Fig.5 shows a DPL image of a large section of the solar farm, containing
approximately 1,200 PV modules in one single image. Only modules located
in the bottom half of this image are connected to the inverter that was
switched, i.e. only half of the modules yield a PL image, the other half
appearing dark in the difference image, as expected. Further, the image
includes a grassy area to the right of the PV array, which also appears
dark. The image was calculated from 20 individual near-infrared camera
images. We observe an average PL signal count in the section containing
the switched modules of 290 counts, whereas the count rate in both the
grassy area and the non-switched section is below 10 counts. Four
modules in the image stand out in red. They exhibit considerably lower
PL counts compared to all other modules in that section of the solar
farm. The PL intensity from these modules is approximately 53% compared
to the
surrounding
modules.
Quantitative analysis of luminescence intensities from solar cells is
based on the Generalised Planck Equation , predicting an exponential
relationship between the emitted luminescence intensity and the diode
voltage of the device. In simplified form that relationship can be
written as
\(I_{\text{PL}}=Ce^{\left(\frac{\text{qV}}{\text{kT}}\right)\ }\)(1)
where IPL is the emitted luminescence intensity,k is Boltzmann’s constant, q is the elementary charge,T is the sample temperature and V is the diode voltage.
The so-called thermal voltage kT/q has a value of about 25.85 mV at room
temperature. The calibration constant C in Eq.1 depends on
various sample specific geometric and optical sample parameters, on the
spectral sensitivity of the detection system and on various measurement
parameters. The accuracy of obtaining implied voltages from luminescence
signals is therefore linked to the ability to accurately determine the
constant C . By contrast, it follows from Eq.1 that the ratio of
two luminescence signals, IPL,1 andIPL,2 respectively, can be converted into a
voltage difference ΔV via
\(V=\frac{\text{kT}}{q}\ln\frac{I_{PL,1}}{I_{PL,2}}\). (2)
provided that the calibration constant C can be assumed to be the
same for the two luminescence measurements. That is the case if
measurements are conducted with the same measurement system (as is the
case here) and with identical measurement parameters (as is also the
case here) and either on the same sample at different times or under
different bias conditions or on different samples which can be assumed
to have similar optical properties.
Eq. 2 predicts 16.5 mV lower voltage per cell for the four modules that
appear in Fig.5 with 53% lower PL counts. During our experiments we
noted some variation in optical properties between modules in this solar
farm in the near infrared spectral range that is defined by the bandpass
filter mounted on the camera lens. A detailed quantitative analysis of
DPL results obtained from this solar farm, including the expected minor
impact of these optical variations is currently underway and will be
reported elsewhere, noting however that the focus of this publication is
the demonstration of the inverter-based DPL image acquisition principle.
We would like to emphasise, that the quantitative analysis of DPL image
variations between cells or modules of the same type in terms of voltage
variations according to Eq.2 is well established. For validation
purposes we performed separate experiments on modules at a UNSW outdoor
test system. These measurements confirmed that the accuracy of implied
voltage differences between modules from PL images is on the order of
+/- 2 mV per cell when PL image data is compared to measured terminal
voltages under well controlled experimental conditions.
The DPL image example in Fig.5 demonstrates, for the first time, the
ability to simultaneously acquire PL images of very large numbers of PV
modules, in similar fashion to what is currently common practice with
RPA-based thermal infrared imaging. RPA-based camera deployment enables
not only the acquisition of images of large sections of the farm from a
high vantage point, but also convenient and rapid change of the camera
field of view by variation of the distance to the target. For the
experiments discussed above positioning of the RPA prior to image
acquisition was achieved with manual RPA flight control. For the near
future we envisage automated mapping of solar farms with pre-programmed
flight paths and way points, as is current practice for thermal
inspection. In this scenario we envisage the use of high-level overview
PL imaging data captured from RPAs, as shown in Fig.5, in combination
with either close-up RPA-based imaging or high-fidelity imaging from the
ground. This will enable rapid overview images to be acquired first,
followed by more detailed inspection of specific modules or farm
sections with suspected defects.
Discussion
The ability to acquire outdoor DPL images of entire PV systems or of
large sections of solar farms using inverter-based modulation is a very
significant step towards cost effective and commercially viable daylight
luminescence imaging of large utility scale solar systems. The analysis
of relative PL intensities is a particularly powerful quantitative tool
to assess voltage variations between and within modules, and to quantify
module degradation. Of particular importance in this context is the
robustness of PL image data against variations in sample temperature.
Zafirovska et al. showed that the temperature coefficient of implied
voltages that are derived from luminescence is about ten times lower
than the temperature coefficient of the terminal voltage. This is
particularly relevant here since measuring temperatures on PV modules in
the field with good accuracy is generally difficult to achieve. While
such accurate temperature measurements are a key requirement for the
analysis of terminal voltage measurements of fielded modules, it is not
required for quantitative PL image analysis due to the low temperature
sensitivity of the PL signal.
The inverter-based manipulation of the operating point of large numbers
of modules that are connected to specific inverters in combination with
RPA-based image acquisition is an elegant solution to inspect large
solar assets, and to find faulty modules or determine module
degradation. No modifications to the system wiring are required, it can
be performed during the day and does not require large and heavy
equipment such as power supplies and generators to be brought on site.
Daylight PL imaging also avoids additional hazards (electrical,
night-work) and administrative difficulties associated with working at
nighttime.
Summary and Conclusions
High quality daylight photoluminescence image acquisition of crystalline
silicon solar modules in both rooftop and utility-scale solar farms
using electrical modulation of the operating point of modules connected
to an inverter was demonstrated. Measurements were performed by
manipulating the operating point of modules using existing functionality
of commercial inverters, without specific customisation of either
hardware or firmware, without any modifications to the PV system’s
electrical wiring and without utilising additional large power
electronic equipment, such as generators and power supplies.
Importantly, measurements were performed in daylight during regular
working hours, avoiding the administrative and practical complications
and risks associated with nighttime operation.
It was shown that close-up high-resolution and high-quality images of
individual modules enable the detection of microcracks and individual
cell level defects. In contrast, RPA based overview images of large
sections of farms, with up to 1,200 modules contained in a single PL
image, as demonstrated here, allow the detection of coarse cell, module,
and system-level defects. Quantitative analysis enables accurate
assessment of voltage variations within and between modules. The work
presented in this paper is a significant step towards commercially
viable, fast and effective daylight photoluminescence imaging becoming a
routine part of large-scale PV plant inspection.