PV-LLM is a novel framework that utilizes multimodal large language models (LLMs) to perform advanced analysis in photovoltaics (PV), including degradation assessment and health monitoring. This project demonstrates how AI can accelerate PV system diagnostics and knowledge extraction from global data sources.
Photovoltaic modules experience various types of degradation and faults during operation. Traditional inspection techniques require significant manual effort and expert interpretation. PV-LLM automates this by using multimodal LLMs to analyze different types of PV images:
- Visibleimages (birddropping)
- Infrared (IR) images (thermal anomalies, like hotspot)
- Electroluminescence (EL) images (microcracks, broken cells)
The system can:
- Classify fault types
- Localize issues
- Describe fault severity in natural language
Figure: Multimodal LLM-based PV Image Analysis Flowchart
PV-LLM also performs large-scale literature analysis using LLMs. The system can:
- Read scientific papers
- Identify and extract degradation rates, technologies, and climates
- Compile global trends in PV degradation
This enables the creation of a unified degradation knowledge base across:
- Regions (e.g., US, EU, Asia)
- Technologies (e.g., mono, poly, thin-film)
- Timescales
Figure: Global PV Degradation Trends Extracted from Literature
This work is supported by the Durable Module Materials (DuraMAT) Consortium, a U.S. Department of Energy initiative focused on PV module reliability and innovation.
For questions or collaboration opportunities, please contact the PV-LLM development team: baojieli@lbl.gov.