This paper examines the suitability and limitations of the Informer deep-learning architecture for forecasting aeroelastic response in high-speed shock-wave/turbulent boundary-layer interaction (SBLI). Dynamic displacement measurements of a thin, compliant panel were obtained in the H2K hypersonic wind tunnel at Mach 5.3 for several cavity-pressure configurations. The Informer, which employs ProbSparse self-attention for long-sequence time-series forecasting, was trained on displacement time histories from a subset of cases and used to predict unobserved scenarios. We systematically varied the forecast horizon (256-768 samples) and data sparsity (1200-4000) and evaluated performance using the normalised root-mean-square error. The model achieved low error (minimum NRMSE 0.19) only for carefully tuned combinations of step size and sparsity, while many other configurations yielded NRMSE values of O(0.5-1) and exhibited systematic underor overestimation. In particular, the Informer tended to underestimate displacement following recent sharp decays and to overestimate signals that contained an initial near-steady phase. These behaviours reveal a strong sensitivity to hyperparameters and difficulty in capturing the non-stationary, thermally buckled, shock-induced dynamics of the panel. The results indicate that, in its standard form, the Informer is not robust enough to serve as a stand-alone predictor for aeroelastic design. Still, it provides a valuable diagnostic benchmark for long-sequence attention models in hypersonic fluid-structure interaction. The study highlights the need for adaptive or hybrid architectures and suggests future applications in virtual sensing and safety monitoring for high-cost hypersonic testing.

Modern regression datasets often contain a mix of weakly informative predictors and purely noisy features. We establish, both theoretically and empirically, that discarding variables whose population covariance with the target is zero does not increase expected generalisation risk, up to a vanishing O(log n/n) term under weight decay, and typically enhances performance in small-sample regimes. Risk bounds show that pruning reduces finite-sample estimation error O(log n/n) without affecting approximation error, while gradients associated with noise dimensions vanish asymptotically. Controlled synthetic benchmarks spanning 27 different configurations confirm these predictions: at n = 10(2) samples with 18/20 informative features (correlation strength of 0.5), pruning reduces MSE by 44 % and increases R-2 from 0.426 to 0.678; as n grows to 104, the benefit tapers, reflecting the growing influence of implicit regularisation. Attribution analysis via SHAP corroborates the oracle-level identification of relevant features. These conclusions are further validated on real-world data using the Boston Housing Dataset, where pruning just two statistically less informative features yields consistent gains in both full-sample and small-sample regimes, despite the presence of negatively correlated predictors-supporting our theoretical claim that correlation strength, not sign, determines informativeness. The safety guarantee is conditional on informativeness defined by nonzero population covariance with the target; variables that are informative only through purely non-monotonic or symmetric effects fall outside this scope. Our results caution against the automatic retention of high-p-value variables, provide sample-size-aware guidelines for feature filtering, and extend to any shrinkage-based learner, including ridge regression, kernel methods, and Gaussian processes. Code and data are publicly released to ensure full reproducibility.

Urban heat islands exacerbate thermal stress in dense cities, where limited green spaces intensify microclimate extremes. This study examines the impact of urban greening on urban flow and thermal fields, using the densely populated city of Athens, the capital of Greece, as a case study. The study utilizes high-resolution computational fluid dynamics simulations. Employing terrain-resolved topography and land-use indices, we investigated and compared the effects of urban greening in densely populated areas. Urban flow and thermal fields are compared between the two case scenarios under identical meteorological forcing. Large-eddy simulations indicate that the redeveloped landscape will reduce near-surface mean air temperatures by almost 1 degrees C on hot summer days, with the effect extending further downstream during moderate prevailing south winds. The cooling effect extends for several kilometers downstream, but at a gradually decreasing rate as the cooled air begins to warm again due to the dense urban landscape. Translated into operational terms, a reduction of 1 degrees C on a hot day near city-specific thresholds corresponds to approximately 3% lower heat-related mortality risk, reduced time in higher universal thermal climate index heat-stress categories, and approximately 2.6% peak-demand relief for the Greek electricity system. These results establish the microclimatic, public health, and energy-relief relevance of single-degree cooling from urban greening, while highlighting the spatial heterogeneity that current synoptic stations underdetect, motivating denser urban monitoring to further evaluate this phenomenon. The above co-benefits, along with the improved ventilation shown by the computational fluid dynamics study, establish 1 degrees C as a policy-relevant microclimate gain rather than a trivial fluctuation.

This paper presents the development and application of a Transformer deep-learning model to fluid–structure problems induced by shock-turbulent boundary layer interaction. The model was trained on data from experiments conducted at a hypersonic wind tunnel under flow conditions that allowed for a Mach number of 5.3 and a Reynolds number of ∼19.3×106/m. The shock-wave turbulent boundary layer interaction occurred over an elastic panel. The Transformer was trained using panel deformation measurements taken at different probe locations and the pressure in the cavity beneath the panel. The trained Transformer was subsequently applied to unseen data corresponding to various mean cavity pressures and panel deformations. The capability of the Transformer to capture aeroelastic trends is promising, with interpolation accuracy shown to depend on the volume of data used in training and the location to which the model is applied. The practical implications of this study for aeroelastic research are significant, offering new insights and potential solutions to real-world aeroelastic challenges.

Accurate simulation of compressible turbulent flows and shock-vortex interactions remains a core challenge in computational fluid dynamics, especially when resolving fine-scale vortical structures alongside strong discontinuities. This paper introduces a hybrid weighted essentially non-oscillatory (WENO) scheme aimed at balancing the demands of shock-capturing and turbulence resolution in compressible flows. The method delivers improved accuracy in capturing both classical flow discontinuities and complex vortical structures typical of turbulent flows. The proposed scheme is tested across core case studies, including one-dimensional shock-entropy wave interactions, two-dimensional double-vortex pairing, and three-dimensional Taylor-Green vortex transition to turbulence. Results show that this hybrid scheme provides sharper resolution of discontinuities and better captures fine-scale turbulent structures. In the double-vortex pairing case, the method reduces numerical dissipation by nearly 20%, compared to earlier versions of WENO schemes, enabling a more precise depiction of vortex dynamics and mixing. For the Taylor-Green vortex, the scheme detects more turbulent structures than the 11th-order method, improving predictions of kinetic energy dissipation and enstrophy evolution. These advancements are vital for applications in science and engineering involving compressible turbulence and shock-boundary layer interactions, where accurately resolving both discontinuities and vortical features is essential.

This study explores, for the first time, the application of vector-quantized variational autoencoders to reconstruct and analyze flow images from fluid dynamics simulations across varying resolutions. The method demonstrates effective reconstruction performance, achieving meaningful quantitative results, such as mean squared error, peak signal-to-noise ratio, and structural similarity index measure. These metrics indicate that the method effectively captures and preserves key flow patterns relative to reference images. However, experiments with latent space interpolation reveal that direct linear interpolation between coarse and fine latents produces lower-quality medium-resolution reconstructions than the reference data, highlighting the need for additional conditioning or constraints. Despite this limitation, the interpolation results outperform naive pixel-based methods, suggesting that the technique encodes meaningful and structured latent representations of flow images. The potential applications of the proposed method in fluid dynamics are significant. By conditioning latent representations with specific flow parameters, the technique can synthesize flow images for novel fluid dynamics scenarios, including intermediate states or parameterized datasets. This capability opens avenues for generating flow states that may not be achievable through direct simulations or experiments, such as scenarios involving dynamically changing boundary conditions-a situation typical in atmospheric, oceanographic, or engineered flows. Moreover, generating such synthetic data could considerably expand the accessible parameter space while reducing the computational cost of running extensive simulations or experiments. These features position the proposed method as a promising tool for compressing, reconstructing, and synthesizing flow data for fluid dynamics research and applications.

Accuracy limitations in solving the compressible Navier–Stokes equations at low speed/Mach numbers have drawn research interest for many decades. Despite numerous attempts to overcome these limitations, a comprehensive solution remains elusive, particularly regarding turbulent flows. This study introduces an innovative numerical approach called the Numerical Enhancement for Hyposonic Accuracy, aimed at effectively addressing the challenges of solving the compressible Navier–Stokes equations in turbulent boundary layer flows within the hyposonic limit. The new method locally adjusts the reconstructed flow velocity, significantly reducing the dissipation of low Mach number features while imposing minimal additional computational cost and enabling straightforward implementation. The effectiveness of the proposed method is validated through implicit large eddy simulations of weakly compressible turbulent channel flow. This validation includes detailed evaluations of the friction Reynolds number, streamwise velocity profiles, and higher-order turbulence statistics. The analysis conclusively demonstrates that the proposed method significantly reduces numerical dissipation in subsonic turbulent boundary layers while preventing the emergence of any artificial noise in numerical simulations at the hyposonic limit.

Virus outbreaks on cruise ships pose significant challenges due to their enclosed environments and high passenger densities. Managing these outbreaks has become even more critical as cruise ships have increased in size and passenger capacity. This study uses numerical simulations to investigate the dispersion of airborne respiratory droplets and aerosols within a passenger cabin on a cruise ship, focusing on the angle of the mechanical ventilation jet's influence. Although previous research primarily focused on larger respiratory droplets that quickly settle, this study emphasizes aerosols under 10 μm that can remain airborne for extended periods. The findings demonstrate that variations in the inflow angle from the ventilation unit can significantly affect aerosol dispersion. The results show that the travel distance of the larger droplets is more effectively restricted at the larger 75° inlet angle. Smaller droplets with a diameter ranging between 1–10 and 10–50 μm can remain airborne 1.4 m above ground 20% and 40% longer, respectively. These insights underscore the importance of tailored air circulation strategies to reduce transmission risks in confined spaces, such as cruise ship cabins, highlighting the need for optimized ventilation design to manage infectious disease outbreaks.

This study investigates the effect of natural ventilation on the distribution of airborne pathogens in narrow, low-ceiling corridors typical of hotels, offices, or cruise ships. Two scenarios are examined: a milder cough at 6 m/s and a stronger cough at 12 m/s. A reference baseline case with no airflow is compared to cases featuring an incoming airflow velocity of 1 m/s (3.6 km/h), examining differences in the dispersal of respiratory droplets from two individuals coughing spaced 5 meters apart. Both individuals cough in the direction of the airflow, assuming one-way traffic to minimize airborne pathogen transmission. Findings indicate that airflow accelerates past the door, exceeding 3 m/s, with gusts reaching 4 m/s due to interaction with recirculation zones. This acceleration affects droplet dispersal. Larger droplets (>150 μm) maintain a ballistic trajectory, traveling 2–4 m, potentially increasing transmission risk but suggesting that a 5-m distancing policy could suffice for protection. Smaller droplets (<150 μm), especially those <100μm, spread extensively regardless of cough strength while containing the most viral mass overall. Thus, distancing alone is insufficient. The study recommends that additional safety measures be enforced, such as wearing masks, stricter usage protocols for corridors by limiting corridor use to one person every 20–30 s, or eliminating natural ventilation when feasible to effectively mitigate transmission risks in such environments.

This study examines the application of Large Reasoning Model (LRM)-based artificial intelligence (AI) agents to accelerate scientific discovery, with a specific focus on the rapid prototyping of numerical algorithms. The research demonstrates that current-generation LRM-based AI agents, when collaborating with human experts, can significantly expedite the development of complex algorithms by formalizing a Human-in-the-Loop (HIL) plus Chain-of-Thought (CoT) workflow and introducing, to our knowledge, the first quantitative benchmark of LRMs on a CFD algorithm derivation task. We test the hypothesis that CoT prompting plus domain-expert oversight reduces the derivation error rate and development time of high-order numerical schemes relative to typical prompting, and we instantiate this hypothesis in a cross-model evaluation suite with end-to-end feasibility, from symbolic derivation (polynomials and smoothness indicators) to automated code generation and solver-level validation. The novelty lies in the use of advanced reasoning artificial intelligence (AI) models to assist in the algorithm development process. To this end, the derivation of key formulae within the widely utilized Weighted Essentially Non-Oscillatory (WENO) algorithm -a high-order algorithm applicable to various fields such as fluid dynamics, astrophysics, and medical imaging- serves as a case study. We employ the WENO algorithm as a test case to help evaluate and demonstrate the capabilities of several AI models in this context, thereby laying the foundation for future research and development in this field. The interaction between a human expert and an LRM was examined in the context of designing and deploying a WENO scheme for simulating vortical flows. Initial AI-generated responses, while generally accurate, required iterative refinement guided by expert knowledge and a CoT approach to correct minor errors and optimize performance. This iterative process demonstrated the importance of user involvement, fostering both deeper engagement and a clearer understanding of the algorithm's intricacies. Optimal performance was achieved through a collaborative partnership that leverages the AI's computational speed and the human's ability to perform logical decomposition and error detection. This collaborative approach facilitates the rapid development of tailored solutions. This study highlights the transformative potential of AI copilots in scientific research, showing that their effectiveness is maximized through synergy with domain experts. The findings suggest that artificial intelligence is poised to significantly accelerate research and development, driving scientific innovation.

The effect of hyperparameter selection in deep learning (DL) models for fluid dynamics remains an open question in the current scientific literature. Many authors report results using deep learning models. However, better insight is required to assess deep learning models' behavior, particularly for complex datasets such as turbulent signals. This study presents a meticulous investigation of the long short-term memory (LSTM) hyperparameters, focusing specifically on applications involving predicting signals in shock turbulent boundary layer interaction. Unlike conventional methodologies that utilize automated optimization techniques, this research explores the intricacies and impact of manual adjustments to the deep learning model. The investigation includes the number of layers, neurons per layer, learning rate, dropout rate, and batch size to investigate their impact on the model's predictive accuracy and computational efficiency. The paper details the iterative tuning process through a series of experimental setups, highlighting how each parameter adjustment contributes to a deeper understanding of complex, time-series data. The findings emphasize the effectiveness of precise manual tuning in achieving superior model performance, providing valuable insights to researchers and practitioners who seek to leverage long short-term memory networks for intricate temporal data analysis. The optimization not only refines the predictability of the long short-term memory in specific contexts but also serves as a guide for similar manual tuning in other specialized domains, thereby informing the development of more effective deep learning models.

This paper concerns the interaction of an impinging shock wave with a supersonic turbulent boundary layer over several distinct and permanently deformed surfaces, resulting in differences in the shock–boundary-layer interaction and the surface acoustic loading. High-order numerical simulations featuring two-dimensional surface deformations typically encountered in experiments are performed. The deformation amplitudes are up to half the incoming turbulent boundary-layer thickness. The results show that the high-pressure region about the shock impingement is significantly altered and can become narrower or wider depending on the local surface inclination of the deformed panel mode. The surface curvature is found to not significantly affect the separation and reattachment locations of the recirculation bubble. The power spectrum analysis of the pressure fluctuations along the panel’s midspan, where the surface attains the largest deformation amplitude, exhibits a rich and varied response. The pressure power spectrum is amplified in all of the surface deformation modes examined, with the magnitude of the amplification varying in the frequency domain, depending on the location and mode.

Upscaling flow features from coarse-grained data is paramount for extensively utilizing computational physics methods across complex flow, acoustics, and aeroelastic environments where direct numerical simulations are computationally expensive. This study presents a deep learning flow image model for upscaling turbulent flow images from coarse-grained simulation data of supersonic shock wave–turbulent boundary layer interaction. It is shown for the first time that super-resolution can be achieved using only the coarsest-grained data as long as the deep learning training is performed using hundreds of fine-grained data. The unsteady pressure data are used in training due to their importance in aeroelasticity and acoustic fatigue occurring on aerospace structures. The effect on the number of images and their resolution features used in training, validation, and prediction is investigated regarding the model accuracy obtained. It is shown that the deep learning super-resolution model provides accurate spectra results, thus confirming the approach's effectiveness.

The dynamic coupling between a Mach 1.94 shock wave/turbulent boundary layer interaction (SBLI) and a flexible panel is investigated. High-order numerical simulations are performed for distinctly different dynamic panel motions and rigid snapshots of their maximum deflected shape. They are compared with a baseline interaction over a rigid planar wall. The panel’s dynamic surface motions were obtained from the Air Force Research Laboratory (AFRL) wind tunnel experiments. The primary aim of the study was to determine whether there were any differences in the flow pressure loading on the compliant panel due to the various rigid and dynamic deformations considered. The results show that the examined panel deformations increase the SBLI size near the panel midpoint, where the deformation amplitude tends to be the largest. Relative to the rigid planar case, the examined surface deformations cause the mean-flow high-pressure surface loading caused by the impinging shock wave to shift downstream along the compliant panel midspan, albeit by a small amount. The spectrogram of the dynamic deformation and the flow surface pressure response suggests that the two are strongly coupled at the dominant (primary) mode but less so at the secondary modes. Although the primary mode frequencies overlap, they do not exactly match, with the pressure response frequency always being slightly higher in all three cases. The rigid deformations did not enhance the pressure power content at the SBLI. However, pre-SBLI and near the panel leading edge, the pressure power spectrum weakly increased throughout the resolved frequency range and overlapped with the onset of the amplification found in the dynamic deformation cases. Post-SBLI, the rigid deformations cause a weak enhancement at frequencies below 1 kHz, which closely match the dominant and secondary pressure response frequencies obtained in the dynamic cases.

This paper presents the development of a novel algorithm for unsupervised learning called RUN-ICON (Reduce UNcertainty and Increase CONfidence). The primary objective of the algorithm is to enhance the reliability and confidence of unsupervised clustering. RUN-ICON leverages the K-means++ method to identify the most frequently occurring dominant centres through multiple repetitions. It distinguishes itself from existing K-means variants by introducing novel metrics, such as the Clustering Dominance Index and Uncertainty, instead of relying solely on the Sum of Squared Errors, for identifying the most dominant clusters. The algorithm exhibits notable characteristics such as robustness, high-quality clustering, automation, and flexibility. Extensive testing on diverse data sets with varying characteristics demonstrates its capability to determine the optimal number of clusters under different scenarios. The algorithm will soon be deployed in real-world scenarios, where it will undergo rigorous testing against data sets based on measurements and simulations, further proving its effectiveness.

This paper concerns the interaction of an oblique shock wave with a supersonic turbulent boundary layer over a thin panel surface, leading to shock–boundary layer interaction and panel buckling. We have performed high-order numerical simulations featuring various static two-dimensional surface deformations typically encountered in experiments. The deformation amplitudes we examined were at least half the height of the incoming turbulent boundary layer thickness. The results show that along the panel midspan, where the maximum deformation amplitude is located, the mean and root mean square pressure are affected by about 10%. Cases for which the pressure at the shock–boundary layer interaction was increased relative to the planar case showed to decrease downstream, and vice versa. Despite the weak response to the mean pressure amplitude, the mean pressure surface contour plots reveal that the streamwise, particularly the spanwise distribution, is affected more noticeably. For example, the surface deformation modes are shown to disrupt the spanwise constant mean pressure, forming higher (or lower) values at either the panel's midspan or edges, depending on the mode. Moreover, the surface curvature leads to a characteristic bending of the spanwise distribution, which can be concave or convex depending on the deformation mode. Analysis of the Reynolds stress anisotropy componentality at different heights from the buckled surface reveals a similar spanwise response of the turbulent velocity fluctuations. The results suggest that the deformation rate plays an important role alongside the deformation amplitude in the turbulent layer and shock–boundary layer interaction.

We present a wavelet analysis of supersonic shock-boundary-layer interaction. We have used direct numerical simulation data for supersonic flow over a compression ramp and performed orthogonal anisotropic wavelets. The wavelet-based method of extracting coherent structures is applied to the flow vorticity field, decomposed into coherent and incoherent contributions using thresholding of the wavelet coefficients. The statistics of the coherent part of vorticity are close to the statistics of the total field. The study aims to improve our understanding of the shock-boundary-layer interaction, the role of vorticity, and the relationship between the flow's coherent and incoherent vorticity components with the near-wall sound. Our analysis shows a substantial correlation between the incoherent part of vorticity components and wall-pressure fluctuations.

This paper investigates deep learning methods in the framework of convolutional neural networks for reconstructing compressible turbulent flow fields. The aim is to develop methods capable of up-scaling coarse turbulent data into fine-resolution images. The method is based on a parallel computational framework that accepts five image sets of various resolutions, trained to correspond to the respective fine resolution. The network architecture mainly consists of convolutional layers, constructing an encoder/decoder network. Based on the U-Net scheme, three different implementations are presented, with residual and skip connections. The methods are implemented in a supersonic shock-boundary-layer interaction problem. The results suggest that simple networks perform better when trained on limited data, and this can be a practical and fast solution when dealing with turbulent flow data, where the computational burden is most of the time difficult to decrease. In such a way, a coarse simulation grid can be upscaled to a fine grid.

The long short-term memory deep-learning model is applied to supersonic shock–boundary layer interaction flow. The study aims to show how near-wall pressure fluctuations can be reconstructed from reduced (under-sampled) datasets of pressure signals. Predicting pressure fluctuations from reduced datasets could allow predictions using less expensive simulations and experiments. The training of the deep learning model is based on direct numerical simulations of supersonic ramp flows, focusing on the regions upstream of and around the shock–boundary layer interaction region. During the pre-processing stage, cubic spline functions increase the fidelity of the sparse signals and feed them to the long-short memory model for an accurate reconstruction. Comparisons are also carried out for different sparsity factors and assess the model's accuracy both qualitatively through the pressure signals and quantitatively using the root mean square error and the power spectra. The deep learning predictions are promising and can be extended to include other aerodynamic or aeroelastic parameters of interest.