Pith. sign in

REVIEW 5 major objections 7 minor 41 references

Bearing fault diagnosis with few samples reaches 90% accuracy

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

DAC-FCF combines conditional GAN data augmentation, contrastive learning, and 1D Fourier convolution to report large accuracy gains in bearing fault diagnosis when only 20 to 200 labeled samples per class are available.

T0 review reviewed 2026-08-04 challenge →

load-bearing objection Large claimed gains from a three-part few-shot bearing fault diagnosis framework, but a key ablation row exactly duplicates the full model and missing GAN/data-split details make the central augmentation claim hard to verify. the 5 major comments →

arxiv 2509.11053 v1 pith:CR7QTVTU submitted 2025-09-14 cs.LG cs.AIcs.CE

An Advanced Convolutional Neural Network for Bearing Fault Diagnosis under Limited Data

Shengke Sun , Shuzhen Han , Ziqian Luan , Xinghao Qin , Jiao Yin , Zhanshan Zhao , Jinli Cao , Hua Wang This is my paper
classification cs.LG cs.AIcs.CE MSC 68T0768T10
keywords bearing fault diagnosislimited datafew-shot learninggenerative adversarial networkcontrastive learningFourier convolutiondata augmentationvibration signals
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper proposes a framework, DAC-FCF, that combines three components to diagnose bearing faults when only tens to hundreds of labeled vibration samples per class are available: a conditional GAN that generates class-specific synthetic samples, a contrastive learning objective that pulls same-class features together, and a one-dimensional Fourier convolutional network that extracts global features. On the CWRU dataset and a self-collected two-stage gearbox bench, the framework reports average accuracies of about 85% and 80%, respectively, across sample sizes from 20 to 200 per class, outperforming conventional CNNs, transfer learning baselines, and transformers. The authors argue that the synthetic data from the conditional GAN is the largest contributor, with ablation showing up to a 50% accuracy drop when it is removed.

Core claim

The central claim is that under limited labeled data, a carefully stabilized conditional GAN can supply enough class-specific synthetic vibration samples to make a deep classifier effective, and that a contrastive pairing loss plus a Fourier-domain global path further sharpen the learned features. The paper demonstrates this with average accuracy gains of up to 32 percentage points over a domain-adaptation baseline on CWRU and 10 points on its own test bench, while showing that removing the augmentation collapses performance.

What carries the argument

The load-bearing components are three. First, CCLR-GAN, a conditional variant of a consistent-latent-representation GAN, injects label embeddings into each upsampling stage of the generator through cascade cross-attention, aiming to produce diverse, fault-specific synthetic samples. Second, a contrastive loss treats pairs of samples with the same label as positive and pairs with different labels as negative, using cosine similarity, to compact the feature space. Third, a 1D-FCNN splits the input into a local convolution path and a global Fourier path, using adaptive strides so that the model sees both time-domain details and global frequency structure.

Load-bearing premise

The method assumes a GAN trained on 20 to 200 labeled samples per class can generate 500 additional class-specific samples that lie on the true distribution of each fault type, and that adding these synthetic samples improves rather than biases the classifier.

What would settle it

Measure the distribution distance between generated and real vibration samples per class (e.g., FID or a statistical two-sample test) or retrain with augmentation removed on a third bearing dataset; if generated samples are far from real data or the accuracy drop vanishes, the central claim fails.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • At 20 samples per class, the framework keeps CWRU accuracy above 74% while all tested baselines fall below 45%.
  • Removing the CCLR-GAN augmentation causes up to a 50% drop in accuracy, indicating synthetic data is the main driver.
  • Contrastive learning and the Fourier global path each improve accuracy on both datasets in ablation.
  • The proposed CCLR-GAN maintains generator-discriminator balance and avoids mode collapse better than a standard DCGAN, especially at small sample sizes.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the method scales to other 1D sensor diagnostics (e.g., gearboxes, pumps), the Fourier global path may be the portable insight, since spectral structure is a general property of rotating machinery signals.
  • The strong dependence on GAN-generated data suggests the reported numbers should be read as conditional on the synthetic samples being distributionally faithful; a validation of generated-sample quality (e.g., distribution distance to real samples) would strengthen the claim.
  • The contrastive pairing assumes label-identical samples share feature similarity; for noisy vibration waveforms this could over-constrain intra-class variance, a risk that grows as sample size shrinks.
  • A direct test of the framework on a third dataset with different sensor placement or operating speeds would clarify whether the gains are specific to the two tested benches.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

5 major / 7 minor

Summary. The paper proposes DAC-FCF, a framework for bearing fault diagnosis with very few labeled vibration samples. It combines three mechanisms: CCLR-GAN, a conditional variant of CLR-GAN that uses cascade cross-attention to inject label information into the generator; a contrastive-learning objective on feature pairs; and a 1D adaptation of Fourier convolution networks for global feature extraction. Experiments on CWRU and a self-collected gearbox dataset at 20–200 samples per class compare against several baselines and report large gains (e.g., 74.02% vs 41.64% for DDC at 20 samples/class on CWRU). Ablation and GAN-stability experiments are also presented.

Significance. If the reported results are reliable, the paper addresses an important practical problem and the reported gains are substantial. The headline comparison tables are internally consistent with the claimed improvements, and the Fourier-global argument is a standard, correctly applied result. However, the central evidence depends on the quality of GAN-generated samples under extreme data scarcity, and the manuscript does not yet provide the details needed to verify or reproduce this. The duplication in Table 6 also weakens the ablation narrative. The potential significance is high, but the current evidence is incomplete.

major comments (5)
  1. [§4.1–4.3] The train/test partitioning is never described. It is stated that 1024-point non-overlapping windows are extracted from each dataset, but it is not specified whether train and test windows are disjoint recordings, disjoint time segments, or random splits within the same recording. With vibration signals, even non-overlapping windows from the same continuous recording can be highly correlated; if train and test are drawn from the same files, the accuracies in Tables 4 and 5 could reflect temporal correlation rather than generalization. The sample sizes (20–200 per class) make the split construction especially consequential. Please specify exactly how training and test sets were generated, including stratification across operating conditions and whether the GAN was trained only on the training split.
  2. [§3.2, Eq. (11); §4.1–4.2] CCLR-GAN is load-bearing: Section 4.4 reports a drop of up to 50% without augmentation. Yet the manuscript gives no GAN training protocol (number of iterations, convergence criterion, early stopping, learning-rate schedule) and no validation of synthetic-sample fidelity or diversity (e.g., nearest-neighbor distance, FID-like measures, visual inspection, mode-collapse diagnostics). Training a GAN on 20–200 real samples per class to generate 500 samples per class is precisely the regime where memorization and mode collapse are expected. In addition, Eq. (11) is dimensionally unclear: F_i v^T / sqrt(d_v) mixes a feature map with a label embedding without defining the projections or the softmax axis, so the conditioning mechanism is not reproducible as written.
  3. [Table 6 vs Table 4; §4.4] The row labeled 'CCLR-GAN+FCNN' in Table 6 is numerically identical to the full DAC-FCF row in Table 4 (74.02, 84.97, 87.73, 87.96, 90.04, 84.94). Since DAC-FCF includes contrastive learning in addition to CCLR-GAN and 1D-FCNN, this identity implies either that contrastive learning has zero effect in the final configuration or that the table entry is a copy-paste error. As written, it directly contradicts the claim in Section 4.4 that the contrastive and Fourier components are 'also important.' Please correct the table or provide the corresponding results without contrastive learning.
  4. [§4.4, Fig. 8] The ablation study is presented only as a figure, without a table of numeric accuracies, error bars, or exact configurations. The text states that removing augmentation causes 'a 50% drop at most,' but the reader cannot verify this from the text. Given that this ablation is one of the three stated contributions, numeric results for each removed component at each sample size should be included.
  5. [Tables 4 and 5] All accuracy numbers are reported without variance, confidence intervals, or number of repeated runs. At 20–50 samples per class, differences between methods can be sensitive to the random split and GAN initialization. For example, on the self-collected dataset, the 63.24% vs 53.80% advantage over DDC at 20 samples/class is about 10 points; without repeated trials it is unclear whether this is significant. Please report mean and standard deviation over multiple random seeds/splits.
minor comments (7)
  1. [§3, contribution 3] Typo: '1D-FCCN' should be '1D-FCNN'.
  2. [§2.3 and §3.2] Typos: 'contrastive learing' in the Section 2.3 heading; 'conditonal CLR-GAN' in Section 3.2; Section 2.1 has garbled text 'y train ∈R R includesRhealth states'.
  3. [Eq. (15)] Cosine similarity Sij can be negative or equal to 1; the log(1−Sij) term is undefined at Sij=1 and ill-suited for negative similarities. Use a bounded similarity or clarify clamping.
  4. [Table 6] The parenthetical improvements (e.g., '+56.89') are not defined in the caption; specify they are relative to DCGAN+FCNN.
  5. [§2.4] The statement that Katznelson [2004] proves the specific global-perturbation claim is inaccurate; the book is a general harmonic-analysis reference. The derivation in Eqs. (8)–(10) is sufficient.
  6. [Abstract] The abstract claims 'up to 32%' and '10%' but does not specify the baseline and sample size; please state these (e.g., CWRU, 20 samples/class, vs DDC).
  7. [Fig. 9] The loss curves would benefit from axis labels and a description of how 'fairness' is quantified.

Circularity Check

0 steps flagged

No significant circularity: the reported accuracy gains are benchmark measurements rather than derived predictions, and the single self-citation (CLR-GAN) is a published external component used as a building block, not as the proof of the paper's claims.

full rationale

I find no equation-level or definitional circularity. The central claim ('DAC-FCF achieves significant improvements, outperforming baselines by up to 32% on CWRU and 10% on a self-collected test bench') is supported by measured accuracies in Tables 4 and 5, not by a derivation that reduces to the method's own definitions. The Fourier global-awareness argument in Section 2.4 is a standard theorem illustrated with a Dirac-delta perturbation; it is not a circular prediction. The conditional injection in Eq. (11) is a cross-attention construction, not a self-definition. The contrastive loss in Eq. (15) is a standard formulation rather than a renamed target result. The paper does cite the authors' own CLR-GAN (Sun et al., 2025) as the basis for CCLR-GAN, but this is a background architectural reference, and the reported gains come from experiments that compare against an external GAN (DCGAN) in Table 6; the central claim is not forced by that self-citation. Several reporting weaknesses exist—no train/test split description in Section 4.1/4.2, no distributional validation of generated samples, and an unexplained duplicate numerical row in Table 6—but these are correctness/validation risks, not circularity. The manuscript itself acknowledges GAN scalability limits in Section 5, which also does not indicate circularity. The honest finding is therefore: no central derivation reduces to its inputs; the minor self-citation is not load-bearing.

Axiom & Free-Parameter Ledger

3 free parameters · 4 axioms · 0 invented entities

The proposed method assumes, without independent validation, that a GAN trained on extremely few samples produces class-conditional synthetic data that improves classifiers, and that the evaluation protocol has no leakage. It also relies on the standard Fourier global-support property and on standard optimizer assumptions. The hand-set loss weights, augmentation count, and stride schedule are not justified by sensitivity analysis.

free parameters (3)
  • lambda_1, lambda_2 (CCLR-GAN loss weights) = not reported
    Eqs. (9)-(10) introduce two weighting coefficients for the consistency and reconstruction losses; no values or sensitivity study are given, so their contribution to the reported gains cannot be assessed.
  • Number of generated samples per class = 500
    Sec. 4.2 sets 500 generated samples per class for all experiments; no ablation over augmentation amount is reported, so the headline accuracy depends on an arbitrary hand-chosen quantity.
  • Adaptive stride schedule in 1D-FCNN = not specified
    Sec. 3.4 says the global path first uses large strides then small strides, but no kernel sizes or stride values are given, so the architecture cannot be reproduced or isolated.
axioms (4)
  • domain assumption A GAN trained on as few as 20 labeled samples per class can learn the conditional distribution P(x|y) well enough that 500 generated samples per class improve a downstream classifier.
    The whole data augmentation component rests on this empirical premise. The paper reports no distributional validation of generated samples and no sensitivity analysis for the 500-sample augmentation count (Secs. 3.2, 4.2, 4.4).
  • domain assumption The training and test split is contamination-free: the CCLR-GAN and the classifier see only the limited training samples, and the test set is never used to select augmentation hyperparameters.
    The paper never states how train/test splits are constructed at each sample size or whether the GAN is trained on the full dataset. If leakage occurred, the reported gains would be spurious.
  • domain assumption A point perturbation in the frequency domain changes the whole time-domain signal, therefore frequency-domain convolution yields global awareness and better limited-data diagnosis.
    The Fourier property in Sec. 2.4 is standard math, but the causal step from 'Fourier transform is global' to 'Fourier convolution improves fault diagnosis' is an empirical assumption tested only through accuracy tables.
  • domain assumption The optimizer and hyperparameter choices (Adam, lr 1e-4/2e-4, batch sizes) from Sec. 4.2 are sufficient for the CCLR-GAN and classifier to converge.
    No convergence analysis or hyperparameter search is reported; the architecture assumes these standard settings work.

reviewed 2026-08-04 · how reviews work

0 comments
Cite this review

Pith. "Pith review of An Advanced Convolutional Neural Network for Bearing Fault Diagnosis under Limited Data." pith.science (2026). https://pith.science/paper/CR7QTVTU

@misc{pith2026250911053,
  author       = {Pith},
  title        = {Pith review of: An Advanced Convolutional Neural Network for Bearing Fault Diagnosis under Limited Data},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CR7QTVTU}},
  note         = {Machine review of arXiv:2509.11053}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

In the area of bearing fault diagnosis, deep learning (DL) methods have been widely used recently. However, due to the high cost or privacy concerns, high-quality labeled data are scarce in real world scenarios. While few-shot learning has shown promise in addressing data scarcity, existing methods still face significant limitations in this domain. Traditional data augmentation techniques often suffer from mode collapse and generate low-quality samples that fail to capture the diversity of bearing fault patterns. Moreover, conventional convolutional neural networks (CNNs) with local receptive fields makes them inadequate for extracting global features from complex vibration signals. Additionally, existing methods fail to model the intricate relationships between limited training samples. To solve these problems, we propose an advanced data augmentation and contrastive fourier convolution framework (DAC-FCF) for bearing fault diagnosis under limited data. Firstly, a novel conditional consistent latent representation and reconstruction generative adversarial network (CCLR-GAN) is proposed to generate more diverse data. Secondly, a contrastive learning based joint optimization mechanism is utilized to better model the relations between the available training data. Finally, we propose a 1D fourier convolution neural network (1D-FCNN) to achieve a global-aware of the input data. Experiments demonstrate that DAC-FCF achieves significant improvements, outperforming baselines by up to 32\% on case western reserve university (CWRU) dataset and 10\% on a self-collected test bench. Extensive ablation experiments prove the effectiveness of the proposed components. Thus, the proposed DAC-FCF offers a promising solution for bearing fault diagnosis under limited data.

Figures

Figures reproduced from arXiv: 2509.11053 by Hua Wang, Jiao Yin, Jinli Cao, Shengke Sun, Shuzhen Han, Xinghao Qin, Zhanshan Zhao, Ziqian Luan.

Figure 1
Figure 1. Figure 1: The flowchart of a conventional GAN. 2.3 Deep Contrastive Learning Extracting recognizable feature representations from data is a fundamental task in deep learning. Many researchers have proposed that there is a strong positive correlation between the model’s feature extraction ability and its performance. Among various training paradigms, contrastive learning has been proved to have a stronger ability to … view at source ↗
Figure 2
Figure 2. Figure 2: The feature space before (left) and after (right) the contrastive learning. [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: The overall architecture of DAC-FCF 3.2 Conditional Consistent-Latent-Representation Generative Adversarial Networks As we mentioned in Sec.2.2, in the conventional training paradigm GANs, the discriminator always trains better than the generator, making the gradient for the generator vanish. In practice, the discriminator can easily identify the fake samples from a relatively early stage of the training a… view at source ↗
Figure 4
Figure 4. Figure 4: The overall architecture of CCLR-GAN unconditional generative model, has a significant limitation: it can only generate samples according to the overall data distribution P(x), without considering the conditional distribution P(x | y), where y represents the fault labels. To enable CLR-GAN to generate fault-specific samples, we proposed a conditonal CLR-GAN (CCLR￾GAN) that injects the label information int… view at source ↗
Figure 5
Figure 5. Figure 5: The overall structure of Fourier Convolutional Neural Network. [PITH_FULL_IMAGE:figures/full_fig_p010_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Experimental environment of the self-collected test bench. [PITH_FULL_IMAGE:figures/full_fig_p012_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Per-class diagnosis accuracy of DDC (left) and the proposed DAC-FCF (right). [PITH_FULL_IMAGE:figures/full_fig_p013_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Accuracy with/without CCLR-GAN, contrastive learning, and Fourier Convolution under [PITH_FULL_IMAGE:figures/full_fig_p014_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: The training loss of the proposed CCLR-GAN and DC-GAN [PITH_FULL_IMAGE:figures/full_fig_p015_9.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

41 extracted references · 2 canonical work pages

  1. [1]

    Multiscale convolutional neural network with feature alignment for bearing fault diagnosis

    Junbin Chen, Ruyi Huang, Kun Zhao, Wei Wang, Longcan Liu, and Weihua Li. Multiscale convolutional neural network with feature alignment for bearing fault diagnosis. IEEE Transactions on Instrumentation and Measurement, 70: 0 1--10, 2021

    2021

  2. [2]

    Wasserstein distance based deep adversarial transfer learning for intelligent fault diagnosis with unlabeled or insufficient labeled data

    Cheng Cheng, Beitong Zhou, Guijun Ma, Dongrui Wu, and Ye Yuan. Wasserstein distance based deep adversarial transfer learning for intelligent fault diagnosis with unlabeled or insufficient labeled data. Neurocomputing, 409: 0 35--45, 2020

    2020

  3. [3]

    Fast fourier convolution

    Lu Chi, Borui Jiang, and Yadong Mu. Fast fourier convolution. Advances in Neural Information Processing Systems, 33: 0 4479--4488, 2020

    2020

  4. [4]

    Neighbor-aware review helpfulness prediction

    Jiahua Du, Jia Rong, Hua Wang, and Yanchun Zhang. Neighbor-aware review helpfulness prediction. Decision Support Systems, 148: 0 113581, 2021

    2021

  5. [5]

    Distributed memetic algorithm for outsourced database fragmentation

    Yong Feng Ge, Wei Jie Yu, Jinli Cao, Hua Wang, Zhi Hui Zhan, Yanchun Zhang, and Jun Zhang. Distributed memetic algorithm for outsourced database fragmentation. IEEE Transactions on Cybernetics, 51 0 (10): 0 4808--4821, 2020

    2020

  6. [6]

    Mdde: multitasking distributed differential evolution for privacy-preserving database fragmentation

    Yong Feng Ge, Maria Orlowska, Jinli Cao, Hua Wang, and Yanchun Zhang. Mdde: multitasking distributed differential evolution for privacy-preserving database fragmentation. The VLDB Journal, 31 0 (5): 0 957--975, 2022 a

    2022

  7. [7]

    An information-driven genetic algorithm for privacy-preserving data publishing

    Yong Feng Ge, Hua Wang, Jinli Cao, and Yanchun Zhang. An information-driven genetic algorithm for privacy-preserving data publishing. In International Conference on Web Information Systems Engineering, pages 340--354. Springer, 2022 b

    2022

  8. [8]

    Distributed cooperative coevolution of data publishing privacy and transparency

    Yong Feng Ge, Elisa Bertino, Hua Wang, Jinli Cao, and Yanchun Zhang. Distributed cooperative coevolution of data publishing privacy and transparency. ACM Transactions on Knowledge Discovery from Data, 18 0 (1): 0 1--23, 2023 a

    2023

  9. [9]

    Evolutionary dynamic database partitioning optimization for privacy and utility

    Yong Feng Ge, Hua Wang, Elisa Bertino, Zhi Hui Zhan, Jinli Cao, Yanchun Zhang, and Jun Zhang. Evolutionary dynamic database partitioning optimization for privacy and utility. IEEE Transactions on Dependable and Secure Computing, 2023 b

    2023

  10. [10]

    Generative adversarial nets

    Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial nets. Advances in neural information processing systems, 27, 2014

    2014

  11. [11]

    Deep residual multiscale convolutional neural network with attention mechanism for bearing fault diagnosis under strong noise environment

    Shuzhen Han, Shengke Sun, Zhanshan Zhao, Ziqian Luan, and Pingjuan Niu. Deep residual multiscale convolutional neural network with attention mechanism for bearing fault diagnosis under strong noise environment. IEEE Sensors Journal, 24 0 (6): 0 9073--9081, 2024

    2024

  12. [12]

    A novel adversarial learning framework in deep convolutional neural network for intelligent diagnosis of mechanical faults

    Te Han, Chao Liu, Wenguang Yang, and Dongxiang Jiang. A novel adversarial learning framework in deep convolutional neural network for intelligent diagnosis of mechanical faults. Knowledge-based systems, 165: 0 474--487, 2019

    2019

  13. [13]

    Deep-learning-based guided wave detection for liquid-level state in porcelain bushing type terminal

    Xiaobin Hong, Bin Zhang, Yuan Liu, Hongchang Qi, and Weihua Li. Deep-learning-based guided wave detection for liquid-level state in porcelain bushing type terminal. Structural Control and Health Monitoring, 28 0 (1): 0 e2651, 2021

    2021

  14. [14]

    Cross-domain fault diagnosis of rolling element bearings using dcgan and dann

    R Hu, M Zhang, and W Xu. Cross-domain fault diagnosis of rolling element bearings using dcgan and dann. J. Vib. Shock, 41 0 (06): 0 21--29, 2022

    2022

  15. [15]

    A deep adversarial diagnosis method for wind turbine main bearing fault with imbalanced small sample scenarios

    Nantian Huang, Xuehang Yang, Guowei Cai, X Song, QZ Chen, and WG Zhao. A deep adversarial diagnosis method for wind turbine main bearing fault with imbalanced small sample scenarios. Proceedings of the CSEE, 40 0 (02): 0 563--574, 2020

    2020

  16. [16]

    Convolutional neural network based fault detection for rotating machinery

    Olivier Janssens, Viktor Slavkovikj, Bram Vervisch, Kurt Stockman, Mia Loccufier, Steven Verstockt, Rik Van de Walle , and Sofie Van Hoecke . Convolutional neural network based fault detection for rotating machinery. Journal of Sound and Vibration, 377: 0 331--345, 2016. ISSN 0022-460X. doi:https://doi.org/10.1016/j.jsv.2016.05.027. URL https://www.scienc...

    work page doi:10.1016/j.jsv.2016.05.027 2016

  17. [17]

    An introduction to harmonic analysis

    Yitzhak Katznelson. An introduction to harmonic analysis. Cambridge University Press, 2004

    2004

  18. [18]

    Imagenet classification with deep convolutional neural networks

    Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hinton. Imagenet classification with deep convolutional neural networks. In F. Pereira, C.J. Burges, L. Bottou, and K.Q. Weinberger, editors, Advances in Neural Information Processing Systems, volume 25. Curran Associates, Inc., 2012

    2012

  19. [19]

    An optimized vmd method and its applications in bearing fault diagnosis

    Hua Li, Tao Liu, Qing Chen, and Xing Wu. An optimized vmd method and its applications in bearing fault diagnosis. Measurement, 166: 0 108185, 2020 a . ISSN 0263-2241. doi:https://doi.org/10.1016/j.measurement.2020.108185

    arXiv 2020

  20. [20]

    A bearing fault diagnosis method based on enhanced singular value decomposition

    Hua Li, Tao Liu, Xing Wu, and Qing Chen. A bearing fault diagnosis method based on enhanced singular value decomposition. IEEE Transactions on Industrial Informatics, 17 0 (5): 0 3220--3230, 2021. doi:10.1109/TII.2020.3001376

    arXiv 2021

  21. [21]

    Rolling bearing fault diagnosis based on improved adaptive parameterless empirical wavelet transform and sparse denoising

    Jimeng Li, Hui Wang, Xiangdong Wang, and Yungang Zhang. Rolling bearing fault diagnosis based on improved adaptive parameterless empirical wavelet transform and sparse denoising. Measurement, 152: 0 107392, 2020 b . ISSN 0263-2241. doi:https://doi.org/10.1016/j.measurement.2019.107392. URL https://www.sciencedirect.com/science/article/pii/S026322411931259X

    arXiv 2020

  22. [22]

    Case western reserve university bearing data center

    Kenneth Loparo. Case western reserve university bearing data center. Bearings Vibration Data Sets, Case Western Reserve University, pages 22--28, 2012

    2012

  23. [23]

    Domain adaptation via transfer component analysis

    Sinno Jialin Pan, Ivor W Tsang, James T Kwok, and Qiang Yang. Domain adaptation via transfer component analysis. IEEE transactions on neural networks, 22 0 (2): 0 199--210, 2010

    2010

  24. [24]

    Rafiee, F

    J. Rafiee, F. Arvani, A. Harifi, and M.H. Sadeghi. Intelligent condition monitoring of a gearbox using artificial neural network. Mechanical Systems and Signal Processing, 21 0 (4): 0 1746--1754, 2007. ISSN 0888-3270. doi:https://doi.org/10.1016/j.ymssp.2006.08.005. URL https://www.sciencedirect.com/science/article/pii/S0888327006001750

    work page doi:10.1016/j.ymssp.2006.08.005 2007

  25. [25]

    Clr-gan: Improving gans stability and quality via consistent latent representation and reconstruction

    Shengke Sun, Ziqian Luan, Zhanshan Zhao, Shijie Luo, and Shuzhen Han. Clr-gan: Improving gans stability and quality via consistent latent representation and reconstruction. In Ale s Leonardis, Elisa Ricci, Stefan Roth, Olga Russakovsky, Torsten Sattler, and G \"u l Varol, editors, Computer Vision -- ECCV 2024, pages 210--227, Cham, 2025. Springer Nature S...

    2024

  26. [26]

    Deep domain confusion: Maximizing for domain invariance

    Eric Tzeng, Judy Hoffman, Ning Zhang, Kate Saenko, and Trevor Darrell. Deep domain confusion: Maximizing for domain invariance. arXiv preprint arXiv:1412.3474, 2014

    Pith/arXiv arXiv 2014

  27. [27]

    A high-accuracy intelligent fault diagnosis method for aero-engine bearings with limited samples

    Zhenya Wang, Qiusheng Luo, Hui Chen, Jingshan Zhao, Ligang Yao, Jun Zhang, and Fulei Chu. A high-accuracy intelligent fault diagnosis method for aero-engine bearings with limited samples. Computers in Industry, 159-160: 0 104099, 2024. ISSN 0166-3615

    2024

  28. [28]

    Bearing fault diagnosis via fusing small samples and training multi-state siamese neural networks

    Chuanbo Wen, Yipeng Xue, Weibo Liu, Guochu Chen, and Xiaohui Liu. Bearing fault diagnosis via fusing small samples and training multi-state siamese neural networks. Neurocomputing, 576: 0 127355, 2024. ISSN 0925-2312

    2024

  29. [29]

    Research on bearing fault diagnosis method based on transformer neural network

    Zhuohong Yang, Jian Cen, Xi Liu, Jianbin Xiong, and Honghua Chen. Research on bearing fault diagnosis method based on transformer neural network. Measurement Science and Technology, 33 0 (8): 0 085111, may 2022

    2022

  30. [30]

    Sspenet: Semi-supervised prototype enhancement network for rolling bearing fault diagnosis under limited labeled samples

    Xuejian Yao, Xingchi Lu, Quansheng Jiang, Yehu Shen, Fengyu Xu, and Qixin Zhu. Sspenet: Semi-supervised prototype enhancement network for rolling bearing fault diagnosis under limited labeled samples. Advanced Engineering Informatics, 61: 0 102560, 2024. ISSN 1474-0346

    2024

  31. [31]

    Knowledge-driven cybersecurity intelligence: Software vulnerability coexploitation behavior discovery

    Jiao Yin, MingJian Tang, Jinli Cao, Mingshan You, Hua Wang, and Mamoun Alazab. Knowledge-driven cybersecurity intelligence: Software vulnerability coexploitation behavior discovery. IEEE transactions on industrial informatics, 19 0 (4): 0 5593--5601, 2022

    2022

  32. [32]

    Empowering vulnerability prioritization: A heterogeneous graph-driven framework for exploitability prediction

    Jiao Yin, Guihong Chen, Wei Hong, Hua Wang, Jinli Cao, and Yuan Miao. Empowering vulnerability prioritization: A heterogeneous graph-driven framework for exploitability prediction. In International Conference on Web Information Systems Engineering, pages 289--299. Springer, 2023

    2023

  33. [33]

    Tlef: Two-layer evolutionary framework for t-closeness anonymization

    Mingshan You, Yong Feng Ge, Kate Wang, Hua Wang, Jinli Cao, and Georgios Kambourakis. Tlef: Two-layer evolutionary framework for t-closeness anonymization. In International Conference on Web Information Systems Engineering, pages 235--244. Springer, 2023 a

    2023

  34. [34]

    A knowledge graph empowered online learning framework for access control decision-making

    Mingshan You, Jiao Yin, Hua Wang, Jinli Cao, Kate Wang, Yuan Miao, and Elisa Bertino. A knowledge graph empowered online learning framework for access control decision-making. World Wide Web, 26 0 (2): 0 827--848, 2023 b

    2023

  35. [35]

    A new deep learning model for fault diagnosis with good anti-noise and domain adaptation ability on raw vibration signals

    Wei Zhang, Gaoliang Peng, Chuanhao Li, Yuanhang Chen, and Zhujun Zhang. A new deep learning model for fault diagnosis with good anti-noise and domain adaptation ability on raw vibration signals. Sensors, 17 0 (2): 0 425, 2017

    2017

  36. [36]

    A deep convolutional neural network with new training methods for bearing fault diagnosis under noisy environment and different working load

    Wei Zhang, Chuanhao Li, Gaoliang Peng, Yuanhang Chen, and Zhujun Zhang. A deep convolutional neural network with new training methods for bearing fault diagnosis under noisy environment and different working load. Mechanical systems and signal processing, 100: 0 439--453, 2018

    2018

  37. [37]

    Fault diagnosis of rotating machinery based on recurrent neural networks

    Yahui Zhang, Taotao Zhou, Xufeng Huang, Longchao Cao, and Qi Zhou. Fault diagnosis of rotating machinery based on recurrent neural networks. Measurement, 171: 0 108774, 2021. ISSN 0263-2241. doi:https://doi.org/10.1016/j.measurement.2020.108774

    arXiv 2021

  38. [38]

    Transfer fault diagnosis of bearing installed in different machines using enhanced deep auto-encoder

    He Zhiyi, Shao Haidong, Jing Lin, Cheng Junsheng, and Yang Yu. Transfer fault diagnosis of bearing installed in different machines using enhanced deep auto-encoder. Measurement, 152: 0 107393, 2020

    2020

  39. [39]

    Intelligent manufacturing in the context of industry 4.0: a review

    Ray Y Zhong, Xun Xu, Eberhard Klotz, and Stephen T Newman. Intelligent manufacturing in the context of industry 4.0: a review. Engineering, 3 0 (5): 0 616--630, 2017

    2017

  40. [40]

    A residual denoising and multiscale attention-based weighted domain adaptation network for tunnel boring machine main bearing fault diagnosis

    Tao Zhong, ChengJin Qin, Gang Shi, ZhiNan Zhang, JianFeng Tao, and ChengLiang Liu. A residual denoising and multiscale attention-based weighted domain adaptation network for tunnel boring machine main bearing fault diagnosis. Science China Technological Sciences, 67 0 (8): 0 2594--2618, 2024

    2024

  41. [41]

    A bearing fault diagnosis method based on l1 regularization transfer learning and lstm deep learning

    Daiie Zhu, Xudong Song, Jie Yang, Yuyang Cong, and Lijuan Wang. A bearing fault diagnosis method based on l1 regularization transfer learning and lstm deep learning. In 2021 IEEE International Conference on Information Communication and Software Engineering (ICICSE), pages 308--312. IEEE, 2021

    2021

This paper was first reviewed by deepseek-v4-flash on August 4, 2026.