Reliable CFD and Simple Computational Grids
Massachusetts Institute of Technology
Computational Fluid Dynamics (CFD) is a technology that supports innovation in a number of industries. Aerospace, automotive, energy, construction, and biomedical are just some of the industries that benefit from detailed simulations of complex flows. However, despite its widespread use, CFD is still restricted to a relatively small number of engineers. In this presentation I will talk about two fronts of research in which I am engaged to make CFD more accessible: (i) data-driven models, and (ii) embedded grid methods.
Data-driven models can improve the predictive capability of CFD, and also quantify uncertainties due to model inadequacy. Physical formulations used in CFD are often based on reduced versions of the flow dynamics. In my research I use recent advances in computer and information science and machine learning to create data-driven models that complement these reduced physical formulations. The combination of predictive models and quantification of model inadequacy leads to CFD results that are reliable, and that can be interpreted with confidence by users that are not CFD experts. I will discuss examples of a data-driven model for the boundary layer in the laminar and incompressible regime, and a data-driven wall model for large-eddy simulation.
Embedded grid methods simplify the tasks of grid generation and simulation of moving boundaries. Most CFD tools are based on body-fitted grids, i.e., grids that conform to the geometry of the boundaries. Generating body-fitted grids around complex geometries is a labor-intensive and difficult task. In contrast, in embedded grid methods the boundaries are embedded in a relatively simple computational grid and the boundary conditions are enforced through modifications to the discrete equations. As a consequence, embedded grid methods are also robust to simulate problems with moving boundaries, as is the case in flapping wings, multi-phase flows, and flows around flexible membranes. In particular, I will present the Correction Function Method (CFM), which is an embedded grid method capable of producing accurate solutions up to the boundaries.
Monday, March 6, 2017
Hughes Aircraft Electrical Engineering Center, Rm 132 (EEB 132)