Reliability and Degradation of Solar PV Modules—Case Study of a 19-Year Old Polycrystalline Modules in Ghana

Fourteen (14) rack-mounted polycrystalline modules installed on the concrete roof of the solar energy applications laboratory at the Kwame Nkrumah University of Science and Technology (KNUST) in Ghana, a hot humid environment, were assessed after 19 years of continuous outdoor exposure. The physical state of the modules was documented using a visual inspection checklist. They were further assessed by current-voltage (I-V) characterization and thermal imaging. The modules were found to be in good physical state, except some bubbles on front side and minor discolouration/corrosion at edge of the cells. Compared with reference values, the performance decline of the modules observed over the exposure period was: nominal power (Pnom), 21% to 35%; short circuit current (Isc), 5.8% to 11.7%; open circuit voltage (Voc) 3.6% to 5.6% and 11.9% to 25.7% for fill factor (FF). It is hoped that this study will provide some helpful information to project developers, manufacturers and the research community on the long-term performance of PV modules in Ghana.


INTRODUCTION
The phenomenal and sustained growth of solar photovoltaics (PV) in recent years is welldocumented [1] [2] [3] [4] [5] [6] [7] [8] [9]. Indeed, this growth extends to other renewable energy technologies such as wind, bioenergy, hydropower and geothermal. The drivers for this growth are widely acknowledged to include, climate policy, technology improvements and energy security considerations. Research and development (R&D) efforts have paid off in the form of increased efficiencies at both cell and module levels. For example, cell efficiencies reported in major laboratories for crystalline silicon have increased from 13% in the 1970s to 25.6% in 2016 [10], almost doubling in the period, and thus approaching the theoretical thermodynamic limit of 34% (Shockley-Queisser limit) [11] [12] for single crystal silicon cells at standard test conditions. Advances in thin-film and multi-junction cell technologies have been even more rapid. The cellto-module efficiencies for these cell technologies have also improved to 99% in 2015 and is projected to exceed 103% by 2026 because of improvement in light management techniques [13].
As research laboratories and industry players focus on improving solar-to-electric conversion efficiencies, there is increasing acknowledgement of the need for data on the performance of systems that have already been deployed across the various climates and geographical regions of the world. A characteristic feature of solar PV (and other renewables) is the high upfront cost per installed power. Once installed, reliable performance and durability of the system enables it to generate electricity (kWh) which represents benefit to the investor/system owner. Reliability is understood as the probability that an item, in this case the solar PV module, will continue to perform its intended function without failure for a specified period of time under stated conditions [14] [15]. Durability, on the other hand, has to do with how long the product will last under normal operating conditions [16].
For the solar PV investor, reliable operation of the PV modules and their durability as per warranty conditions, are important for the realization of the expected return on investment. Module manufacturers, similar to other product manufacturers, attempt to assure buyers of their products by providing warranties. Warranty on PV modules have evolved from 5 years in the 1980s, through 10 -20 years in the 1990s to current industry standard practice of 80% of peak power for 25 years [17]. Some manufacturers provide 2-stage (and other multi-stage) warranties. A warranty typically in the range of 90-95% peak power in the first 5-10 years and thereafter, 80% -87% of peak power up to year 25 [18]. REC Solar [19], for example, provides a 97% performance guarantee on its modules within the first year of exposure and 80.2% of peak power by the end of the 25th year.
For current standard 25 year warranty (80% of peak power) to hold, modules must, on the average, degrade at not more than 0.8%/year ( Figure 1). However, industry players are seeking to extend warranty periods to 30 years [20], which would imply a maximum average annual performance decline of no more than 0.65% (Figure 1). Solar PV warranties typically encounter two major hurdles. First, is the enforceability of the warranties [21]. This is largely attributable to the rapidly evolving landscape that has led to the extinction of many manufacturers, with others filing for bankruptcy. This extinction was most noticeable in the period 2010 -2013, when, in the US alone, over 50 companies either collapsed or filed for bankruptcy protection [22]. Warranties provided by defunct companies on PV modules sold therefore become difficult to enforce, particularly if the need for warranty claims arise after official liquidation. To address this first problem with the warranties, some manufacturers are now   [26]. These insurance products are, however, certain to add to the cost of modules to the buyer.
The second difficulty with the warranties, is the more fundamental question of physical basis. Current testing and qualification regimes such as the IEC 61215 [27] for crystalline Silicon modules and IEC 61646 [28] for thin film modules employ techniques such as accelerated ageing and stress tests with the view to detecting the presence of known failure or degradation modes in the intended environments [29]. These qualifications tests can however not predict or guarantee product life under field conditions. They are indeed not designed for such a purpose [29] [30].
Qualification tests rather seeks, among others, to find design and process flaws and have been credited with significantly reducing infant mortality among PV modules [31]. It has been suggested [17], that, recent studies based on field exposure is providing some basis for and validation of 25-year warranties; however, many an overwhelming majority of modules have hardly lasted 25 years on the field to prove the validity of the warranties. Figure 2 shows that, over 80% of modules installed today have been installed in the last 5 to 6 years. To compound this problem further, the technologies themselves are undergoing significant material level changes. For example, [32] [33] [34] reported that crystalline Silicon (c-Si) wafer thickness has reduced from 400 µm in 1990 to 180 µm in 2015, while silicon usage has declined from 16g/Wp in 2004 to less than 6g/Wp in 2015. The metallization is also moving away from silver to less expensive options such as copper, nickel and zinc [33].  Precipitation in tropical savanna climate is less than 60 mm in the driest month and also less than h [45]. Further microclimate sub-classifications of the country have also been made based on agro-ecological characteristics of various zones and is used by organizations such as the UN Food and Agriculture Organization (FAO) [46]. These sub-classifications are: Rain Forest, Deciduous Forest, Forest-Savannah Transition, Coastal Savannah and Northern (Interior) Savannah which comprises Guinea and Sudan Savannahs [46].
It should be noted, however, that, whereas the Koppen-Geiger classification scheme is based on temperature and precipitation, for solar PV performance and durability issues additional parameters such as humidity, temperature variation (thermal cycling), altitude and air salinity are important as well [42]. For consistency and comparability of analysis, consolidation of climate categories sometimes becomes necessary. In the work of Jordan et al [43], tropical climates such as Aw and Af are broadly classified together with Cfa (Temperate hot summer without dry season) and consolidated as hot and humid. This present work adopts this climate categorization. Another climate categorization of interest is that which was used for the All-India Survey of Photovoltaic Module Degradation: 2013 [47].
This paper seeks to contribute to filling this gap (long-term performance of PV modules and degradation in Africa) and presents results of assessment conducted on an 19-year old polycrystalline solar PV installation at the Kwame Nkrumah University of Science and Technology (KNUST) in Ghana.

Site description and climate
As shown in Figure 3, Kumasi is a hot and humid climate with average daily temperature ranging from 24.4 o C in July to 27.8 o C in March and relative humidity of 65% in January to 83.5% in July [48]. Hence, accordingly, Kumasi is climate condition can be classified as Aw using Koppen classification. This climate would be categorized as "warm and humid" per classifications used in ref [47].

Description of installation
The system under study was installed in 1998 and comprises of fourteen (   The modules were qualified under the CEC ESTI 1 503 Standard. The cell design comprises two bus bars and 40 grid lines, organized in a rectangular pattern around four (4) centres on the two bus bars ( Figure 5). In addition to the rectangular grid patterns, four (4) grid lines are also observed to run across the cells in both horizontal and vertical directions ( Figure 5).  During the measurements, the modules were electrically isolated to permit access to the terminals of the modules. Even though there was no cleaning schedule in place for the modules, the rains had done some cleaning of the module surface during the period of measurement. Nonetheless, to further eliminate impact of dust on the measurement, the module surfaces were cleaned with clean water and allowed to dry before measurements were made.

Inspection and measurements
Field assessment of solar PV modules usually employ techniques such as visual and physical inspection, I-V (current-voltage) characterization, bypass diode test, insulation testing and thermal imaging to help detect defects in the modules [40]. Some defects, such as micro-cracks in cell structure may not be visible to the eye and require more advanced techniques such as electroluminescence imaging to detect. However, the I-V curve tracing is the most widely used technique in outdoor electrical characterization of solar PV modules [50].
In this study, three methods, which are visual inspection, I-V characterisation and thermal imaging were used. The visual inspection of the modules was undertaken using the widely-applied field visual inspection template by NREL/IEA [31]. This sought to systematically document any visually observable defects on the various components of the modules (such as front glass, back sheets, junction boxes, metallization, encapsulation and frames) and interconnections following  Impp, Vmpp, Isc, and Voc. It must be noted, that Pmax, which is the product of Isc and Voc is never generated, as the I-V curve is never rectangular. The peak power may therefore be viewed as a fraction of the maximum power. This fraction is the fill factor.
A Fluke R TI400 thermal camera was used to obtain thermal images of the modules in forward bias mode at Isc. The emissivity of the camera was set to 0.85 since the module surface is glass. This was to help assess temperature inhomogeneity and possible hotspots in the modules.

Calculations and data analysis
Where: I is current (A); Isc is short-circuit current (A); G is irradiance (W/m 2 ); T is module is dimensionless temperature coefficient of Voc (default value is -0.004) a is dimensionless irradiance correction factor (default value is 0.06) and Rs is series resistance (Ω) (default value is 0) Although a number numeric and algebraic approaches exist (e.g. IEC [53], Smith et al [54], Tsuno et al [55] and Marion [56]) for translating I-V data from measured to desired reference conditions (such as STC), many of these require the determination of coefficients, which in turn require controlled conditions such as constant irradiance and constant temperature; conditions which are difficult to obtain under field conditions, making the equations difficult to implement. The approach proposed by JRC therefore presents a balance between accuracy and practicability and has uncertainty of about 4%.
Standard manufacturer reference electrical parameters are normally reported at STC conditions. However, outdoor conditions differ, and sometimes the irradiance is well below this reference. Work by Anderson [57], has shown that translating to STC from low irradiance values and high temperature comes with lower levels of accuracy. In this present study, effort was, therefore, made to obtain I-V measurements at irradiance levels that were as close as possible to 1000 W/m 2 . The standardization could in principle also be undertaken to other reference conditions, such as to the nearest irradiance level for which reference I-V data is available.

Visual and physical inspection.
The visual inspection documented the visually observable condition of the modules. The summary is presented in Table 3. Accompanying pictures are shown in Figures 8 to 11. Generally, no major visually observable defects were seen on the front glass, backsheets, wires/connectors, the junction box, frame and metallization. Some minor corrosion/discolouration at the edge of some cells was also observed (Figure 9).

Component Observation(s)
Front Glass Front glass was smooth and dusty prior to cleaning with water; no damage or cracks were observed. Minor occurrence of bubbles were observable on the front side of the module (Figure 8).

Backsheet
Minor discoloration observable from front glass (Figure 8) -this might be due to moisture ingress. Generally was like new. No wavy texture was observed, no chalking, burn marks or other damage was visible.

Wires/Connectors no embrittlement or burns was observed
Junction Box Intact and well-attached, no loss of adhession was observed, Figure 9 (opened by authors during study).

Frame
Minor discoloration was observed; no corrosion was found; frame adhesive was not visible and showed no signs of degradation; the bottom section of the frame was soiled and had accumulated over the years.

Performance Measurements (I-V curve)
Module temperatures during the measurements period ranged from 47.

Peak Power
Overall on module-by-module, the power output had declined from 49.5 W to 32.2 W -39 W (with a median of 37.6 W) over the 19-year period of exposure ( Figure 12). This represents a decline of 21.1% -35.0%, with a median of 24%. On an annual basis, the median degradation rate is 1.3%. The advantage of using the median instead of the mean, is that, it minimizes the impact of outliers. Modules of this era (the 1990s) came with 10-year warranties; within which period 80% of the peak power was guaranteed [17].
This implied a maximum average annual degradation of 2% (see Figure 1).
Assuming a linear degradation rate for the modules, as suggested by [59], all the modules could be said to have met and exceeded warranty provisions. In addition, modules of the time came with tolerance of +/-10% [17]; which implied that the peak power less 10%, (in this case 44.5 W) would still be within the purchase agreement. If this lower limit of the tolerance were used, 80% of peak power will translate to 35.6 W, in which case over 85% of the modules will still be operating above the minimum guaranteed output after 19 years of exposure. While some authors have defined module failure to mean an effect that degrades the module power, which is not reversed by normal operation or creates a safety issue [31], module life on the other hand, as noted by [29], does not lend itself to easy definition, as they could mean different things depending on the perspective. From a manufacturer's perspective, module life may be viewed in terms of financial liability period, whereas the user is not necessarily bound by this and may keep the module in service as long as it is safe, and meets or contributes to meeting their needs. peak power is a function of short-circuit current, open-circuit voltage and fill factor; changes in power output must be explainable in terms of these parameters.

Short-circuit current
As shown in Figure 13, the short-circuit current of the modules declined from 3.1 A to between 2.74 A and 2.92 A, with a median value of 2.86 A. These represent, respectively declines of 5.8% to 11.7% and median of 7.9% over the 19-year period. On an annual basis, the median decline in the short-circuit current is 0.4%.
Of particular note is module PWX6 (Figure 13), which showed a much sharper decline in short circuit current (0.6%/year). Bubble formation and delamination is suspected to be the cause of this observation.
Overall however, coefficient of variation was 2%. The short circuit current has a strong dependence on irradiance and hence a decline in short-circuit current could be due to optical transmission problems caused by factors such as module soiling, browning/yellowing of encapsulant and delamination which causes optical decoupling (mismatch of refractive indices). Figure 13: short-circuit current of modules at STC

Open-circuit voltage
The open-circuit voltage (V oc ) of the modules (Figure 14  Short-circuit current, A

Thermal Imaging
Thermal imaging of the modules, forward-biased at I sc , generally did not reveal much inhomogeneity in module surface temperature. Temperature distribution on the module surface showed difference of less than 5 o C for most modules, the exception being module PWX 1 which had cell temperature difference of about 10 o C at some sections of the module (Figure 17). The general uniformity of temperature distribution and non-occurrence of hot-spots agrees with the point made earlier on the uniform ageing of the modules based on the co-efficient of variation of peak power and other key performance parameters.

Accounting for power loss
As noted in Section 3.2.1, the peak power of the module is a function of short-circuit current, open-circuit voltage and fill factor. A plot of annual loss in peak power versus these parameters (Isc, Voc and FF) indicate that they are positively correlated ( Figure 18). A regression analysis indicates that Isc and FF, together explain 97.5% of the variability in peak power (adjusted R-squared), Table 3. The loss of power is therefore attributable to losses in shortcircuit current and fill factor, and to a lesser extent, on the open circuit voltage.

General Discussion
To put the current results into perspective, it is helpful to compare with similar studies. Jordan and Kurtz [17], conducting an analytical review of alomost 2000 reported degradation rates found in the literature, determined a median value of 0.61% degradation for multicrystalline Silicon modules installed prior to the year 2000, a total of 409 installations. In a more recent and updated work by an expanded team, Jordan et al [59] based on literature review, found that a median degradation rate of 0.6% (based on 683 observations) was reported for crystalline silicon modules installed in hot humid climates. Even though the median degradation (peak power) of the modules in this present study is higher (1.3%/year) than the median of reported in these reviews, it is consistent with rates reported of installations in the period 1990 -2000, which were about 1%/year [59]. It should be noted also, that, the data used in the reviews do not have balanced geographical representation as acknowledged by the authors of the review. This geographical imbalance underscores the need for continuing effort to study systems that are installed in regions for which data is under-reported or not reported at all.
The dominance of short-circuit current and fill factor in driving the power loss of the modules is consistent with previous studies which have reported on this, including: Ndiaye et al [37] , Bandou et al [36] and Jordan and Kurtz [17]. Bandou et al [36] studied a 28-year old systen installed in the desert of Algeria and reported annual degradation of 1.22% in peak power, arising mainly from 0.78% in short circuit current and 0.56% in fill factor. Even though the study by Ndiaye et al [37] had a rather short observation period (1.3 -4 years), the results also confirm the dominance of short-circuit current and fill factor in power degradation.

CONCLUDING REMARKS
This study has examined the performance of fourteen (14)  and thermal imaging were used to assess the modules. This paper concludes as follows: • The physical condition of the modules was good, with no major visually observable defects on the front glass, back sheets, wires/connectors, junction box, frame and metallization.
Some bubble on front side and minor discolouration at the edge of some cells were, however, observed upon magnification of images acquired.
• The module had a warranty of 10 years, with a tolerance of +/-10%. At the median annual power degradation rate of 1.3%, the module has met and exceeded warranty expectations, even if the lower end of the tolerance was ignored.
• The reduction in nominal power is dominated by reduction in fill factor and short circuit current.
To contribute to filling the data gap on long-term PV module performance degradation in Ghana, future work will study modules installed in various micro-climatic zones of Ghana.