Diagnosing motor eccentricity faults is essential for rotational machinery stability. However, these faults in the initial stage exhibit weak characteristics that are easily submerged by noise. Additionally, the scarcity of fault samples and imbalanced data distribution make it difficult for fault diagnosis models to learn key features, thereby limiting their generalization ability. To alleviate this issue, this paper proposes a small-sample motor fault diagnosis method based on data augmentation via a multi-scale statistical feature optimized diffusion model. First, fault samples are sampled across multiple scales, followed by the extraction of their statistical features. The error of features between generated samples and real samples is further estimated to optimize the loss function of the diffusion model, thereby improving the quality of synthetic data. Secondly, a residual network architecture for learning data features and a framework for diffusion timesteps are specifically designed for motor eccentric faults, thereby enabling high-quality data augmentation. Finally, comparative experiments on a motor eccentricity fault datasets demonstrate that, compared to traditional data augmentation methods, the fault samples generated by our proposed method exhibit superior performance in terms of statistical features and similarity to the original samples. When the real data and generated data are mixed in different proportions for training, the fault diagnosis accuracy can be improved by 10% over the real-sample-only baseline. These results further validate that the proposed data augmentation method significantly enhances the generalization ability of diagnostic models, underscoring its practical value in data-scarce scenarios.

Gear fault diagnosis in heavy-duty rotational transfer devices is challenging because vibration signals are strongly affected by intermittent motion, load variation, structural coupling, and external disturbances. Under such conditions, discriminative fault information is usually distributed across localized transients, temporal evolution patterns, and long-range contextual dependencies, which limits the effectiveness of single-view models and fixed fusion strategies. To address this issue, this study proposes a sample-adaptive multi-cue diagnosis framework that explicitly organizes fault diagnosis around three complementary cue types: local, temporal, and global cues. Dedicated cue encoders are used to extract these heterogeneous fault-related representations, and a gated fusion mechanism is introduced to adaptively determine their contributions for each input sample. To improve deployment feasibility, a compact three-channel subset is further constructed from the original ten-channel sensing system. Experiments on a self-developed rotational transfer device gear dataset show that the proposed framework achieves reliable classification under known operating conditions and maintains 97.1% accuracy under unseen operating conditions. The method also remains robust under moderate noise corruption and outperforms representative baselines and ablation variants. Additional validation on the SEU bearing dataset further confirms its cross-scenario applicability and interpretable fusion behavior. These results demonstrate that the proposed framework offers an adaptive and interpretable soft-computing solution for robust fault diagnosis under complex industrial conditions.

Due to the weak fault signals of motor bearings and the extreme difficulty in obtaining sufficient data samples, the requirements for training fault diagnosis models cannot be met. Moreover, current machine learning-based fault diagnosis methods lack interpretability. To address this issue, this paper proposes a small-sample bearing fault diagnosis method based on interpretable component energy features (ICEF). By combining the component energy values of bearing vibration signal samples with interpretable artificial intelligence techniques, an interpretable feature selection method is proposed. This method extracts effective features of bearing faults as training features for the fault diagnosis model, enhancing the model's ability to learn and identify fault category characteristics. Finally, experiments demonstrate that the fault diagnosis accuracy reaches 99.5% and 97.5% when component energy is used as the learning feature. Moreover, by mixing data from two bearing datasets and dividing them into training and test sets at different ratios, an accuracy of 91.7% is achieved, validating the effectiveness of the proposed method under small-sample conditions.

Gear is the critical transmitting component of a reciprocating manipulator, whose failure can cause changes in the output current of the drive motor. Many fault diagnosis methods based on convolutional neural networks take the current as a monitor signal. However, there are some shortcomings in the application of convolutional neural networks. On the one hand, the local receptive field limits the traditional convolutional layer, which cannot efficiently learn long-term spatial dependencies in sequences. On the other hand, the loss function fails to pay proper attention to imbalance fault patterns, leading to a lack of generalization ability for the diagnosis model. This article proposed a robust diagnosis method using a global convolution network (GCN) and sample weighted (SW) loss (dubbed SW-GCN) for gear fault in a reciprocating manipulator. Therein, a novel convolutional module, termed the global depthwise separable convolution (GDSConv) module, is designed to improve the convolution operation's receptive field by introducing the Fourier transform. Moreover, a sample imbalance loss design guideline that considers the sample information is proposed to enhance the robustness of the diagnostic model. A series of experimental studies using data from a laboratory environment are carried out to verify the performance of the proposed method for gear fault diagnosis. Experimental results show that the proposed method can effectively diagnose six types of gear faults. Besides, comparing it with other methods shows that the proposed method performs strongly robustly under imbalanced sample and noise conditions.

Object detection in specialized fields-such as industrial inspection, autonomous driving, and remote sensing-presents significantly greater challenges than in general scenarios, particularly when detecting small targets in complex and cluttered environments. These scenarios often involve challenges such as low resolution, occlusion, scale variation, and class imbalance, all of which can significantly degrade the performance of conventional detection models. To effectively improve the detection performance of small objects in complex environments, we propose the information-coordinated fusion attention mechanism (ICFAM), which extracts features from multiple spatial directions and effectively preserves the essential positional information of targets. Building on this, we incorporate an inverted bottleneck structure and depthwise separable convolutions to develop the large-kernel correlation module (LKCM). This module incorporates ICFAM, which capitalizes on the synergy between expansive receptive fields and attention mechanisms. Furthermore, we design an efficient multi-branch network architecture known as the coordinative efficient layer aggregation network (CELAN). This architecture replaces traditional stacked convolutions with depthwise separable convolutions, thereby reducing computational load. It also introduces an independent pathway that integrates with LKCM, ensuring noninterfering computation and enabling deeper computational blocks to extract richer, multi-level semantic information. Using CELAN as the core module and the YOLOv9 framework, we develop a novel object detector YOLO-CELAN. Extensive testing on two public datasets substantiates the effectiveness of the proposed modules. This work demonstrates that the YOLO-CELAN series (N/S/M/L) significantly improves mAP50 by 1.2 %, 2.9 %, 1.9 %, and 2.3 %, respectively, compared to the YOLOv9 models (S, M, C, and E).

The ammunition loading system manipulator is susceptible to gear failure due to high-frequency, heavy-load reciprocating motions and the absence of protective gear components. After a fault occurs, the distribution of fault characteristics under different loads is markedly inconsistent, and data is hard to label, which makes it difficult for the traditional diagnosis method based on single-condition training to generalize to different conditions. To address these issues, the paper proposes a novel transfer discriminant neural network (TDNN) for gear fault diagnosis. Specifically, an optimized joint distribution adaptive mechanism (OJDA) is designed to solve the distribution alignment problem between two domains. To improve the classification effect within the domain and the feature recognition capability for a few labeled data, metric learning is introduced to distinguish features from different fault categories. In addition, TDNN adopts a new pseudo-label training strategy to achieve label replacement by comparing the maximum probability of the pseudo-label with the test result. The proposed TDNN is verified in the experimental data set of the artillery manipulator device, and the diagnosis can achieve 99.5%, significantly outperforming other traditional adaptation methods.

In response to the tight gear arrangement and the high accuracy required for recognition in gear transmission systems, this paper proposes an improved lightweight model for gear detection, called HPF-YOLO. Built upon YOLOv11, the model incorporates modules such as wavelet transform and multi-scale feature fusion, and optimizes key components like feature extraction and fusion. Experiments are conducted on a dedicated dataset simulating a gear transmission system. The results show that the proposed model outperforms the benchmark YOLOv11 in both accuracy and speed.

In most industrial processes, rotating machinery often operates under time-varying speed conditions, making it difficult to extract fault-related features using conventional diagnostic techniques. To address this challenge, a novel diagnostic framework is developed that integrates high-resolution time-frequency analysis with interpretable causal learning. First, a synchrosqueezed continuous wavelet transform (SSQ-CWT) is employed to generate a clean time-frequency representation, upon which a specialized Adaptive Rotational Frequency Ridge Extraction (ARFRE) algorithm operates to accurately estimate the instantaneous speed without a tachometer. Subsequently, features extracted from the resulting order spectrum are screened by a Causal Forest to identify a subset with true causal impact on the fault state. Finally, an interpretable decision tree is constructed using this causally validated feature set. The framework was comprehensively validated on two public benchmarks (SQV, Ottawa) and a proprietary industrial dataset. The results demonstrate superior diagnostic performance, achieving accuracies of 98.49%, 100.00%, and 97.44% respectively, and consistently outperforming baseline methods. This work presents a robust, interpretable, and tachometer-free solution for fault diagnosis in variable-speed machinery, with significant potential for practical industrial application.

Fault diagnosis occupies a pivotal position within the domain of machine and equipment management. Existing methods, however, often exhibit limitations in their scope of application, typically focusing on specific types of signals or faults in individual mechanical components while being constrained by data types and inherent characteristics. To address the limitations of existing methods, we propose a fault diagnosis method based on graph neural networks (GNNs) embedded with multirelationships of intrinsic mode functions (MIMF). The approach introduces a novel graph topological structure constructed from the features of intrinsic mode functions (IMFs) of monitored signals and their multirelationships. Additionally, a graph-level based fault diagnosis network model is designed to enhance feature learning capabilities for graph samples and enable flexible application across diverse signal sources and devices. Experimental validation with datasets including independent vibration signals for gear fault detection, mixed vibration signals for concurrent gear and bearing faults, and pressure signals for hydraulic cylinder leakage characterization demonstrates the model's adaptability and superior diagnostic accuracy across various types of signals and mechanical systems.

Gear fault diagnosis plays a crucial role in detecting gear faults promptly, enabling timely repairs or replacements to minimize potential losses. While deep learning-based methods have been widely used for gear fault diagnosis, their effective application in real-world industrial environments remains a challenge. In this paper, we introduce a novel approach that combines a multi-scale convolutional neural network with multiknowledge distillation (MSCNN-MKD). This approach is deployed on edge computing nodes for online gear fault diagnosis. Experiments results validate the effectiveness of our proposed methodology. The proposed method achieves an accuracy of 98.76% in recognizing 7 fault categories. Additionally, the number of parameters is 18.54K, the floating-point operations are 41.54M, and the average inference time is 1.07ms. The methodology has great potential for practical industrial applications in gear faults diagnosis.

Inter-turn short circuit fault is one of the common faults in permanent magnet synchronous motor (PMSM). It can lead to changes in the time domain output current of a PMSM and also manifest corresponding fault characteristics in the frequency domain. However, there are several problems with the present motor fault diagnosis based on deep learning. First, traditional methods ignore information from the frequency domain, which causes poor diagnostic accuracy. Second, there are few good solutions to the problem of noise interference in the model diagnosis. In response to this situation, this paper proposes a novel neural network, named Dual-domain Convolution Neural Network, which composed of Dual-domain Convolution (DC) mod-ules. In the proposed module, the frequency-domain branch adopts Fourier Separable Convolution to explore the frequency information, the time-domain branch to extract the time features with Depthwise Separable Convolution. Experiments results show that the proposed method can effectively diagnose inter-turn short circuit faults and locate fault positions. Besides, compared with other methods, the proposed model performs strong robustness under different degrees of noise interference.

Object detection is one of the research hotspots in computer vision. However, most existing object detectors struggle with the identification of small targets. Therefore, the paper proposes two modules: the MDFFAM (Multi-Directional Feature Fusion Attention Mechanism) and the LKSPP (Large Kernel Spatial Pyramid Pooling), to enhance the detector's effectiveness in identifying subtle faults on the surface of mechanical equipment. LKSPP aims to expand the receptive field to capture high-level semantic features through large kernels. Meanwhile, the MDFFAM allows the network to efficiently utilize spatial location information and adaptively recognize detection priorities. In the detection task, MDFFAM effectively captures feature information in three spatial directions: width, height, and channel, with the location information fully utilized to establish stable long-range dependencies. Moreover, LKSPP boasts a larger receptive field and imposes less computational burden compared to the SPPCSPC by YOLOv7. Finally, experiments demonstrate that the proposed module effectively improves the detection accuracy for small targets, surpassing the state-of-the-art object detector, YOLOv7. Remarkably, MDFFAM incurs almost negligible computational overhead.

Inverter open-circuit faults can cause significant changes in the output current of permanent magnet synchronous motor (PMSM), so many fault diagnosis methods based on convolutional neural networks (CNNs) take the current as monitor signals. However, there are some shortcomings in the application of CNNs for fault diagnosis. On the one hand, most of the traditional pooling methods confuse the different frequency components of the feature map due to ignoring the classical sampling theorem, which decreases the accuracy of fault diagnosis. On the other hand, the generalization performance of the diagnostic model is not good enough when the data distribution changes due to variations in the operating conditions of the drive system. To address these issues, this article proposes an open-circuit fault diagnosis method using an adaptive sparse attention wavelet network (ASAWN), where a novel pooling layer, termed as adaptive sparse attention wavelet pooling (ASAW-pooling) layer, is designed using discrete wavelet transform (DWT) and adaptive sparse attention (ASA) mechanism. The pooling layer improves the robustness of the diagnostic model by filtering out the high-frequency components of the current data and the redundant information of the attention mechanism. Experimental results show that the proposed method can effectively diagnose 22 types of open-circuit faults. Besides, a comparison with other models shows that the proposed model not only has high accuracy but also performs strong robustness.

Gear fault diagnosis is an important part of rotating machinery maintenance. Most state-of-the-art research methods in the field of gear fault diagnosis combine deep learning methods with multisensor data fusion, big data analysis, and physical models to achieve more accurate fault diagnosis techniques; however, there are still some shortcomings in the method of gear fault diagnosis based on deep network models. First, the impact of noise on model diagnostics is rarely considered, and second, there is a conflict between a large number of parameters and calculations of deep networks and the computing resources of current embedded platforms. To address these issues, this article proposes a multilayer fusion convolutional neural network (AMFCNN) and a multilayer fusion module-relational knowledge distillation (MFM-RKD) module based on an attention mechanism. The multilayer fusion network uses an inception module to extract multiscale features from multisensor information and adopts an attention mechanism to extract features from the network in different periods. The MFM-RKD module activates the extracted features of different periods through a fully connected layer and transfers structural knowledge to a lightweight student network through RKD to build a lightweight multisensor gear fault diagnosis model. Experiments show that the lightweight model has excellent robustness in different noise environments. Compared with the teacher network, the model performance is reduced by less than 2%, while the calculations of the model are reduced by 92 times and the running memory is reduced by 7.74 times, which provides an efficient solution to address the robustness and model deployment issues in gear fault diagnosis.

The gear fault diagnosis technology based on the signal is crucial for maintaining the normal operation of the gear in the motor drive chain. In some cases, it is challenging to add sensors on the unit of the motor transmission chain for collecting vibration signals in practical engineering applications. However, the current signal can be collected. Nonetheless, due to the long distance between the collection point and the fault source, it becomes difficult to extract the features of the weak gear fault from the current signal. In order to solve the aforementioned problems efficiently, an optimized principal neighborhood aggregation (OPNA) graph neural network (GNN) was proposed to diagnose gear faults in the motor drive chain. First, the current signal is reconstructed to obtain the topological data graph sample by the graph sample construction method proposed in this article. Second, OPNA, an architecture that combines multiple message aggregators with a degree scaler, was designed to extract the features of nodes and edges. Subsequently, the embedding and the particular pooling improvement were used to reduce the number of nodes and achieve steady and rapid classification. Finally, the experimental studies, based on the current signal of the gear dataset, were conducted to validate the effectiveness of the proposed method and its superiority over the traditional methods.

The open-circuit fault is one of the most common faults of the automatic ramming drive system (ARDS), and it can be categorized into the open-phase faults of Permanent Magnet Synchronous Motor (PMSM) and the open-circuit faults of Voltage Source Inverter (VSI). The stator current serves as a common indicator for detecting open-circuit faults. Due to the identical changes of the stator current between the open-phase faults in the PMSM and failures of double switches within the same leg of the VSI, this paper utilizes the zero-sequence voltage component as an additional diagnostic criterion to differentiate them. Considering the variable conditions and substantial noise of the ARDS, a novel Multi-resolution Network (MrNet) is proposed, which can extract multi-resolution perceptual information and enhance robustness to the noise. Meanwhile, a feature weighted layer is introduced to allocate higher weights to characteristics situated near the feature frequency. Both simulation and experiment results validate that the proposed fault diagnosis method can diagnose 25 types of open-circuit faults and achieve more than 98.28% diagnostic accuracy. In addition, the experiment results also demonstrate that MrNet has the capability of diagnosing the fault types accurately under the interference of noise signals (Laplace noise and Gaussian noise).

The fault diagnosis of gear is indeed a crucial aspect of maintaining rotating machinery, as it helps in ensuring the safe and efficient operation of industrial equipment. Deep learning models have gained significant attention for gear fault diagnosis due to their ability to automatically extract features from raw data, but they also come with their own set of challenges. One major limitation of existing methods is the insufficient consideration given to the impact of environmental noise at industrial field on the diagnostic effectiveness of the models. Additionally, there is a contradiction between the week computational resources of current embedded platforms for industrial field device applications and the large number of parameters and computations required for deep learning models. This may hinder the deployment of complex models in industrial field devices. To address these issues, a novel approach to multi-path convolutional neural network with dual branch attention (AMPCNN) has been proposed. This approach aims to enhance the recognition of different fault types and maintain high accuracy in noisy environments by extracting multi-scale features of the original vibration signal using multi-path convolution and dual branch attention mechanisms. Furthermore, a multi-knowledge distillation (MKD) method has been introduced to construct lightweight multi-sensor gear fault diagnosis models. This approach facilitates the transfer of multiple knowledge from a complex teacher network to a simpler student network, resulting in a lightweight model that exhibits excellent robustness in various noise environments. The experimental results show that the lightweight model achieves high accuracy while requiring significantly fewer floating-point operations and parameter quantities compared to the original teacher network.

Defect detection on the gear surface is crucial for preventing faults in mechanical systems. However, most detection models are not extremely effective on embedded platforms. To address this issue, we present a lightweight detection model called CDC-YOLO, which is based on YOLOv8 and specifically designed for embedded platforms. It utilizes our proposed CDC module as a residual structure for extracting multi-scale features, allowing for better adaptation to different platforms. Additionally, we achieve model lightweighting by using a dual convolutional architecture. Experimental results on both computers and embedded platforms demonstrate that our proposed method outperforms the baseline YOLOv8.

To enhance the reliability and safety of industrial systems, gear fault diagnosis in manipulators is crucial. However, the operation of manipulators often occurs under variable working conditions, resulting in vibration signals with characteristics such as nonlinearity, non-stationarity, and strong environmental noise. Additionally, studying vibration signals traditionally necessitates the use of extra vibration sensors, thereby increasing operational costs for the manipulator. In response to these challenges, this paper introduces a multiscale kernel convolutional neural network (MKCNN) that relies on three-phase stator current data for effective gear fault diagnosis. First, Z-score normalization is used to preprocess the data to enhance the training efficiency of the network model for current data under complex working conditions. Then, a convolutional neural network architecture with multi-scale convolution kernels was constructed to extract multi-level sensory information through convolution kernels of different scales to improve the feature extraction capabilities of the model. Additionally, an attention mechanism module is introduced, assigning higher weights to features in proximity to feature frequencies. Finally, to validate the effectiveness of the proposed method, current signals are collected from a real manipulator gear test platform. Experimental results show that MKCNN can fully extract multi-resolution information and has a high accuracy in gear fault detection, which is higher than other comparison methods.

In activity recognition, Wi-fi-based Channel State Information (CSI) has the advantage of capturing fine-grained information. In the process of action judgment, it is necessary to identify the interval in which each action occurs. For the detection of the motion interval, many methods in the past set statistical thresholds according to experience, find out special areas through sliding Windows, or determine the dynamic changes of signals at the start and end of actions through dynamic time warping. However, relying solely on empirical thresholds can be challenging due to noise interference. To address these challenges, researchers propose a novel approach using a one-dimensional convolutional neural network (CNN) to traverse the data, achieving results similar to sliding windows. This allows the CNN to consider all available data for improved performance. However, CSI data are often the distortion changes of each channel in Wi-Fi over a period of time. If the convolutional neural network is simply used, the time characteristics of the data are ignored. In this paper, conv-LSTM is used to extract the time characteristics of each layer of the data extraction network, and an overall fusion of each feature is carried out in the data fusion network. To improve the processing effect of CSI data. Besides, directly classifying actions using a convolutional network may yield low accuracy. To overcome this limitation, in this paper, we combine the bidirectional long short-term memory network to scan and classify the action interval, and improve the overall classification accuracy