Quantitative Assessment of the Influence of Camera and Parameter Choice for Outdoor Electroluminescence Investigations of Silicon Photovoltaic Panels

2.1 Cameras

The cameras used for image acquisition were two with a Si (Si_A and Si_B) and two with an InGaAs (In_C and In_D) detector. Their main parameters are shown in Table 1. Because of the low EQE of the silicon detectors between 1000 to 1300 nm, significantly higher integration times are typically necessary. In contrast, the InGaAs detectors exhibit a high EQE in the near-infrared NIR region between 900 to 1600 nm for the In_C camera or even an extended InGaAs detector from 400 to 1600 nm for the In_D camera. For both detector types, the quantum efficiency is shown in Figure 1 in comparison with the EL spectra of an SSC.

Figure 1:

Quantum efficiencies of a typical Si and InGaAs detector in comparison with an EL spectra of a SSC [12].

2.2 Indoor and Outdoor EL Image Setups

Indoor EL images were taken inside a dark chamber with four different cameras, each of them with different integration times. The PV panels were forward fed with 90 %, 75 %, 50 %, 10 %, and 0 % of their short circuit current (I_SC) at standard testing conditions (STCs). The integration times used differ for each camera and are mentioned separately in Section 3.2. PV panels were fixed in a vertical rack. Reference cell and black spot were set next to it. Cameras were mounted in an automated xyz stage focusing the PV panels perpendicularly. Power supplies and cameras are controlled from outside the dark chamber.

Outdoor EL images were taken at nighttime to avoid high noise signals from the sun. The PV panels were fixed to a 40° tilted rack and forward fed with 90 % of their respective I_SC. Cameras were attached to a tripod and focusing the PV panel making an effective focus angle of 60° between camera and PV panel.

2.3 EL Signal Angle Dependency

Additionally, in this study, the influence of the relative image acquisition angle on the images was studied. In Figure 2, the influence on the EL signal and resolution is shown. With the angle of tilt at 90°, the module is perpendicular to the camera view direction. The resolution follows the trigonometric functions and is sufficient for angles >50°. On the other hand, the mean of the EL intensity signal dramatically drops for angles lower than 40°, mainly due to internal reflections in the glass layer of the module. Figure 3 shows the SSC module at a tilted angle of 40° where the perspective distortion is still acceptable for image evaluation.

Figure 2:

EL signal and resolution dependence on the tilted angle being defined as the angle between the SSC and camera view direction with the trigonometric sine function.

Figure 3:

EL image at a tilted angle of 40° taken with the Si_A camera with an integration time of 40 s.

2.4 Reference Cell Characterisation

The goal to position a reference area in each EL image was realised with a high-power back-contacted monocrystalline 5” Si PV cell, with a peak power of 3.2 Wp. Additionally, the EL signal was homogenised with a plexiglas layer on top of the front surface. In Figure 4, the colourised EL image of the reference cell is shown. The mean value within the dashed rectangle was 3128 counts with a standard deviation of 188 counts under 6.4 A forward current. For further investigations, this dashed rectangle is called reference spot (RS).

Figure 4:

Homogenised EL reference cell image acquired with the In_C camera, with 6.4 A forward current and 20 ms integration time. The dashed rectangle indicates the area of interest.

Furthermore, the temperature was logged at the backside of the RS to determine the dependence of the EL signal to the cell temperature. The temperature was measured with an Almemo 2590 data logger, and the EL signal was averaged from the mean of the RS acquired with the In_C camera for 20 ms every 10 s. In Figure 5, the behaviour of the temperature influence on the EL signal for indoor and outdoor is shown. The rise of the EL signal is caused by a change of the non-radiant and radiant recombination probability and its temperature dependency for the indirect semiconductor material [13]. It clearly reveals that a minimum of 30 min warm-up thermalisation time is needed for precise and reproducible measurements.

Figure 5:

EL signal dependence from temperature change for (a) outdoor conditions at around 5 °C and (b) indoor conditions at around 25 °C. The EL signal is measured with the luminescence intensity collected from the camera In_C for 20 ms.

2.5 Normalisation Process

The qualitative and quantitative comparison among the EL images is achievable by analysing them in the same intensity range and without external and internal influences. The recorded EL intensity (I_EL) detected by the cameras is shown with constant gain and ISO value in (1), simplified from [14], [15], [16]:

where I_S is the pure signal intensity and the other intensities are additive noise intensities for background/stray light I_BG,N, and dark counts of the detectors I_DC,N depending on the integration time T_int and temperature T and the readout noise I_RO,N depending on the camera. The background/stray light includes all types of radiation from the same direction as the SSC except from the SSC itself.

The signal intensity I_S depends on the variables in (2),

I_C is the SSC powering current, Ω_cam,SC and d_cam,SC descript the angle and the distance between the camera and the SSC plane, A is the aperture number, and f_K is a correction factor for material properties and behaviour of the SSCs.

For the quantitative comparison, all noise, operator-related, and geometric influences have to be eliminated. Therefore, a need exists to recalculate the luminescence intensities recorded with the different detector technologies. In order to achieve this recalculation, two spots in the SSC plane for each EL image are necessary. The black spot (BS), a metal plate covered with black plastic sheet, provides a mean intensity value IBS¯ for subtraction of large area background noises. For the second one IRS,BS¯, the RS is used with its mean intensity value after the subtraction of the IBS¯.

Finally, the normalisation process is conducted according to the following equation:

where the intensity values are normalised, between the values 0 and 2000 of the newly defined unit luminescence units (L.U.). Those values were chosen based on experience indicating that a depth of 2000 is sufficient for analysis of typical mono- or multicrystalline SSC. In Figure 6, the setup for the reference spots and PV modules is shown.

Figure 6:

Setup of the reference spot (RS) marked blue, the black spot (BS) marked green, and the PV module marked yellow for indoor EL images.

After the normalisation process, the influences to the external EL intensity are eliminated such that the normalised EL intensity is equal to signal intensity with the mean value m between 0 and 2000 L.U.:

2.6 Slope Comparison and SNR

The normalised images are analysed on the one hand by determining the slope of the normalised cumulative histogram and on the other hand by the signal to noise ratio (SNR) of the intensity values. For the slope analysis, the slope between the median and the 25 %-percentile provides a numeric value for image comparison. Furthermore, the SNR determines the ratio between useful EL signal and the noise in an image, see (5):

where m_signal is the mean value of a determined signal area and the m_noise and σ_noise are the mean value and the standard deviation of the BS area [15].

2.7 Comparability of Histograms

Different camera systems or their different onboard image processing algorithms lack for quantitative image comparability for defect recognition.

Figure 7 gives a good example for the comparability problem. Camera Si_B and In_D show a high gain and ISO values. Those values could either be controlled or semi-automated. The In_C camera has a fixed gain, but due to manual non-uniformity correction, which was conducted before using, value comparability between different measurements is difficult. For the Si_A and In_C cameras, very steep slope values, which correspond to a lower intensity distribution range of the collected data, are observable. Thus, the EL signal intensity levels are harder to separate. The integration times of each camera were chosen to be as long as possible for the highest quality images, just limiting by saturation at the RS and the maximum possible integration time. Other integration times are analysed in Section 3.2.

Figure 7:

Comparison of the normalised cumulative histograms for typical indoor images of the four different cameras without normalisation with the corresponding integration times.

3.1 Normalisation Process

The comparability of all cameras can be increased by externally levelling the camera system with reference spots: One bright RS and one dark reference spot for postprocessing normalisation.

The normalisation process shifts the EL intensity captured from the cameras to a range of 0 to 2000 L.U. It does not modify the EL signal behaviour. In Figure 8, the normalised cumulative histograms after the normalisation process for all the cameras are shown. Compared with Figure 7, the histograms show similar behaviour and are in the same luminescence intensity range, except for the Si_B camera. The explanation can be given due to the low linearity of the detector in the NIR region, which was not produced for low-light NIR imaging and therefore not optimised in this region.

Figure 8:

Comparison of the normalised cumulative histograms for the SSC module area in typical indoor images of the four different cameras after normalisation with the corresponding integration times.

3.2 Comparability

As an example, the quantitative comparability of the normalised cumulative histograms for a single camera is shown for a silicon detector Si_A in Figure 9 and for an InGaAs detector In_D in Figure 10. For each camera, EL images with different integration times T_int and different currents were taken. All cameras showed similar behaviour. For higher integration times, the normalised cumulative histograms are nearly identical. But at low integration times of 10 s for Si_A and 20 ms for In_D, the slope is reduced, and the intensity level shifts to higher values. The reasons for those changes might be due either to camera internal increases of the readout noise influence for smaller integration times or to nonlinear image on-board processing like automated gain controls.

Figure 9:

Normalised cumulative histograms for camera Si_A with different integration times T_int and currents I in percent of the module’s I_SC.

Figure 10:

Normalised cumulative histogramms for camera In_D with different integration times T_int and currents I in percent of the module’s I_SC.

This clearly indicates the limit of the normalisation process for not properly chosen integration times.

Additionally, Figure 11 shows the slope values and the SNR of all cameras. High quality images will share a high SNR value and a low slope value. Higher noise levels for low integration times reduce the SNR levels for all cameras but with different factors. In our experience, SNR values higher than two to three are sufficient, which is a bit lower than the Rose criteria [17].

Figure 11:

Quality analysis of the different cameras for the SNR and slope value of the EL cumulative histograms for different currents. The dashed lines indicate linear detector behaviour. The currents of 90 %, 75 %, 50 %, and 10 % of the I_SC starting with the lowest slope value and highest SNR on the left. In_D camera produces EL images of the highest quality, whereas camera Si_B shows the images with the lowest quality.

As already shown in Figures 9 and 10 with sufficiently high integration times and currents, the image quality and reproducibility of the normalisation procedure is high and robust. Nonetheless, the dashed linear fitting lines in Figure 11 clearly indicate that camera In_D is the most suitable for PV EL measurements, because the highest SNR and slope values are achieved. Cameras Si_A and In_C show good images for laboratory cameras. As expected, the rebuild camera Si_B shows the lowest performance.

3.3 Outdoor EL Imaging

The main motivation to improve the quantitative comparison is the need to compare EL images taken in different environments. In this Section, we prove the robustness of the normalisation method for comparisons of indoor and outdoor EL images.

In Figure 12, both images are shown for camera In_C. Due to the tilt of the outdoor modules, the perspective changed for outdoor EL images. Figure 13 shows the normalised cumulative histograms of all the cameras with the slope values. The SNR in outdoor images suffers of a decrement of 60 % to 40 % in respect to the SNR in indoor images, stemming from higher noise levels in outdoor conditions. The normalisation does not eliminate the noise but reduces its intensity levels.

Figure 12:

Comparison between indoor (a) and outdoor (b) images. Images were taken with In_C camera for 20 ms integration time.

Figure 13:

EL image comparison between 60° angle indoor (a) and 60° angle outdoor (b) measurements. The images were taken with camera Si_A with 40 s integration time.

In order to perform a more accurate comparison between indoor and outdoor images, the acquisition angle was taken into count. Therefore, two images one from indoor and one from outdoor conditions, with the same tilt angle, forward current, and camera parameters, were compared (see Fig. 13).

Both EL images in Figure 13 are quite similar, as proven by the cumulative histograms in Figure 14, which indicates the robustness of our method even at different perspectives. The small difference at 50 to 250 L.U. might be the influence of the stray light, which increases the intensity level for outdoor measurements.

Figure 14:

Comparison of Si_A camera’s cumulative histograms for indoor and outdoor EL images under 60° angle. For the outdoor measurements, the module was shifted 40° and the camera was shifted 20°.

Further comparison is done in Table 2. Firstly, the slope value of 4.42 ⋅ 10³ from the indoor image is close to 4.06 ⋅ 10³ from the outdoor image. The outdoor image contains more counts with less EL intensity, also proven by the 228 L.U. in comparison with the 249 L.U. in EL mean value from the indoor image. This can be a factor for a variation in the different image resolutions, because of the different perspective and distance for indoor and outdoor acquisition.

Moreover, the difference in temperature between indoor (∼25 °C) and outdoor (∼5 °C) conditions after 30 min warm-up time still might cause a decrement in the EL intensity (see Fig. 5).

Consequently, the higher noise level in the images might cause an increase of EL intensity in the image, and it leads to a decrement in the SNR. On the other hand, in order to be able to compare and analyse the information, the use of the reference cell to perform the experiment and obtain the data was critical.

3.4 Black Spot Subtraction

One major interest in the investigations is whether BS subtraction already enables quantitative comparison of EL images obtained from different sources or if background (BG) subtraction is necessary. For movie recording, the BS subtraction method might increase the throughput. In Figure 15, both EL images are shown.

Figure 15:

Comparison of the outdoor EL images for camera Si_A with the different subtractions. BG subtraction is shown in (a), and BS subtraction is shown in (b).

The background is clearly visible without background subtraction, but the EL signal has the same intensity as it can be seen in Figure 16. Thus, it is sufficient to acquire just one EL image and increase the measurement throughput [15].

Figure 16:

Comparison of the outdoor EL images with cumulative histograms for the different subtractions, BG and BS.

3.5 Quantitative Defect Analysation

The main finding in our work is that quantitative defect analysis is possible in the same intensity range for all four different cameras. The line scans in Figure 17 indicate that the EL intensity for all cameras is in the same range after the normalisation procedure, which enables quantitative analysis. The higher resolution for the silicon cameras gives more detailed information, especially visible at the bus bar positions with deeper drops at 0.26 and 0.77. Furthermore, the highest resolution of camera Si_A results in a steeper rise at the right edge at first sight of the inactive area at position 0.65 as compared with camera In_C.

Figure 17:

Line scans of the same solar cell with large inactive area in the center between the two busbars, normalised over the cell length for all cameras after the normalisation steps.

In Figure 18, the line scans over a single crack are shown. The negative deviation from the trend lines indicate the easily detectable crack at position zero. The deviation between crack and trend line can easily be detected by automated defect detection methods. Together with the same intensity range, automated analysis methods can determine energy output and module degradation change for iterated EL measurements [18], [19].

Figure 18:

Line scans with trend lines over the same crack centered in the zero point for all cameras after the normalisation steps.