< Photo 1. (From left) Doctoral candidate Youngho Kim, Professor Wonho Choe and Doctoral candidate Jaehong Park of the Department of Nuclear and Quantum Engineering >
Hall thruster is a high-efficiency propulsion device using plasma* that is used for various difficult space missions such as SpaceX's constellation satellites, Starlink and NASA's asteroid probe, Psyche, and is one of the core space technologies. KAIST researchers announced that they will be verifying the performance of the Hall thruster for CubeSats developed using artificial intelligence techniques by loading it onto the CubeSat K-HERO during the fourth launch of Nuri scheduled for November of this year.
*Plasma is one of the four states of matter in which gas is heated with high energy and separated into ions and electrons with electric charges. It is widely used in space electric propulsion as well as in semiconductor and display manufacturing processes and sterilization devices.
KAIST (President Kwang Hyung Lee) announced on the 3rd that the research team led by Professor Wonho Choe of the Department of Nuclear and Quantum Engineering developed an artificial intelligence technique that can predict the thrust performance of Hall Effect ion thrusters (i.e. Hall thrusters), which are engines for satellites or space probes, with high accuracy.
Hall thrusters have high fuel efficiency, so they can greatly accelerate satellites or spacecraft using less propellant (fuel), and can generate large thrust relative to the power consumed. Based on these advantages, it is widely used for various missions such as maintaining formation flight of satellite clusters in space environments where propellant conservation is important, orbital deorbit maneuvers for reducing space debris, and providing propulsion for deep space exploration such as comet or Mars exploration.
Recently, as the space industry has expanded in the era of Newspace, space missions are becoming more diverse and the demand for Hall thrusters is increasing accordingly. In order to quickly develop high-efficiency Hall thrusters optimized for each unique mission, a technique to accurately predict the performance of the thruster from the design stage is essential.
However, existing methods have limitations such as not being able to precisely handle the complex plasma phenomenon occurring in the Hall thruster or being limited to specific conditions, resulting in low performance prediction accuracy.
The research team developed a highly accurate thruster performance prediction technique based on artificial intelligence that drastically reduces the time and cost required for repetitive work of designing, manufacturing, and testing the Hall thruster.
Professor Wonho Choe's team, which started the first domestic electric thruster development research in 2003 and has been leading related research and development, introduced an artificial neural network ensemble structure based on 18,000 Hall thruster learning data generated using a self-developed electric thruster computer analysis tool and applied it to predicting thrust performance.
The computer analysis tool developed to secure high-quality learning data models plasma physics and thrust performance. The accuracy of the computer analysis tool was verified to be high, with an average error of less than 10% compared to approximately 100 experimental data performed with 10 Hall thrusters developed for the first time in Korea by the research team.
< Figure 1. The paper on the results of this study was selected to be the cover of the third issue of 2025 Advanced Intelligent Systems (Vol. 7), an internationally recognized multidisciplinary journal on artificial intelligence. >
The learned artificial neural network ensemble model operates as a digital twin model that can predict thruster performance in a short period of time, within a few seconds, with high accuracy depending on the design variables of the Hall thruster.
In particular, it can analyze in detail the changes in performance indicators such as thrust and discharge current according to design variables such as fuel flow rate and magnetic field that were difficult to analyze with previously known scaling laws.
The research team showed that the AI neural network model developed this time showed an average error of less than 5% for the 700W and 1kW class Hall thrusters developed in-house, and an average error of less than 9% for the 5kW class high-power Hall thruster developed by the US Air Force Research Laboratory. This study proved that the AI prediction technique developed can be widely applied to Hall thrusters of various power sizes.
Professor Wonho Choe said, "The AI-based performance prediction technique developed by the research team has high accuracy and is already being used to analyze the thrust performance of Hall thrusters, which are engines for satellites and spacecraft, and to develop high-efficiency, low-power Hall thrusters. This AI technique can be applied not only to Hall thrusters, but also to the research and development of ion beam sources used in various industries such as semiconductors, surface treatment, and coating."
< Figure 2. The AI-based technique to predict Hall Effect Ion Source developed by the research team can predict the performance of the thrust according to design variables with high accuracy, rendering it highly useful in the development of high-efficiency Hall thrusters. The artificial neural network ensemble receives design variables such as the shape of the channel where plasma is generated and magnetic field information, and provides performance values such as thrust and prediction accuracy. This enables the design of high-efficiency Hall thrusters, and is also highly useful in the analysis of thrust performance characteristics. >
In addition, Professor Choe explained, "The Hall thruster for the cube satellite developed using AI techniques in collaboration with Cosmo Bee Co., Ltd., an electric propulsion specialist and a laboratory startup of the research team, will be installed on the 3U (30x10x10 cm) cube satellite K-HERO in the 4th launch of Nuri scheduled for November of this year to verify its performance in space."
The results of this study, in which Ph.D. student Jaehong Park of KAIST Department of Nuclear and Quantum Engineering (Space Exploration Engineering Interdisciplinary Major) participated as the first author, were published online on December 25, 2024 in the internationally renowned academic journal in the field of multidisciplinary AI research and development, Advanced Intelligent Systems, and were recognized for their innovation by being selected as the journal's front cover paper.
< Figure 3. This shows the operation of a 150 W-class low-power Hall thruster for (ultra)small satellites that is being developed by the research team of Professor Wonho Choe of KAIST in collaboration with Cosmobee, an electric propulsion specialist company founded by the laboratory. This Hall thruster is scheduled to be installed on the CubeSat, K-HERO, that will be launched into space through the 4th launch of Nuriho in the 4th quarter of 2025 to perform in-orbit verification. >
< Figure 4. Graphs that shows the actual thrust and discharge current of the 700 W-class Hall thruster developed by KAIST and the predictions by the AI model (HallNN) developed by the research team. The photo on the left shows the Hall thruster in operation in the Electric Propulsion Vacuum Chamber at KAIST, and the graphs to the right show the results of thrust and discharge current according to the anode flow rate. In the graph, the red line represents the results predicted by the AI model, and the blue dots represent the experimental results. The predicted values showed high accuracy with an average error of less than 5% compared to the experimental values. >
This study was conducted with the support of the National Research Foundation of Korea's Space Pioneer Project for the Development of a 200mN-class High-performance Electric Propulsion System.
Paper Title: "Predicting Performance of Hall Effect Ion Source Using Machine Learning" DOI: https://doi.org/10.1002/aisy.202400555
