Optimisation of panel component regions subject to hot stamping constraints using a novel deep-learning-based platform

• Published in: IOP conference series. Materials Science and Engineering Vol. 1270; no. 1; pp. 12123 - 12132
• Main Authors: Attar, H R, Foster, A, Li, N
• Format: Journal Article
• Language: English
• Published: Bristol IOP Publishing 01.12.2022
• Subjects: Cold working; Complexity; Deep learning; Design; Hot stamping; Neural networks; Optimization; Redesign; Stiffness; Thinning
• ISSN: 1757-8981; 1757-899X
• DOI: 10.1088/1757-899X/1270/1/012123

The latest hot stamping processes can enable efficient production of complex shaped panel components with high stiffness-to-weight ratios. However, structural redesign for these intricate processes can be challenging, because compared to cold forming, the non-isothermal and dynamic nature of these processes introduces complexity and unfamiliarity among industrial designers. In industrial practice, trial-and-error approaches are currently used to update non-feasible designs where complicated forming simulations are needed each time a design change is made. A superior approach to structural redesign for hot stamping processes is demonstrated in this paper which applies a novel deep-learning-based optimisation platform. The platform consists of the interaction between two neural networks: a generator that creates 3D panel component geometries and an evaluator that predicts their post-stamping thinning distributions. Guided by these distributions the geometry is iteratively updated by a gradient-based optimisation technique. In the application presented in this paper, panel component geometries are optimised to meet imposed constraints that are derived from post-stamping thinning distributions. In addition, a new methodology is applied to select arbitrary geometric regions that are to be fixed during the optimisation. Overall, it is demonstrated that the platform is capable of optimising selective regions of panel component subject to imposed post-stamped thinning distribution constraints.