A systematic domain reduction strategy is proposed to mitigate the high computational cost of performing a large-eddy simulation (LES) for predicting underwater radiated noise from a marine propeller. The key concept of this strategy is to determine where detailed analysis is necessary beforehand and accordingly reduce the computational domain. The strategy comprises the following steps: (1) performing a Reynolds-averaged Navier–Stokes (RANS) simulation in the full domain, including the entire hull; (2) quantitatively analyzing the region affected by the propeller; (3) reducing the domain to encompass most of the propeller’s effect; and (4) conducting an LES in the reduced domain, using the flow field from the RANS simulation as an inlet boundary condition. By implementing the proposed strategy, the LES domain length is reduced by up to twice the propeller diameter in the upstream direction from the propeller, representing only 7% of the hull’s total length. The strategy’s effectiveness is validated by comparing the simulation results with the experimental data obtained at the Korea Research Institute of Ships and Ocean.

Turbine blade tip leakage flows occur between the unshrouded rotor and stationary casing. Such flow is one of the main sources of aerodynamic loss in a turbine. Many different methods have been proposed to reduce this loss, including the squealer tip and winglet design. In this study, numerical analysis showed that a large fraction of the leakage flow occurs after 50% of the axial chord. Thus, an additional rim was installed at the mid-point of the cavity, in an attempt to block the leakage flow to the suction side and redirect it back to the pressure side. The effect of an extended pressure-side winglet was also investigated and compared with the base squealer tip. The passage velocity field between the blades of a linear cascade was measured at 0, 25, 50, 75, and 100% of the axial chord using a 5-hole probe to assess the development of flow structures responsible for the tip leakage loss. The total pressure loss coefficient distribution was measured downstream of the cascade. Performance results from the different tip geometries were experimentally compared with each other, and also used to validate the numerical results. The winglet design showed the best performance. This design does not have a pressure-side rim, and thus the tip leakage flow has less hindrance passing over the suction-side rim, which creates a strong coherent tip leakage vortex compared to that of the squealer tip. This leads to an increase in loss beyond 90% of the span, but the counter-rotating tipwall passage vortex underneath interacts with the tip leakage vortex to reduce the loss in the 70–80% span region by a greater extent. Therefore, the pressure-side winglet design has an overall loss that is 28.3% smaller than that of the baseline squealer tip. The additional cavity rim did not show any noticeable improvements, possibly due to the angle at which it was placed, and thus needs further investigation.

An artificial neural-network-based subgrid-scale (SGS) model, which is capable of predicting turbulent flows at untrained Reynolds numbers and on untrained grid resolution is developed. Providing the grid-scale strain-rate tensor alone as an input leads the model to predict a SGS stress tensor that aligns with the strain-rate tensor, and the model performs similarly to the dynamic Smagorinsky model. On the other hand, providing the resolved stress tensor as an input in addition to the strain-rate tensor is found to significantly improve the prediction of the SGS stress and dissipation, and thereby the accuracy and stability of the solution. In an attempt to apply the neural-network-based model trained for turbulent flows with a limited range of the Reynolds number and grid resolution to turbulent flows at untrained conditions on untrained grid resolution, special attention is given to the normalisation of the input and output tensors. It is found that the successful generalization of the model to turbulence for various untrained conditions and resolution is possible if distributions of the normalised inputs and outputs of the neural network remain unchanged as the Reynolds number and grid resolution vary. In a posteriori tests of the forced and the decaying homogeneous isotropic turbulence and turbulent channel flows, the developed neural-network model is found to predict turbulence statistics more accurately, maintain the numerical stability without ad hoc stabilisation such as clipping of the excessive backscatter, and to be computationally more efficient than the algebraic dynamic SGS models.

Higher annual energy production can be obtained by joint optimization which considers active yaw control in the layout design stage. Although accurate representation of a non-centrosymmetric three-dimensional yawed wake is necessary for the joint optimization of a realistic wind farm, it has not been considered. Furthermore, non-convexity in the joint optimization becomes severe because the layout and yaw angles have to be optimized for all wind directions considering non-centrosymmetric three-dimensional yawed wakes, leading to a not globally but locally optimal layout. To tackle the difficulty, a particle-swarm-optimization-based method which is capable of large-scale non-convex joint optimization is developed. In the present method, a farm layout is globally optimized with simultaneous consideration of yaw angles for various wind speeds and directions. The use of random initial particles which consist of the layout and yaw angles of wind turbines prevent from obtaining a locally optimal layout caused by non-convexity of the problem. The improvement in the annual energy production by the present simultaneously optimized layout is demonstrated.

A second-order total variation diminishing (TVD) method with variable flux limiters is proposed to overcome the non-realizability issue, which has been one of major obstacles in applying the conventional second-order TVD schemes to the moment transport equations. In the present method, a realizable moment set at a cell face is reconstructed by allowing the flexible selection of the flux limiter values within the second-order TVD region. Necessary conditions for the variable flux limiter scheme to simultaneously satisfy the realizability and the second-order TVD property for the third-order moment set are proposed. The strategy for satisfying the second-order TVD property is conditionally extended to the fourth- and fifth-order moments. The proposed method is verified and compared with other high-order realizable schemes in one- and two-dimensional configurations, and is found to preserve the realizability of moments while satisfying the high-order TVD property for the third-order moment set and conditionally for the fourth- and fifth-order moments.

Growth of a Cu(In, Ga)Se2 (CIGS) layer during a selenization process is numerically studied to understand mechanisms for formation of stains on large-area substrates batched. CIGS layers need to be uniformly deposited onto the substrates to obtain even conversion efficiency. However, it is difficult to control growth of large-area CIGS layers due to turbulent thermal-fluid flow leaving stains on the substrates. In the present research, the selenization process for an industrial-scale substrates of which sizes are order of square-meters is considered with integrated simulations of detailed key physical processes such turbulent convection, convective-radiative-conductive heat transfer, and chemical reactions. Ascending or descending gas generated by heaters is identified by the time-averaged velocity fields. Descending flow in the passages between substrates produces uneven flow rates across the substrates leading to inhomogeneous supply of heat energy and gas species to the surface chemical reactions. The uneven temperature distribution is the major cause for the stain formation on the substrates. Gross shapes of the stains are found to be well matched with the predicted velocity contour of gas flow above the substrate. The stains are expected to be alleviated by rectifying gas flow such that flow rates become uniform across substrates before entering the passages.

Large-eddy simulation of turbulent flow around a 1:8 scale 15-ton heavy vehicle model is performed at Reynolds number of 9.1×105 to study unsteady characteristics of the surface forces. The drag coefficient and profiles of the mean streamwise velocity in the wake region are in good agreement with the experimental counterparts. Time histories of the drag, lift, and side forces are collected to analyze the frequency characteristics of forces applied on the vehicle surface. Velocity fluctuations near the vehicle surface are also obtained to study spectral features of the flow field. Results show that fluctuations of the streamwise velocity in the cab roof region are closely related to the drag frequency, whereas wake shedding heavily affects the frequency characteristics of the side force.

A novel multi-condition multi-objective optimization method that can find Pareto front over a defined condition space is developed using deep reinforcement learning. Unlike the conventional methods which perform optimization at a single condition, the present method learns correlations between conditions and optimal solutions. The exclusive capability of the developed method is examined in solutions of a modified Kursawe benchmark problem and an airfoil shape optimization problem. The solutions include nonlinear characteristics which are difficult to be resolved using conventional optimization methods. Pareto front with high resolution over a condition space is successfully determined in both problems. Compared with multiple operations of a single-condition optimization method for multiple conditions, the present multi-condition optimization method shows a greatly accelerated search of Pareto front by reducing the required number of function evaluations. An analysis of aerodynamic performance of optimally designed airfoils confirms that multi-condition optimization is indispensable to avoid significant degradation of target performance for varying flow conditions.

To save lives and reduce damage from the destructive impacts of a typhoon, an accurate and fast forecast method is highly demanded. Particularly, predictions for short lead times, known as nowcasting, rely on fast forecasts allowing immediate emergency plannings in the affected areas. In this paper, we propose a generative adversarial network that operates on a single graphics processing unit, to predict both the track and intensity of typhoons for short lead times within fractions of a second. To investigate the effects of meteorological variables on typhoon forecasts, we conducted a parameter study for 6-h track predictions. The results of the study indicate that learning velocity, temperature, pressure, and humidity along with satellite images have positive effects on prediction accuracy. To address the limited access to observational data and facilitate predictions for 12-h intervals, we replaced satellite images with reanalysis data of the total cloud cover and vorticity fields. This replacement led to an increase in data from 76 to 757 typhoons, and it reduced the error of the 6-h track forecasts by 23.5%. The best combination of the parameter study yields track predictions in intervals of 6 and 12 h with the corresponding averaged absolute errors of 44.5 and 68.7 km. Typhoon intensities are predicted by extracting information from generated velocity fields with averaged hit rates of 87.3% and 83.2% for 6- and 12-h interval forecasts, respectively. For typhoons after 1994, tracks and intensities for 12-h intervals are compared to forecasts from the Joint Typhoon Warning Center and Regional Specialized Meteorological Center Tokyo.

The relationship between inertial particle pair dynamics in the dissipation range and local turbulence characteristics is investigated using direct numerical simulations in homogeneous isotropic turbulence at a Reynolds number of 110 based on the Taylor microscale. Seventeen million sub-Kolmogorov-sized particles with Stokes numbers of 0.5, 1.0, 1.5, and 2.0 are tracked backward in time to investigate the formation of the relative velocity of inertial particles in the dissipation range. The numerical experiment shows that particle pairs take different paths and sample different underlying flows depending on the intensity of the local turbulence activity at final locations where the distance between particles is much smaller than the Kolmogorov length scale. Taking different paths depending on the intensity of the local turbulence activity is described by the sling time, which is defined as the time when particle pairs are slung out from the underlying flow. Particle pairs with shorter sling time values than the characteristic time scale of particles are detached from the dissipation range of flow and approach each other in a ballistic fashion. In contrast, the large-scale flow guides particle pairs with longer sling time than the characteristic time scales. In particular, particle pairs forming caustics with the same sling time are detached from similar flows regardless of the magnitude of particle relative velocities. Samplings of similar flows at sling moments cause the scaling relationship between relative velocities of particles and sling time. These behaviors of particle pairs, especially for pairs with caustics in relative velocities, are also observed in the range of considered Stokes numbers. The relationship between the intensity of the local turbulence activity and the paths taken by particles highlights the importance of considering a preferential sampling of flow in understanding inertial particle pair dynamics.