KAIST announced that a new spacecraft propulsion system it has developed will be launched into space aboard a satellite on South Korea’s Nuri launch vehicle. (Image courtesy of KAIST)
SEOUL, Feb. 4 (Korea Bizwire) — The Korea Advanced Institute of Science and Technology (KAIST) announced on February 3 that a new spacecraft propulsion system it has developed will be launched into space aboard a satellite on South Korea’s Nuri launch vehicle.
A research team led by Choe Won-ho, a professor in KAIST’s Department of Nuclear and Quantum Engineering has developed groundbreaking technology that significantly improves the performance of Hall thrusters — engines used in satellites and spacecraft — through artificial intelligence.
Unlike chemical thrusters that rely on combustion, Hall thrusters generate propulsion by using electrical energy to create and accelerate plasma, a fourth state of matter where gas is heated to such high energy that it separates into ions and electrons.
These thrusters are valued for their high thrust-to-power ratio, making them ideal for various space missions where propellant efficiency is crucial, including formation flying of satellite constellations, orbital maneuvers for space debris reduction, and deep space exploration of comets and Mars.
The technology is already being used in ambitious space missions, including SpaceX’s Starlink satellite constellation and NASA’s Psyche asteroid explorer.
While optimizing Hall thrusters for specific missions requires accurate performance predictions during the design phase, traditional methods have struggled with either the complexity of plasma phenomena or limited operational conditions.
The KAIST team developed an AI-based performance prediction method that dramatically reduces the time and cost associated with thruster design, manufacturing, and testing cycles. Researchers implemented an artificial neural network ensemble structure based on 18,000 Hall thruster training data points generated using their proprietary computational analysis tools.
The researchers explained that the ensemble technique, which averages predictions from independently trained neural networks, allows for intuitive and effective estimation of prediction uncertainty by calculating the standard deviation of neural network predictions.
Testing showed the system achieved an average error rate of less than 9%, surpassing the accuracy of conventional methods.
The technology was developed in collaboration with Cosmo Bee, a company specializing in electric propulsion. It will undergo space verification aboard the K-HERO CubeSat during the fourth launch of the Nuri rocket, scheduled for November of this year.
“Our AI-based performance prediction method is already being used to analyze thrust performance and develop high-efficiency, low-power Hall thrusters,” said Choe. “The technology has potential applications beyond Hall thrusters, extending to semiconductors, surface treatment, coating, and a number of other industrial fields.”
The research was published as the cover article in the third issue of volume seven of the international journal “Advanced Intelligent Systems” this year.
Kevin Lee (kevinlee@koreabizwire.com)
