Direct numerical simulation and modelling of thermodiffusively-unstable lean premixed hydrogen flames for carbon-free internal combustion engines

Abstract

Thermodiffusively (TD) unstable lean hydrogen flames, a promising alternative to the combustion of hydrocarbons, have been studied using direct numerical simulation with complex chemistry over a wide range of reactant and turbulent conditions to develop a better understanding of the fundamental behaviour of TD-unstable flames, with a specific focus on the flame speed, which will be used to inform turbulent flame speed models used for device scale simulations. Firstly the thesis investigates the fundamental behaviour of the TD-instability for 3-dimensional laminar freely-propagating flames over a wide range of reactant conditions. It is shown that the instability parameter ω2 extends well to 3-dimensions and a model is proposed that has good predictive capabilities for local flame acceleration and thinning arising from the reactant conditions. Next, the work is extended to turbulent flames at a scale over a wide range of reactant and turbulent conditions. It is found that turbulence intensifies the TD-response and is found that the mean local flame speed scales well. This the square-root of the Karlovitz number, provided the laminar TD-instability is accounted for using ω2. Next, the integral length scale as well as the Karlovitz number over a range of reactant conditions are considered to evaluate the combined effect of reactant conditions, turbulent intensity and length scale on both the local and global flame statistics with a focus on the turbulent flame speed. It is found that length scale does not effect the local flame statistics, but does effect the global flame statistics where the turbulent flame speed adheres to Damkohler’s ¨ small scale limit. Lastly, a method for a well resolved (DNS-style) G-equation capable of simulating TD-unstable flames is proposed and presented, which could be used in future studies to simulate larger domains due to its lower computational cost.