Intelligent fault diagnosis (IFD) has emerged as a powerful paradigm for ensuring the safety and reliability of industrial machinery. However, traditional IFD methods rely heavily on abundant labeled data for training, which is often difficult to obtain in practical industrial environments. Constructing a digital twin (DT) of the physical asset to obtain simulation data has therefore become a promising alternative. Nevertheless, existing DT-assisted diagnosis methods mainly transfer diagnostic knowledge through domain adaptation techniques, which still require a considerable amount of unlabeled data from the target asset. To address the challenges in few-shot scenarios where only extremely limited samples are available, a bi-directional DT prototype anchoring method with multi-periodicity learning is proposed. Specifically, a framework involving meta-training in the DT virtual space and test-time adaptation in the physical space is constructed for reliable few-shot model adaptation for the target asset. A bi-directional twin-domain prototype anchoring strategy with covariance-guided augmentation for adaptation is further developed to improve the robustness of prototype estimation. In addition, a multi-periodicity feature learning module is designed to capture the intrinsic periodic characteristics within current signals. A DT of an asynchronous motor is built based on finite element method, and experiments are conducted under multiple few-shot settings and three working conditions. Comparative and ablation studies demonstrate the superiority and effectiveness of the proposed method for few-shot fault diagnosis.

In modern industry, fault diagnosis is critical for ensuring production safety and efficiency. Intelligent fault diagnosis (IFD) methods often suffer performance degradation under varying real-world operating conditions, primarily due to distribution shifts. Domain generalization (DG) techniques have been introduced to address this issue, aiming to enhance model performance under unseen working conditions. However, existing methods often overlook the significance of multilevel feature representations in learning generalizable information. Furthermore, most methods disregard the fact that not all source domains contribute equally to generalization, where certain domains provide more transferable features than others. To address these challenges, this article proposes a self-distillation-based domain weighted generalization method for fault diagnosis. With the Vision Transformer (ViT) as the backbone, the method first incorporates a self-distillation mechanism that distills class-specific knowledge within class tokens from the final Transformer block to intermediate blocks, enabling the learning of more generalized features. Second, a dynamic domain weighting strategy based on self-distillation loss is proposed, which assigns higher weights to domains with more generalized features. Finally, these weights are integrated into domain adversarial learning and fault classification process. Extensive experiments on motor fault diagnosis are conducted, covering multiple conditions, including constant conditions, start-up and shut-down conditions, and the New European Driving Cycle (NEDC) conditions. Experimental results on 33 cross-condition tasks, including constant-to-constant, constant-to-variable, and variable-to-variable tasks, demonstrate that the proposed method achieves superior accuracy and robustness compared with several representative methods. Ablation studies further validate the effectiveness of each component in the proposed method.

Reliable health assessment of hydraulic power components is essential for autonomous construction machinery. As the core power component, the health of axial piston pumps directly impacts the safety of automated process. However, traditional single-source data-driven methods are limited by incomplete information and vulnerable to sensor malfunctions or data loss. To address this, we propose a dual-path physics-informed heterogeneous information fusion (DP-PIHIF) framework, which consists of a Virtual Health Indicator (VHI) path and a Physical Health Indicator (PHI) path, integrating high-frequency vibration and relatively low-frequency physical data. In the VHI path, multi-domain vibration features are extracted and compressed using a Variational Autoencoder to characterize dynamic fault signatures. In the PHI path, a Physics-Informed Transformer is developed to infer interpretable leakage-related coefficients from multi-sensor hydraulic measurements, with physical consistency enforced through a volumetric flow loss model. The resulting VHI and PHI representations are integrated using an adaptive heterogeneous fusion module with dynamic weight allocation, followed by a multi-task learning scheme for simultaneous health state classification and ordinal health-level regression. Experimental results show superior assessment accuracy and regression stability, outperforming single-source baselines. This indicates the promising potential of the proposed framework for remote health assessment, which may support the predictive maintenance and operational continuity in autonomous construction machinery.

Acoustic sensing provides a non-intrusive pathway for motor health monitoring, yet its application to motor fault diagnosis has been extremely limited. Existing acoustic studies on rotating machinery have mainly employed single microphones and have rarely addressed induction motors themselves, leaving the spatial and metrological aspects of acoustic measurements largely unexplored. This paper introduces a multi-channel microphone-array framework that integrates beamforming-based spatial filtering with a lightweight Transformer classifier. The array is calibrated and its spatial response characterized to ensure traceability and repeatability. Beamforming enhances the signal-to-noise ratio and angular selectivity of fault-induced acoustic components, while the Transformer captures temporal–spectral dependencies in the beamformed Mel-spectrograms. Experiments on a drivetrain testbed under varying loads and background noise demonstrate that the proposed system achieves 96.87% average diagnostic accuracy with high robustness and computational efficiency (0.026 GFLOPs per sample). Beyond accuracy, the framework is examined from a metrological perspective—its calibration, measurement repeatability, and uncertainty contributors are analyzed to establish the reliability of acoustic array sensing for reproducible, non-contact motor fault diagnosis.

Intelligent fault diagnosis has become an indispensable technique for ensuring machinery reliability. However, existing methods suffer significant performance decline in real-world scenarios where models are tested under previously unseen working conditions, while domain adaptation approaches are limited by their dependence on target-domain samples. In addition, most existing studies rely on single-modal sensing signals, overlooking the complementary nature of multi-modal information for improving model generalization. To address these limitations, this paper proposes a multi-modal cross-domain mixed fusion model with dual disentanglement for fault diagnosis. A dual disentanglement framework is developed to decouple modality-invariant and modality-specific features, as well as domain-invariant and domain-specific representations, enabling both comprehensive multi-modal representation learning and enhanced generalization to working conditions. A cross-domain mixed fusion strategy is designed to increase modality and domain diversity by randomly mixing modality information across domains. Furthermore, a triple-modal fusion mechanism is introduced to adaptively integrate multi-modal heterogeneous information. Extensive experiments are conducted on induction motor fault diagnosis under both unseen constant and time-varying working conditions. The results demonstrate that the proposed method consistently outperforms advanced methods, and comprehensive ablation studies further verify the effectiveness of each proposed component and the advantage of multi-modal fusion.

The accurate fault diagnosis of asynchronous motors is critical in industry, and intelligent diagnosis methods have demonstrated notable performance. However, the scarcity of labeled data in real-world scenarios constrains model performance, while efficiently extracting fault-related features from three-phase stator current signals remains a persistent challenge. In this article, a new method of knowledge distillation from high-fidelity electromagnetic simulations to actual measurement data is proposed. First, the empirical wavelet transform recurrence plot (EWT-RP) is proposed to adaptively decompose fault-induced components from current signals and convert them into 2-D colored images, emphasizing variances across both time and phases. A self-attention distillation strategy incorporating feature and attention distributions is designed to enable the model to inherit the capability of attending to fault-related representations from simulations. In addition, a novel class-aware Sinkhorn distance (CASD) is proposed to achieve intraclass knowledge transport. Comparative and ablation experiments conducted on an asynchronous motor under two operating conditions validate the effectiveness of the proposed method.

Recently, significant progress has been made in the research on the diagnosis of diesel engine faults. However, the detection of diesel engine misfire faults under noise conditions often faces the issue of limited available samples. This paper proposes a novel hybrid-attention and residual mechanism-based transfer network (HARTN) for small-sample misfire fault diagnosis under noise conditions. In the feature extraction module, residual blocks and the introduced channel attention and spatial attention mechanism together create a dual-path feature extractor. The hybrid attention mechanism can adjust the channel weights and position parameters, and the original rich information can be preserved through residual blocks. For the classification module, specially designed diverse convolution kernel sizes and layers achieve a fusion of multiscale features before classification. With a target domain sample size of 180 for fine-tuning the model, the average accuracy of HARTN was 99.446% under -10 dB, whereas that of CNN, ResNet, RTN, HATN, and HARN_DNN was 94.727%, 81.985%, 97.526%, 75.898%, and 95.565%, respectively. For variable sample diagnosis tasks, HARTN achieved an accuracy of over 97% with a target domain sample size of only 10, which was better than compared models.

Motor fault diagnosis is a fundamental aspect of ensuring the reliability of industrial equipment. However, industrial scenarios exhibit an inherent data scarcity problem, which imposes significant restrictions on the practical application of traditional deep learning-based intelligent fault diagnosis (IFD) methods. Typically, only a small volume of labeled data along with limited informative unlabeled data are available from industrial motors. Effectively utilizing informative unlabeled samples in the context of few-shot fault diagnosis poses a substantial challenge. In this article, a prototype refinement method for semi-supervised few-shot fault diagnosis based on deep reinforcement learning (DRL) is proposed. First, we propose to formalize a Markov decision process (MDP) of an iterative semi-supervised meta-learning strategy involving the selection of informative unlabeled samples and the refinement of category prototypes. Subsequently, we develop a mirror prototypical network (ProtoNet) structure for interaction with a DRL agent, which learns to adaptively select valuable samples to supervise the diagnosis process. Moreover, a state space involving feature embedding and category information is designed, and a comprehensive reward taking into account selection confidence, effectiveness, and representative is proposed. Extensive experiments on several motor experimental datasets verify the method's effectiveness in few-shot diagnosis of unseen faults and new working conditions.

Incremental fault diagnosis refers to overcoming catastrophic forgetting by constantly learning new knowledge from the collected data stream, thus ensuring that the model can adapt to the changing environment. However, in practical applications, especially in the case of few-shot samples and increasing fault categories, gearbox fault diagnosis faces significant challenges in stability learning ability. To address these challenges, we proposed a convolutional-attention fusion network (CAFNet) based approach to optimize gearbox fault diagnosis capabilities. First, we built a knowledge base to store the information continuously from the data stream. This knowledge base is used to store historical data and includes data preprocessing to ensure data quality and consistency. Secondly, an adaptive weight updating algorithm is designed to dynamically adjust the weights according to the actual performance of the model to improve the adaptive ability of the model. At the same time, the L2 regularization parameter is introduced into the loss function to effectively prevent overfitting. The validity and reliability of this method are verified by two datasets, which provide a new solution for mechanical fault diagnosis.

Deep learning technology has made significant progress in fault diagnosis. However, in real-world industrial settings, most existing methods require substantial labeled data for training, while harsh operating conditions and data collection constraints often result in scarce fault samples. This limitation significantly impairs their diagnostic performance in practical applications. To address this challenge, we propose a few-shot fault diagnosis approach based on a time-frequency contrastive learning (TF-CL) framework. The TF-CL framework adopts a pre-training and downstream task pipeline, enabling the model to automatically learn and extract multi-perspective features from unlabeled data in self-supervised conditions. During the pre-training, dedicated encoders separately extract time-domain and frequency-domain feature representations from abundant unlabeled samples. The extracted features are then projected into a shared time-frequency space using a projector. To ensure that multi-perspective features can be extracted from unlabeled data, this paper introduces a time-frequency consistency loss function, constructed using novel positive and negative sample pairs. In the downstream task, the TF-CL model is combined with a multilayer perceptron classifier and optimized fine-tuned end-to-end using the limited labeled data. Gradient updates during downstream training further refine the learned feature representations, enhancing their adaptability to target classification tasks. The superiority of TF-CL was demonstrated through a variety of fault diagnosis experiments conducted on both public and self-collected datasets.

Diesel engines are critical power sources widely used in marine, transportation, and industrial applications, where reliable operation is essential for safety and economic efficiency. However, traditional signal processing and many machine learning methods face challenges in extracting generalized fault features and accurately diagnosing misfires under complex, noisy operating conditions. To address these challenges, this paper proposes a novel multi-scale bottleneck attention and mixup- based domain adaptation network (MBA-MDAN) for reliable misfire detection across varying noise levels and working conditions. The approach integrates a denoising convolutional neural network (DnCNN) to suppress noise unrelated to fault characteristics, enabling clearer fault signal extraction. A parallel multi-scale convolution module captures fault features at different time scales, while a bottleneck attention module (BAM) selectively emphasizes critical features for deep fault representation. To preserve important information, time-domain statistical features are also incorporated. During training, metric learning minimizes feature discrepancies between source and target domains, and adversarial training between a domain discriminator and the fault classifier enhances domain adaptation. Additionally, domain mixup is applied to augment discriminator samples, further improving diagnostic performance. On the real-world datasets, compared to several state-of-the-art methods—including ShuffleNetV2, DenseNet, ANMCNN, MCBACNN, DANN, and WDAN—MBA-MDAN improves average diagnostic accuracy by 30.372

Convolutional neural network (CNN) was widely applied to the data-driven-based fault diagnosis. However, it often needs to artificially transform the signal into a 2-D image with the help of time-frequency transformation; furthermore, the alternative 1-D CNN can only extract single-scale features but cannot adaptively reveal the relationship between scales even if 1-D convolution kernels of multiple scales are applied at the same time. Accordingly, this article proposes a mixed CNN model, namely, a multiscale dynamic weighted 1-D to 2-D (1D-2D) CNN model (1D-2D MDWCNN), which resembles multiwavelet-based CNN method but uses different scale convolution kernels in 1-D convolution neural network to facilitate the multiscale feature extraction and accomplish adaptive features fusion in the constructed 1D-2D seamless joint network framework. Specially, to reflect the global contribution of different convolution kernels, a dynamic weighted (DW) network layer is constructed to adaptively adjust the global weight of each convolution kernel, so as to improve the fault diagnosis ability. The validity of the model is verified by motor bearing fault diagnosis experiment and gearbox fault diagnosis experiment. The diagnosis results proved the developed 1D-2D MDWCNN model superior to the latest CNN-based models.

Under the dual influence of time-varying working conditions and noise interference, accurately faults diagnosing in mechanical equipment poses significant challenges. Therefore, this paper proposes a limited sample fault diagnosis method using dilation kernel gated recurrent dropout attention unit (GRDAU) for time-varying speed based on interference suppression. Firstly, dilation kernel parameters were integrated into traditional convolution to suppress high-frequency noise. Secondly, to enhance the model’s robustness against noise interference and variations in speed, a GRDAU was developed based on gated recurrent unit (GRU). Additionally, a global cyclic dynamic decay learning strategy was implemented within the GRDAU to better adapt to complex speed variation conditions. Finally, two case studies were conducted to validate the robustness and interference suppression capabilities of the GRDAU. When compared to a range of existing advanced diagnostic methods, it demonstrated superior performance and stronger generalization ability.

A single type of sensor signal cannot fully represent the operational status of mechanical equipment, leading to incomplete state characterization and inaccurate diagnostics. This paper proposes an innovative fault diagnosis method based on Convolutional AutoEncoders combined with multivariate information fusion to accurately identify the overall health status of bearings by analyzing various sensor data. Our approach leverages the Convolutional AutoEncoders to effectively integrate heterogeneous sensor data from multiple sources, including vibration and sound signals, with data augmentation and normalization techniques for preprocessing, thereby improving the model’s generalization capability and accuracy. Furthermore, the integration of K- means clustering and a Sparse Attention mechanism enables precise recognition of critical fault features. The model’s effectiveness is validated through comprehensive performance evaluation using confusion matrices and visualization techniques. Experimental results demonstrate that our method achieves high accuracy and robustness in fault diagnosis tasks, offering a significant advancement in intelligent maintenance and fault prediction of rolling bearings by addressing the limitations of traditional methods.

Autonomous underwater vehicles (AUVs) are widely used in ocean exploration, scientific research, and other fields that perform tasks in complex underwater environments. Since propellers fault samples are very scarce and difficult to collect in AUV practice, traditional fault diagnostics face the challenge of insufficient data. For this reason, a channel attention residual transfer learning (ECRTN) model based on dual-loss nonlinear independent component estimation (DLNICE) is proposed to expand data for few-shot fault diagnosis of AUVs. Specifically, DLNICE is first used to augment fault samples by combining both time and frequency-domain information of AUV propellers vibration signals with a dual-loss function. The augmented fault samples and normal data are fed into the channel-attention residual network, which is fine-tuned by the large language model. The fine-tuned model is then used to diagnose real vibration samples in the target domain. Experimental results show that the proposed ECRTN achieves an average diagnosis accuracy of 96.81 %, outperforming other state-of-the-art fault diagnosis models for dealing with few-shot fault diagnosis tasks. The present method provides an effective technical solution for the few-shot fault diagnosis of AUV propellers and has a better prospect for practical applications.

The human-centered paradigm of intelligent manufacturing is reshaping traditional production models. The fully mechanized coal mining face production system (FMCMFPS) represents a typical application of complex manufacturing systems (MS) in the coal mining industry, with an increased emphasis on the safety of personnel within the system. Among its key components, the real-time monitoring of the scraper conveyor S-shaped bending is critical to ensuring safe and efficient coal mining operations. This monitoring also aligns with the principles of Industry 5.0, which emphasize human-centric, safe and intelligent manufacturing practices. However, the underground environment is harsh, with moisture, dust and high vibrations. These conditions, along with the structure of the scraper conveyor, complicate the accurate capture of data. Issues such as data misalignment, missing information and frame retraction errors lead to inconsistencies between sensor data and actual conditions. To tackle these challenges, the paper presents a virtual-real mapping method for the scraper conveyor S-shaped bending using digital twin technology. A framework and mechanistic model for the S-shaped bending are established, and techniques like the Interquartile Range (IQR), sliding window and filtering algorithms are used to handle anomalous data. The proposed approach is validated through a prototype system at the Ulanmulun Mine in Inner Mongolia.

For aero-engine rotor system, the interface load-bearing performance is crucial for its connection stability. In order to effectively ensure the connection performance of aero-engine in strong service environments, this paper proposed an intelligent assembly-adjustment method that integrates key feature measurement, performance prediction and process feedback. Taking the typical multi-bolts connection structure in aero-engine rotor system as the research object, firstly, by using the intelligent fastener, the precise acquisition of preload distribution after assembly process has been achieved. Then, for the issue of interface load-bearing performance weakening represented by non-uniform slip behavior, a CON-Delta e agent model was constructed for predicting the non-uniformity of interface slip. Finally, based on CON reduction principle, a local preload feedback-adjustment method for improving the interface load-bearing performance has been proposed. Through a simulated rotor assembly case, the effectiveness of the proposed intelligent assembly-adjustment method in improving interface load-bearing performance was demonstrated. Overall, the research content has a better application prospect in connection stability improvement of mechanical system.

Bearing and reducer are the key transmission components of rotating machinery, timely and accurate fault diagnosis is an important guarantee for the safe operation of rotating machinery transmission system. The collected signals of bearing and reducer have typical non-stationary and nonlinear characteristics. In order to make full use of the spatial and temporal features contained in the running signals, combining with bidirectional Long short-term memory network, this paper proposes a one-dimensional convolutional neural network fault diagnosis model SAM-1DCNN-BiLSTM (Spatial attention model 1DCNN BiLSTM) based on attention mechanism with no pooling layer. Specifically, omitting pooling layer, the model uses spatial attention model (SAM) weighted 1DCNN network to extract spatial features, then, the BiLSTM layer is following to extract the time series features, and finally the device running state features that integrate spatial and temporal feature information are obtained, and the intelligent fault identification of key transmission components of rotating machinery are conducted. The performance of the developed method is verified with the experiments of the key components of bearing and reducer. For bearings, the average diagnostic accuracy reaches 99.77% with a variance of 0.03. For reducer, the average diagnostic accuracy reaches 99.00% with a variance of 0.11. Compared with the state-of-the-art deep learning diagnostic models, the proposed SAM-1DCNN-BiLSTM achieves better diagnostic performance.

Accurate health assessment of axial piston pumps is vital to overall system performance, yet the pumps' intricate architecture, subtle fault signatures, and limited fault data challenge conventional diagnostics. This paper proposes a health-assessment method that couples a physics-informed neural network (PINN) with a Flow Loss Model. By fusing multi-sensor signals with flow measurements, the method inverts three physically interpretable parameters-k(1) (compression-loss coefficient capturing flow lost to fluid compressibility and chamber elasticity),k(2) (laminar-leakage coefficient quantifying clearance leakage in the low-Reynolds regime), and k(3) (turbulent-leakage coefficient)-to characterize pump health. The network is trained under a maximum a posteriori (MAP) objective that integrates measurement fit, physics-residual constraints, and priors to stabilize the inversion. Validation on a labeled experimental dataset shows that the proposed PINN surpasses an otherwise identical purely data-driven model in prediction accuracy and health-state recognition, while achieving superior physical consistency. Notably,k(2) increases monotonically with health degradation, providing a salient, physics-grounded indicator for distinguishing health states and evidencing strong generalization potential.

[Objective]To address the challenge of accurately capturing weak features in vibration signals under strong noise interference,a joint filtering method combining ensemble empirical mode decomposition(EEMD),fast kurtogram(FK),and adaptive maximum correlation kurtosis deconvolution(AMCKD)was proposed.[Methods]Firstly,the vibration signal was decomposed into multiple intrinsic mode functions(IMF)via EEMD for multiscale analysis.The IMF components were then screened using cross-correlation coefficients and kurtosis as evaluation metrics,followed by signal reconstruction.Next,the fast kurtogram algorithm was employed to determine the carrier frequency,bandwidth,and the layer with the maximum kurtosis value of the reconstructed signal,enabling the design of a bandpass filter for noise reduction.Subsequently,particle swarm optimization(PSO)was utilized to adaptively determine the MCKD parameters,and the AMCKD algorithm was applied to enhance the features of the filtered signal.Finally,the fault characteristic frequency was extracted via envelope demodulation and compared with the theoretical value to achieve fault diagnosis.[Results]The results demonstrate that the proposed method effectively extracts weak features under strong noise interference,exhibiting robust noise resistance.This approach provides valuable reference for research on identifying bearing weak features in high-background-noise environments.