The use of image processing to determine cell defects in polycrystalline solar modules

This research aims to use image processingtodetermine cell defects in polycrystalline solar modules. Image processing is a process of enhancing images for differentapplications. One domain that seems to not yet utilise the use of image processing, is photovoltaics. An increased use of fossil fuels is damaging the earth and a call to protect the earth has resulted in the emergence of pollutant-free technologies such as polycrystalline photovoltaic (PV) cells, which are connected to make up solar modules. However, defects often affect the performance of PV cells and consequently solar modules. Electroluminescence (EL) images are used to examine polycrystalline solar (PV) modules to determine if the modules are defective. The main research question that this research addressed is“How can an image processing technique be used to effectively identify defective polycrystalline PV cells from EL images of such cells?“. The experimental research methodology was used to address the main research question. The initial investigation into the problem revealed that certain sectors within industry, as well as the Physics Department at Nelson Mandela University(NMU), do not currently utiliseimage processing when examining EL images of solar modules. The current process is a tedious, manual process whereby solar modules are manually inspected. An analysis of the current processes enabled the identification of ways in which to automatically examine EL images of solar modules. An analysis of literatureprovided a better understanding of the different techniques that are used to examine solar modules, and it was identified how image processing can be applied to EL images. Further analysis of literatureprovided a better understanding of image processing and how image classification experiments using Deep Learning (DL) as an image processing technique can be used to address the main research question. The outcome of the experiments conducted in this research weredifferentadaptive models(LeNet, MobileNet, Xception)that can classify EL images of PV cellsaccording to known standardsused by the Physics Department at NMU. The known standards yielded four classes; normal, uncritical, critical and very critical, which were used for the classification of EL images of PV cells. The adaptive models were evaluated to obtain the precision, recall and F1–scoreof the models.The precession, recall, and F1–score were required to determine how effective the models were in identifying defective PV cells from EL images.The results indicated that an image processing technique canbe used to identify defective polycrystalline PV cells from EL images of such cells. However, further research needs to be conducted to improve the effectiveness of the adaptive models