Artificial neural network models for phase equilibrium predictions under engine trans/supercritical spray conditions

Abstract

Engine spray models based on phase equilibrium have made great progress in simulating trans/supercritical engine spray processes, but there are inherent weaknesses in terms of efficiency and stability for the conventional phase equilibrium algorithm due to the iterative schemes for solving complex nonlinear equations. The low efficiency of the conventional algorithm limits the amount of detail that can be considered in the simulation, while the instability may lead to unphysical results or even simulation divergence. In this work, a method based on artificial neural networks (ANNs) was developed as a potential alternative to the conventional algorithm applied in the engine spray models to achieve fast and robust phase equilibrium calculations. Three ANNs were constructed, including isothermal-isobaric-ANN (TPn-ANN), isenthalpic-isobaric-ANN (HPn-ANN) and adiabatic-mixing-temperature-ANN (AMT-ANN). The latter two models combined with TPN-ANN can be applied to heat and mass transfer flow and adiabatic mixing problems, respectively, to achieve the prediction of phase equilibrium temperature, phase stability and phase splitting. The current work shows that the ANN method leads to significant efficiency improvements while maintaining almost the same accuracy as the conventional algorithm. Analysis of execution time for a high-fidelity n-dodecane spray simulation shows that the conventional phase equilibrium calculation can take up to 70% of total computational time, which can be reduced to negligible levels with the current proposed ANN approach.