Abstract:
This paper details a computational methodology for conducting reliable residual strength and statistical fatigue analysis of full-scale aircraft components, such as the C-130 center wing box, which require Finite Element models with up to a billion degrees of freedom (GDoF). The approach utilizes a highly optimized Finite Element software called STRIPE, which employs a mathematical multi-scale splitting scheme based on domain decomposition and sub-structuring to make these massive problems solvable. This technique breaks the single GDoF analysis into thousands of smaller, million-DoF local problems and one large global problem, which are then solved on high-performance computing systems like NAVO/BABBAGE and AFRL/HAWK using a custom, non-iterative direct solver designed to handle the tens of thousands of right-hand sides generated by the multiple crack and contact analyses. The paper discusses the challenges in achieving scalability with this method, particularly the massive I/O bottleneck inherent to direct solvers, and outlines the techniques used to mitigate it, including a mixed MPI/OpenMP implementation and optimized memory usage. The methodology is demonstrated by solving a generic wing model with 13.3 million hexahedral elements and 1.24 GDoF, representing the largest known structural analysis of its kind and enabling, for the first time, detailed statistical analysis that accounts for uncertainties in geometry, materials, and crack locations.
