SAN FRANCISCO – Miles Space has demonstrated a water-fueled electric thruster with unusually lower power demands.
“We have a water-vapor-based plasma thruster that works at power levels so low, people don’t believe it,” Miles Space CEO Brad Berkson told SpaceNews.
During testing conducted in September 2024 on a European satellite, Miles Space’s Poseidon M1.5 thruster, which fits in a one-unit cubesat, demonstrated its utility for applications like descent from low-Earth orbit where thrusters fire for a long time, said an engineer who does not work for Miles Space but reviewed raw telemetry data.
The 10-centimeter-cube M1.5 produced thrust of 37.5 millinewtons over five minutes at a specific impulse of 4,800 seconds, while drawing power of 1.5 watts.
By comparison, hydrazine propulsion offers ten times more thrust but operates at a lower specific impulse. And Hall-effect thrusters can provide similar force at a lower specific impulse but may need hundreds of times the power.
“This enables some high Delta V missions to be performed that wouldn’t be possible otherwise due to either excessive fuel mass or thruster power requirements,” said the engineer who asked not to be identified because he was not authorized by his employer to discuss the propulsion.
Neural Network
Wesley Faler, chief technology officer for Florida-based Miles Space, developed the M1.5 thruster over a couple of years with the help of artificial intelligence. After establishing parameters for reaction chamber length and width, inlet locations, exit voltage and timing, Faler used genetic algorithms to simulate various combinations.
“Out would come thrust data and fuel-economy data,” Faler said. “I would decide which data points were better and the system used a neural net to understand my responses. Then, it would use that trained neural net to run simulations when I wasn’t around.”
Every couple of days, Faler reviewed the most promising simulations and provided feedback.
“Genetic algorithms were coming up with the shape of the thruster,” Faler said. “They evaluated themselves against a three-layer neural net that learned from talking to the human. Neither rise to the level of today’s AI.”
