Two approaches for constructing multivariate injection models for prefilming airblast atomizers

Abstract

More realistic droplet starting conditions for Euler–Lagrangian simulations enable e.g. more precise soot prediction in jet engines. Up to now, mainly the droplet size distribution of sprays is considered, but not the multivariate dependence structure of droplet size, starting position and initial velocity. A novel concept for extracting multivariate spray data efficiently from detailed simulations of the atomizing process into high fidelity Euler–Lagrangian simulations of spray combustion is presented in this paper. Therefore, simulations of a prefilming airblast atomizer using the Smoothed Particle Hydrodynamics method are considered. The multivariate dependence structure in the spray is identified using rank transformation. Two models of different nature are proposed which are able to reproduce the multivariate dependence structure. The first model follows a data-driven approach using vine copulas and marginal distributions. In contrast, the second model is based on human knowledge and assumptions enabling deeper insights into the atomization process. Both models demonstrated to reproduce the multivariate character of the spray data effectively. An assessment of their capabilities reveals that the first model might be more suitable for spray data of annular injectors.