A Machine Learning Framework Using Neural Networks for Conceptual Design of First Stage Engine

This paper presents a fast and efficient methodology for estimating the thrust during the conceptual design phase of a first stage engine, leveraging a neural network model. The proposed approach shows strong potential for integration into digital twin frameworks and multidisciplinary design optimization processes. The methodology begins with the generation of a dataset derived from physics-based engine simulations. Engine characteristics are encoded as signal-like representations, which are then used to train a Multi-Layer Perceptron (MLP) neural network to predict thrust profiles over time based on a given set of engine parameters. The model’s accuracy is assessed by comparing its predictions with results from high-fidelity physics-based simulations. Finally, two key applications are demonstrated: inverse engine design and propulsion modelling within a multidisciplinary design optimization environment.