Jun 2026· Aerospace· Vol 13, pp. 590· 0 citations· 13 references
TL;DR
This research provides direct visual evidence bridging black-box AI decisions with classical fluid mechanics, proposing a “Mechanism-Guided Verification” framework that offers a trustworthy pathway for the future certification of AI in safety-critical aerospace systems.
Abstract
The Vortex Ring State (VRS) is a critical aerodynamic hazard for rotorcraft, characterized by highly unsteady fluid–structure interactions and severe low-frequency vibrations. While data-driven deep learning models have shown promise in aviation state monitoring, their inherent “black-box” nature fundamentally contradicts the stringent interpretability requirements of airworthiness certification. To address this, we propose an “AI for Science” paradigm, investigating whether advanced Vision Transformers (ViT) can autonomously discover underlying aerodynamic mechanisms without human physical priors. First, to ensure absolute data fidelity, flight test datasets of a coaxial unmanned aerial vehicle were rigorously labeled using cross-validation from high-fidelity Computational Fluid Dynamics (CFD) simulations and wind tunnel tests. One-dimensional vibration signals were then transformed into two-dimensional Continuous Wavelet Transform (CWT) spectrograms. By employing Target-Layer Gradient Adaptation (Grad-CAM) techniques, we conducted a systematic comparison between traditional Convolutional Neural Networks (ResNet50) and ViT. The results demonstrate that while CNNs suffer from diffuse attention caused by high-frequency noise, the frozen-backbone ViT model achieves a physically interpretable accuracy of 93.24%, while autonomously locking its global attention onto a perfectly horizontal feature band centered at 41.7 Hz. Crucially, this autonomously discovered feature precisely aligns with the theoretically derived once-per-revolution (1P) fundamental frequency of the rotor’s flap-lag coupling response under VRS aerodynamic turbulence. This research provides direct visual evidence bridging black-box AI decisions with classical fluid mechanics, proposing a “Mechanism-Guided Verification” framework that offers a trustworthy pathway for the future certification of AI in safety-critical aerospace systems.
Aerodynamic analysis, a core aspect of studying physical phenomena, relies on governing equations traditionally solved via computational fluid dynamics (CFD). While CFD enables high-fidelity simulations of complex flows, including the Navier–Stokes equations, it often requires simplifying assumptions and substantial computational resources, limiting real-time analysis. Physics-Informed Neural Networks (PINNs) offer a promising alternative by embedding physical laws into the training process, ensuring predictions remain consistent with aerodynamic principles. This study presents a systematic implementation and validation of PINNs for predicting lift (Cl) and drag (Cd) coefficients across NACA airfoil geometries under varying Reynolds numbers and angles of attack. The methodology leverages CFD-generated data for training and incorporates physics-based loss functions enforcing linear lift theory, symmetry, positive drag, and drag–lift relationships. The trained PINN is integrated into an interactive Digital Twin environment, enabling real-time aerodynamic analysis, design exploration, and optimization. Results demonstrate rapid convergence, with total loss decreasing four orders of magnitude and mean absolute errors of 0.049 for Cl and 0.0019 for Cd, corresponding to 2.4% and 12.5% relative errors, respectively. Lift predictions show excellent agreement with CFD (R² = 0.993), while drag predictions capture trends with moderate scatter (R² = 0.738). Case studies of symmetric, low-camber, and high-camber airfoils confirm accurate reproduction of expected aerodynamic behavior. This work highlights the feasibility of combining CFD, physics-informed AI, and Digital Twin technologies to accelerate and enhance aerodynamic workflows.
Maryam Mohammed Elbanna, Mahmoud Yousef, S. El-Bahloul et al.· Journal of Physics, Conferen...· 0 citations
Hydrodynamic models underpin Autonomous Underwater Vehicle (AUV) design, motion control, and performance evaluation. Existing methods face two critical bottlenecks: (1) conventional explicit CFD requires predefined trajectories, which fails to capture true motion responses under combined rudder-propeller action and creates a disconnect between simulation and real operations; (2) the widely adopted Standard Submarine Motion Equations (SSME) suffer from high parameter redundancy, while high-precision non-parametric models incur prohibitive computational costs, hindering embedded deployment. To address these gaps, this paper proposes an implicit CFD-driven framework for fully appended AUVs equipped with through-body thrusters. It requires no preset trajectories, directly coupling periodic propeller thrust and rudder angle excitations to achieve 5-degree-of-freedom (5DOF) spatial motion simulations aligned with real navigation states. Parametric and non-parametric models are identified via Least Squares (LS) and Neural Networks (NN), respectively. Sobol global sensitivity analysis reduces SSME dimensionality, yielding a Basic Submarine Motion Equation (BSME) with only 25 key parameters—cutting the parameter count by 55% with negligible accuracy loss. Validation shows the non-parametric NN model reduces prediction error by over 10% compared to its parametric counterpart, while the streamlined BSME enables real-time forecasting in low-power computing scenarios. This approach balances accuracy and efficiency for rapid hydrodynamic prediction during early AUV design and embedded controller deployment.
Yingjie Guan, Xiao-Yang Deng, Yougang Bian et al.· Journal of Marine Science an...· 0 citations
The Boundary Layer Transition (BOLT) series of flight experiments was initiated by the United States Air Force Office of Scientific Research in 2017 to study flow phenomena related to hypersonic boundary-layer transition and turbulent flow on low-curvature concave lifting surfaces with highly swept leading edges. This paper details the preflight aerodynamic modeling carried out in support of the BOLT-1B flight experiment, which was launched in September 2024. The modeling approach presented here includes multifidelity analysis meant to anchor a large database of low- to medium-fidelity predictions based on inviscid and correlation-based methods by leveraging a subset of high-fidelity, viscous computations. The efforts of this study were primarily focused on anchoring vehicle drag, static stability, roll-driving, and roll-damping predictions. The resulting aerodynamic database was then used to inform predictions of the scientific conditions for the flight experiment, including trajectory modeling and prediction of flight dynamics.
Cameron S. Butler, Marius Franze, G. McKiernan et al.· Journal of Spacecraft and Ro...· 0 citations
Unsteady aerodynamic phenomena, such as gusts, turbulence, and fluid-structure interactions affect an aircraft during flight. For design, optimisation and certification, it is indispensable to quantify such unsteady aerodynamic effects. Industry-standard computational fluid dynamics methods, such as solving the unsteady Reynolds-averaged Navier-Stokes equations or the linearized frequency domain method, are either computationally expensive or restricted by assumptions like linearity. Once trained, machine learning methods are capable of computing non-linear relationships very fast, making them suitable as surrogate models. By autoregressively applying graph neural networks (GNNs), operating on a discretised spatial domain, spatio-temporal predictions can be made. However, autoregressive GNNs suffer from error accumulation leading to unstable rollouts over time. Here we show that combining GNNs with augmented Neural Ordinary Differential Equations yields temporally stable predictions of the surface forces on a pitching airfoil. We found that our approach, called GNODE, based on Graph Neural Ordinary Differential Equations, provides temporally more stable, spatially smoother, and overall more accurate results than an autoregressive GNN baseline. Tests are conducted on a dataset consisting of a simulations of a pitching airfoil, including transonic shocks, transient behaviour and dynamic non-linearities. Augmenting GNODEs with additional latent dimensions improves the expressivity and accuracy by capturing underlying history effects. The developed method demonstrates an approach that is suitable to model non-linear spatio-temporal systems with exogenous inputs.
Henrik Lange, R. Thormann, P. Bekemeyer· 0 citations
This study presents an extended Blade Element Momentum Theory (BEMT) framework for predicting the hover performance of an EC135-class light-twin helicopter rotor while quantifying the impact of sectional aerodynamic uncertainty on global rotor metrics. Because the proprietary EC135 airfoils are not publicly available, a reproducible surrogate blade based on the ONERA OA213 and OA209 airfoils is adopted. The airfoil substitution is explicitly treated as an epistemic modelling assumption, and its effect on rotor-level hover predictions is assessed through a dedicated geometric comparison and bounded sensitivity analysis. The classical BEMT formulation is enhanced with Prandtl tip-loss corrections, Mach-dependent sectional aerodynamics, and an iterative non-uniform inflow model. Aerodynamic coefficients are obtained from Gaussian Process (GP) surrogate models trained on XFOIL-generated databases and calibrated using cross-validation techniques. The calibrated GP models are coupled with the rotor solver and their predictive uncertainty is propagated through Monte Carlo simulations. For the nominal hover trim condition, the rotor was trimmed to CT=0.00607, while the model predicted a power coefficient of 0.00041 and a figure of merit of 0.819. The propagated GP/XFOIL-conditioned uncertainty yields a 95% confidence interval of 0.8074–0.8285 for the figure of merit, indicating limited sensitivity of rotor performance to sectional aerodynamic uncertainty. The influence of compressibility and tip-loss effects is also quantified. In a separate Caradonna–Tung solver-verification case, the thrust-coefficient error is reduced from 32.57% to 7.78% when finite-aspect-ratio corrections are included. The proposed framework provides a fast, reproducible, and uncertainty-aware approach for helicopter rotor hover analysis suitable for preliminary design and performance assessment.
Florin Mihaila, Ion Fuiorea, G. Cican· Engineer· 0 citations
A framework based on the twin-delayed deep deterministic policy gradient algorithm is coupled with a validated numerical simulation of a two-bladed VAWT, aimed at maximising power generation.
Jarno Platenburg, Brice Martin, Thierry Jardin et al.· Journal of Fluid Mechanics· 0 citations