pith. sign in

arxiv: 2606.04073 · v1 · pith:CWDMIA7Gnew · submitted 2026-06-02 · 💻 cs.LG · cs.AI· stat.ML

TPA-AD: A Two-Stage Pseudo Anomaly-Guided Method for Bearing Time-Series Anomaly Detection

Pith reviewed 2026-06-28 11:13 UTC · model grok-4.3

classification 💻 cs.LG cs.AIstat.ML
keywords anomaly detectiontime seriespseudo anomaliescontrastive learningbearing fault detectionunsupervised learningmixed variablesdegradation monitoring
0
0 comments X

The pith

A two-stage method generates pseudo-anomalies near the normal boundary to learn anomaly-sensitive representations for time-series detection using only normal training data.

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

The paper presents TPA-AD, a method for detecting anomalies in bearing time-series data when only normal samples are available during training. It first builds pseudo-anomalous windows close to the normal data boundary by using a reconstruction model with controlled per-feature errors. These synthetic examples then drive a contrastive learning stage that produces representations distinguishing normal from anomalous patterns. Finally, k-nearest neighbors scoring yields both window-level and point-level anomaly measures. The approach targets clearer separation at the normal boundary and handles datasets mixing continuous and discrete features, with tests on bearing fault, degradation, and public time-series benchmarks.

Core claim

The central claim is that constructing pseudo-anomalous windows near the normal boundary using a reconstruction model and per-feature target-error control, followed by contrastive learning between normal and pseudo-anomalous windows and KNN-based scoring, improves the separability of the normal boundary and enables effective anomaly detection in mixed-variable time-series scenarios without requiring real anomaly examples or known fault categories.

What carries the argument

The pseudo-anomaly generation step that places synthetic anomalies in boundary neighborhoods through reconstruction and per-feature error control, which then supplies the contrastive pairs for learning anomaly-sensitive representations.

If this is right

  • The method produces relatively stable anomaly responses across bearing fault detection and degradation process datasets.
  • It remains sensitive to the evolution of degradation in the time-series signals.
  • It extends to public time-series anomaly detection benchmarks and real high-speed train bearing data with a degree of broader applicability.
  • It jointly processes continuous and discrete features within the same mixed-variable time series.

Where Pith is reading between the lines

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

  • The boundary-neighborhood construction of pseudo-anomalies might reduce reliance on domain-specific fault knowledge in other industrial monitoring settings.
  • If the contrastive stage proves robust, similar pseudo-anomaly guidance could be tested on non-bearing time series where labeled anomalies remain scarce.
  • The per-feature error control during pseudo-anomaly creation may offer a route to adapt the method when feature scales or types differ sharply across datasets.

Load-bearing premise

That pseudo-anomalous windows created near the normal boundary serve as effective proxies for real anomalies so that contrastive learning produces representations that generalize to actual anomalies.

What would settle it

Running the trained model on a labeled bearing dataset with known real anomalies and finding no gain in detection precision or recall compared with a standard reconstruction baseline that does not use pseudo-anomaly contrastive learning.

Figures

Figures reproduced from arXiv: 2606.04073 by Lin Wang, Minghang Zhao, Ran Li, Rui Wang, Shisheng Zhong, Xiancheng Wang, Zhibo Zhang.

Figure 1
Figure 1. Figure 1: Schematic comparison between conventional fault injection and the pseudo-anomaly generation [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The proposed two-stage framework. The upper part shows reconstruction-driven pseudo-anomalous [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Stage 1 pipeline. The left part shows data normalization in the continuous subspace and the [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Stage 2 pipeline. The left side shows normal–pseudo-anomalous contrastive training, where normal [PITH_FULL_IMAGE:figures/full_fig_p012_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Illustration of the CWRU dataset. A shows the experimental device, B shows the long sequence [PITH_FULL_IMAGE:figures/full_fig_p023_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Fault-detection results on the CWRU dataset. A shows the window-energy heatmap together [PITH_FULL_IMAGE:figures/full_fig_p024_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Illustration of the HTBF dataset. A shows the experimental setup, B shows the long sequence [PITH_FULL_IMAGE:figures/full_fig_p026_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Fault-detection results on the HTBF dataset. A shows the window-energy heatmap and [PITH_FULL_IMAGE:figures/full_fig_p027_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Illustration of the PHM2009 dataset. A shows the experimental setup, B shows the long sequence [PITH_FULL_IMAGE:figures/full_fig_p029_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Fault-detection results on the PHM2009 dataset. A shows the window-energy heatmap and [PITH_FULL_IMAGE:figures/full_fig_p030_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Illustration of the REALBOX field-measurement data. A shows the real data-acquisition scenario, [PITH_FULL_IMAGE:figures/full_fig_p032_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Fault-detection results on the REALBOX field-measurement data. A shows the window-energy [PITH_FULL_IMAGE:figures/full_fig_p033_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Illustration of the XJTU-SY degradation data. A shows the experimental setup, B shows the [PITH_FULL_IMAGE:figures/full_fig_p035_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Degradation-detection results for XJTU-SY Bearing1-1. The proposed method maintains low [PITH_FULL_IMAGE:figures/full_fig_p037_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Degradation-detection results for XJTU-SY Bearing1-3. The anomaly-score curve and [PITH_FULL_IMAGE:figures/full_fig_p038_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Degradation-detection results for XJTU-SY Bearing2-2. The proposed method shows continuous [PITH_FULL_IMAGE:figures/full_fig_p039_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Degradation-detection results for XJTU-SY Bearing2-5. The figure shows anomaly scores, metric [PITH_FULL_IMAGE:figures/full_fig_p041_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: Degradation-detection results for XJTU-SY Bearing3-3. The model shows progressively stronger [PITH_FULL_IMAGE:figures/full_fig_p042_18.png] view at source ↗
Figure 19
Figure 19. Figure 19: Degradation-detection results for XJTU-SY Bearing3-5. The proposed method yields stable high [PITH_FULL_IMAGE:figures/full_fig_p043_19.png] view at source ↗
Figure 20
Figure 20. Figure 20: Illustration of the IMS bearing degradation data. The figure shows the experimental setup, long [PITH_FULL_IMAGE:figures/full_fig_p045_20.png] view at source ↗
Figure 21
Figure 21. Figure 21: Degradation-detection results on the IMS bearing data. The proposed method captures part of [PITH_FULL_IMAGE:figures/full_fig_p046_21.png] view at source ↗
Figure 22
Figure 22. Figure 22: Summary of the core metrics across multiple datasets. A–E compare the results of [PITH_FULL_IMAGE:figures/full_fig_p048_22.png] view at source ↗
Figure 23
Figure 23. Figure 23: Ablation on pseudo-anomaly injection strategies. The upper row corresponds to HTBF and [PITH_FULL_IMAGE:figures/full_fig_p050_23.png] view at source ↗
Figure 24
Figure 24. Figure 24: Ablation on the shared encoder design and the coverage relation with CARLA. Panels A and [PITH_FULL_IMAGE:figures/full_fig_p053_24.png] view at source ↗
Figure 25
Figure 25. Figure 25: Ablation on the Stage 2 contrastive loss and anomaly score. Panels A and B show the anomaly [PITH_FULL_IMAGE:figures/full_fig_p055_25.png] view at source ↗
Figure 26
Figure 26. Figure 26: Ablation results for the RL step-size controller. The left column reports the relative gains of RL [PITH_FULL_IMAGE:figures/full_fig_p058_26.png] view at source ↗
Figure 27
Figure 27. Figure 27: Analysis of the empirical basis of the Stage 1 target reconstruction-error interval. The figure com [PITH_FULL_IMAGE:figures/full_fig_p061_27.png] view at source ↗
Figure 28
Figure 28. Figure 28: Comparison of score curves for CARLA-style pseudo-anomaly injection and the original CARLA [PITH_FULL_IMAGE:figures/full_fig_p063_28.png] view at source ↗
Figure 29
Figure 29. Figure 29: Hyperparameter sensitivity analysis of the target-error quantile interval. Each subplot shows the [PITH_FULL_IMAGE:figures/full_fig_p066_29.png] view at source ↗
Figure 30
Figure 30. Figure 30: Anomaly-detection results on simulated TSAD data. The top row shows the input time series and [PITH_FULL_IMAGE:figures/full_fig_p070_30.png] view at source ↗
Figure 31
Figure 31. Figure 31: Example anomaly scores on public TSAD data I. The figure shows the anomaly-score curves and [PITH_FULL_IMAGE:figures/full_fig_p072_31.png] view at source ↗
Figure 32
Figure 32. Figure 32: Example anomaly scores on public TSAD data II. The figure further shows the model’s detection [PITH_FULL_IMAGE:figures/full_fig_p073_32.png] view at source ↗
read the original abstract

This paper proposes a two-stage pseudo anomaly-guided anomaly detection method (\textbf{T}wo-stage \textbf{P}seudo \textbf{A}nomaly-guided \textbf{A}nomaly \textbf{D}etection, \textbf{TPA-AD}) for axle-box bearing time-series anomaly detection (time series anomaly detection, TSAD) under the setting where only normal samples are available for training. The method first generates pseudo-anomalous windows near the normal boundary using a reconstruction model and per-feature target-error control. It then learns anomaly-sensitive representations through contrastive learning between normal and pseudo-anomalous windows, and finally produces window-level and point-level anomaly scores using k-nearest neighbors (KNN). Compared with existing methods that rely on known fault categories, real anomaly priors, or random anomaly injection, TPA-AD improves the separability of the normal boundary by constructing pseudo-anomalies in boundary neighborhoods and can jointly handle continuous and discrete features in mixed-variable scenarios. The main experiments are conducted on bearing fault detection datasets and degradation-process datasets, with an additional exploratory extension on $13$ public TSAD datasets. The results show that the proposed method yields relatively stable anomaly responses, is sensitive to degradation evolution, and demonstrates a certain degree of broader applicability on public TSAD benchmarks and real high-speed-train-related bearing data.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper proposes TPA-AD, a two-stage method for unsupervised time-series anomaly detection on axle-box bearing data (only normal samples available). Stage 1 generates pseudo-anomalous windows near the normal boundary via a reconstruction model with per-feature target-error control. Stage 2 applies contrastive learning on normal vs. pseudo-anomalous windows to obtain anomaly-sensitive representations, followed by KNN-based window- and point-level scoring. The method claims improved normal-boundary separability, ability to handle mixed continuous/discrete features, stable anomaly responses, sensitivity to degradation evolution, and a degree of applicability on 13 public TSAD benchmarks plus real high-speed-train bearing data.

Significance. If the pseudo-anomalies generated via reconstruction error and per-feature thresholds prove to be representative proxies for real bearing faults, the approach could advance unsupervised TSAD by avoiding reliance on known fault categories, real anomaly priors, or random injection. The extension to degradation-process datasets and 13 public benchmarks is a positive step toward assessing broader utility in industrial mixed-variable settings.

major comments (2)
  1. [Abstract] Abstract: the claim of performance gains, stable responses, and broader applicability is stated without any quantitative results, error bars, baseline comparisons, or statistical tests; this prevents assessment of the central claim that pseudo-anomaly construction improves separability.
  2. [Method] Method (pseudo-anomaly generation stage): the construction relies on reconstruction error distribution plus independent per-feature target-error thresholds, but provides no analysis showing that the resulting windows capture correlated temporal structure or specific fault signatures (e.g., periodic impulses or degradation trends) present in real axle-box bearing data; this assumption is load-bearing for the subsequent contrastive-learning stage producing generalizable representations.
minor comments (1)
  1. Notation for the reconstruction model, target-error control, and contrastive loss could be made more explicit with equations or pseudocode to improve reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments. We address each major comment below and note planned revisions.

read point-by-point responses
  1. Referee: [Abstract] Abstract: the claim of performance gains, stable responses, and broader applicability is stated without any quantitative results, error bars, baseline comparisons, or statistical tests; this prevents assessment of the central claim that pseudo-anomaly construction improves separability.

    Authors: We agree that the abstract would benefit from quantitative support to allow readers to assess the claims. In the revised manuscript we will add concise quantitative highlights, including key performance metrics and baseline comparisons from the bearing and degradation experiments. revision: yes

  2. Referee: [Method] Method (pseudo-anomaly generation stage): the construction relies on reconstruction error distribution plus independent per-feature target-error thresholds, but provides no analysis showing that the resulting windows capture correlated temporal structure or specific fault signatures (e.g., periodic impulses or degradation trends) present in real axle-box bearing data; this assumption is load-bearing for the subsequent contrastive-learning stage producing generalizable representations.

    Authors: The pseudo-anomaly stage is intentionally designed to produce boundary-near samples via per-feature error control. The manuscript validates the overall approach through downstream contrastive learning and detection results on real axle-box data, including degradation sensitivity. We acknowledge the absence of a dedicated analysis of temporal structure or fault signatures in the generated windows and will add visualizations and comparisons of pseudo-anomaly characteristics versus real fault patterns in the revision. revision: yes

Circularity Check

0 steps flagged

No circularity: TPA-AD is a constructive two-stage pipeline without self-referential reductions

full rationale

The paper describes a method that generates pseudo-anomalies using a reconstruction model plus per-feature target-error control, then applies contrastive learning between normal and pseudo-anomalous windows, and finally computes KNN-based scores. No equations, fitted parameters renamed as predictions, self-citations as load-bearing premises, or ansatzes imported from prior author work appear in the provided text. The central claim rests on the empirical behavior of this explicit construction on external datasets rather than any derivation that reduces outputs to inputs by definition. This is the standard case of a non-circular algorithmic proposal.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on standard machine-learning assumptions about the utility of synthetic anomalies and the availability of only normal data; no explicit free parameters, axioms, or invented entities are stated in the abstract.

axioms (2)
  • domain assumption Normal samples alone are sufficient to define a boundary that can be meaningfully perturbed to create useful pseudo-anomalies.
    Invoked in the first stage of pseudo-anomaly generation described in the abstract.
  • domain assumption Contrastive learning between normal and pseudo-anomalous windows produces representations that transfer to real anomalies.
    Central to the second stage and final scoring step.

pith-pipeline@v0.9.1-grok · 5790 in / 1353 out tokens · 20879 ms · 2026-06-28T11:13:31.534335+00:00 · methodology

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

68 extracted references · 2 canonical work pages

  1. [1]

    A comprehensive review of deep learning-based fault diagnosis approaches for rolling bearings: Advancements and challenges

    Jiangdong Zhao, Wenming Wang, Ji Huang, and Xiaolu Ma. A comprehensive review of deep learning-based fault diagnosis approaches for rolling bearings: Advancements and challenges. AIP Advances, 15 0 (2), 2025

  2. [2]

    Rolling bearing intelligent fault diagnosis towards variable speed and imbalanced samples using multiscale dynamic supervised contrast learning

    Yutong Dong, Hongkai Jiang, Renhe Yao, Mingzhe Mu, and Qiao Yang. Rolling bearing intelligent fault diagnosis towards variable speed and imbalanced samples using multiscale dynamic supervised contrast learning. Reliability Engineering & System Safety, 243: 0 109805, 2024 a

  3. [3]

    Time-frequency supervised contrastive learning via pseudo-labeling: An unsupervised domain adaptation network for rolling bearing fault diagnosis under time-varying speeds

    Bin Pang, Qiuhai Liu, Zhenduo Sun, Zhenli Xu, and Ziyang Hao. Time-frequency supervised contrastive learning via pseudo-labeling: An unsupervised domain adaptation network for rolling bearing fault diagnosis under time-varying speeds. Advanced Engineering Informatics, 59: 0 102304, 2024

  4. [4]

    Dynamic model-assisted disentanglement framework for rolling bearing fault diagnosis under time-varying speed conditions

    Yuhui Xu, Yimin Jiang, Tangbin Xia, Dong Wang, Zhen Chen, Ershun Pan, and Lifeng Xi. Dynamic model-assisted disentanglement framework for rolling bearing fault diagnosis under time-varying speed conditions. Mechanical Systems and Signal Processing, 230: 0 112588, 2025

  5. [5]

    Ccft: The convolution and cross-fusion transformer for fault diagnosis of bearings

    Tantao Lin, Yongsheng Zhu, Zhijun Ren, Kai Huang, and Dawei Gao. Ccft: The convolution and cross-fusion transformer for fault diagnosis of bearings. IEEE/ASME Transactions on Mechatronics, 29 0 (3): 0 2161--2172, 2023

  6. [6]

    A novel method for rolling bearing fault diagnosis based on gramian angular field and cnn-vit

    Zijun Zhou, Qingsong Ai, Ping Lou, Jianmin Hu, and Junwei Yan. A novel method for rolling bearing fault diagnosis based on gramian angular field and cnn-vit. Sensors, 24 0 (12): 0 3967, 2024 a

  7. [7]

    A deep learning-based method for bearing fault diagnosis with few-shot learning

    Yang Li, Xiaojiao Gu, and Yonghe Wei. A deep learning-based method for bearing fault diagnosis with few-shot learning. Sensors, 24 0 (23): 0 7516, 2024

  8. [8]

    Msff-net: Multi-sensor frequency-domain feature fusion network with lightweight 1d cnn for bearing fault diagnosis

    Miao Dai, Hangyeol Jo, Moonsuk Kim, and Sang-Woo Ban. Msff-net: Multi-sensor frequency-domain feature fusion network with lightweight 1d cnn for bearing fault diagnosis. Sensors, 25 0 (14): 0 4348, 2025

  9. [9]

    Multi-sensor fusion rolling bearing intelligent fault diagnosis based on vmd and ultra-lightweight googlenet in industrial environments

    Shouqi Wang and Zhigang Feng. Multi-sensor fusion rolling bearing intelligent fault diagnosis based on vmd and ultra-lightweight googlenet in industrial environments. Digital Signal Processing, 145: 0 104306, 2024

  10. [10]

    Multi-sensor data fusion-enabled lightweight convolutional double regularization contrast transformer for aerospace bearing small samples fault diagnosis

    Yutong Dong, Hongkai Jiang, Mingzhe Mu, and Xin Wang. Multi-sensor data fusion-enabled lightweight convolutional double regularization contrast transformer for aerospace bearing small samples fault diagnosis. Advanced Engineering Informatics, 62: 0 102573, 2024 b

  11. [11]

    Mrcfn: A multi-sensor residual convolutional fusion network for intelligent fault diagnosis of bearings in noisy and small sample scenarios

    Maoyou Ye, Xiaoan Yan, Xing Hua, Dong Jiang, Ling Xiang, and Ning Chen. Mrcfn: A multi-sensor residual convolutional fusion network for intelligent fault diagnosis of bearings in noisy and small sample scenarios. Expert Systems with Applications, 259: 0 125214, 2025

  12. [12]

    Agfcn: A bearing fault diagnosis method for high-speed train bogie under complex working conditions

    Deqiang He, Jinxin Wu, Zhenzhen Jin, ChengGeng Huang, Zexian Wei, and Cai Yi. Agfcn: A bearing fault diagnosis method for high-speed train bogie under complex working conditions. Reliability Engineering & System Safety, 258: 0 110907, 2025

  13. [13]

    Zero-sample fault diagnosis of rolling bearings via fault spectrum knowledge and autonomous contrastive learning

    Guoqiang Li, Meirong Wei, Defeng Wu, Yiwei Cheng, Jun Wu, and Jin Yan. Zero-sample fault diagnosis of rolling bearings via fault spectrum knowledge and autonomous contrastive learning. Expert Systems with Applications, 275: 0 127080, 2025

  14. [14]

    Frequency pattern graph spectrum model and its applications in rolling bearing fault diagnosis

    Yanlei Liu, Yonggang Xu, Miaorui Yang, Hong Jiang, and Kun Zhang. Frequency pattern graph spectrum model and its applications in rolling bearing fault diagnosis. Mechanical Systems and Signal Processing, 240: 0 113426, 2025 a

  15. [15]

    Intelligent diagnosis of rolling bearings under cross-domain missing data: A lightweight complex domain imputation and unsupervised time--frequency alignment approach

    Xuan Wang, Zhanqiang Hou, Gao Liu, Junwei Xue, Qingsong Li, Xuezhong Wu, and Dingbang Xiao. Intelligent diagnosis of rolling bearings under cross-domain missing data: A lightweight complex domain imputation and unsupervised time--frequency alignment approach. Mechanical Systems and Signal Processing, 241: 0 113504, 2025

  16. [16]

    Deep learning for time series anomaly detection: A survey

    Zahra Zamanzadeh Darban, Geoffrey I Webb, Shirui Pan, Charu Aggarwal, and Mahsa Salehi. Deep learning for time series anomaly detection: A survey. ACM Computing Surveys, 57 0 (1): 0 1--42, 2024

  17. [17]

    The elephant in the room: Towards a reliable time-series anomaly detection benchmark

    Qinghua Liu and John Paparrizos. The elephant in the room: Towards a reliable time-series anomaly detection benchmark. Advances in Neural Information Processing Systems, 37: 0 108231--108261, 2024

  18. [19]

    Learning multi-pattern normalities in the frequency domain for efficient time series anomaly detection

    Feiyi Chen, Yingying Zhang, Zhen Qin, Lunting Fan, Renhe Jiang, Yuxuan Liang, Qingsong Wen, and Shuiguang Deng. Learning multi-pattern normalities in the frequency domain for efficient time series anomaly detection. In 2024 IEEE 40th International Conference on Data Engineering (ICDE), pages 747--760. IEEE, 2024

  19. [20]

    Unraveling the ‘anomaly’in time series anomaly detection: a self-supervised tri-domain solution

    Yuting Sun, Guansong Pang, Guanhua Ye, Tong Chen, Xia Hu, and Hongzhi Yin. Unraveling the ‘anomaly’in time series anomaly detection: a self-supervised tri-domain solution. In 2024 IEEE 40th International Conference on Data Engineering (ICDE), pages 981--994. IEEE, 2024

  20. [21]

    Temporal-frequency masked autoencoders for time series anomaly detection

    Yuchen Fang, Jiandong Xie, Yan Zhao, Lu Chen, Yunjun Gao, and Kai Zheng. Temporal-frequency masked autoencoders for time series anomaly detection. In 2024 IEEE 40th international conference on data engineering (ICDE), pages 1228--1241. IEEE, 2024

  21. [24]

    Interdependency matters: graph alignment for multivariate time series anomaly detection

    Yuanyi Wang, Haifeng Sun, Chengsen Wang, Mengde Zhu, Jingyu Wang, Wei Tang, Qi Qi, Zirui Zhuang, and Jianxin Liao. Interdependency matters: graph alignment for multivariate time series anomaly detection. In 2024 IEEE International Conference on Data Mining (ICDM), pages 869--874. IEEE, 2024

  22. [25]

    Catch: Channel-aware multivariate time series anomaly detection via frequency patching

    Xingjian Wu, Xiangfei Qiu, Zhengyu Li, Yihang Wang, Jilin Hu, Chenjuan Guo, Hui Xiong, and Bin Yang. Catch: Channel-aware multivariate time series anomaly detection via frequency patching. In International conference on learning representations, volume 2025, pages 17017--17045, 2025

  23. [26]

    Towards a general time series anomaly detector with adaptive bottlenecks and dual adversarial decoders

    Qichao Shentu, Beibu Li, Kai Zhao, Yang Shu, Zhongwen Rao, Lujia Pan, Bin Yang, and Chenjuan Guo. Towards a general time series anomaly detector with adaptive bottlenecks and dual adversarial decoders. In International Conference on Learning Representations, volume 2025, pages 81358--81381, 2025

  24. [27]

    Learn hybrid prototypes for multivariate time series anomaly detection

    Ke-Yuan Shen. Learn hybrid prototypes for multivariate time series anomaly detection. In The Thirteenth International Conference on Learning Representations, 2025

  25. [28]

    Carla: Self-supervised contrastive representation learning for time series anomaly detection

    Zahra Zamanzadeh Darban, Geoffrey I Webb, Shirui Pan, Charu C Aggarwal, and Mahsa Salehi. Carla: Self-supervised contrastive representation learning for time series anomaly detection. Pattern Recognition, 157: 0 110874, 2025

  26. [30]

    Rtdetector: Deep transformer networks for time series anomaly detection based on reconstruction trend

    Xinhong Liu, Xiaoliang Li, Yangfan Li, Fengxiao Tang, and Ming Zhao. Rtdetector: Deep transformer networks for time series anomaly detection based on reconstruction trend. In Proceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence, IJCAI-25, J. Kwok, Ed. International Joint Conferences on Artificial Intelligence Organiza...

  27. [31]

    Bearing data center: Apparatus & procedures

    Case Western Reserve University Bearing Data Center . Bearing data center: Apparatus & procedures. https://engineering.case.edu/bearingdatacenter/apparatus-and-procedures, 2024. Accessed: 2026-05-30

  28. [32]

    2009 phm challenge competition data set

    PHM Society . 2009 phm challenge competition data set. https://phmsociety.org/public-data-sets/, 2009. Accessed: 2026-05-30

  29. [33]

    Xjtu-sy bearing datasets

    Biao Wang. Xjtu-sy bearing datasets. https://biaowang.tech/xjtu-sy-bearing-datasets/, 2021. Accessed: 2026-05-30

  30. [34]

    Ims bearings

    NASA Open Data Portal . Ims bearings. https://data.nasa.gov/dataset/ims-bearings, 2023. Experiments on bearings provided by the Center for Intelligent Maintenance Systems (IMS), University of Cincinnati. Accessed: 2026-05-30

  31. [35]

    Krishna, Marine Bruel, Qinghua Liu, Mingyi Huang, Themis Palpanas, Ruey S

    Paul Boniol, Ashwin K. Krishna, Marine Bruel, Qinghua Liu, Mingyi Huang, Themis Palpanas, Ruey S. Tsay, Aaron Elmore, Michael J. Franklin, and John Paparrizos. Vus: effective and efficient accuracy measures for time-series anomaly detection. The VLDB Journal, 34: 0 32, 2025. doi:10.1007/s00778-025-00907-x

  32. [36]

    Tsay, Themis Palpanas, and Michael J

    John Paparrizos, Yuhao Kang, Paul Boniol, Ruey S. Tsay, Themis Palpanas, and Michael J. Franklin. Tsb-uad: An end-to-end benchmark suite for univariate time-series anomaly detection. Proceedings of the VLDB Endowment, 15 0 (8): 0 1697--1711, 2022. doi:10.14778/3529337.3529354

  33. [37]

    AIP Advances , volume=

    A comprehensive review of deep learning-based fault diagnosis approaches for rolling bearings: Advancements and challenges , author=. AIP Advances , volume=. 2025 , publisher=

  34. [38]

    Reliability Engineering & System Safety , volume=

    Rolling bearing intelligent fault diagnosis towards variable speed and imbalanced samples using multiscale dynamic supervised contrast learning , author=. Reliability Engineering & System Safety , volume=. 2024 , publisher=

  35. [39]

    Advanced Engineering Informatics , volume=

    Time-frequency supervised contrastive learning via pseudo-labeling: An unsupervised domain adaptation network for rolling bearing fault diagnosis under time-varying speeds , author=. Advanced Engineering Informatics , volume=. 2024 , publisher=

  36. [40]

    IEEE/ASME Transactions on Mechatronics , volume=

    CCFT: The convolution and cross-fusion transformer for fault diagnosis of bearings , author=. IEEE/ASME Transactions on Mechatronics , volume=. 2023 , publisher=

  37. [41]

    Sensors , volume=

    A novel method for rolling bearing fault diagnosis based on Gramian angular field and CNN-ViT , author=. Sensors , volume=. 2024 , publisher=

  38. [42]

    Sensors , volume=

    A deep learning-based method for bearing fault diagnosis with few-shot learning , author=. Sensors , volume=. 2024 , publisher=

  39. [43]

    Sensors , volume=

    MSFF-Net: Multi-sensor frequency-domain feature fusion network with lightweight 1D CNN for bearing fault diagnosis , author=. Sensors , volume=. 2025 , publisher=

  40. [44]

    Digital Signal Processing , volume=

    Multi-sensor fusion rolling bearing intelligent fault diagnosis based on VMD and ultra-lightweight GoogLeNet in industrial environments , author=. Digital Signal Processing , volume=. 2024 , publisher=

  41. [45]

    Advanced Engineering Informatics , volume=

    Multi-sensor data fusion-enabled lightweight convolutional double regularization contrast transformer for aerospace bearing small samples fault diagnosis , author=. Advanced Engineering Informatics , volume=. 2024 , publisher=

  42. [46]

    Reliability Engineering & System Safety , volume=

    AGFCN: A bearing fault diagnosis method for high-speed train bogie under complex working conditions , author=. Reliability Engineering & System Safety , volume=. 2025 , publisher=

  43. [47]

    Expert Systems with Applications , volume=

    Zero-Sample fault diagnosis of rolling bearings via fault spectrum knowledge and autonomous contrastive learning , author=. Expert Systems with Applications , volume=. 2025 , publisher=

  44. [48]

    Mechanical Systems and Signal Processing , volume=

    Frequency pattern graph spectrum model and its applications in rolling bearing fault diagnosis , author=. Mechanical Systems and Signal Processing , volume=. 2025 , publisher=

  45. [49]

    Expert Systems with Applications , volume=

    MRCFN: A multi-sensor residual convolutional fusion network for intelligent fault diagnosis of bearings in noisy and small sample scenarios , author=. Expert Systems with Applications , volume=. 2025 , publisher=

  46. [50]

    Mechanical Systems and Signal Processing , volume=

    Dynamic model-assisted disentanglement framework for rolling bearing fault diagnosis under time-varying speed conditions , author=. Mechanical Systems and Signal Processing , volume=. 2025 , publisher=

  47. [51]

    Mechanical Systems and Signal Processing , volume=

    Intelligent diagnosis of rolling bearings under cross-domain missing data: A lightweight complex domain imputation and unsupervised time--frequency alignment approach , author=. Mechanical Systems and Signal Processing , volume=. 2025 , publisher=

  48. [52]

    ACM Computing Surveys , volume=

    Deep learning for time series anomaly detection: A survey , author=. ACM Computing Surveys , volume=. 2024 , publisher=

  49. [53]

    Advances in Neural Information Processing Systems , volume=

    The elephant in the room: Towards a reliable time-series anomaly detection benchmark , author=. Advances in Neural Information Processing Systems , volume=

  50. [54]

    arXiv preprint arXiv:2506.18046 , year=

    Tab: Unified benchmarking of time series anomaly detection methods , author=. arXiv preprint arXiv:2506.18046 , year=

  51. [55]

    2024 IEEE 40th International Conference on Data Engineering (ICDE) , pages=

    Learning multi-pattern normalities in the frequency domain for efficient time series anomaly detection , author=. 2024 IEEE 40th International Conference on Data Engineering (ICDE) , pages=. 2024 , organization=

  52. [56]

    2024 IEEE 40th International Conference on Data Engineering (ICDE) , pages=

    Unraveling the ‘anomaly’in time series anomaly detection: a self-supervised tri-domain solution , author=. 2024 IEEE 40th International Conference on Data Engineering (ICDE) , pages=. 2024 , organization=

  53. [57]

    2024 IEEE 40th international conference on data engineering (ICDE) , pages=

    Temporal-frequency masked autoencoders for time series anomaly detection , author=. 2024 IEEE 40th international conference on data engineering (ICDE) , pages=. 2024 , organization=

  54. [58]

    arXiv preprint arXiv:2307.00754 , year=

    Imdiffusion: Imputed diffusion models for multivariate time series anomaly detection , author=. arXiv preprint arXiv:2307.00754 , year=

  55. [59]

    arXiv preprint arXiv:2312.16831 , year=

    Meter: A dynamic concept adaptation framework for online anomaly detection , author=. arXiv preprint arXiv:2312.16831 , year=

  56. [60]

    2024 IEEE International Conference on Data Mining (ICDM) , pages=

    Interdependency matters: graph alignment for multivariate time series anomaly detection , author=. 2024 IEEE International Conference on Data Mining (ICDM) , pages=. 2024 , organization=

  57. [61]

    International conference on learning representations , volume=

    Catch: Channel-aware multivariate time series anomaly detection via frequency patching , author=. International conference on learning representations , volume=

  58. [62]

    International Conference on Learning Representations , volume=

    Towards a general time series anomaly detector with adaptive bottlenecks and dual adversarial decoders , author=. International Conference on Learning Representations , volume=

  59. [63]

    The Thirteenth International Conference on Learning Representations , year=

    Learn hybrid prototypes for multivariate time series anomaly detection , author=. The Thirteenth International Conference on Learning Representations , year=

  60. [64]

    Pattern Recognition , volume=

    CARLA: Self-supervised contrastive representation learning for time series anomaly detection , author=. Pattern Recognition , volume=. 2025 , publisher=

  61. [65]

    arXiv preprint arXiv:2411.00278 , year=

    KAN-AD: Time series anomaly detection with Kolmogorov-Arnold networks , author=. arXiv preprint arXiv:2411.00278 , year=

  62. [66]

    Proceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence, IJCAI-25, J

    Rtdetector: Deep transformer networks for time series anomaly detection based on reconstruction trend , author=. Proceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence, IJCAI-25, J. Kwok, Ed. International Joint Conferences on Artificial Intelligence Organization , volume=

  63. [67]

    2024 , howpublished=

    Bearing Data Center: Apparatus & Procedures , author=. 2024 , howpublished=

  64. [68]

    2009 , howpublished=

    2009 PHM Challenge Competition Data Set , author=. 2009 , howpublished=

  65. [69]

    2021 , howpublished=

    XJTU-SY Bearing Datasets , author=. 2021 , howpublished=

  66. [70]

    2023 , howpublished=

    IMS Bearings , author=. 2023 , howpublished=

  67. [71]

    Proceedings of the VLDB Endowment , volume=

    TSB-UAD: An End-to-End Benchmark Suite for Univariate Time-Series Anomaly Detection , author=. Proceedings of the VLDB Endowment , volume=. 2022 , doi=

  68. [72]

    The VLDB Journal , volume=

    VUS: effective and efficient accuracy measures for time-series anomaly detection , author=. The VLDB Journal , volume=. 2025 , doi=