Pith. sign in

REVIEW 4 major objections 6 minor 37 references

FE-MCFormer: a novel time-frequency interpretable architecture for machinery fault diagnosis under strong noise environments

T0 review · 4 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

Pith's one-line read FE-MCFormer proposes a learnable Fourier reconstruction layer plus multiscale time-frequency fusion to keep rotating-machinery fault diagnosis accurate and interpretable at signal-to-noise ratios down to -10 dB, outperforming seven…

desk verdict Solid engineering paper with a load-bearing metric problem: Eq. (19) defines accuracy as precision, so the headline accuracy tables are unverifiable. read the letter →

arxiv 2505.06285 v3 pith:QISOHWBR submitted 2025-05-07 eess.SP cs.CV

classification eess.SPcs.CV
keywords faultdiagnosisrotatingmachinerytime-frequencytransformerFourieradaptivereconstructionnoiserobustnessinterpretabilitymultiscaleconvolutionvibrationsignalprocessing
verification ladder T0 review T1 audit T2 compute T3 formal

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 claims that fault diagnosis of rotating machinery under strong noise can be made both accurate and interpretable by coupling a learnable frequency-domain reconstruction step with a multiscale time-frequency transformer. It proposes FE-MCFormer, whose Fourier adaptive reconstruction embedding layer (FAREL) multiplies the signal's complex spectrum by a learned weight, suppresses noise-dominated frequency bins, and maps the result back to time. On three datasets — a rolling bearing benchmark, a gearbox benchmark, and a real centrifugal compressor — the method reports higher accuracy than seven deep-learning baselines at signal-to-noise ratios down to -10 dB, with best bearing accuracy 99.11% at -2 dB and 94.74% at -6 dB. The authors also argue FAREL works as a plug-and-play module, improving a WDCNN baseline whenever added.

What carries the argument

The load-bearing mechanism is FAREL (Fourier adaptive reconstruction embedding layer), a learnable spectral gate: after a $1\times 63$ convolution, the signal is transformed with the discrete Fourier transform, multiplied element-wise by a learnable complex weight $W$, transformed back with the inverse DFT, and added back to the original feature map with a scaling factor $\gamma$. This makes the network learn which frequency bins carry fault information and which carry noise. The same Fourier adaptive reconstruction appears inside the time-frequency fusion network (TFFN), and the multiscale time-frequency fusion (MSTFF) module's multiscale convolutional attention layer (MSCAL) supplies the local time-domain branch with $1\times 3$ and $1\times 5$ kernels, softmax attention, and residual connections; distillation layers halve the time dimension to widen the receptive field.

What would settle it

Recompute every reported result with the standard accuracy formula $\frac{TP+TN}{TP+TN+FP+FN}$ on the same test splits; if the corrected percentages fall substantially or the method no longer leads at low SNR, the central claim fails. A second, complementary test is to run FE-MCFormer on field vibration data with colored or non-stationary noise; if its margin over baselines disappears, the robustness claim is an artifact of white-noise injection.

Watch

Extended reading notes

Core claim

FE-MCFormer establishes that a transformer for machinery fault diagnosis does not need conventional multi-head self-attention to excel under heavy noise. The paper's central discovery is that replacing multi-head self-attention with a multiscale convolutional attention layer and inserting Fourier adaptive reconstruction at the embedding and feed-forward stages lets the network learn which frequency components carry fault information: it attenuates noise-dominated bins, preserves fault harmonics, and concentrates attention on impulsive fault regions. The evidence is the reported accuracy at SNR values from -2 to -10 dB on the bearing and gearbox datasets, plus feature-quality metrics and t-SNE and heatmap visualizations showing tighter clustering and more focused attention than baselines. On the real compressor data, the method reports 96.89% mean accuracy despite only 240 samples per class.

Load-bearing premise

The headline percentages are only as strong as the accuracy metric in Eq. (19), which defines Acc as $TP/(TP+FP)$ — precision, not accuracy — and the noise-robustness results are only as strong as the assumption that Gaussian white noise added via Eq. (20) represents real industrial strong noise.

Editorial extensions

If this is right

  • If the reported results hold, FE-MCFormer is the best among the compared methods on all three datasets at every SNR from -2 dB to -10 dB, with the largest gains at the lowest signal-to-noise ratios.
  • FAREL is claimed to be a general plug-and-play component: adding it to a WDCNN baseline raised mean accuracy by 4.31 to 9.92 percentage points across SNR levels on the bearing dataset.
  • The method maintains usable accuracy in extreme noise, for example 72.99% mean accuracy on the bearing dataset at -10 dB SNR, where the strongest transformer baseline reaches 60.65%.
  • On the small-sample real compressor dataset, FE-MCFormer reports 96.89% mean accuracy, suggesting the architecture transfers to field data with limited training samples.
  • The interpretability claims imply the model's attention and reconstructed spectra can be read as physical evidence: noise bands are attenuated and harmonics at $1\times f$, $2\times f$, and $3\times f$ are enhanced.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Although the paper only demonstrates FAREL as a plug-in for WDCNN, the same mechanism could be inserted into other CNN or transformer backbones; a systematic study across architectures would establish how general the improvement is.
  • The learned frequency mask that FAREL produces could be extracted and applied as a standalone pre-processing denoiser before any classifier, which the paper hints at but does not formalize.
  • Because all noise injections are Gaussian white noise, a direct test would add colored or non-stationary noise from variable-speed machinery; the 'strong noise' claim is only shown for white-noise contamination.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper proposes FE-MCFormer, a transformer-inspired architecture for rotating machinery fault diagnosis under strong noise. It introduces a Fourier Adaptive Reconstruction Embedding Layer (FAREL) that performs learnable spectral filtering, and a Multiscale Time-Frequency Fusion (MSTFF) module combining multiscale convolutional attention with frequency-domain reconstruction. The method is evaluated on three datasets (PU bearing, SEU gearbox, and a real-world centrifugal compressor dataset) with Gaussian white noise added at SNR levels from -2 to -10 dB, compared against seven deep learning baselines, and is supported by ablation studies and interpretability visualizations. The reported results indicate consistent accuracy improvements, particularly from the FAREL component, which is also presented as a plug-and-play enhancement for existing CNN models.

Significance. If the reported results are accurate, the paper makes a useful empirical contribution: a novel architecture that improves noise robustness, and a plug-and-play component (FAREL) that consistently improves existing models such as WDCNN. The experiments are extensive, include a real-world dataset, and the architecture is described in sufficient detail for reproduction. The inclusion of the FAREL ablation with WDCNN-FAREL provides a falsifiable claim about the component's generality. However, the central metric-definition issue and the lack of variance reporting currently prevent full verification of the quantitative claims, and the interpretability evidence is post-hoc rather than independent verification of physical mechanisms.

major comments (4)
  1. [Section 3.1, Eq. (19)] The evaluation metric is defined as Acc = TP/(TP+FP) × 100%, which is precision, not accuracy. All reported mean-acc and max-acc values in Tables 3, 4, 6, and 9, as well as the abstract headline numbers, depend on this definition. In the multi-class settings of this paper (18 classes in Case 1, 5 classes in Cases 2 and 3), accuracy should be computed as (TP+TN)/(TP+TN+FP+FN). Unless the implementation actually used a correct accuracy formula, the claimed superiority over the seven baselines is unverified. Please correct the definition or provide code, per-class confusion matrices, or TP/FP counts to resolve this discrepancy.
  2. [Section 3.1, experimental settings] Only the mean and maximum over five repeated runs are reported, with no standard deviation, confidence intervals, or significance tests. Several comparative margins are small at high SNR (e.g., Table 3, SNR=-2: FE-MCFormer 99.11% vs. Li-convformer 97.53%; Table 6, SNR=-8: 98.52% vs. 97.86%). Without variance information or significance testing, the claim of consistent superiority is not statistically supported. Please report per-run results, error bars, or perform statistical tests for the key comparisons.
  3. [Section 3.3.2, after Table 6] The text states that 'FE-MCFormer achieves diagnosis accuracies of 100%, 100%, 94.93%, 98.83%, and 95.51% in five trials,' but Table 6 lists mean accuracies of 95.51%, 98.52%, 99.79%, 100%, and 100% for SNR values from -10 dB to -2 dB. The five numbers in the text do not match the table entries, creating an internal inconsistency in the reporting of the main Case 2 result.
  4. [Section 3.2.3, Table 4] The claim that 'smaller scaling factors γ achieve the better performance across all noise levels' is contradicted by Table 4, where FE-MCFormer-0.2 (73.58%) and FE-MCFormer-0.3 (86.41%) outperform FE-MCFormer-0.1 (72.99% and 86.31%) at SNR=-10 and SNR=-8, respectively. The text acknowledges these exceptions but still draws the opposite conclusion, so the sensitivity analysis interpretation needs to be reconciled with the reported data.
minor comments (6)
  1. [Section 2.1, after Eq. (5)] The sentence fragment 'various tasks and data properties' appears without a verb; please rewrite the sentence.
  2. [Sections 2.2.1 and 3.2.3] The abbreviation 'MSA' is used to refer to the multiscale attention mechanism, while the Introduction uses 'MHSA' for multi-head self-attention; please disambiguate these terms.
  3. [Table 7] The table header contains formatting artifacts ('Methodsssssss', '𝐽1ssssss𝐽2'); please clean up the table formatting.
  4. [Section 3.4.1] The phrase 'the 7200 and 3300 eddy current probes' should specify what these numbers refer to (e.g., model numbers or measurement ranges).
  5. [References] Reference [3] has a publication year of 2026, which is in the future; please verify the citation details.
  6. [Abstract and Section 3.2.2] The abstract claims 'stable and interpretable diagnostic performance ... down to -10 dB SNR', but in Case 1 the mean accuracy at -10 dB is 72.99%, which is substantially lower than at higher SNR; please clarify what is meant by 'stable'.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; the central claims rest on held-out empirical tests, with Eq. (19) a correctness risk rather than a derivational circle.

full rationale

The claimed derivations in this paper are architectural and empirical, not first-principles predictions. FAREL and TFFN are defined by Eqs. (1)-(16) as convolutional/DFT operations with learnable weights W; no statement in the paper asserts that these equations produce the reported accuracies by construction. The accuracy numbers in Tables 3, 6, and 9 are held-out test results on external benchmarks (PU bearing, SEU gearbox, real compressor) with fixed train/test splits, and the WDCNN-FAREL comparison is an independent empirical ablation on the same held-out protocol. There is no load-bearing self-citation chain: Refs. [15] and [16] overlap with the authors but are cited only as background examples of multiscale attention networks, not as justification of the proposed method or as uniqueness theorems. The interpretability figures use the trained model's own reconstructions and attention maps; that is post-hoc introspection, not a derivation that assumes what it claims to show. One serious correctness risk, but not a circularity, is Eq. (19), which literally defines 'accuracy' as TP/(TP+FP), i.e., precision. If the implementation followed Eq. (19), then all 'mean-acc' and 'max-acc' values are precision values and the headline accuracy claims are mislabeled; this cannot be resolved without code or corrected metrics. This is a definitional and verification flaw in the evaluation, not a case of the model's output being equivalent to its input by construction. Hence the circularity score is 0.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The central claim rests on tuned constants (gamma and architecture choices), on modeling assumptions about noise and frequency-localized fault signatures, and on a misdefined evaluation metric. There are no invented physical entities. The count of free parameters is modest, but the metric error is a significant unaccounted assumption.

free parameters (3)
  • gamma (scaling factor) = 0.1
    Controls the contribution of the reconstructed frequency-domain signal in Eq. (5) and Eq. (14). Selected by sweeping gamma on the Case 1 test set (Table 4) and fixed for all cases; performance is sensitive to this value, so it is a tuned constant.
  • Architecture hyperparameters (kernel sizes, channel widths, number of MSTFF blocks) = 63-tap embedding kernel, 1x3 and 1x5 kernels, 4 blocks, channels 32/64/128/256
    Listed in Table 1 without ablations or derivation; these hand-chosen values affect capacity and receptive field but are not justified.
  • SNR range = -10, -8, -6, -4, -2 dB
    Experimental choice for the noise conditions; no physical basis or link to the actual noise level of the compressor dataset is given.
assumptions (5)
  • domain assumption Fault-related periodic impulsive characteristics correspond to identifiable frequency components that can be preserved while noise components are suppressed.
    Underlies the FAREL and TFFN designs (Sections 2.1, 2.2.2). If fault signatures are not frequency-localized, the frequency reconstruction cannot separate them from noise.
  • domain assumption Additive Gaussian white noise is an adequate model for strong industrial noise.
    All noisy experiments add Gaussian noise via Eq. (20) (Section 3.1). Real noise may be colored, non-stationary, or machine-specific, which would affect generalization.
  • domain assumption Random train/test splits do not introduce leakage between training and test samples.
    The datasets are split randomly (Sections 3.2.1, 3.4.1), but no segmentation details are given; overlapping windows from the same continuous recording would inflate accuracy.
  • ad hoc to paper TP/(TP+FP) is a valid measure of diagnostic accuracy.
    Eq. (19) defines Acc as TP/(TP+FP), which is precision. The paper treats this quantity as accuracy, so the reported tables inherit this assumption.
  • standard math Standard deep learning primitives (DFT, convolutions, GELU, softmax, batch norm) behave as commonly implemented.
    The model relies on standard operations; no formal verification is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of FE-MCFormer: a novel time-frequency interpretable architecture for machinery fault diagnosis under strong noise environments." pith.science (2026). https://pith.science/paper/QISOHWBR

@misc{pith2026250506285,
  author       = {Pith},
  title        = {Pith review of: FE-MCFormer: a novel time-frequency interpretable architecture for machinery fault diagnosis under strong noise environments},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QISOHWBR}},
  note         = {Machine review of arXiv:2505.06285}
}
read the original abstract

Interpretable fault diagnosis (FD) plays a critical role in industrial manufacturing, as it improves human-machine understanding and operational efficiency. However, harsh operating environments often introduce strong background interference or noise, which weakens the discriminative capability and interpretability of existing FD methods. To address this issue, this paper proposes FE-MCFormer, a time-frequency fusion framework for robust and time-frequency interpretable fault diagnosis under strong noise conditions. A frequency adaptive learning layer (FALL) is developed to perform learnable spectral reconstruction, which explicitly suppresses noise-dominated frequency responses while preserving fault-sensitive harmonic structures. Furthermore, a multiscale time-frequency fusion (MSTFF) architecture is designed to jointly capture localized impulsive characteristics and structured global spectral interactions. Extensive experiments on a rolling bearing dataset and a real-world centrifugal compressor dataset demonstrate that the proposed method achieves stable and interpretable diagnostic performance under severe noise environments down to -10 dB SNR. The results indicate that FE-MCFormer provides an effective framework for turbomachinery fault diagnosis in complex noisy environments.

Figures

Figures reproduced from arXiv: 2505.06285 by the authors.

Figure 1
Figure 1. Diagnosis framework for the rotary machine with FE-MCFormer. where 𝐵𝑟𝑎𝑛𝑐ℎ𝑖 [⋅] denotes the two feature branches with 1×3 and 1 × 5 convolutional operations, which are sensitive to local fault features. Further, a softmax function is applied to scale multiscale features in the range of [0, 1], and another 1 × 1 convolution kernel is utilized to fuse the multiscale information. The Hadamard product is performed betwee… view at source ↗
Figure 2
Figure 2. Rolling bearing test rig. (Case 1) feature uses a 1 × 1 convolutional operation to to transform it back into the time-domain, given by: 𝑋̌ ∈ 𝑅 𝐶×𝐿 = 𝑐𝑜𝑛𝑣𝑠𝑒 1×1[𝐹 −1(𝑋̂)] (15) where 𝑐𝑜𝑛𝑣𝑠𝑒 1×1(⋅) denotes the 1 × 1 convolutional opera￾tion with channel squeeze, and 𝐹 −1[⋅] stands for the inverse DFT. Finally, the reconstructed features are added element￾wise to the original feature 𝑋 as the final output of TFFN: 𝑋 𝑟 =… view at source ↗
Figure 3
Figure 3. Test results of rolling bearing dataset in Case 1. FE-MCFormer under noisy environment, different degrees of Gaussian white noise are added to the raw mechanical signals to construct noisy signals with various signal-to￾noise ratios (SNR). The SNR is defined as follows: 𝑆𝑁𝑅 = 10 lg ( 𝑃𝑠 𝑃𝑛 ) (20) 𝑃𝑠 = 1 𝑁 ∑ 𝑁 𝑛=1 | | | 𝑣(𝑛) 2| | | (21) where 𝑃𝑛 and 𝑃𝑠 denote the power of the raw signal and noisy signal, and 𝑣(𝑛) rep… view at source ↗
Figures from the paper (9 more)
Figure 4
Figure 4. Figure 4: Results of 2D T-SNE visualization on rolling bearing dataset under SNR = -6 (Case 1). noisy environments, the multiscale kernels in the multiscale convolutional neural networks (MSCNN) tend to amplify both high frequency noise patterns and key local features, resulting…
Figure 5
Figure 5. Figure 5: Reconstruction results of bearing fault signal under SNR = −4 dB (Case 1). consistently achieves higher accuracy under all noise con￾ditions, indicating the superiority of other components in FE-MCFormer. Compared with Resnet50 and Densenet, FE-MCFormer has the higher …
Figure 6
Figure 6. Figure 6: Attention heat map of vibration signals from three states under SNR = −8 dB: (a) KA01, outer ring fault. (b) KA07, outer ring fault. (c) KI01, inner ring fault. 3.2.3. Sensitivity analysis and ablation study To assess the influence of each components in FE￾MCFormer, fo…
Figure 7
Figure 7. Figure 7: Gearbox fault test rig. (Case2) [PITH_FULL_IMAGE:figures/full_fig_p011_7.png]
Figure 8
Figure 8. Figure 8: Test results on SEU dataset (Case 2). shows highly concentrated attention on the fault impulse regions. This suggests that the introduction of TFFN and FAREL not only enhance the global dependency of FE￾MCFormer, but also enhance the model interpretability un￾der stron…
Figure 9
Figure 9. Figure 9: Confusion matrix of gearbox dataset under SNR = −10 dB (Case 2). 3.3.2. Result analysis The experimental results of the SEU dataset are shown in [PITH_FULL_IMAGE:figures/full_fig_p012_9.png]
Figure 10
Figure 10. Figure 10: Reconstruction results of gear fault signals (Case 2) [PITH_FULL_IMAGE:figures/full_fig_p013_10.png]
Figure 11
Figure 11. Figure 11: Illustration of the centrifugal compressor for methodology implementation (Case 3) [PITH_FULL_IMAGE:figures/full_fig_p014_11.png]
Figure 12
Figure 12. Figure 12: Reconstruction results of the compressor dataset (Case 3). where the blue curves represent the original spectra, while the red ones correspond to the reconstructed signals. It can be clearly observed from [PITH_FULL_IMAGE:figures/full_fig_p015_12.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

37 extracted references · 24 canonical work pages

  1. [1]

    Appli- cationsofmachinelearningtomachinefaultdiagnosis:Areviewand roadmap

    Lei, Y., Yang, B., Jiang, X., Jia, F., Li, N., Nandi, A.K., 2020. Appli- cationsofmachinelearningtomachinefaultdiagnosis:Areviewand roadmap. Mech. Syst. Signal Process. 138, 106587

  2. [2]

    Consistency-regularized-label-aware contrastive learning with uncertainty-aware periodic pseudo-labeling for machinery fault diagnosis under limited labeled data

    Ahmad, H., Cheng, W., Wang, W., Zhang, S., Nie, Z., Liu, H., Chen, X., 2025. Consistency-regularized-label-aware contrastive learning with uncertainty-aware periodic pseudo-labeling for machinery fault diagnosis under limited labeled data. Adv. Eng. Inform. 68, 103656

  3. [3]

    Physics modeling-driven interpretable data augmentation method for bear- ing fault diagnosis under imbalanced data

    Zhao, L., Qi, J., Wu, F., Wang, Y., Qin, Y., 2026. Physics modeling-driven interpretable data augmentation method for bear- ing fault diagnosis under imbalanced data. Adv. Eng. Inform. 70, 104189. URL:https://www.sciencedirect.com/science/article/pii/ S1474034625010821, doi:https://doi.org/10.1016/j.aei.2025.104189

  4. [4]

    A time- frequencyspectralamplitudemodulationmethodanditsapplications in rolling bearing fault diagnosis

    Jiang, Z., Zhang, K., Xiang, L., Yu, G., Xu, Y., 2023. A time- frequencyspectralamplitudemodulationmethodanditsapplications in rolling bearing fault diagnosis. Mech. Syst. Signal Process. 185, 109832. doi:10.1016/j.ymssp.2022.109832

  5. [5]

    Weak fault detection of rolling bearing using a ds-based adaptive spectrum reconstruction method

    Xu, Y., Hu, J., 2019. Weak fault detection of rolling bearing using a ds-based adaptive spectrum reconstruction method. J. Instrum. 14, P03022

  6. [6]

    Feature extractionbasedonhierarchicalimprovedenvelopespectrumentropy for rolling bearing fault diagnosis

    Chen, Z., Yang, Y., He, C., Liu, Y., Liu, X., Cao, Z., 2023. Feature extractionbasedonhierarchicalimprovedenvelopespectrumentropy for rolling bearing fault diagnosis. IEEE Trans. Instrum. Meas. 72, 1–12. doi:10.1109/TIM.2023.3277938

  7. [7]

    Application of ICEEMDAN energy entropy and AFSA-SVM for fault diagnosis of hoist sheave bearing

    Kou, Z., Yang, F., Wu, J., Li, T., 2020. Application of ICEEMDAN energy entropy and AFSA-SVM for fault diagnosis of hoist sheave bearing. Entropy 22, 1347. doi:10.3390/e22121347

  8. [8]

    A novel rolling bearing fault diagnosis method based on continuous hierarchical fractional range entropy

    Zhou, J., Chen, C., Guo, J., Wang, L., Liu, Z., Feng, C., 2023. A novel rolling bearing fault diagnosis method based on continuous hierarchical fractional range entropy. Measurement 220, 113395. doi:10.1016/j.measurement.2023.113395

Show all 37 references
  1. [9]

    Meta-learning with distributional similarity preference for few-shot fault diagnosis under varying working conditions

    Ren, C., Jiang, B., Lu, N., Simani, S., Gao, F., 2024. Meta-learning with distributional similarity preference for few-shot fault diagnosis under varying working conditions. IEEE Trans. Cybern. 54, 2746–

  2. [10]

    A meta-learning method for electric machine bearing fault diagnosis undervaryingworkingconditionswithlimiteddata

    Chen, J., Hu, W., Cao, D., Zhang, Z., Chen, Z., Blaabjerg, F., 2023. A meta-learning method for electric machine bearing fault diagnosis undervaryingworkingconditionswithlimiteddata. IEEETrans.Ind. Inf. 19, 2552–2564. doi:10.1109/TII.2022.3165027

  3. [11]

    Semi- supervised fault diagnosis of gearbox based on feature pre-extraction mechanismandimprovedgenerativeadversarialnetworksunderlim- ited labeled samples and noise environment

    Zhang, L., Wang, B., Liang, P., Yuan, X., Li, N., 2023. Semi- supervised fault diagnosis of gearbox based on feature pre-extraction mechanismandimprovedgenerativeadversarialnetworksunderlim- ited labeled samples and noise environment. Adv. Eng. Inform. 58, 102211. doi:10.1016/...

  4. [12]

    Ahybridcross-domainfew- shot bearing fault diagnosis method combining multi-scale feature association and physical information

    He,C.,Gai,N.,Yan,K.,Shao,H.,2026. Ahybridcross-domainfew- shot bearing fault diagnosis method combining multi-scale feature association and physical information. Adv. Eng. Inform. 69, 104077

  5. [13]

    Multiscale deep attention q network: A new deep reinforcement learning method for imbalancedfaultdiagnosisingearboxes

    Wang, H., Zhou, Z., Zhang, L., Yan, R., 2024. Multiscale deep attention q network: A new deep reinforcement learning method for imbalancedfaultdiagnosisingearboxes. IEEETrans.Instrum.Meas. 73, 1–12. doi:10.1109/TIM.2023.3338664

  6. [14]

    Xu,Y.,Yan,X.,Feng,K.,Sheng,X.,Sun,B.,Liu,Z.,2022.Attention- basedmultiscaledenoisingresidualconvolutionalneuralnetworksfor fault diagnosis of rotating machinery. Reliab. Eng. Syst. Saf. 226, 108714. doi:10.1016/j.ress.2022.108714

  7. [15]

    Multiscale residual attention convolutional neural network for bearing fault diagnosis

    Jia, L., Chow, T.W.S., Wang, Y., Yuan, Y., 2022. Multiscale residual attention convolutional neural network for bearing fault diagnosis. IEEETrans. Instrum. Meas.71, 1–13. doi:10.1109/TIM.2022.3196742

  8. [16]

    Remain- ing useful life prediction for the harmonic reducer of industrial robots via in-situ current signal and lightweight multiscale atten- tion deep networks

    Yuan, Y., Han, Y., Xiao, K., Xu, Z., Jiang, X., 2025. Remain- ing useful life prediction for the harmonic reducer of industrial robots via in-situ current signal and lightweight multiscale atten- tion deep networks. Journal of Manufacturing Systems 83, 322–

  9. [17]

    Variational Attention-Based Interpretable Transformer Network for Rotary Ma- chine Fault Diagnosis

    Li, Y., Zhou, Z., Sun, C., Chen, X., Yan, R., 2024. Variational Attention-Based Interpretable Transformer Network for Rotary Ma- chine Fault Diagnosis. IEEE Trans. Neural Netw. Learning Syst. 35, 6180–6193

  10. [18]

    A novel time–frequency Transformer based on self–attention mechanism and its application infaultdiagnosisofrollingbearings

    Ding, Y., Jia, M., Miao, Q., Cao, Y., 2022. A novel time–frequency Transformer based on self–attention mechanism and its application infaultdiagnosisofrollingbearings. MechanicalSystemsandSignal Processing 168. Publisher: Elsevier BV

  11. [19]

    Wd-kantf: An interpretable intelligent fault diagnosis framework for rotating machinery under noise environments and small sample conditions

    Zhang, Y., Zhao, X., Peng, Z., Xu, R., Chen, P., 2025. Wd-kantf: An interpretable intelligent fault diagnosis framework for rotating machinery under noise environments and small sample conditions. Adv. Eng. Inform. 66, 103452

  12. [20]

    Convformer- NSE: A novel end-to-end gearbox fault diagnosis framework under heavy noise using joint global and local information

    Han, S., Shao, H., Cheng, J., Yang, X., Cai, B., 2023. Convformer- NSE: A novel end-to-end gearbox fault diagnosis framework under heavy noise using joint global and local information. IEEE/ASME Trans. Mechatron. 28, 340–349. doi:10.1109/TMECH.2022.3199985

  13. [21]

    YOLOv11: An overview of the key architectural enhancements

    Khanam, R., Hussain, M., 2024. YOLOv11: An overview of the key architectural enhancements. URL:http://arxiv.org/abs/2410.17725, arXiv:2410.17725 [cs]

  14. [22]

    Inception-v4, inception-resnet and the impact of residual connections on learning

    Szegedy,C.,Ioffe,S.,Vanhoucke,V.,Alemi,A.,2017. Inception-v4, inception-resnet and the impact of residual connections on learning. AAAI 31

  15. [23]

    DRSwin- ST: An intelligent fault diagnosis framework based on dynamic threshold noise reduction and sparse transformer with shifted win- dows 250, 110327

    Zhou, T., Yao, D., Yang, J., Meng, C., Li, A., Li, X., 2024. DRSwin- ST: An intelligent fault diagnosis framework based on dynamic threshold noise reduction and sparse transformer with shifted win- dows 250, 110327

  16. [24]

    Longformer: The long- document transformer

    Beltagy, I., Peters, M.E., Cohan, A., 2020. Longformer: The long- document transformer

  17. [25]

    Attention is not all you need: Pure attention loses rank doubly exponentially with depth

    Dong, Y., Cordonnier, J.B., Loukas, A., 2023. Attention is not all you need: Pure attention loses rank doubly exponentially with depth. URL:http://arxiv.org/abs/2103.03404, doi:10.48550/arXiv. 2103.03404,arXiv:2103.03404 [cs]

  18. [26]

    One wide feedforward is all you need, in: Proceedings of the Eighth Conference on Machine Translation, Association for Computational Linguistics

    Pires, T., Vilarinho Lopes, A., Assogba, Y., Setiawan, H., 2023. One wide feedforward is all you need, in: Proceedings of the Eighth Conference on Machine Translation, Association for Computational Linguistics. pp. 1031–1044

  19. [27]

    FNet: Mixing tokens with fourier transforms

    Lee-Thorp, J., Ainslie, J., Eckstein, I., Ontanon, S., 2022. FNet: Mixing tokens with fourier transforms. URL:http://arxiv.org/abs/ 2105.03824, doi:10.48550/arXiv.2105.03824,arXiv:2105.03824 [cs]

  20. [28]

    org/abs/2105.14103,doi:10.48550/arXiv.2105.14103,arXiv:2105.14103 [cs]

    Zhai, S., Talbott, W., Srivastava, N., Huang, C., Goh, H., Zhang, R., Susskind,J.,2021.Anattentionfreetransformer.URL:http://arxiv. org/abs/2105.14103,doi:10.48550/arXiv.2105.14103,arXiv:2105.14103 [cs]. Yuhan Yuan et al.:Preprint submitted to ElsevierPage 15 of 16 FE-MCFormer...

  21. [29]

    Lessmeier, C., Kimotho, J.K., Zimmer, D., Sextro, W., 2016. Con- dition monitoring of bearing damage in electromechanical drive sys- tems by using motor current signals of electric motors: A benchmark data set for data-driven classification, in: PHM society European conference

  22. [30]

    Bearing fault diagnosis base on multi-scale cnn and lstm model

    Chen, X., Zhang, B., Gao, D., 2021. Bearing fault diagnosis base on multi-scale cnn and lstm model. J. Intell. Manuf. 32, 971–987

  23. [31]

    Understanding and learning discriminant features based on multiattention 1dcnn for wheelset bearing fault diagnosis

    Wang, H., Liu, Z., Peng, D., Qin, Y., 2020. Understanding and learning discriminant features based on multiattention 1dcnn for wheelset bearing fault diagnosis. IEEE Trans. Ind. Inf. 16, 5735–

  24. [32]

    Deeplearningalgorithmsforrotatingmachineryintelligentdiagnosis: An open source benchmark study

    Zhao, Z., Li, T., Wu, J., Sun, C., Wang, S., Yan, R., Chen, X., 2020. Deeplearningalgorithmsforrotatingmachineryintelligentdiagnosis: An open source benchmark study. ISA Trans

  25. [33]

    LiConvFormer: A lightweight fault diagnosis framework using separable multiscale convolutionandbroadcastself-attention

    Yan,S.,Shao,H.,Wang,J.,Zheng,X.,Liu,B.,2024. LiConvFormer: A lightweight fault diagnosis framework using separable multiscale convolutionandbroadcastself-attention. ExpertSystemswithAppli- cations 237, 121338. doi:10.1016/j.eswa.2023.121338

  26. [34]

    Highly accurate machine fault diagnosis using deep transfer learning

    Shao, S., McAleer, S., Yan, R., Baldi, P., 2018. Highly accurate machine fault diagnosis using deep transfer learning. IEEE Trans. Ind. Inf. 15, 2446–2455. Yuhan Yuan et al.:Preprint submitted to ElsevierPage 16 of 16

  27. [336]

    URL:https://www.sciencedirect.com/science/article/pii/ S0278612525002353, doi:https://doi.org/10.1016/j.jmsy.2025.09.008

  28. [2756]

    doi:10.1109/TCYB.2023.3338768

  29. [5745]

    doi:10.1109/TII.2019.2955540

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.