Thesis
Fluid thermal structural predictive modeling of a high-speed airframe using machine learning
University of Iowa
Master of Science (MS), University of Iowa
Spring 2026
DOI: 10.25820/etd.008520
Abstract
High-speed flight is on the bleeding edge of current aerospace research. Moving at such high-speeds introduces an airframe to complex, turbulent flow conditions from shock regions that form around the body throughout a given trajectory. In high-speed travel, the thermal and structural responses of a given airframe to the fluid surrounding it are deeply coupled. Material properties such as the thermal expansion coefficient or conductivity for a given airframe are subject to change as heat increases, which in turn amplifies the structural response as it bends and deforms under load. In the early design phase of a high-speed airframe, it is vital for engineers to analyze and account for these affects. Failing to consider extreme thermal and mechanical loading conditions during high-speed travel could lead to, at minimum, an inefficient, poorly constructed vehicle, or at worst, catastrophic vehicle failure. However, current modeling techniques for coupled thermomechanical high-speed analysis are cumbersome, slow, and not easily generalizable for quick changes to the input parameters without requiring significant computational time.
Current analysis of the fluid-thermal-structural response for high-speed flow for an airframe utilizes Computational Fluid Dynamics (CFD) software to analyze the fluid behavior of the environment surrounding the body. Conventional analysis of the thermal and structural response of the airframe is carried out with Finite Element Analysis (FEA) software. Completing CFD analysis for high-speed flow and then coupling with the FEA software is slow, and can only be performed in series; a parameter change in the trajectory requires a complete rerun of CFD simulation, followed by accompanying thermal and structural FEA simulations. This process can take hours or even days to complete. The computational time for current CFD and FEA software presents a glaring issue as it pertains to the necessary work required for the design optimization of a high-speed airframe. This work presents an approach to address this issue.
The main objective of this study is to evaluate the effectiveness of using machine learning frameworks in reducing the computational overhead of CFD and FEA for high-speed airframe analysis while maintaining the fidelity of these simulations. This work develops a machine learning-based surrogate modeling framework to approximate the coupled fluid-thermal-structural analysis currently completed via CFD and FEA software, allowing for greater optimization of the input space. This objective is accomplished in four parts: (1) model high-speed flow using CFD, (2) model thermal response to CFD input of airframe using FEA, (3) model structural response to fluid and thermal loads using FEA, (4) create a machine learning surrogate framework to estimate resulting fluid, thermal and structural response using data obtained in parts (1)-(3). All CFD results obtained during this study were then approximated using a machine learning model (MLM), allowing for the functionality and accuracy of CFD simulations to be maintained while reducing the run time dramatically. The thermal response to several unique trajectories created by the CFD MLM were then analyzed via FEA. After a sufficient amount of data was collected from the thermal FEA portion of the analysis, another separate MLM was created to approximate the generalized thermal response of an airframe during high-speed flight. In a similar manner to the thermal MLM, structural response data to both thermal and structural loading was collected over several trajectories before a structural MLM was created to estimate the structural response an airframe in a high-speed trajectory via machine learning. The validity of each respective MLM created in this work was ensured by maintaining an error margin with $5\%$ or lower of the total range of the given data range. By estimating the fluid and thermal response during a high-speed trajectory using a machine learning framework, computational time for a single trajectory is reduced from days to minutes. However, further work is required to improve the accuracy of the structural MLM to fully capture the complexity of the coupled thermal and structural response of the airframe. With these modeling capabilities, design engineers will gain the capacity to optimize airframe in minutes as opposed to weeks or hours using conventional CFD and FEA techniques. In turn, this will allow for the construction of efficient, durable, effective high-speed airframes capable of traveling at the extreme boundaries of the sonic envelope.
Details
- Title: Subtitle
- Fluid thermal structural predictive modeling of a high-speed airframe using machine learning
- Creators
- Mitchell Harris Leschensky
- Contributors
- Phillip Deierling (Advisor)Albert Ratner (Committee Member)Shaoping Xiao (Committee Member)
- Resource Type
- Thesis
- Degree Awarded
- Master of Science (MS), University of Iowa
- Degree in
- Mechanical Engineering
- Date degree season
- Spring 2026
- DOI
- 10.25820/etd.008520
- Publisher
- University of Iowa
- Number of pages
- xiii, 96 pages
- Copyright
- Copyright 2026 Mitchell Harris Leschensky
- Language
- English
- Date submitted
- 04/27/2026
- Description illustrations
- Illustrations, graphs, charts, tables
- Description bibliographic
- Includes bibliographical references (pages 87-91).
- Public Abstract (ETD)
Aircraft operating at extreme speeds are subject to complex thermal and structural loading. In order to make these aircraft suitable for high speeds, a considerable amount of design analysis is required to ensure success despite these conditions. Engineers generally rely on software tools such as computational fluid dynamics (CFD) and finite element analysis (FEA) to model the fluid, thermal, and structural behaviors of aircraft during high-speed flight. However, these current analysis tools are slow and inefficient, limiting the early design process. A solution is provided in this work through the use of machine learning. For this work, a characteristic trajectory is proposed for the airframe, and subsequently analyzed using conventional CFD and FEA software. Using large sets of CFD and FEA data computed on a high-speed airframe, a machine learning model was then trained to predict the thermal and structural response of the airframe as it interacts with the chaotic flow around it. This replacement model achieves comparable levels of accuracy to the CFD and thermal FEA analysis of the airframe, while drastically speeding up the design process compared to conventional methods. However, further work is required to refine the structural portion of the machine learning model to ensure higher levels of accuracy in the prediction of the stress profile of the airframe throughout high-speed flight.
- Academic Unit
- Mechanical Engineering
- Record Identifier
- 9985177076502771
Metrics
1 File views/ downloads
1 Record Views