BayeSAF: A Bayesian Framework for Modeling Physicochemical Surrogates of Sustainable Alternative Fuels

Mid-term climate neutrality cornerstone policies by international governments and institutions target a climate-neutral aviation system by mid-century \cite{Afonso2023}. In this regard, blending conventional and “drop-in” sustainable aviation fuels (SAFs) currently represents a key enabling technology to foster an effective transition of the aviation sector towards net zero carbon. However, the unusual properties of alternative jet fuels may profoundly impact jet engines' performance and safe operability in terms of altitude relight, lean blow-out, and cold-start ignition, as well as emission levels of specific pollutants. To assess the actual technology readiness level (TRL) of these drop-in options, computational fluid dynamics (CFD) offers a pivotal active support tool, partially or entirely replacing vast, expensive, and practically complex experimental campaigns. Yet, the numerical characterization of combustion systems fueled with alternative jet fuel blends inherently exhibits modeling challenges, including the representation of real fuels through physicochemical surrogate mixtures, and resulting in scarce CFD analyses of non-conventional aviation fuel spray and combustion characteristics available in the open literature. In this sense, existing strategies to formulate surrogate mixtures typically hinge on genetic optimization algorithms, which address complex combustion behaviors by replicating lumped, non-fully representative, one-parameter combustion property targets (CPTs), such as the cetane number. In the present research study, we illustrate a novel Bayesian framework, named BayeSAF, which fosters employing polynomial chaos expansion (PCE) representations of fully representative combustion observables of conventional and alternative aviation fuels under jet-engine-relevant conditions - such as the ignition delay time, the laminar flame speed, and the S-shaped curve denoting flame response to aerodynamic straining - instead of lumped CPTs. This way, a Bayesian-based emulation of physicochemical properties paves the way toward using ad-hoc surrogate mixtures in CFD codes.