Hybrid Convolutional Neural Network on Fault Detection in Electroluminescence Images of Photovoltaic Cells

The expansion of installed capacity in photovoltaic generation systems demands automated methods for fault detection in its constituent cells. This paper proposes a hybrid convolutional neural network model for fault detection in electroluminescence images of photovoltaic panels. The model leverages the convolutional neural networks ResNet50 and VGG16 for feature extraction and the support vector machine classifier to detect faulty cells. Adjusting the model's settings using a genetic algorithm achieved accuracy rates of 98.2% and 99.7% in tests with two public datasets. The challenges that this dataset's heterogeneity imposed on training the model were addressed by data augmentation and contrast enhancement techniques. These results support hybrid convolutional neural networks as a promising solution for automatically detecting defects in photovoltaic cells, which is important for keeping energy conversion efficiency high and extending the lifespan of photovoltaic systems.