The reliable operation of Unmanned Aerial Vehicles (UAVs) as mobile Internet of Things (IoT) assets is critically dependent on robust, real-time anomaly detection within their high-dimensional flight data. Conventional methods based on reconstruction error often fail to capture the strong spatial-temporal dependencies and nonlinear dynamics inherent in time-series flight data. To address these limitations, this paper proposes the Nonlinear Latent Dynamic Consistency Probabilistic Variational Auto-Encoder (NLDC-VAE), a novel framework that synergistically integrates a hybrid convolutional-recurrent encoder for spatio-temporal feature extraction with a nonlinear, GRU-based transition network that enforces latent dynamic consistency. Furthermore, a probabilistic decoder enables robust uncertainty quantification by reconstructing a full distribution with mean and variance. A dual anomaly score (DAS) mechanism is introduced, which decouples the assessment of anomalies into two distinct components: the observation anomaly score measures the Negative Log-Likelihood of reconstruction, and the dynamic anomaly score quantifies latent inconsistency via Kullback-Leibler divergence. Experimental results on both simulation and real-world UAV flight data demonstrate that NLDC-VAE achieves superior anomaly detection performance and robustness, confirming its effectiveness for safety-critical UAV monitoring applications.

With the rapid development of power systems, electricity theft has become a major threat to grid security and economic stability. Existing methods have advanced in periodic modeling, spatiotemporal fusion, and class imbalance handling, but fail to preserve cross-scale structural invariance (CSSI), limiting accuracy and robustness. This article proposes DKGT-Net, a novel detection framework addressing class imbalance, temporal feature insufficiency, and inadequate global dependence modeling. First, a distribution-preserving generative adversarial network (GAN) generates high-quality theft samples with statistical calibration to mitigate imbalance. Second, a CSSI-oriented dual-branch architecture combines the Kolmogorov–Arnold network for global nonlinearities and periodicity with a gated recurrent unit for local and mid-term dynamics, adaptively fused by attention. Third, a consistency-driven training strategy embeds the prior that normal users remain consistent while theft users display degradation, reducing false alarms and amplifying structural inconsistency. Experiments on real-world data from the State Grid Corporation of China show DKGT-Net achieves an F1-score of 95.13% and an area under the receiver operating characteristic (ROC) curve of 97.6%, outperforming existing methods with clear separability in visualization. The framework enhances detection accuracy and robustness while offering potential for lightweight real-time deployment and applications in load forecasting, demand response, renewable energy integration, and industrial anomaly detection.

Aero-engines operate in complex environments, where accurate prediction of their remaining useful life (RUL) is critical to flight safety and operational efficiency. The rapid development of the Internet of Things (IoT) enables more accurate real-time monitoring of engines, providing valuable sensor data for RUL prediction. However, existing studies have largely overlooked the impact of frequently varying operating conditions on degradation patterns. To address this issue, this paper proposes an innovative prediction framework termed the Causal Adaptive Residual Time Network (CART-net). The framework first treats operating conditions as causal modulating factors, introducing Conditional Adaptive Dynamic Feature Enhancement (CADFE) and the Causal Adaptive Attention Mechanism (CAAM). CADFE effectively decouples the influence of operating conditions by dynamically weighting sensor data, while CAAM integrates condition-guided strategies with causal masking to enhance temporal consistency and capture dynamic correlations in the degradation process. Subsequently, a Dynamic Temporal Convolutional Network is employed, leveraging a multi-layer causal convolutional structure, dynamic temporal embedding, and adaptive residual connections to effectively capture long-term dependency patterns and nonlinear temporal evolution characteristics during degradation. Experimental results on the CMAPSS and N-CMAPSS datasets demonstrate that CART-net significantly outperforms existing mainstream methods in key metrics such as RMSE and Score, offering a novel theoretic perspective and technical approach for RUL prediction of aero-engines under varying operating conditions.

The X-ray tube is the core component of computed tomography (CT) equipment, directly affecting imaging resolution and diagnostic accuracy. Degradation and failure prediction ensures the safe and reliable operation of X-ray tube. This article proposes a filament degradation prediction method for X-ray tubes based on Internet of Medical Things (IoMT) time-series data. First, this article analyzes the degradation mechanism of the filament and construct a health indicator based on filament current. Subsequently, key setting parameters are fixed to filter the original data, obtaining pure degradation information. Then, a multiscale attention prediction (MSAP) model is constructed to learn the filament degradation process from historical filament current data, and an ensemble epistemic uncertainty capture method is proposed to ascertain the uncertainty of prediction results. Finally, a failure threshold determining method is designed to predict the remaining useful life of the tube. Supported by the IoMT platform of West China Hospital, clinical monitoring data from four X-ray tubes that failed due to filament burnout were collected. Experimental results demonstrate that the proposed method outperforms existing state-of-the-art methods, achieving root mean square error and score values of 0.0249 and 0.0016, respectively. The proposed maintenance strategy is anticipated to yield economic benefits of 126 000–31 500 yuan per X-ray tube, significantly reducing downtime, and ensuring timely treatment for patients.

Distributed Acoustic Sensing (DAS) instrument connected to dark fibers that widely deployed near roads can collect quasi-static strain signals generated by vehicles over a large range. This approach addresses the high deployment and maintenance costs and limited coverage of traditional roadside sensing technologies, making DAS a highly promising vehicle detection technology for intelligent transportation systems. However, using existing communication cables rather than specially laid sensing cables as the DAS sensing medium, while offering ultra-low deployment and maintenance cost advantages, poses significant challenges for detecting lightweight, low-speed vehicles that are far from the fiber cable. The quasi-static strain generated by such vehicles are low and easily overwhelmed by environmental noise and DAS fading noise. In this paper, we analyze the causes of weak vehicle quasi-static signals and propose a Unet image segmentation network, trained to recognize these weak signals using a large window with a small step size for data input. In a typical campus test scenario containing numerous lightweight low-speed vehicles, we tested various vehicle quasi-static signals using Unet. The results demonstrated that our method has high recognition accuracy and excellent resistance to DAS fading noise.

Incorrect labels as well as the discrepancy between training and test domain data distributions can significantly affect the effectiveness of supervised data-driven models in machine fault diagnosis applications. Such a challenge can be characterized as the noisy label-domain generalization (NL-DG) problem. In this article, the extended invariant risk minimization (EIRM) is developed, which incorporates flat minima seeking to address the NL-DG challenge. The ability of handling NL-DG is realized by shifting the gradient penalty base from the dummy classifier to the entire model. EIRM is shown to be closely related to locating a flat minimum, which is crucial for label noise (LN) robustness and model generalization. Explorations on function smoothness and algorithm convergence are offered to understand EIRM from the theoretical aspect. An efficient implementation of EIRM is also developed to construct the fault diagnosis model. The EIRM-based fault diagnosis method is compared with strong benchmarks on multiple NL-DG tasks using actuator and gearbox fault datasets. Results indicate that the EIRM-based method on average is more effective than the benchmarks. The code is available at https://github.com/mozhenling/doge-eirm.

The Surface Wave Investigation and Monitoring (SWIM) aboard the China-France Oceanic Satellite (CFOSAT) provides the ocean wave spectrum (70–50 m wavelength range). However, the accuracy of this data is affected by speckle noise, low-frequency parasitic peaks, and missing information in the short wavelength range. To improve the accuracy of the SWIM wave spectrum, this letter introduces a vision transformer (ViT) deep learning (DL) model combined with a deconvolution block, which leverages buoy wave spectrum and full wavenumber wind wave spectrum to improve the SWIM wave spectrum with high precision and wide wavelength range. The results show that the linear correlation coefficient of the improved wave spectrum has increased from 0.510 to 0.833. Furthermore, the accuracy of spectrum parameters is enhanced. Particularly, compared with the original SWIM spectrum, the root mean square error (RMSE) for the mean wave period (MWP) and peak wave period (PWP) decreased by 70.19% and 71.68%, respectively.

Machine learning has been widely applied to study AI-informed machinery fault diagnosis. This work proposes a sparsity-constrained invariant risk minimization (SCIRM) framework, which develops machine-learning models with better generalization capacities for environmental disturbances in machinery fault diagnosis. The SCIRM is built by innovating the optimization formulation of the recently proposed invariant risk minimization (IRM) and its variants through the integration of sparsity constraints. We prove that if a sparsity measure is differentiable, scale invariant, and semistrictly quasi-convex, the SCIRM can be guaranteed to solve the domain generalization problem based on a few predefined problem settings. We mathematically derive a family of such sparsity measures. A practical process of implementing the SCIRM for machinery fault diagnosis tasks is offered. We first verify our theoretical exploration of the SCIRM by using simulation data. We further compare SCIRM with a set of state-of-the-art methods by using real machinery fault data collected under a variety of working conditions. The computational results confirm that the machinery fault diagnosis model developed by the SCIRM offers a higher generalization capacity and performs better than the other benchmarks across the different testing datasets.

With the development of unmanned aerial vehicle (UAV) technology, a UAV is gradually applied to a variety of civil fields, such as photography, power line inspection, and environmental monitoring. At the same time, the safety and reliability of a UAV also attract wide attention. Anomaly detection is one of the key technologies to improve the safety of an UAV. The structure of the UAV system is complex, and there are complex spatio-temporal correlations among the high-dimensional flight data with many parameters. However, the existing methods often ignore the spatio-temporal correlation of data and lack parameter selection, which is used to abandon the parameters without a positive impact on anomaly detection results. This article proposes a spatio-temporal correlation based long short-term memory (LSTM) method for anomaly detection and recovery prediction of UAV flight data. First, an artificial neural network correlation analysis is proposed to preliminarily mine the spatio-temporal correlation in flight data and to obtain the correlation parameter sets. Second, the LSTM model is established, and the mapping among different parameters is realized. Finally, anomaly detection and recovery prediction are carried out based on parameter sets mapping model. The effectiveness of the proposed method is verified by generating sample sets with anomaly injection on real UAV flight data.

Due to the fact that single sensor data contains only partial information about complex systems, multiple sensors are often embedded to simultaneously monitor the health state and predict the remaining useful life (RUL). This brings new challenges to traditional approaches that focus on single sensor in terms of data fusion, optimization, and uncertainty management. To address these challenges, this article proposes a novel RUL prediction and uncertainty management framework for multisensor systems. In this framework, a composite 1-D health indicator (1-D HI) is obtained from multiple sensors by optimizing some HI characteristics, including monotonicity, robustness, fitting error, and range information, to better describe the underlying degradation process. A multiobjective grasshopper optimization algorithm is used to achieve the optimal weight vector of the fusion model. Then, an unscented particle filter is introduced to predict the RUL by combining the degradation model constructed from HI and the composite 1-D HI as measurement. To manage the uncertainty in prognosis, a probability distribution of failure threshold and noise parameter adjustment are developed. Experimental results on aircraft turbine engine degradation and comparison with state-of-the-art methods are presented to demonstrate the effectiveness of the proposed framework in RUL prediction of multisensor systems.

Mechanical rotating components such as bearing and gear are widely used in various industrial occasions. Fault diagnosis of rotating component can guarantee operational reliability and reduce maintenance cost of system. Although many fault diagnosis methods have been developed based on artificial intelligence methods, most of them ignore the existing signal processing knowledge in rotating element fault diagnosis domain. Meanwhile, these methods just assume that working conditions such as load or rotating speed keep constant, which may cause low accuracy once operating condition changes. To address these problems, an enhanced intelligent fault diagnosis method for rotating component is proposed based on multifeature and convolutional neural network (CNN). The vibration signals are first transformed into angular domain by resampling technique. Then, the angular domain signals are converted to obtain corresponding envelope and squared envelope spectrum features, which are fused into red–green–blue color image form to enhance sample features and enlarge differences among various health states. Finally, a CNN is constructed to accomplish fault recognition. Experimental results show that the average diagnosis accuracies of planetary gearbox and rolling element bearing datasets can reach 96.7% and 95.65%, which demonstrate that the multifeature approach is effective in fault diagnosis of rotating components in scenarios with different rotating speeds.

Lithium-ion rechargeable batteries are widely used in various electronic products and equipment due to their immense benefits in power supplying. The exact remaining useful life (RUL) prediction of lithium-ion batteries has shown excellent achievements in preventing severe economic and security consequences incurred in failing to provide necessary power levels. Recently, the nonlinear-drifted fractional Brownian motion made quite a splash in RUL prediction, since its first hitting time distribution can be approximated by weak convergence theorem and time-space transformation. However, the previous RUL prediction methods based on fractional Brownian motion only considered current state measurement. In this paper, a prediction framework based on nonlinear-drifted fractional Brownian motion with multiple hidden state variables is put forward to estimate RUL. Specifically, all the parameters of nonlinear function are defined as specific hidden state variables of lithium-ion battery degradation model, and all the state measurements are used to posteriorly estimate the distribution of the multiple hidden state variables by unscented particle filter algorithm. Four sets of lithium-ion battery degradation data provided by NASA Ames Research Center are used to validate the proposed prediction framework. According to comparison study with other methods, the proposed prediction framework demonstrates greater precision in the RUL prediction.

A hard disk drive (HDD) failure may cause serious data loss and catastrophic consequences. Online health monitoring provides information about the degradation trend of the HDD, and hence the early warning of failures, which gives us a chance to save the data. This paper developed an approach for HDD anomaly detection using Mahalanobis distance (MD). Critical parameters were selected using failure modes, mechanisms, and effects analysis (FMMEA), and the minimum redundancy maximum relevance (mRMR) method. A self-monitoring, analysis, and reporting technology (SMART) data set is used to evaluate the performance of the developed approach. The result shows that about 67% of the anomalies of failed drives can be detected with zero false alarm rate, and most of them can provide users with at least 20 hours during which to backup the data.

Health monitoring is used to analyze and predict the battery health status. However, no matter what health monitoring methods and parameters are, a major aim is to improve the battery reliability through surveillance and prognostics. Hence, the latest known methods of state estimation and life prediction based on battery health monitoring are discussed in this paper. Through comparing their characteristics respectively, a prognostics-based fusion technique is proposed that combines physics-of-failure (PoF) with data-driven technology. The fusion approach not only investigates battery failure mechanism caused by environmental and internal characteristics, but also assesses parameters with aid of real-time health monitoring. The specific method is presented to realize the estimation on remaining useful life (RUL) of batteries.

Evidence conflict that may cause the counter-intuitive results is one of the most concerns for information fusion by Dempster-Shafer's (D-S) evidence theory. To deal with the issue and manage evidence conflict greatly for the improvement of belief co