.png){kind=icon}
Return
Deep Learning Platform for Optimizing Forming Tool Design
.png){kind=icon}
PRE
AI
DOI：10.1007/s12666-026-03925-7.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Stamping process is widely used in automotive industries to manufacture complex shapes. A proper die design is required to ensure quality of the stamped parts. It is common to use Computer Aided Design (CAD) and Finite Element Analysis (FEA) to modify die design for achieving defects free components production. This iterative process is costly and time consuming. Deep learning has emerged as a powerful tool for assisting the tool/die design by reducing computational and physical effort needed for hundreds of sequential simulations. However, achieving both computational efficiency and high predictive accuracy, particularly in geometrically intricate regions, remains a challenge. This paper enhances the previously introduced deep learning platform for sheet metal forming tool optimization by incorporating curvature-based data sampling technique that prioritizes data acquisition from critical regions of the die or punch surfaces. The platform has two neural networks: 1) geometry generator to represent tool shape and 2) surrogate model trained on simulation data to assess forming feasibility. Generator model uses latent vector for geometry representation and is trained on Signed Distance Function (SDF). Marching cube is used to generate tool shape from the SDF values. Surrogate model takes height map from the generated shape and predict the manufacturing performance. Gradient-based optimization is employed to get desired manufacturing performance after both models are trained, by changing only the latent vector of generator model. This optimized latent vector gives us the final tool shape. This paper presents the platform, framework followed and the optimization results for defect-free forming of cross die.
Keywords:
Sheet metal forming
Deep learning
FEM
Optimization
Curvature-based data sampling
Journal
IF:
1.6
Papers:
209
Citations:
4.7K
Organization
Cited Papers
No cited papers available