Marine propellers are studied in design and off-design modes of operation like crashback, where the propeller rotates in reverse while the vehicle is in forward motion. Past experiments (Jessup et al., Proceedings of the 25th Symposium on Naval Hydrodynamics, St John's, Canada, 2004; Proceedings of the 26th Symposium on Naval Hydrodynamics, Rome, Italy, 2006) studied the marine propeller David Taylor Model Basin 4381 in the open-jet test section of the 36-inch variable pressure water tunnel (VPWT). In crashback, a significant discrepancy with unclear sources exists between the mean propeller loads from the VPWT and open-water towing tank (OW) experiments (Ebert et al., 2007 ONR Propulsor S & T Program Review, October, 2007). We perform large-eddy simulation at $Re=561\,000$ and advance ratios $J=-0.50$ and $-0.82$ with the VPWT geometry included, contrasting to the unconfined (OW) case at those same $J$ and $Re=480\,000$. We identify and delineate the water tunnel interference effects responsible, and demonstrate that these effects resemble those of a symmetric solid model or bluff body. Solid blockage due to jet expansion and nozzle blockage due to proximity to the tunnel nozzle are identified as the primary interference effects. Their impact varies with the advance ratio $J$ and strengthens for higher magnitudes of $J$. The effective length scale to assess the severity of interference effects is found to be larger than the vortex ring diameter.

This paper investigates the impact of large-scale turbulence on the aerodynamic characteristics of a pitching wind turbine blade at Reynolds number 135 000, whose cross-section is a NACA0012 airfoil with constant chord length. Large-eddy simulations at reduced frequencies, $k_{red} = 0.05$ and 0.1, were validated against reference data from the literature. An efficient method capable of generating synthetic large-scale turbulence at the inlet was applied by using two streamwise integral length scales $L_{x} = 1c$ and 1.5$c$, which represent energetic turbulence eddies at the height where the wind turbine operates. For $k_{red}= 0.1$, the change in the maximum lift coefficient at the dynamic stall angle near the maximum angle of attack is on average 20 % lower and during the downstroke it is on average 22 % lower, compared with the smooth inflow. A higher reduced frequency ($k_{red} = 0.2$) apparently does not further change the lift, drag and moment coefficients, and the inflow turbulence disordered leading-edge vortices. The turbulent shear stress and the phase-averaged dispersive shear stress in the wake are of the same magnitude, but with negative and positive signs, respectively, suggesting that the large-scale phase-averaged fluctuations transfer momentum in the opposite direction compared with the turbulent fluctuations, reducing the drag on the suction-side flow, and subsequently increasing the averaged lift coefficient. This demonstrates the critical importance of the large-scale non-turbulent unsteadiness in the wake of a pitching wing.

Compressible anisothermal flows, which are commonly found in industrial settings such as combustion chambers and heat exchangers, are characterized by significant variations in density, viscosity, and heat conductivity with temperature. These variations lead to a strong interaction between the temperature and velocity fields that impacts the near-wall profiles of both quantities. Wall-modeled large-eddy simulations (LESs) rely on a wall model to provide a boundary condition, for example, the shear stress and the heat flux that accurately represents this interaction despite the use of coarse cells near the wall, and thereby achieve a good balance between computational cost and accuracy. In this article, the use of graph neural networks for wall modeling in LES is assessed for compressible anisothermal flow. Graph neural networks are a type of machine learning model that can learn from data and operate directly on complex unstructured meshes. Previous work has shown the effectiveness of graph neural network wall modeling for isothermal incompressible flows. This article develops the graph neural network architecture and training to extend their applicability to compressible anisothermal flows. The model is trained and tested a priori using a database of both incompressible isothermal and compressible anisothermal flows. The model is finally tested a posteriori for the wall-modeled LES of a channel flow and a turbine blade, both of which were not seen during training.

As the Reynolds number increases, the large-eddy simulation (LES) of complex flows becomes increasingly intractable because near-wall turbulent structures become increasingly small. Wall modeling reduces the computational requirements of LES by enabling the use of coarser cells at the walls. This paper presents a machine-learning methodology to develop data-driven wall-shear-stress models that can directly operate, a posteriori, on the unstructured grid of the simulation. The model architecture is based on graph neural networks. The model is trained on a database which includes fully developed boundary layers, adverse pressure gradients, separated boundary layers, and laminar–turbulent transition. The relevance of the trained model is verified a posteriori for the simulation of a channel flow, a backward-facing step and a linear blade cascade.

Large-eddy simulation (LES) using an unstructured overset numerical method is performed to study the flow around a ducted marine propeller for the highly unsteady off-design condition called crashback. Known as one of the most challenging propeller states to analyse, the propeller rotates in the reverse direction to yield negative thrust while the vehicle is still in forward motion. The LES results for the marine propeller David Taylor Model Basin 4381 with a neutrally loaded duct are validated against experiments, showing good agreement. The simulations are performed at Reynolds number of 561 000 and an advance ratio $J=-0.82$. The flow field around the different components (duct, rotor blades and stator blades) and their impact on the unsteady loading are examined. The side-force coefficient $K_S$ is mostly generated from the duct surface, consistent with experiments. The majority of the thrust and torque coefficients $K_T$ and $K_Q$ arise from the rotor blades. A prominent contribution to $K_Q$ is also produced from the stator blades. Tip-leakage flow between the rotor blade tips and duct surface is shown to play a major role in the local unsteady loads on the rotor blades and duct. The physical mechanisms responsible for the overall unsteady loads and large side-force production are identified as globally, the vortex ring and locally, leading-edge separation as well as tip-leakage flow which forms blade-local recirculation zones.

The interaction between turbulence in a minimal supersonic channel and radiative heat transfer is studied using large-eddy simulation. The working fluid is pure water vapour with temperature-dependent specific heats and molecular transport coefficients. Its line spectra properties are represented with a statistical narrow-band correlated-k model. A grey gas model is also tested. The parallel no-slip channel walls are treated as black surfaces concerning thermal radiation and are kept at a constant temperature of 1000 K. Simulations have been performed for different optical thicknesses (based on the Planck mean absorption coefficient) and different Mach numbers. Results for the mean flow variables, Reynolds stresses and certain terms of their transport equations indicate that thermal radiation effects counteract compressibility (Mach number) effects. An analysis of the total energy balance reveals the importance of radiative heat transfer, compared to the turbulent and mean molecular heat transport.

To investigate the flame stability in a solid-fuel ramjet combustor, time-accurate calculations using a compressible flow solver with a modified Godunov flux-splitting scheme have been performed on high Reynolds number turbulent non-premixed reacting flows over a backward-facing step with mass bleed on one wall. The combustion process considered was a one-step, irreversible, and finite rate chemical reaction. The numerical results for reacting flows show that the two-dimensional (2-D) simulations can provide reasonable predictions on the dimensionless particle number decay rate and residence time in the flame holding recirculation zone, evolutions of both axial and transverse mean velocity profiles, and critical characteristic exhaust velocity separating the sustained combustion from the non-sustained combustion. In addition to the validation of 2-D reacting flow calculations, two- and three-dimensionally computed mean-velocity profiles are compared with existing experimental data for isothermal flows to check the suitability of 2-D simulations on capturing the large-scale mean flows.