Plasma, the fourth state of matter, is created when gases are energized to a state where they separate into charged ions and electrons. This technology is widely applied beyond space propulsion, including in semiconductor manufacturing, display production, and sterilization.

On February 3rd, KAIST's Electric Propulsion Laboratory, under the Department of Nuclear and Quantum Engineering and led by Professor Wonho Choe, announced the development of an AI-based predictive model that significantly improves the accuracy of Hall thruster performance estimations.

Hall thrusters are highly efficient, requiring minimal propellant while delivering significant thrust relative to power consumption. Their advantages make them ideal for a range of applications, including satellite constellations, debris deorbiting maneuvers, and deep-space exploration.

With the expanding space industry in the NewSpace era, demand for mission-specific Hall thrusters is increasing. Developing optimized thrusters efficiently requires precise performance prediction at the design stage. Traditional predictive methods struggle with the complex plasma dynamics within Hall thrusters, limiting their effectiveness.

KAIST's AI-based prediction model overcomes these limitations by providing high-accuracy performance forecasts while significantly reducing the time and cost associated with iterative design, prototyping, and testing. Since 2003, Professor Choe's team has been a leader in Korea's electric propulsion research. Their AI system is trained on 18,000 Hall thruster data points generated from their in-house numerical simulation tool.

This simulation tool, designed to model plasma physics and thrust behavior, ensures high-quality training data. Its accuracy was validated against experimental data from ten KAIST Hall thrusters, achieving an average prediction error of less than 10%.

The trained neural network ensemble acts as a digital twin, rapidly predicting Hall thruster performance based on design variables within seconds. It provides detailed assessments of critical performance parameters, including thrust and discharge current, and accounts for variables such as propellant flow rate and magnetic field - factors difficult to analyze using conventional scaling laws.

The AI model demonstrated an average prediction error of less than 5% for KAIST's in-house 700 W and 1 kW Hall thrusters, and under 9% for a 5 kW high-power thruster developed by the University of Michigan and the U.S. Air Force Research Laboratory. This confirms the model's applicability across different thruster power levels.

Professor Wonho Choe stated, "The AI-based prediction technique developed by our team is highly accurate and is already being utilized in the analysis of thrust performance and the development of highly efficient, low-power Hall thrusters for satellites and spacecraft. This AI approach can also be applied beyond Hall thrusters to various industries, including semiconductor manufacturing, surface processing, and coating, through ion beam sources."

Professor Choe also highlighted, "The CubeSat Hall thruster, developed using the AI technique in collaboration with our lab startup - Cosmo Bee, an electric propulsion company - will be tested in orbit this November aboard the K-HERO 3U (30 x 10 x 10 cm) CubeSat, scheduled for launch on the fourth flight of the KSLV-2 Nuri rocket."

The research findings were published online in *Advanced Intelligent Systems* on December 25, 2024. PhD candidate Jaehong Park was the lead author of the paper, which was also selected as the journal's cover article.

Research Report:Predicting Performance of Hall Effect Ion Source Using Machine Learning

Related Links
Korea Advanced Institute of Science and Technology
Rocket Science News at Space-Travel.Com