A New Pathway for Prediction of Gasoline Sprays using Machine-Learning Algorithms

Abstract

<div class="section abstract"><div class="htmlview paragraph">The fuel spray process is of utmost importance to internal combustion engine design as it dominates engine performance and emissions characteristics. While designers rely on computational fluid dynamics (CFD) modeling for understanding of the air-fuel mixing process, there are recognized shortcomings in current CFD spray predictions, particularly under super-critical or flash-boiling conditions. In contrast, time-resolved optical spray experiments have now produced datasets for the three-dimensional liquid distribution for a wide range of operating conditions and fuels. By utilizing such a large amount of detailed experimental data, the machine learning (ML) techniques have opened new pathways for the prediction of fuel sprays under various engine-like conditions. The ML approach for spray prediction is promising because (1) it does not require phenomenological spray models, (2) it can provide time-resolved spray data without time-stepping simulation, and (3) its evaluation has only a tiny fraction of the computational cost of a CFD simulation. In this study, an Artificial Neural Network (ANN) was applied for gasoline spray prediction under realistic engine conditions. Experimental data obtained under seven different fuels and three ambient conditions, totaling 21 different cases, were fed into a training procedure to investigate fuel effects on spray morphology. The quantitative validation results showed that the ANN is capable of predicting spray performance with nine input features, including fuel properties and ambient conditions. The ANN model fully trained on the experimental dataset showed greater accuracy in capturing the details of plume dynamics especially under flash-boiling conditions than the current state-of-the-art CFD model. While the ANN model cannot yet function or replace CFD in a full engine simulation, the ANN can be used now as a convenient design tool incorporating vast physical conditions.</div></div>