This presentation was held at the 2020 NAFEMS UK Conference "Inspiring Innovation through Engineering Simulation". The conference covered topics ranging from traditional FEA and CFD, to new and emerging areas including artificial intelligence, machine learning and EDA.

Resource Abstract

Simulation and modelling, enabled by high performance computing, have transformed the way aero-engines are designed and engineered. However, next generation engines will place demands on simulation that cannot be met by incremental changes to current techniques. There is a strong industry need to deliver efficient engines to the aerospace market in significantly reduced timescales, but with a much higher level of maturity. High Performance Computing (HPC) offers a means of achieving this by providing fast, accurate, simulation of operational behaviour at all stages of a products lifecycle. This includes the accurate prediction of blade tip and seal clearances using very large, high fidelity, coupled thermo-mechanical models at whole engine level. Under the EPSRC funded ASiMoV project, Rolls Royce Plc is working with a number of academic and industrial partners, to develop an advanced, large scale multi-physics simulation system capable of running multi-billion DoF models on HPC systems.

This paper describes the complex process of model generation, meshing, application of complex boundary conditions and creation of complex flight profiles at whole engine level. The paper also covers the development of a Message Passing Interface (MPI) based distributed parallel iterative finite element solver, implemented on the public domain FEniCS code and an in-house finite element system. The three benchmarks under consideration are a large component model, a whole engine level static structure model and finally a large high fidelity model that is a close representation of a production aero engine. The largest model consists of six thousand components with complex thermal boundary conditions, structural constraints and loads. The loading conditions are representative of complex flight cycles. Multi-billion element meshes have been generated with advanced mesh generation processes. This results in the generation of large finite element binary input data for these models. To handle the large amounts of input data, mesh related data has been separated from data related to boundary conditions and loadings, which are linked to the external geometry. These geometry faces are tagged to finite element faces and this information is stored in an HDF5 formatted mesh file.

The major challenge in developing scalable speed and accuracy is the development of solver technology, particularly iterative solvers for implicit coupled thermo-mechanical finite element analysis. After briefly presenting the algorithmic aspects of the solver, this paper will focus on the performance aspects of the benchmark models under various solver parameters and pre-conditioners. In order to improve the scalability of the solver, codes are parallelised throughout the analysis, from reading input HDF5 files to writing results data after the parallel iterative solve. The input mesh data on multiple MPI processes have been partitioned using parallel partitioning algorithms and distributed to multiple processes. In the transient time stepping process, parts of large linear system are assembled from evaluated element matrices in corresponding MPI processes and solved using a pre-conditioned iterative solver. The iterative solution process is based on the efficient usage of the public domain Portable, Extensible Toolkit for Scientic Computation (PETSc) and related codes. For each test case, the results, especially deflections, are compared between the in-house system and the FEniCS code.

In the future, the focus will be to develop an advanced process to build large whole engine models representing better physics, faster generation of large meshes and implementation of advanced algorithms to improve parallel scalability in solution process in HPC systems.