Rapid Aerodynamic Analysis through Deep Learning: FY23 Engineering Research Technical Investment Program

Abstract

MIT Lincoln Laboratory designs and fabricates airborne sensors to solve problems across a range of disciplines. These sensors are flown on Lincoln Laboratory Flight Test Facility aircraft. In order to certify airworthiness of an aircraft modification before it is flown, substantial analysis is performed. Of interest for this work is the aerodynamic performance of modifications made to the aircraft. Graph Neural Network, is performed in order to estimate quantities of interest for a given aircraft configuration. We developed a tool that predicts aerodynamic properties of arbitrary geometries either directly or by first generating a flowfield that can be post-processed for quantities of interest. After training the model, these predictions are conducted in a matter of seconds with the direct-prediction model demonstrating 10 percent to 14 percent mean absolute error for real-world aircraft that were not included in the training data. Predictions are fundamentally limited by the diversity of the training data and the quality of predictions is inherently bound to the quality of that data. We sought to improve realism in predictions through the incorporation of a physics-informed neural network; however, the present implementation does not show major improvement. We believe that the formulation of the network should be rewritten as a Graph Neural Network, so that the model can better learn the intricacies of wall-bounded shear flows by having more detail in the regions of interest. This in turn can help to provide better predictions of the flowfield and, subsequently, better airworthiness assessments.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Jan 24, 2024
Accession Number
AD1220012

Entities

People

  • Cameron Cubra
  • James Crouse
  • Matthew C. Jones
  • Steven H. Spreizer

Organizations

  • Massachusetts Institute of Technology

Tags

Fields of Study

  • Physics

Readers

  • Aerodynamics/Aeronautics.
  • Computational Modeling and Simulation
  • Neural Network Machine Learning.

Technology Areas

  • AI & ML
  • AI & ML - Bayesian Inference
  • AI & ML - Neural Networks