본문 바로가기 대메뉴 바로가기

KAIST

NEWS

홈페이지 통합검색

-
KOREAN

%65%6c%65%63%74%72%69%63%20%70%72%6f%70%75%6c%73%69%6f%6e

KAIST Develops AI-Driven Performance Prediction Model to Advance Space Electric Propulsion Technology < (From left) PhD candidate Youngho Kim, Professor Wonho Choe, and PhD candidate Jaehong Park from the Department of Nuclear and Quantum Engineering > Hall thrusters, a key space technology for missions like SpaceX's Starlink constellation and NASA's Psyche asteroid mission, are high-efficiency electric propulsion devices using plasma technology*. The KAIST research team announced that the AI-designed Hall thruster developed for CubeSats will be installed on the KAIST-Hall Effect Rocket Orbiter (K-HERO) CubeSat to demonstrate its in-orbit performance during the fourth launch of the Korean Launch Vehicle called Nuri rocket (KSLV-2) scheduled for November this year. *Plasma is one of the four states of matter, where gases are heated to high energies, causing them to separate into charged ions and electrons. Plasma is used not only in space electric propulsion but also in semiconductor manufacturing, display processes, and sterilization devices. On February 3rd, the research team from the KAIST Department of Nuclear and Quantum Engineering’s Electric Propulsion Laboratory, led by Professor Wonho Choe, announced the development of an AI-based technique to accurately predict the performance of Hall thrusters, the engines of satellites and space probes. Hall thrusters provide high fuel efficiency, requiring minimal propellant to achieve significant acceleration of spacecrafts or satellites while producing substantial thrust relative to power consumption. Due to these advantages, Hall thrusters are widely used in various space missions, including the formation flight of satellite constellations, deorbiting maneuvers for space debris mitigation, and deep space missions such as asteroid exploration. As the space industry continues to grow during the NewSpace era, the demand for Hall thrusters suited to diverse missions is increasing. To rapidly develop highly efficient, mission-optimized Hall thrusters, it is essential to predict thruster performance accurately from the design phase. However, conventional methods have limitations, as they struggle to handle the complex plasma phenomena within Hall thrusters or are only applicable under specific conditions, leading to lower prediction accuracy. The research team developed an AI-based performance prediction technique with high accuracy, significantly reducing the time and cost associated with the iterative design, fabrication, and testing of thrusters. Since 2003, Professor Wonho Choe’s team has been leading research on electric propulsion development in Korea. The team applied a neural network ensemble model to predict thruster performance using 18,000 Hall thruster training data points generated from their in-house numerical simulation tool. The in-house numerical simulation tool, developed to model plasma physics and thrust performance, played a crucial role in providing high-quality training data. The simulation’s accuracy was validated through comparisons with experimental data from ten KAIST in-house Hall thrusters, with an average prediction error of less than 10%. < Figure 1. This research has been selected as the cover article for the March 2025 issue (Volume 7, Issue 3) of the AI interdisciplinary journal, Advanced Intelligent Systems. > The trained neural network ensemble model acts as a digital twin, accurately predicting the Hall thruster performance within seconds based on thruster design variables. Notably, it offers detailed analyses of performance parameters such as thrust and discharge current, accounting for Hall thruster design variables like propellant flow rate and magnetic field—factors that are challenging to evaluate using traditional scaling laws. This AI model demonstrated an average prediction error of less than 5% for the in-house 700 W and 1 kW KAIST Hall thrusters and less than 9% for a 5 kW high-power Hall thruster developed by the University of Michigan and the U.S. Air Force Research Laboratory. This confirms the broad applicability of the AI prediction method across different power levels of Hall thrusters. 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.” < Figure 2. The AI-based prediction technique developed by the research team accurately predicts thrust performance based on design variables, making it highly valuable for the development of high-efficiency Hall thrusters. The neural network ensemble processes design variables, such as channel geometry and magnetic field information, and outputs key performance metrics like thrust and prediction accuracy, enabling efficient thruster design and performance analysis. > Additionally, Professor Choe mentioned, “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.” This research was published online in Advanced Intelligent Systems on December 25, 2024 with PhD candidate Jaehong Park as the first author and was selected as the journal’s cover article, highlighting its innovation. < Figure 3. Image of the 150 W low-power Hall thruster for small and micro satellites, developed in collaboration with Cosmo Bee and the KAIST team. The thruster will be tested in orbit on the K-HERO CubeSat during the KSLV-2 Nuri rocket’s fourth launch in Q4 2025. > This research was supported by the National Research Foundation of Korea’s Space Pioneer Program (200mN High Thrust Electric Propulsion System Development). (Paper Title: Predicting Performance of Hall Effect Ion Source Using Machine Learning, DOI: https://doi.org/10.1002/aisy.202400555 ) < Figure 4. Graphs of the predicted thrust and discharge current of KAIST’s 700 W Hall thruster using the AI model (HallNN). The left image shows the Hall thruster operating in KAIST Electric Propulsion Laboratory’s vacuum chamber, while the center and right graphs present the prediction results for thrust and discharge current based on anode mass flow rate. The red lines represent AI predictions, and the blue dots represent experimental results, with a prediction error of less than 5%. >
2025.02.03 View 15786