Research

CFDG is currently pursuing research in a number of areas, as described below.

Mesh Generation and Adaptation

Mesh generation and adaptation continue to be significant bottlenecks in the CFD workflow, and this becomes more problematic as increasing computational power allows us to run higher resolution simulations. In fact, mesh generation is often the most time-consuming and user-intensive task in setting up CFD simulations. Furthermore, even when adaptive mesh refinement strategies offer significant benefits, they are not widely used in production-level CFD due to lack of robustness or software complexity issues. Some research topics of interest in this area are as follows:

• • Mesh anisotropy detection

  • Robust mesh adaptation strategies for complex geometries

  • New mesh refinement techniques to further improve performance of high-order finite element methods

  • Automated adaptive mesh generation techniques

Error Estimation

Fast, reliable mesh generation and adaptivity is not possible without accurate error estimation capabilities. However, current error estimation techniques are proving inadequate for increasingly many applications. This is especially true for unsteady flows, which may become chaotic. Research topics in this area includes:

• • Error estimation using adjoint-based output error calculations

  • Unsteady output-based error estimation

Quantifying Uncertainty and Improving Robustness

Errors and uncertainties in current CFD simulations are not always well-understood or well-quantified. Sources of errors include temporal and spatial discretization errors and incomplete convergence, while uncertainties include modeling errors based on simplifying assumptions or unknown parameters. The lack of error quantification raises the risk that engineering decisions are based on inaccurate and/or uncertain results. Moreover, applying CFD to ever-more challenging simulations comes with a new liability: ensuring that the computed solutions are sufficiently accurate. Methods to handle these problems are being tackled, such as

• • Quantifying and reducing numerical errors arising from the discretization of the governing equations

  • Developing stochastic-space adaptive methods for uncertainty quantification

  • Using probabilistic approaches to solve large-scale uncertainty quantification and inverse problems

High Performance Computing (HPC)

The development of CFD algorithms cannot be separated from the development of algorithms that effectively use HPC resources. As HPC hardware evolves rapidly, algorithmic developments that will enable us to exploit emerging hardware capabilities become paramount. Research topics in this area include:

• • Designing robust and scalable solvers that can take advantage of modern HPC architectures, including the anticipation of exascale computing.

  • Developing massively parallel CFD algorithms that balance computation and communication cost

  • Increasing automation in the CFD workflow to efficiently use computational power