Submitting Campus

Daytona Beach

Student Status



Graduate Student Works

Advisor Name

Dr. Riccardo Bevilacqua


The future of spaceflight is threatened by the increasing amount of space debris, especially in the near-Earth environment. To continue operations, accurate characterization of hypervelocity fragment propagation following collisions and explosions is imperative. While large debris particles can be tracked by current methods, small particles are often missed. This paper presents a method to estimate fragment fly-out properties, such as fragment, velocity, and mass distributions, using machine learning. Previous work was performed on terrestrial data and associated simulations representing space debris collisions. The fragmentation of high-velocity fragmentation can be modeled by terrestrial fragmentation tests, such as static detonations. Recently, stereoscopic imaging techniques have become an addition to static arena testing. Collecting data with this method provides position vector and mass information faster and more accurately than previous manual-collection methods. Additionally, there is limited space debris data of similar accuracy on individual fragments. Therefore, this imaging technique was used as the primary collection method for the previous research data. Now, two-line element (TLE) sets for Iridium 33 are also used. Machine learning methodologies are leveraged to predict fragmentation fly-out from the collision event with Cosmos 2251. First, gaussian mixture models (GMMs) are used to model the probability distribution of the particles for a given desired characteristic at Julian dates following the event. Once this training data is generated, regression techniques can be used to predict these characteristics. K-nearest neighbor (K-NN) regressors are used to estimate the spatial distribution of the satellite fragments. Monte Carlo simulations are then used to validate the results, finding that this technique accurately estimates the total number of fragments expected to intersect a region of interest at a given time. Following this work, the same technique can be used to estimate the velocity and mass distributions of the debris. This information can then be used to estimate the kinetic energy of the particle and classify it to avoid future debris collisions.

Document Type


Publication/Presentation Date


Publication Title

Using Machine Learning to Predict Hypervelocity Fragment Propagation of Space Debris Collisions


Baku, Azerbaijan

Grant or Award Name

U.S. Air Force Office of Scientific Research (award number FA9550-20-1-0200), SMART Scholarship Program and the Intuitive Machine and Columbia Sportswear Advancing Women in Technology Fellowship