Long‐term outdoor study of an organic photovoltaic module for building integration

Organic photovoltaics (OPV) has attracted tremendous attention as a promising alternative to silicon wafer‐based technologies for building integration. While significant progress has been achieved on the power conversion efficiency of OPV technologies, their field stability is rarely studied. This work investigates the field performance and reliability of a large‐area OPV module designed for building integration in tropical Singapore for 4.5 years. The device suffered more than 14% degradation in power at the standard testing conditions from the initial performance, largely due to losses in fill factor (−12% relative). During the monitoring period, it exhibited comparable performance to more conventional silicon PV technologies, with an average specific energy yield of about 4 kWh/kWp/day and an average performance ratio of 0.96. Excellent performance at low light conditions was also observed. However, its field performance was heavily impacted by soiling, which typically led to a 5 to 10% loss in the current output after several months. Further, the device's outdoor performance also showed a three‐stage degradation process, including (1) an initial slow degradation in the first 2 years (about −1%/year), (2) a stable period with negligible performance loss from Years 2 to 3.5, and (3) a rapid degradation in the last year (about −5%/year).


| INTRODUCTION
Photovoltaics (PV) plays an important role in today's energy transition to decarbonize the power sector.The current PV market is dominated by crystalline silicon (c-Si) wafer-based PV technologies with about 95% market share in 2020 [1].However, several other PV technologies have emerged as promising alternatives in specific market segments.Among them, organic photovoltaics (OPV) has the potential to revolutionize building-integrated photovoltaics (BIPV) owing to its unique advantages of low production cost, mechanical flexibility, easy fabrication, lightweight, semi-transparency, and esthetics.These properties are highly desirable for BIPV applications that are difficult to achieve from c-Si wafer-based technologies.
In recent years, tremendous progress has been made in improving the power conversion efficiency (PCE) of OPV [2][3][4][5], surpassing the 20% threshold at the cell level.The PCE of large-area OPV modules (>200 cm 2 ) has also been boosted over the 10% milestone with the current world record efficiency at 11.7% [6].While PCE is a main driver for commercialization, it is not the sole requirement.To be commercially viable, OPV devices also need to be stable and durable under real-world operating conditions.
It is well known that both intrinsic (e.g., morphology changes) and extrinsic factors (e.g., light, water and oxygen) could cause significant performance degradation to OPV devices [7][8][9].However, the level of research effort in OPV device stability is noticeably less than PCE enhancement [10].In particular, studies on long-term field stability of OPV devices are rare, since such studies may take several years to produce meaningful and conclusive results.Based on a recent review, the longest outdoor study of OPV technology is only slightly over 2 years [7,11].Comparing to indoor accelerated stress testing that typically targets a single degradation mode, outdoor reliability studies are more representative and informative, as the interactions of different stress factors are considered.Therefore, more research is urgently needed to fill this crucial gap.Furthermore, PCEs of PV devices are measured under predescribed steady-state conditions and may not translate equivalently into actual performance outdoors.Hence, it is also important to understand field performance of OPV modules, which benefits the future developments of this technology.This paper aims to address these two issues.The field performance and reliability of a large-area commercial OPV module designed for BIPV application are investigated in Singapore from December 2016 to May 2021.The hot and humid climatic conditions (plus high ultraviolet exposure) throughout the year in Singapore present a great reliability challenge for the OPV module (a useful list of abbreviations and nomenclature is supplied as supporting information).

| EXPERIMENTAL DETAILS
The all-small-molecule tandem OPV module with dimensions of 1.60 m in length and 0.32 m in width was commercially obtained (but the exact material composition is not known to the authors).This flexible device using polyethylene terephthalate (PET) foils was then pasted onto a 12-mm glass (2.04 m by 0.36 m) to simulate BIPV applications.It was installed (rack-mounted) on the rooftop of a building at the National University of Singapore (NUS) campus (1.3 N, 103.8 E) in December 2016.The South-East Asia city-state has a tropical rainforest climate, with Köppen climate classification Af.
Facing south, the module was tilted at 10 to allow self-cleaning.
Its outdoor performance was monitored with a high-precision maximum power point (MPP) tracker (ISET-mpp meter V4.10 from Papendorf Software Engineering GmbH).The I-V characteristics were measured at 10-s intervals, and the module operated at its MPPs between the I-V sweeps.The in-plane irradiance was measured with a silicon sensor (ISET sensor from IKS-Photovoltaik GmbH), and module operating temperature was measured on the front surface (no shading caused) with a PT1000 temperature sensor.Both the MPP tracker and irradiance sensor were calibrated every 2 years, as per manufacturer's requirement.The irradiance sensor was cleaned weekly, whereas the module was intermittently cleaned.Weekly checking of the test setup was also conducted to ensure a smooth operation and minimize downtime (if any).the module were measured with the same solar simulator.The power matrix of the device after the long-term outdoor exposure was also measured.For the power matrix measurement, the module temperature was increased from 25 C to 60 C with a step size of 5 C at seven different irradiance levels (i.e., 100, 200, 400, 600, 800, 1,000, and 1,100 W/m 2 ).
Table 1 lists the detailed specifications of the measurement equipment used in this work and their uncertainties (more details on uncertainties can be found in Section 3.3).(relative) and 2.5%, respectively, whereas the open-circuit voltage (V OC ) remained nearly unchanged.The shape of the I-V curves also suggests an increased series resistance (looking at the slope near V OC ), which is likely the culprit for FF loss.In addition, about 4% soiling loss in power was recovered upon cleaning the module front surface, as shown in Figure 1.The module was soiled for over 3 months in the field before the measurements (uncertainty considerations for these and the following measurements are reported in Section 3.3).

| Indoor characterization
The power matrix of the OPV module was also measured after the outdoor exposure, which enables comparisons with the outdoor measurements.Figure 2a-c

| Soiling analysis
Soiling mainly impacts the current and power outputs of PV devices, so either of them can be used to quantify the soiling effects.Herein, current generation is preferred, as it is less sensitive to temperature variations (see Figure 2b).Moreover, the device's I SC was much more stable than P mpp over the monitoring period (see Figure 1), hence using I SC removes most of the effects from device degradation.The field performance of PV devices is affected by many factors such as AOI and spectral effects.As a result, the performance trend is often noisy.Filters are therefore applied to extract a good soiling signal.An irradiance filter (≥600 W/m 2 ) and AOI filter (AOI >45 ) are used to minimize bias from low light conditions (as illustrated in the inset of Figure 4) and AOI effects.AOI is modeled under clear-sky conditions, using PVLIB [14].Data with an irradiance over 1,100 W/m 2 are also removed since these are often due to cloud reflections [15].Moreover, a local filter is applied to data collected in each day; ratio of P mpp to irradiance outside of the 1.5 times of interquartile range (IQR) is considered as an outlier.Global filterers (based on 1.5 times of IQR) on ratio of I SC to irradiance, ratio of P mpp to irradiance, V OC and FF are applied to detect further outliers.
The I SC values at each timestamp are corrected for temperature (to 25 C) and then proportionally translated to the 1,000 W/m 2 condition; the translated I SC values of each day are aggregated into a daily index (i.e., soiling signal), using insolation as the weights (a list of the used formulas is supplied as supporting information).Figure 5a plots these daily indexes (normalized by the median of the first three data points in the soiling signal), where the manual cleaning events are indicated by the gray dash lines (13 in total).As illustrated in Figure 5a, these cleaning events coincide with the peaks in the soiling signal; however, the performance improvement from cleaning varies at times.The rainfall data were also collected from two nearby meteorological stations, and the average values are shown in Figure 5b.It is, though, difficult to conclude that raining alleviates the soiling effects on the OPV device.On the other hand, soiling effects were minimal on nearby c-Si PV modules, which were installed in the same configuration [16].This may suggest that the deposited dust particles on the front PET sheet cannot be easily washed off by rain or the residual dust particles have more impact on the OPV device.The soiling signal was subsequently separated into soiling intervals (between two consecutive cleaning events) for soiling rate estimation, as shown in Figure 6.Soiling rate describes the percentage reduction in current output per day.For each soiling interval, the data were fitted by a piecewise linear function with two segments.The data shown in the Figure 6a-n were normalized so that the fitted lines pass the point [0, 1].In general, the model describes the soiling profile satisfactorily, but it is true that a more complex function may fit the data better for some soiling intervals (as in the case of Figure 6h).The estimated soiling rates for each interval are also provided in the subplots, where the two values in square brackets represent the soiling rates (%/day) for the first and second phases, respectively.In most of soiling intervals, a rapid decrease in the OPV device's current output was observed in the first 1 or 2 months after cleaning, followed by a steady period during which the extent of soiling did not increase substantially.In some cases (e.g., Figures 6c,h), even an upward trend was detected after reaching the minimum.The soiling rates also varied greatly for different intervals with the most severe period at À0.26%/day.In general, if the OPV device is not cleaned, the current output will be reduced by 5% to 10% due to soiling after several months.
To further validate the results, the current output gain from each cleaning is also estimated by comparing the average soiling signal last cleaning was done indoors, so the gain was measured under STC (4.8%I SC improvement).A general pattern is obvious from Figure 7: the larger the current output loss due to soiling is, the greater the improvement is from the subsequent cleaning.The average current output gain from cleaning outdoors is 4.9%, which is in line with the value measured at STC (i.e., cleaning events 14).Several factors may have contributed to the difference in the current output gain and loss, such as measurement uncertainties, device degradation, and modeling inaccuracy.

| Degradation analysis
As shown in Figure 3c, there is a non-linear decreasing trend in the weekly PR.Herein, the degradation rate of the field performance is calculated.To minimize the soiling effects, only data collected within 7 days of cleaning are used.Two methods are deployed for the degradation analysis, evaluating different performance metrics.In the first method, the device's PI values of each 7-day interval are aggregated using insolation as weights into a single value (i.e., PI clean ).Data are pre-filtered before the aggregation, using the same criteria for the soiling analysis.An extra filter on module temperature (between 25 C and 60 C) is applied.PR is used instead of PI in the second method, producing (PR clean ).In addition, the filtering criterion for the lower irradiance threshold is relaxed to 200 W/m 2 to include some low irradiance conditions (a list of the used formulas is supplied as supporting information).
a rapid degradation in the final year.It is postulated that moisture and/or oxygen ingress into the module may have accelerated/ triggered some degradation processes, which highlights the importance of encapsulation for OPV.Table 2 summarizes the change points (CP) of the piecewise function and degradation rates (DR) of different stages.The device's field performance after 4.5 years of outdoor operation dropped by over 7%.If the trend continues, it will cross the 20% threshold in less than 3 years.It is also possible that its field performance will stabilize again, so outdoor monitoring of the device performance will be continued.Nonetheless, a service lifetime of 5 years or more in tropics is achievable with OPV technologies.It is also important to note that the field performance degradation is less significant than the STC degradation; this requires further investigation in future.

| Uncertainty considerations
The OPV module has spectral responsivity (SR) in the visible range from 400 to 800 nm.Spectral mismatch effects on narrow band and multi-junction PV devices can be relevant and can represent the dominant uncertainty contribution in indoor measurements on Class A F I G U R E 7 A direct compassion of the current output loss at the end of the soiling interval and the current output gain after manual cleaning + A + A + solar simulators.Spectral correction is usually performed to mitigate the effects on I SC and P mpp : however, equipment for SR of multi-junction modules was not available at the time the exposure campaign started, and the initial measurement at STC was therefore not spectrally corrected.Post exposure, after SR measurement was performed, the spectral mismatch analysis confirmed that the indoor STC measurements on the A+A+A+ with an unfiltered silicon reference cell may give rise to an additional uncertainty contribution due to spectral mismatch in I SC and P mpp as high as ±1.7%, which can be mitigated down to about ±0.5% when a filtered reference cell is used  (The overall uncertainty in I SC and P mpp with unfiltered silicon raises then to more than ±3%, while it can be mitigated down to less than ±2% if spectral mismatch correction is performed).However, under the reasonable assumption of same lamp spectral irradiance (typical of the region).However, the difference in SR between the OPV testing device and the (unfiltered) silicon sensor causes remarkable spectral mismatch effects on I SC (hence also on P mpp ) measurements, when combined with the difference between the outdoor real spectral irradiance and the standard AM1.5G at which the sensor is calibrated.Spectral mismatch analysis under various irradiance conditions and time of the day showed that the spectral mismatch effects observed caused an overestimation in the testing device I SC ranging in general around À3 to +0%.It can be assumed that all over the outdoor exposure period, this effect may have contributed to the total uncertainty in I SC and P mpp generally in the range of ±5%.However, spectral mismatch effect is generally a cyclic effect, varying with time of the day and weather conditions: the overall effect on the relative P mpp and I SC trend is can therefore in great part be ignored; in particular, the impact of spectral mismatch on I SC and P mpp largely cancels out in FF, which is therefore not affected by spectral mismatch related uncertainties.
In this work, the field performance and reliability of a large-area OPV module designed for building integration are investigated in the tropical Singapore.The device's STC power degraded by more than 14% after 4.5 years of outdoor operation; its FF and I SC dropped by approximately 12% (relative) and 2%, respectively, whereas V OC remained nearly the same.In the field testing during the monitoring period, it exhibited comparable performance to conventional silicon PV technologies with an average specific energy yield of about 4 kWh/kWp/day and an average PR of 0.96.Spectral mismatch effect in outdoor measurements may have caused an overestimation in absolute I SC and P mpp measurements.Excellent performance at low light conditions was also observed, under which the actual power output in the field exceeded the expectations from indoor characterization.However, its field performance was heavily affected by soiling, reducing the current output by up to 0.26%/day.While manual cleaning improved its performance momentarily, the device's current output dropped again quickly in the next 1 or 2 months, followed by a steady period during which the soiling impact did not increase substantially.In general, if the OPV module is not cleaned, the current output will be reduced by 5% to 10% because of soiling after several months.
Furthermore, the device's field performance showed a threestage degradation process, including (1) an initial slow degradation (about À1%/year) in the first 2 years, (2) a stable period with negligible performance loss, and (3) a rapid degradation (about À5%/year) in the last year after the stable period.It is however noticed that the field performance degradation is less severe than the STC degradation, which needs further investigation in future.Nonetheless, a service lifetime of 5 years or more outdoors is achievable from OPV devices, based on both the STC measurements and field performance.
These findings are valuable for the future commercialization and large deployment of OPV technologies.

After 4 .
5 years of operation in the field, the module was taken into laboratory in May 2021 for I-V measurements at the standard testing conditions (STC: 1,000 W/m 2 total irradiance, AM1.5G spectral irradiance and 25 C cell temperature) with a class A + A + A + h.a.l.m cetisPV 100 ms solar simulator.The pre-exposure STC ratings of T A B L E 1 Equipment specification and their measurement uncertainties Equipment Description Model Measurement uncertainty (k = 2, $95% confidence) Traceability h.a.l.m. indoor solar simulator 100 ms pulsed solar simulator, class A + A + A+ cetisPV moduletest3 On multi-j PV modules,

Figure 1
Figure 1 presents the I-V characteristics of the OPV device at STC before and after the long-term outdoor operation.The OPV module survived after 4.5 years of operation in tropics, as it retained roughly 86% of the initial maximum power (P mpp ).Indeed, typical warranties for c-Si PV modules often include an 80% threshold over 20 years or Figures 3a shows the probability distribution of the irradiance (only >50 W/m 2 are considered) received by the OPV module over its 4.5 years of operation, where the median of 336 W/m 2 is highlighted by the dash line.Figure3bshows the corresponding probability distribution of the module operating temperature; the median is slightly lower than 38 C (highlighted by the dash line), and the maximum exceeds 65 C (but it rarely occurs).During the monitoring period, the average daily specific yield is about 4 kWh/kWp/day, and the average performance ratio (PR) is around 0.96; these values are comparable with the conventional c-Si PV technologies[12,13].Furthermore, Figure3cplots the evolution of the weekly PR with time.There were some sharp drops and increases in the weekly PR, which were caused by soiling and manual cleaning.Overall, the module's performance

F
I G U R E 5 (a) Normalized soiling signal (normalized by the median of the first data points in the soiling signal); (b) the average daily rainfall amount from two nearby meteorological stations

3
days before and after cleaning.Meanwhile, the current output loss on the last day of soiling intervals is calculated by the piecewise linear model.A direct comparison of these values is shown in Figure 7.The F I G U R E 6 (a-n) Normalized soiling signal in each soiling interval.The data are normalized so that the fitted lines pass the point [0, 1] for all intervals.The values in each subplot represent the soiling rates (unit: %/day) for their respective phases.

Figure
Figure 8a,b shows the evolution of PI clean and PR clean , respectively.The line in blue represents the smoothed trend using the Locally

F
I G U R E 8 (a) The evolution of normalized PI clean and (b) the evolution of the normalized PR clean .The dash-dot line indicates the time when the module was taken indoors.T A B L E 2 Summary of the change points of the linear piecewise function and the degradation rates (DR) of different stages

(
or negligible variations), and same SR negligible variations) of the testing device before and after exposure, the measurement bias due to spectral mismatch cancels out and the deviation observed in the relative comparison of performance degradation is not affected.Concerns may arise from the assumption of same SR before and after the 4.5-year exposure.Hence, spectral mismatch analysis was performed also both on post-exposure data based on the SR measurements at the time of writing and on pre-exposure data based on SR data provided by the manufacturer.The result showed that the deviation in I SC can indeed be highly uncertain.However, the 12% degradation observed in the FF is not affected by spectral mismatch, and that result is robust, arising from module degradation and confirmed by the field performance study.For what concerns measurement uncertainty in outdoor exposure measurements, the global irradiance was recorded by a silicon sensor (model ISET Sensor by IKS Photovoltaik GmbH), with a measurement uncertainty of ±3%, which represents the dominant uncertainty contribution.Another important uncertainty arises even in outdoor measurements from the spectral mismatch effect (other uncertainty contributions, e.g., data acquisition and temperature sensors and give minor contributions to the overall uncertainty).Using silicon sensors is the most recommended choice in tropical climates, since it has the advantage to be more sensitive to the frequent irradiance fluctuations