Addressing Discrepancies between Synthetic and Real-World Data of Deep-Drawing Manufacturing Systems in AI Applications

While simulated data has facilitated substantial advancements in AI applications for sheet metal manufacturing systems, the direct transfer of these achievements into real-world scenarios remains challenging. Simulated environments offer high-fidelity data, yet fail to encompass the complexities and limitations inherent in real-world settings. The differences in data characteristics, such as resolution discrepancies in 3D meshes, pose hurdles when models trained on synthetic data face data of real-world manufacturing scenarios. This paper explores the divergence between simulated and real-world data in the context of deep drawing with focus on preprocessing such data, highlighting the necessity to address this gap for a proficient AI model architecture, capable of predicting the springback effect.