Surrogate Neural Network Model for Integrated Ascent Trajectory Optimization of Throttleable Hybrid Rockets

In this paper, the integrated optimization of a pump-fed two-stage hybrid launch vehicle with throttling capabilities on the first stage is analyzed. The novel approach of utilizing neural network-based surrogate models to predict hybrid rocket engines performance and masses is adopted, thereby extending the state-of-the-art to include throttleable hybrid rockets. The primary aim of this methodology is to significantly reduce the computational cost of the optimization process by substituting the complex calculations of hybrid rockets internal ballistics with a lightweight neural network model. Three throttling modes are analyzed and compared: a bi-level throttling, which entails a single-step oxidizer mass flow rate shift during engine operation; linear throttling, which features a continuous linear variation of the oxidizer mass flow rate throughout the engine burn; and an unthrottled configuration. Surrogate models for each configuration are trained using a 0-D hybrid rocket engine model to map a reduced number of design parameters to the engine performance and mass budget. Following an evaluation of their accuracy in predicting crucial data for hybrid rockets, the trained neural networks serve as surrogates for the throttleable first stage hybrid rocket and are integrated into a multi-disciplinary optimization process. The aim is ascent trajectory which maximize the payload capacity to the target orbit.