Thermal image-driven CNN for predicting solar photovoltaic module lifespan from hotspots

Abstract

Reliability research on photovoltaic (PV) modules is still underdeveloped. Environmental factors involved in decreasing system performance are studied. They depend on environmental conditions, technology, design and materials used. It is essential therefore, a detailed study on these factors to then be able to quantify module degradation. Current challenges stem from thermal-induced degradation, where hotspot formation accelerates aging processes and reduces module lifespan, directly impacting system economics and reliability. Existing inspection methods lack predictive capabilities for quantitative lifetime assessment, limiting effective maintenance planning and investment decision making. This research proposes a modified convolutional neural network (Mod-CNN) that uses thermal images to predict the lifespan of solar PV based on degradation techniques. The proposed model integrates thermal hotspot images with the corresponding average temperature data to train a specialized CNN architecture that features enhanced attention mechanisms for thermal pattern recognition. This novel approach combines physics-based degradation modeling via the Peck model with advanced machine learning techniques, achieving exceptional performance with R² (0.97), significantly surpassing conventional methodologies. Comprehensive validation across multiple datasets confirms the effectiveness of the model in accurately predicting solar module operational lifespan. In this research work, the average hotspot temperature and its corresponding degradation data for thermal images serve as the training foundation for the deep learning model implemented in MATLAB. The results demonstrate the effectiveness of the model in predicting the lifespan of solar module. This research work is especially important from a practical point of view in the solar PV installation field for planners, installers, consumers, and financers.