Uncertainty Quantification Analysis of RANS of Spray Jets

Abstract

Parametric uncertainty is propagated through Reynolds-averaged Navier-Stokes (RANS) computations of a prototypical acetone/air aerosol stream flowing in a dry air environment. Two parameters are considered as uncertain: the inflow velocity dissipation and a coefficient that blends the discrete random walk and the gradient-based dispersion models. A Bayesian setting is employed to represent the degree of belief about the parameters of interest in terms of probability theory, such that uncertainty is described with probability density functions. Random variables are represented by means of polynomial chaos expansions. The sensitivity of mean axial velocity and mean vapor mass fraction to the uncertain parameters is discussed.

Uncertainty quantification analysis of Reynolds-averaged Navier–Stokes simulation of spray swirling jets undergoing vortex breakdown

“…[20][21][22][23] Once the PDFs are known, the submodel uncertainty may be propagated to the QoIs through low-fidelity RANS simulations, employing a polynomial chaos expansion (PCE) representation of the random variables (RVs) being involved to reduce the number of required simulations. [24][25][26][27][28] This way, CFD results become supported by reliability measures, such as error bars or confidence intervals, similar to the usually adopted representations of experimental results. Moreover, the derivation of a surrogate model in terms of a PCE naturally offers the opportunity to assess the sensitivity of the output variance to each uncertain parameter, e.g., in terms of the so-called Sobol indices.…”

Impact of scalar mixing uncertainty on the predictions of reactor-based closures: Application to a lifted methane/air jet flame

“…We focus on the latter source of uncertainty. Uncertainty quantification (UQ) techniques were already employed to propagate uncertainties associated with chemical kinetics rate coefficients [11] and subfilter models' coefficients, e.g., Smagorinksy constant and turbulent Prandtl and Schmidt numbers [12] or spray dispersion models [13].…”

Direct numerical simulation of Vortex Breakdown in Evaporating Dilute Sprays

“…With respect to the latter, Bayesian calibration techniques MacKay (2005) can be employed to statistically characterize the uncertainty sources which affect the sub-models embedded into the lower-fidelity approaches, i.e., LES and RANS, addressing liquid phase behavior in multiphase reacting flows, e.g., droplet dispersion. In this regard, the impact of spray sub-models' uncertain parameters on the major observables can be assessed through nonintrusive spectral projection techniques Ciottoli et al (2020b); Liberatori et al (2021); Cavalieri et al (2023); Liberatori et al (2023), thus returning an overview of those model uncertainties that require further investigation through high-fidelity campaigns.…”