During the conceptual design phase of aircraft development, wing structural optimization is frequently omitted or excessively simplified due to the intensive manual labor involved in geometry modeling, meshing, and finite-element analysis. These challenges are exacerbated for innovative configurations like tilt-duct aircraft, which must resolve conflicting requirements from vertical takeoff and landing (VTOL) and cruise modes. This paper presents an automated framework for conceptual-stage wing structural design and optimization, tailored for tilt-duct aircraft. The primary innovation is the incorporation of Bayesian optimization (BO) driven by a novel dynamic expected improvement (DEI) strategy. DEI adaptively manages the exploration–exploitation trade-off, enabling efficient navigation of a mixed-variable design space encompassing discrete variables (e.g., rib and spar counts) and continuous variables (e.g., rib and spar positions and thicknesses). This approach facilitates global search and rapid convergence in computationally expensive black-box evaluations. The framework integrates parametric geometry generation, high-fidelity meshing, and multi-mode finite-element analysis. Applied to a 1000 kg-class laboratory-developed tilt-duct aircraft wing, the DEI-based BO achieved a 22.61

Transonic aircraft design requires repeated aerodynamic evaluation across varying geometries, yet integral coefficients cannot localize component-interference effects or concentrated surface loads, and high-fidelity simulations remain too expensive for large-scale design-space exploration. Existing field-prediction surrogates generally rely on prescribed output discretizations, limiting reuse of geometry representations across resolutions and query sets. We propose the Geometry-Conditioned Adaptive Neural Operator (GeoCANO), which takes sampled surface coordinates as geometric input and learns a mapping from a complete aircraft configuration to a surface-field function. A cross-attention encoder extracts a configuration-level geometry state that modulates a multiresolution axis-factorized coordinate representation, while a query-wise sparse mixture-of-experts decoder allocates predictive capacity according to local approximation difficulty. The resulting set-to-function construction separates geometry encoding from coordinate evaluation, enabling a single encoded configuration to be queried directly on different surface point sets and local regions without reconstructing the output topology or retraining the model. Experiments on a Reynolds-averaged Navier–Stokes dataset containing 10,484 wing–body–nacelle configurations demonstrate improved complete-surface prediction, high-gradient pressure reconstruction, and pressure-drag recovery for unseen geometries within the sampled design family. GeoCANO reduces the \(C_p\) mean absolute error by 77.9\% relative to the best baseline and retains a substantial advantage over a finite near-wall range. Ablation and routing analyses further show that the geometry-modulated coordinate representation achieves higher accuracy with substantially fewer parameters than conventional three-dimensional hash encoding, while the sparse experts develop structured, query-dependent allocation. These results establish GeoCANO as a reusable continuous-coordinate interface for transonic aerodynamic-field evaluation.

Supercritical airfoils operating in transonic flows present an inherent multiscale challenge, where sharp shock discontinuities are coupled with delicate, attached boundary layer structures. Conventional neural operators often suffer from inherent spectral bias, where global smooth features are prioritized over high-frequency singularities or encounter non-physical Gibbs oscillations when attempting to resolve shocks via global encodings. To address this limitation, we propose the physics-guided hash-encoding dual-branch neural operator (PHD-NO). Its central design is a dual-branch hash-encoding architecture tailored to transonic aerodynamic fields, where the background flow and shock-dominated corrections are handled in a scale-decoupled manner. By employing multi-resolution hash grid encoding as a unified feature basis, the framework leverages the locally compact support of fine-resolution grids to achieve exceptional spatial precision at shock locations while maintaining global consistency through coarse-resolution levels. Theoretically, we analyze PHD-NO through the Neural Tangent Kernel framework and argue that the multi-scale hash sub-kernels can compensate for high-frequency eigenvalue decay over a broader frequency range, thereby mitigating the spectral-bias gap. On a high-fidelity supercritical-airfoil dataset, PHD-NO attains mean L1 errors at the 10-3 level, preserves accurate shock localization and surface-pressure reconstruction, and delivers up to about 474 & times; faster CPU-side inference than the CFD solver on the same hardware basis.

Neural operators have emerged as powerful deep learning frameworks for approximating solution operators of parameterized Partial Differential Equations (PDE). However, current methods predominantly rely on Multilayer Perceptron (MLP) for mapping inputs to solutions, which impairs training robustness in physics-informed settings due to inherent spectral biases and fixed activation functions. To overcome the architectural limitations, we introduce the Physics-Informed Chebyshev Polynomial Neural Operator (CPNO), a novel mesh-free framework that leverages a basis transformation to replace unstable monomial expansions with the numerically stable Chebyshev spectral basis. By integrating parameter dependent modulation mechanism to main net, CPNO constructs PDE solutions in a near-optimal functional space, decoupling the model from MLP-specific constraints and enhancing multi-scale representation. Theoretical analysis demonstrates the Chebyshev basis’s near-minimax uniform approximation properties and superior conditioning, with Lebesgue constants growing logarithmically with degree, thereby mitigating spectral bias and ensuring stable gradient flow during optimization. Numerical experiments on benchmark parameterized PDEs show that CPNO achieves superior accuracy, faster convergence, and enhanced robustness to hyperparameters. The experiment of transonic airfoil flow has demonstrated the capability of CPNO in characterizing complex geometric problems.

The transition to green aviation, particularly in Urban Air Mobility (UAM), demands highly energy-efficient aircraft like electric tilt-ducts. A central challenge is designing structures that are lightweight enough for both vertical takeoff-landing (VTOL) and high-speed cruise. This study tackles the complex structural optimization problem arising from these multi-modal demands, where design variables are often a mix of continuous and discrete parameters. We develop a deep reinforcement learning (DRL) framework based on the Proximal Policy Optimization (PPO) algorithm to efficiently navigate this design space without explicit gradients. When applied to a critical load-bearing component, our framework discovered designs with significant mass reduction that satisfied all constraints across flight modes. On a benchmark wing, it achieved a 17.28

Meshfree methods require a pre-processed particle distribution that is suitable for the body to obtain high-precision solutions. Currently, there remains a deficiency in point distribution approaches that can simultaneously satisfy the isotropic point distribution requirements of body-fitted uniform distributions while also addressing the computational speed, scale, and robustness demands of point distribution calculations. In this paper, we propose a novel system for generating body-fitted and isotropic point cloud for meshfree methods applied to complex geometries using High-Performance Computing (HPC). Modular design is applied to this system. Firstly, this system employs an external wall particle distribution module designed to facilitate uniform point distribution of wall particles along with boundary condition marking for various surfaces. Subsequently, an internal fluid particle filling module is utilized for large-scale internal fluid particle filling at variable filling spacings with enhanced robustness. Finally, the body-fitted adjustment of the internal fluid particle module is implemented using an improved Discrete Element Method (DEM) proposed in this paper to refine the internal fluid particle arrangement and ensure compliance with isotropic body-fitting distributions. During computations, this system employs CUDA and MPI technologies to enable accelerated computing. Numerical experiments have conclusively shown that our system is adept at generating a robust and rational isotropic point distribution for complex geometries. This capability satisfies the pre-processing demand for meshfree methods, ensuring both rapid computation speed and enhanced robustness for large-scale point cloud generation.

The nozzle is a critical component responsible for generating most of the net thrust in a scramjet engine. The quality of its design directly affects the performance of the entire propulsion system. However, most turbulence models struggle to make accurate predictions for subsonic and supersonic flows in nozzles. In this study, we explored a novel model, the algebraic stress model k-kL-ARSM+J, to enhance the accuracy of turbulence numerical simulations. This new model was used to conduct numerical simulations of the design and off-design performance of a 3D supersonic asymmetric truncated nozzle designed in our laboratory, with the aim of providing a realistic pattern of changes. The research indicates that, compared to linear eddy viscosity turbulence models such as k-kL and shear stress transport (SST), the k-kL-ARSM+J algebraic stress model shows better accuracy in predicting the performance of supersonic nozzles. Its predictions were identical to the experimental values, enabling precise calculations of the nozzle. The performance trends of the nozzle are as follows: as the inlet Mach number increases, both thrust and pitching moment increase, but the rate of increase slows down. Lift peaks near the design Mach number and then rapidly decreases. With increasing inlet pressure, the nozzle thrust, lift, and pitching moment all show linear growth. As the flight altitude rises, the internal flow field within the nozzle remains relatively consistent due to the same supersonic nozzle inlet flow conditions. However, external to the nozzle, the change in external flow pressure results in the nozzle exit transitioning from over-expanded to under-expanded, leading to a shear layer behind the nozzle that initially converges towards the nozzle center and then diverges.

Physics-informed neural operators have emerged as a powerful paradigm for solving parametric partial differential equations (PDEs), particularly in the aerospace field, enabling the learning of solution operators that generalize across parameter spaces. However, existing methods either suffer from limited expressiveness due to fixed basis/coefficient designs, or face computational challenges due to the high dimensionality of the parameter-to-weight mapping space. We present LFR-PINO, a novel physics-informed neural operator that introduces two key innovations: (1) a layered hypernetwork architecture that enables specialized parameter generation for each network layer, and (2) a frequency-domain reduction strategy that significantly reduces parameter count while preserving essential spectral features. This design enables efficient learning of a universal PDE solver through pre-training, capable of directly handling new equations while allowing optional fine-tuning for enhanced precision. The effectiveness of this approach is demonstrated through comprehensive experiments on four representative PDE problems, where LFR-PINO achieves 22.8%-68.7% error reduction compared to state-of-the-art baselines. Notably, frequency-domain reduction strategy reduces memory usage by 28.6%-69.3% compared to Hyper-PINNs while maintaining solution accuracy, striking an optimal balance between computational efficiency and solution fidelity.

Variable-sweep aircraft can actively change the sweep angle of the wings during flight, causing corresponding changes in the flight envelopes. Aerodynamic forces and moments are both time-varying and dynamic during different sweeping processes. The static aerodynamic characteristics are insufficient to meet the design requirements of modern intelligent variable-sweep aircraft. In this work, a numerical simulation method for studying the unsteady aerodynamic characteristics of variable-sweep aircraft is proposed. An overset grid method is used to numerically simulate a variable-sweep aircraft geometry. The numerical simulation results are obtained, which are well compared with the experimental data. The flow field state at the initial position of the aircraft is analyzed, and the lift coefficient and pitch moment coefficient show dynamic hysteresis loops during the unsteady process. At the same sweep angle, the lift coefficient and pitch moment coefficient during forward sweep are both large, while those during the backward sweep are small. The mechanism of unsteady aerodynamic characteristics is studied. It has been found that during the sweep, the additional velocity generated by the wing movement is the key factor leading to the generation of dynamic hysteresis loops. The additional velocity causes changes in the flow field vorticity during forward and backward sweep, resulting in changes in the pressure distribution on the upper and lower surfaces of the wing.

Low emission combustion technology for gas turbine and aero-engine industrial has attracted more and more attention by the aeronautic community. Hydrogen fuel is believed being a promising alternative fuel for gas turbines and aeroengines in the next generation of propulsion system. In this work, a novel type of annular combustor with multi-point direct injection (MDI) burners for burning pure hydrogen is proposed for gas turbines and aero-engines. The MDI burner is featured with several multi-point injectors arranged in inclined direction, which forms a swirling flow to enhance the mixing process of hydrogen and air. Meanwhile, the MDI burners are beneficial for flame stability and preventing flashback. The flow structure, combustion process and NOx emission of a single MDI burner are numerically studied with hydrogen fuel under different air flow rate Qair and the equivalence ratio phi. Then, the cold flow and combustion characteristics of the annular combustor with 12 MDI burners are investigated by the steady RANS method to understand the flow structure, temperature field and NOx emission under different operation conditions. Furthermore, the thermoacoustic property of the annular combustor is investigated to obtain the acoustic modes, including the longitudinal modes and azimuthal modes.

This paper proposes a new configuration of the dual-channel inlet with an isometric isolation section and uses numerical simulation to investigate its performance concerning off-design conditions. The starting performances of the dual-channel inlet are compared with a single-channel one under variable Mach numbers. The RANS method with a new non-linear turbulence model, the Quadratic Constitutive Relation (QCR) model, is utilized to calculate the inlet's flow field and performance. Numerical results show that the designed single- and dual-channel inlets significantly broaden the starting range, allowing for starting at incoming flow speed below M = 4. The dual-channel inlet achieves start at the angle of attack (AOA) of − 8°, 0°, and 8°, and the angle of sideslip (AOS) of 0°, 4°, and 8°. However, the blunt head between the dual-channel inlet induces detached shock waves, which coupled with the leading edge shock waves, can alter the incoming flow conditions in the inner runner. The single-channel inlet outperforms the dual-channel inlet, as the latter's S-transition section has a significant impact on its performance. This section exhibits complicated flow phenomena with secondary flows, vortex, and shock wave/boundary layer interactions (SBLI), leading to a lower total pressure recovery coefficient (TPR). Moreover, blunt heads and S-transition sections are the predominant factors contributing to considerable total pressure losses, consequently leading to a general decrease in the inlet system performance. It is therefore imperative to meticulously design and optimize the guide mechanism at the entrance and S-transition sections to take full advantage of the outstanding performance of the dual-channel inlet.

For the spectral characteristics of thermochemical nonequilibrium flow field gases and wall coating radiation, as well as the aerodynamic thermal loads of re-entry capsule Fire II, a finite-rate catalytic model combined with the multi-component chemical nonequilibrium N–S equation is developed. Based on this model, the effects of dissociation and recombination reactions of gas components on the wall coating temperature and heating flux are simulated. In addition, we investigate and analyze further the influence of the coating’s catalytic properties on the radiation spectral characteristic of the high-temperature nonequilibrium flow field and the wall coating radiation. The results show that the catalytic properties intensify the non-homogeneous recombination and exothermic processes near the wall surface. It results in a significant decrease in the number of component particles contributing to the radiative spectral intensity near the wall, which slightly reduces the flow field gas spectral radiation intensity. Meanwhile, the catalytic properties also increase the wall temperature and heating flux, significantly increasing the intensity of the wall radiation spectral in the near-infrared region.

Tilt-duct Unmanned Aerial Vehicles (UAV) combine the high-speed efficiency of fixed-wing aircrafts with vertical takeoff and landing (VTOL) and the hover capabilities of rotary-wing aircrafts while maximizing the advantages of ducted fans in terms of noise reduction, efficiency, and safety, making it a pivotal direction for the future of aviation such as urban air mobility. This paper concentrates on the design and optimization of the primary structures of a laboratory-designed reference tilt-duct UAV. Firstly, the general data of the reference tilt-duct UAV are presented. According to the load conditions, the overall structural layout design for the wing, fuselage, and empennage is carried out, where special attention has been paid to account for the requirements of VTOL/hover and cruise flight modes. Based on the structural layout, finite element models (FEM) are established and static analyses are performed. The results indicate that the design can fulfill the structural requirements during a flight mission. Furthermore, based on the Method of Feasible Directions (MFD) algorithm, we have carried out the optimization of the composite wing box that incorporates manufacturing constraints. Via optimization, the total mass of the wing box is reduced by 38.6%, i.e., from 3.73 kg to 2.29 kg. The results indicate that the combination of composite materials with a tilt-duct configuration holds significant potential for future high-efficiency and environmentally friendly aviation.

In this work, a data-driven framework based on digital model is proposed to predict the remaining useful life (RUL) of aero-engines. An encoder-decoder structure is applied to generate a hidden representation of the degradation status of aero-engines. The digital model achieves high similarity between real sensor data and the virtual sensor data generated by the digital model. A hidden representation with lower dimension is generated by data-driven model, which contains the vital features of the aero-engines. A RUL prediction block is also proposed to map the hidden representation into the final RUL result. The NASA C-MAPSS dataset is used to evaluate the performance of the proposed method. The loss between the virtual sensor data from the digital model and the actual sensor data indicates that our proposed data-driven model well represents the physical engine model. Evaluation results of RMSE and Score show that the framework achieves accurate prediction of the remaining useful life. In addition, the proposed framework improves the interpret-ability in the process of data processing and feature extraction, and has high applicability in predictive maintenance.

Supersonic and hypersonic flows have gained considerable attention in the aerospace industry in recent years. Flow control is crucial for refining the quality of these high-speed flows and improving the performance and safety of fast aircraft. This paper discusses the distinctive characteristics of supersonic flows compared to low-speed flows, including phenomena such as boundary layer transition, shock waves, and sonic boom. These traits give rise to significant challenges related to drag, noise, and heat. Therefore, a review of several active and passive control strategies is provided, highlighting their significant advancements in flow transitions, reducing drag, minimizing noise, and managing heat. Furthermore, we provide a comprehensive analysis of various research methodologies used in the application of flow control engineering, including wind tunnel testing, flight testing, and computational fluid dynamics (CFD). This work gives an overview of the present state of flow control research and offers insights into potential future advancements.

Prediction of aero-engine remaining useful life (RUL) is crucial for the aeronautic industrial. In recent years, the deep learning enhanced methods for the aero-engine RUL prediction has attracted more and more attention by the community. However, how to correctly process information, extract features and carry out feature fusion from cluttered sensors is still a challenging problem. In this paper, a hybrid framework containing a method of data reduction and feature extraction based on physics-informed self-attention encoder (PISAE) is proposed to predict the RUL of aero-engines. The self-attention mechanism is used to process the data. In the part of feature extraction, the encoder-decoder architecture is applied to reduce the data dimensions and extract the features. In the encoder section, data features are embedded in vectors of low dimensions. Then, a neural network based on the self-attention mechanism is constructed to make predictions of the RUL. Physical information is embedded into the training process. The proposed framework is experimentally evaluated on the Commercial Modular Aeronautical Propulsion System Simulation (C-MAPSS) dataset. Results show that the framework can effectively improve the prediction accuracy of the remaining service life by information processing and feature extraction. In addition, the proposed framework improves the interpretability in the process of data processing and feature extraction, and has high applicability in predictive maintenance.

Aiming at processes of radiation emission and absorption of high-temperature non-equilibrium flow field gas at various energy level states, the present work extracts the flow field radiation spectral characteristics of the re-entry orbital test vehicle and analyzes the radiation signals at different altitudes and observation angles during reentry. Using an aircraft configuration similar to the X37B, its flow field is modeled using the multicomponent full N-S equations, and the radiation is calculated using the line-by-line method and the quasi-steady-state assumption in combination with the calculation of the high-temperature gas physical properties. Based on this, we analyze the changes in the flow field's gas radiation spectrum characteristics of the vehicle with altitude and observation angle. The effects of the angle of attack and slip wall assumption on aerodynamic heating and radiation characteristics are provided. Additionally, as a validation of the prediction method, this work compares the results of several techniques for predicting nonequilibrium flow field parameters and gas radiation using experimental data from the FIRE II standard model. The results show that both the distribution of the emitted radiation spectrum and the position of the strongest emission peak are in the ultraviolet band, and X37B is more likely to emit a strong radiation emission peak at higher altitudes. The cumulative radiation intensity of most lines of sight increases as the reentry process proceeds, which is consistent with the trend of heating flux. Compared with the slip boundary, the no-slip boundary assumption underestimates the thermal load and radiation intensity of the vehicle to a certain extent.

Hypersonic vehicles create high-temperature environments, thus causingair molecules to undergo chemical reactions. As a result, the nonlinearity of the map-ping relationship between heat flux and inflow condition is strengthened. Accuratelypredicting the heat flux of real gases using machine learningmethods relies on anamount of training data of real gas, but the computational cost is often unacceptable.To solve the issue stated, we propose a novel neural network named Physically GuidedNeural Network based on Transfer Learning (TL-PGNN). We design a new networkarchitecture to depict physical laws and use heat flow data ofideal gases to enhancethe network's capability to depict heat flow from only a few real gas samples. Exper-iments of sphere demonstrate that, compared to ordinary DNNand PGNN, applyingTL-PGNN decreases the mean L1 error by 72.66% and 24.53% on the test set, respec-tively

For the nonequilibrium flowfield and high-temperature gas radiation spectral characteristics of the hypersonic reentry capsule, a spectral prediction method based on quasi-steady-state assumption and line-by-line method is introduced. Meanwhile, the characteristics of the spectrum are studied to investigate the influence of viewing angles and trajectory points during reentry based on this method. We use Fire II flight test data to verify the accuracy of the gas radiation model and update the prediction values of the cumulative radiation intensity in the wavelength range of 2.2–4.1 eV by previous methods. The results show that the radiation signal is most significant at the stagnation line and gradually decreases with the deviation of the line of sight from the stagnation line. The cumulative radiation intensity of the various trajectory points exhibits a profile that increases and then decreases during reentry; it rises to its peak at 50 km. The emission spectrum of gas radiation is mainly distributed in the ultraviolet and near-infrared bands. The transition processes of atoms N and O contribute more than 90% of the total radiation intensity, and the contribution of the [Formula: see text] first negative series in the molecular spectrum is relatively considerable compared to other molecular components.