Modeling In-Flight Ice Accretion Under Uncertain Conditions

Giulio Gori, Pietro Marco Congedo, Olivier Le Maı̂tre, Tommaso Bellosta, Alberto GuardoneJournal of Aircraft 2021 · DOI

TL;DR

This paper tackles the challenge of predicting ice accretion on aircraft under uncertain environmental conditions. By developing a novel non-linear regression approach, the authors significantly improve the accuracy of ice accretion models, particularly in critical regions near ice accretion limits, leading to more reliable aircraft safety and design.

What Problem Does This Paper Address?

In-flight ice accretion is a serious threat to aircraft safety, impacting performance and potentially leading to accidents. Predicting this phenomenon accurately is crucial for designing effective anti-icing systems and for aircraft certification. However, real-world flight conditions are inherently uncertain, with variations in factors like air temperature, liquid water content, and angle of attack. Traditional computational models struggle to capture the complex, non-linear behavior of ice accretion, especially in regions where ice buildup is minimal or absent. This difficulty arises because standard uncertainty quantification (UQ) techniques, like linear polynomial chaos (PC) expansion, fail to accurately approximate the model's behavior in these "scarcely affected" regions, leading to significant errors in predictions.

Methods

The authors employ a multi-disciplinary computational framework that couples a CFD solver (SU2) with a particle tracking code (PoliDrop) and an ice accretion solver (PoliMIce). To address the limitations of standard UQ techniques, they propose a novel non-linear regression approach for approximating the ice thickness. This method combines a polynomial chaos expansion with a thresholding function. This thresholding function ensures that the predicted ice thickness remains non-negative, a crucial physical constraint that linear methods often violate, especially near ice accretion limits. The non-linear regression problem is solved using a coordinate descent strategy with elastic net regularization. This approach is compared against the standard linear PC method to demonstrate its improved accuracy.

Datasets

The study utilizes experimental data from three test cases conducted at the NASA Glenn Icing Research Tunnel (IRT) facility. These cases involve a two-dimensional subsonic airfoil under different icing conditions:

  • Test Case A: Focuses on collection efficiency prediction for a GLC-305 airfoil under warm weather conditions.
  • Test Case B: Investigates rime ice accretion on a NACA0012 airfoil, propagating uncertainties to ice thickness predictions.
  • Test Case C: Examines glaze ice accretion on a NACA0012 airfoil, also focusing on ice thickness predictions.

The paper does not introduce new datasets but rather uses existing experimental data for validation.

Key Insights and Ideas

  • The critical role of non-linearity: The paper highlights that the non-linear nature of ice accretion, particularly the sharp transition at ice accretion limits, is a major challenge for standard UQ methods.
  • Physical constraints matter: Forcing physical constraints, like non-negative ice thickness, is essential for accurate modeling, especially in regions with low ice buildup. The proposed thresholded PC approach effectively enforces this.
  • Dominant uncertainties vary by regime: In rime icing, angle of attack and cloud liquid water content are key drivers of uncertainty. In glaze icing, ambient temperature becomes the most significant factor.
  • Suction side sensitivity: The airfoil's suction side is identified as a critical region for water collection, exhibiting high variability in collected water despite a relatively stable impinging limit. This has implications for anti-ice system design.

Key Results

  • The proposed non-linear regression approach significantly reduces the RMSE of ice thickness predictions, especially near ice accretion limits, compared to the standard linear PC method.
  • In Test Case B (rime ice), the thresholded formulation improved accuracy near accretion limits, where linear PC struggled to predict non-negative ice thickness.
  • In Test Case C (glaze ice), the non-linear approach showed substantial error reduction, particularly at the suction side accretion limit.
  • Sensitivity analyses revealed that AoA and MVD are the most influential uncertain parameters for rime ice accretion, while ambient temperature dominates for glaze ice.
  • The study successfully reproduced experimental measurements from the NASA IRT, demonstrating the predictive capabilities of the computational framework.

Limitations and Future Work

  • Approximation of the full icing model: The authors acknowledge that approximating the full icing model, especially in regions with limited ice buildup, remains a critical challenge.
  • Cloud representation: The current model uses a simplified representation of clouds (e.g., constant MVD). Future work aims to introduce a more detailed description of cloud properties, including a probability distribution of droplet diameters.
  • Reduced order models: The uncertainty inherent in reduced-order models for droplet drag and wall interaction parameters will be investigated.
  • Aerodynamic flow field uncertainty: The role of uncertainty in the aerodynamic flow field (e.g., turbulence) needs further assessment.
  • Heat transfer contributions: The evaluation of heat transfer contributions due to convection, kinetic release, and evaporation under uncertain conditions requires further research.
  • Scalable methods: Developing efficient and scalable methods for propagating uncertainty in in-flight ice applications is identified as a key research opportunity due to the "curse of dimensionality."

Summary updated

Abstract

In-flight ice accretion under parametric uncertainty is investigated. Three test cases are presented that reproduce experiments carried out at the NASA’s Glenn Icing Research Tunnel facility. A preliminary accuracy assessment, achieved by comparing numerical predictions against experimental observations, confirms the robustness and the predictiveness of the computerized icing model. Besides, sensitivity analyses highlight the variance of the targeted outputs with respect to the different uncertain inputs. In rime icing conditions, a predominant role is played by the uncertainty affecting the airfoil angle of attack, the cloud liquid water content, and the mean volume diameter of droplets. In glaze icing conditions, the sensitivity analysis shows instead that the output variability is due mainly to the ambient temperature uncertainty. Moreover, this paper exposes a challenge inherent in the approximation of the full icing model by means of standard (linear) polynomial chaos regression techniques. The complexity is related to the approximation of the model behavior in domain regions scarcely affected by ice buildup. To mitigate this issue, a nonlinear regression method is proposed and applied.
BibTeX
@article{gori2021modeling,
  author = {Giulio Gori and Pietro Marco Congedo and Olivier Le Maı̂tre and Tommaso Bellosta and Alberto Guardone},
  title = {{Modeling In-Flight Ice Accretion Under Uncertain Conditions}},
  journal = {Journal of Aircraft},
  year = {2021},
  doi = {10.2514/1.c036545},
}
Figure