High-fidelity Computational Fluid Dynamics Simulations Research Articles

Heat exchange performance evaluation inside a lattice panel using CFD analysis for an innovative aerospace anti-icing system

As the global push for sustainable development intensifies, the aerospace industry is exploring innovative solutions to reduce greenhouse gas emissions and optimize resource consumption. This paper investigates the performance of an advanced anti-icing system that integrates an onboard function within the primary structure of an aircraft, offering potential reductions in weight, cost, and energy consumption. To evaluate the heat exchange performance of this innovative system, we examine various lattice structures featuring different cell topologies, sizes, and densities. High-fidelity Computational Fluid Dynamics (CFD) simulations are utilized to assess the performance of the constituent cells within the lattice panel, with the goal of identifying the optimal solution that meets design requirements for heat transfer efficiency, pressure, and overall mass of the components. A statistical approach based on Design of Experiments (DoE) is employed to elucidate the relationship between the cell design parameters and the resulting thermal properties. Our findings contribute valuable insights for the development and implementation of advanced aerospace anti-icing systems, supporting the ongoing evolution of sustainable and efficient aviation technologies.

Critical effect of fore-aft tapering on galloping triggering for a trapezoidal body

The critical effect of the windward interior angles of elastically mounted trapezoidal bodies on a galloping instability is numerically investigated in this paper using two methodologies of high-fidelity computational fluid dynamics simulations and data-driven stability analysis using the eigensystem realization algorithm. A micro exploration of the dynamical response is processed to understand the mechanism underpinning the structural amplification at the initial stage of the galloping instability and the competition between wake and structural modes. It is observed that very small changes in the windward interior angle of an isosceles-trapezoidal body can provoke or suppress galloping – indeed, a small decrease or increase (low to $1^\circ$ ) of the windward interior angle from a right angle ( $90^\circ$ ) can result in a significant enhancement and complete suppression, respectively, of the galloping oscillations. This supports our hypothesis that the contraction and/or expansion (viz., fore-aft tapering and/or widening) of the cross-section in the streamline direction has potential influences on galloping triggering from the geometrical perspective. The data-driven stability analysis is also applied to verify and analyse this phenomenon from the perspective of modal analysis. The experimental measurements are also conducted in the wind tunnel to support this hypothesis.

Unsteady reduced order model with neural networks and flight-physics-based regularization for aerodynamic applications

Numerical simulation of unsteady fluid flow plays an important role in several areas of the aeronautical industry. Since high-fidelity computational fluid dynamics simulations could be prohibitive in terms of computational cost, data-driven reduced order models become a suitable alternative for efficiently predicting flow variables as long as the accuracy of such models is comparable to that obtained by the full order model counterpart. This is especially important for iterative design purposes, where a few target variables must be evaluated on a large number of possible parameters. Therefore, we propose a neural network based methodology to develop an unsteady reduced order model of the subsonic/transonic flow field on 2D aerodynamic profiles trained on high-fidelity computational fluid dynamics data. For the purpose of dimensionality reduction, either proper-orthogonal decomposition or autoencoders are employed. For the regression task, a gated recurrent unit neural network is used to map an unsteady Schroeder multi-sine signal of angle of attack along with its first and second time-derivatives to the solution of surface variables, such as coefficients of pressure and friction. In order to shed light on the inner workings of data-driven methods so it could be employed in the aircraft design process, we introduce a flight-physics-based regularization term to incorporate information about the calculation of integral coefficients, like drag and lift, into the machine learning training workflow. Using our method, airfoil flow variables of interest can be predicted at a fraction of the cost of classical methods without any considerable accuracy loss. We also provide a comparison between reduction methods and we show evidence that supports the use of the proposed flight-physics-based regularization for building unsteady reduced order models based on machine learning.

Investigation on the influence of vortex generator on particle resuspension

The vortex generator (VG) and its well-known effect in flow optimization are widely studied and employed across different engineering sectors. However, while the same working principles of VG may be well suited for the applications on surface-cleaning technologies, such promising potential is hardly, if any, explored in the published literature. Therefore, in the present study, the influence on flow-induced particle resuspension brought by a rectangular VG in a channel flow is investigated with the help of high-fidelity computational fluid dynamics simulations. Substantial increases of particle removal forces and resuspension rates are discovered in long, strip-like regions with reduced boundary-layer thickness resulted from the VG-induced vortices, and the enhancement effect is especially significant for configurations with the VG installed at a greater angle of attack. It is also shown that while the resuspension enhancements on the lower and the upper surfaces of the channel exhibit distinct statistical characteristics, having a VG in the channel improves the overall particle-removing capability of the channel flow by introducing higher surface-averaged removal forces and particle resuspension rates. Last but not least, the increase of resuspension rate is especially significant for the smaller, micron-scale particles which are otherwise hardly disturbed by a VG-less channel flow, and such resuspension-enhancement effect generally subsides with increasing particle size.

Detailed Simulations of a Three-Stage Supercritical Carbon Dioxide Axial Compressor with a Focus on the Shrouded Stator Cavities Flow

This paper describes the findings of detailed simulations performed to investigate the impact of seal teeth cavity leakage flow on the aerodynamic and thermal performance of a three-stage supercritical CO2 axial compressor. The study compares a shrouded stator configuration (with cavities) to a cantilevered stator configuration (without cavities) to highlight their differences. High-fidelity computational fluid dynamics simulations were performed using non-linear harmonic (NLH) and mixing plane assumptions, considering various possible rotor/stator interface configurations for mixing plane calculations. The key performance parameters for each case were compared, and the best-performing configuration selected for further analysis. The individual stage performance parameters are also examined and compared between the cantilevered and shrouded configurations. It was observed that in the shrouded case, the leakage flow enters the cavity downstream of the stator trailing edge and gets entrained into the primary flow upstream of the stator, leading to boundary layer changes at the hub and degradation of stator and downstream rotor performance. Vortical flow structures were also observed in the stator wells, which tended to change the flow angles around the region, thereby affecting mixing and velocity distribution, resulting in a slight deterioration of compressor performance. Additionally, the study examines windage heating due to shear work from rotating walls, including the seal teeth surface. The amount of shear work done on the leakage flow and the corresponding rise in fluid temperature were quantified, tabulated, and further compared with a simple analytical model, showing good agreement between them and, hence, validating the numerical approach used.

대형 비행갑판을 갖는 함정과 풍동시험 결과를 활용한 고신뢰도 함정 Airwake 예측

Developing high-fidelity Computational Fluid Dynamics (CFD) simulation methods used to evaluate the airwake characteristics along a flight deck of a large ship, the various kind of data such as actual ship measurement and wind tunnel results are required to verify the accuracy of CFD simulation. Inflow velocity profile at the bow, local unsteady flow field data around the flight deck, and highly reliable wind tunnel data which were measured after reviewing Atmospheric Boundary Layer (ABL) simulation and Reynolds Number effects were also used to determine the key parameters such as turbulence model, time resolution and accuracy, grid resolution and type, inflow condition, domain size, simulation length, and so on in STAR CCM+. Velocity ratio and turbulent intensity difference between Full-scale CFD and actual ship measurement at the measurement points show less than 2% and 1.7% respectively. And differences in velocity ratio and turbulence intensity between wind tunnel test and small-scale CFD are both less than 2.2%. Based upon this fact, the selected parameters in CFD simulation are highly reliable for a specific wind condition.

Optimisation and Analysis of Streamwise-Varying Wall-Normal Blowing in a Turbulent Boundary Layer

Skin-friction drag is a major engineering concern, with wide-ranging consequences across many industries. Active flow-control techniques targeted at minimising skin friction have the potential to significantly enhance aerodynamic efficiency, reduce operating costs, and assist in meeting emission targets. However, they are difficult to design and optimise. Furthermore, any performance benefits must be balanced against the input power required to drive the control. Bayesian optimisation is a technique that is ideally suited to problems with a moderate number of input dimensions and where the objective function is expensive to evaluate, such as with high-fidelity computational fluid dynamics simulations. In light of this, this work investigates the potential of low-intensity wall-normal blowing as a skin-friction drag reduction strategy for turbulent boundary layers by combining a high-order flow solver (Incompact3d) with a Bayesian optimisation framework. The optimisation campaign focuses on streamwise-varying wall-normal blowing, parameterised by a cubic spline. The inputs to be optimised are the amplitudes of the spline control points, whereas the objective function is the net-energy saving (NES), which accounts for both the skin-friction drag reduction and the input power required to drive the control (with the input power estimated from real-world data). The results of the optimisation campaign are mixed, with significant drag reduction reported but no improvement over the canonical case in terms of NES. Selected cases are chosen for further analysis and the drag reduction mechanisms and flow physics are highlighted. The results demonstrate that low-intensity wall-normal blowing is an effective strategy for skin-friction drag reduction and that Bayesian optimisation is an effective tool for optimising such strategies. Furthermore, the results show that even a minor improvement in the blowing efficiency of the device used in the present work will lead to meaningful NES.

E-POD investigations of turbulent premixed flame dynamics approaching lean blow-out conditions

Thanks to the continuous computational power increase, the use of high-fidelity computational fluid dynamics (CFD) simulations is nowadays customary, especially in the gas turbines design process. The extraordinary temporal and spatial detail of such analyses generate large datasets, which must be carefully studied to correlate different quantities and gain information to characterize the behavior of combustor designs. Several advanced post-processing tools have been proposed; however, the Extended-POD (E-POD) holds the greatest potential for turbulent combustion applications when the mutual influence of different quantities is the main goal of the investigation. The present work investigates the application of the E-POD to an LES model of a perfectly-premixed, swirl-stabilized, methane-air flame approaching Lean-Blow-Out. Leveraging the validation against the experimental data at two different operating conditions on a laboratory test case, the numerical model has been used to collect several quantities of interest for shedding light on the flow-flame interaction near the blow-out. The post-processing algorithm has been used to highlight the differences between two conditions approaching the extinction at distinct air-flow velocities. It has been found that, when the burner is operated with a higher velocity, the flame is subjected to a cyclic low-frequency breakdown around the internal recirculation zone, leading to an ingestion of cold products from the external parts of the combustor toward the center. Although other local effects acting on the flame brush have been found in both conditions, they are related mainly to higher order coherent structures with a lower energy content. As a result, their impact onto flame stability is found to be of secondary importance since their limited interaction with flame stabilization. The work shows that E-POD represents a powerful tool for investigating the key features of flame dynamics even at near-blow-out conditions, constituting a valid algorithm for interpreting the results of CFD analyses on gas turbines combustors.

NM-MF: Non-Myopic Multifidelity Framework for Constrained Multi-Regime Aerodynamic Optimization

The exploration and trade-off analysis of different aerodynamic design configurations requires solving optimization problems. The major bottleneck to assess the optimal design is the large number of time-consuming evaluations of high-fidelity computational fluid dynamics (CFD) models, necessary to capture the non-linear phenomena and discontinuities that occur at higher Mach number regimes. To address this limitation, we introduce an original non-myopic multifidelity Bayesian framework aimed at including expensive high-fidelity CFD simulations for the optimization of the aerodynamic design. Our scheme proposes a novel two-step lookahead policy to maximize the improvement of the solution quality considering the rewards of future steps, and combines it with utility functions informed by the fluid dynamic regime and the information extracted from data, to wisely select the aerodynamic model to interrogate. We validate the proposed framework for the case of a constrained drag coefficient optimization problem of a NACA 0012 airfoil, and compare the results to other popular multifidelity and single-fidelity optimization frameworks. The results suggest that our strategy outperforms the other approaches, allowing to significantly reduce the drag coefficient through a principled selection of limited evaluations of the high-fidelity CFD model.

Quantifying Equiaxed vs Epitaxial Solidification in Laser Melting of CMSX-4 Single Crystal Superalloy

The competition between epitaxial vs. equiaxed solidification has been investigated in CMSX-4 single crystal superalloy during laser melting as practiced in additive manufacturing. Single-track laser scans were performed on a powder-free surface of directionally solidified CMSX-4 alloy with several combinations of laser power and scanning velocity. Electron backscattered diffraction (EBSD) mapping facilitated identification of new orientations, i.e., “stray grains” that nucleated within the fusion zone along with their area fraction and spatial distribution. Using high-fidelity computational fluid dynamics simulations, both the temperature and fluid velocity fields within the melt pool were estimated. This information was combined with a nucleation model to determine locations where nucleation has the highest probability to occur in melt pools. In conformance with general experience in metals additive manufacturing, the as-solidified microstructure of the laser-melted tracks is dominated by epitaxial grain growth; nevertheless, stray grains were evident in elongated melt pools. It was found that, though a higher laser scanning velocity and lower power are generally helpful in the reduction of stray grains, the combination of a stable keyhole and minimal fluid velocity further mitigates stray grains in laser single tracks.

Accelerating High-Fidelity Thermal Process Simulation of Laser Powder Bed Fusion via the Computational Fluid Dynamics Imposed Finite Element Method (CIFEM)

• A finite element (FE) method is proposed to accelerate high-fidelity scanwise thermal process simulation in laser powder bed fusion (L-PBF) • Computational fluid dynamics (CFD) thermal fields are predicted by a dimensionality reduction based deep learning framework • Predicted thermal fields are imposed on the finite element (FE) solution domain within a relatively small region encompassing the melt pool, while heat diffusion effects elsewhere are solved via the FE method • The FE solution precisely matches the CFD solution across all melting regimes • The FE model running on a GPU card with 4,608 CUDA cores is 28.2x more efficient than the CFD simulations running on 24 CPU cores in parallel The current work proposes a finite element method (FEM) to accelerate scanwise thermal process simulation of the laser powder bed fusion (L-PBF) process with computational fluid dynamics (CFD) resolution near the melt pool. Termed the CFD imposed FEM (CIFEM), the transient thermal fields from a high-fidelity CFD simulation and inferred by deep learning are imposed as temperature values rather than utilizing a conventional heat source model as in existing FEM-based process simulations. These fields are enforced only within a relatively small computational region encompassing the melt pool, while heat diffusion effects elsewhere are solved via the FEM. For a wide range of laser power and scan speeds covering the conduction, transition, and keyhole melting regimes, 29 of the 30 total CIFEM-simulated melt pool sizes lie within two standard deviations of the experimental melt pool sizes. Compared with the CFD simulations, the thermal fields obtained by CIFEM possess 7.44% mean absolute relative error (MARE), significantly less than the 43.76% MARE on three representative test cases simulated using the Goldak heat source model calibrated to the measured melt pool dimensions. In terms of computational efficiency, the CIFEM model running on a GPU card with 4,608 Compute Unified Device Architecture (CUDA) cores is 28.2x more efficient than the CFD simulations running on 24 CPU cores in parallel.

High-fidelity simulation of ship and linear shear current interaction

In this paper, the hydrodynamic performance of KCS (KRISO Container Ship) in the presence of a non-uniform shear current is investigated for the first time using high-fidelity CFD (Computational Fluid Dynamics) simulations. Various shear current strength conditions Frs with different directions were considered and results were compared to the ones with no shear current. The studies were conducted for the ship advancing at Fr = 0.26, 0.367. The focus is laid on the shear current effects on the ship resistance as well as the free surface elevation, pressure distribution, vortex structures, boundary layer, and nominal wake. Results showed that the presence of the shear current affects the hydrodynamic performance of the ship profoundly such that it increases/decreases the ship resistance in heading/following shear current with the largest change for the cases with the strongest shear current. The viscous resistance is dominant at Fr = 0.26 for all shear current conditions while the wave resistance is dominant for most cases at Fr = 0.367. The wave resistance reduces as the shear current direction varies from head to following shear current, similar to the theoretical model predictions. The viscous component follows a similar trend. The lateral resistance generated by the cross flow in beam shear current increases with the increase of Frs and the reduce of Fr and it is strongly dominated by the lateral pressure resistance. The Kelvin angle reduces in the head shear current and increases in the following shear current while the beam shear current reduces the Kelvin angle on the windward side of the ship and increases on the leeward side. The shear current also altered the transverse wavelength such that it increased in head shear current and reduced in following shear current. The predicted Kelvin angles and transverse wavelengths confirmed the finding of simplified theories on ship-shear current interaction. The presence of shear current also significantly affects the boundary layer, vortical structures, and the wake field.

Influence of morphological parameters on hemodynamics in internal carotid artery bifurcation aneurysms

Recent advancements in medical imaging techniques have enabled the accurate identification of unruptured intracranial aneurysms. To facilitate a proper patient management strategy, it is important to develop suitable mathematical models for their rupture prediction. To this end, the development of high-fidelity computational fluid dynamics (CFD) simulations with patient-specific boundary conditions will be useful in providing reliable hemodynamic parameters. In recent review articles, researchers have pointed out that, among several clinical and image-based indicators, morphological parameters, such as aspect ratio (AR) and size ratio (SR) of the aneurysm, correlated consistently with the rupture mechanism. However, it is not clear how these morphological indicators influence the hemodynamics-based CFD predictions. In the present work, the effect of these top-ranked morphological parameters on aneurysm hemodynamics and rupture prediction is investigated. Three patient-specific models have been used for analysis with the patient-specific inlet boundary conditions. We found that with an increase in AR and SR, the maximum value of wall shear stress (WSS) near the aneurysm neck is increased. Oscillatory shear index and relative residence time values are also increased with an increase in AR and SR. Furthermore, it was observed that an aneurysm with a multilobed structure shows complex flow, low WSS, and higher residence time over the secondary lobe. The turbulent kinetic energy and vorticity near the aneurysm neck are also increased with an increase in AR and SR.

Investigating Errors Observed during UAV-Based Vertical Measurements Using Computational Fluid Dynamics

Unmanned Aerial Vehicles (UAVs) are a popular platform for air quality measurements. For vertical measurements, rotary-wing UAVs are particularly well-suited. However, an important concern with rotary-wing UAVs is how the rotor-downwash affects measurement accuracy. Measurements from a recent field campaign showed notable discrepancies between data from ascent and descent, which suggested the UAV downwash may be the cause. To investigate and explain these observed discrepancies, we use high-fidelity computational fluid dynamics (CFD) simulations to simulate a UAV during vertical flight. We use a tracer to model a gaseous pollutant and evaluate the impact of the rotor-downwash on the concentration around the UAV. Our results indicate that, when measuring in a gradient, UAV-based measurements were ∼50% greater than the expected concentration during descent, but they were accurate during ascent, regardless of the location of the sensor. These results provide an explanation for errors encountered during vertical measurements and provide insight for accurate data collection methods in future studies.

Response Surface Optimisation of Vertical Axis Wind Turbine at low wind speeds

The Vertical Axis Wind Turbines (VAWTs) have an increasing global market and this emphasis the need for to improve the performance of VAWTs, especially at relatively low wind speed. This paper utilises the Response Surface methodology to optimise the performance of a VAWT. A three bladed VAWT configuration was considered with a NACA0015 profile. Three significant input parameters were selected including the tip speed ratio, the turbine solidity, and the pitch angle. An extended range of each input parameter was chosen in order to gain a good insight into how these input parameters affect the performance of the VAWT. The high-fidelity Computational Fluid Dynamics (CFD) simulations were carried out for the modelling of the turbine. The use of the Response Surface Optimisation based on Multi-Objective Genetic Algorithm (MOGA) along with the CFD simulations is found to be useful in the selection of the optimal design of VAWT. Moreover, the 3D aspects of the VAWT geometry are investigated and these include the turbine aspect ratio and the effect of the blade tip geometry. The implementation of an optimised winglet at the tip of the turbine blades is found to provide a significant enhancement of the cycle averaged power coefficient, especially at low aspect ratios.

Wake interaction in offshore wind farms with mesoscale derived inflow condition and sea waves

Numerical simulation is an indispensable tool for the design and optimization of wind farms layout and control strategies for energy loss reduction. Achieving consistent simulation results is strongly related to the definition of reliable weather and sea conditions, as well as the use of accurate computational fluid dynamics (CFD) models for the simulation of the wind turbines and wakes. Thus, we present a case study aiming to evaluate the wake-rotor interaction between offshore multi-MW wind turbines modelled using the Actuator Line Model (ALM) and realistic wind inflow conditions. In particular, the interaction between two DTU10 wind turbines is studied for two orientations of the upstream turbine rotor, simulating the use of a yaw-based wake control strategy. Realistic wind inflow conditions are obtained using a multi-scale approach, where the wind field is firstly computed using mesoscale numerical weather prediction (NWP). Then, the mesoscale vertical wind profile is used to define the wind velocity and turbulence boundary conditions for the high-fidelity CFD simulations. Sea waves motion is also imposed using a dynamic mesh approach to investigate the interaction between sea waves, surface boundary layer, and wind turbine wakes and loads.

Wing Kinematics and Unsteady Aerodynamics of a Hummingbird Pure Yawing Maneuver.

As one of few animals with the capability to execute agile yawing maneuvers, it is quite desirable to take inspiration from hummingbird flight aerodynamics. To understand the wing and body kinematics and associated aerodynamics of a hummingbird performing a free yawing maneuver, a crucial step in mimicking the biological motion in robotic systems, we paired accurate digital reconstruction techniques with high-fidelity computational fluid dynamics (CFD) simulations. Results of the body and wing kinematics reveal that to achieve the pure yaw maneuver, the hummingbird utilizes very little body pitching, rolling, vertical, or horizontal motion. Wing angle of incidence, stroke, and twist angles are found to be higher for the inner wing (IW) than the outer wing (OW). Unsteady aerodynamic calculations reveal that drag-based asymmetric force generation during the downstroke (DS) and upstroke (US) serves to control the speed of the turn, a characteristic that allows for great maneuvering precision. A dual-loop vortex formation during each half-stroke is found to contribute to asymmetric drag production. Wake analysis revealed that asymmetric wing kinematics led to leading-edge vortex strength differences of around 59% between the IW and OW. Finally, analysis of the role of wing flexibility revealed that flexibility is essential for generating the large torque necessary for completing the turn as well as producing sufficient lift for weight support.

Optimization of Turbine Blade Aerodynamic Designs Using CFD and Neural Network Models

Optimization methods have been widely applied to the aerodynamic design of gas turbine blades. While applying optimization to high-fidelity computational fluid dynamics (CFD) simulations has proven capable of improving engineering design performance, a challenge has been overcoming the prolonged run-time due to the computationally expensive CFD runs. Reduced-order models and, more recently, machine learning methods have been increasingly used in gas turbine studies to predict performance metrics and operational characteristics, model turbulence, and optimize designs. The application of machine learning methods allows for utilizing existing knowledge and datasets from different sources, such as previous experiments, CFD, low-fidelity simulations, 1D or system-level studies. The present study investigates inserting a machine learning model that utilizes such data into a high-fidelity CFD driven optimization process, and hence effectively reduces the number of required evaluations of the CFD model. Artificial Neural Network (ANN) models were trained on data from over three thousand two-dimensional (2D) CFD analyses of turbine blade cross-sections. The trained ANN models were then used as surrogates in a nested optimization process alongside a full three-dimensional Navier–Stokes CFD simulation. The much lower evaluation cost of the ANN model allows for tens of thousands of design evaluations to guide the search of the best blade profiles to be used in the more expensive, high-fidelity CFD runs, improving the progress of the optimization while reducing the required computation time. It is estimated that the current workflow achieves a five-fold reduction in computational time in comparison to an optimization process that is based on three-dimensional (3D) CFD simulations alone. The methodology is demonstrated on the NASA/General Electric Energy Efficient Engine (E3) high pressure turbine blade and found Pareto front designs with improved blade efficiency and power over the baseline. Quantitative analysis of the optimization data reveals that some design parameters in the present study are more influential than others, such as the lean angle and tip scaling factor. Examining the optimized designs also provides insight into the physics, showing that the optimized designs have a lower amount of pressure drop near the trailing edge, but have an earlier onset of pressure drop on the suction side surface when compared to the baseline design, contributing to the observed improvements in efficiency and power.

Numerical and experimental investigation of thermoelectric materials in direct contact membrane distillation

The high energy penalty associated with the direct contact membrane distillation (DCMD) technology is a major challenge precluding its commercial deployment. Although DCMD requires less aggressive operating conditions than reverse osmosis (∼1 bar vs 55 bar), its throughput is way lower (∼10 L/m2-h vs 70 L/m2-h) and its energy consumption is more (∼1700 kWh/m3 vs 7 kWh/m3). One approach to improve these metrics is the use of direct thermal energy sources inside the membrane module. A Peltier device installed inside the DCMD module to provide the needed thermal energy can reduce heat exchanger losses to near zero as the Peltier device is sandwiched between the hot feed and cold permeate in a thermoelectric-DCMD system. Previous experimental results have indicated gains in performance metrics using this approach. Given these results, robust high-fidelity computational fluid dynamics (CFD) simulations validated with experiments are developed to predict the performance of the thermoelectric-DCMD module whilst conducting a parametric analysis of the operational conditions and geometrical configurations. The Peltier device’s heat transfer rates are predicted from a simple analysis using the manufacturer’s performance datasheet and used as boundary conditions for the CFD simulations.

Geometrical deep learning for performance prediction of high-speed craft

Computational Fluid Dynamics (CFD) has become an indispensable tool in the field of engineering design evaluation and optimisation in the maritime industry. Accurate modelling of flow around a ship’s hull and superstructure requires high fidelity CFD simulations due to complex nonlinear phenomena. On the other hand, high fidelity CFD simulations are computationally expensive, memory demanding and time-consuming, thus limiting design space exploration when used in design optimisation. In order to overcome these challenges, we propose using geometrical deep learning-based surrogate modelling trained with CFD simulations for maritime applications combined with automatic hyperparameters tuning. Our proposed framework can predict both flow fields (e.g pressure field) on the surface of the geometry and any overall scalar parameters (e.g drag force) given a three-dimensional shape input. Furthermore, proposed deep learning model is extended to provide uncertainty quantification over its predictions (assuming CFD as ground truth). Finally, our proposed deep learning-based surrogate model is applied to predict pressure distribution and drag force for wind drag and calm water-resistance of high-speed craft. Despite the common drawback of deep learning-based algorithms to require large datasets for training, our proposed solution can maintain high prediction accuracy even when trained with a very small dataset which makes it highly suitable for use during design optimisations.