Data Analytics and Machine Learning for Exascale Computational Fluid Dynamics

PI Ken Jansen, University of Colorado Boulder
Jansen Aurora ESP

Flow over a vertical tail/rudder assembly (grey surface) with 24 active synthetic jets which introduce vortical structures (visualized by isosurfaces of vorticity colored by local flow speed) that alter the flow, reducing separation, improving rudder performance, and thereby allowing future aircraft designs to lower drag and fuel consumption. Image: Jun Fang, Argonne National Laboratory

Project Summary

This project will develop data analytics and machine learning techniques to greatly enhance the value of flow simulations with the extraction of meaningful dynamics information. A hierarchy of turbulence models will be applied to a series of increasingly complex flows before culminating in the first flight-scale design optimization of active flow control on an aircraft’s vertical tail.

Project Description

This project seeks to accomplish four new goals:

  • Advance our ability to extract meaningful insight into the dynamics from exascale computational fluid dynamics (CFD) simulations of complex flows through data compression in both space and time;
  • Advance machine learning algorithms to mine the data from exascale simulations for turbulence model improvements to benefit lower-fidelity (and therefore less computationally intensive) modeling capability;
  • Extend existing work in uncertainty quantification and multi-fidelity modeling to exascale; and
  • Extend these new components into our growing stable of in situ data analytics. Collectively, these objectives also benefit other areas of partial differential equation simulation through a dramatic advance in computational modeling and associated scientific and engineering insight.

This work creates a series of stepping-stone simulations that can be carried out alongside the four new objectives so as to provide testing and assessment at each stage. These stepping stones will be aligned to reach the aerodynamic flow control task which has built upon previous ESP and INCITE efforts in which computational models of a vertical tail and rudder assembly with 12 synthetic jets were validated against experiments at a Reynolds number 53 times smaller than flight conditions.

These simulations will be the first design optimization under uncertainty quantification of a full-scale aerodynamic component using detached eddy simulation. The vertical tail is sized to handle an engine-out condition which requires it to be much larger than what is needed for all other conditions in the flight envelope. The economic impact is directly related to the size of the stabilizer since it is a significant contributor to drag in cruise where much of the fuel is expended. If flow control can achieve the same side force with a vertical tail 25% smaller than current sizes, then the jet fuel usage can be substantially reduced.


Project Type