StampFormer: a physics-guided material-geometry-coupled multimodal model for rapid prediction of physical fields in sheet metal stamping
File(s)
Published version
Author(s)
Type
Journal Article
Abstract
Traditional sheet metal forming relies on computationally expensive Finite Element Analysis (FEA), significantly prolonging design cycles. While surrogate models accelerate iteration, existing approaches either provide limited scalar outputs or ignore crucial material properties by focusing solely on geometry. To address this gap, we develop StampFormer, a physics-guided deep learning framework that simultaneously processes component geometry and material stress-strain responses to predict FEA outcomes. Its architecture employs a Material-Augmented Geometric Network (MAGN) for initial multimodal fusion, followed by a Hierarchical Material Embedding Injection Unit (HMEIU) that progressively integrates material features into an adapted Swin-UNet backbone. Evaluated on steel and aluminium crossmember stamping datasets, StampFormer predicts critical physical fields—including thinning, major/minor strain, plastic strain, and displacement—in under a second. Compared with the reference FEA, our model achieves a mean representative-maximum relative error below for each of the four 2D fields in both material datasets and a masked displacement mean squared error below mm2. Within the investigated component family and fixed process conditions, this multimodal framework provides sub-second predictions for rapid early-stage material–geometry screening and design exploration.
Date Issued
2026-12-01
Date Acceptance
2026-10-17
Citation
Materials & Design, 2026, 272
ISSN
0264-1275
Publisher
Elsevier
Journal / Book Title
Materials & Design
Volume
272
Copyright Statement
© 2026 Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
Publication Status
Published
Article Number
117298
Date Publish Online
2026-10-21