MultiSolSegment: Multi-channel segmentation of overlapping features in electroluminescence images of photovoltaic cells

A research team from Sandia National Laboratories, Case Western Reserve University, and the University of Colorado Boulder has developed MultiSolSegment, a novel AI model that revolutionizes how solar panel defects are detected and analyzed. Published in Solar Energy, this multi-channel U-Net architecture addresses a critical limitation in existing machine learning methods for electroluminescence (EL) imaging: the inability to assign multiple labels to the same pixel. By outputting independent probability maps for four key features—cracks, busbars, dark areas, and non-cell regions—the model can accurately classify overlapping degradation features that commonly occur in real-world photovoltaic modules.

The model achieved an impressive 98% accuracy and demonstrated strong generalization capabilities when tested on unseen datasets, a crucial requirement for practical deployment in large-scale solar farms. This breakthrough enables precise quantification of interacting defects, such as cracks that intersect with busbars, which was previously challenging or impossible with single-label segmentation approaches. The framework provides solar energy operators with a scalable, extensible tool for automated inspection that goes beyond simple defect detection to enable more accurate lifetime predictions and maintenance planning.

By automating what has traditionally been a manual, time-intensive inspection process, MultiSolSegment represents a significant advancement in photovoltaic quality control and reliability assessment. The model's architecture is designed to be extensible, allowing for the addition of new defect categories as they're identified, making it a future-proof solution for the rapidly evolving solar energy industry. This technology could dramatically reduce inspection costs while improving the accuracy of performance predictions for solar installations worldwide.