Hydrofoils operating in a stable super-cavitating regime present great promise in designing high-performance marine vehicles with cruising speeds exceeding 100 knots. Due to the inherent complexity and multi-phase nature of the turbulent flow, assessing the performance of such hydrofoils for a wide range of operating conditions becomes a formidable task. Here we address this challenge by putting forth a data-driven multi-fidelity framework that is able to combine simplified computational models with a small number high-fidelity simulations and/or experimental data. The compositional synthesis of these variable fidelity information sources leads to significant computational expediency gains, and enables the construction of accurate predictive surrogates for a given hydrofoil performance metric. Using a Bayesian nonparametric approach based on Gaussian process priors, the resulting multi-fidelity surrogates can also naturally quantify uncertainty due to noisy or incomplete data. We demonstrate the effectiveness of the proposed framework by building stochastic response surfaces for the lift over drag ratio of a wedge-shaped super-cavitating hydrofoil operating in different flow regimes, as controlled by the angle of attack to the incoming flow and the cavitation index. In particular, we consider three information sources: 2D RANS simulations (low-fidelity), a 2D RANS solver corrected for spanwise effects through a reformulated lifting line theory (intermediate fidelity), and a fully 3D RANS solver (high-fidelity). We also show how noisy experimental data can be seamlessly incorporated in the workflow, and we validate our predictions against the cavitation tunnel experiments of Kermen et al.
展开▼
