Zandamela, FrankPratt, Lawrence EMay, Siyasanga IMkasi, Hlaluku WMabeo, Reuben T2024-05-062024-05-062023-11Zandamela, F., Pratt, L.E., May, S.I., Mkasi, H.W. & Mabeo, R.T. 2023. Towards improved solar PV module characterisation: Correlating electroluminescence image defects with I-V curve characteristics using a semantic segmentation based multi-defect detection algorithm. http://hdl.handle.net/10204/13668 .978-0-7972-1907-6http://hdl.handle.net/10204/13668There has been significant research on the relationship between current-voltage (I-V) curve characteristics and electroluminescence (EL) module defects. Current methods use EL image pixels to develop features, which are then correlated with module I-V curve characteristics. In most cases, image thresholding is used to gather pixel information. These approaches have two major limitations. First, they lack generalisability, as imaging conditions may vary from module to module, and thresholding algorithms are often developed for specific types of defects or imaging conditions. Second, the correlation between specific types of defects and I-V features cannot be studied because all defects are grouped into one highlevel defect detected by a sharp change in pixel intensity. In this paper, we conduct a correlation study between EL defects and IV curve characteristics of photovoltaic (PV) modules that were exposed to accelerated stress testing. We correlate power loss and two common EL defects. The defects are detected and quantified using a prediction model based on semantic segmentation in which each pixel is assigned to one of multiple classes. Results obtained indicate that the defect detection tool can be used to correlate power loss with dark cells and cell cracks. A significant amount of variability in output power delta can be explained by defects detected by the prediction model (r2= 72%).FulltextenCell cracksElectroluminescence image defect detectionI-V curve characteristicsDeep learningPV moduleSemantic segmentationConference PresentationZandamela, F., Pratt, L. E., May, S. I., Mkasi, H. W., & Mabeo, R. T. (2023). Towards improved solar PV module characterisation: Correlating electroluminescence image defects with I-V curve characteristics using a semantic segmentation based multi-defect detection algorithm. http://hdl.handle.net/10204/13668Zandamela, Frank, Lawrence E Pratt, Siyasanga I May, Hlaluku W Mkasi, and Reuben T Mabeo. "Towards improved solar PV module characterisation: Correlating electroluminescence image defects with I-V curve characteristics using a semantic segmentation based multi-defect detection algorithm." <i>Southern African Sustainable Energy Conference (SASEC), Gqeberha, Port Elizabeth, 15-17 November 2023</i> (2023): http://hdl.handle.net/10204/13668Zandamela F, Pratt LE, May SI, Mkasi HW, Mabeo RT, Towards improved solar PV module characterisation: Correlating electroluminescence image defects with I-V curve characteristics using a semantic segmentation based multi-defect detection algorithm; 2023. http://hdl.handle.net/10204/13668 .TY - Conference Presentation AU - Zandamela, Frank AU - Pratt, Lawrence E AU - May, Siyasanga I AU - Mkasi, Hlaluku W AU - Mabeo, Reuben T AB - There has been significant research on the relationship between current-voltage (I-V) curve characteristics and electroluminescence (EL) module defects. Current methods use EL image pixels to develop features, which are then correlated with module I-V curve characteristics. In most cases, image thresholding is used to gather pixel information. These approaches have two major limitations. First, they lack generalisability, as imaging conditions may vary from module to module, and thresholding algorithms are often developed for specific types of defects or imaging conditions. Second, the correlation between specific types of defects and I-V features cannot be studied because all defects are grouped into one highlevel defect detected by a sharp change in pixel intensity. In this paper, we conduct a correlation study between EL defects and IV curve characteristics of photovoltaic (PV) modules that were exposed to accelerated stress testing. We correlate power loss and two common EL defects. The defects are detected and quantified using a prediction model based on semantic segmentation in which each pixel is assigned to one of multiple classes. Results obtained indicate that the defect detection tool can be used to correlate power loss with dark cells and cell cracks. A significant amount of variability in output power delta can be explained by defects detected by the prediction model (r2= 72%). DA - 2023-11 DB - ResearchSpace DP - CSIR J1 - Southern African Sustainable Energy Conference (SASEC), Gqeberha, Port Elizabeth, 15-17 November 2023 KW - Cell cracks KW - Electroluminescence image defect detection KW - I-V curve characteristics KW - Deep learning KW - PV module KW - Semantic segmentation LK - https://researchspace.csir.co.za PY - 2023 SM - 978-0-7972-1907-6 T1 - Towards improved solar PV module characterisation: Correlating electroluminescence image defects with I-V curve characteristics using a semantic segmentation based multi-defect detection algorithm TI - Towards improved solar PV module characterisation: Correlating electroluminescence image defects with I-V curve characteristics using a semantic segmentation based multi-defect detection algorithm UR - http://hdl.handle.net/10204/13668 ER -27605