To address the challenge of real-time fault diagnosis for brushless motors in scenarios where real fault data is scarce, this paper proposes a efficient neural network framework oriented toward embedded deployment. Using the dataset generated by a Semi-Physical simulation platform, this study constructs a complete methodology covering data preprocessing, model design, training optimization, and embedded deployment verification. At the data processing stage, a dataset encompassing 19 fault states is established through timefrequency signal analysis and feature extraction. In terms of network architecture design, traditional complex deep models are discarded, and an innovative minimalist three-layer fully connected network is proposed, which ensures efficient model construction under limited computing resources. The superiority of the proposed method is verified through systematic comparative experiments. Experimental results demonstrate that, while maintaining diagnostic accuracy, the proposed network achieves a 15-fold improvement in computational efficiency and a 92% reduction in storage requirements, thereby providing a quantifiable evaluation benchmark and design paradigm for deep learning applications on resource-constrained platforms.

Fault diagnosis plays an important role in the safety and reliability of aero-engine bearings. Aero-engine bearings operate under highly variable conditions, placing stringent demands on the generalization capability of diagnostic models. Numerous approaches have sought to integrate physical mechanisms and domain knowledge to enhance model generalizability. However, most existing methods directly combine knowledge with data without adequate selection or adaptation of relevant information, which can undermine the robustness of the diagnostic framework. Therefore, this paper proposes an adaptive bearing physical information neural network that combines order domain processing with learnable prior selection and refinement. We first use computed order tracking to map the spectrum to orders and remove the effect of speed. A lightweight weighting module takes bearing fault frequencies and their harmonics as anchors, learns sparse gates to keep the most useful components, and allows small center shifts to tolerate parameter and speed errors. The weighted spectrum is then fed into a simple encoder trained with cross entropy plus two weak constraints that encourage few active harmonics and limited shifts. This design keeps the model interpretable and delivers stable diagnosis across speeds and under strong noise. The effectiveness of this method is validated using a real aero-engine bearing dataset.

Reliability evaluation of high-reliability, long-life products is challenging due to limited test duration, small sample sizes, and the absence of failure data. To address this issue, this paper proposes an improved Bayesian reliability evaluation framework. First, Pseudo-lifetimes are derived through degradation modeling under accelerated stress conditions. To better utilize limited run-to-failure information, a cubic B-spline-based smoothed Bootstrap method is developed to construct continuous empirical distributions and generate informative priors. Then, Bayesian posterior inference for model parameters is via Markov Chain Monte Carlo sampling, and an acceleration model is applied to extrapolate lifetime characteristics to normal operating conditions. Simulation studies show that the proposed method enhances the stability and accuracy of parameter estimation under limited data conditions. Finally, real-world case study demonstrates the practical applicability of the proposed framework, confirming its effectiveness for reliability evaluation.

This paper presents a multivariate degradation modeling framework for operational reliability assessment of high-reliability aerospace systems. The methodology integrates fractional Brownian motion with unit-to-unit variability modeling to simultaneously capture long-term dependency and quantify individual uncertainties in degradation processes. A drawable vine Copula hierarchically fuses marginal remaining useful life distributions into a joint probability framework at the system level. Validated on aerospace electromechanical servomotors, the proposed framework shows that: (1) physically interpretable and physics-guided degradation parameters can provide a consistent statistical description of multi-component degradation behaviors; (2) feature-level conditional failure probabilities offer quantitative information that can be used to inform maintenance decision-making; and (3) the resulting system reliability curves exhibit stage-dependent characteristics that are consistent with different phases of degradation and may support condition-based maintenance planning. The framework aims to facilitate condition-dependent and dynamically updatable operational reliability assessment under practical constraints of data scarcity and environmental variability.

Weakly supervised anomaly detection aims to identify anomalies and localize their temporal boundaries using coarse-grained labels in industrial monitoring scenarios. However, most existing methods simplify this task as binary classification, while real-world industrial systems often involve multiple anomaly types and complex anomaly associations. To address this issue, we propose a weakly supervised multiclass anomaly detection method based on anomaly association. The method first generates pseudoanomaly proposals, including pseudolabels for anomaly categories and temporal boundaries, and then uses them to model intraclass and interclass anomaly associations. Specifically, a state-oriented adaptive segmentation algorithm is introduced to ensure state consistency within each segment. An anomaly attention branch further separates anomalies, anomaly contexts, and normal data to reduce contextual interference. Based on the generated pseudolabels, intraclass and interclass associations are exploited to improve anomaly classification and temporal localization. Experiments on two real-world datasets demonstrate that the proposed method effectively enhances weakly supervised multiclass anomaly detection performance.

Component-stacked rotor systems are coupled through shared load paths and vibration transmission. Degradation in any component can change system-level structural parameters and introduce uncertainty into their evolution, thereby reshaping the measured features and apparent fault patterns. Fault identification should therefore account for degradation effects. In this study, we refer to this objective as degradation-induced fault identification. A key challenge is that failure histories capturing progressive degradation are often scarce. Purely data-driven models trained on such samples may learn feature distributions that do not adequately characterize degradation evolution and its associated fault modes. To address this issue, we develop a mechanism-informed generative modeling framework. In the physical model, structural parameters are modeled as stochastic variables following specified probability distributions, enabling the augmented data to better cover the underlying distribution of degradation-induced fault states. Additionally, we develop an uncertainty-guided attention mechanism that concentrates on long-term dependencies in high-uncertainty feature regions. It quantifies the uncertainty propagation from structural parameters to the learned feature space, and provides interpretable insights into degradation-induced fault manifestations. By integrating distributional parameter modeling with physical knowledge, the framework characterizes degradation variability and evolution within a unified model, demonstrating promising applicability in multi-component systems.

Current data-driven fault diagnosis methods suffer from poor transferability. It is challenging to apply a model effective on one device directly to another. Many methods now employ domain adaptation algorithms to align their fault distributions for model transferability. However, most methods focus only on aligning either marginal or pseudo-labels-based conditional distributions, ignoring cases where both label and conditional distributions change, along with the unreliable nature of pseudo-labels. This oversight can lead to transfer failures. To tackle this, this article introduces an information theory-based joint distribution alignment model. The algorithm starts by maximizing mutual information between predicted categories and input samples for conditional alignment without pseudo-label involvement. Simultaneously, the model introduces virtual adversarial training with a penalty term to improve the robustness of prediction results. When label distribution changes, the model uses entropy values to assign data in categories unique to the target domain to "outliers," thus preventing misalignment of these data. In experiments, this algorithm outperformed other domain adaptation-based methods.

Heterogeneous unmanned systems (HUS) consist of multiple types of unmanned sub-systems, and when one or more sub-systems experience failures, it can severely impact the overall system’s operation. Therefore, establishing effective fault diagnosis(FD) methods is crucial for ensuring the safety and reliability of heterogeneous unmanned systems. This paper proposes a federated fault diagnosis method based on data fusion, which combines visual images and multi-sensor information to enhance the fault identification capability of heterogeneous unmanned systems in complex environments. By using an offline broad reinforcement learning strategy, we propose a Federated Broad Reinforcement Learning fault diagnosis method. It achieves high-precision fault diagnosis under various fault conditions by iteratively reconstructing fused data and knowledge. Finally, the proposed method is validated on a hardware-in-the-loop (HIL) simulator in large-scale heterogeneous unmanned systems. Experimental results show that the proposed method improves fault diagnosis accuracy and enhances the safety and reliability of the system.

In data-driven intelligent fault diagnosis, due to the diversity of actual data distribution, more researches focus on cross-domain fault diagnosis. Current research mainly focuses on single sensor data. However, in complex industrial systems, it is necessary to make comprehensive use of multi-source heterogeneous data to monitor the operation status of the system from multiple perspectives. To address the limitation of existing cross-domain fault diagnosis methods in effectively utilizing the complementary information from multi-source heterogeneous data, this paper proposes a novel multi-source heterogeneous information fusion model, called feature enhancement variant-swin transformer (FEV-Swin), based on the swin-transformer framework for intelligent cross-domain fault diagnosis. The proposed model incorporates a series of components designed to enhance cross-domain fault diagnosis performance. It includes a data augmentation module to increase the diversity of source domain data, a shifted windows mechanism to strengthen horizontal and vertical feature extraction capabilities, and a combination of a pyramid feature fusion module with a domain adaptation module to achieve efficient multi-source heterogeneous data fusion and distribution alignment. Experimental results demonstrate that the application of the FEV-Swin model in rotor systems significantly improves fault diagnosis accuracy and applicability, highlighting its broad potential for deployment in complex industrial scenarios.

Unmanned Aerial Swarms are attracting widespread interest in fields such as disaster response, environmental monitoring, and agriculture. However, there is still a lack of effective recovery strategies and comprehensive performance metrics for UAV swarms facing communication attacks, especially in capturing dynamic recovery. The aim of this study is to recover the split and disconnected UAV swarm under attacks. A dynamic recovery method is proposed under attacks by establishing the relationship between algebraic connectivity and consensus speed. The proposed recovery method enables each UAV to selectively establish communication links with responsive UAVs based on the proposed recovery method to reduce communication cost, rather than linking with all neighbours within communication range. Based on this, a set of performance indexes is introduced, considering factors such as consensus ability, communication efficiency, mission execution, and resource consumption. Furthermore, a resilience metric is proposed to quantitatively assess the efficiency of recovery and consensus transition, providing a comprehensive measure of the ability to reach consensus after attacks. Simulations utilizing the second-order consensus protocol and dynamics validate that the consensus speed of the proposed recovery method is 18.88% faster than random recovery. The proposed resilience metric captures the change in the time from recovery to new consensus state, and the resilience of the proposed recovery method is 66.99% higher than random recovery.

Rotor skidding behavior can exacerbate bearing wear and failure, leading to equipment instability, making research on root-cause tracing and regulation of rotor skidding essential. However, rotor skidding behavior is coupling influenced by multiple factors, posing significant challenges in precisely identifying the root causes and formulating effective regulation measures. To address these challenges, a rotor skidding root-cause tracing and regulation method based on supervised fine-tuning of large language models (LLMs) is proposed in this paper. Firstly, nearly 200,000 skidding text knowledge is constructed based on the established rotor skidding parameter classification rules and skidding-domain-specific prompt. Subsequently, the supervised fine-tuning of LLM is implemented using the constructed skidding knowledge combined with the QLoRA algorithm. Finally, the fine-tuned LLM is employed to achieve root-cause tracing and regulation under the coupling effects of multiple factors for rotor skidding. Specifically, the effectiveness of the proposed supervised fine-tuning strategy is further validated using the designed semantic knowledge similarity metrics.

Manufacturing processes frequently experience fault propagation when an initial fault triggers widespread alarms through informational or physical relationships. Effective fault propagation analysis is, therefore, essential to guarantee production safety by accurately identifying fault propagation pathways, thereby preventing cascading failures in process monitoring systems. However, the majority of existing methodologies in the relevant literature provide unreliable fault propagation monitoring outcomes, primarily because they overlook the accuracy of causality estimators and fail to adequately consider the significance of causal variables. To address these limitations, this article introduces a nonlinear multivariate Lasso Granger (LG) approach specifically designed for fault propagation analysis. To reduce ambiguous causal pathways, a many-to-one neural network architecture, rather than conventional pairwise networks, is employed for extracting LG causality. Furthermore, an input attention mechanism is integrated into the network to adaptively identify and prioritize relevant driving variables. Subsequently, a Monte Carlo-based significance test is devised to determine an appropriate threshold for the LG causality estimator, effectively eliminating spurious causal paths. The effectiveness and practical applicability of the proposed fault propagation analysis approach are validated through numerical simulations and computer numerically controlled machining case studies. Finally, the obtained results demonstrate the practicality and superior performance of the developed methodology.

Test selection represents a crucial step in testability modeling, ensuring diagnostic metrics while reducing testing costs. Nevertheless, in practice, the reliability of each test influences key indicators such as miss alarm rate and false alarm rate. Currently, most test selection methods rarely consider test reliability. This paper proposes a multi-objective optimization test selection method that accounts not only for the optimization of test costs, but also for the optimization of miss alarm rate and false alarm rate under the influence of test reliability. The effectiveness of the method is verified in several testing models.

This paper presents a self-attention-based framework for predicting the remaining useful life of aerospace equipment. This study performs statistical analysis to select sensors and set initial parameters. The proposed framework employs self-attention mechanism for both feature extraction and regularization, with the latter facilitated by physics-informed neural networks. The results demonstrate that the framework provides stable and accurate predictions during the late degradation stage, which is more relevant in real engineering applications. Feature distribution visualization in the latent space reveals a clear degradation path, highlighting the high interpretability. Based on the experimental results, the proposed framework outperforms traditional methods, enabling precise condition-based maintenance.

Escalators, as vital components of modern urban transportation and commercial buildings, necessitate stable operation. However, due to prolonged usage and environmental influences, key components of escalators are prone to failures, leading to operational anomalies. Traditional periodic inspections and manual maintenance methods suffer from low efficiency, high cost, and poor real-time performance. To address these issues, this paper proposes an unsupervised escalator anomaly detection method based on feature optimization and an improved Gaussian Mixture Model (GMM) algorithm.The method initially employs the Pareto principle for feature selection, enhancing computational efficiency and generalization capabilities. Subsequently, to overcome the limitations of traditional GMMs in determining the number of clusters and their application to anomaly detection, a bidirectional Fuzzy C-Means-Expectation Maximization (FCM-EM) algorithm is proposed. This algorithm determines the optimal number of clusters and optimizes the model’s fit to the data, thereby enabling accurate identification of anomalous states in escalators.Experimental results demonstrate that the proposed method achieves high diagnostic accuracy across four escalator fault modes. Furthermore, comparative experiments show that the proposed method outperforms the comparison methods, providing reliable technical support for intelligent maintenance and safe operation of escalators.

UAV swarms are sensitive to cascading failures due to dynamic topology, limited communication resources, and unstable link conditions. These challenges significantly affect the performance of UAV swarms. Hence, a nonlinear load-capacity model is proposed, considering the limits of communication resources. This model helps analyze how load fluctuations influence cascading failures. The dynamic features of UAV communication links are also included. Different model parameters are simulated to see how they affect failure spread and survivability. θ determines the activation of cascading failure, while α influences the underlying load dynamics.

Cascading failures represent a significant issue in production lines, as they can lead to process defects and safety incidents. An accurate risk assessment of cascading failures is crucial for ensuring both safety and operational efficiency. However, existing methods for assessing cascading failures typically focus only on exposed failures, neglecting hidden failures. Hidden failures are functional faults not apparent under normal operating conditions; they often remain undetected until triggered by another failure event. Considering solely exposed failures thus provides an incomplete picture, insufficient for accurately assessing cascading failure risks. To address this limitation, this article proposes a novel virtual node-based framework designed to assess cascading failure risks explicitly accounting for hidden failures. A Bayesian network approach, enhanced by leveraging connectivity information, is employed to effectively model the structure of the production line. Within this Bayesian network, a virtual node is integrated, thus representing the background impact of hidden failures. Specifically, the interactions between this virtual node and other network nodes explicitly capture the dynamics and mechanisms underlying hidden failures. Building upon this framework, we propose the virtual node-assisted inverse PageRank algorithm. The algorithm is rigorously defined, with mathematically guaranteed properties including positivity, convergence, and an analytical solution. The methodology is validated using a real-world case study involving an aerospace impeller production line. Experimental results demonstrate that the proposed algorithm successfully identifies hidden failures, delivering superior performance compared to traditional risk assessment approaches.

A novel adaptive recovery method is proposed for DAV swarms considering the consensus speed with the dynamics subjected to attacks. Depending on the objective of consensus speed after recovery, an optimization problem is proposed and solved through maximization the algebraic connectivity integrating local and global information with attacks. The UAV swarm is muddled by a second-order system and dynamics of each DAV are considered. The efficiency of the proposed recovery method is demonstrated with faster consensus speed compared with a random recovery method.