Segmenting unmanufacturable regions in CAD designs through deep learning and domain adaptation
- Author
- Toon Van Camp (UGent) and Dries Benoit (UGent)
- Organization
- Abstract
- Design for Manufacturing (DfM) systems aim to provide feedback on manufacturability to avoid costly redesigns. Existing studies use deep learning models to classify whether a design is manufacturable, and apply post-hoc XAI techniques to localize issues. We propose a graph neural network that directly segments unmanufacturable regions, trained on synthetic data. To ensure effective domain adaptation to real-world designs, we employ a contrastive learning framework that maps synthetic and limited real examples in a shared latent space. By training only on synthetic samples closest to real-world counterparts, this novel method bridges the gap between synthetic data and real-world applicability.
Citation
Please use this url to cite or link to this publication: http://hdl.handle.net/1854/LU-01KFVDN6A5XG0T8K2RJ0VVMC4Z
- MLA
- Van Camp, Toon, and Dries Benoit. “Segmenting Unmanufacturable Regions in CAD Designs through Deep Learning and Domain Adaptation.” 56th Annual Conference of the Decision Sciences Institute, Abstracts, 2025.
- APA
- Van Camp, T., & Benoit, D. (2025). Segmenting unmanufacturable regions in CAD designs through deep learning and domain adaptation. 56th Annual Conference of the Decision Sciences Institute, Abstracts. Presented at the 56th Annual Conference of the Decision Sciences Institute (DSI 2025), Orlando, USA.
- Chicago author-date
- Van Camp, Toon, and Dries Benoit. 2025. “Segmenting Unmanufacturable Regions in CAD Designs through Deep Learning and Domain Adaptation.” In 56th Annual Conference of the Decision Sciences Institute, Abstracts.
- Chicago author-date (all authors)
- Van Camp, Toon, and Dries Benoit. 2025. “Segmenting Unmanufacturable Regions in CAD Designs through Deep Learning and Domain Adaptation.” In 56th Annual Conference of the Decision Sciences Institute, Abstracts.
- Vancouver
- 1.Van Camp T, Benoit D. Segmenting unmanufacturable regions in CAD designs through deep learning and domain adaptation. In: 56th Annual Conference of the Decision Sciences Institute, Abstracts. 2025.
- IEEE
- [1]T. Van Camp and D. Benoit, “Segmenting unmanufacturable regions in CAD designs through deep learning and domain adaptation,” in 56th Annual Conference of the Decision Sciences Institute, Abstracts, Orlando, USA, 2025.
@inproceedings{01KFVDN6A5XG0T8K2RJ0VVMC4Z,
abstract = {{Design for Manufacturing (DfM) systems aim to provide feedback on manufacturability to avoid costly redesigns. Existing studies use deep learning models to classify whether a design is manufacturable, and apply post-hoc XAI techniques to localize issues. We propose a graph neural network that directly segments unmanufacturable regions, trained on synthetic data. To ensure effective domain adaptation to real-world designs, we employ a contrastive learning framework that maps synthetic and limited real examples in a shared latent space. By training only on synthetic samples closest to real-world counterparts, this novel method bridges the gap between synthetic data and real-world applicability.}},
author = {{Van Camp, Toon and Benoit, Dries}},
booktitle = {{56th Annual Conference of the Decision Sciences Institute, Abstracts}},
language = {{eng}},
location = {{Orlando, USA}},
title = {{Segmenting unmanufacturable regions in CAD designs through deep learning and domain adaptation}},
year = {{2025}},
}