Common Research Model Research Articles

Supervised kernel principal component analysis-polynomial chaos-Kriging for high-dimensional surrogate modelling and optimization

Surrogate-based optimization (SBO) approach is becoming more and more popular in the expensive aerodynamic design of aircraft. However, with increasing number of design variables required for parameterizing a complex shape, SBO is suffering from the serious difficulty of the curse of dimensionality. To ameliorate this issue, a supervised nonlinear dimensionality-reduction surrogate modelling method was proposed. Such a method combines the supervised kernel principal component analysis (SKPCA) and polynomial chaos-Kriging (PCK) techniques into the jointly surrogate modelling process and adaptively establishes the accurate mapping from the high-dimensional inputs to the output of the system. This SKPCA-PCK method, which fully considers the effect of inputs on outputs and adaptively trains these hyper-parameters in the surrogate modelling process, escapes from the low prediction accuracy and instability of the surrogate model in conjunction with current linear or unsupervised dimensionality-reduction methods. Further, an efficient SKPCA-PCK-based global optimization method for high-dimensional aerodynamic design was developed. The performance of the proposed method is examined by investigating two numerical examples, the transonic RAE2822 airfoil and the wing of the NASA Common Research Model. Results demonstrate that the proposed SKPCA-PCK method significantly improves the modelling efficiency and accuracy compared to the unsupervised linear PCA-Kriging method. More importantly, the proposed SKPCA-PCK-based optimization method provides better performance and an appreciably higher optimization efficiency for expensive single-point and robust aerodynamic design involving high-dimensional design variables compared to the Kriging-based optimization method. These results provide further evidence that the proposed method provides a promising approach for mitigating the curse of dimensionality in SBO.

Sensor placement for optimal aerodynamic data fusion

Aircraft design is recently evolving towards a digital twin representation that involves many heterogeneous data sources. The aerodynamic development of aircraft usually incorporates data from computational fluid dynamics simulations, wind tunnel testing, and flight tests. All of these data sources have their advantages and disadvantages, which can optimally be combined using data fusion methods. However, the quality of the data fusion result strongly depends on the experimental design, i.e. the placement of discrete sensors. Therefore, an optimized sensor placement is essential for data fusion applications, as the number of sensors is limited. This work presents a sensor placement strategy for the widely used Gappy proper orthogonal decomposition data fusion methodology. The sensor placement relies on a Bayesian formulation of the data fusion, allowing accurate error estimates. Based on the Bayesian posterior, a utility function characterizes the quality of the fused result by quantifying the expected information gain for the proper orthogonal decomposition coefficients. As the optimization of the sensor locations involves a complex combinatorial problem, we introduce an efficient genetic algorithm for this task. The method is demonstrated on a two-dimensional airfoil and the NASA Common Research Model with synthetic measurement errors. For both test cases, an optimal sensor placement results in smaller reconstruction errors than a conventional layout. The Bayesian approach leads, in most cases, to more accurate reconstructions and is more versatile than other well-established sensor placement methods. The proposed genetic algorithm finds better optima with significantly fewer function evaluations than the widely used greedy algorithms.

Wall-Modeled Large-Eddy Simulation Method for Unstructured-Grid Navier–Stokes Solvers

This paper reports on the implementation and assessment of a wall-modeled large-eddy simulation (WMLES) methodology in an unstructured-grid, node-centered flow solver, FUN3D, that is developed and supported at the NASA Langley Research Center. Finite-volume (FV) and finite-element (FE) discretization schemes considered in the study provide formal second-order spatial accuracy. Large-eddy simulations (LES) resolve large-scale turbulent-flow features, and small-scale effects are modeled using the Vreman subgrid-scale model. At solid-wall boundaries, a shear-stress model is employed to provide a proper boundary-flux closure. The nonlinear equations are integrated in time using either an optimized backward difference formula or an implicit multistage Runge–Kutta temporal scheme. The implicit equations at each time step are solved by strong nonlinear iteration schemes. WMLES demonstrations are shown for two high-lift configurations, namely, the McDonnell Douglas 30P30N multi-element airfoil and a NASA High-Lift Common Research Model. Results show that the WMLES approaches implemented in the FV and FE discretization methods produce consistent solutions and are capable of capturing key aerodynamic characteristics and flow structures for high-lift configurations at a wide range of angles of attack, including maximum-lift conditions. In the 30P30N example, correct trends in the variations of integrated aerodynamic forces and moments, surface pressure distributions, and boundary-layer profiles are captured as the Reynolds number is increased.

Parametric Optimization Study of Novel Winglets for Transonic Aircraft Wings

This paper deals with the topic of reducing drag force acting on aircraft wings by incorporating novel winglet designs, such as multi-tip, bird-type, and twisted. The high-speed NASA common research model (CRM) was selected as the baseline model, and winglet designs were retrofitted while keeping the projected wingspan constant. Computational analysis was performed using RANS coupled with the Spalart–Allmaras turbulence model to determine aerodynamic coefficients, such as CL and CD. It was observed that the multi-tip and bird-type designs performed exceptionally well at a low angle of attack (0°). A parametric study was conducted on multi-tip winglets by tweaking the parameters such as sweep angle (Λ), tip twist (Є), taper ratio (λ), and cant angle (Φ). The best combination of parameters for optimal aerodynamic performance while maintaining the wing root bending moment was determined using both the Taguchi method and Taguchi-based grey relational analysis (T-GRA) coupled with principal component analysis (PCA). Also, the percentage contribution of each parameter was determined by using the analysis of variance (ANOVA) method. At the design point, the optimized winglet design outperformed the baseline design by 18.29% in the Taguchi method and by 20.77% in the T-GRA coupled with the PCA method based on aerodynamic efficiency and wing root bending moment.

Combined Reynolds-averaged Navier-Stokes/Large-Eddy Simulations for an aircraft wake until dissipation regime

A new methodology has been devised to simulate the aerodynamic wake of an airliner during cruise flight using Reynolds-averaged Navier-Stokes (RANS) and Large Eddy Simulations (LES). The wake evolution is simulated considering the full geometry of the aircraft and spans from the jet regime to the dissipation regime. First, the jet regime is computed using RANS modeling and state-of-the-art anisotropic mesh adaptation techniques. Second, the vortex and dissipation regimes are solved using temporal LES with flow field initialization from the previous RANS calculation. RANS information on turbulent quantities is transferred to the LES domain using synthetic turbulence methods. The methodology is first tested on a NACA-0012 wing. It is then applied to the NASA Common Research Model (CRM) geometry on cruise flight conditions. Jet/vortex interaction is studied during the jet regime up to twenty wingspans behind the aircraft. The influences of atmospheric turbulence, jet turbulence, and stable atmospheric stratification are investigated for the vortex and dissipation regimes. It is observed that atmospheric turbulence intensity and stratification both accelerate the vortex decay. Moreover, the secondary wake structure is found to be much more turbulent for numerical RANS initialization than for classical analytic initialization, showcasing the importance of early aerodynamics on wake evolution.

Surface roughness effects on transonic aircraft performance: Experimental/numerical comparisons

This work highlights the necessity of taking into account surface roughness when conducting experimental tests, and when using numerical simulations to precisely calculate the turbulent lift and drag of wind-tunnel models or real aircraft in transonic conditions. The present article is a continuation of “Turbulent drag induced by low surface roughness at transonic speeds: Experimental/numerical comparisons,” Physics of Fluids, Vol. 32, 045108 (2020) by Hue and Molton. The outcomes of this former study, which was focused on flat plate samples, are here applied to a three-dimensional aircraft configuration: the Common Research Model used as a reference in the recent international Drag Prediction Workshops. Experimental campaigns have been performed in the largest ONERA wind tunnels S1MA and S2MA involving models with average surface roughness heights Ra close to 0.5 micrometers, wingspans up to 3.5 meters, Mach and Reynolds numbers up to 0.95 and 5 million respectively. Reynolds-averaged Navier–Stokes computations based on the wind-tunnel tests have then been carried out, using the equivalent sand-grain roughness height approach as well as a Musker-type correlation to determine relevant ks values. The results of both the experimental and numerical campaigns have demonstrated that the aerodynamic coefficients of the aircraft can be significantly affected by the surface roughness, even with roughness Reynolds numbers ks+ potentially below the usual threshold values sometimes considered in engineering applications (i.e. in the order of 3.5 to 5). In particular, the surface roughness effects on lift and drag have been studied using far-field analyses to evaluate the responses of friction, viscous pressure, wave and lift-induced drag components. Finally, the numerical studies have been extended to the full-scale geometry in flight conditions in order to assess the roughness effects and potential gains in realistic aircraft operating conditions.

High-fidelity fluid-structure interaction applied to static aeroelasticity in transonic flows

Transonic flows at high Reynolds numbers can lead to high dynamic pressures and, consequently, to aerostructural deflections of aircraft structures. This study aims to develop and validate a high-fidelity static aeroelastic analysis environment that is efficient and that can be used in an industrial setting. The aerodynamics is represented by numerical solutions of the Reynolds-averaged Navier-Stokes equations with appropriate turbulence closures. The load transfer process uses finite element shape functions in order to distribute the aerodynamic loads into the structural discretization. The structural analysis employs a modal basis approach, and a wingtip deflection convergence study is performed to find an adequate modal basis size. Radial basis functions are used for the fluid mesh displacement, and the influence of the support radius is evaluated to determine the optimal values relative to the wing mean aerodynamic chord. The capability is tested using the static aeroelastic benchmarks of the High Reynolds Aerostructural Dynamics Project (HIRENASD) and NASA's Common Research Model (CRM). The static aeroelastic results demonstrate robustness and consistency for the aerodynamic coefficients, pressure distributions, and structural deflection predictions at different normalized dynamic pressure values and grid refinement levels.

On the Generalization Capability of a Data-Driven Turbulence Model by Field Inversion and Machine Learning

This paper discusses the generalizability of a data-augmented turbulence model with a focus on the field inversion and machine learning approach. It is highlighted that the augmented model based on two-dimensional (2D) separated airfoil flows gives poor predictive capability for a different class of separated flows (NASA wall-mounted hump) compared to the baseline model due to extrapolation. We demonstrate a sensor-based approach to localize the data-driven model correction to tackle this generalizability issue. Furthermore, the applicability of the augmented model to a more complex aeronautical three-dimensional case, the NASA Common Research Model configuration, is studied. Observations on the pressure coefficient predictions and the model correction field suggest that the present 2D-based augmentation is to some extent applicable to a three-dimensional aircraft flow.

Utilizing global-local neural networks for the analysis of non-linear aerodynamics

In addressing the computational challenges pervasive in engineering where time and cost limitations are key concerns, particularly within the Computational Fluid Dynamics (CFD) domain, Reduced Order Models (ROMs) have emerged as instrumental tools. Focused on reducing computational complexity without intrusively modifying the computational model, this study centres on the strategic application of aerodynamic ROMs, which provide efficient computation of distributed quantities and aerodynamic forces. This work presents ROMs for non-linear aerodynamic applications, integrating principal component analysis (PCA) with Global Local Neural Networks (GLNN). The effectiveness of the proposed methodology is demonstrated by leveraging dependency on the parameter space created with non-linear high-fidelity CFD data, incorporating viscous simulation for a comprehensive approach. Results are first presented for a two-dimensional airfoil case and then for a three-dimensional test case featuring a transonic wing-body-tail transport aircraft configuration (NASA Common Research Model). In transonic flows, the proposed ROMs demonstrate the ability to accurately capture both the location and strength of shocks, as well as forces and moments for unseen prediction points. This highlights the efficiency of the proposed method in navigating complex aerodynamic scenarios, achieving comparable accuracy to full-order modelling but at orders of magnitude less computational time, for unseen parameters outside the ROM training set within the parameter space.

Towards Wall-Resolved Large-Eddy Simulation of the High-Lift Common Research Model

In the present study, we performed a sixth-order wall-resolved large-eddy simulation (WRLES) of the high-lift Common Research Model (CRM-HL) on the Leadership Class Computing Cluster Summit. During the 4th High-Lift Prediction Workshop (HLPW-4), wall-modeled LES (WMLES) and hybrid Reynolds-averaged Navier–Stokes (RANS)/LES approaches showed promise in predicting the maximum lift and large flow separations near stall. In those simulations, however, laminar and transition regions were not resolved since the flow was assumed to be fully turbulent. If one wants to resolve these regions, the only viable approach is WRLES. The main purpose of the present study is to demonstrate the feasibility of WRLES for a complex real-world configuration at the wind tunnel Reynolds number. A high-order unstructured mesh LES solver called hpMusic was employed in the study. The simulation at solution polynomial order P=5 has more than 14 billion degrees of freedom per equation. Computational results are compared to experimental data in the form of surface oil flows and pressure coefficient profiles. We highlight lessons learned from attempting a grand-challenge type large-scale simulation: 1) such a simulation is feasible but very expensive, estimated to be at least three orders more expensive than WMLES, 2) oil flows produced with WRLES agree very well with experimental oil flows in fine detail, not observed in those produced with WMLES, and 3) flow visualization is a severe bottleneck, and parallel postprocessing capability is needed.

High-Order Wall-Modeled Large-Eddy Simulation of High-Lift Configuration

This paper presents the assessment of several recent enhancements for a high-order wall-modeled large-eddy simulation (WMLES) approach and demonstrates order independence with a fixed data exchange location in the wall model. The two enhancements include the use of isotropic tetrahedral elements to improve accuracy and an explicit subgrid-scale model, the Vreman model, to improve accuracy and robustness. The P-refinement study focused on the high-lift Common Research Model (HL-CRM) at the angle of attack of 19.57 deg, a benchmark problem from the 4th AIAA High-Lift Prediction Workshop. Solution polynomial orders of P=2, 3, 4, and 5 were used in the study. The study demonstrated p independence in integrated forces, pitch moment, velocity profile in the wall-normal direction, and surface flow topology. It also showed that a P order of at least 3 (P3) was needed to correctly predict the external inviscid flow and the surface flow topology. Thereafter, P3 simulations over several other angles of attack demonstrated that the high-order WMLES approach can correctly predict the maximum lift and flow separation regions for HL-CRM with about 40 million degrees of freedom (DOF) compared to at least 250 million DOF required by second-order methods.

Recycling Krylov subspaces for efficient partitioned solution of aerostructural adjoint systems

Robust and efficient solvers for coupled-adjoint linear systems are crucial to successful aerostructural optimization. Monolithic and partitioned strategies can be applied. The monolithic approach is expected to offer better robustness and efficiency for strong fluid-structure interactions. However, it requires a higher implementation cost and convergence may depend on appropriate scaling and initialization strategies. On the other hand, the modularity of the partitioned method enables a straightforward implementation while its convergence may require relaxation. In addition, a partitioned solver often leads to a higher number of iterations to get the same level of convergence as the monolithic one.The objective of this paper is to accelerate the fluid-structure coupled-adjoint partitioned solver by considering techniques borrowed from approximate invariant subspace recycling strategies adapted to sequences of linear systems with varying right-hand sides. Indeed, in a partitioned framework, the structural source term attached to the fluid block of equations affects the right-hand side with the nice property of quickly converging to a constant value. For the fluid block of equations, we also consider deflation of approximate eigenvectors in conjunction with advanced inner-outer Krylov solvers.To demonstrate the effectiveness of our proposed recycling strategy, we compute the coupled derivatives for two emblematic problems in transonic viscous flow. The first one is an aeroelastic configuration of the ONERA-M6 fixed wing. For this exercise, the fluid grid was coupled to a structural model specifically designed to exhibit a high flexibility. The second one is a wing-body aeroelastic configuration of the Common Research Model. All computations are performed using a fully linearized one-equation Spalart-Allmaras turbulence model. Numerical simulations show up to 39% reduction in matrix-vector products for GCRO-DR and up to 25% for the nested FGCRO-DR solver.

Characterization of Humanized Mouse Model of Organophosphate Poisoning and Detection of Countermeasures via MALDI-MSI.

Organophosphoate (OP) chemicals are known to inhibit the enzyme acetylcholinesterase (AChE). Studying OP poisoning is difficult because common small animal research models have serum carboxylesterase, which contributes to animals' resistance to OP poisoning. Historically, guinea pigs have been used for this research; however, a novel genetically modified mouse strain (KIKO) was developed with nonfunctional serum carboxylase (Es1 KO) and an altered acetylcholinesterase (AChE) gene, which expresses the amino acid sequence of the human form of the same protein (AChE KI). KIKO mice were injected with 1xLD50 of an OP nerve agent or vehicle control with or without atropine. After one to three minutes, animals were injected with 35 mg/kg of the currently fielded Reactivator countermeasure for OP poisoning. Postmortem brains were imaged on a Bruker RapifleX ToF/ToF instrument. Data confirmed the presence of increased acetylcholine in OP-exposed animals, regardless of treatment or atropine status. More interestingly, we detected a small amount of Reactivator within the brain of both exposed and unexposed animals; it is currently debated if reactivators can cross the blood-brain barrier. Further, we were able to simultaneously image acetylcholine, the primary affected neurotransmitter, as well as determine the location of both Reactivator and acetylcholine in the brain. This study, which utilized sensitive MALDI-MSI methods, characterized KIKO mice as a functional model for OP countermeasure development.

Aerostructural Wing Optimization Using a Structural Surrogate in a Coupled Adjoint Formulation

As the number of disciplines included in the multidisciplinary design-optimization process continues to increase, it is envisioned that some of the disciplinary tools will take the form of surrogate models, whereas others remain physics-based, depending on the requirements and stage of the design process. To simulate this in the context of an aerostructural optimization of an aircraft wing, the work presented herein features a high-fidelity aerodynamic flow solver, while a surrogate is employed to model the wing structure. This approach includes the evaluation of the sensitivities of both the aerodynamic and structural disciplines, using a coupled-adjoint formulation to enable gradient-based optimization. An important aspect of the method is that the surrogate is trained only once, prior to the optimization, and held fixed throughout. The surrogate in effect parameterizes the structural design process, and outputs the weight and stiffness of an optimized structure, given inputs of geometry parameters and sizing loads. To minimize the number of surrogate inputs and enable the representation of the entire structural design space, parameterized loads are used to build the surrogate. The method is applied to the optimization of the NASA Common Research Model, illustrating the effectiveness of the new approach.

A mid-range approximation method assisted by trust region strategy for aerodynamic shape optimization

This paper presents an efficient solution for high-fidelity large-scale aerodynamic shape optimization problems based on several developments in the mid-range approximation method within a trust region optimization framework. The mid-range approximation method is an iterative optimization technique that utilizes mid-range approximations to replace the physical experiments or simulations during the optimization based on the selected trust region strategy. It could transform the original optimization problem into a sequence of approximate sub-optimization problems. In this work, an improved trust region strategy is proposed to contain more optimization states with a flexible and controllable performance to suit different types of problems. A metamodel assembly technique and its gradient-enhanced version are developed to further relax the requirements of computational costs in the mid-range approximation method. Its performance is discussed through a detailed comparison of metamodel performance using a mathematical benchmark case named Vanderplaats scalable beam. The wing only of the Common Research Model is offered to the proposed method for aerodynamic shape optimization. The optimization has 1 design objective, 232 design variables, and 135 design constraints. With all constraints satisfied, the optimized configuration has a 4.85 % improvement in wing drag performance. The shock region is greatly reduced and the wing pressure distribution is smooth and nearly parallel. These results show that the proposed method could achieve the design goal successfully within a reasonable computational cost for large-scale problems.

Aerodynamic Prediction of the High-Lift Common Research Model with Modified Turbulence Model

Aerodynamic simulations were carried out in the study presented in this paper focusing on the stall performance of the High-Lift Common Research Model obtained from the fourth AIAA High-Lift Prediction Workshop. Various turbulence models of Reynolds-averaged Navier–Stokes simulations are analyzed. A modified version of the transitional k−v2¯−ω model was developed to enhance stall prediction accuracy for high-lift configurations with a nacelle chine. The vortex generator, three-element airfoil, and high-lift model are numerically simulated. The results reveal that implementing a k−v2¯−ω model with the separation shear layer fixed notably enhances the stall prediction behavior for both the three-element airfoil and high-lift configuration without affecting the prediction of the vortex strength of a vortex generator. Moreover, incorporating rotation correction into the SPF k−v2¯−ω model improves the prediction of vortex strength and further enhances stall prediction for the high-lift configuration. The relative error in predicting the maximum lift coefficient is less than 5% of the experimental data. The study also investigated the impact of the nacelle chine on the stall behavior of the high-lift configuration. The results demonstrate that the chine vortex can mitigate the adverse effects of the nacelle/pylon vortex system and increase the maximum lift coefficient.

Influence of structural elasticity on the aerodynamic characteristics of the common research model with deflected control surfaces

Numerical aerodynamic and aeroelastic investigations are conducted for the Common Research Model (CRM) configuration with deployable aileron control surfaces and a trimmable horizontal stabilizer at cruise and off-cruise conditions. The numerical methodology which is implemented in the multi-disciplinary simulation environment SimServer is based on CFD and coupled CFD-CSM simulations. The DLR Tau Code is applied in order to solve the Reynolds-Averaged-Navier-Stokes (RANS) equations. The Spalart-Allmaras turbulence model with Edwards modification is used for the closure modeling of the three-dimensional RANS equations. A linear modal solver is utilized to calculate structural displacements and rotations of control surfaces are modeled by a Chimera approach. A grid independence study is conducted by using the rigid aerodynamic model of the CRM. The main focus of the analysis lies on the evaluation of the stationary CFD-CSM simulations addressing the influence of structural deformation on aileron effectiveness. Influences of various flow conditions and different fuel tank filling levels on the resulting local and global aerodynamic properties of the CRM are investigated. The reduction of aileron effectiveness as well as the difference in horizontal trim deflection angle if structural elastic effects are considered is investigated for the considered Mach number range and aircraft mass configurations.

Wall-Modeled Large-Eddy Simulation of Transonic Buffet over NASA-CRM Using FFVHC-ACE

This study investigates the capability of wall-modeled large-eddy simulation (WMLES) in predicting the transonic buffet phenomenon over three-dimensional aircraft configurations at high Reynolds numbers. The WMLES is conducted using the Cartesian-grid-based flow solver FrontFlow/Violet Hierarchical Cartesian for Aeronautics Based on Compressible-Flow Equations (FFVHC-ACE). To extend the capability of FFVHC-ACE to transonic flow simulations with shock waves, a hybrid kinetic energy and entropy preserving/upwind scheme is implemented in FFVHC-ACE. To validate the employed simulation framework, the flow around the NASA Common Research Model in the near-cruise condition is simulated. In this case, the surface pressure distributions and force coefficients are reasonably predicted and show the trend of grid convergence. Furthermore, the WMLES at the buffet conditions reproduces the wavy structures of the shock wave (i.e., buffet cells) propagating outboard. The obtained propagation speed of the buffet cells and surface pressure spectra show reasonable agreement with the wind tunnel experiments.

Evaluating large language models for selection of statistical test for research: A pilot study

Abstract Background: In contemporary research, selecting the appropriate statistical test is a critical and often challenging step. The emergence of large language models (LLMs) has offered a promising avenue for automating this process, potentially enhancing the efficiency and accuracy of statistical test selection. Aim: This study aimed to assess the capability of freely available LLMs – OpenAI’s ChatGPT3.5, Google Bard, Microsoft Bing Chat, and Perplexity in recommending suitable statistical tests for research, comparing their recommendations with those made by human experts. Materials and Methods: A total of 27 case vignettes were prepared for common research models with a question asking suitable statistical tests. The cases were formulated from previously published literature and reviewed by a human expert for their accuracy of information. The LLMs were asked the question with the case vignettes and the process was repeated with paraphrased cases. The concordance (if exactly matching the answer key) and acceptance (when not exactly matching with answer key, but can be considered suitable) were evaluated between LLM’s recommendations and those of human experts. Results: Among the 27 case vignettes, ChatGPT3.5-suggested statistical test had 85.19% concordance and 100% acceptance; Bard experiment had 77.78% concordance and 96.3% acceptance; Microsoft Bing Chat had 96.3% concordance and 100% acceptance; and Perplexity had 85.19% concordance and 100% acceptance. The intra-class correction coefficient of average measure among the responses of LLMs was 0.728 (95% confidence interval [CI]: 0.51–0.86), P < 0.0001. The test–retest reliability of ChatGPT was r = 0.71 (95% CI: 0.44–0.86), P < 0.0001, Bard was r = −0.22 (95% CI: −0.56–0.18), P = 0.26, Bing was r = −0.06 (95% CI: −0.44–0.33), P = 0.73, and Perplexity was r = 0.52 (95% CI: 0.16–0.75), P = 0.0059. Conclusion: The LLMs, namely, ChatGPT, Google Bard, Microsoft Bing, and Perplexity all showed >75% concordance in suggesting statistical tests for research case vignettes with all having acceptance of >95%. The LLMs had a moderate level of agreement among them. While not a complete replacement for human expertise, these models can serve as effective decision support systems, especially in scenarios where rapid test selection is essential.

Graph convolutional multi-mesh autoencoder for steady transonic aircraft aerodynamics

Calculating aerodynamic loads around an aircraft using computational fluid dynamics is a user’s and computer-intensive task. An attractive alternative is to leverage neural networks (NNs) bypassing the need of solving the governing fluid equations at all flight conditions of interest. NNs have the ability to infer highly nonlinear predictions if a reference dataset is available. This work presents a geometric deep learning based multi-mesh autoencoder framework for steady-state transonic aerodynamics. The framework builds on graph NNs which are designed for irregular and unstructured spatial discretisations, embedded in a multi-resolution algorithm for dimensionality reduction. The test case is for the NASA common research model wing/body aircraft configuration. Thorough studies are presented discussing the model predictions in terms of vector fields, pressure and shear-stress coefficients, and scalar fields, total force and moment coefficients, for a range of nonlinear conditions involving shock waves and flow separation. We note that the cost of the model prediction is minimal having used an existing database.