Machine-learning enabled prediction of 3D spray under engine combustion network spray G conditions

Spray and air–fuel mixing in gasoline direct-injection (GDI) engines play a crucial role combustion emission characteristics. While variety of phenomenological spray models computational fluid dynamics (CFD) simulations have been applied to identify mixture distribution, most research efforts so far were concentrated on single axial-nozzle injectors limited range ambient conditions. Especially, the prediction flash-boiling sprays multi-hole remains great challenge due lack understanding complicated two-phase flow dynamics. For specific conditions, question can arise concerning capability machine-learning algorithms predict complex sprays. We developed algorithm, as simple variant linear regression, that is capable predicting 3D topology for various fuels A series experiments carried out constant-flow vessel coupled with high-speed diffused back-illumination extinction imaging produce data set algorithm training. Nine different test fuels, including component iso-octane (ic8) multi-component EEE gasoline, cover wide fuel properties injected using Engine Combustion Network (ECN) G injector under ECN G2 (50 kPa absolute), G3 (100 G3HT (G3 393 K temperature) Among ic8ib2 (ic8 80%, iso-butanol 20% v/v) specified target by thus they not included training data. The macroscopic analysis based projected liquid volume (PLV) computed tomographic (CT) reconstruction showed excellent agreement true values from experimental maximum differences penetration 3.6 mm (7.3% error) 1.3 (2.32% error), respectively. predicted had consistent trend showing slight plume movement but complete collapsing fuel. direction angle enabled CT up 2° compared during injection period. quantitative validation results performance nine input features (fuel conditions), actually superior CFD these same number parameters.