A Data-Driven Surrogate Model for Predicting Springback in Deep Drawing Processes

Abstract

The design of deep drawing processes for high-quality sheet metal forming products requires substantial expertise in product design, process optimisation, and equipment selection. Here, a special focus lies on predicting the material behaviour during the forming process, particularly the springback behaviour after deep drawing, which remains a major challenge due to stochastic process deviations. Currently, the prediction of springback is associated with complex and time-consuming simulations. Hence, in this context, this paper proposes a data-driven surrogate model utilising simulation data from synthetic finite element analysis (FEA) to compensate for the computational effort in predictive manufacturing. The simulation contains surface data on the deep-drawing and trimming process for both the loaded and unloaded part. This provides necessary information on the springback of the part. By representing simulation data through graph-based representations with multidimensional edge features, this model allows complex relationships between part geometry, process parameters and springback effects to be identified without relying on restrictive domain assumptions. The proposed methodology demonstrates significant potential for reducing simulation-based prediction times and enhancing deep-drawing tool design and optimisation. Furthermore, the modelling approach incorporates differentiable properties, enabling the application of gradient-based optimisation techniques. This facilitates the precise prediction of springback effects.