Content uploaded by Michael Gostein
Author content
All content in this area was uploaded by Michael Gostein on Aug 16, 2024
Content may be subject to copyright.
Content uploaded by Michael Gostein
Author content
All content in this area was uploaded by Michael Gostein on Nov 26, 2018
Content may be subject to copyright.
MARS™ SOILING SENSOR
Michael Gostein, Stan Faullin, Keith Miller, Jason Schneider, Bill Stueve
Atonometrics, 8900 Shoal Creek Blvd. Suite 116, Austin, TX, 78757, USA, info@atonometrics.com
ABSTRACT: We evaluate a new proprietary optical soiling measurement technology that does not require washing.
The compact, water-free, and maintenance-free sensor technology is designed to permit PV system soiling
measurement with lower cost and labor requirements compared to traditional soiling measurement systems using a
clean/soiled pair of PV reference devices. These attributes make the sensor suitable for a wide range of sites including
not only utility but also commercial-scale PV installations. Laboratory test results using soiled glass test coupons
demonstrate excellent correlation between the optical sensor and a PV cell, and field testing is underway.
1 INTRODUCTION
Soiling of PV modules is one of the principal energy
loss factors for PV systems. Accordingly, in recent years
there has been a growing interest in measuring PV soiling
at both pre-production prospecting and operational sites
[1]. Motivations include performance prediction,
monitoring of plant operations, verifying compliance
with performance guarantees, and optimizing wash
schedules for best return-on-investment of washing
expense.
Traditional soiling measurement systems compare the
outputs of two PV devices, one of which is allowed to
soil naturally and the other of which is routinely cleaned
to serve as a reference [1][2]. Daily or weekly cleaning is
performed either manually by onsite personnel or by an
automated [3] washing system.
Eliminating the requirement for washing a reference
device in order to measure soiling would lower the cost
of soiling monitoring, and be especially favorable for
commercial-scale PV sites that typically have smaller
budgets and no onsite personnel.
In this work we evaluate a new proprietary all-optical
soiling measurement concept developed by Atonometrics
(Mars Soiling SensorTM, [4][5]), conceived as a solution
for maintenance-free measurement of soiling loss. The
Mars™ sensor is designed to be deployed at a PV
installation site in plane-of-array and to operate
unattended and without water. Soiling particles
accumulate on the sensor window just as on any nearby
PV modules and the sensor directly detects the soiling
level on the sensor window.
2 OPTICAL MEASUREMENT OF SOILING
Consider a PV device covered by a transparent
window upon which soiling particles have accumulated.
The soiling particles block incident sunlight through
reflection and absorption of light, and therefore result in
lost power production by the PV device relative to a
clean state of the window. The soiling loss SL, the
fractional loss in power due to the soiling particles, can
be expressed as
SL = 1 – T = R + A (1)
where T is the fractional transmission of light through the
layer of soiling particles, R is the fractional reflection (or
scattering) of light, and A is the fractional absorption of
light. Here we consider T, R, and A as properties of the
layer of soiling particles only, neglecting any reflection
or absorption by the glass, i.e. SL is only the additional
loss due to the soiling particles. For simplicity, in eq. (1)
we also neglect the spectral irradiance distribution of
incident light, spectral transmissivity of the soiling layer,
and spectral response of the PV device, although these
can be considered in a more complete treatment.
Our aim is to determine SL or, equivalently, T.
Recently another optical soiling measurement device
has been developed [6] which uses a reflection
measurement to measure a signal proportional to the R
term in eq. (1). A beam of light is directed from inside
the sensor to the sensor window and light that is back-
reflected/scattered by soiling particles is detected. The A
term is neglected and SL is assumed to be proportional to
the reflectance signal or R term. The device therefore
requires separate calibrations for different colors of dust
with different reflectance/absorbance. This is a
significant difficulty since the type and color of dust
encountered at a particular site are not necessarily known
and may even change over time with weather conditions.
In contrast, our sensor concept directly measures both
the T and R terms in eq. (1) using a microscopic imaging
approach, similar to the soiling microscope used in [7] to
study particle deposition dynamics, although with
important differences.
Figure 1: Mars Soiling Sensor™ concept. A microscope
system captures an image of a soiling particle layer on a
window using either external (sunlight) or internal (LED)
illumination. Black and white reference marks provide
calibration features for image analysis.
Preprint, to appear in proceedings of 35th EU PVSEC, September 2018
3 MARS™ SENSOR CONCEPT
Fig. 1 illustrates the Mars Soiling Sensor™ concept.
Soiling particles accumulate on a glass window mounted
on the outside of the sensor enclosure. A camera
assembly, consisting of a lens and image sensor, is
focused on the outer layer of the window and captures a
microscopic image of the soiling particle layer, which can
be illuminated by sunlight and/or by internal LEDs. A
diffuser spreads the LED illumination evenly across the
window. Black and white reference marks screen-printed
on the inside of the window are used as self-calibration
features for image analysis, as well as for detection and
rejection of invalid images. Note that the window edge is
not covered, to prevent trapping moisture and dust. Fig. 2
is a photograph of the actual sensor head, which better
shows the reference marks which are formed as two
concentric circles. The sensor is focused to obtain images
from a field of view on the window surface with diameter
up to approximately 15 mm.
Fig. 3 shows an example acquired image when the
sensor head is covered with a layer of ISO 12103-1, A2
fine test dust (“Arizona road dust”), which has a large
silica content and is somewhat white and reflective. The
sensor is illuminated externally by simulated diffuse
sunlight. The dust particles, as well as both the black and
white reference marks, appear dark, as they cast shadows
blocking light entering from outside the sensor. To
analyze the image, we use a proprietary background
correction algorithm to correct for non-uniformities in
illumination, identify a reference pixel brightness
corresponding to local regions of maximum transmission,
and then compute the average brightness of all pixels,
excluding the reference marks, relative to the reference
pixel brightness. The result directly yields the fractional
transmission T and therefore SL. Note that this technique
does not require that the camera resolve individual dust
particles, unlike the method in [7], and that it does not
assume any optical properties of the dust itself, which
could be either opaque or partially translucent.
The image analysis method is further illustrated in
Fig. 4, which shows pixel intensity histograms of Mars™
images, similar to that of Fig. 3, of two samples, one
clean and one soiled by dust. Note that pixels
corresponding to the circular reference mark pattern are
not included in the histograms. For the clean sample the
histogram distribution is very narrow, with most pixels
having intensity near a value corresponding to clean glass
and maximum light transmission. For the soiled sample,
the histogram shows a wide range of pixel intensity
values, indicating that many of the pixels have reduced
light transmission. As described above, soiling loss is
calculated by determining from each image a reference
pixel brightness corresponding to clean glass and then
comparing all pixels to the reference brightness to
calculate average transmission and therefore soiling loss.
Fig
ure 2: Mars Soiling Sensor™
measurement head,
showing reference marks on inside of window.
Fig
ure 3: View through the sensor head with
external
illumination and
a layer of dust on the sensor window.
Figure
4:
Pixel intensity distribution of Mars™ images
of two samples, one clean and one soiled.
Figure
5: View through the sensor head
with internal
illumination
and simulated snow on the sensor window.
By using internal illumination from the LEDs,
additional information can be gained. When illuminated
from the inside (for example, at night), the camera
captures an image of the reflection of material on the
window. This mode of operation can be used to measure
the R term in eq. 1, thus setting a lower limit on SL and
providing an independent check. In addition, the
measured reflectance can be used to provide a small
correction to the external-illumination determination of T
to account for daylight which enters the sensor, scatters
upwards, and illuminates the particles from below. Thus
analysis may combine both externally illuminated (i.e.
daytime) and internally-illuminated (e.g. nighttime)
images of the soiling layer.
Figure 5 shows another potential application of
internal illumination, the detection of snow or other
reflective substances (e.g. bird droppings) on the sensor
which block external light. The image in Fig. 5 is taken
with the sensor covered with simulated snow (a white
foam sheet) and illuminated from the inside by LEDs.
Note that the white reference marks reflect the internal
illumination and that reflection from the simulated snow
layer is also apparent in the image and can be referenced
to the reflection from the white marks.
4 LABORATORY TESTING
Laboratory evaluation was performed using soiled
test coupons. Coupons were prepared using three
different dust materials, of different colors, representative
Fig
ure 6: Dust powders (top) and microscope images (bottom) of representative soiled glass test coupons
using each of the
dust
powders, Arizona road dust (left), carbon black (center), and red iron oxide (right)
. The clear glass soiled coupons were
overlaid on a PV cell for the
images
; although appearance varies due to dust reflectivity, soil levels of each sample are similar
(~12%). Image widths are 5 mm.
Figure 7
:
Mars™ images of representative soiled glass test coupons, illustrating the effects of varying deposition conditions.
Each image has width ~11 mm.
of airborne soiling particles. These included fine Arizona
road dust (ISO 12103-1, A2, Powder Technology), 30 um
filtered carbon black (Inoxia), and 30 um filtered red iron
oxide (Alpha Chemicals). The three materials are
respectively white, black, and red in appearance, as
shown in Fig. 6. The coupons consisted of 2” x 2”
borosilicate glass slides. Dust was uniformly applied to
the coupons by placing them, together with a selected
dust material, in a custom-built dusting box, which uses a
high speed recirculation fan to maintain a dust-laden air
stream. Prior to dust exposure, the coupons were chilled
to -10 °C to 5 °C, causing moisture condensation during
exposure which promotes dust adhesion. Fig. 6 illustrates
the varying appearance of the three different materials
once applied to the coupons. The figure shows
microscope images of coupons with each of the three
materials in which the clear glass coupon is overlaid on a
PV cell, simulating the appearance of a soiled PV device.
Although all three coupons have similar soiling level of
~12% (measured as discussed below) their appearance is
quite different: the white Arizona road dust particles are
bright and reflective, while the carbon black is barely
visible against the dark background and the red iron
oxide has an intermediate appearance. This illustrates the
difficulty of measuring particle coverage by appearance
or by reflectivity.
By varying the deposition conditions, including pre-
chilling temperature, exposure time, position within the
dusting box, and other parameters, we achieved a wide
variety of different dust coverages and morphologies, as
illustrated in Fig. 7. The figure shows close-ups of
images of various coupons as acquired with the Mars™
sensor, similar to Fig. 3, where each image is an image of
the shadows of the dust particles against the external
lighting. Note in some images that particles have
agglomerated together into larger collections, due to the
moisture level, which may be similar to natural soiling
processes.
As a control, the soiling loss of each coupon was
measured with a PV cell illuminated with a halogen
lamp. A 10 mm diameter aperture was placed in front of
the cell to limit the spot size of the measurement. The
ratio of the PV cell short-circuit current for each coupon
to its short-circuit current with clean glass was recorded,
and one minus the ratio was logged as the soiling loss for
the coupon. A small correction was performed to account
for reflection of light from the multiple interfaces of the
soiled coupons. The characteristic uniformity of soiling
loss within the central region of the samples was
estimated by translating selected samples across the PV
cell aperture and measuring maximum and minimum
soiling loss.
Following PV measurement, each sample was
positioned on the window of the Mars™ sensor and
images were acquired with both external and internal
illumination. Images were processed as described above
and the results were correlated with the control
measurements using the PV cell.
Results are shown in Fig. 8. Error bars represent the
uncertainty in the control measurements due to lamp
instability as well as non-uniformity of the soiling in the
central region of each sample. White, black, and red
symbols indicate the white, black, and red dust particles.
Blue symbols show other materials, which include salt, a
mixture of the three dusts, and clean glass. The
correlation between the optical soiling sensor and the PV
cell is very good, with R2 of 0.97 and standard error of
correlation of 1.1%. Furthermore, all three different dust
materials fall on the same trend line, suggesting a near-
universal response of the sensor independent of dust
Fig
ure 8: Correlation of soiling loss measured by the Mars Soiling Sensor™ versus results with
a PV cell, for various
representative
dust materials including those with white, black, and red particles as indicated by the colored symbols.
R2 = 0.97
Std. Err = 1.1%
color.
The results presented in Fig. 8 are for soiling levels
up to 25%, which is a practical range for PV system
operation. Additional testing has been performed up to
75% soiling levels, showing adequate performance
although with larger errors. Work is ongoing in this area.
5 OUTDOOR TESTING
Outdoor testing of the Mars™ sensor is underway,
with several units deployed or in process of deployment
at locations in Texas and California, together with
traditional soiling measurement techniques for data
correlation. Due to low soiling and frequent rains at the
already installed test sites, data with significant soiling
levels for correlation are not yet available, although
initial results demonstrate the ability of the sensor to
operate properly in the outdoor environment and in
diffuse sunlight illumination. Further work is ongoing.
6 DISCUSSION AND CONCLUSIONS
Laboratory data show promising results for the new
water-free and maintenance-free Mars Soiling Sensor™
concept. In particular, the results show very little, if any,
dependence on dust color, in contrast to [6].
Note that the results presented in Fig. 8 do not
include any arbitrary calibration constants. The slope of
the best-fit line to all data points is 0.95, indicating that
without calibration the technique already measures
soiling loss close to the control method. By applying a
calibration factor, the results can of course be improved.
One potential reason for the non-unity slope of the
correlation could be spectral mismatch between the PV
cell and the image sensor. Further work is needed to
assess this. Also, while to date we have processed
acquired images in grayscale only, use of the color
sensitivity of the image sensor could in the future allow
automatic compensation for a spectral mismatch effect.
Our method for determining T does not depend on
dust color, but it does depend on the morphology of the
dust layer, since we must be able to observe some gaps
between dust particles in order to determine reference
intensities corresponding to local regions of clean glass.
Therefore the method may be less effective for highly
soiled samples with very small and highly uniformly
distributed particles. However, in such cases our
complementary method of determining R using internal
illumination could provide a check and lower limit on SL.
Determining the accuracy of the Mars™ technique in
actual practice will require long term outdoor testing with
naturally-occurring soiling in a variety of conditions.
Finally, we note that our optical sensor measures
soiling loss only at a relatively small region on the sensor
itself, while PV module power loss can be significantly
affected by the distribution pattern of dust across an
entire module [8][9]. We envision potential extensions of
the technique that could account for these effects [10].
Overall, the promising results obtained in laboratory
testing so far suggest that the Mars Soiling Sensor™
concept could be an effective, compact, low-cost, and
unattended soiling sensor practical not only for utility-
scale but also commercial-scale PV projects.
7 ACKNOWLEDGEMENTS
We wish to thank Christina Fonda for her help in
preparing the dust coupons used for this study.
8 REFERENCES
[1] M. Gostein, J. R. Caron, B. Littmann, “Measuring
soiling losses at utility-scale PV power plants,” in
40th IEEE Photovoltaic Specialists Conference,
Denver, CO, 2014.
[2] “Photovoltaic system performance – part 1:
monitoring,” IEC standard 61724-1, 2017.
[3] M. Gostein, et al, “Soiling measurement station to
evaluate anti-soiling properties of PV module
coatings,” in 43rd IEEE Photovoltaic Specialists
Conference, Portland, OR, 2016.
[4] Patents Pending.
[5] M. Gostein, et al, “Mars Soiling Sensor™,” in 45th
IEEE Photovoltaic Specialists Conference, Hawaii,
2018.
[6] M. Korevaar, et al, “Novel soiling detection system
for solar panels,” in 33rd European Photovoltaic
Solar Energy Conference, Amsterdam, the
Netherlands, 2017.
[7] B. Figgis, et al, “Outdoor soiling microscope for
measuring particle deposition and resuspension,”
Solar Energy, v. 137, pp. 158-164, November 2016.
[8] M. Gostein, B. Littmann, J. R. Caron, L. Dunn,
“Comparing PV power plant soiling measurements
extracted from PV module irradiance and power
measurements,” in 39th IEEE Photovoltaic
Specialists Conference, Tampa, FL, 2013.
[9] M. Gostein, B. Stueve, M. Chan, “Accurately
measuring PV soiling losses with soiling station
employing PV module power measurements,” in
44th IEEE Photovoltaic Specialists Conference,
Washington, DC, 2017.
[10] Patents Pending.
