Neural Network-Based Prediction of Liquid-Phase Diffusion Coefficient to Model Fuel-Oil Dilution on Engine Cylinder Walls

Nowadays the role played by passenger vehicles on the greenhouse effect is of great value. To slow down both the global warming and the fossil fuel wasting, the design of high-efficiency engines is compulsory. Downsized Turbocharged Gasoline Direct Injection engines comply with both high-efficiency and power demand requirements. Nevertheless, the application of Direct Injection inside downsized combustion chambers may result in the fuel wall impingement, depending on the operating conditions. The impact of the fuel on the cylinder liner leads to the mixing of the fuel and the lubricant oil on the cylinder wall. When the piston moves, the piston top ring scraps the non-evaporated fuel-oil mixture. Then the scraped fuel-oil mixture may be scattered into the combustion chamber, becoming a source of diffusive flames in all conditions and abnormal combustions known as Low-Speed Pre-Ignitions at the highest loads. To analyze these phenomena, accurate predictions of the liquid-phase diffusion between fuel and oil are needed. Currently, no experimental data are available for the diffusion between fuel and oil and the available correlations are characterized by a high inaccuracy (errors around 20%-40% are reported). In this work, a Deep Neural Network methodology was developed and validated against engine-like fluids. Furthermore, the diffusion coefficients for different gasoline surrogates/SAE oils are provided, and the effect of gasoline-ethanol blending is discussed.