相关参考文献
注意：仅列出部分参考文献，下载原文获取全部文献信息。
Article
Computer Science, Artificial Intelligence
Ruiyi Ma et al.
Summary: A meta learning intelligent fault diagnosis method is proposed to address the problem of new faults not being identified due to lack of training data in the process of equipment operation. The method utilizes multi-scale dilated convolution and relation module for feature extraction and fault diagnosis. The training set is transformed into multiple tasks using meta learning strategy to train the proposed method, which is validated through bearing and gearbox experiments.
KNOWLEDGE-BASED SYSTEMS
(2023)
Article
Engineering, Multidisciplinary
Ning Zhou et al.
Summary: This paper proposes a method called CFFsgram based on candidate fault frequencies (CFFs) for the optimal demodulation frequency band (DFB) identification of axle-box bearings. The vibration signal is divided into different narrowbands using a 1/3-binary tree filter bank constructed by empirical wavelet transform. The local features of the squared envelope spectra of the narrowband signals are used to identify the CFFs, which are frequencies most likely associated with bearing faults. An indicator calculated on the narrowband signals is designed to guide the selection of the DFB. The CFFsgram is shown to be superior in resisting strong noise and random impulses through experiments with challenging datasets.
Article
Computer Science, Artificial Intelligence
Xue Han et al.
Summary: This paper provides a comprehensive account of the opportunities and challenges of Transformer-based multimodal pre-trained models in various domains. It reviews representative tasks of multimodal AI applications and analyzes state-of-the-art Transformer-based multimodal models from different aspects. The paper concludes with key challenges in the field and suggests future research directions.
Review
Computer Science, Artificial Intelligence
Xiaoxu Li et al.
Summary: This paper provides an up-to-date review of deep metric learning methods for few-shot image classification from 2018 to 2022. The methods are categorized into three groups based on three stages of metric learning. The trends suggest a shift towards learning task-specific features, computing task-dependent prototypes or learning prototypes, and learning similarities through convolutional or graph neural networks. The paper also discusses the challenges and future directions of few-shot deep metric learning and summarizes its applications to real-world computer vision tasks.
PATTERN RECOGNITION
(2023)
Article
Computer Science, Information Systems
Xiaoyan Zhu et al.
Summary: This study proposes a new method called MLDE for solving the multi-label classification problem. It selects the most competent ensemble of base classifiers to predict each unseen instance, effectively utilizing label correlation and achieving better performance.
INFORMATION SCIENCES
(2023)
Article
Engineering, Multidisciplinary
Junbo Long et al.
Summary: Time-frequency analysis is widely used to detect fault signals in rotating machinery. The vibration signals from mechanical bearing faults are non-stationary and non-Gaussian, following an alpha stable distribution with a characteristic index of 1 < alpha < 2. Existing linear chirplet transform methods fail to handle alpha stable distribution noise. This paper proposes robust time-frequency analysis methods, such as FLOCWT, FLOLCT, FLOGLCT, and FLOVSLCT, to overcome the influence of impulse noise. Simulation results demonstrate the advantages of these new methods compared to existing methods based on second order statistics in both Gaussian and alpha distribution environments. The improved methods are applied to analyze bearing fault data contaminated by alpha stable distribution noise, successfully extracting the fault signature and showcasing their performance.
Article
Engineering, Mechanical
Jing Yuan et al.
Summary: This study proposes a multichannel feature synchronous extraction tool named Msegram for detecting rolling bearing faults. The method first utilizes high order singular value decomposition (HOSVD) to preprocess the multichannel signals by tensor synchronization denoising. Then, a multi-layer K-value multivariate variational mode decomposition (MVMD) is designed for synchronous adaptive filtering and decomposition. Finally, a tower-shaped crest factor of envelope spectrum (EC) diagram is used to visualize the output of multichannel bearing fault feature results and extract the optimal analytic results.
MECHANICAL SYSTEMS AND SIGNAL PROCESSING
(2023)
Article
Engineering, Mechanical
Shen Liu et al.
Summary: This paper proposes a fault diagnosis method for time series based on Attentional Contrastive Calibrated Transformer (ACCT), which captures both low-level and global features using multiple convolutional layers and the transformer. It also improves diagnostic accuracy through data augmentation and unsupervised contrastive learning.
MECHANICAL SYSTEMS AND SIGNAL PROCESSING
(2023)
Article
Engineering, Mechanical
Yanfei Jin et al.
Summary: This paper investigates the stochastic resonance and mean-first passage time of a quad-stable potential in the presence of Gaussian white noise and periodic forcing. The analytical expressions of mean-first passage time and spectral amplification are obtained, respectively. It is found that noise-assisted hopping and accelerated escape process occur in the system under different noise intensities. Moreover, the paper proposes a multi-stable stochastic resonance method for fault diagnosis and demonstrates its superior performance compared to existing methods.
PROBABILISTIC ENGINEERING MECHANICS
(2023)
Article
Engineering, Industrial
Ruxue Bai et al.
Summary: In this paper, a novel data representation method based on fractional Fourier transform (FRFT) and recurrence plot transform is proposed for machinery fault diagnosis. Experimental results show that the proposed method outperforms conventional methods such as Fourier spectrum and short time Fourier transform. The fusion of maximum kurtosis based fractional Fourier domain recurrence plot and time domain recurrence plot achieves the best performance, making the trained convolutional neural network adaptive to variable working conditions.
RELIABILITY ENGINEERING & SYSTEM SAFETY
(2023)
Article
Automation & Control Systems
Kai Zheng et al.
Summary: Group-sparse mode decomposition (GSMD) is a decomposition method that takes into account the sparse property of signals in the frequency domain. It has shown high efficiency and robustness against noise, making it a promising approach for diagnosing bearing faults. However, limitations such as not considering the impulsiveness and periodicity of bearing faults and difficulties in locating the informative frequency band hinder its application in extracting incipient fault features. To address these limitations, an adaptive group sparse feature decomposition (AGSFD) method is proposed, which models harmonics, large-amplitude random shocks, and periodic transient features as limited bandwidth signals. The AEDOHNR indicator is introduced to guide the construction and optimization of the filter bank in AGSFD, and the regularization parameters are adaptively determined. The AGSFD method, with an optimized filter bank, is able to decompose the original bearing fault into components, including the sensitive fault-induced periodic transient component. Simulation and experimental studies demonstrate the feasibility and superiority of the AGSFD method, showing its capability to identify early failure under heavy noise, strong harmonics, or random shocks and its improved decomposition efficiency.
Article
Energy & Fuels
Ling Xiang et al.
Summary: A new method is proposed in this study to extract multidirectional spatio-temporal features of SCADA data for wind turbine condition monitoring, using convolutional neural network and bidirectional gated recurrent unit with attention mechanism. This method can effectively detect early abnormal operation and identify failed components of wind turbine, showing better feasibility for practical wind energy application.
Article
Biology
Chenxin Li et al.
Summary: Medical imaging datasets often experience domain shift due to various factors, leading to concerns about the generalization capacity of machine learning models. Domain generalization (DG) and meta-learning methods have been introduced to address these challenges. However, the limited availability of annotated source domains in clinical practice poses a risk of overfitting. This paper proposes a novel DG scheme with episodic training and task augmentation to enhance the variety of training tasks and mitigate task-level overfitting. Exploiting a meta-objective to regularize deep embedding further validates the efficacy of the proposed method through experiments on histopathological and abdominal CT images.
COMPUTERS IN BIOLOGY AND MEDICINE
(2022)
Article
Automation & Control Systems
Shen Liu et al.
Summary: This study introduces a Subspace Network with Shared Representation learning (SNSR) based on meta-learning for fault diagnosis under speed transient conditions with few samples. The method demonstrated superior performance in bearing fault diagnosis.
Article
Engineering, Industrial
Hongru Cao et al.
Summary: This paper proposes an unsupervised domain-share convolutional neural network method for efficient fault transfer diagnosis of machines from steady speeds to time-varying speeds. By improving the efficiency and robustness of feature adaptation and simultaneously extracting domain invariant features from the source domain and target domain, the proposed method aims to improve diagnosis accuracy and transferability.
JOURNAL OF MANUFACTURING SYSTEMS
(2022)
Article
Computer Science, Artificial Intelligence
Yong Feng et al.
Summary: The recent advances in intelligent fault diagnosis demonstrate the strong capabilities of deep learning in automatic feature extraction and accurate identification of fault signals, yet challenges such as data scarcity and varying working conditions may impact model performance. The tool proposed to address these challenges is meta-learning, which quickly adapts to new tasks using small samples and has great potential in few-shot and cross-domain fault diagnosis. The lack of a survey to summarize existing work and look into the future is noted, with this paper comprehensively investigating deep meta-learning in fault diagnosis from three perspectives.
KNOWLEDGE-BASED SYSTEMS
(2022)
Article
Automation & Control Systems
Yong Feng et al.
Summary: In this paper, a semi-supervised meta-learning network with attention mechanism is proposed for few-shot fault diagnosis in mechanical systems. The method utilizes unlabeled data to improve fault recognition and achieves outstanding adaptability in different situations, as demonstrated through experiments with bearing vibration datasets.
Article
Engineering, Mechanical
Jacob Hendriks et al.
Summary: This paper investigates the use of the CWRU bearing dataset for benchmarking CNNs in a domain shift problem. It identifies a potential flaw in the accepted procedure and proposes an alternative benchmarking framework. The results indicate that the original framework allows CNNs to learn features related to specific bearings and may not be able to generalize for different bearings. Additionally, using existing state-of-the-art deep CNNs from other fields may be a more efficient option when large fault datasets are unavailable.
MECHANICAL SYSTEMS AND SIGNAL PROCESSING
(2022)
Article
Engineering, Mechanical
Hao Su et al.
Summary: This paper proposes a novel method for bearing fault diagnosis with small samples under different working conditions. The method includes data reconstruction and meta-learning stages to extract useful information and train the model using a recurrent meta-learning strategy. Experimental results demonstrate the effectiveness of this method in intelligent bearing fault diagnosis.
MECHANICAL SYSTEMS AND SIGNAL PROCESSING
(2022)
Article
Engineering, Electrical & Electronic
Long Tian et al.
Summary: Learning from multi-modality high-resolution range profiles (HRRPs) in radar target recognition is a challenge. This work proposes a multi-modality prototypical network (MMPN) that learns a modality-aware network using metric learning and FiLM layers, addressing the problem of rapid learning from a few HRRPs.
Article
Engineering, Multidisciplinary
Tianyuan Yang et al.
Summary: This paper proposes a novel cross-domain fault diagnosis method based on MAML to address the challenges of highly variable working conditions and limited data in industrial scenarios. Through data preprocessing and optimized training strategies, the model achieves improved accuracy and generalization performance.
Article
Computer Science, Artificial Intelligence
Chu Han et al.
Summary: Cells/nuclei deliver massive information of microenvironment. An automatic nuclei segmentation approach can reduce pathologists' workload and allow precise investigation of the microenvironment for biological and clinical research. This paper proposes a generalized nuclei segmentation model with less data dependency and more generalizability. The model combines meta multi-task learning and contour-aware multi-task learning to improve model generalization under limited training samples.
MEDICAL IMAGE ANALYSIS
(2022)
Article
Computer Science, Artificial Intelligence
Yuhong Jin et al.
Summary: This paper proposes a new method based on the Time Series Transformer for fault diagnosis of rotating machinery. The effects of structural hyperparameters on diagnosis performance are analyzed through experiments, and the feature vectors are visualized. The results show that the proposed method has better fault identification capability than traditional models.
Review
Automation & Control Systems
M. A. Ganaie et al.
Summary: This paper provides a comprehensive review of state-of-art deep ensemble models, their applications in different domains, and potential research directions.
ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE
(2022)
Article
Engineering, Mechanical
Rujie Hou et al.
Summary: In practical mechanical fault diagnosis, obtaining fault data is difficult and there is a great imbalance between normal and fault data. To address this issue, this study proposes a two-stage training approach and utilizes a contrastive-weighted self-supervised model with augmented vision transformer to handle the imbalance learning strategies. The pre-training stage effectively learns a good initialized encoder for better classification of long-tailed data in downstream tasks.
MECHANICAL SYSTEMS AND SIGNAL PROCESSING
(2022)
Article
Computer Science, Artificial Intelligence
Orhun Bugra Baran et al.
Summary: This paper investigates a key issue in few-shot learning, which is handling ambiguities resulting from having too few training samples per class. To address this problem, the authors propose a meta-learning model based on semantic knowledge, which leverages feature attention and sample attention mechanisms to improve the synthesis of classifiers.
Article
Computer Science, Interdisciplinary Applications
Xuanyuan Su et al.
Summary: This paper proposes an end-to-end framework to improve the forecasting performance of Remaining Useful Life (RUL), utilizing methods such as feature pre-extraction and adaptive transformer to enhance accuracy and efficiency in prediction.
COMPUTERS & INDUSTRIAL ENGINEERING
(2021)
Article
Computer Science, Artificial Intelligence
Zenghui An et al.
Summary: Transfer learning has potential applications in intelligent machinery fault diagnosis, but current methods fail to fully consider the unlabeled and imbalanced characteristics of real mechanical data. This study introduces a self-learning transferable neural network (STNN) with three loss terms to achieve self-belief, doubt, and rectification in predicting health conditions, demonstrating its effectiveness and superiority in experimental cases of rotating machinery with unlabeled and imbalanced data.
KNOWLEDGE-BASED SYSTEMS
(2021)
Article
Computer Science, Artificial Intelligence
Xiaoan Yan et al.
Summary: The performance of complex rotor-bearing system deteriorates with time, making it difficult to identify fault categories and severities throughout the entire life-cycle. The proposed DRVAE method utilizes deep learning to effectively improve identification accuracy and feature learning performance.
KNOWLEDGE-BASED SYSTEMS
(2021)
Article
Engineering, Mechanical
Duo Wang et al.
Summary: The article proposes a Feature Space Metric-based Meta-learning Model (FSM3) to address the challenge of few-shot fault diagnosis. Experimental results demonstrate that the method outperforms baseline methods in fault diagnosis tasks under various limited data conditions. Additionally, the time complexity and implementation difficulty have been analyzed to show the relatively high feasibility of the method.
MECHANICAL SYSTEMS AND SIGNAL PROCESSING
(2021)
Review
Engineering, Mechanical
Yaguo Lei et al.
MECHANICAL SYSTEMS AND SIGNAL PROCESSING
(2020)
Article
Automation & Control Systems
Zhibin Zhao et al.
Article
Acoustics
Chang Yan et al.
JOURNAL OF SOUND AND VIBRATION
(2020)
Article
Thermodynamics
Changchang Che et al.
ADVANCES IN MECHANICAL ENGINEERING
(2019)
Review
Engineering, Mechanical
Mariela Cerrada et al.
MECHANICAL SYSTEMS AND SIGNAL PROCESSING
(2018)
Article
Engineering, Multidisciplinary
Zong Meng et al.
Article
Engineering, Mechanical
Shao Haidong et al.
MECHANICAL SYSTEMS AND SIGNAL PROCESSING
(2017)
