Fault Diagnosis of PV Modules Using Deep Convolutional Neural Networks

Abstract

Conventional methods of fault diagnosis for PV Systems are quite challenging and inefficient, particularly with regard to large-scale PV arrays. Early and effective diagnosis of system faults is also imperative in order to minimize cost and sustainable damage. Hence, over recent years, numerous effective and efficient monitoring and diagnostic techniques to detect defects in PV systems have been studied and propositioned. Over the last few years, various deep learning frameworks have been studied and proposed for the detection & classification of faults in PV modules with the aid of thermal images. This study involves the utilization of Convolutional Neural Networks (CNN), namely, VGG-16/VGG-19 and EfficientNet, in order to assess their performance and reliability in diagnosing module defects through significant hotspots within PV modules by employing pre-processed thermal images. The result shows that VGG-16 had significant superiority over other models in terms of performance and accuracy.