par Amaduzzi, Ruggero
Président du jury Coussement, Axel
Promoteur Parente, Alessandro
Publication Non publié, 2023-09-29
Thèse de doctorat
Résumé : In order to meet the required CO2 emissions targets by 2050 the global energy production, distribution and consumption framework needs to drastically change. While most decarbonisation will come from the electrification of most end-uses, for the so-called hard-to-abate sectors decarbonisation will be met through a combination of increased efficiency of existing processes and usage of alternative fuels. Moreover, chemical storage of surplus electricity will be key for a sustainable energy system based on renewable sources. The development of efficient, flexible and clean combustion technologies is necessary. In recent years, flameless, or Moderate or Intense Low-oxygen Dilution (MILD) combustion has been studied to this end. However, literature on its use in combination with alternative fuels such as hydrogen and ammonia in industrial-relevant conditions is still scarce. The objective of this work is threefold. First, to investigate flameless combustion of non-conventional fuel blends, such as methane/hydrogen and ammonia/hydrogen, to assess the robustness and accuracy of state-of-the-art numerical models for a variety of operating conditions. Second, to quantify the uncertainty associated with such combustion model predictions. Third, to provide tools to reduce the computational cost associated to turbulencechemistry interaction models for non conventional regimes. Experimental and numerical campaigns on the ULB furnace were carried out to assess MILD conditions in the furnace for different methane/hydrogen mixtures in different working conditions. Operating in MILD conditions above 50% content in H2 proved to be quite challenging, even while varying the equivalence ratio. Different solutions were investigated, aiming at delaying the injection of the fuel to allow mixing with exhausts before chemical reactions could take place. This was accomplished by modifying the injection system by increasing the fuel lance length, allowing MILD conditions up to a hydrogen content of 75%. Varying the air injector geometry had no significant effect on the combustion regime: co-axial and different multi-hole injectors were tested to modify the inlet momentum flux ratio without a noticeable effect on the combustion regime. Fuel dilution with CO2 or H2O was also found to lead to smoother temperature fields. An experimental campaign was then performed to assess the fuel flexibility of the ULB furnace fired with ammonia/hydrogen blends. Optimal operating configurations in terms of trade-off between NOx emissions and ammonia slip were identified. Ammonia slip was found to be negligible in lean conditions, while it became relevant in stoichiometric conditions. The optimal working point was identified for the equivalence ratio f = 0.95, which enabled reduced NO emissions with respect to leaner conditions while keeping NH3 slip below 10 ppm. In stoichiometric conditions, peak NO production was observed for both tested air injectors (internal diameter ID 16 mm and 25 mm) at 10% ammonia in volume and 90% hydrogen in volume (N10H90) fuel composition. Emissions then decreased up to extinction, which occurred above N80H20 fuel composition. The larger air injector (ID25) helped control pollutants emissions, as the resulting increased residence time enhanced NO conversion to N2. Experimental data were employed to validate the combustion model for ammonia/hydrogen oxidation. Temperature predictions were found to be satisfactory and nearly insensitive to different kinetic mechanisms. On the other hand, NOx emissions showed substantial differences between predictions made with different mechanisms. Because of this, an uncertainty quantification study on the kinetic scheme employing a well-stirred reactor network representing the ULB furnace was performed to assess the propagation of the kinetic sub-model uncertainty on the NOx emissions predictions. Sensitivity and rate of production analyses identified influential reactions for NO production, and two kinetic mechanisms were determined, representing the minimum and maximum NO distribution. Over-prediction of NO was still noticeable, therefore the characterisation of said reactions requires improvement, especially for diluted and stoichiometric conditions. To quantify the uncertainty associated with the combustion model predictions in nonconventional combustion regimes, an uncertainty quantification study for RANS simulations of the Cabra flame was then carried out. Polynomial Chaos Expansion was employed to build surrogate models to assess the effect of parametric uncertainty of the Partially-Stirred Reactor combustion model. In particular, the parameters controlling the local mixing time scale were investigated. Results indicate that that the two dissipation coefficients have the strongest influence on the variability of the model response across all regions of the flame. An optimisation procedure which employs experimental results as targets was proposed to determine an optimal set of parameters which yield improved predictions of the quantities of interest. Last, a framework to reduce the computational cost of finite-rate models in LES calculations was presented. The Sample-Partitioning Adaptive Chemistry (SPARC) methodology was advanced with automated algorithms for clustering and mechanism reduction and it was applied to LES of the Adelaide Jet-in-Hot-Coflow burner. The procedure begins with the generation of a sample composition space from cheap 1D simulations. The composition space is clustered and then for each cluster a reduced kinetic mechanism is built. Once these pre-processing steps are concluded, during the CFD simulation, each cell is classified and assigned to a cluster and the respective reduced mechanism is employed. High accuracy with respect to the reference LES computed with the full detailed mechanism was observed, and the chemistry integration time step was more than halved.

Documents en relation

DI-fusion