Sensitivity of ice accretion and aerodynamic performance degradation to critical physical and modeling parameters affecting airfoil icing

Modeling In-Flight Ice Accretion Under Uncertain Conditions

“…Recently, Prince et al [30] presented the evaluation of the effects of some input uncertainty e.g., ice surface roughness and density, w.r.t. specific features characterizing the ice shape resulting from the accretion process i.e., ice accretion limits, horns position, horns angle and others.…”

Sensitivity Study of Ice Accretion Simulation to Roughness Thermal Correction Model

“…The points for the 190 sample simulations are plotted and show that the metamodel M1 for the HAX model has a good precision, with an absolute mean error between the RANS-calculated values and the NIPCE-predicted values of 0.90 W/m 2 /K and a linear regression coefficient of 0.9999. This regression coefficient value is higher than the ones commonly accepted in the literature for sensitivity analysis [32]. The LOO error for this metamodel, M1, at x = 11.36 cm is 4.25 × 10 −4 .…”

“…These results highlight the importance of the roughness model in the simulation, whose parameters have to be carefully selected, since their variation leads to bigger changes to the ice accretion shape than the temperature variation within 1 K. This relative importance of roughness on the accretion sensitivity compared to other factors related to ambient conditions has also been noticed in the literature. In [32], Raj et al showed that the roughness plays a bigger role in the accretion shape than the ice density, the droplet distribution and the evaporation.…”

“…Despite UQ's wide use in aeronautics and heat transfer applications, no published paper has yet thoroughly characterized, from a statistical point of view, the sensitivity of ice accretion to surface roughness parameters. The closest related work is from Raj et al [32], where the sensitivity of the ice accretions to the equivalent sand grain roughness is compared to the sensitivity to ice density, droplet distribution and number of ice layers. However, no comparison is made for the roughness parameters between them.…”

“…The NIPCE has low sample requirements compared to the traditional Monte-Carlo methods used in the past [33]. In similar applications, Raj et al [32] use the radial basis function (RBF) metamodeling. The performance of polynomial-type metamodels, like the NIPCE, compared to RBF methods, is better for small sample designs, giving higher regression coefficients [34].…”