Search Results (101)

Fatigue health monitoring of engineering structures requires continuous degradation assessment, yet ground-truth health labels are unavailable during run-to-failure tests. Existing self-supervised approaches rely on monotonic degradation assumptions that are violated by the structured non-monotonic behaviour of acoustic emission signals during fatigue. A self-supervised framework called Cross-Modal Degradation Rivalry (CMDR) is proposed, which introduces the Modal Rivalry Index (MRI) as a directional measure of cross-modal predictability between heterogeneous sensor modalities. CMDR comprises a label-free representation-learning stage trained via the Cross-Modal Prediction Asymmetry (CMPA) pretext task, followed by a lightweight supervised stage that maps MRI features to scalar health indicators (HIs) using normalised lifecycle labels. The MRI is conceptually related, under the stated assumptions only loosely met in practice, to the Transfer Entropy difference between sensor latent channels. Experiments on a structural fatigue dataset with seven specimens under two loading conditions demonstrate that CMDR achieves competitive trendability and prognosability, as well as the lowest remaining useful life (RUL) error in three of four scenarios. RUL evaluations are additionally repeated under a fully online estimator that uses only training specimens. A strictly inductive ablation that re-pre-trains the self-supervised stage within each leave-one-specimen-out fold confirms a bounded transductive-vs-inductive gap, and CMDR remains the best against three further self-supervised baselines on the within-condition and mixed-condition scenarios. Ablation studies confirm the necessity of directional asymmetry, bottleneck architecture, and momentum-updated target encoders. Full article

IoT devices operate as integrated systems spanning hardware, firmware/software layers, and communication layers. In operational settings, many faults and performance degradations are emergent: they arise from cross-layer interactions, workload changes, and telemetry artifacts, rather than a single physics-of-failure mechanism. These realities make traditional supervised fault classification difficult because labeled fault data are rarely available during deployment, and the fault surface is unknown and a priori. This paper presents a practitioner-oriented, label-free fault detection and diagnosis (FDD) pattern based on Dynamic Time Warping (DTW) for rapid implementation in production IoT telemetry. The method represents a device as a sequence of overlapping episodes and organizes telemetry into interpretable layers (hardware sensors, communication health proxies, and software/firmware-derived KPIs). A reference library of regular episodes is built from an assumed-healthy training window; new episodes are scored using constrained DTW distances against this library, while retaining per-layer and per-channel contributions for attribution. We show that production performance depends strongly on operational parameterization, including episode length, DTW constraints, robust threshold learning, and temporal validation. Within a verified-healthy evaluation window, the tuned configuration achieves an AUROC of 0.97 for the temporally structured faults DTW is suited to (bias, drift, and interaction faults, with spikes detected at an AUROC of 0.93), detecting 100% of injected faults, with a mean delay under 25 min. We further show that constant-value (stuck-at) and missing-data (dropout) faults fall outside DTW’s shape-matching scope (AUROC about 0.66) and are better served by complementary variance- and missingness-based detectors, a consequence of DTW’s shape-matching scope rather than a parameter choice. This work contributes a system-level methodological framework for deploying DTW as an IoT fault-detection-and-diagnosis capability: an episode-and-layer architecture aligned with hardware, communication, and software/firmware ownership; a label-free reference library requiring only assumed-healthy data; per-layer and per-channel attribution for cross-domain triage; and a reproducible operational tuning procedure. Together, these deliver a fast-to-deploy, scalable, and accurate first-line detector for label-scarce IoT systems. Full article

Multivariate time-series anomaly detection on physical sensor networks faces three challenges that generic deep learning models inadequately addressed: heterogeneous sensor reliability, context-dependent anomaly scoring, and inactionable binary outputs lacking per sensor attribution. We propose an uncertainty-aware Temporal Convolutional Network (TCN) framework built on two tightly integrated uncertainty-driven components: (i) an Adaptive Uncertainty-Aware Attention (AUAA) mechanism that gates temporal attention weights by per sensor predictive uncertainty obtained from Monte Carlo dropout; and (ii) a Dynamic Weight Adapter that learns context-sensitive blending of reconstruction error and uncertainty via a GRU over weight history. The architecture also includes an exploratory per sensor attribution head, which we audit rather than claim: a controlled-perturbation test shows it is not yet causally faithful. We complement the empirical architecture with two distribution-free theoretical results: a Chebyshev-type false-positive bound on the hybrid anomaly score, and a Monte Carlo posterior moment convergence result at rate $O\left(M^{-1/2}\right)$. Evaluated on four-month indoor air quality sensor data, the Full Enhanced model achieves $R^2=0.9988$ and MSE $1.65\times {10}^{-4}$, a $25.2\%$ MSE reduction over the Base TCN ($R^2=0.9984$, MSE $2.20\times {10}^{-4}$). Because the IAQ stream is unlabeled, the primary quantitative detection evaluation uses the labeled Skoltech Anomaly Benchmark (SKAB), a publicly available industrial water-circulation corpus disjoint from the IAQ training distribution; it yields an 8.8 × F1 advantage ($0.477$ vs. $0.054$) and a 14.4 × recall advantage ($0.418$ vs. $0.029$) for the proposed model configuration over the Base TCN at a validation-calibrated threshold applied without retuning. Against twelve established detectors under a unified protocol, the proposed model attains the best F1 and recall, while the strongest reconstruction baselines retain higher precision and a marginally higher ROC-AUC, a recall-driven trade-off. Ablation isolates each component’s contribution, the detector degrades gracefully under channel masking and noise, and the distribution-free false-positive bound is empirically respected. The framework retains a low inference cost ($0.16$ ms per window at $M=20$ Monte Carlo samples, including the uncertainty pass). Full article

Laser beam welding (LBW) of aluminium busbar-to-terminal connections for electric-vehicle battery packs requires precise in-process monitoring. Membrane-free optical microphones provide a high-bandwidth (DC–MHz) acoustic channel that captures keyhole, melt-pool, and plume dynamics. This study proposes Acoustic Signal Intelligence (ASI), a deep learning framework for LBW state classification from a single optical microphone, evaluated on an open dataset (183 AA1050 welds, $f_s$ = 2.5 MHz) under a five-class taxonomy: lack of fusion, lack of connection, sound, marginal, and piercing. The contributions are: (i) a compact 1-D CNN encoder on a mel-scale STFT spectrogram, reaching the highest macro-F1 (0.72 mean across three-fold replicate-out cross-validation) and 100% piercing recall in every fold—a multi-representation fusion variant adding a wavelet-packet decomposition and a 24-feature library targeting the 8, 63 and 110 kHz keyhole-resonance peaks was evaluated as an ablation arm and did not survive cross-validation, so the proposed model is mel-only; (ii) a systematic benchmark against six classical-ML and four deep learning baselines in which Transformer-hybrid ablations and ACGAN-style augmentation underperform compared to the compact CNN on the 122-sample training set, with the Transformer underperformance confirmed by a 30-configuration grid search over learning rate, weight decay, and dropout (best tuned macro-F1 = 0.441 vs. CNN 0.724); and (iii) a Grad-CAM analysis that recovers the keyhole-resonance bands without prior knowledge. A single optical microphone is thus a viable real-time alternative to multi-sensor stacks for battery-pack laser welding. Full article

By leveraging the complementary advantages of free-space optical (FSO) and radio frequency (RF) links, hybrid FSO/RF systems exhibit broad application prospects. However, maintaining robustness while performing trade-off optimization between reliability and transmission efficiency under dynamic conditions with power constraints remains challenging. To address this, we propose a weather-aware multi-objective adaptive power allocation approach for hybrid FSO/RF systems based on the Non-dominated Sorting Genetic Algorithm II (NSGA-II) and a deep neural network (DNN). Closed-form expressions for the average bit-error rate (ABER) and average channel capacity (ACAP) are derived and used to evaluate NSGA-II objectives, generating a labeled optimal allocation dataset across diverse scenarios. A DNN is then trained on the dataset to learn adaptive power allocation strategies for dynamic environments. Numerical results demonstrate that the proposed scheme effectively achieves adaptive power allocation and significantly outperforms existing benchmark schemes. In dynamic scenarios, it reduces the ABER by 1–2 orders of magnitude and substantially lowers the outage probability (OP), while improving the overall ACAP by more than 0.5 Gbps under the same transmit power. Full article

In Low Earth Orbit Beam-Hopping Satellite Systems (L-BHSS), co-channel interference among beams severely degrades communication quality. To address the inter-beam co-channel interference avoidance problem, this paper proposes a Parameter-Sharing Multi-Agent Deep Deterministic Policy Gradient-Based Beam Geometry Multi-Parameter Optimization Algorithm (PS-MADDPG-BGMPOA) for the joint optimization of satellite beam geometric parameters. The effects of free-space path loss, atmospheric impairments, and Rician fading are comprehensively considered, and a beam geometric multi-parameter optimization model is formulated with the objective of maximizing the long-term Signal-to-Interference-plus-Noise Ratio (SINR), incorporating beamwidth, beam center offset from the satellite nadir direction angle, inter-beam separation angle, and beam activation states. To tackle the resulting high-dimensional mixed action space, the proposed algorithm employs parameter sharing and grouped decision-making, which alleviates the dimensionality explosion problem and decouples the network scale from the number of beams, enabling efficient cooperative optimization with reduced training complexity. Simulation results show that, under various channel conditions and beam configurations, the proposed method effectively enhances communication quality and spectral efficiency while exhibiting superior real-time performance and stability. Full article

In ubiquitous Wi-Fi sensing, human motion interval segmentation is crucial for applications ranging from basic intrusion detection to advanced activity understanding. Existing methods often treat the Channel State Information (CSI) primarily as time series, overlooking its rich information in the spatial and frequency domains. To address this, we propose a training-free motion segmentation method that exploits the spatiotemporal features of CSI. We first analyze the discriminative spatial distributions of the CSI Ratio on the complex plane and construct a spatiotemporally dual-constrained local density estimator to characterize motion-induced perturbations. To overcome subcarrier selection challenges, we introduce a packet-level asymmetric truncation-based fusion algorithm, which yields a feature representation with a pronounced bimodal histogram. This enables the automatic determination of the optimal segmentation threshold based on the distribution characteristics of the truncated density image. Experiments in typical indoor environments demonstrate that the proposed method achieves high accuracy in both motion event detection and interval localization. Full article

With the advancement of remote sensing technology, image acquisition has become more convenient and the amount of information captured has significantly increased, playing a vital role in numerous fields. However, cloud cover often results in missing image data, severely affecting data usability. In recent years, although deep learning methods have made progress in cloud removal tasks, the complexity of modeling multispectral band relationships and the scarcity of paired data remain major challenges. To address this, this paper proposes a two-stage codebook fusion vector-quantized generative adversarial network (TCF-VQ GAN) and a training framework. The first stage employs synthetic aperture radar (SAR), MODIS, and cloud-free data for unsupervised training; the second stage performs fusion fine-tuning using SAR and MODIS on paired cloudy/cloud-free data. The model incorporates a space-channel jointed gated convolution (SCGC) module to model irregular cloud cover and combines channel attention for band selection, while a dynamically weighted wavelet alignment loss function (DW²A) is designed to enhance multiscale feature representation. Experiments on the SEN12MS-CR and SMILE-CR datasets demonstrate that the proposed method outperforms existing methods across all metrics: on SEN12MS-CR, PSNR is 31.0397 and SAM is 4.7243; they are 33.5191 and 2.1663, respectively, on SMILE-CR. Furthermore, under fixed paired data conditions, simply adding auxiliary and cloud-free data further improves performance, validating the method’s effectiveness in data-scarce scenarios. Full article

Rapid localisation of trapped victims after urban disasters is essential but challenging because Bluetooth Low Energy (BLE) beacons are intermittent, radio propagation is obstructed by rubble, UAVs are energy-constrained, and real-world multi-UAV training is impractical in high-risk search-and-rescue (SAR) environments. This study formulates post-disaster victim localisation as a cooperative Dec-POMDP and adapts a model-aided federated multi-agent reinforcement learning framework based on FedQMIX. The proposed pipeline combines a lightweight LoS/NLoS surrogate channel model, PSO-based victim-position estimation, return-to-base and map-feasibility safety checks, an SAR-aligned shaped reward, and a leakage-free centralised training state based on estimated rather than ground-truth victim locations. Each UAV trains locally inside a learned digital-twin simulator and periodically shares only QMIX network parameters, avoiding the exchange of raw trajectories or RSSI logs. The framework is evaluated on two synthetic post-earthquake urban maps representing a compact return-to-base scenario and a larger reach-to-destination scenario. Across five independent seeds per method and map, Model-Aided FedQMIX achieves the highest and most stable victim-localisation performance, with the clearest advantage observed in the larger long-horizon scenario. Additional diagnostic tests examine reward-weight sensitivity, RF channel-shift robustness, BLE/smartphone hardware heterogeneity, non-IID client-data variation, and partial-client FedAvg under missing client updates. The results indicate that combining model-aided localisation cues, decentralised value factorisation, SAR-aligned objective design, and federated parameter sharing can improve the robustness of UAV-based victim-localisation policies. The framework also clarifies deployment considerations for federated SAR coordination, including communication payload, privacy boundaries, heterogeneous client experience, device variability, and intermittent connectivity. This study remains simulation-based, and future validation with real UAVs, BLE devices, and rubble-inspired testbeds is required before operational deployment. Full article

In impedance imaging, the incompatibility and nonlinearity of the inverse problem lead to problems such as blurred boundaries and severe artifacts in the reconstructed images, making it difficult to meet the requirements for precise identification of multi-layer tissue structures in the legs. To this end, this paper proposes a post-processing algorithm for leg EIT that integrates the boundary attention mechanism, with a Wasserstein generative adversarial network as the training framework, cyclic residual U-Net as the generator, and the boundary attention module embedded in the RecurrentBlock. This leads to adaptive enhancement of the ability to extract organizational boundary features through a three-path fusion of spatial attention, channel attention, and learnable Laplacian edge enhancement. A leg anatomy prior constraint loss function was designed, integrating six constraints—pixel loss, edge loss, hierarchical tissue constraint, total variation regularization, structural similarity loss, and histogram matching—to guide the reconstruction results to conform to the multi-layered tissue structure features of the leg. A simulation dataset of leg sections containing multiple tissues such as skin, fat, muscle, bone, blood vessels, and nerves was constructed, and the pre-reconstructed images were obtained using the hybrid total variation regularization algorithm as the network input. The simulation results show that, under noise-free and different signal-to-noise ratio conditions, the proposed BAM-R2UNet algorithm achieves the best performance in RMSE, SSIM and PSNR metrics compared with HTV, DnCNN and standard U-Net algorithms, can remove artifacts, accurately restore the boundary and conductivity distribution of leg tissues, and has stronger anti-noise robustness. Full article

Continuous in situ marine holographic observation generates data volumes that challenge onboard storage and transmission. Parallel phase-shifting digital holography (PPSDH) is especially sensitive to compression because phase retrieval depends on consistent four-channel demodulation. We present a training-free spatiotemporal compression framework for sparse-particle marine PPSDH sequences based on background–residual decomposition and a shared four-channel processing path. The background is coded once per temporal window by a discrete wavelet transform (DWT) followed by principal component analysis (PCA), and the dynamic residual is decorrelated by temporal principal component analysis before quantization and entropy coding. The framework is evaluated on three primary 64-frame marine PPSDH sequences using a common reconstruction-and-evaluation pipeline with wrapped-phase root-mean-square error (PhaseRMSE) as the primary metric and amplitude peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) as secondary references; expanded supplementary checks are also reported for nine additional selected 64-frame groups spanning sparse to transitional occupancy. On the primary sequence and within the high-fidelity achieved-rate overlap with the JPEG Pleno anchor codec INTERFERE, the proposed framework reduces PhaseRMSE by about 3.3-fold to 3.4-fold while increasing amplitude PSNR by about 11 dB and preserving amplitude SSIM above 0.99997. Lower-bitrate sweeps further quantify the rate–fidelity trade-off rather than claiming universal low-rate superiority. These results support BG–Res spatiotemporal coding as a practical phase-fidelity-oriented option for the tested sparse-to-transitional marine PPSDH conditions; extension to dense scenes, broader marine conditions, and downstream biological tasks requires separate validation. Full article

Maliciously deployed disco intelligent reflecting surfaces (DIRSs) introduce active channel aging (ACA) to achieve fully passive jamming without requiring channel state information or jamming power. To enhance this capability, we propose a distributed DIRS framework for downlink multi-user multiple-input single-output (MU-MISO) systems. By distributing multiple panels, this framework increases independent reflection paths and introduces inter-panel cascaded reflections, severely exacerbating precoder mismatch. We develop a comprehensive near- and far-field cascaded channel model, deriving closed-form expressions for the interference variance and a sum-rate lower bound in the large-antenna regime. Both pilot training (PT) phase-on and phase-off scenarios are investigated to evaluate the jamming impact under different operational states. Analytical and simulation results reveal that DIRS-induced interference scales with transmit power, imposing a strict rate ceiling. Specifically, at 10 dBm transmit power per LU, the proposed framework not only reduces the achievable sum-rate by over 57% relative to the interference-free scenario, but also improves the jamming impact by approximately 36% compared to the conventional single-panel DIRS, demonstrating superior and robust fully passive jamming capability. Full article

Modeling non-stationary systems with dynamically evolving data distributions remains a fundamental challenge in modern learning and optimization problems. In this work, we adopt a generalized notion of non-stationarity, where distribution shifts across tasks and domains are treated as forms of non-stationary processes. This perspective allows us to study non-stationary behavior in controlled settings such as Few-Shot Learning (FSL) and Source-Free Domain Adaptation (SFDA), where data distributions vary across episodes or domains. Conventional normalization and feature modulation strategies often rely on batch-level statistics, leading to unstable behavior under small-batch, streaming, and distribution-shifted conditions. To address these limitations, we propose Meta-LSTM-Affine, a memory-based meta-adaptive affine modeling (normalization) framework that unifies recurrent temporal memory and meta-learning for robust feature modulation. Unlike batch-statistics-driven normalization, our method employs an LSTM-based affine parameter generator (APG) to dynamically produce channel-wise scale and shift parameters based on both current inputs and historical context. To further enhance task-level adaptability, we introduce three lightweight meta-learning mechanisms—Meta-Initialization, Meta-Conditioning, and Meta-Update—that enable rapid cross-task adaptation without modifying the backbone. A bi-level training strategy with temporal smoothness regularization ensures stable affine parameter dynamics under distributional shifts. We validate Meta-LSTM-Affine on FSL and SFDA benchmarks, including Omniglot, MiniImageNet, TieredImageNet, Office-31, MNIST, SVHN, and USPS. Experimental results show that our method consistently outperforms existing approaches such as BN, MetaBN, MetaAFN, and LSTM-Affine, achieving improved stability and adaptation performance with minimal additional computational overhead. Overall, Meta-LSTM-Affine provides a stable and efficient affine modeling mechanism for learning under generalized non-stationary conditions without relying on batch-level statistics. This generalized formulation of non-stationarity allows us to study distributional changes in controlled and widely used benchmark settings, while maintaining relevance to real-world scenarios such as streaming data, continual learning, and time-evolving environments. Full article

This study proposes a sensor-free approach for indoor fire detection that leverages existing Wi-Fi infrastructure as a passive sensing modality. By extracting Channel State Information (CSI) from prevalent 802.11n Wi-Fi signals and applying an unsupervised deep anomaly detection model, the approach conceptualizes the wireless environment as a sensorless detection field capable of identifying combustion-induced perturbations without requiring any physical sensors. CSI data were collected in both normal and flame-induced states under three combustion conditions (gasoline, wood, plastic), each introducing unique signal perturbations. These data, which exhibit diverse signal perturbations, were used as input to four unsupervised deep anomaly detection architectures: a variational autoencoder (VAE), a 1D convolutional autoencoder (CNN-AE), a long short-term memory autoencoder (LSTM-AE), and a hybrid CNN-LSTM autoencoder. Each architecture was trained exclusively on baseline data to learn compact latent representations of normal signal patterns. Among the evaluated architectures, CNN-AE achieved perfect detection across all scenarios, showing high responsiveness to signal disruptions. LSTM-AE tracks prolonged combustion but struggles with fast-onset anomalies. VAE maintains low error during baseline but misses sharp deviations. These findings validate that Wi-Fi CSI encodes latent combustion features. The method requires no additional sensors and operates on existing signals, enabling scalable smart building integration via lightweight software updates. Full article

Two-phase displacement between cement slurry and drilling fluid in wellbore systems is inherently nonlinear, interface-dominated, and strongly affected by geometric confinement, posing substantial challenges to efficient and stable numerical simulation. Conventional CFD approaches rely on mesh discretization and explicit interface tracking, which become computationally demanding and sensitive to grid quality in complex geometries and convection-dominated regimes. To address these limitations, this study develops a unified physics-informed neural network (PINN) framework for directly solving the coupled incompressible Navier–Stokes and Volume of Fluid (VOF) equations governing pressure-driven displacement. The framework is first validated against canonical transient flows and then applied to two-phase displacement in parallel-plate channels, semicircular bends, and a casing–annulus geometry representative of well cementing operations. The predicted velocity, pressure, and volume fraction fields exhibit strong agreement with ANSYS Fluent (2024R1) results, with relative errors generally around 5%, thereby demonstrating physical consistency and numerical stability without mesh generation or pressure–velocity splitting, while also showing favorable computational efficiency for the cases considered. Sensitivity analyses demonstrate that a smoother casing-shoe geometry significantly enhances PINN convergence, while higher Péclet numbers deteriorate training stability by increasing convection-dominated stiffness and optimization difficulty. The results demonstrate that the proposed PINN framework, with its mesh-free and geometrically flexible characteristics, is a promising approach for modeling multiphase displacement in cementing applications. Full article