Efficient Bridging Temporal Scales for Fatigue Damage Simulations via Compressive Sensing Approach

Fatigue damage simulations based on physical models and numerical simulation, e.g. via Finite Element Methods (FEM) remains a challenging task. Major issues are the computational effort required for the non-linear computation of only one time instant with FEM. For high-cycle fatigue (HCF) simulations about 10⁹ cycles have to be computed. And, following the well-established Shannon-Nyquist criteria, a cyclic load sequence has to be treated corresponding the highest excitation frequency of interest. Besides own previous experience to circumvent that issue via Latin-PGD Method and jump-cycling, we believe that there is still open space for research.

The paradigm of compressive sensing (CS), see promises to circumvent that by reconstructing information from sparse or missing information. Thus, goal of this research project is to investigate, to what content CS could speed up fatigue simulation based on physical models. In contrast to own prior work, here we intend to switch from intrusive Latin-PGD to non-intrusive POD-RB methods, whose investigations in own prior work also shows potential for this application. Besides only construction of efficient ROM for HCF-simulation for only one time instant, the capacity of fast predictions in time domain using sparse information on each time step will be investigated.

Doctoral Researcher: Jannik Jarms

Scientific Advisors:  Udo Nackenhorst