EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

ML‐Accelerated Framework for Fatigue Life‐Timescale Homogenization and Cyclic Damage Evolution Under Physics Constraints

Elsayed S. Elsayed, Abedulgader Baktheer, Fadi Aldakheel

International Journal for Numerical Methods in Engineering · 15.09.2026 · 127 (17)

DOI: 10.1002/nme.70394 ↗

Kurz gefasst

Fachartikel in der Zeitschrift International Journal for Numerical Methods in Engineering, online erschienen am 03.09.2026. Verfasst von Elsayed S. Elsayed und 2 weiteren Autor:innen. Die Arbeit wurde bisher 2-mal zitiert (Crossref) und nennt 104 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Predicting structural lifetime under complex loading remains a major challenge in computational mechanics, especially for high cycle fatigue in large scale systems such as aircraft, bridges, and wind turbines. This difficulty arises from the need to capture fatigue damage accumulation, crack initiation, crack growth across scales, and life prediction, making direct simulation over millions of load cycles computationally infeasible. To overcome this limitation, the presented work introduces a physics‐based machine learning () framework to accelerate fatigue life‐timescale homogenization and cyclic damage evolution. The first model predicts damage evolution over the full cyclic loading history, while the second operates at the lifetime scale by estimating accumulated damage over blocks of load cycles through the development of macro‐ and micromechanically motivated degradation mechanisms. Both models are built on double feed‐forward neural networks and are guided by physical constraints, including energy balance, damage evolution, and degradation laws, thereby ensuring physically consistent, interpretable, and generalizable predictions for unseen materials and loading conditions. Validation studies of the proposed framework demonstrate that the models accurately reproduce nonlinear damage evolution under diverse loading histories and material parameters. Most importantly, the lifetime‐scale model directly predicts damage over large blocks of load cycles, thereby eliminating the need for costly cycle‐by‐cycle simulations. Quantitative comparisons show computational speed‐ups of up to approximately 2700 times compared with conventional incremental fatigue analyses, while maintaining excellent agreement with reference solutions. As a result, this acceleration makes large‐scale high‐cycle fatigue simulations computationally feasible under physics constrains, providing a pathway toward scalable, reliable, and interpretable lifetime assessment of structural components subjected to cyclic loading.

Zitieren

APA

Elsayed, E. S., Baktheer, A., & Aldakheel, F. (2026). ML‐Accelerated Framework for Fatigue Life‐Timescale Homogenization and Cyclic Damage Evolution Under Physics Constraints. International Journal for Numerical Methods in Engineering, 127(17). https://doi.org/10.1002/nme.70394

Vancouver

Elsayed ES, Baktheer A, Aldakheel F. ML‐Accelerated Framework for Fatigue Life‐Timescale Homogenization and Cyclic Damage Evolution Under Physics Constraints. International Journal for Numerical Methods in Engineering. 2026;127(17). doi:10.1002/nme.70394

BibTeX
@article{Elsayed2026,
  author = {Elsayed, Elsayed S. and Baktheer, Abedulgader and Aldakheel, Fadi},
  title = {ML‐Accelerated Framework for Fatigue Life‐Timescale Homogenization and Cyclic Damage Evolution Under Physics Constraints},
  journal = {International Journal for Numerical Methods in Engineering},
  year = {2026},
  volume = {127},
  number = {17},
  doi = {10.1002/nme.70394}
}
Kostenlos starten →