A Multi-Modal Machine Learning Framework for Improved Health Assessment of End-of-Life PV Modules

The rapid increase in photovoltaic (PV) installations worldwide has raised growing concerns about the management of end-of-life (EOL) PV modules, with waste volumes projected to rise significantly in the coming decades. Robust and accurate health assessment of EOL PV modules is essential to enable their reuse as second-life products and to optimize recycling processes. Several studies have explored machine learning (ML) methods for assessing PV module health using current-voltage (IV) data and electroluminescence (EL) imaging. While successful, most of these techniques have been applied to single data sources, either EL images for crack detection or IV curves for failure classification. Such approaches fail to use the full potential of integrating multiple characterization methods. This paper proposes a conceptual framework for combining diverse data sources, such as EL images and IV curves, using ML approaches for EOL PV module health assessment at waste management facilities. It integrates both EL images and IV curves and uses the correlations between visual and electrical data to provide a more robust and reliable assessment of module health. This can identify degradation patterns and performance issues with better accuracy than single-data approaches, ultimately improving the sorting of EOL PV modules for reuse or recycling. The proposed framework can support circular economy objectives and sustainability in PV systems by reducing environmental impact and improving management of growing EOL module volumes.