Bearing skidding is the primary factor restricting the development of aeroengines toward ultrahigh speed, low friction, and lightweight. Compared to typical bearing faults, analysis of bearing skidding presents greater challenges due to the weak signal properties, significant time-varying characteristics and coupling influence of multiple factors. It is crucial to fully utilize multisource signals to enhance skidding features and capture time-varying characteristics. This article proposes a prior knowledge-embedded dual feedback spatial-temporal graph convolutional network (DFSTGCN) for skidding assessment. Unlike existing adjacency matrix construction strategies, the correlation between multisource signals is described based on multiple prior knowledge, which includes dynamic model, structural dynamics, and expert experience. Furthermore, a DFSTGCN is designed to simultaneously focus on the spatial and temporal dependencies of time-varying skidding data. Specifically, a dual feedback mechanism that includes prediction error ratio and uncertainty loss function is employed to improve the generalization performance of skidding prediction model. The effectiveness of the proposed strategy is validated under different working conditions.

The multifactor coupling influence on the skidding behavior of aeroengine rotors presents significant challenges in locating the skidding causes and developing effective skidding suppression measures. However, ongoing research into fault mechanism and knowledge graph (KG) facilitates the accurate tracing of complex faults. We propose a heterogeneous KG inference-assisted skidding tracing and regulation strategy for aeroengine rotor. First, a skidding heterogeneous KG is constructed based on the text and data knowledge, in which the skidding level classification rules are determined for the first time. Second, we design an adaptive distributed metalearning algorithm to extract data features by combining the structural characteristics of the skidding KG. Third, few-shot knowledge inference is performed using the relation-metalearning graph convolutional network. Finally, we develop skidding suppression measures by tracing the input knowledge under unknown working states, enabling effective regulation of skidding behavior.

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.

In recent years, intelligent manufacturing has integrated industrial data and artificial intelligence technology, which has been a widely concerned development direction in the manufacturing industry. Industrial data integrity is the key factor for the successful implementation of intelligent manufacturing. However, in flexible manufacturing systems with multivariety and small-batch, it is hard to collect production data from all working conditions. Actually, for the purpose of status monitoring, data acquisition and annotation on complex mechanical components is also time-consuming and labor-intensive, which requires the assistance of professional domain knowledge. Faced with the challenge of incomplete data quantity and quality, federated learning is a promising paradigm of collaborative modeling, which ensures data privacy and fully utilizes distributed data information from different industrial users. However, due to the heterogeneity of data among industrial users, cooperation benefits cannot satisfy all industrial users. In this article, a novel federated learning cooperation framework is proposed to guide participants to choose the appropriate coalition and improve the benefits of participants. In this framework, the self-organizing incremental neural network is employed to generate prototypes that can effectively capture the distributional characteristics of raw data, obviating the necessity for industrial users to provide their raw data and labels. It offers recommendations for industrial users to foster collaboration by assessing the similarity among these prototypes. The collaborative tool wear prediction experiments demonstrate the effectiveness of the framework on industrial data.

The process data is strongly nonlinear due to variations in system performance caused by machine operation or condition; many nonlinear methods for operating performance assessment, which focus on capturing information about the global of each performance grade while ignoring local information, have difficulty dealing with this strong nonlinearity. Aiming at the problems of strong nonlinearity and incomplete information extraction, an operating performance assessment method based on kernel locally linear embedding partial least squares (KLLEPLS) is proposed. Firstly, the local and global information of the dataset is mined using kernel locally linear embedding (KLLE) and partial least squares (PLS). Secondly, to simplify the computation, a new method, embedding KLLE into PLS, is introduced to make the solution similar to KPLS. Then, an assessment index is introduced into online assessment; when the performance is non-optimal, the contribution rates of the variables are calculated based on the extracted local and global information to determine the possible cause variable. Finally, the practice and effectiveness of the proposed algorithm are verified by the dense medium coal preparation and the Tennessee Eastman (TE) process. Note to Practitioners—Due to the presence of noise and interference, operating performance of the process is non-optimal and the control effect is difficult to meet the actual production requirements. The motivation of this paper is to find the non-optimal operation points in time, identify the reasons, feedback to the operator, and make adjustments to maximize the comprehensive economic benefits of the enterprise. The process data is strongly nonlinear due to variations in system performance caused by machine operation or condition, and some nonlinear methods are not complete to extract feature information for this strong nonlinearity. Therefore, this paper proposes the kernel local linear embedding partial least squares (KLLEPLS) algorithm, and uses it to build a model to evaluate whether the current operation is in an optimal performance; when it is in the non-optimal performance, possible variable is identified and the process is adjusted in time to ensure that it operates under optimal conditions. This method extracts more comprehensive information compared with the traditional methods, which can provide more accurate adjustment basis for operators.

Few-shot fault diagnosis is a challenging problem for complex engineering systems due to the shortage of enough annotated failure samples. This problem is increased by varying working conditions that are commonly encountered in real-world systems. Meta-learning is a promising strategy to solve this point, open issues remain unresolved in practical applications, such as domain adaptation, domain generalization, etc. This article attempts to improve domain adaptation and generalization by focusing on the distribution-shift robustness of meta-learning from the task generation perspective. In fact, few-shot fault diagnosis under varying working conditions allows to address the distribution shift problem in a natural way. An unsupervised across-tasks meta-learning strategy with distributional similarity preference is proposed, where the core is the distribution-distance-weighting mechanism. Differently from the naive random meta-train task generation strategy used in existing meta-learning methods, the source instances that present a more similar distribution with respect to the target instances gain larger weightings in the task generation. This strategy leads to a meta-task training set that is enough diverse, and at the same time can be easily learned due to the distribution similarity features of the source tasks. The proposed method introduces the concept of maximum mean discrepancy that is applied to derive the distribution distance of the measurements. Moreover, a model-agnostic meta-learning is applied to realize few-shot fault diagnosis under varying working conditions. The proposed solutions are verified and compared by considering two public datasets used for bearing fault diagnosis. The results show that the proposed strategy outperforms different related few-shot fault diagnosis methods under varying working conditions. Moreover, it is thus proved that, meta-learning with distribution similarity feature represents an effective approach for domain adaptation and generalization.

Heterogeneous Multi-Agents System (MAS) has been attracting increasing attention in many application areas, but the safety and reliability of MAS are still challenging issues. Fault diagnosis is a necessary technology to ensure the safety and reliability of heterogeneous MAS. According to the characteristics of high dispersion in MAS, strong local perception ability and weak global perception ability, this paper proposes a distributed hybrid knowledge-based and data-driven fault diagnosis, which realizes dynamic re-construction of data and knowledge through reinforcement learning and fuzzy broad learning. In the meantime, we also consider communication network topology to realize distributed collaborative diagnosis, which can effectively improve the diagnostic performance. Then, we develop a high-fidelity heterogeneous MAS software-in-the-loop and hardware-in-the-loop fault simulators to simulate different types of failures (i.e., actuator failure, sensor and communication failure). Finally, through the cross-validation on the above developed simulators, this work verifies the effectiveness of the proposed distributed intelligent fault diagnosis.

Recent advances in the wind energy industry have stimulated the demand for automated condition monitoring mechanisms capable of mitigating the cost of operations and avoiding tremendous economic losses due to unplanned downtime. To this end, a wide range of normal behavior models have been developed to monitor wind turbine performance. However, since most models are tailored to a single target at a time, a separate model is required for each target and are thus deemed unwieldy and expensive to implement, particularly in large-scale wind farms. Therefore, this article advocates for a multitarget normal behavior model which is capable of monitoring multiple targets simultaneously. The proposed model is specifically based on heterogeneous stacked regressions, trained with normal data curated via kernel density estimation. The distinct targets are monitored through a control chart based on an exponentially weighted moving average chart and a change-point detection (CPD) method via binary segmentation for wind turbine suboptimal performance detection. Extensive experiments based on real-world wind farm data are carried out and the results are compared with state-of-the-art methods. The attained results indicate that the proposed model is highly effective in not only reducing the number of models required for monitoring wind turbines, but also in improving model accuracy significantly.

Accelerated degradation testing (ADT) data typically exhibit a time-stress-dependent structure, as well as random uncertainties due to time-varying effects and unit-to-unit variations. Existing ADT models based on Brownian motion with drift have successfully represented the fault/failure-based degradation behavior and random uncertainty by assuming that the drift parameter follows a Gaussian distribution. However, these models often lack robustness to outliers, leading to distorted analysis, affecting parameter estimation, model accuracy, decision-making, risk assessment, and potentially overlooking the influence of stress factors. A novel robust ADT model based on the Wiener process and its corresponding lifetime analysis method are proposed to address these issues. The proposed approach improves upon traditional ADT models by making the drift parameter follow a$t$-distribution rather than a Gaussian distribution, which can reduce sensitivity to outliers in real degradation processes. In addition, the proposed method allows for the simultaneous consideration of time-stress-dependent factors in the ADT model, facilitating the derivation of a closed-form robust ADT formulation. Subsequently, the lifetime is analyzed based on the ADT model using the first hitting time method in a probabilistic framework. The proposed method is applied to stress relaxation data of electrical connectors and compared to three other common methods.

The demand for bearing skidding diagnosis is widely present in aeroengines operating at high-speed and light-load conditions. However, the weak and time-varying characteristics of skidding signal raise challenges for accurate diagnosis. To address these issues, we propose a synergistic TransGCN strategy to extract rich feature information from time-varying weak bearing skidding signals. Unlike existing methods, the prior knowledge obtained from bearing skidding analysis and the alternate integration and synergistic optimization of various advantages are used to enhance algorithm performance. First, an adaptive chirplet transform is designed to measure the time-varying cage slip rate. Second, the skidding sensitive characteristics are determined, and the variation ranges of slip rate sensitivity are employed as prior knowledge to calculate the fusion weights of multisource information. Then, an unsupervised deep feature representation network is constructed to analyze the complex correlation of bearing skidding signals. Finally, a synergistic TransGCN is developed by alternately integrating and synergistic optimizing Bayesformer and graph convolutional network. The superiority of the proposed strategy has been verified.

The absence of fault data in certain categories presents a significant challenge in data-driven fault diagnosis, as obtaining a complete fault dataset is often unfeasible. Zero-shot learning has emerged as a viable solution to this problem. Nonetheless, it often encounters problem of unreliable diagnosis results due to domain shift. In this article, a feature generating network with attribute-consistency is developed for zero-shot fault diagnosis, which introduces the attribute consistency constraint and feature transformation with attribute information. The implementation process comprises two parts, unseen fault class generation and discriminative feature transformation. The attribute consistency constraint adopted in data generation can make the generated data represent their attribute well. For feature transformation, a concatenation operation is used to transforming the generated samples into more discriminative representations. The effectiveness of the proposed method is verified using a public dataset for fault diagnosis purpose. Results indicate that the proposed method outperforms the state-of-art zero-shot diagnosis method.

The data generated by modern industrial processes often exhibit high-dimensional, nonlinear, timing, and multiscale characteristics. Presently, most of the fault diagnosis methods based on deep learning only consider the part of the characteristics of industrial data, which will cause the loss of part of the feature information during training, thereby affecting the final diagnosis effect. In order to solve the above problems, this article proposes an end-to-end multiscale feature learning method based on model fusion, which can simultaneously extract multiscale spatial features and temporal features of data, effectively reducing the loss of feature information. First, this article combines the convolutional neural network (CNN) with residual learning and designs a multiscale residual network (MRCNN) to extract high-dimensional nonlinear spatial features of different scales in the data. Then, the extracted features are input into the long and short-term memory (LSTM) network to further extract the temporal features of the data. After the fully connected layer, it is input into the classifier for final fault classification. The residual learning in MRCNN can effectively avoid the problem of model degradation and improve the training efficiency of the model. Through the fusion of MRCNN and LSTM, we can significantly improve the feature extraction ability of the model, thereby greatly improving the diagnosis effect. In the final case experiment, the method improved the comprehensive diagnostic accuracy of the Tennessee-Eastman (TE) process and industrial coking furnace datasets to 94.43% and 97.80%, respectively, which was significantly better than the existing deep learning model and proves the effectiveness and superiority of this method.

With the rise of wind energy production in global power generation, wind farm facilities are becoming increasingly attractive targets for malicious attacks, in particular those affecting wind farm network operators’ cybersubsystems and functionalities. Given the significance of this problem, this article proposes a novel anomaly-based intrusion detection and diagnosis system to carry out in-line monitoring as with firewalls. Also, an innovative cyberattack-resilient active power control is designed to responsively mitigate the impacts of cyberattacks on the safe regulation of active power from wind farms. An offshore wind farm benchmark is used to implement and demonstrate the effectiveness of the proposed solutions in the presence of wind turbulences, measurement noises and realistic smart cyberattack scenarios.

Data-driven fault detection and isolation (FDI) depends on complete, comprehensive, and accurate fault information. Optimal test selection can substantially improve information achievement for FDI and reduce the detecting cost and the maintenance cost of the engineering systems. Considerable efforts have been worked to model the test selection problem (TSP), but few of them considered the impact of the measurement uncertainty and the fault occurrence. In this article, a conditional joint distribution (CJD)-based test selection method is proposed to construct an accurate TSP model. In addition, we propose a deep copula function which can describe the dependency among the tests. Afterward, an improved discrete binary particle swarm optimization (IBPSO) algorithm is proposed to deal with TSP. Then, application to an electrical circuit is used to illustrate the efficiency of the proposed method over two available methods: 1) joint distribution-based IBPSO and 2) Bernoulli distribution-based IBPSO.

Remaining useful life (RUL) prediction is a reliable tool for the health management of components. The main concern of RUL prediction is how to accurately predict the RUL under uncertainties. In order to enhance the prediction accuracy under uncertain conditions, the relevance vector machine (RVM) is extended into the probability manifold to compensate for the weakness caused by evidence approximation of the RVM. First, tendency features are selected based on the batch samples. Then, a dynamic multistep regression model is built for well describing the influence of uncertainties. Furthermore, the degradation tendency is estimated to monitor degradation status continuously. As poorly estimated hyperparameters of RVM may result in low prediction accuracy, the established RVM model is extended to the probabilistic manifold for estimating the degradation tendency exactly. The RUL is then prognosticated by the first hitting time (FHT) method based on the estimated degradation tendency. The proposed schemes are illustrated by a case study, which investigated the capacitors' performance degradation in traction systems of high-speed trains.

Degradation prognostics of aero-engine are a well-recognized challenging issue. Data-driven prognostic techniques have been receiving attention because they rely on neither expert knowledge nor mathematic model of the system. But they are highly dependent on the quantity and quality of degradation data. To solve the problems caused by unlabeled, unbalanced condition monitoring (CM) data and uncertainties of the prognostics process, a novel data-driven aero-engine degradation prognostic strategy is proposed in this article. First, two indicators are defined to remove redundant degradation features. Then, the number of discrete states of health is determined by a fuzzy c -means algorithm, and the health state labels can be automatically assigned for health state estimation, where the uncertain initial condition and the uncertainty of health state's transition are fully considered. Finally, a multivariate health estimation model and a multivariate multistep-ahead long-term degradation prediction model are proposed for remaining useful life estimation for aero-engines. Verification results using the aero-engine data from NASA can show that the proposed data-driven degradation prognostic strategy is effective and feasible.

An effective strategy for analyzing and diagnosing multiple faults is developed, based on the concise causality structure obtained by multisignal modeling and the fault source signals extracted by blind source separation (BSS). The key idea is enlightened by the need to handle the redundant test signals and the multiple fault ambiguity groups when applying multisignal modeling for multiple fault diagnosis. Considering that BSS is inherently suitable to extract the independent source information, it is integrated into the multisignal model to reconstruct the causality structure that will have superior diagnosticability. Preliminary study on test point design is also presented in the proposed strategy. The proposed multiple fault diagnosis strategy has been verified on a hydraulic automatic gauge control simulation system in a cold rolling mill. Results show that it can use less test information to effectively diagnose all simulated single and multiple faults.

Incipient faults in high-speed trains are usually masked by noises and disturbances from process and sensors, which severely increases the difficulty of incipient fault detection and diagnosis. By introducing Hellinger distance into multivariate statistical analysis framework, this paper develops a robust detection and diagnosis method for incipient faults under the principal component analysis. The proposed method can detect all incipient sensor faults in traction systems of high-speed trains in real time by comparing reference probability density functions (PDFs) with the online estimated PDFs. According to the fault detection information, an accurate fault diagnosis can be achieved online through Bayesian inference. Key advantages of the proposed method are its salient robustness to unknown noises and disturbances, as well as the high sensitivity to incipient faults. In addition, the proposed method does not require any information on system models of high-speed trains or any human intervention. The effectiveness of the proposed method has been firstly proven by mathematical derivations and then been verified by numerical simulations. Finally, the proposed method has been applied to the practical experiment platform of the high-speed trains.

In this paper, a novel descriptor estimator-based incipient fault estimation scheme is designed for Lipschitz nonlinear descriptor systems with process disturbances and measurement output noises. By using the proposed estimator, incipient sensor faults, abrupt actuator faults, and measurement noises can be estimated asymptotically. Application results conducted on a three-phase inverter system of China railway high-speed trains are given to illustrate the effectiveness of the developed approach. The main contributions are summarized in two aspects: 1) the unified framework designed for actuator and sensor fault reconstruction/estimation problem is capable of dealing with actuator and sensor faults at the same time and 2) a linear matrix inequality optimization method is formulated to achieve optimal performance of signal estimation.

A fault detection and diagnosis (FDD) method is proposed by combining transfer entropy (TE) and signed digraph (SDG). Given process historical data, transfer entropy is used to construct a SDG model to represent causal relationship between process variables. A fault severity evaluation method is then proposed based on the modified SDG model, where the nodes can take values of (0), (±1), (±3) and (±6). Then, an index named DoF is developed to measure fault severity. The application results can verify the effectiveness and feasibility of the proposed method.