Solar-NIRT: Identification of PV-module backsheets in the field with natural sunlight

Reliable and durable solar power plants require PV modules with high-grade polymer encapsulants and backsheets (BSs). For performance analyses of PV installations fast, reliable and non-destructive methods for determining composition and degradation state of polymer components need to be developed. Here, we show that the structure of some common polymer BSs can be determined in the field in real time by analyzing near-infrared transmission (NIRT) spectra collected under illumination with natural sunlight. The potential of this “ Solar-NIRT ” method was probed by field measurements on a multi-MW PV power plant where four major BS types were identified by multispectral cross-sectional Raman imaging. Additionally, degradation of a particular BS type was found to result in distinct changes in NIRT spectra allowing the degraded BSs to be classified as a separate type. Principal component analysis (PCA) applied to a collection of 62 Solar-NIRT spectra allowed to create a map of five clusters, each corresponding to a particular BS type. The feasibility of using the PCA cluster map for the identification of unknown samples was shown on a test set of 13 different BSs.

with natural sunlight. The potential of this "Solar-NIRT" method was probed by field measurements on a multi-MW PV power plant where four major BS types were identified by multispectral cross-sectional Raman imaging. Additionally, degradation of a particular BS type was found to result in distinct changes in NIRT spectra allowing the degraded BSs to be classified as a separate type. Principal component analysis (PCA) applied to a collection of 62 Solar-NIRT spectra allowed to create a map of five clusters, each corresponding to a particular BS type. The feasibility of using the PCA cluster map for the identification of unknown samples was shown on a test set of 13 different BSs. The Solar-NIRT is relatively fast, non-invasive, selective, can be upgraded to a non-contact regime making it a promising tool for high-throughput characterization.

| INTRODUCTION
Degradation of polymer components of silicon PV modules is currently recognized as one of the major issues resulting in module under-performance, failures, and shortened lifetime. [1][2][3][4][5] Both organic encapsulant surrounding the silicon wafers and organic module backsheets (BSs) protecting the module from adverse climatic effects can be subject to chemical and mechanical degradation caused by UV irradiation, broad and rapid temperature variations, and attacks by atmospheric gases. The degradation processes in the encapsulant and the underlying BS mutually affect and accelerate each other. 2,4-6 As mechanisms of degradative processes can be specific for different encapsulant and BS materials, any proper analysis or monitoring of the degradation status of PV modules should start with the determination of the chemical composition of these materials. Such analysis is also important for the initial quality check of the PV modules before their commissioning because manufacturers typically do not reveal bill of materials used in their products.
The analysis of the chemical composition of encapsulants and BSs should comply with a set of requirements, including a non-contact (or at least non-destructive) character, high speed, reliability, and feasibility of large-scale field measurements. In view of these requirements, infrared spectroscopic methods, such as Fourier-transform (FTIR) and Raman spectroscopy provide unique analytic capabilities, allowing most of the PV-relevant polymer types to be identified and their degradation state to be evaluated with no mechanical or chemical intrusions into the PV modules. [6][7][8][9][10][11] However, both FTIR and Raman are mostly laboratory techniques, with only a few recent successful examples of deployment for large-scale field measurements. 12 Both methods are limited by the depth within the material from which data can be gained and can consequently not provide accurate information on multi-layer BS materials. The Raman spectroscopic identification of polymers in aged PV modules can also be impeded by fluorescence from the BSs and encapsulants and is mostly confined to new modules. 12 By all these reasons, there is still a demand for new reliable spectroscopic approaches for fast and nondestructive field identification/inspection of polymer components of PV modules.
Recently, we have reported on a high potential of near-infrared absorption (NIRA) spectroscopy for chemical identification of polymer components of PV modules. 13 This method can be applied to the operating modules directly in the field with no particular preliminary preparations and can provide analytical information on the entire polymer BS and encapsulant layers from a single measurement. [14][15][16][17] By combining NIRA with multi-spectral Raman imaging, we have shown the feasibility of applying this technique for the identification of the structure and composition of more than 10 different configurations of multi-layer BSs based on polyethylene terephtalate (PET) core insulation layers. 18 Here, we further extend this approach and show that NIRA can be used for the spectroscopic analysis of PV module BSs with no additional excitation light sources, by only using natural sunlight transmitted through the modules. As the characteristic vibration bands of many PV-relevant polymers do not overlap with the absorption bands of atmospheric gases (H 2 O, CO 2 ), we were able to distinguish between four exemplary BS types found on a multi-MW power plant used for testing the method. Additionally, degradation signs were detected for one of the BS types showing the potential of such "Solar-NIRT" spectroscopy for evaluating the degradation status of the module BSs. We expect the presented results to constitute a basis for future development of Solar-NIRT into a non-contact and highthroughput characterization tool.

| METHODS
Raman spectra and multispectral maps were collected using a WITec alpha700 confocal Raman microscope coupled to an UHTS 300 spectrometer in a spectral range of 130-3,700 cm À1 with 532-nm laser excitation. 13 Spectral maps were constructed by scanning 1-mm 2 sample sections using characteristic spectral Raman ranges of BS constituents. 13 Raman mapping was performed on BS cross-section samples produced by detaching a piece of BS from a PV module and cutting the cross-section edge along the BS length. FTIR transmission spectra were recorded from the BS air-sides with a Vertex 70 spectrometer (Bruker) with an attenuated total reflection (ATR) diamond accessory.
The Solar-NIRT spectra were recorded with a FT-NIR Rocket 2.6 spectrometer (Arcoptix) in a spectral range of 900-2,600 nm 13 using natural sunlight. Typically, 20 NIRT spectra were collected consecutively by placing an optical fiber end as a probe to a module air-side spot where sunlight can penetrate the BS and averaged for noise reduction.

| General description of measurements
The present data were collected during a field measurement campaign

| Raman characterization of the BS crosssections
The polyamide type-A BS cross-section is composed of a single references can be found in previous studies. 8,13,18,21 The core of the type-A BS (point 2 in Figure 1 Table 1).
The air-side Raman spectrum of type-D BSs (Figure 1, sample ID) is very similar to the spectrum of PET with some additional minor T A B L E 1 Summary of the BS cross sections studied in the present work F I G U R E 1 Microphotographs of BS cross-sections (row I), Raman spectra of BS components (row II), and multispectral cross-sectional Raman maps (row III) for BSs of different type A-D. In IIA, spectra were taken close to the air-side (curve 1) and in the middle of the BS (curve 2); the insert shows an enlarged section of these spectra and Raman spectrum of PP (curve 3). In IIB-D curves 1 correspond to air-side layers, 2 to core PET layers, and 3 to inner PE/PP layers. In IID, curve 4 is the air-side FTIR transmittance spectrum peaks and distinct rutile series (Figure 1, IID, curve 1), indicating that this layer is too thin to be reliably identified by Raman spectroscopy.
Additional probing of the air-side layer with FTIR spectroscopy (Figure 1, IID, curve 4) revealed the presence of two C═O species (bands at 1,720 and 1,680 cm À1 ) assigned to free and amide carbonyls as well as CF 2 and CF 3 fragments seen by C-F vibrations at 1,146/1,240 cm À1 . 8,21,26 Basing on these observations, the air-side polymer was identified as a rutile-filled fluorinated copolymer coating (FC). 18 The core layer of type D BSs is typical PET (Figure 1, IID, curve 2), while the inner layer shows characteristic signs of rutile-filled PP.

| Solar-NIRT spectra
Sunlight passes through the BS and is collected by a NIR spectrometer. The fraction of the transmitted light depends on thickness and scattering capacity of each BS ( Figure 2). Typically, a cross point between four neighboring Si wafers was used to collect the transmittance spectrum as indicated by points marked "1" in Figure 2.
A reference solar irradiation spectrum was measured together with each transmittance spectrum, typically using a gap between adjacent PV modules (point 2 in Figure 2).
Additionally, a conventional NIR reflectance spectrum was measured for each BS by using the lamp excitation and converted into a NIRA spectrum. In this way, for each BS, a set of three spectra was collected including NIRA, Solar-NIRT, and reference solar spectrum.
The   Figure 3D and D*).
These degradation-induced changes can also be observed in the Solar-NIRT spectra (compare curve 2 in Figure 3D and D*). The Solar- NIRT spectrum of D* shows a much more pronounced ═C H feature at 1,660 nm indicating a higher contribution from aromatic species in

| Multivariate analysis of NIRT data
The Solar-NIRT spectra of different multi-layer PET-based BSs show a similar structure, only varying in relative intensities of aromatic ═C H and aliphatic C H vibration bands (Figure 3). These differences are difficult to evaluate directly from the spectra but can be uncovered using multivariate analysis. 27 In particular, principal component (PC) analysis allows evaluating the variance among Solar-NIRT spectra and allows presenting them as a set of clusters.
The PC analysis was performed on a collection of 75 Solar-NIRT spectra for BSs from types A to D and D*. A portion of the Solar-NIRT spectra were selected as a test set, totally 13 spectra from each of five above-described different BS types (A, B, C, D, and D*) to verify the applicability of the PC analysis results for the BS identification. The Solar-NIRT spectra were analyzed in the range of 1,400-2,300 nm with a step of 5 nm taking each wavelength as a variable. A set of PCs was calculated, reflecting the variance between the Solar-NIRT spectra and was arranged by significance as PC1, PC2, etc.
The "PC1 versus PC2" plot presented in Figure 4a shows that all tested BS types can be reliably distinguished by the position of their respective clusters. Each point on the graph represents a single Solar-NIRT spectrum and belongs to one of the clusters according to their specific spectral features. A scree plot presented in Figure 4b shows that the first two principal components, PC1 and PC2, are responsible for almost 90% of variance observed between the tested Solar-NIRT spectra. In this view, we can neglect further PCs and analyze the Solar-NIRT spectra by using 2D "PC1 versus PC2" plots without major losses in variance. A loading plot of PC1 and PC2, that is, the distribution of variance as a function of wavelength The feasibility of the Solar-NIRT identification using the principle "library+1" was probed by using a test set of 13 PV modules representative of all five BS types studied in the present work. The results of the test (Figure 4d) showed all samples within the expected clusters. Further work is in progress to collect more statistical information on the studied BS types by field measurements.
If the geometric place of the tested sample differs from all presented clusters, we can conclude that we encounter a new BS type or a strongly degraded A-D BSs. In this case, we need to perform an additional multi-spectral Raman study of the new BS cross-section, associate the structure of such BS with its Solar-NIRT spectrum, and add the spectrum to the presented PC plot. The library of cataloged BS types can be constantly expanded increasing the accuracy of the Solar-NIRT analysis.
The Solar-NIRT identification is performed using ambient sunlight with no additional NIR light source required to probe the sample, and therefore, it may potentially be carried out in a non-contact way by collecting the sunlight coming through the backsheets distantly by a system of lenses. A combination of this capacity with a relatively high acquisition and processing speed of the Solar-NIRT spectra (totally less than 1 min per PV module), a high selectivity of the PC analysis evidenced by the present report and portable character of the measuring equipment makes the Solar-NIRT a promising tool for the non-invasive and non-contact high-throughput characterization of PV power plants.

| CONCLUSIONS
The composition and structure of some of common polymer BSs of commercial silicon PV modules can be determined in the field by analyzing NIR transmission spectra collected using illumination with the natural sunlight ("Solar-NIRT" spectra).
The Solar-NIRT spectra were found to provide as much analytical information as conventional NIRA spectra registered by using NIR light sources in a reflectance mode. The applicability of the Solar-NIRT analysis is shown on an example of a large multi-MW PV power plant where four major BS types were identified and their composition and cross-sectional structure determined by using Raman spectrocopy. In particular, we focused on polyamide-based single-layer BSs as well as on multi-layer PET-based BSs having different air-side fluoropolymer (PVF, PVDF, FC) and inner polyolefine (PE, PP) layers. These four BS types revealed characteristic differences in the Solar-NIRT spectra, which can be uncovered by applying a multivariate PC analysis allowing all four types to be reliably distinguished.
We found that photoinduced browning of FC-based BSs results in distinct spectral changes in NIR range allowing this particular degradation event to be tracked by the proposed analysis as well.
The proposed Solar-NIRT approach is expected to evolve into a selective, non-destructive and non-contact method for humanassisted or automated high-throughput characterization of large PV installations.