Back to search

West Virginia University

Physics-Informed Neural Network Based Aerodynamic Modeling Framework

Abstract

dc:description.abstract

Significant advances have been made in developing aerodynamic models over the years. While these advances span many modeling techniques and data collection methods, certain aerodynamic regimes still pose significant challenges. These regimes commonly occur when an aerodynamic body is under extreme flight conditions. Many of these conditions occur simultaneously in the rare and poorly understood case of a tumbling aerodynamic body. The flight of a tumbling aerodynamic body goes through the entire range of aerodynamic angles, alpha and beta, causing significant non-linearities and time-dependent effects associated with flow separation. During tumbling, the aerodynamic body simultaneously rotates about all three axes at high rotational rates, introducing additional non-linearities. This work, motivated by a need to understand the underlying dynamics of tumbling aerodynamic bodies, develops a Physics-Informed Neural Network Aerodynamic Modeling Framework (PINN-AMF) purposely developed to improve aerodynamic models in challenging regimes. PINNs are built on the concept of introducing physical knowledge into the training process of neural networks while benefiting from their universal approximation capabilities. At the center of the PINNAMF are the modifications made to the original PINN architecture to ensure the technique is suitable for aerodynamic modeling. These modifications allowed PINNs to be used on unbounded flight data for the application of aerodynamic modeling for the first time and were demonstrated on increasingly complex simulated case studies. The remaining components of the PINN-AMF consist of the data processing, flight propagation, and uncertainty quantification techniques. The primary purpose of any aerodynamic model is its use to propagate the motion of an aerodynamic body in flight. Therefore, the aerodynamic models can easily be integrated into a flight propagator to provide trajectory predictions. Small errors in the aerodynamic model can lead to significant errors in the propagated trajectories. Therefore, ensemble-based uncertainty quantification weighting schemes were developed to provide calibrated uncertainty estimates. For all techniques, accurate modeling is dependent on good data. Historically, data for aerodynamic modeling has come from multiple sources, including flight testing, wind tunnel testing, and computational fluid dynamics. While the primary data source for the PINN-AMF is assumed to be flight test data, techniques have been developed to allow multiple data sources to be used in the modeling process. This work covers the entire development of the PINN-AMF while demonstrating its applicability on both simulated case studies and real flight test data of a tumbling aerodynamic body.

Degree

thesis:*
Name thesis:degree_name
PhD
Level thesis:degree_level
Dissertation
Discipline thesis:degree_discipline
Mechanical and Aerospace Engineering

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Michek, Nathaniel E
Contributors dc:contributor
  • Piyush Mehta
  • Wade Huebsch
  • Christopher Griffin
  • Jason Gross
  • Alric Rothmeyer

Subjects

dc:subject × 7

Identifiers

dc:identifier.*
Identifier
<p><a href="https://orcid.org/0000-0003-0091-9802">https://orcid.org/0000-0003-0091-9802</a></p>
https://researchrepository.wvu.edu/etd/12843

Chain of custody

source
Harvested from
West Virginia University
Base URL
researchrepository.wvu.edu/do/oai/
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
citation

Michek, Nathaniel E. Physics-Informed Neural Network Based Aerodynamic Modeling Framework. Dissertation thesis, https://doi.org/10.33915/etd.12843