Shape Optimization and Flow Analysis of Supersonic Nozzles Using Deep Learning

Abstract

Shape optimisation of supersonic nozzles is of crucial importance in designing propulsion systems and space thrusters. In order to optimise the profile of a supersonic nozzle, the properties of the flow inside the nozzle should be obtained. This paper proposes and verifies a new methodology for analysing flows and designing supersonic nozzles. Flow analysis has been conducted using the method of characteristics, Ansys Fluent and convolutional neural networks. It is shown that deep convolutional neural networks can reach high levels of accuracy in predicting supersonic flow behaviour inside the nozzle. Also, shape optimisation of the supersonic nozzle has been conducted using the genetic algorithm in Ansys Fluent and artificial neural networks. The proposed ANN can optimise the shape of a supersonic nozzle for the given throat diameter, outlet diameter and nozzle length with high accuracy.