A novel deep learning framework for PV module thermal condition monitoring

The increasing consumption of solar energy has generated a requirement for efficient techniques to monitor and evaluate the condition of photovoltaic modules. This research approaches the difficulty by developing a novel transfer learning framework that employs thermographic images and deep convolutional neural networks (DCNNs) for non-intrusive and reliable photovoltaic module monitoring. The framework analyzes variations in temperatures over time with the help of a thermal imager, further leading to the categorization of photovoltaic modules as either functional or faulty. Three fault scenarios, namely shadowing causing less than 15% power loss, dust hotspots causing 15–20% power loss and cracking causing above 20% power loss, lead to the classification of faults based on power loss. These criteria were determined by empirical data analysis where the power was recorded by the PV analyzer 9018BT. Furthermore, the images captured under photovoltaic module condition monitoring were used to train the different pretrained DCNN transfer learning models, such as ResNet-50, GoogLeNet and VGG-19, for photovoltaic module health classification, whether healthy or not, and compare the model’s efficacy with the proposed VGG-16-based architecture. The findings indicate that the suggested framework reaches an accuracy of 95.5%, surpassing ResNet-50 (83.3%), GoogLeNet (73.5%) and VGG-19 (90.9%). The results underscore the efficacy of the suggested method in precisely identifying and categorizing photovoltaic module defects. This study provides a significant contribution by presenting an economical, precise and scalable approach for the monitoring of photovoltaic module conditions. The novel use of transfer learning for thermographic data offers a powerful tool for early problem diagnosis, enhancing maintenance strategies and prolonging the lifetime of photovoltaic systems.