turbulence diagnostics

This study addresses the urgent need for low-cost, reliable structural health monitoring of large flexible wind turbine blades by proposing a damage detection method based on a non-intrusive aerodynamic pressure sensing system (Aerosense). By integrating convolutional neural networks with interpretable machine learning and embedding physical priors from structural dynamics, the approach enables real-time damage detection and quantification of severity using only aerodynamic pressure signals under varying operational conditions and mild turbulence. The method overcomes the limitations of conventional black-box models, significantly enhancing the interpretability and physical consistency of damage identification while maintaining high robustness and practicality.

For computationally expensive rare failure event analysis in nonlinear stochastic systems—such as tall steel frames subjected to stochastic wind excitation—this paper proposes an adaptive multi-fidelity hierarchical sampling method. The approach integrates high- and low-fidelity model data to construct an adaptive deep learning surrogate, embedded within a hierarchical importance sampling and multi-fidelity Monte Carlo framework, ensuring unbiased estimation while substantially improving computational efficiency. Key contributions include: (i) adaptive, data-driven updating of the surrogate training process; and (ii) fidelity-coordinated modeling to alleviate bottlenecks associated with high-accuracy finite element simulations. In predicting exceedance probabilities of wind-induced nonlinear structural responses, the proposed method reduces computational cost by over one order of magnitude compared to single-fidelity approaches, while preserving accuracy in failure probability estimation.

Existing metrics, such as relative L² error, often fail to comprehensively assess the numerical plausibility of learned PDE simulators. To address this limitation, this work proposes the first architecture-agnostic, post-hoc diagnostic framework that systematically audits a model’s behavior as an approximate evolution operator through structural indicators—including semigroup consistency, energy dynamics, and response to perturbations. Requiring only reference trajectories, predictions, equation metadata, and simulation configurations, the framework uniformly evaluates diverse architectures such as FNOs, DeepONets, U-Nets, and ResNets. Experiments across five canonical PDE benchmarks demonstrate that even when L² errors are low, structural metrics can exhibit significant degradation, thereby underscoring the necessity and efficacy of the proposed approach.

This study addresses the challenge of efficiently localizing faulty modules in automotive system-level 0D simulations following model updates—a process that traditionally incurs high verification costs and prolonged cycles. To overcome this, the authors propose a novel diagnostic approach based on graph-structured modeling, which uniquely integrates Dynamic Mode Decomposition (DMD), linear programming, and autoencoders to embed system simulation behaviors into a graph representation. This framework enables automatic fault module identification with only a minimal number of simulation runs. The proposed method substantially reduces computational overhead, enhances fault localization efficiency, and seamlessly integrates into existing engineering validation workflows, offering both practical utility and strong scalability.

Laboratory plasma diagnostic data are inherently complex, susceptible to hardware degradation, and embedded in a vast configuration space, rendering traditional analysis methods inefficient. This work introduces, for the first time in this domain, an energy-based model (EBM) that integrates convolutional neural networks with attention mechanisms. Trained on synthetically generated device parameters paired with corresponding diagnostic time series, the model enables unified multi-task learning, achieving high-fidelity signal reconstruction, probe position inversion, conditional inference, and unconditional multimodal generation. The architecture effectively captures inherent data symmetries, facilitates anomaly detection, and demonstrates strong performance on real-world data: incorporating auxiliary diagnostic information significantly reduces reconstruction error and accurately reproduces the multimodal nature of the underlying data distribution.

This study addresses the challenge of reliably distinguishing genuine structural anomalies from benign disturbances in infrastructure health monitoring under varying environmental and operational conditions. To this end, the authors propose a physics-informed digital twin framework that performs real-time anomaly inference by quantifying inconsistencies between a healthy-structure twin model and actual sensor responses. The approach integrates an intrinsically compressed dynamic representation, residual-enhanced spectral features, and a persistence-based decision mechanism. Notably, it is the first method to unify the detection of diverse damage mechanisms—including abrupt changes, gradual degradation, nonlinear behavior, and damping loss—within a single framework. Evaluated across seven numerical benchmarks, the method achieves zero sustained false alarms in undamaged states and limited detection latency under damage scenarios, substantially enhancing monitoring robustness, reliability, and generalization capability.

This study addresses the challenge of accurately identifying impact parameters—velocity, mass, and energy—in aerospace composites under data degradation or noise interference. To this end, the authors propose a unified modeling framework that synergistically integrates physical priors with data-driven learning. By constructing an input space based on a physics-informed energy metric, designing a decoupled surrogate model, and incorporating a hybrid physics-constrained loss function within a neural network architecture, the approach systematically embeds observational, inductive, and learning-based physical biases. This enables decoupled inference of impact parameters while ensuring kinetic energy consistency. Experimental results demonstrate that the method achieves mean absolute percentage errors below 8% for both impact velocity and mass, and below 10% for energy, exhibiting robust generalization and stability even under data sparsity, high noise levels, and damage-induced conditions.

This work addresses the challenge of high-resolution turbulence modeling, which is hindered by the prohibitive cost of direct numerical simulation and the scarcity of full-resolution data, while existing compression methods struggle to preserve physical fidelity and generalize across resolutions. To overcome this, the authors propose PPLC, a local patch-based latent-space compression framework that, for the first time, enables zero-shot cross-resolution transfer without retraining. Leveraging scale similarity in the inertial subrange, PPLC integrates a shared variational autoencoder, exact mean preservation, zero-mean fluctuation encoding, an invertible Haar wavelet front-end, translation-consistency regularization, and overlap-aware reconstruction to rigorously enforce turbulence physics in the latent representation. Evaluated on 1024³ turbulence fields, PPLC substantially outperforms both classical and learning-based baselines, accurately preserving key diagnostic quantities—such as dissipation rate, vorticity, energy spectrum, and incompressibility—in zero-shot reconstructions.

This study addresses the scarcity of damage-state data in bridge health monitoring—particularly for critical failure mechanisms such as deck deterioration and foundation settlement induced by flood scour—by integrating a high-fidelity physics-based finite element model with full-scale experimental measurements from a test bridge. Leveraging Bayesian model updating, key mechanical parameters are calibrated to construct a functional digital twin. This approach represents the first integration of a full-scale physical test bridge with a Bayesian-driven digital twin, effectively bridging the scale gap between laboratory research and real-world bridge monitoring. The resulting framework enables high-fidelity simulation of structural responses under multiple loading scenarios, offering a robust decision-support system for infrastructure safety assessment and service-life extension.

This study investigates whether the internal mechanisms of the scientific foundation model Walrus align with physical principles when reproducing continuum dynamics, and examines the relationship between its representations and performance. By introducing sparse autoencoders (SAEs) at specific layers, the work pioneers the use of enstrophy—the integral of squared vorticity—for physically grounded filtering and prioritization of large-scale features, complemented by comparative numerical simulations. The findings reveal that while the model’s feature activations exhibit segment-wise consistency, they do not correspond to physically meaningful decompositions. Notably, certain output inaccuracies, such as excessive energy dissipation, can be traced to variations in specific SAE features. The study underscores fundamental challenges in achieving representational fidelity and interpretability in scientific foundation models.