A Novel End-to-End Solution for Solving Super Large Structural Simulation Problems

This paper was produced for the 2019 NAFEMS World Congress in Quebec Canada

Resource Abstract

The demand for solving super large structural simulation problems has been significantly increased in recent years. Ten years ago, it was a dream for finite-element analysis specialists to solve structural problems with the sizes of 50 million to 100 million degrees of freedom. Today, this is no longer a dream but has become a reality because of the evolutions of the software and its supporting hardware. This paper presents an end-to-end solution for solving super large structural simulation problems that can be more than 200M degrees of freedom. The ultimate goal is to reduce manual workloads of simplifying and abstracting geometrical details and to achieve an unprecedented high level of simulation fidelity and accuracy. This novel end-to-end solution covers modeling, solving, and result analysis, making “simulate as designed” possible.



The challenges involving solving super large problems come from different sides. It requires fully- or semi-automated robust meshers, an efficient linear equation solver with fast speed, lean memory management, and good scalability, and effective visualization tools for quickly processing large results data.



This paper presents an end-to-end high fidelity solution for solving super large structural simulation problems and demonstrates through three real-world industrial problems — sequentially coupled heat transfer, thermal stress and creep analyses of a turbine blade, structural analysis of an inner combustion engine assembly, and thermal stress analysis of ball-grid-array solder joints. Through these examples, the high fidelity simulation solution is demonstrated with the capabilities of high-performance modeling and meshing for generating high-quality detailed meshes, high-performance computing for solving systems of equations as large as 220 million degrees of freedom, and high-performance visualization for displaying and analyzing large simulation data of 10—50 GB. The performance results have shown that these capabilities provide an effective and holistic approach to solving super large structural simulation problems.