Interaction mechanisms of flash boiling spray with air-wall-time in a SIDI engine and implications on flame propagation and emissions using experimental investigation and machine learning

Flash boiling spray is a promising method for improving performance and efficiency of spark-ignition direct-injection (SIDI) engines and reducing emissions. However, controlling it is challenging due to its complex interactions with the air, cylinder wall, piston, and time. This study investigates the flame propagation and particulate number (PN) emissions in an optical SIDI engine using high speed camera under different spray superheat degrees ranging from 4.5 to 0.065, and injection timings of 350, 300, and 180 °bTDC. Additionally, a machine learning (ML) approach applied to perform importance analysis and confusion matrices based on 26 extracted combustion and flame features. The findings revealed severe interactions between subcooled spray and piston top surface and cylinder wall, resulting in central or near wall sooty flames and high PN emissions. However, flash boiling spray reduces sooty flames and PN emissions, except when excessive superheat was combined with late injection, resulting in luminous sooty flames near the injector nozzle. Therefore, combining moderate superheat degree and airflow can improve the combustion process and reduce emissions. The ML algorithm demonstrated up to 86 % accuracy in classifying flame images. The importance analysis highlighted features that give high classification efficiency to maintain feasible research efforts. Under the optimal flash boiling spray control, IMEP has enhanced by10 %–19 % and PN emissions decreased by 50∼90 %.