Assessing Field Degradation of Photovoltaic Modules by Near‐Infrared Absorption Spectroscopy of Ethylene Vinyl Acetate Encapsulant

The degradation status of polymer encapsulants in photovoltaic (PV) modules with backsheets of various compositions is evaluated by near‐infrared absorption (NIRA) spectroscopy. The method is used to identify backsheet types and to assess the content of carbonyl species in the ethylene vinyl acetate copolymer encapsulant expressed as a carbonyl index (CI). The CI distributions are collected for PV modules of different ages and in different climatic conditions from twelve PV installations in Europe (EU) and South Africa (ZA) and the peak of the index distribution is found to be related exponentially to the total dose of solar irradiation. The CI growth rate with the irradiation dose in EU is around two times larger than in ZA, most probably due to more massive moisture ingress into PV modules installed in a more humid EU climate. For similar total solar exposures, the CI is found to depend on the backsheet structure. Repeated measurements on the same plant after an additional 2 years of module operation confirm the capacity of NIRA spectroscopy to track the temporal progression of encapsulant degradation using CI as a quantifier depending on the encapsulant‐backsheet bill of materials, plant age, and climatic zone.


Introduction
[3][4][5] The operating status of a PV plant as a whole or on the inverter or string The degradation status of polymer encapsulants in photovoltaic (PV) modules with backsheets of various compositions is evaluated by near-infrared absorption (NIRA) spectroscopy.The method is used to identify backsheet types and to assess the content of carbonyl species in the ethylene vinyl acetate copolymer encapsulant expressed as a carbonyl index (CI).The CI distributions are collected for PV modules of different ages and in different climatic conditions from twelve PV installations in Europe (EU) and South Africa (ZA) and the peak of the index distribution is found to be related exponentially to the total dose of solar irradiation.The CI growth rate with the irradiation dose in EU is around two times larger than in ZA, most probably due to more massive moisture ingress into PV modules installed in a more humid EU climate.For similar total solar exposures, the CI is found to depend on the backsheet structure.Repeated measurements on the same plant after an additional 2 years of module operation confirm the capacity of NIRA spectroscopy to track the temporal progression of encapsulant degradation using CI as a quantifier depending on the encapsulant-backsheet bill of materials, plant age, and climatic zone.
decisive impact of polymer components on PV module performance, field-ready tools for material identification or evaluation of the degradation status of polymers in PV modules are still lacking.
Recently, we have shown that near-infrared absorption (NIRA) spectroscopy can be applied for nondestructive identification of the composition of polymer components of PV modules. [9,10]his application of NIRA becomes possible due to a large penetration depth of NIR light, allowing multilayer BS materials and transparent encapsulants of PV modules to be probed from either side of a module.So far, NIRA studies have been focused on the identification of BS BOMs and the evaluation of the BS degradation state.At the same time, the spectral tools for the examination of the degradative processes in transparent PV encapsulants are still underdeveloped.
Detailed investigations of the artificial degradation of ethylene vinyl acetate (EVA) copolymer encapsulants indicate that many individually applied tests, such as the standard ultraviolet (UV) irradiation or damp-heat tests, [11,12] are insufficient to reproduce adequately the effects of the field degradation.The combination of several stress factors, such as UV irradiation accompanied by a thermal treatment, can yield more realistic degradation patterns which are still to be proven to be relevant for the field-aged PV modules. [12]For this reason, the development of noninvasive tools for the direct field evaluation and tracking of EVA degradation is instrumental in providing field-relevant data and building a bridge between field-aging scenarios and advanced artificial testing of polymer encapsulants.
In the present report, we advance the NIRA-based approach and apply it to the spectral characterization of EVA encapsulants.An analysis of spectral data collected from EVA encapsulants of PV modules with different BOMs, ages, and climatic conditions of installation provides quantitative information on the oxidation state of EVA in the form of a carbonyl index (CI).The CI of EVA correlates with the "wet" leakage resistance (WLR) of corresponding PV modules, being, therefore, a reliable quantifier of their degradation status.A comparison of average carbonyl indices derived for PV plants with different operational conditions in Europe (EU) and South Africa (ZA) brings unique insights into the PV degradation dynamics depending on climatic conditions (in particular, total solar irradiation dose) and BS materials, allowing the temporal evolution of degradative processes to be continuously tracked.

Instruments
NIRA spectra were recorded using a fiber-coupled fouriertransform-NIR Rocket spectrometer (ArcOptix, Switzerland) in a spectral range of 900-2600 nm (3800-11 000 cm À1 ) with a spectral resolution of ≈8 cm À1 .The samples were illuminated with a broad-emitting stabilized fiber-coupled NIR lamp (Thorlabs, Germany).The spectrometer and the lamp were connected to a probe block (Thorlabs) at an angle of 90°via 1000 μm-thick optical fibers transmitting in the range of 400-2600 nm (Thorlabs).The probe was placed on the inspected surface and held manually, and a series of 10 NIR reflectance (NIRR) spectra were collected and averaged.The original NIRR spectra were then converted into NIRA spectra by dividing the reflectance spectrum of the lamp taken on a polished aluminum mirror and the reflectance spectrum of the sample.

Module Inspection
Typically, a pair of NIRA spectra of BS and EVA encapsulant were taken from each inspected module.NIRA spectra of BSs were taken from the air side behind one of the lowest Si cells of the module, regardless of the module orientation (Figure 1a, insert 1).NIRA spectra of the EVA encapsulant were taken above metal interconnects close to the points where BS spectra were registered (Figure 1a, insert 2).As a rule, EVA was probed over a section of metal interconnect located above a Si cell in the left upper corner of the PV module when looked at from the frontal side.In rare cases of strong corrosive damage of this section of metal interconnect (below 1% of inspected modules) the measurements were done on less corroded sections.Only the lowest rows of modules were used for the spectral data collection.The spectra were collected on clear days with no rain at ambient temperatures of 22-28 °C from dry modules with a rate of 30-40 pairs of BS/EVA spectra per hour allowing the data for each particular field to be collected in steady ambient conditions.
The WLR of PV modules was measured in the lab by immersing the modules into an aqueous electrolyte bath, biasing them at 1000 V, and measuring the leakage current (resistance) between the module and a counter electrode placed in the bath at a certain distance. [8]The modules were kept in the climatecontrolled lab (22 °C, relative humidity (RH) 35%) for 24 h before measurements to eliminate possible differences in the moisture distribution among the tested samples.During the measurements of WLR the modules were submerged for identical periods.

Analysis of Spectra
NIRA spectra of BSs and EVA were processed using an in-house programmed Python script, which automatically performed preprocessing (cutting and baseline subtraction), identification of the BS type, and determination of the carbonyl and water indices from corresponding NIRA spectra of EVA encapsulant.The BS type was identified from a ratio of NIRA peaks at 1660 nm and 1730 nm corresponding to ═C─H and ─C─H vibrations in polyethylene terephthalate (PET) and other components of BSs as described in detail elsewhere. [9,10]The CI of the EVA encapsulant was calculated as a ratio of integral intensities of the carbonylrelated band peak at 2140 nm and the reference ─C─H-related band peak at 1730 nm.To account for the possible distribution of C═O species contributing to the NIRA spectrum, the C═Orelated band was integrated with a broader range between 2120 and 2170 nm.The water index (WI) was calculated similarly as a ratio of integral intensities of the water-related band peak at 1910 nm (integrated with a broader range of 1880-1920 nm) and the reference band peak at 1730 nm.

Results and Discussion
3.1.Structure of NIRA Spectrum of EVA Encapsulant NIRA provides a unique opportunity to probe the spectral properties of EVA encapsulants directly through the glass without any intervention in the operation of the PV module.At that, the most highly resolved and informative NIRA spectra of EVA were collected by measuring over the shiny uncorroded metal interconnects, using them as a mirror.An example of a NIRA spectrum of an EVA encapsulant measured in this way is presented in Figure 1a, showing many well-resolved spectral bands both in the first-overtone/combination range (1600-2500 nm) and in the second-overtone range (1000-1600 nm).In particular, NIRA spectra of EVA show a pronounced peak at 2140 nm which is assigned to the first overtone of a combination of C═O and C─H vibrations [9,10,13,14] and can be used to quantify the presence of carbonyl groups.The band centered at 1910 nm originates from the first overtone of O─H vibration in H 2 O [9,10,13,14] and can be used to evaluate the water content.The range of 1700-1800 nm shows a series of peaks related to the vibrations of C─H bonds in aliphatic hydrocarbon fragments of EVA. [9,10,13,14]n the present work, an automated procedure for processing NIRA spectra was developed, allowing the sections of the spectrum at 2120-2170 nm, 1880-1920 nm, and 1720-1750 nm to be cut, corrected for the baseline, and integrated.This procedure yields integral intensities of C═O-related absorbance (I CO ), water-related absorbance (I W ), and a reference absorbance I ref , the latter being noncharacteristic and originating from C─H vibrations in methylene fragments of EVA, respectively.The CI and WI were calculated from these values as CI = I CO /I ref and WI = I W /I ref and used to characterize the relative content of carbonyl species and water in the EVA encapsulants.
The presence of inorganic components (glass, silicon cells, rutile titania pigments in BSs) does not alter the positions and relative intensity of the bands in NIRA spectra, only affecting the absolute intensity of the NIRA spectra due to light scattering effects.In particular, the NIRA spectra measured over the highly reflective metal interconnects and over close-by light-scattering silicon cells were found to yield very similar values of CI and WI, but with much higher uncertainty in the latter case due to light scattering and a suppressed signal-to-noise ratio.
The carbonyl indices calculated from C═O-related features in Fourier transform infrared spectroscopy [15][16][17] and Raman [13] spectra are frequently used to track the degradation of polymer compounds under the conditions of photochemical [15][16][17] or thermal degradation. [18]The present article further extends this methodology, deriving CI values from the first-overtone/combination NIRA spectra.It should be noted that the relevance of the spectral NIRA feature observed at 2140 nm for the assessment of the carbonyl population in EVA encapsulants was confirmed earlier by an almost linear relationship between the relative intensities of the 2140 nm band in NIRA spectra and the carbonyl-related 1740 cm À1 band in Raman spectra of EVA samples. [9,19]onsidering the high tolerance of NIRA spectroscopy to sample state and preparation, [9,13,14] as well as the current availability of portable NIR spectrometers, the present report provides a solid argument for using NIRA-derived carbonyl indices for the field analysis of polymer materials.
As reported earlier, [19] NIRA is not sensitive enough to detect the presence of additives (UV filters, antioxidants, adhesives, etc.) in EVA encapsulants because of their small content (below 1 w%).For this reason, NIRA spectra of EVA have the same structure for all tested modules showing only spectral features typical for the bulk polymer matrix and revealing no variances that can be related to differences in the additive composition of the EVA encapsulant in the inspected modules.

Identification of BS Type
The automated procedure for processing the spectra, realized as a Python script, can also perform preprocessing and analysis of NIRA spectra, identifying the most probable BS type as described in detail elsewhere, [10] and grouping the CI and WI values of corresponding EVA encapsulants according to the BS type.In this way, BS-specific sets of CI/WI values were collected for each of the inspected fields.Five major BS types were identified and named by the composition of the air layer (Table 1).The BS identification is performed by comparing NIRA spectra collected in the field with a library of NIRA spectra of representative BS samples for which the structure was identified independently, using cross-sectional Raman microscopy. [9,10]ypically, the BSs showed a layered-stack structure with the air layer composed of different materials, including PET, polyvinyl fluoride (PVF), polyvinylidene fluoride (PVDF), and a copolymeric fluorinated coating (FC).The core layer of such BSs is made of PET, while the inner layer in contact with EVA can be either polypropylene (PP) or polyethylene (PE).Both the air layer and the inner layer are filled with microcrystalline rutile titania pigments that can be readily observed in Raman spectra. [9,10]Along with multilayer (laminated) BSs, coextruded polyamide (PA), BSs were also identified, showing distinct rutile-enriched air and inner layers and a core layer composed of a mixture of PA and PP (Table 1).More details on the BS identification and structure can be found in recent reports. [9,10,20]3.Relevance of CI for the Field Degradation of PV Modules Carbonyl species are present in any EVA sample and originate from acetate groups of vinyl acetate component of EVA copolymer.New PV modules used as references in the present article (see Discussion below) show the presence of carbonyl in acetate groups in EVA unaffected by the cross-linking step at the PV module manufacturing process.All inspected field-aged modules showed CI values higher than those observed for corresponding reference modules, indicating that new carbonyl species were formed during the field exposure of the modules as a result of oxidative degradation.[5,7,21] For example, a nonexposed FC-based module manufactured in 2012 showed CI = 0.148, while an average CI value of 0.220 and maximal CI of 0.243 were observed in 2023 for a PV field in Northern Germany (NG) for this BS type, as discussed below.Hence, the CI can be used as a quantitative parameter to characterize the field EVA degradation.
To illustrate the relevance of the CI parameter as a general quantifier for the PV module degradation, we subjected a series of 60 field-aged 60-cell modules with PVF-type BS and ≈10 years of field exposure to a WLR test as described in detail elsewhere. [8,20]The WLR is a generally accepted indicator of the degradation of the polymer insulation materials in PV modules, decreasing as the barrier properties of BS and EVA layers deteriorate due to the field degradation.For the given set of field-aged modules, the CI of the EVA encapsulant was found to be inversely proportional to the WLR, with higher CIs corresponding to lower WLRs (Figure 1b), showing a distinct negative correlation (Pearson correlation coefficient ρ = -0.95) between the population of carbonyl species in EVA and the barrier properties of the BS/encapsulation of the tested modules.
A similar dependence was also observed for a series of FC-type modules (data not shown), attesting to a general character of the WLR-CI correlation for different BS/EVA compositions.These data indicate that spectral CI measurements in the field can support WLR measurements, enhancing the reliability of the assessment of the PV module insulation state.
Based on the analysis of a spectral dataset collected during the inspection of several PV plants (see Discussion below) we found a positive correlation (ρ = 0.90) between water content (WI) in EVA encapsulant and carbonyl content (CI) for several BS types (Figure 1c).This observation indicates a direct relationship between the oxidation state of the EVA encapsulant and the permeability of the BS/encapsulant layers to environmental moisture.At that, the EVA layers with higher a CI show a higher moisture content.The field-exposed modules stay in equilibrium with the environment at given climatic conditions resulting in a certain saturation with moisture and air oxygen.As the BS and EVA degrade and lose their barrier properties, the permeation of environmental gases increases resulting in a new and deeper saturation state that is characterized by a higher amount of moisture in the EVA layers.It can be assumed that air oxygen penetrating the modules together with moisture contributes to deeper oxidation of the EVA layers, making them ever more hydrophilic and increasing the amount of moisture that can be retained in the EVA encapsulant.This continuous oxidative degradation of EVA and accumulation of higher amounts of water inside the modules results in a direct correlation between WI and CI parameters.Both indicators, CI and WI, are derived from the same NIRA spectra of EVA encapsulants, and they can both be used to evaluate the current degradation state of the modules.We note that even though both indices originate from the same spectra, the bands used for the calculations relate to different chemical species, allowing CI and WI to be taken as independent variables.According to our observations in different weather conditions and seasons, WI can dynamically vary reflecting variations in the external humidity, while CI is a more conservative and weather-independent parameter.As such, CI can be used to evaluate the extent of general degradation of PV modules, while tracking WI can show the susceptibility of PV modules to current weather conditions. [22]A deeper analysis of CI-WI correlations goes beyond the topic of the present article.
We note that the CI/WI values typically measured in the lower left corner of the modules cannot be representative of other parts of the same module far from the measurement spot, for example, in the middle or on the upper rim of the same module.However, if collected in the same spots by the same procedure in similar weather conditions for many PV modules of identical type and size (as done in the present work), the CI/WI values can be adopted as quantitative degradation markers to enable comparisons among the modules on the same field and among different fields with the same type of modules.
Some of the inspected modules, especially those with FC and PA BSs, showed very distinct signs of oxidative corrosion of the metal interconnects (see exemplary photograph in Figure 1d).NIRA measurements along the partially corroded metal interconnects revealed a clear relationship between the visual signs of corrosion and the local CI and WI values.In particular, CI and WI detected on relatively intact fragments of interconnects were found to be much lower than the values collected on strongly corroded fragments (Figure 1d).These data show a direct relationship between the visually observed progress of metal corrosion in particular spots of the interconnects and the population of moisture and carbonyl species in EVA close to these spots.At that, the higher local concentration of water and content of C═O groups in EVA correspond to a higher local density of the metal corrosion products.Considering the fact that NIRA spectra of EVA are routinely collected above the metal interconnects, the CI and WI values can be used directly as quantitative indicators of the corrosion depth of metal components of PV modules.
We note that NIRA is not sensitive enough to identify directly the presence of additives in EVA encapsulants.The additives, such as UV filters, antioxidants, adhesives, etc., though present in a low amount (below 0.5%), can still contribute to the measured CI and introduce some uncertainty that was not controlled in the reported experiments.

Collection of BS-Specific CI Data from Different PV Plants
As described above, NIRA spectra were collected pair-wise, both for BS and for EVA encapsulant of each of the inspected modules.The BS type was then identified from the NIRA BS spectrum, while the CI values were calculated from the NIRA spectra of corresponding EVA encapsulants and related to the specific BS types.
In this way, we inspected 12 PV power plants located in NG, Southern Germany (SG), Southern France (SF), and ZA.NIRA data were collected during the spring-summer seasons 2020-2023.The size of power plants ranged from 5 to 90 MWp, while the service time of the installations varied from 1 to 12 years.At each of the plants, specific sites for spectral inspection were selected to include PV modules with several BS types within the same power class and located close to each other, i.e., in the same row or several neighboring rows.Two of the fields in NG were inspected twice-in the summer of 2021 and the spring of 2023, with the spectral data collected mostly for the same PV modules on both inspections.Table 2 provides an overview of the inspected PV installations.
The BS-specific sets of CI values collected on different plants typically show normal Gaussian distributions (Figure 2).The position of the peak of CI distribution, CI P , was adopted in the present work as a cumulative indicator, which can be used to make comparisons between different BOMs, plants, and climates (characterized by Köppen-Geiger climate class).Additionally to CI P , the full width on half-maximum (FWHM) was calculated for each Gaussian fit to illustrate the width of CI distribution (Table 2).
Several conclusions can be made from the CI distributions as discussed below: 1) Modules with the same BS material, the same age of PV plant, and situated in a similar climate show strongly overlapping CI distributions and almost the same CI P values.This case is exemplified by Figure 2a which shows CI distributions collected for modules with PVF-type BS in SG and SF, both belonging to the same Cfb Köppen-Geiger climate class.2) Modules with the same BS material and the same geographic location, but with a different age of plant show distinctly different CI P values (Figure 2b).This observation is exemplified by PET-type PV modules showing a CI P shift from ≈0.14 for a 6-year-old PV field to 0.18-0.19for a 10-year-old PV plant.For the dataset collected in NG (Table 2), CI P shifts from ≈0.17 for PVF BSs to 0.18-0.19for PET to ≈0.19 for PA and to 0.20-0.21for FC BSs.We note that the BS-dependent CI P shifts are typically smaller than FWHMs of corresponding distributions (Table 2) and, therefore, cannot be considered statistically significant.At the same time, the shift shows the same trend of BS susceptibility to environmental degradation, PVF ≈ PVDF > PET > PA > FC, observed in our previous combined lab/field studies. [8,20]The dataset collected in ZA also showed a small difference between CI P values for PV modules with FC and PVDF BSs (Figure 2d) attesting that the trend in the BS-type dependence of CI P is general for different climates.
4) The same modules inspected twice, in 2021 and 2023, show small changes in CI distributions with CI P increasing with the age of the plant.At that, the increment of CI P distinctly depends on the BS type of inspected modules (Figure 2e,f ).We note again that the CI P variation illustrated in Figure 2e,f is still too small to be considered statistically reliable and requires longer observations, but it shows a trend of increasing CI P with longer exposure for the same set of tested modules.Köppen-Geiger climate classes taken from ref. [28], photovoltaic Köppen-Geiger classes (given in parenthesis) taken from ref. [29]; b) The total irradiation dose was calculated as a product of the field age and the averaged by-year solar irradiance calculated from public meteorological data. [27] The dependence of the carbonyl indices of EVA encapsulant on the BS composition of the field-aged modules can be rationalized by considering the mutual dependence of the degradative processes in the BS and encapsulant.At that, the chain of degradative events starts with the photochemical degradation of EVA encapsulant under solar UV irradiation. [23,24]The photochemical degradation of EVA can occur by many pathways including Norrish type-II elimination of acetic acid [23,24] and formation of C═C species which are considered to be responsible for the fluorescence typically observed for field-aged EVA encapsulants. [6,24]he acetic acid permeates BS layers resulting in various degradative processes accelerated by high environmental temperatures and the presence of water. [25]As a result, BS degradation by photogenerated acetic acid is expected to depend on climatic conditions. [5,7,21]The degradation of BSs promotes the gradual loss of their mechanical stability and resistivity to environmental agents, including water and air oxygen.The more damaged BS is, the higher saturation of the PV module with environmental water vapor and air oxygen can be achieved.
The penetration of O 2 and H 2 O from the environment through the damaged BS into the encapsulant layers results in secondary degradation, including oxidation of fluorescent species in EVA which can be observed by oxidative fluorescence quenching and formation of characteristic quenching patterns on the cell edges and over cracks in Si cells. [6,24]We assume that this secondary degradation is the main source of added carbonyl species shown by an increased CI of field-aged modules as compared to the reference ones.
In this very simplified model, the initiating actor which starts the degradative processes and to some extent controls their depth is the solar irradiation absorbed by the EVA encapsulant.In general, the proposed sequence of events resembles a typical autocatalytic reaction [26] with a slow induction period required for the formation of a catalyst (in the present case-acetic acid) followed by a fast acceleration period showing maximal reaction rates.At this period, the kinetic curve of the process can be described by exponential growth. [26]The acceleration phase is followed by a saturation phase with the reaction rate eventually dropping almost to zero.
To evaluate the applicability of such a kinetic model to the development of EVA degradation in PV modules we plotted, the peak of CI distribution, CI P , as a function of the total irradiation dose (total solar exposure), that is, the total dose of solar illumination received by a PV plant during its lifetime (Figure 3, Table 2).Total solar exposure was calculated as a product of the age of installation and the averaged by-year solar irradiance was calculated from publicly available meteorological data for all PV locations. [27]xcept for zero abscissae, each point in Figure 3a represents a peak of CI distribution determined by Gaussian fitting of a set of multiple CI values collected for each BS type on each particular PV field.The point x = 0 corresponds to reference modules, which should be of the same type and manufacturing year as the modules inspected in the field but with zero exposure to solar irradiation.Considering the age of inspected PV plants (up to 12 years) and the fast progress in the design and manufacturing of silicon PV modules, access to such reference modules is very challenging.Five reference modules held in stock (storage rooms) of corresponding power plant operators were tested, one per BS type, including FC, PA, PET, PVF, and PVDF.This set of reference samples cannot pretend to be representative in the statistical sense, but it provides a notion about the range of CI values expected for the modules with zero solar exposure.As mentioned above, the CI values of reference modules are in all cases lower than the values measured for the corresponding field-aged modules, being in the range of ≈0.14-0.15for all tested BS types.The absolute CI values reported in the present work are valid only for the specific set of experimental conditions, that is, the spectrometer used for the measurements and the method applied for the CI calculation.Our results indicate that the standardization of CI measurements is an important and, at the same time, challenging issue, which needs to be addressed for the possible utilization of CI as a universal degradation marker for reporting by different groups for different climates and different PV module architectures.
The data presented in Figure 3 show that CI P values measured for different plants and BS types generally constitute a uniform dependence on the total irradiation dose, growing with an increase in solar exposure.This dependence allows the progression of the module degradation status to be seen for the PV plants with different ages and yearly irradiances (Table 2).
The dependence of CI P on the total irradiation dose can be fitted by an exponential function (Figure 3b), indicating an accelerated character of the degradation development.Figure 3b shows a relation of CI P on the total exposure exclusively for PVF-type modules with a corresponding single-exponential fit showing a growth constant of 1 / τ = 0.17 m 2 MWh À1 .
For the cases of artificial photodegradation of polymers, sigmoidal shapes of the dependence of the CI on the total irradiation dose were reported with a saturation state achieved after the initial accelerated degradation. [15]Most probably, the present observations refer to the initial accelerated section of the dependence and the saturation is still to be expected in the years to come.Our observations of the focusing of CI distributions with time, which indicate a lower degradation rate for a deeper degradation state, also support the expectation of a saturation event.As we have observed no saturation so far, it is impossible to fit the entire "CI Pirradiation dose" dependence by a sigmoidal function with acceptable accuracy.Still, the intermediate fast phase of the CI P evolution is satisfactorily described by an exponential function, indicating that a sigmoidal function will become applicable to the present dataset when the state of saturation is finally achieved.
In addition to the exponential dependence of the CI P on the total exposure, a clear correlation with the BS type can be observed in the datasets for the same total solar exposure.For example, an inspection made in 2021 in NG (fields 2,3 in Table 2) showed a definitive growth of CI P from PVF to PVDF to PA to FC (see also dashed rectangle for 2021 in Figure 3a).This BS order was confirmed also in 2023 by the second inspection of the same modules (dashed square for 2023 in Figure 3a), FC-type modules showing a faster degradation, as discussed below.The order of BSs by increasing CI P , FC » PA > PVDF > PVF is also in agreement with our previous reports on BS relations observed on different levels, from the BS-dependence of water penetration and insulation resistance on the single-module-level [8,20] to BS-dependent inverter performance on the system level. [8] NIRA dataset collected in a strongly different climate of ZA also showed a clear dependence on the total irradiation dose, indicating the universal character of this relationship for different climatic zones (Figure 3a).In this case, the "CI Pirradiation dose" dependence can also be fitted with an exponential function (Figure 3b), however, producing a much lower growth constant, 1 / τ = 0.07 m 2 MWh À1 .The slower development of degradation signs in ZA can be related to low humidity in the zones of inspection and needs more detailed investigations.Preliminary analysis of climatic data relevant to the inspected sites showed indeed a generally higher annual distribution of relative humidity in Germany (45.7-94.3%)as compared to ZA (11.6-63.7%), the ranges evaluated for 2%-and 98%-percentiles for daily values.[27] Additionally, the development of degradation is expected to depend on the daily humidity profile.The humid PV modules can dry completely during hot sunny days in ZA (Köppen-Geiger climate class BSk) while retaining some residual humidity continuously trapped in a more moderate and cooler mid-European climate (climate class Cfb).
Similar to EU, the South African PV plants of the same age revealed a dependence of CI P on the BS type, showing a shift of CI distribution to higher values for FC-type modules as compared with PVDF-type modules (Figure 2d and 3a).We note that the data array collected so far for ZA-located plants remains small as compared to the data assembled in EU.Currently, we perform further investigations in ZA aimed at providing a broader perspective and a more firmly substantiated interpretation of differences in the degradation behavior of PV modules in strongly different climates.
Apart from a general evaluation of the degradation status of a PV field, the inspection of carbonyl indices can potentially be used for regular monitoring of the degradation progress.Here, we provide preliminary proof of the feasibility of such an approach by presenting two sets of CI values measured on the same PV plants for the same set of modules in 2021 and 2023 (fields #2 and #3, Table 2).For the particular case of CI measurements collected for FC-type modules, a small, but distinct shift of the CI distribution can be observed in 2023 as compared with 2021 (Figure 2e).For the case of PVDF-based modules measured at the same PV plant, only negligible changes in the CI distribution can be detected (Figure 2f ).The changes in CI P for four BS types found in fields #2 and #3 are summarized in Figure 3c.The temporal trend can be seen for all four types of BS, with the most pronounced changes detected for FC-type samples.We stress again that these observations are still below the statistical significance and have a preliminary character requiring additional collection of data.However, they illustrate the potential of NIRA-based spectral inspection for the evaluation of the current state of PV plants and the evolution of degradative processes depending on the BS type and climatic conditions.

Conclusion
We introduce a new NIRA-based spectroscopic approach for the noninvasive field evaluation of the oxidative degradation of EVA encapsulants.The CI derived from the NIRA spectra as a quantifier of the EVA degradation correlates with the WLR, water content, and the metal corrosion state of tested modules.These observations indicate that the CI of EVA encapsulants can be used as a quantitative measure of the general degradation state of PV modules.
A field study was performed showing the CI to depend on the bill-of-materials, in particular, on the backsheet composition, as well as on the age of PV modules, and the climatic operation conditions of PV plants.In this study, twelve multi-MWp PV plants located in EU and ZA with installation ages varying between 1 and 12 years were inspected.
We found that the collected dataset can be presented in a unified way by relating CI distributions to the corresponding total irradiation dose.In particular, by analyzing CI data from eight PV plants in EU, we found that the peak of CI distribution, CI P , increases with total solar exposure.The same trend can be observed for CI values collected on four PV fields in ZA, however, showing a much slower growth constant (0.17 m 2 MWh in EU vs 0.07 m 2 MWh in ZA) which correlates with a lower humidity observed in the latter case.
For the same solar exposures, CI P values were found to distinctly depend on the BS structure.The highest CI P values were observed for FC-type BSs, the lowest-for PVF-type BSs, and the inspected BS formed the following order by decreasing CI P : FC > PA > PET > PVDF > PVF.This order corresponds to our previous observations on BS reliability both on the single-module and inverter levels.
By repeating CI measurements on the same plant after a 2-year exposure, we showed the feasibility of using NIRA spectroscopy to track the temporal evolution of the degradation status of PV modules.In total, the present observations show that spectroscopic NIRA inspection of EVA/BS combinations can potentially be used for comparative and comprehensive evaluation of the degradation status of PV plants depending on their BOM, age, and climatic zone.
As an outlook, we note that further development of this method for large-scale PV installations will require an upscaling from manual to automated high-throughput collection and analysis of spectral data, e.g., using drone-assisted data acquisition.At the same time, to enable a broader use of this method and to facilitate the comparison of spectral data produced by different groups with different equipment in different climates, standard protocols for data acquisition and processing have to be developed.As the long-term degradation of PV encapsulants is mostly oxidative, we expect the proposed spectroscopic approach to apply not only to EVA copolymers but also to other types of PV encapsulants, such as polyethylene-based copolymers, which can develop carbonyl functionalities due to the oxidative degradation.

Figure 1 .
Figure 1.a) NIRA spectrum of EVA encapsulant.Photographs in inserts 1 and 2 illustrate the process of NIRA spectra acquisition.b) Relationship between the WLR and CI of a series of PV modules with PVF-type BS (scatter).A solid line represents a linear fit of the experimental data, corresponding to the equation CI = 0.233(AE0.005)-5.80Â 10 À4 (AE0.06Â 10 À4 )ÂWI.c) Relationship between carbonyl and water indices collected from NIRA spectra of EVA encapsulants for PV modules with three different BS types-PVF (green triangles, 169 modules), PA (blue circles, 70 modules), and FC (red rectangles, 160 modules).d) Water (WI) and carbonyl (CI) indices at different points along a partially corroded interconnect (point numbers indicated on the photograph) of a PV module with FC-type BS.Abbreviations and symbols: PVF-polyvinyl fluoride; PA-polyamide; FC-fluorinated coating; ρ-Pearson correlation coefficient.

3 )
Modules with different BS materials found in PV plants of the same age and the same geographic position show slightly different CI P values as exemplified by two datasets collected in different climates of NG (Figure 2c, Köppen-Geiger climate class Cfb) and ZA (Figure 2d, Köppen-Geiger climate class BSk).

Figure 2 .
Figure 2. a-c,e,f ) Distributions of carbonyl indices CI collected on different fields in EU and d) ZA and grouped by the BS type (field number and BS type noted on figures, see Table 2 for details).Solid lines represent Gaussian fits of CI distributions.Abbreviations: PVF/PVDF-polyvinyl/polyvinylidene fluoride; PET-polyethylene terephthalate; PA-polyamide; FC-fluorinated coating.
c) Fields 2,3 in NG and 5,6 in SF are located in the same respective areas and composed of PV modules of the same type, age, and manufacturer; d) Abbreviations: SG/NG-Southern/Northern Germany; SF-Southern France; ZA-South Africa; PVF/PVDF-polyvinyl/polyvinylidene fluoride; PA-polyamide; FC-fluorinated coating; PET-polyethylene terephthalate.

Figure 3 .
Figure 3. a) A summary of collected peak CI P values for different fields in EU and ZA.The BS type and location (EU, ZA) are noted in the figure.Colored traces are presented only to guide the eye.b) Dependences of peak CI P for PVF-and PVDF-type modules on the total solar exposure collected in EU (circles) and ZA (balls).Dashed lines represent corresponding single-exponential fits.c) Peaks CI P values were determined in 2021 and 2023 for the same fields (#2,3 in Table 2) for PV modules with different BS types.Abbreviations: PVF/PVDF-polyvinyl/polyvinylidene fluoride; PET-polyethylene terephthalate; PA-polyamide; FC-fluorinated coating.

Table 1 .
A summary of identified BS types.Layer thickness (in μm) is indicated in parentheses.

Table 2 .
A summary of inspected PV fields.