The Deflector Selector: A machine learning framework for prioritizing hazardous object deflection technology development. (May 2018)

• Record Type: Journal Article
• Authors: Nesvold, E.R.; Greenberg, A.; Erasmus, N.; van Heerden, E.; Galache, J.L.; Dahlstrom, E.; Marchis, F.
• Is Part Of: Acta astronautica. Volume 146(2018)
• Page Start: 33
• Page End: 45
• Publication Date: 2018-05
• Subjects: Planetary defense -- Orbital mechanics -- Machine learning; Astronautics -- Periodicals; Outer space -- Exploration -- Periodicals; Astronautics; Periodicals; 629.405
• DOI: 10.1016/j.actaastro.2018.01.049
• Languages: English
• ISSNs: 0094-5765

Abstract

Abstract: Several technologies have been proposed for deflecting a hazardous Solar System object on a trajectory that would otherwise impact the Earth. The effectiveness of each technology depends on several characteristics of the given object, including its orbit and size. The distribution of these parameters in the likely population of Earth-impacting objects can thus determine which of the technologies are most likely to be useful in preventing a collision with the Earth. None of the proposed deflection technologies has been developed and fully tested in space. Developing every proposed technology is currently prohibitively expensive, so determining now which technologies are most likely to be effective would allow us to prioritize a subset of proposed deflection technologies for funding and development. We present a new model, the Deflector Selector, that takes as its input the characteristics of a hazardous object or population of such objects and predicts which technology would be able to perform a successful deflection. The model consists of a machine-learning algorithm trained on data produced by N -body integrations simulating the deflections. We describe the model and present the results of tests of the effectiveness of nuclear explosives, kinetic impactors, and gravity tractors on three simulated populations of hazardous objects. Highlights: Discovery time significantly constrains the probability of deflection success. Object semi-major axis has the largest effect on the success of a deflection attempt. Chance of success of a deflection technology varies with object population parameters. Machine learning can accelerate predictions of deflection technology success.