| Beschreibung |
The fatigue life of an additively manufactured metal part produced by PBF-LB/M mainly depends on the roughness of the outer surface. It has been shown that capturing the surface exactly by means of surface scanning can predict the failure location and has been validated by cyclic testing of specimen. The results are assembled into an S-N curve to predict the fatigue life of future parts. However, a robust prediction of fatigue life depends on hundreds of parts being manufactured and tested in different load regions. Often this is unfeasible due to the large number of required specimens. This is where numerical testing can assist cutting down on the number of required experiments.To this end, the generation of artificial surfaces to enhance the experiments that consider the characteristics of existing surface scans, is a promising approach. Each student is assigned one family of deep generative models (e.g. variational autoencoder, diffusion model, generative adversarial network) and works as a team member to answer the following questions: - How big and what characteristics do the surface features have? How can these characteristics be described?
- How realistic are the generated surfaces, and how can this be measured?
- Which model family reproduces the variation of the real surfaces best, without copying the training data?
- Can the generation of 1-D profiles be expanded to higher dimensions?
Over twelve weeks each student first delves into the theory of their model family on a toy example, programming it from scratch before relying on libraries. The students will then be given 1-D profiles of scanned surfaces, analyse them statistically, generate new ones, and measure how realistic their generations are. Finally, the students investigate how the generation can be expanded to higher dimensions. Deliverables are reproducible code and a final presentation. |
| Literatur |
[1] M. Horstjan et al., ”Scan-based fatigue prediction of DED-Arc additively manufactured steel components,” Additive Manufacturing, 125, 105243, 2026. [2] L. Herrmann, M. Jokeit, O. Weeger, and S. Kollmannsberger, Deep Learning in Computational Mechanics: An Introductory Course. Springer, 2025. [3] E. Hüllermeier and W. Waegeman, ”Aleatoric and epistemic uncertainty in machine learning: an introduction to concepts and methods,” Machine Learning, 110, 457–506, 2021. [4] C. E. Rasmussen and C. K. I. Williams, Gaussian Processes for Machine Learning. MIT Press, 2006. |
