DEVELOPMENT OF A TEST INFRASTRUCTURE FOR A NEURAL NETWORK CONTROLLED GREEN PROPELLANT THRUSTER

Abstract

This paper describes the setup of a test facility for various modern control methods for a green propellant thruster. The overarching goal is the intelligent control of a rocket engine, for example with a neural network. Therefore, a flexible and secure training and testing environment is needed that offers the possibility to test and compare different control approaches. In this paper, particular attention is paid to the interface between the test facility and the neural network. A reinforcement learning based controller is described in more detail and in the last section the results of a first test run with a neural network-based controller at the test facility M11.5 are discussed.