Strengthening Ion Thrusters with CFD and Reinforcement Learning

Ion thrusters have been explored by many scientists, from hobbyists to space exploration agencies, and they usually have low thrust output / thrust-to-weight ratio, despite the creation of many brilliant designs. We have been exploring methods to use computational fluid dynamics to simulate ion thrusters, with the intention of finding novel designs worthy of experimental implementation. Reinforcement learning often does well at discovering non-intuitive options for complicated design problems, when given a rich parameter space to explore, which in our case includes electrode voltage, electrode separation, chassis geometry, electrode geometry, etc. Our hope is to find a design that offers appreciable thrust output / thrust-to-weight ratio for manufacturable ion thrusters.