Author ORCID Identifier
Semester
Fall
Date of Graduation
2025
Document Type
Dissertation
Degree Type
PhD
College
Statler College of Engineering and Mineral Resources
Department
Mechanical and Aerospace Engineering
Committee Chair
Piyush M. Mehta
Committee Member
Hang Woon Lee
Committee Member
Andrew Rhodes
Committee Member
Jason Gross
Committee Member
Scott A. Zemerick
Abstract
Recent advances in hardware and software technology have made it possible to implement more resource-demanding deep learning algorithms in constrained hardware environments. This creates opportunities to use deep learning for aerospace applications on increasingly smaller aerospace vehicles. This work presents the implementation of a Neural Network Execution Framework (NNEF), which aims to provide a cross-platform and reusable framework to deploy and execute trained neural networks for deep learning aerospace applications. The NNEF executes any neural network inference process regardless of the original deep learning framework in which it was created, for supported flight software platforms, and space-like computer boards. Users and organizations can utilize the framework to create reusable deployment and execution solutions for deep learning, rather than implementing one-off solutions each time they need to develop a specific aerospace application. This approach allows developers to focus on their deep learning research objectives, rather than the implementation and deployment process on the flight platform. This work shows the design, implementation, and testing of the NNEF for deploying and executing neural networks on supported flight software, so far, Core Flight System (cFS) and F Prime (F’), two of the main NASA flight software frameworks for space missions. Through the implementation of general, reusable, and cross-platform classes, components, and applications, the NNEF executes neural networks developed and trained in PyTorch and TensorFlow, using the interfaces offered by the LibTorch, TensorFlow, and TensorFlow Lite libraries. For prototype testing and technology validation, we deployed a basic Convolutional Autoencoder applied to MNIST images for the three technologies, as well as one of our laboratory’s novel networks (ASSIST lab), a neural-based compression algorithm used to process images from NASA’s Solar Dynamics Observatory. Additionally, we will also describe a stepby- step guideline to show two case studies of integration and deployment of a new NN model for an aerospace application. We used two different low-cost development boards to emulate a space-like hardware-software platform, to deploy both cFS and F’ implementations, and to show how all integrated models perform. With this, the NNEF’s cross-platform, portability between flight software, and multi-DL framework support capabilities are validated, contributing to a novel implementation paradigm for deep learning model inference in aerospace applications.
Recommended Citation
Polanco Segovia, Rafael, "Implementation of a Neural Network Execution Framework for Generalized and Cross-platform Deep Learning Deployment and Inference on Spacecraft Systems" (2025). Graduate Theses, Dissertations, and Problem Reports. 13160.
https://researchrepository.wvu.edu/etd/13160
Included in
Aeronautical Vehicles Commons, Computer and Systems Architecture Commons, Hardware Systems Commons, Other Aerospace Engineering Commons, Other Computer Engineering Commons, Space Vehicles Commons