Infrared thermography based defect analysis of photovoltaic modules using machine learning

Abstract

Solar photovoltaic (PV) has gained much attention throughout the world for clean energy production. Faults in the PV modules cause the reduction of the amount of the electricity gain from the PV systems. Detecting faults of PV modules could help to take necessary measures. In this study, Infrared thermography (IRT) is used in order to take images of PV modules which may indicate the hotspot. Later, these images are converted into datasets for a classifier to detect the hotspot of PV modules. Besides, I-V characteristics of the PV modules are also analyzed to find out the relation between hotspot & defected area. The further part of this study has been conducted by using a machine learning tool called ‘YOLO: You Only Look once’. After training & testing the learner with the datasets, the outputs are validated with the IRT images of PV modules. The major gain of this study is to apply a modified real time object detection tool to understand and detect the defect of the PV module. The algorithm is capable of detecting the hotspot using the weightage file of the training phase. Result shows that with more diversified datasets the accuracy of detecting hotspot increases.