve of S2 is not shown in its entiretyin Fig. 6, but is visible in Fig. 4. A precisecomparison between S1 and S2 should take intoaccount that the S2 sensor was further from theedge, but it can be remarked that desorption in S2occurs approximately at the same rate as absorp-tion, taking about 260 h to return from r ¼ 1.05 to1.35. A clear desorption plateau is seen in S2 andalso in both desorption curves from S1. Thisbehaviour may be either due to the desorptioncharacteristics of water from EVA or to sensorbehaviour. Absorption and desorption processes arecompared in Table 1, where the times needed for thetransition between r ¼ 1.05 and 1.35 are presentedfor each absorption/desorption phase. At a 50 1Cmodule temperature, desorption from S1 occurs 16times faster than absorption at ambient temperatureand desorption at 35 1C is at least four times asquick as ambient absorption. The question of howrepeated cycling affects the desorption rate was notstudied in this work.

4. Summary and discussion

The purpose of this paper was to quantify thedifference in water absorption and desorption ratesat the edge of photovoltaic modules encapsulatedwith EVA without edge sealing. Absorption occursin conditions where the module is at ambienttemperature, desorption when module temperatureis above ambient temperature. In a basic glass/EVA/glass encapsulation without framing, even a10–13 1C difference between the module and ambi-ent temperatures is enough to drive out water four

times faster than the rate of absorption at ambienttemperature. When the module temperature is25–28 1C higher than the surroundings, the deso-rption rate is 16 times higher than the rate ofabsorption in ambient conditions. For this reason itis plausible to assume that unsealed, EVA-encapsu-lated modules should frequently dry out in thesummer when high module temperatures arereached repeatedly for long time periods. The waterexposure of a module will depend strongly on howoften wet and sunny conditions alternate in theclimate where it is used.

The method used in this paper can be directlyapplied to the study of water absorption/desorptionin laminates with other encapsulation polymersand/or different edge sealing alternatives. Mostencapsulation polymers can be expected to behavesimilarly to EVA, except that the area of the modulewhich is exposed to repeated moisture penetrationand escape will depend on the water diffusioncoefficient. The design of the PV module edge,especially with regard to the distance of the nearestphotovoltaic cells from the module edge and to thesensitivity of the entire module to edge-cell damage,is an important determinant of its moisture stability.Edge sealing further extends the possibilities ofencapsulation design. Measurement of water ab-sorption and desorption in field conditions shouldprovide interesting comparisons. It is possible thatthe optimal encapsulation solution for PV moduleswhich are used in locations with long and wetwinters is significantly different from the optimumfor sunnier climates. Finally, it should be remarkedthat the results obtained in this paper were fromunaged samples. In long-term outdoor usage agingmechanisms [1,6] may change the absorption/desorption characteristics of EVA and other en-capsulants. This is an important question to addressin future studies on this topic.

References

[1] A.W. Czanderna, F.J. Pern, Sol. Energy Mater. Sol. Cells 43

(2) (1996) 101–181.

[2] T. McMahon, Progr. Photovolt. 12 (2–3) (2004) 235–248.

[3] T. Carlsson, J. Halme, P. Lund, P. Konttinen, Sens. Actuat.

A 125 (2) (2006) 281–287.

[4] S. Marais, Y. Hirata, D. Langevin, C. Chappey, T. Nguyen,

M. Metayer, Mater. Res. Innovat. 6 (2) (2002) 79–88.

[5] M. Kempe, in: Conference Record of the 31st IEEE Photo-

voltaic Specialist Conference, Orlando, 2005, pp. 503–506.

[6] P. Klemchuk, M. Ezrin, G. Lavigne, W. Holley, J. Galica, S.

Agro, Polym. Degrad. Stab. 55 (3) (1997) 347–365.