An extensive near-field noise prediction of a subsonic jet using data-driven surrogate model based on neural networks

The use of surrogate models has become essential in modern design processes, with machine learning algorithms increasingly adapting to active meta modelling techniques. In this specific investigation, the formulation focuses on Artificial Neural Networks (ANNs) as data-driven nonlinear models. Within this study, ANNs formulation is used as a data–driven nonlinear model aimed at describing the dynamics and the noise emitted by a single-stream subsonic jet, trained with a large numerical database that includes a wide range of parameters. What sets this model apart is its pioneering incorporation of variables such as nozzle exhaust turbulence intensity and nozzle–exhaust boundary–layer thickness. These parameters significantly influence noise emissions and are challenging to model using traditional analytical methods. The training dataset was formulated using 80% of the information sourced from the numerical database derived from large-eddy simulations (LESs) of a jet flow operating at M=0.9 and Re = 105.Pressure time series were gathered from virtual probes placed at various radial and axial positions within the near field. The model has been properly validated and it is shown to predict well the pressure spectra over the entire range of frequencies of interest.