Thesis title: Toward Climate-Neutral Aviation: Uncertainty Quantification, Bayesian Inference, and Optimization supporting Multi-Fidelity CFD
Mid-term climate neutrality cornerstone policies by international governments and institutions target a climate-neutral aviation system by 2050. In this context, sustainable aviation fuels (SAFs) represent a drop-in key enabling technology to foster an effective transition of the aviation sector towards net zero carbon, with blending ratios of conventional and alternative jet fuels bound to increase. Nonetheless, peculiar and unexplored properties of unconventional fuel blends may profoundly impact the performance and safe operability of jet engines in terms of altitude relight, lean blow-out, and cold-start ignition. To assess the actual technology readiness level of these drop-in options, computational fluid dynamics (CFD) offers a pivotal active support tool, partially or entirely replacing vast, expensive, and practically difficult experimental campaigns. Yet, characterizing turbulent reacting multiphase flows in combustion systems fueled with alternative jet fuel blends via CFD inherently exhibits uncertainties concerning turbulence closure in complex flow fields, distinctive oxidation pathways modeled via finite-rate chemistry, spray characteristics and gas-liquid interaction, and the representation of real fuels - typically mixtures of thousands of hydrocarbons - through physicochemical surrogate mixtures. Thus, to develop cost-efficient, still accurate CFD models, barely affected by any modeling bias and potentially driving robust design optimization of large-scale SAF-fueled combustion devices, a multi-fidelity framework is proposed. Notably, the multi-fidelity strategy originally hinges on a restrained amount of reference data from high-fidelity experimental and numerical campaigns. After that, non-intrusive spectral projection uncertainty quantification (UQ) techniques are exploited to assess the impact of any modeling uncertainty on the variability of the outcomes delivered by low-fidelity numerical models, thus driving additional high-fidelity campaigns aimed at reducing or even eliminating the most relevant uncertainty sources via Bayesian inference and optimization algorithms. This way, an augmented- or quantified-fidelity solver is formulated, constituting a reliable instrument to conduct large-scale numerical campaigns providing thorough insights into distinguishing spray and combustion processes in state-of-the-art SAF-fueled aeronautical combustion devices and eventually drive computer-aided engineering (CAE) design optimization processes. In the present research study, advanced mathematical techniques leveraging the Bayesian setting are adopted to investigate how model-embedded and intrinsic modeling uncertainties propagate to quantities of interest in variable-fidelity CFD solvers addressing application studies of relevance to both aeronautical and space propulsion, highlighting their potentiality in enhancing ad-hoc sub-model calibration against high-fidelity data. Nonetheless, a novel kinetic mechanism genetic optimization algorithm driven by computational singular perturbation (CSP) theory is introduced, trading off accuracy for computational effort in compact application-tailored skeletal schemes reproducing primary combustion observables in real-world combustion systems. Lastly, an innovative Bayesian framework, named BayeSAF, for developing physicochemical surrogates of both conventional and non-conventional fuels is presented, extending the surrogate formulation process to include fully representative combustion observables of real fuels, such as the ignition delay time, via polynomial chaos expansion (PCE) representations.