|
|
← Return to agenda
| Meeting | ACGS Committee Meeting 125 - Virtual - November 2020 | | Agenda Location | 4 GENERAL COMMITTEE TECHNICAL SESSION
4.6 Sub-Committee E
4.6.2 A Gaussian Mixture Model and Machine Learning based approach to predict warhead fragmentation in-flight behavior from static arena test data | | Title | A Gaussian Mixture Model and Machine Learning based approach to predict warhead fragmentation in-flight behavior from static arena test data | | Presenter | Riccardo Bevilacqua | | Affiliation | University of Florida | | Available Downloads* | presentation | | *Downloads are available to members who are logged in and either Active or attended this meeting. | | Abstract | Abstract
The detonation of warheads in static arena tests has provided real fragment trajectory data that is used to characterize lethality and collateral damage. In application, however, fragment trajectories do not match static test data because warheads arrive at targets at high velocities. State of the art simulation capabilities for high speed warhead detonations do not fully predict fragment trajectory characteristics that appear in real world tests, such as rigid-body aerodynamic effects on fragments. In this presentation, a framework to predict warhead fragment track characteristics from static arena test data and dynamic simulation data is explored. A model was developed to predict the number of fragments that pass through a defined surface of interest given warhead in-flight terminal conditions. Surfaces are defined by polar and azimuthal angle ranges on a sphere of specified radius. Distributions of fragment-surface intersections are modeled by Gaussian Mixture Models (GMMs), and the use of Deep Neural Networks as well as Random Forest Regressors trained to predict these GMMs from warhead in-flight terminal conditions is investigated. The ability of GMMs to model fragment-surface intersections when directly fit to data is evaluated through Monte Carlo simulations. In addition, the performance of a GMM predicted by a Random Forest is evaluated through Monte Carlo Simulations. The effects of varying Random Forest and GMM hyperparameters on model performance is investigated. The investigations show that the proposed model can accurately predict GMMs that fit well to fragment-surface intersection distributions.
Bio
Dr. Riccardo Bevilacqua is an Associate Professor of the Mechanical and Aerospace Engineering Department, at the University of Florida. He holds a M.Sc. in Aerospace Engineering (2002), and a Ph.D. in Applied Mathematics (2007), both from the University of Rome, "Sapienza", Italy. Dr. Bevilacqua is the recipient of two Young Investigator Awards, from the Air Force Office of Scientific Research (2012) and the Office of Naval Research (2013), of the 2014 Dave Ward Memorial Lecture Award from the Aerospace Controls and Guidance Systems Committee, and of four Air Force Summer Fellowships (2012 and 2015 at AFRL Space Vehicle Directorate, and 2019 and 2020 at AFRL Munitions Directorate). His research interests focus on spacecraft formation flight, space robotics and warheads/spacecraft fragment fly-out predictions. He has authored and co-authored more than 100 journal and conference publications on the topic. He is an AIAA Associate Fellow and IAA Corresponding Member. He is the creator and chair of the IAA conference on Space Situational Awareness. |
|
Copyright © 2025 | Question? webmaster@acgsc.org
|
