PVCracks is the DuraMAT project that investigates the effects of cracks on power loss in photovoltaic (PV) solar cells and tracks crack progression over time. We provide:
- Open-source cell-level imaging and electrical datasets
- MultiSolSegment: segmentation of cracks, busbars, and dark areas in electroluminescence (EL) images
- Variational Autoencoder (VAE): parameterization & clustering of segmented data
- XGBoost model: estimation of power loss (ΔPMPP) per cell
- pvspice_lite: spice toolset for simulating current-voltage (I-V) curves and building synthetic module I-Vs (more features to come)
Full online documentation is available at
https://pvcracks.readthedocs.io/en/latest/index.html#
Below is a schematic of the repo’s architecture:
All datasets, trained model weights, and additional resources are hosted on DuraMAT DataHub:
https://datahub.duramat.org/project/pv-crack-stress-and-power-loss
- Cell-level EL & electrical data (data publication to follow)
- MultiSolSegment training images and masks (DOI: 10.21948/2587738)
- MultiSolSegment model weights (DOI: 10.21948/2997859)
- VAE model weights (DOI: 10.21948/2997860)
-
MultiSolSegment
In revision for Renewable Energies (Elsevier). Link to follow. -
Variational Autoencoder (VAE)
EUPVSEC 2024 conference proceeding
DOI: 10.4229/EUPVSEC2024/3BO.15.6 -
Power-Loss Model (XGBoost)
2025 IEEE 53rd Photovoltaic Specialists Conference (PVSC) proceeding
DOI: 10.1109/PVSC59419.2025.11132966
- Sep. 30th 2025: Please attend our webinar on Nov. 10th. Sign up here:
www.duramat.org/news-and-events/webinars
Clone and install:
git clone git@github.com:sandialabs/pvcracks.git
cd pvcracks
# 1. Upgrade pip, setuptools, and wheel
pip install --upgrade pip setuptools wheel
# 2. Install runtime dependencies (see requirements.txt for full list)
pip install -r requirements.txt
# 3. Install pvcracks in editable mode
pip install -e . # enables intra-project imports