The separate inducer-impeller configuration for rocket turbopumps provides good efficiency and cavitation margin thanks to the sequential loading of the in-series blades. However, it is widely acknowledged that this configuration gives rise to significant flow instabilities and structural issues at start-up, part-load, and overload due to the strong interaction between the inducer and impeller rows. In the context of modern reusable rocket units, addressing these multi-regime interactions has become a compelling consideration for aerospace engineering teams.

This webinar introduces a novel design and optimization methodology that applies 3D inverse design and Physics-Enhanced Machine Learning (PEML) to develop a high-performance, single-component rocket-pump impeller. By doing away with the separate inducer row and using a splittered impeller configuration within the same meridional hub and shroud outline, the design completely avoids row interaction problems. The optimization workflow parameterizes the physics that controls flow in the blade passage using a 13-parameter design space subject to two key constraints: throat area and Stage Head at the design point. To drive the cavitation margin optimization, minimum pressure (Pmin) from 3D inverse design method is utilized as an excellently correlated proxy for cavitation onset, verified post-design via full 2-phase CFD simulations.

The use of Physics-Enhanced Machine Learning made possible by parametrizing geometry with 3D inverse design and the use of the unique Machine Learning algorithm of Reactive Response, a very high accuracy surrogate model is created with only 57 designs. The validation by CFD at multiple operating points for two different designs generated by the machine learning confirms the surrogate model prediction error of less than 0.3% for most parameters. The resulting compact, 4-blade, single-component turbopump matches or surpasses all baseline performance metrics, establishing a new paradigm for next-generation rocket engine design.