Machine Learning Prediction of Defect Types for Electroluminescence Images of Photovoltaic Panels

Research output: Chapter in Book/Report/Conference proceedingArticle in proceedingsResearchpeer-review

167 Downloads (Pure)

Abstract

Despite recent technological advances for Photovoltaic panels maintenance (Electroluminescence imaging, drone inspection), only few large-scale studies achieve identification of the precise category of defects or faults. In this work, Electroluminescence imaged modules are automatically split into cells using projections on the x and y axes to detect cell boundaries. Regions containing potential defects or faults are then detected using Hough transform combined with mathematical morphology. Care is taken to remove most of the bus bars or cell boundaries. Afterwards, 25 features are computed, focusing on both the geometry of the regions (e.g. area, perimeter, circularity) and the statistical characteristics of their pixel values (e.g. median, standard deviation, skewness). Finally, features are mapped to the ground truth labels with Support Vector Machine (RBF kernel) and Random Forest algorithms, coupled with undersampling and SMOTE oversampling, with a stratified 5- folds approach for cross validation. A dataset of 982 Electroluminescence images of installed multi-crystalline photovoltaic modules was acquired in outdoor conditions (evening) with a CMOS sensor. After automatic blur detection, 753 images or 47244 cells remain to evaluate faults. All images were evaluated by experts in PV fault detection that labelled: Finger failures, and three types of cracks based on their respective severity levels (A, B and C). Our results based on 6 data series, yield using Support Vector Machine an accuracy of 0.997 and a recall of 0.274. Improving the region detection process will most likely allow improving the performance.
Original languageEnglish
Title of host publicationProceedings of SPIE
Number of pages9
Volume11139
PublisherSPIE - International Society for Optical Engineering
Publication date2019
DOIs
Publication statusPublished - 2019
Event14th International Conference on Solid State Lighting and LED-based Illumination Systems
: SPIE Optical Engineering + Applications
- San Diego Convention Center, San Diego, United States
Duration: 9 Aug 201513 Aug 2015
Conference number: 14

Conference

Conference14th International Conference on Solid State Lighting and LED-based Illumination Systems
Number14
LocationSan Diego Convention Center
CountryUnited States
CitySan Diego
Period09/08/201513/08/2015
Other9571
SeriesProceedings of SPIE, the International Society for Optical Engineering
Volume11139
ISSN0277-786X

Keywords

  • Machine learning
  • Solar panel
  • Defect detection
  • Fault detection
  • Electroluminescence imaging

Cite this

Mantel, C., Villebro, F., Benatto, G. A. D. R., Parikh, H. R., Wendlandt, S., Hossain, K., ... Forchhammer, S. (2019). Machine Learning Prediction of Defect Types for Electroluminescence Images of Photovoltaic Panels. In Proceedings of SPIE (Vol. 11139). SPIE - International Society for Optical Engineering. Proceedings of SPIE, the International Society for Optical Engineering, Vol.. 11139 https://doi.org/10.1117/12.2528440
Mantel, Claire ; Villebro, Frederik ; Benatto, Gisele Alves dos Reis ; Parikh, Harsh Rajesh ; Wendlandt, Stefan ; Hossain, Kabir ; Poulsen, Peter Behrensdorff ; Spataru, Sergiu ; Sera, Dezso ; Forchhammer, Søren. / Machine Learning Prediction of Defect Types for Electroluminescence Images of Photovoltaic Panels. Proceedings of SPIE. Vol. 11139 SPIE - International Society for Optical Engineering, 2019. (Proceedings of SPIE, the International Society for Optical Engineering, Vol. 11139).
@inproceedings{7eceed1439874b82a5e48897fae0e5f0,
title = "Machine Learning Prediction of Defect Types for Electroluminescence Images of Photovoltaic Panels",
abstract = "Despite recent technological advances for Photovoltaic panels maintenance (Electroluminescence imaging, drone inspection), only few large-scale studies achieve identification of the precise category of defects or faults. In this work, Electroluminescence imaged modules are automatically split into cells using projections on the x and y axes to detect cell boundaries. Regions containing potential defects or faults are then detected using Hough transform combined with mathematical morphology. Care is taken to remove most of the bus bars or cell boundaries. Afterwards, 25 features are computed, focusing on both the geometry of the regions (e.g. area, perimeter, circularity) and the statistical characteristics of their pixel values (e.g. median, standard deviation, skewness). Finally, features are mapped to the ground truth labels with Support Vector Machine (RBF kernel) and Random Forest algorithms, coupled with undersampling and SMOTE oversampling, with a stratified 5- folds approach for cross validation. A dataset of 982 Electroluminescence images of installed multi-crystalline photovoltaic modules was acquired in outdoor conditions (evening) with a CMOS sensor. After automatic blur detection, 753 images or 47244 cells remain to evaluate faults. All images were evaluated by experts in PV fault detection that labelled: Finger failures, and three types of cracks based on their respective severity levels (A, B and C). Our results based on 6 data series, yield using Support Vector Machine an accuracy of 0.997 and a recall of 0.274. Improving the region detection process will most likely allow improving the performance.",
keywords = "Machine learning, Solar panel, Defect detection, Fault detection, Electroluminescence imaging",
author = "Claire Mantel and Frederik Villebro and Benatto, {Gisele Alves dos Reis} and Parikh, {Harsh Rajesh} and Stefan Wendlandt and Kabir Hossain and Poulsen, {Peter Behrensdorff} and Sergiu Spataru and Dezso Sera and S{\o}ren Forchhammer",
year = "2019",
doi = "10.1117/12.2528440",
language = "English",
volume = "11139",
series = "Proceedings of SPIE, the International Society for Optical Engineering",
publisher = "SPIE - International Society for Optical Engineering",
booktitle = "Proceedings of SPIE",

}

Mantel, C, Villebro, F, Benatto, GADR, Parikh, HR, Wendlandt, S, Hossain, K, Poulsen, PB, Spataru, S, Sera, D & Forchhammer, S 2019, Machine Learning Prediction of Defect Types for Electroluminescence Images of Photovoltaic Panels. in Proceedings of SPIE. vol. 11139, SPIE - International Society for Optical Engineering, Proceedings of SPIE, the International Society for Optical Engineering, vol. 11139, 14th International Conference on Solid State Lighting and LED-based Illumination Systems
, San Diego, United States, 09/08/2015. https://doi.org/10.1117/12.2528440

Machine Learning Prediction of Defect Types for Electroluminescence Images of Photovoltaic Panels. / Mantel, Claire; Villebro, Frederik; Benatto, Gisele Alves dos Reis; Parikh, Harsh Rajesh; Wendlandt, Stefan; Hossain, Kabir; Poulsen, Peter Behrensdorff; Spataru, Sergiu; Sera, Dezso; Forchhammer, Søren.

Proceedings of SPIE. Vol. 11139 SPIE - International Society for Optical Engineering, 2019. (Proceedings of SPIE, the International Society for Optical Engineering, Vol. 11139).

Research output: Chapter in Book/Report/Conference proceedingArticle in proceedingsResearchpeer-review

TY - GEN

T1 - Machine Learning Prediction of Defect Types for Electroluminescence Images of Photovoltaic Panels

AU - Mantel, Claire

AU - Villebro, Frederik

AU - Benatto, Gisele Alves dos Reis

AU - Parikh, Harsh Rajesh

AU - Wendlandt, Stefan

AU - Hossain, Kabir

AU - Poulsen, Peter Behrensdorff

AU - Spataru, Sergiu

AU - Sera, Dezso

AU - Forchhammer, Søren

PY - 2019

Y1 - 2019

N2 - Despite recent technological advances for Photovoltaic panels maintenance (Electroluminescence imaging, drone inspection), only few large-scale studies achieve identification of the precise category of defects or faults. In this work, Electroluminescence imaged modules are automatically split into cells using projections on the x and y axes to detect cell boundaries. Regions containing potential defects or faults are then detected using Hough transform combined with mathematical morphology. Care is taken to remove most of the bus bars or cell boundaries. Afterwards, 25 features are computed, focusing on both the geometry of the regions (e.g. area, perimeter, circularity) and the statistical characteristics of their pixel values (e.g. median, standard deviation, skewness). Finally, features are mapped to the ground truth labels with Support Vector Machine (RBF kernel) and Random Forest algorithms, coupled with undersampling and SMOTE oversampling, with a stratified 5- folds approach for cross validation. A dataset of 982 Electroluminescence images of installed multi-crystalline photovoltaic modules was acquired in outdoor conditions (evening) with a CMOS sensor. After automatic blur detection, 753 images or 47244 cells remain to evaluate faults. All images were evaluated by experts in PV fault detection that labelled: Finger failures, and three types of cracks based on their respective severity levels (A, B and C). Our results based on 6 data series, yield using Support Vector Machine an accuracy of 0.997 and a recall of 0.274. Improving the region detection process will most likely allow improving the performance.

AB - Despite recent technological advances for Photovoltaic panels maintenance (Electroluminescence imaging, drone inspection), only few large-scale studies achieve identification of the precise category of defects or faults. In this work, Electroluminescence imaged modules are automatically split into cells using projections on the x and y axes to detect cell boundaries. Regions containing potential defects or faults are then detected using Hough transform combined with mathematical morphology. Care is taken to remove most of the bus bars or cell boundaries. Afterwards, 25 features are computed, focusing on both the geometry of the regions (e.g. area, perimeter, circularity) and the statistical characteristics of their pixel values (e.g. median, standard deviation, skewness). Finally, features are mapped to the ground truth labels with Support Vector Machine (RBF kernel) and Random Forest algorithms, coupled with undersampling and SMOTE oversampling, with a stratified 5- folds approach for cross validation. A dataset of 982 Electroluminescence images of installed multi-crystalline photovoltaic modules was acquired in outdoor conditions (evening) with a CMOS sensor. After automatic blur detection, 753 images or 47244 cells remain to evaluate faults. All images were evaluated by experts in PV fault detection that labelled: Finger failures, and three types of cracks based on their respective severity levels (A, B and C). Our results based on 6 data series, yield using Support Vector Machine an accuracy of 0.997 and a recall of 0.274. Improving the region detection process will most likely allow improving the performance.

KW - Machine learning

KW - Solar panel

KW - Defect detection

KW - Fault detection

KW - Electroluminescence imaging

U2 - 10.1117/12.2528440

DO - 10.1117/12.2528440

M3 - Article in proceedings

VL - 11139

T3 - Proceedings of SPIE, the International Society for Optical Engineering

BT - Proceedings of SPIE

PB - SPIE - International Society for Optical Engineering

ER -

Mantel C, Villebro F, Benatto GADR, Parikh HR, Wendlandt S, Hossain K et al. Machine Learning Prediction of Defect Types for Electroluminescence Images of Photovoltaic Panels. In Proceedings of SPIE. Vol. 11139. SPIE - International Society for Optical Engineering. 2019. (Proceedings of SPIE, the International Society for Optical Engineering, Vol. 11139). https://doi.org/10.1117/12.2528440