A Reinforcement Learning Approach for Transient Control of Liquid Rocket Engines

Nowadays, liquid rocket engines use closed-loop control at most near-steady operating conditions. The of the transient phases is traditionally performed in open loop due to highly nonlinear system dynamics. This situation unsatisfactory, particular for reusable engines. open-loop cannot provide optimal engine performance external disturbances or degeneration components over time. In this article, we study a deep reinforcement learning approach generic gas-generator engine's continuous startup phase. It shown that learned policy can reach different steady-state points and convincingly adapt changing parameters. Compared carefully tuned sequences proportional-integral-derivative (PID) controllers, controller achieves highest performance. addition, it requires only minimal computational effort calculate action, which big advantage approaches require online optimization, such as model predictive control.