Distinguishing between different types of multi‐layered PET‐based backsheets of PV modules with near‐infrared spectroscopy

Degradation of backsheets (BSs) of commercial silicon PV modules is currently recognized as a source of reduced module performance and module failure. Monitoring of the BS state in the field is possible by using non‐destructive and highly informative near‐infrared absorption (NIRA) spectroscopy. Application of NIRA for the analysis of multi‐layer polyethylene terephtalate (PET) based BSs, which dominate the PV module market, is challenging due to a large variety of possible BS configurations that show only small differences in NIRA spectra. In the present work, a spectroscopic tool for the structural identification of PET‐based BSs is introduced. The method is based on a principal component analysis of a database of 250 representative NIRA spectra of BSs of different types. It allows a BS with an unknown structure to be assigned to one of 12 different types based solely on its NIRA spectrum. The identification was successfully validated on a test collection of 45 selected BSs and shown to be feasible for the field deployment. Further automation of NIRA measurements and spectral analysis are expected to elevate the proposed tool to the level of a non‐intrusive high‐throughput field analysis of the BS composition and state in operating PV module grids.


| INTRODUCTION
Degradation of polymer components of commercial silicon PV modules is currently recognized as one of the major factors limiting module performance and lifetime, and causing security and financial risks for the stakeholders of PV installations. [1][2][3] Typically, UV irradiation has the largest impact on polymer encapsulants of Si wafers while climatic stress (high/low temperatures, humidity) affects mostly the polymeric backsheets (BSs) designed to provide a mechanical support to solar cells as well as to insulate and protect the cells from the environmental factors. [1][2][3][4][5][6] The degradation of both encapsulant and BS can influence and accelerate each other, for example via partial BS decomposition by the products of encapsulant hydrolysis and, vice versa, via encapsulant deterioration by moisture and oxygen penetrating through a partially decomposed BS. 1,3,[5][6][7][8][9] As a result, the long-term behavior and degradation stability of PV modules depend on composition and structure of both encapsulant and BS. 1,[3][4][5]8,9 Due to this dependence, as well as the fact that the composition of polymer components is not typically disclosed by the PV module manufacturers, there is a need for fast, non-intrusive, and reliable analytical tools for the identification of encapsulant and BS composition, as well as for the evaluation of their composition-dependent degradation status.
Spectroscopic methods, such as Fourier-transform infrared (FTIR), Raman and fluorescence spectroscopy, were found to be most suitable to address this challenge due to their high chemical specificity and non-destructive character. 1,7,[10][11][12][13][14] However, these techniques are capable of probing only the BS surface layer and cannot provide information on the inner structure of multi-layer BSs that are typically several hundred μm thick. By this reason, the investigations into structure and degradation state of multi-layer BSs using Raman, FTIR and fluorescence spectroscopy typically have an invasive character and need to be performed on BS cross-sections. 7,[13][14][15][16][17] In-depth noninvasive probing by confocal Raman spectroscopy is feasible for new PV modules but typically hindered for field-aged samples due to a strong fluorescence from partially degraded encapsulant and BS materials. 14 Recently, near-infrared absorption (NIRA) spectroscopy has been recognized as a viable alternative to other spectroscopic approaches 14,18 that can be upgraded from lab tests to large-scale field measurements and implemented as a non-destructive and noninvasive high-throughput characterization tool for large PV systems.
NIRA can provide a combination of penetration depth (more than a thousand μm), versatility of analytic information, and field applicability unrivaled by other spectroscopic approaches proposed for the characterization of PV modules so far.
NIRA probes vibration overtones and combinations of them which are orders of magnitude weaker than the fundamental vibrations in organic polymers typically registered by FTIR. [19][20][21][22][23] By this reason, NIRA can be applied to almost any kind of sample without additional preparations, [19][20][21][22][23] it is not sensitive to moderate soiling of PV modules or to inorganic components (cover glass, Si wafers, pigments in BS) 18 allowing the polymer components of PV modules to be probed both from the front side (encapsulant) and the air side (BS) directly on operating modules in reflectance mode.
For most polymers used in PV modules, such as polyethylene terephtalate (PET), polyamide or ethylene vinyl acetate (EVA) copolymer encapsulant, NIRA spectra reveal characteristic features, 19,[21][22][23] allowing their accurate and fast identification, and discrimination from other BS components, such as pristine and fluorinated polyolefins.
Due to relatively large penetration depth of the NIR light (1 mm and more), 19 the entire multi-layer BSs can be probed in a single measurement. Finally, NIRA spectroscopy is fast and can be realized using por- To address this challenge, we propose a method for spectrally identifying PET-based multi-layer BSs by using a combination of NIRA spectroscopy with multivariate principal component analysis. The method is supported by a library of BS cross-sections created by the multi-spectral Raman imaging of different BS types. The approach includes (i) assembling a library of representative BSs from a variety of module types and manufacturers, (ii) identification/mapping of the BS composition and cross-sectional structure by multi-spectral Raman imaging, (iii) performing a multivariate analysis of variance among the NIRA spectra of representative Raman-mapped BSs.
The latter analysis allows reducing each NIRA spectrum to a single point on a principal component (PC) plot assigning all BSs of the same structure to a separate cluster. In this way, a library of clusters was generated using BSs with the known structures and known NIRA spectra. This library can then be used as a basis for independent lab and field identification of BSs in unknown PV modules using solely non-invasive and portable NIRA setup.

| METHODS
Raman spectra and spectral maps were detected on a WITec alpha700 confocal Raman microscope equipped with an UHTS 300 spectrometer in a spectral range of 130-3700 cm À1 and a resolution of 3 cm À1 . The samples were excited by a 532-nm laser with the maximal power of 7 mW. Spectral maps were constructed by scanning square (1000 Â 1000 or 500 Â 500 μm) sections of the sample with a resolution of 100 Â 100 spectral points (100 Â 100 pixels with spectral information on each pixel). Raman spectroscopic measurements were 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 length of the BS.

| General description of tested samples
Recently we have reported on the potential of NIRA spectroscopy for the characterization of composition and degradation state of polymer components of commercial silicon PV modules, in particular BSs and encapsulants. 18 We have found that despite the immense diversity of The NF group is dominated by polyamide, which was found in roughly 25% of modules tested in the present work. In some old modules, PET-based non-fluorinated BSs were also found, while some of the modern modules showed PP-based BSs. The major part (about 75%) of tested modules were constructed with BSs composed of a PET core layer protected by one or two fluorinated polymer layers. Totally, the multi-layer PET-based SF and DF BSs together with polyamide NF BSs accounted for about 95% of all tested modules.
As shown in our previous report, 18 polyamide BSs can be distinguished from PET-based ones by using characteristic NIRA bands without the need for a sophisticated spectral analysis. The PP-based BSs show neither PA not PET features in NIRA spectra and can also be easily identified. In contrast, multi-layer PET-based SF-and DF BSs show similar NIRA spectra that differ only by the number, order and thickness of the constituent layers. As a result, spectral NIRA discrimination among various multi-layer PET-based BSs becomes a challenging task. In the present work, we limit our discussion to PET-based BSs and exclude PA and polyolefine BSs from the analysis.
To gain insights into the composition and structure of PET-based BSs, we collected BS cross-sections from 84 PV modules differing by type, manufacturer, age, and history (new, shelf-stored, and field-aged modules). The BS samples were characterized by Raman spectroscopy and multi-spectral confocal Raman microscopy, allowing major polymer components and polymer fillers to be identified.

| Building multi-spectral Raman maps of BS cross-sections
Confocal Raman microscopy allows probing cross-sections of BSs or BS/EVA assemblies with a spatial resolution of around 1 μm, resulting in 2D maps with a chemical contrast achieved by analyzing characteristic spectral ranges of different BS components. 17,18,24 By using this approach and the above-discussed Raman/FTIR identification of the BS components, we constructed a multispectral Raman map for each particular BS type, visualizing simultaneously composition, order, and thickness of BS layers.
The structure and characteristic features of Raman spectra of the most important polymer BS components and EVA encapsulant as well as details of the multi-spectral Raman mapping were discussed in detail in our previous report. 18  The Raman mapping of BS cross-sections adopted in the present study does not allow any adhesive layers between air-side, core, and inner BS layers to be reliably distinguished and chemically identified.
However, as the reported thickness of adhesive layers are relatively small (on the order of 1 μm) and the chemical composition of such adhesives is close to that of BS layers we do not expect any critical influence of these layers on the accuracy of the spectroscopic NIRA identification of different BS types.
Totally, we collected 84 multi-spectral maps of BS cross-sections from the NF, SF, and DF types and recognized twelve distinctly different BS configurations, summarized in Table 1.
The NF group of BSs revealed three BS configurations. In particular, some old PV modules (more than 20 years of commission) showed NF BSs composed of a single layer of PET filled with inorganic pigments (NF-1 and NF-2 types in Table 1). The NIRA spectra of such BSs are identical to those of pure PET. Additionally, we recognized a third type of multi-layer PET-based NF BSs which can be found in modern PV modules (type NF-3 in Table 1).
The SF group includes BSs with a single rutile-filled fluoropolymer layer, mostly found on the BS air side, a PET core and additional inner

| NIRA spectroscopic analysis of PETbased BSs
One of the advantages of NIRA as compared to other spectroscopic techniques of the BS characterization is a large penetration depth of near-IR light of up to 1.0-1.5 mm. [19][20][21][22]25,26 The BS thickness of commercial silicon solar modules is typically smaller, about 300-500 μm, Adsorbed water reveals a characteristic band at 1910 nm. 22 Other peaks, including a distinctly resolved complex band at 1730 nm, correspond to various overtones and combinations of C-H vibrations. 19,20,22,25 The NIRA spectrum of NF-2 BS ( Figure S10a, curve 2) is expectedly similar to that of NF-1 and shows a more distinct water-related 1910 nm band due to a deeper degradation state of such BSs, most probably because the BaSO 4 pigment is not efficient in absorbing UV irradiation and more prone to water uptake as compared to rutile titania.
The NIRA spectrum of NF-3 BS ( Figure S10a, curve 3) shows the same features; however, the -C-H band at 1730 nm is much more intense as compared to NF-1 and NF-2. We assigned this feature to the presence of an inner PE layer revealed by cross-sectional Raman studies.   Figure 2). As the air side layers of SF-3 and SF-4 are of the same material and thickness the strongly suppressed character of PET band can be assigned to a rough topography of the SF-4 air side, while SF-3 demonstrates a smooth and uniform air side (panel 1 of Figure 4). The uneven morphology of the SF-4 air side is expected to contribute to strong scattering of NIR light resulting in generally low resolution of all spectral features.

| NIRA analysis of group SF
The NIRA spectra of SF-5 BSs were found to be strongly distorted by interference pattern which is specific for this particular BS type. The interference pattern blends with NIRA spectrum changing irregularly for different modules and probing points and making impractical a spectral analysis of such BSs in the NIR reflectance mode.

| NIRA analysis of group DF
A gallery presented in Figure 3 shows that the cross-sections of such BSs are similar differing in the thickness of PVF and PET layers (panels 2 and 3). Accordingly, NIRA spectra of DF-type BSs reveal the same structure only varying in the I 1660nm /I 1730nm ratio. Similar to SF-type BSs, delamination of DF BSs shows the complexity of their NIRA spectra. The NIRA spectrum of a DF BS without the air side R-PVF layer ( Figure S10c, curve 2) reveals a much higher I 1660nm /I 1730nm ratio, while the delaminated R-PVF (curve 3) shows only -C-H-related spectral features. Therefore, the I 1660nm /I 1730nm ratio is expected to reflect variations in the thicknesses of both PVF and core PET layers of DF-type BSs.
In line with expectations, we observe a reduction in the I 1660nm /I 1730nm ratio from DF-1 BSs with the thinnest air side layer (10 μm Summarizing, we found that NIRA spectra of all tested multi-layer PET-based BSs have similar structure differing mainly in relative intensities of PET-and aliphatic polymer-related features. In principle, the type of BS can be evaluated directly from the NIRA spectra using the I 1660nm /I 1730nm ratio. However, the variation of this ratio is relatively small and, considering different signal-to-noise ratio and signal intensity observed for BS of different types, the reliability of such evaluation is not expected to be satisfactory.

| Results of PC analysis
We found that the first two PCs account for 68% of variance between the tested spectra (see a "Scree plot" in Figure S11a) allowing the PC analysis results to be presented as a 2D plot (Figure 4a). Increasing the number of PCs to 3 or 4 adds only 5% of variance and does not affect the analysis in a major way.
The distribution of PC1 across the wavelengths (PC1 "loading By using the spectral database presented in Figure 4a, we can identify the BS type of an unknown sample. To do this, we add its NIRA spectrum to the database, perform the PC analysis for the "database + 1" set of spectra and determine the geometrical position of the unknown sample on the 2D PC plot relative to reference samples from the database, for which the BS structure is known.

| Validation of PC analysis results
We applied this method for the test set of 45 BSs of different types and found that each test sample added to the database appeared within the expected cluster (Figure 4c) We tested the feasibility of using NIRA-based identification for the field measurements on a multi-MW solar power plant in Germany.
The tested PV modules showed only NF-and SF-type BSs but these two types of BSs as well as different configurations of the SF-type BSs were successfully identified (more detail in a previous study 29 ). As the PC analysis can be easily automatized and a single NIRA measurement is fast, taking less than 1 s per BS (for the acquisition of a single spectrum), we expect high feasibility of automated identification of BS structure performed in a high-throughput regime.
The NIRA identification of the BS type for the case field study reported in a previous study 29 allowed certain failure behavior of PV modules to be tentatively associated with specific BS types, including occurrence of mechanical damage, potential-induced degradation, isolation problems, and metal bus corrosion. In view of these data, the automated NIRA analysis can become a potent addition to conventional techniques such as thermography, IV measurements, EL imaging, and so forth. Additionally, NIRA can become indispensable for the discrimination between fluorinated and non-fluorinated BS materials necessary for the selection of proper recycling technologies for discarded PV modules. 30 Our test field deployment of the method showed several limiting factors that currently pose challenges on the way to the highthroughput NIRA characterization. The manual character of the acquisition of NIRA spectra and hindered access to the PV modules in the test field campaign allowed only about 20-25 spectra per hour per operator to be registered and processed. The manual procedure should be performed during daytime and also requires a direct contact of the NIRA probe with the module air side. These and other limitations may be overcome by combining distant NIRA probing from a mobile robotized setup with a synchronous automatized spectra processing and BS identification based on the present BS database.

| CONCLUSIONS
We propose a spectroscopic tool for the identification of structure and composition of multi-layer PET-based BSs of commercial silicon PV modules that can be used both for the lab and field applications. shown to be feasible for the field deployment.
The entire procedure can be potentially automatized and applied for a non-intrusive high-throughput analysis of the BS structure of PV modules in the field by using only NIRA measurements.