Scalable computation of thermomechanical turbomachinery problems

TL;DR

This paper demonstrates that finite element methods can efficiently handle extremely large-scale thermomechanical simulations in turbomachinery, achieving near linear scalability with billions of degrees of freedom.

Contribution

It presents the technology and performance data for large-scale finite element simulations with up to 3.3 billion degrees of freedom, challenging the notion that these methods are unsuitable for such problems.

Findings

Successful simulation of problems with up to 3.3 billion degrees of freedom

Achieved near linear computational complexity and good parallel scaling

Open source software libraries enable large-scale thermomechanical modeling

Abstract

A commonly held view in the turbomachinery community is that finite element methods are not well-suited for very large-scale thermomechanical simulations. We seek to dispel this notion by presenting performance data for a collection of realistic, large-scale thermomechanical simulations. We describe the necessary technology to compute problems with $O(10^7)$ to $O(10^9)$ degrees-of-freedom, and emphasise what is required to achieve near linear computational complexity with good parallel scaling. Performance data is presented for turbomachinery components with up to 3.3 billion degrees-of-freedom. The software libraries used to perform the simulations are freely available under open source licenses. The performance demonstrated in this work opens up the possibility of system-level thermomechanical modelling, and lays the foundation for further research into high-performance formulations for even larger problems and for other physical processes, such as contact, that are important in turbomachinery analysis.