A methodology for estimating hypersonic engine performance by coupling supersonic reactive flow simulations with machine learning techniques

Abstract

We propose a methodology used to estimate the performance of hypersonic engines by coupling some machine learning methods with a generated CFD database and one-dimensional expansion equations. The present work also investigates the effects of fuel struts geometrical parameters on the supersonic combustion which takes place in a dual-mode ramjet/scramjet engine operating at stratospheric flight conditions with a cruise speed of Mach 8. We solved 2D compressible RANS flow equations coupled with the β-pdf method using Ansys Fluent commercial code, and built a CFD database with varying three design parameters which are strut V-settlement angle, strut location, and wedge angle. The numerical code was verified/validated using the data obtained in the DLR scramjet combustor. Three main objective functions in the combustor (combustion efficiency along 14 stations having 10 cm interval between each other, averaged flow variables at exit of combustor, and total pressure recovery factor in the combustor) were selected to evaluate the struts configuration parameters effects on supersonic combustion flow physics and engine performance. These objective functions, i.e. dependent variables were regressed with independent design parameters by three machine learning techniques which are Kernel regression, Gaussian process regression, and Artificial neural network. The machine learning models of the flow variables at the combustor throat provided initial conditions for a given fuel strut configuration to the expansion solutions for the nozzle component in order to calculate the engine performance. The Artificial neural network was found the most successful technique in overall regression of the objective functions. We also discovered that the most significant design variable is the strut wedge angle for all investigated phenomena in the supersonic reactive flow field.