Domain Generalization (DG) has been explored to achieve machine fault diagnosis under previously unseen operating conditions. However, most DG methods assume access to training data collected across multiple conditions, an assumption that rarely holds in industrial practice, where fault data are typically available from only a single operating condition. To address this critical constraint, we propose an attention-throughout latent diffusion model for single-source domain generalization (ATLD-SSDG). The proposed framework learns discriminative fault representations from a single-condition source domain and generalizes robustly to multiple unseen target conditions. First, to effectively capture complementary fault information, vibration signals from three views are fused and projected into a latent space via a collaborative attention fusion mechanism. Next, a dedicated one-dimensional (1D) U-Net is constructed to address information loss in existing approaches and facilitate more effective conditional diffusion. Unlike existing methods that directly adopt computer vision diffusion architectures, the proposed 1D U-Net is specifically designed for vibration signals, preserving localized fault-related details and preventing information loss caused by time–frequency transformations. Moreover, by explicitly regulating self-attention and cross-attention within the diffusion model, the framework preserves fault-relevant characteristics while selectively substituting operating-condition-related factors, thereby enabling controllable and effective domain generalization. Extensive experiments demonstrate superior generalization performance and diagnostic accuracy of the proposed method over state-of-the-art DG methods. These results indicate that latent diffusion, when properly structured for 1D condition-monitoring signals, provides an effective mechanism for single-source domain generalization, helping to close an important gap in DG research for predictive maintenance.

The core objective of graph neural network (GNN)-based remaining useful life (RUL) prediction methods for equipment with multi-source sensors is to learn effective graph representations, and graph pooling is an efficient approach to achieve it. However, existing graph pooling techniques are limited in modeling hierarchical structures and have limitations in embedding space representation. To overcome these limitations, a dynamic curvature pooling graph convolutional network (DCPGCN) is proposed for RUL prediction of equipment with multi-source sensors. DCPGCN develops a hyperbolic hierarchical graph pooling framework. By leveraging the geometric advantages of hyperbolic space for hierarchical representation, the proposed framework more effectively captures multi-level structural information in graphs, significantly improving the overall structural fidelity of the graph representation. Moreover, a curvature predictor driven by pooling path deviation is proposed. By quantifying the geometric distortion along leaf-to-root paths in hyperbolic space, the predictor dynamically adjusts the curvature parameter, improving the embedding space’s adaptability and expressiveness for the graph’s hierarchical structure. Finally, experiments on the CMAPSS dataset demonstrate that the proposed method outperforms multiple state-of-the-art approaches in prediction accuracy, while experiments on real-world wind turbine RUL prediction further confirm its superiority and potential in engineering applications.

Fault diagnosis in distributed device clusters (DDCs) is essential for ensuring safe and reliable operation of modern automated systems. However, practical deployments face two fundamental challenges: severe class imbalance from scarce device-level fault data and strict resource constraints on edge nodes. This paper proposes a Lightweight Diffusion-Enhanced Multi-Domain Fusion (LDEMF) framework that addresses both challenges through integrated data balancing, representation learning, and efficient deployment. A dual-attention conditional diffusion model synthesizes class-balanced vibration samples preserving temporal, spectral, and non-stationary fault characteristics. A multi-domain feature fusion network then integrates time-, frequency-, and time-frequency-domain representations via cross-attention for robust fault characterization. Finally, structured channel pruning, knowledge distillation, and FP16 post-training quantization enable resource-efficient edge deployment. Experimental results on the CWRU bearing dataset and a self-collected industrial robot dataset demonstrate that LDEMF achieves superior diagnostic accuracy under moderate-to-severe class imbalance. The framework reduces model size, computational cost, and inference latency by up to an order of magnitude with negligible performance loss, validating its effectiveness for edge-level fault diagnosis in DDCs.

Zero-shot fault diagnosis (ZSFD) faces significant challenges in aligning time-series signal features and contextual semantic information. Direct projection from feature space to semantic space may suffer from domain bias, while mutual projection approaches require complex tradeoffs among multiple objective functions. This article proposes a multiattribute cross-modal semantic alignment network (MCSANet) for ZSFD. An enhanced feature extractor incorporating a conditional fault severity encoding mechanism is employed to extract discriminative fault features across multiple attributes. The time-series features, and contextual semantic information are then aligned using a novel cross-modal embedding approach, eliminating the need for complex tradeoffs among multiple objective functions. The proposed method was validated on both self-designed and open-source bearing experiments. Experimental results demonstrate that MCSANet achieves robust diagnosis performance even under nonstationary operational conditions and limited distributional diversity in the training phase. Comparative experiments confirm that MCSANet outperforms current state-of-the-art approaches.

Spallation neutron sources are among the rarest and most advanced research infrastructures in the world, with fewer than five large-scale facilities in operation globally. Neutron choppers, as mission-critical components within such systems, must operate continuously under extreme conditions—including strong radiation, low vacuum, and high rotational inertia. These constraints make conventional fault diagnosis approaches ineffective, as sensors cannot be installed near the fault-prone areas (e.g., bearing housings), but instead must be placed remotely due to radiation shielding. This leads to long signal transmission paths, structural discontinuities, and severely degraded signal-to-noise ratios (SNRs), posing substantial challenges for fault diagnosis and predictive maintenance. To address this unique and high-stakes problem, we propose LPWOC (Learnable Parallel Wavelets with Orthogonality Constraints), a noise-robust deep learning model that learns adaptive wavelet filter banks and thresholding functions directly from vibration data. By incorporating conjugate quadrature filters with orthogonality regularization and fully learnable denoising layers, LPWOC offers enhanced feature diversity, low computational complexity, and exceptional resilience to noise. Experiments on a dedicated neutron chopper testbed—featuring realistic sensor placement and seven bearing health statuses—demonstrate 99.21% accuracy under low-SNR conditions, outperforming five state-of-the-art methods. This work provides a scalable and deployable diagnostic solution for one of the most demanding industrial environments in existence.

Anomaly detection task is an open-set challenge, aiming to identify unseen faulty signals using only healthy signals for training. While data reconstruction frameworks are inherently suited for this task, they often struggle with complex signals due to limited feature extraction capabilities. Contrastive learning offers powerful representation learning, but faces challenges in one-class scenarios and requires effective augmentation techniques. To address these limitations, a novel fault detector is proposed, integrating a redesigned one-class contrastive loss into a data reconstruction framework to endow clustering capability. Learnable feature augmentation is incorporated to ensure effective and generalizable contrastive learning while reducing computational costs by performing augmentation in the feature space. A dilated inception network structure is employed to capture long-distance dependencies in complex input signals. A one-class similarity distance-based threshold is introduced to filter outliers in the healthy signal distribution, and an optimal model selection strategy is proposed based on the minimal threshold during training. The approach is evaluated using two case studies: single-fault and multifault scenarios in a reciprocating compressor. Our method achieves balanced accuracies of 98.44% and 97.81%, respectively, outperforming other methods. These results confirm our detector's capacity to balance false alarms and missed detections effectively, even in challenging multifault conditions.

The burgeoning field of the Industrial Internet of Things (IIoT) necessitates advanced fault diagnosis methods capable of navigating the dual challenges of high predictive accuracy and the constraints of edge computing environments. Our study introduces Wave-ConvNeXt, a novel fault diagnosis model that seamlessly integrates the state-of-the-art ConvNeXt architecture with Wavelet Transform. This innovative model stands out for its lightweight design yet delivers exceptional accuracy in fault diagnosis. In Wave-ConvNeXt, we re-engineer the ConvNeXt model for IIoT applications by adopting onedimensional convolution, tailored for processing high-frequency, non-periodic inputs. This adaptation is complemented by replacing the traditional “patchify” layer with a Wavelet transform layer, which simplifies input signals into sub-signals, thereby easing learning complexities and diminishing the dependence on elaborate deep architectures. Further enhancing this model, we incorporate a squeeze-and-excitation module, enriching its ability to prioritize channel-wise feature relevance, akin to self-attention mechanisms. This integration is rigorously validated through an ablation study. Wave-ConvNeXt epitomizes a holistic approach, enabling an end-to-end optimization of feature learning and fault classification. Our empirical analysis on two real-world IIoT datasets demonstrates Wave-ConvNeXt’s superiority over existing models. It not only elevates prediction accuracy but also significantly curtails computational complexity. Additionally, our exploration into the impact of various mother wavelets reveals the effectiveness of using wavelet basis functions with smaller support, bolstering diagnostic precision. The source code of Wave-ConvNeXt is available at https://github.com/leviszhang/waveConvNeXt.

The service performance of an industrial bearing may deteriorate sharply after a sudden transition which is very difficult to be captured prior to its appearance. It is accordingly valuable to estimate this critical transition for avoiding serious catastrophes caused by the faulty bearing. To address this issue, an asynchronous gated recurrent network (AGRN) is proposed to estimate the critical transition of the bearing deterioration. A gated recurrent unit (GRU) is first developed to extract prognostic data as features using few early samples. Chi-square distributions of the squared prediction error of the residuals and the variation of principal components calculated from the features are fused as a statistic series, where the critical threshold is calculated adaptively. Another GRU is proposed to forecast deterioration process using fused statistic series. The forthcoming transition can be estimated by the process with the critical threshold at the initial operation. The present AGRN is evaluated by lifecycle experiments of three bearings, all from public benchmark datasets. Results show that the proposed method is robust in offering maintenance response time for bearings before transiting to critical failures.

In real-world industrial scenarios, fault diagnosis often relies on a significant volume of normal data to detect faults with a few coming samples. The limited sample nature of newly introduced faults can pose challenges for deep learning-based fault diagnosis models. To address this issue, a novel framework named incrementally generative adversarial diagnostics was developed by using few-shot enabled one-class learning (FSEOCL) for effective anomaly detection and fault classification of the new-coming few samples. In the addressed method, a bi-directional generative adversarial network is first trained using only normal data to acquire an encoder for latent representation from time-series data. A few samples of each fault condition in latent space can be then identified and classified correctly through FSEOCL. The effectiveness of the proposed framework is demonstrated through fault diagnosis experiments conducted on both a benchmark bearing and an industrial robot. The results underscore the adaptability of this framework to industrial fault diagnosis scenarios, highlighting its ability to achieve accurate anomaly detection and fault classification compared to state-of-the-art peer methods.

Anomaly detection of rolling element bearing aims at detecting bearing defects based on its monitoring data of normal condition, to prevent unexpected shut-downs of rotating machines. Commonly adopted methods are based on hand-crafted features or data reconstruction, which are indirect to the anomaly detection task. In this work, a bearing anomaly detection method is proposed based on contrastive learning. To extract essential representations of normal condition, the time-domain vibration signal and its frequency spectra are compared and forced to provide common information. Hence the contrastively learned essential representation does not change regarding data transformation and is more faithful to the normal condition. And a straightforward distance-based detector is enough for the anomaly detection task. The proposed method is applied to benchmark bearing experimental data. Results show that the proposed method outperforms other commonly used anomaly detection methods.

Bearing condition is a non-negligible part of mechanical equipment health monitoring. Most of the existing bearing fault diagnosis methods are based on the premise that all data classes are known and lack the capability of incremental diagnosis of fault modes. However, in engineering practice, the initial monitoring data only provide normal condition, and the subsequent data of different classes of faults are collected gradually. To address this practical problem, we propose incremental contrastive learning (CL) of homologous and interclass features for bearing to achieve incremental diagnosis of bearing fault modes from single to multiple classes. Important homologous and interclass features of bearings are first extracted by CL. The obtained features are then employed to establish a distance threshold for the anomaly diagnosis of subsequent samples. Upon appearing anomalies incrementally up to a certain amount, novel classes are upgraded and fed back to the model. In this way, new class data are treated as incremental learning resources. The proposed method was evaluated using both benchmark bearing and gearbox bearing experiments. Results show supreme diagnostic performance compared to peer state-of-the-art approaches. The present method is intrinsic in extracting homologous and interclass features for practical bearing fault diagnostics.

An echo state network (ESN) is a recurrent neural network with low computational complexity. However, a single ESN cannot extract effective features from complex inputs, especially for dealing with low-cost condition signals in machinery fault diagnosis. A novel deep learning model, referred to as the deep fuzzy ESN (DFESN), was proposed to improve the feature extraction capability with less computational burden. In the present method, the output data of the previous ESN reservoir were regarded as abstract feature vectors for the next ESN input. The features were reinforced in each hidden layer by using fuzzy clustering as a tuning step for classification enhancement. In this way, layerwise fuzzy tuning was developed to replace traditional overall feedback fine tuning in deep models. This improved learning efficiency and robustness while overcoming the vanishing gradient problem for deep learning. The superiority of the proposed approach was evaluated by both theoretical analysis and experimental tests. The results showed that the present DFESN features improved classification accuracy and reduced the computational burden. In addition to machinery fault diagnosis, the proposed DFESN also has potential for other deep learning applications.

Echo state network (ESN) is a fast recurrent neural network with remarkable generalization performance for intelligent diagnosis of machinery faults. When dealing with high-dimensional signals mixed with much noise, however, the performance of a deep ESN is still highly affected by the random selection of input weights and reservoir weights, resulting in the optimal design of the deep ESN architecture, which is an open issue. For this reason, a hybrid evolutionary algorithm featuring a competitive swarm optimizer combined with a local search is proposed in this article. An indirect encoding method is designed based on the network characteristics of ESN to make the evolutionary process computationally economical. A layerwise optimization strategy is subsequently introduced for evolving deep ESNs. The results of two experimental cases show that the proposed approach has promising performance in identifying different faults reliably and accurately by comparing with other intelligent fault diagnosis approaches.

An echo state network (ESN) is a type of recurrent neural network that is good at processing time-series data with dynamic behavior. However, the use of ESNs to enhance fault-classification accuracy continues to be challenging when the condition signals are collected by low-cost sensors. In this paper, a deep network algorithm, called a deep hybrid state network (DHSN), is proposed for fault diagnosis of three-dimensional printers using attitude data with low measurement precision. In the DHSN, the output data of a sparse auto-encoder are regarded as the abstract features of a double-structured ESN (DESN). The DESN is designed for feature reinforcement and fault recognition, wherein the first function reinforces the features and the second is used for fault classification. More specifically, feature reinforcement is developed to improve the clustering performance and replace the traditional overall feedback fine-tuning in deep models. This strategy improves learning efficiency and overcomes the vanishing-gradient problem for deep learning. The forecasting performance of the proposed approach is evaluated in experiments, and its superiority is demonstrated through comparison with other intelligent fault-diagnosis technologies.

Bearings are fundamental mechanical components in rotary machines (engines, gearboxes, generators, radars, turbines, etc.) that have been identified as one of the primary causes of failure in these machines. This makes bearing fault diagnosis (detection, classification, and prognosis) an economic very relevant topic, as well as a technically challenging one as evaluated by the extensive research literature on the subject. This paper employs a systematic methodology to identify, summarize, analyze, and interpret the primary literature on fuzzy formalisms for bearing fault diagnosis from 2000 to 2017 (March). The main contribution is an updated, unbiased, and (to a higher extend) repeatable search, review, and analysis (summary, classification, and critique) of the available approaches resorting to fuzzy formalisms in this trendy topic. A discussion on a new promising future research direction is provided. A comprehensive list of references is also included.

Continuous function optimization is ubiquitous in many branches of Science and Technology. Memetic algorithms are a particularly interesting approach to the optimization of continuous, non-linear, multimodal, ill-conditioned or noisy functions as these algorithms do not require derivatives and bala nce global exploratory search with local refinement. The Wang genetic algorithm promotes genetic diversity (exploratory capacities) by applying crossover only to parents with sufficient different chromosomes (genomes). In this work an improvement of the Wang algorithm is proposed that allows for an adaptive evaluation of the genomic difference between individuals in a way that is independent of the optimization problem and takes into account the stage of the evolutionary process. Moreover, the work proposes an original and relevant memetic algorithm combining the improved Wang genetic algorithm, for exploration purposes, with the covariance matrix adaptation evolutionary strategy (CMA-ES) for refinements. The proposed algorithm is empirically evaluated using 25 bench marking functions against five state-of-the-art memetic algorithms revealing superior performance which is a strong evidence on the relevance of proposed algorithm.

Nowadays, the available data to describe real world problems grows in considerable manner, due to the amount of measurable characteristics (features) that can be collected. Machine learning techniques are widely used to extract valuable knowledge from data, but their performance might decrease when the proper features are not selected. Feature selection is introduced to search relations to disclose possible redundant or irrelevant features in a case study; this search is performed either in a supervised or unsupervised manner. In the present work, we propose an unsupervised feature selection algorithm using: (1) relative dependency to search similarities between features, (2) a clustering algorithm to group similar features, and (3) a procedure to select the most representative feature to obtain a reduced feature space. The relative dependency degree between pairs of attributes is used to compute a similarity measure. This measure is used by a clustering algorithm to perform attribute clustering through KNN and prototype based clustering. The proposal is tested with well-known benchmarks, and compared with classic supervised and unsupervised feature selection techniques. Additionally, a real world application in fault diagnosis for rotating machinery is evaluated by our proposal.

UNLABELLED: A new edition of the DelPhi web server, DelPhi web server v2, is released to include atomic presentation of geometrical figures. These geometrical objects can be used to model nano-size objects together with real biological macromolecules. The position and size of the object can be manipulated by the user in real time until desired results are achieved. The server fixes structural defects, adds hydrogen atoms and calculates electrostatic energies and the corresponding electrostatic potential and ionic distributions. AVAILABILITY AND IMPLEMENTATION: The web server follows a client-server architecture built on PHP and HTML and utilizes DelPhi software. The computation is carried out on supercomputer cluster and results are given back to the user via http protocol, including the ability to visualize the structure and corresponding electrostatic potential via Jmol implementation. The DelPhi web server is available from http://compbio.clemson.edu/delphi_webserver.