Modelling of Soot Nanoparticle Formation in Turbulent Flames

Abstract

Soot emission from hydrocarbon fuel combustion is a major source of particulate pollution. The increasingly stringent regulations on emissions have necessitated the developments of soot models to aid designs of combustion devices with cleaner performance. Such models will also make valuable contributions in designing and optimising processes that produce beneficial carbonaceous particulates, e.g. carbon black plants and processes requiring enhanced soot-induced radiation. The present work aims to develop a detailed soot model and implements the model in the sparse multiple mapping conditioning (MMC) - large eddy simulation (LES) framework to form a predictive tool for soot evolution in turbulent flames. The thesis consists of two major parts presenting soot evolution without and with turbulence, respectively. In the first part, the fundamental physics of soot evolution processes is presented, followed by a review of soot formation modelling that focuses on the sectional methods. The current work uses a sectional soot kinetics scheme that approximates solutions to the population balanced equations by lumped species and replaces individual growth/oxidation models by a sequence of equivalent physicochemical reactions with Arrhenius-like rate expression. The soot kinetics is a reduced version derived from a multisectional soot mechanism (Sirignano et al., Energy & Fuels 27, 2013). In this work, a novel generalised model describing the interaction potential well depth between soot particles of any size and composition is proposed for a thermal rebound based coagulation model to account for the probability of combining electrically neutral entities, i.e. nucleation, condensation and coagulation. The coagulation model is then simplified into Arrhenius expression so that the gas and soot kinetics can be integrated into a fully coupled system. The model is tested by comparisons to the experimental data of a series of ethylene burner stabilised stagnation (BSS) premixed flames and a methane laminar coflow diffusion flame, and the sensitivity to model parameters is investigated. Overall, the simulation results show good agreement with the experimental measurement, but strong sensitivity to the parameter λ that accounts for void fractions of soot particles is observed. In the second part, the turbulence and combustion models are reviewed. The focus is placed on the LES and MMC methods and the closure strategies in the methods. As the source terms of a lumped species reflecting different evolution processes are represented by a chemical source term in the sectional soot kinetics model, the filtered transport equation of lumped species can be straightforwardly closed by a joint filtered density function (FDF) of gas-phase species and lumped species. This work employs the sparse generalised MMC-LES, a stochastic FDF type method but uses much fewer notional particles (usually fewer than the number of LES cells) than the traditional FDF method. Combined with the code developments with some emphasis on computational load balancing models, the coupled turbulence-chemistry-soot model is shown to provide detailed soot evolution solutions, such as particle size distribution (PSD), with reasonable computational costs. The model is examined by comparison to the experimental data of the Delft Adelaide flame. Although discrepancies exist, the numerical predictions on soot volume fraction and intermittency are in reasonable accuracy. Detailed investigations on the probability density functions of the soot volume fraction show that the model captures the key features of soot formation but predicts significantly more soot in the range between 0.1 ppb and 6.4 ppb than the experimental measurements. Lastly, the predicted soot PSDs are presented. The results suggest that the PSDs in the turbulent flame simulation are in mixed unimodal and bimodal distributions.