An application of machine learning for geometric optimization of a dual-throat bent nozzle

Abstract

The optimal thrust shape for a dual-throat bent nozzle (DTBN), designed as a hybrid thrust vectoring nozzle, is derived through machine learning. A compressible, steady-state numerical analysis using the $k-\omega$ SST model is employed for model construction. The main geometric parameters that determine the shape of the DTBN are selected as the convergence angle ${\theta }_c$, divergence angle ${\theta }_d$, and cavity length $l_c$. By varying these parameters, DTBN models with a total of 600 different geometries are generated, and the axial and normal forces at the nozzle exit are observed to derive the thrust magnitude and thrust vectoring angle. A model that can accurately predict the correlation between input and output parameters is built by comparing various machine learning algorithms. The model using the random forest regression algorithm shows the best performance. Based on this developed machine learning model, optimized shapes of the DTBN are presented. The optimally designed DTBNs are expected to contribute to the development of a new system with more convenient thrust control.