Speaker
Description
Traditional tool development processes in mechanical forming, especially for pressing tools, are often limited by time-consuming trial-and-error cycles that rely heavily on expert knowledge. These cycles can create bottlenecks in production and increase costs. Machine learning can be imployed here to greatly speed up these processes by using advanced models. In this regard, a Denoising Diffusion Model (DDM) has been proposed and employed in the previous work to inversely model effective tool surfaces from final product geometries [1]. This model leverages spatial and temporal attention mechanisms to enhance its predictive capabilities. While DDM achieved commendable overall predictions, it struggled with accuracy in reconstructing individual denoised frames during later forming steps. To address these challenges without significantly increasing computational demands, we explore integrating Pixel Adaptive Convolutional Neural Networks (PAC) [2]. PAC refines predictions from pre-trained models using high-resolution guidance images and adaptive kernels, enhancing both final results and intermediate frame accuracy. Additionally, the refinement could also be enhanced by incorporating reliability estimates for each pixel prediction. PAC has already been demonstrated in benchmark tasks like optical flow and semantic segmentation and thus aligns well with the forming process simulation by capturing motion and deformation over time. By incorporating these techniques, we aim to refine stress distributions and enhance the overall prediction reliability of the DDM, ultimately supporting a more efficient and accurate tool development process.
[1] Hupfeld, H. K., Teshima, Y., Ali, S. S., Dröder, K., Herrmann, C., \& Hürkamp, A. (2024). Accelerating the design of the effective surface of pressing tools with probabilistic inverse modeling approaches. Proceedings in Applied Mathematics and Mechanics, 24, e202400177.
[2] Su, H., Jampani, V., Sun, D., Gallo, O., Learned-Miller, E.G., \& Kautz, J. (2019). Pixel-Adaptive Convolutional Neural Networks. 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 11158-11167.
