Predicting the Complex Dependence of Ignition Delay Times on Pressure Using Machine Learning

Abstract Accurate knowledge of ignition times is crucial for the development of combustion systems. Reliable tools are therefore needed for the prediction of ignition delay times at various conditions. Although detailed chemical kinetic simulations can be used to predict ignition times under various conditions, these simulations require special solvers and they can be expensive, with respect to associated computational cost and required time. Simplified correlations of ignition delay times, based on perceived dependence on pressure, temperature, and composition can be used to overcome these challenges. But the resulting correlation parameters tend to vary based on initial conditions. One of these complex variations is encountered in the relation between ignition time and pressure, which is not easily captured by a single exponent over a wide range of conditions. Machine learning can be used to deal with these complexities without having to assign fixed parameters. In this study, validated chemical kinetic models are used to simulate the ignition times of kerosene surrogates and hydrogen fuel/air mixtures. The data from these simulations are first used to investigate the complex dependence of ignition times on pressure over a wide range of pressures. A machine learning algorithm is then trained to accurately predict ignition times, based on initial conditions of pressure, temperature, and fuel/air compositions. Further analyses are carried out to elucidate the kinetic origin of the varying pressure dependence of ignition delay times. This work sheds light on the dependence of ignition delay times and demonstrates the use of machine learning to capture this complexity, thereby enhancing prediction of ignition delay times over a wide range of conditions.