Tracking an untracked space debris after an inelastic collision using physics informed neural network

Area of Science:

• Space Debris Dynamics
• Astrodynamics
• Artificial Intelligence in Space Surveillance

Background:

• Increasing satellite deployment in Low Earth Orbits (LEO) elevates collision risks from untracked space debris.
• Small-sized space debris (<10 cm) poses a significant tracking challenge for current state-of-the-art methods.
• Accurate trajectory prediction of space debris is vital for preventing future orbital collisions.

Purpose of the Study:

• To develop and evaluate a Physics Informed Neural Network (PINN) approach for estimating the trajectory of untracked space debris post-collision.
• To compare the performance of PINN-based methods against classical optimization and deep neural network approaches for debris state estimation.

Main Methods:

• Simulated 8565 inelastic collision events between active satellites (Starlink, LEMUR) and space debris.
• Utilized TLE data for active satellite states and proposed novel methods for debris velocity initialization and coefficient of restitution sampling.
• Applied classical optimization (Lagrange multipliers), Deep Neural Networks (DNNs), and Physics Informed Neural Networks (PINNs) to estimate post-collision debris states (position, velocity, mass, coefficient of restitution).

Main Results:

• Classical optimization methods proved unsatisfactory due to the under-determined nature of the system.
• PINN-based methods demonstrated superior performance in estimating the position, velocity, mass, and coefficient of restitution of untracked space debris.
• Performance was quantitatively assessed using Root Mean Square Error (RMSE) and interquartile range, highlighting PINNs' improved accuracy.

Conclusions:

• Physics Informed Neural Networks offer a robust and accurate solution for tracking and estimating the states of small, untracked space debris after collisions.
• The PINN approach significantly enhances the ability to predict debris trajectories, contributing to improved space situational awareness and collision avoidance strategies in LEO.