Pith. sign in

REVIEW 4 major objections 3 minor 59 references

This paper claims that machinery fault detection can be cast as adversarial inverse reinforcement learning over observed state transitions—using the next sensor reading as a proxy action—so that a 'health' reward learned from healthy data a

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 ↗

T0 review · deepseek-v4-flash

2026-08-01 03:55 UTC pith:UNU55EBE

load-bearing objection A useful conceptual critique of contextual-bandit MFD, but the headline PDC claim collapses under the paper's own recommended deployment threshold. the 4 major comments →

arxiv 2607.22987 v1 pith:UNU55EBE submitted 2026-07-25 cs.LG cs.AI

Label-free Industrial Fault Detection via Adversarial Inverse Reinforcement Learning: A System for Run-to-Failure Prognostics

Dhiraj Neupane , Mohamed Reda Bouadjenek , Richard Dazeley , Sunil Aryal This is my paper
classification cs.LG cs.AI
keywords machinery fault detectionadversarial inverse reinforcement learningrun-to-failure prognosticslabel-free anomaly detectionstate-only imitationpost-detection consistencytemporal credit assignmentindustrial sensor monitoring
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 paper claims that treating machinery fault detection as an adversarial inverse reinforcement learning problem—learning a reward that scores how plausibly each sensor reading follows from the previous one, with the next observation serving as a proxy action—lets a model detect incipient faults from healthy-only training data. Across three run-to-failure benchmarks (a helicopter gearbox and two bearing datasets), the authors report that this method is the only one whose post-detection consistency remains non-saturated on all three datasets, avoiding the two failure modes that dominate baselines: constant false-alarm saturation or complete detection collapse. If correct, this provides a label-free prognostic path that requires no fault examples, no manual reward design, and only a few hundred parameters at deployment.

Core claim

On the paper's own terms, the discovery is that the temporal structure of degradation is the information that matters, and that structure is accessible without labels. By learning a reward over consecutive observation pairs from healthy data alone—rather than classifying individual segments as normal or faulty—AIRL produces an anomaly score that behaves differently from both reconstruction-error methods and contextual-bandit classifiers. The paper argues that baselines fail in two characteristic ways: reconstruction and one-class methods saturate, flagging nearly every file so that high detection rates are artifacts of false-alarm persistence, while the bandit formulation collapses to a stat

What carries the argument

The load-bearing mechanism is the AIRL discriminator, which separates expert (healthy) transitions from policy-generated ones using the identity D(s,a,s′) = exp(r(s,a)+γV(s′)−V(s)) divided by exp(r(s,a)+γV(s′)−V(s)) plus π(a|s), where the 'action' a is the next sensor observation x_{t+1} (the state-only imitation trick), r is the learned reward, V a value function, π a stochastic Gaussian policy, and γ=0.99. The discriminator's confidence is converted into an anomaly score 1−D, and thresholds are calibrated on healthy training scores. All three networks are two-layer MLPs with 64 hidden units and a total of 644 parameters, so scoring a test file amounts to one forward pass over a single tran

Load-bearing premise

The headline cross-dataset consistency claim assumes that a global threshold fitted to the entire test-score distribution is a valid measure of post-detection reliability; under the paper's own deployment-valid threshold (Max−2SE computed from healthy training scores alone), AIRL flags only a single IMS file across the whole test set, so the 'only method with non-saturated PDC on all three datasets' claim does not survive the paper's recommended deployment protocol.

What would settle it

On the IMS bearing dataset, compute post-onset detection consistency for AIRL using the Max−2SE threshold calibrated strictly from healthy training scores (the paper's deployment recommendation) instead of a threshold fitted to the full test set. If the number of post-onset flagged files collapses from thousands to one, the claim of consistent non-saturated tracking across all three datasets fails.

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

If this is right

  • Label-free deployment: a prognostic system could be trained on healthy-only vibration data and score new files at runtime without fault labels or manual reward engineering.
  • The contextual-bandit critique generalizes: 'RL-based' fault detectors that ignore state transitions and discounting are methodologically equivalent to supervised classifiers and cannot be expected to track gradual degradation.
  • Saturation-aware evaluation: post-detection consistency must be read together with false-alarm rate; AUROC computed against a rigid onset boundary is structurally inappropriate for run-to-failure data.
  • Edge compatibility: the 644-parameter model and constant per-file inference cost make the approach compatible with controllers that cannot buffer full windows.

Where Pith is reading between the lines

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

  • The paper's own deployment recommendation (Max−2SE from healthy scores) is not the threshold used for its headline consistency claim (a global adaptive threshold fitted to the full test distribution); reconciling this gap is necessary before the headline result transfers to practice.
  • Because the proxy action is deterministic (the next observation), the learned reward is effectively a transition-density estimate; comparing AIRL to a simple density model over consecutive pairs would isolate what the adversarial training adds.
  • A natural transfer test the paper does not run: train on healthy windows of one bearing run and score a different bearing run of the same type under varying load; the claimed invariance of the reward to dynamics predicts such transfer would work, and it is directly testable.
  • The PDC metric is vulnerable to a detector that triggers early and never switches off; the saturation flag (FAR>90%) is a heuristic patch, and a per-file alert-rate normalization or a precision-recall-style summary might be a more principled fix.

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

4 major / 3 minor

Summary. The paper proposes an adversarial inverse reinforcement learning (AIRL) framework for label-free machinery fault detection, treating raw vibration scalar transitions as state-action pairs and recovering a health reward from healthy training trajectories. It evaluates AIRL on three run-to-failure datasets (HUMS2023, IMS, XJTU-SY) and compares against one-class, reconstruction, and contextual-bandit baselines. The headline claim is that AIRL is the only method achieving meaningful, non-saturated post-detection consistency (PDC) across all three datasets, supported by Table 4 which reports PDC values of 79.09%, 90.15%, and 68.38% under an Otsu threshold. The paper also provides a threshold sensitivity analysis, recommends max−2SE for deployment, and includes an ablation on the discount factor.

Significance. If the headline claim were valid, the paper would offer a practically interesting demonstration of IRL for prognostics: a compact (644-parameter) model that requires no fault labels, with a systematic comparison to bandit-style RL baselines and a useful critique of the static CB formulation. The threshold sensitivity analysis and the explicit discussion of saturation failure are also valuable contributions. However, the central PDC result is not supported by the paper's own deployment recommendations: Table 5 shows that under the recommended max−2SE threshold, AIRL flags exactly one IMS test file across the entire 8,940-file test set, so the claimed 90.15% PDC on IMS is an artifact of a test-derived Otsu threshold, not a property of the learned reward. This invalidates the abstract's 'only method' claim and the main comparative conclusion.

major comments (4)
  1. [Sec. 5.2 vs. Sec. 3.4] The PDC values in Table 4 are computed using Otsu's threshold 'which partitions test scores globally' (Sec. 5.2). This contradicts Sec. 3.4, which states that 'all thresholds are computed once from healthy training scores and fixed at test time.' A global Otsu threshold requires observing the full test score distribution, including future failures, and is therefore unavailable in a label-free deployment. The central PDC claim, and the abstract's assertion that AIRL is the only method with non-saturated PDC, rest on this hindsight threshold and are not reproducible under a training-only calibration.
  2. [Sec. 5.3, Table 5] Under the paper's own recommended deployment threshold (max−2SE, Sec. 5.4), Table 5 shows that on IMS AIRL flags exactly one anomalous file (Nov 17) across the entire 8,940-file test set. The IMS post-onset period includes thousands of files; one positive implies PDC of roughly 0.03%, not the 90.15% reported in Table 4. The claimed cross-dataset PDC consistency therefore collapses under a threshold that can actually be used in practice. The text in Sec. 5.2 acknowledging the two-threshold framework does not resolve this: Table 4 is not a deployable evaluation.
  3. [Sec. 6, Limitations] The Limitations paragraph states that 'AIRL remains the only method maintaining non-saturated FAR across all three datasets.' This is contradicted by Table 4, where VAE has FAR 89.1%, 50.0%, 46.8% and TCAE has FAR 41.8%, 88.3%, 86.3% on HUMS, IMS, and XJTU-SY respectively — all below the paper's own 90% saturation ceiling. The claim is therefore false under the authors' definition. This is not a minor wording issue; it is part of the concluding summary of the method's advantage.
  4. [Sec. 5.3 vs. Appendix A(iii)] Table 5 shows that on IMS, all max-based thresholds (Max, Max−SE, Max−2SE) produce exactly one anomalous file, reflecting a compressed score range. Appendix A(iii) claims the sequential model 'maintained dense, stable detection across the full threshold hierarchy.' These statements are mutually inconsistent: one point cannot be described as dense or stable. Additionally, the IMS FAR of 88% under Otsu (Table 4) is only two percentage points below the paper's own saturation ceiling, making the 'non-saturated' designation fragile and operationally questionable.
minor comments (3)
  1. [Sec. 4.1 vs. Sec. 5.2] Sec. 4.1 reports the XJTU-SY test split as 'Remaining 18,295' files, but Sec. 5.2 states that CTQN is evaluated on a 16,295-file subset because the last 2,000 test files are withheld as training data. The discrepancy should be reconciled or clarified in the experimental setup.
  2. [Sec. 3.3, Eq. (3)] The notation σ is used both for the logistic sigmoid in Eq. (3) and for the standard deviation of the Gaussian policy in Sec. 3.2. This is confusing; consider using a distinct symbol such as sigmoid or logistic.
  3. [General] The reference to AAMAS 2026 in Sec. 1 and Ref. [26] should be checked for publication status and exact venue details to avoid citing a non-existent or future volume.

Circularity Check

2 steps flagged

Headline 'only non-saturated PDC' claim is built on an Otsu threshold fitted to the test scores; under the paper's own deployment threshold, AIRL's IMS PDC collapses to one flagged file.

specific steps
  1. fitted input called prediction [Section 5.2 (Table 4) vs. Section 3.4 and Table 5]
    "Table 4 PDC uses Otsu's adaptive threshold, which partitions test scores globally to maximize between-class variance. ... All thresholds are computed once from healthy training scores and fixed at test time. ... all three max-based strategies (Max, Max-SE, Max-2SE) converge to exactly one anomalous file across the entire 8,940-file test set: Nov 17 12:22:30."

    The PDC headline (79.09/90.15/68.38%) is generated with a threshold fitted to the full test-score distribution, whereas Section 3.4 promises thresholds computed from healthy training scores only. The 90.15% IMS figure is therefore not a prediction from the learned reward: under the paper's own recommended max-2SE deployment threshold, Table 5 shows exactly one IMS file is flagged after onset, so the PDC ranking is an artifact of using test data to set the evaluation threshold.

  2. other [Section 5.1, XJTU-SY paragraph]
    "Given that the ~500-min reference is itself an approximate onset annotation rather than a hard cutoff, both detections fall within a 45-min precursor window of the verified degradation transition, and are best interpreted as operationally reasonable early warnings rather than false alarms."

    The '45-min precursor window' is introduced only after observing AIRL's detection at minute 459, while baseline detections more than 150 minutes before onset are classified as false alarms. The window is not a pre-specified evaluation criterion but is retrospectively sized to include AIRL's detection, making the 'near-onset' claim true by construction rather than by an independent benchmark.

full rationale

The core IRL derivation (transition construction, AIRL discriminator, reward learning) is self-contained and does not reduce to a fitted output or self-citation chain; the self-citations [23-26] are background/prior-work citations, not load-bearing. However, the central comparative claim — that AIRL alone achieves non-saturated PDC across all three datasets — is not a free prediction. It is computed under an Otsu threshold that the paper itself says is fitted to the global test-score distribution, contradicting the Section 3.4 promise that all thresholds are fixed from healthy training data. Under the paper's own deployment-recommended max-2SE threshold, AIRL flags exactly one IMS file across 8,940, so the 90.15% PDC is an evaluation artifact. The XJTU-SY 'precursor window' is similarly defined post hoc to include AIRL's pre-onset detection. These issues make the headline result partially circular even though the learning method itself is not.

Axiom & Free-Parameter Ledger

6 free parameters · 6 axioms · 0 invented entities

The paper introduces no new physical entities. Its main postulates are modeling and evaluation choices: proxy actions, healthy-only training, ground-truth onset annotations, and test-derived thresholds. The free parameters listed are choices that affect the central PDC claim.

free parameters (6)
  • discount factor gamma = 0.99
    Chosen by hand; central to the claim that temporal credit assignment helps; the ablation uses 0.
  • entropy bonus beta = 0.01
    Chosen by hand; affects exploration in policy rollouts.
  • max-2SE threshold = recommended default
    Selected after seeing cross-dataset detection results; under this threshold IMS has only one detection.
  • Otsu global threshold = computed from test score distribution
    Used for Table 4 PDC; not a label-free deployment threshold.
  • GPD tail fraction = top 5%
    One of seven threshold strategies; arbitrary choice.
  • XJTU-SY precursor window = 45 minutes before onset
    Post hoc reinterpretation of the 459-min detection as reasonable rather than a false alarm.
axioms (6)
  • standard math AIRL/GAN training converges and the discriminator approximates the density ratio between expert and policy transitions.
    Assumed from Fu et al. [9]; no convergence guarantees are checked in this paper.
  • domain assumption Next-observation proxy action (SOIL) adequately captures fault progression dynamics.
    Section 3.1 constructs a_t = s_{t+1} = x_{t+1}; the method's validity depends on this proxy being a meaningful action.
  • domain assumption Healthy training partitions are fault-free and representative.
    Section 4.1 uses Day 17-20 for HUMS, Oct for IMS, first 2000 segments for XJTU-SY; contamination would invalidate the reward.
  • domain assumption Ground-truth onset annotations (HUMS Day 24 #264; IMS mid-November; XJTU-SY ~500 min) are correct.
    Section 4.1 relies on external annotations; HUMS is also described as 'onset spans Days 22–24', creating ambiguity.
  • ad hoc to paper FAR < 90% defines non-saturation.
    Section 5.2 uses this cutoff to label bracketed entries as saturation failure; it is not derived from any operational requirement.
  • ad hoc to paper XJTU-SY detections within 45 minutes before the ~500-min onset are operationally reasonable early warnings.
    Section 5.1 introduces this precursor framing after the fact to avoid classifying AIRL's 459-min detection as a pre-onset false alarm.

pith-pipeline@v1.3.0-alltime-deepseek · 14384 in / 14370 out tokens · 143877 ms · 2026-08-01T03:55:56.222643+00:00 · methodology

0 comments
read the original abstract

Machinery fault detection (MFD) remains heavily reliant on supervised learning, which struggles with the scarcity of fault labels in real-world settings. While reinforcement learning (RL) offers a framework to model the sequential nature of degradation, current ``RL-based'' MFD methods reduce the problem to a static contextual bandit (CB) formulation: by ignoring state transitions and discarding the temporal discount factor, they collapse to standard supervised classification. We propose an adversarial inverse reinforcement learning (AIRL) framework that treats MFD as an offline IRL problem. Unlike reconstruction-based approaches that rely on static error margins, or CBs that ignore dynamics, our method recovers an intrinsic "health" reward directly from observational state transitions, requiring neither manual reward engineering nor fault labels. On three run-to-failure benchmarks (HUMS2023, IMS, XJTU-SY), AIRL is the only method achieving non-saturated post-detection consistency across all datasets, while CB baselines fail to detect gradual degradation and reconstruction models collapse into always-anomalous states. Code and data: https://github.com/dhirajneupane/AIRL-MFD-DN.

Figures

Figures reproduced from arXiv: 2607.22987 by Dhiraj Neupane, Mohamed Reda Bouadjenek, Richard Dazeley, Sunil Aryal.

Figure 1
Figure 1. Figure 1: Cracked planet gear: crack path indicated by red [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Comparison of Earliest Fault Detection for [PITH_FULL_IMAGE:figures/full_fig_p004_2.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

59 extracted references · 1 canonical work pages

  1. [1]

    Julien Audibert, Pietro Michiardi, Frédéric Guyard, Sébastien Marti, and Maria A Zuluaga. 2020. USAD: Unsupervised anomaly detection on multivariate time series. InProceedings of the 26th ACM SIGKDD International Conference on Knowl- edge Discovery & Data Mining. 3395–3404

    2020

  2. [2]

    Shaojie Bai, J Zico Kolter, and Vladlen Koltun. 2018. An empirical evaluation of generic convolutional and recurrent networks for sequence modeling.arXiv preprint arXiv:1803.01271(2018)

    Pith/arXiv arXiv 2018

  3. [3]

    Andrew G Barto and Thomas G Dietterich. 2004. Reinforcement learning and its relationship to supervised learning.Handbook of learning and approximate dynamic programming10 (2004), 9780470544785

    2004

  4. [4]

    Tarek Berghout, Mohamed Benbouzid, and Leïla-Hayet Mouss. 2021. Leveraging label information in a knowledge-driven approach for rolling-element bearings remaining useful life prediction.Energies14, 8 (2021), 2163. Neupane et al

    2021

  5. [5]

    Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. 2020. A simple framework for contrastive learning of visual representations. InInter- national conference on machine learning. PmLR, 1597–1607

    2020

  6. [6]

    Junyoung Chung, Kyle Kastner, Laurent Dinh, Kratarth Goel, Aaron C Courville, and Yoshua Bengio. 2015. A recurrent latent variable model for sequential data. Advances in neural information processing systems28 (2015)

    2015

  7. [7]

    Saurabh Deshpande, Rahee Walambe, Ketan Kotecha, Ganeshsree Selvachandran, and Ajith Abraham. 2025. Advances and applications in inverse reinforcement learning: a comprehensive review.Neural Computing and Applications(2025), 1–53

    2025

  8. [8]

    Yu Ding, Liang Ma, Jian Ma, Mingliang Suo, Laifa Tao, Yujie Cheng, and Chen Lu

  9. [9]

    Justin Fu, Katie Luo, and Sergey Levine. 2018. Learning Robust Rewards with Ad- versarial Inverse Reinforcement Learning. InInternational Conference on Learning Representations (ICLR). https://openreview.net/forum?id=rkHywl-A-

    2018

  10. [10]

    Bo-Ruei Huang, Chun-Kai Yang, Chun-Mao Lai, Dai-Jie Wu, and Shao-Hua Sun. 2024. Diffusion Imitation from Observation. InAdvances in Neural In- formation Processing Systems, A. Globerson, L. Mackey, D. Belgrave, A. Fan, U. Paquet, J. Tomczak, and C. Zhang (Eds.), Vol. 37. Curran Associates, Inc., 137190–137217. https://proceedings.neurips.cc/paper_files/pa...

    2024

  11. [11]

    Fasikaw Kibrete, Dereje Engida Woldemichael, and Hailu Shimels Gebremedhen

  12. [12]

    Diederik P Kingma and Max Welling. 2013. Auto-encoding variational Bayes. arXiv preprint arXiv:1312.6114(2013)

    Pith/arXiv arXiv 2013

  13. [13]

    Yaguo Lei, Naipeng Li, Liang Guo, Ningbo Li, Tao Yan, and Jing Lin. 2018. Ma- chinery health prognostics: A systematic review from data acquisition to RUL prediction.Mechanical systems and signal processing104 (2018), 799–834

    2018

  14. [14]

    Guoqiang Li, Jun Wu, Chao Deng, Xuebing Xu, and Xinyu Shao. 2021. Deep reinforcement learning-based online domain adaptation method for fault diagno- sis of rotating machinery.IEEE/ASME Transactions on Mechatronics27, 5 (2021), 2796–2805

    2021

  15. [15]

    Xiang Li, Qian Ding, and Jian-Qiao Sun. 2018. Remaining useful life estimation in prognostics using deep convolution neural networks.Reliability Engineering & System Safety172 (2018), 1–11

    2018

  16. [16]

    Zhenning Li, Hongkai Jiang, and Yutong Dong. 2025. A convolutional- transformer reinforcement learning agent for rotating machinery fault diagnosis. Expert Systems with Applications271 (2025), 126669

    2025

  17. [17]

    Zhenning Li, Hongkai Jiang, and Xin Wang. 2025. A novel reinforcement learning agent for rotating machinery fault diagnosis with data augmentation.Reliability Engineering & System Safety253 (2025), 110570

    2025

  18. [18]

    Fei Tony Liu, Kai Ming Ting, and Zhi-Hua Zhou. 2008. Isolation Forest. In2008 Eighth IEEE International Conference on Data Mining. IEEE, 413–422

    2008

  19. [19]

    Pankaj Malhotra, Anusha Ramakrishnan, Gaurangi Anand, Lovekesh Vig, Puneet Agarwal, and Gautam Shroff. 2016. LSTM-based encoder-decoder for multi- sensor anomaly detection.arXiv preprint arXiv:1607.00148(2016)

    Pith/arXiv arXiv 2016

  20. [20]

    Wentao Mao, Zongtao Chen, Yanna Zhang, and Zhidan Zhong. 2024. Harmony better than uniformity: A new pre-training anomaly detection method with tensor domain adaptation for early fault evaluation.Engineering Applications of Artificial Intelligence127 (2024), 107427

    2024

  21. [21]

    Wentao Mao, Di Zhang, Siyu Tian, and Jiamei Tang. 2020. Robust detection of bearing early fault based on deep transfer learning.Electronics9, 2 (2020), 323

    2020

  22. [22]

    James B McQueen. 1967. Some methods of classification and analysis of multi- variate observations. InProc. of 5th Berkeley Symposium on Math. Stat. and Prob. 281–297

    1967

  23. [25]

    Dhiraj Neupane, Mohamed Reda Bouadjenek, Richard Dazeley, and Sunil Aryal

  24. [26]

    InPHM Society European Conference, Vol

    A Comparative Study of Semi-Supervised Anomaly Detection Methods for Machine Fault Detection. InPHM Society European Conference, Vol. 8. 10–10

  25. [27]

    Andrew Y Ng, Stuart Russell, et al. 2000. Algorithms for inverse reinforcement learning.. InIcml, Vol. 1. 2

    2000

  26. [28]

    InPHM Society European Conference, Vol

    Machinery Fault Detection using Advanced Machine Learning Techniques. InPHM Society European Conference, Vol. 8. 4–4

  27. [29]

    P Nunes, J Santos, and E Rocha. 2023. Challenges in predictive maintenance–A review.CIRP Journal of Manufacturing Science and Technology40 (2023), 53–67

    2023

  28. [30]

    Nobuyuki Otsu et al. 1975. A threshold selection method from gray-level his- tograms.Automatica11, 285-296 (1975), 23–27

    1975

  29. [31]

    Dhiraj Neupane, Richard Dazeley, Mohamed Reda Bouadjenek, and Sunil Aryal. 2026. Learning Rewards, Not Labels: Adversarial Inverse Reinforce- ment Learning for Machinery Fault Detection. InProceedings of the 25th In- ternational Conference on Autonomous Agents and Multiagent Systems (AAMAS). doi:10.65109/AXYX4522

    work page doi:10.65109/axyx4522 2026

  30. [32]

    James Pickands III. 1975. Statistical inference using extreme order statistics.the Annals of Statistics(1975), 119–131

    1975

  31. [33]

    Rui Nian, Jinfeng Liu, and Biao Huang. 2020. A review on reinforcement learn- ing: Introduction and applications in industrial process control.Computers & Chemical Engineering139 (2020), 106886

    2020

  32. [34]

    José Miguel Ramírez-Sanz, Jose-Alberto Maestro-Prieto, Álvar Arnaiz-González, and Andrés Bustillo. 2023. Semi-supervised learning for industrial fault detection and diagnosis: A systemic review.ISA Transactions(2023)

    2023

  33. [35]

    Maria Andrea Romo-Chavero, Jose Antonio Cantoral-Ceballos, Jesus Arturo Pérez-Díaz, and Carlos Martinez-Cagnazzo. 2024. Median absolute deviation for BGP anomaly detection.Future Internet16, 5 (2024), 146

    2024

  34. [36]

    Cédric Peeters, Wenyi Wang, David Blunt, Timothy Verstraeten, and Jan Helsen

  35. [37]

    Fatigue crack detection in planetary gears: Insights from the HUMS2023 data challenge.Mechanical Systems and Signal Processing212 (2024), 111292

    2024

  36. [38]

    Nader Sawalhi, Wenyi Wang, and David Blunt. 2024. Helicopter planet gear rim crack diagnosis and trending using cepstrum editing enhanced with deconvolu- tion.Sensors24, 8 (2024), 2593

    2024

  37. [39]

    Gensheng Qian and Jingquan Liu. 2022. Development of deep reinforcement learning-based fault diagnosis method for rotating machinery in nuclear power plants.Progress in Nuclear Energy152 (2022), 104401

    2022

  38. [40]

    Sutton and Andrew G

    Richard S. Sutton and Andrew G. Barto. 2020.Reinforcement Learning: An Intro- duction(second ed.). MIT Press, Cambridge, Massachusetts London, England

    2020

  39. [41]

    Faraz Torabi, Garrett Warnell, and Peter Stone. 2018. Generative adversarial imitation from observation.arXiv preprint arXiv:1807.06158(2018)

    Pith/arXiv arXiv 2018

  40. [42]

    Diletta Sacerdoti, Matteo Strozzi, and Cristian Secchi. 2023. A comparison of signal analysis techniques for the diagnostics of the IMS rolling element bearing dataset.Applied Sciences13, 10 (2023), 5977

    2023

  41. [43]

    M Saied, N Tahan, K Chreif, C Francis, and Z Noun. 2025. Development of deep reinforcement learning-based fault diagnosis method for actuator faults in unmanned aerial vehicles.The Aeronautical Journal(2025), 1–17

    2025

  42. [44]

    Rik Vaerenberg, Alex Ricardo Mauricio, and Konstantinos Gryllias. 2025. Detect- ing planet gear crack propagation using FRESH filters. In14th Defence Science & Technology (DST) International Conference on Health and Usage Monitoring HUMS2025 Proceedings

    2025

  43. [45]

    Bernhard Schölkopf, Robert C Williamson, Alex Smola, John Shawe-Taylor, and John Platt. 1999. Support vector method for novelty detection.Advances in neural information processing systems12 (1999)

    1999

  44. [46]

    Wenyi Wang, David Blunt, and J Kappas. 2023. Helicopter main gearbox planet gear crack propagation test dataset

    2023

  45. [47]

    W Wang, K Vos, J Taylor, C Jenkins, B Bala, L Whitehead, and Z Peng. 2023. Is deep learning superior to traditional techniques in machine health monitoring applications.The Aeronautical Journal127, 1318 (2023), 2105–2117

    2023

  46. [48]

    Faraz Torabi, Garrett Warnell, and Peter Stone. 2019. Adversarial imitation learning from state-only demonstrations. InProceedings of the 18th International Conference on Autonomous Agents and MultiAgent Systems. 2229–2231

    2019

  47. [49]

    David Umsonst, Justin Ruths, and Henrik Sandberg. 2022. Finite sample guaran- tees for quantile estimation: An application to detector threshold tuning.IEEE Transactions on Control Systems Technology31, 2 (2022), 921–928

    2022

  48. [50]

    Daoguang Yang, Hamid Reza Karimi, and Marek Pawelczyk. 2023. A new intelli- gent fault diagnosis framework for rotating machinery based on deep transfer reinforcement learning.Control Engineering Practice134 (2023), 105475

    2023

  49. [51]

    Biao Wang, Yaguo Lei, Naipeng Li, and Ningbo Li. 2020. A Hybrid Prognostics Approach for Estimating Remaining Useful Life of Rolling Element Bearings.IEEE Transactions on Reliability69, 1 (2020), 401–412. doi:10.1109/TR.2018.2882682

    arXiv 2020

  50. [52]

    Shen Zhang, Shibo Zhang, Bingnan Wang, and Thomas G Habetler. 2020. Deep learning algorithms for bearing fault diagnostics—A comprehensive review.IEEE access8 (2020), 29857–29881

    2020

  51. [53]

    Zhibin Zhao, Qiyang Zhang, Xiaolei Yu, Chuang Sun, Shibin Wang, Ruqiang Yan, and Xuefeng Chen. 2021. Applications of Unsupervised Deep Transfer Learning to Intelligent Fault Diagnosis: A Survey and Comparative Study.IEEE Transactions on Instrumentation and Measurement70 (2021), 1–28. doi:10.1109/ TIM.2021.3116309

    arXiv 2021

  52. [54]

    Shuai Xiao, Qingsheng Feng, Xue Li, and Hong Li. 2024. Research on intelligent fault diagnosis for railway point machines using deep reinforcement learning. Transportation Safety and Environment6, 4 (2024), tdae007

    2024

  53. [55]

    Jiehui Xu, Haixu Wu, Jianmin Wang, and Mingsheng Long. 2022. Anomaly Transformer: Time series anomaly detection with association discrepancy. In International Conference on Learning Representations (ICLR)

    2022

  54. [57]

    Xue Yang, Enda Howley, and Michael Schukat. 2025. Agent-based dynamic thresholding for adaptive anomaly detection using reinforcement learning.Neu- ral Computing and Applications37, 23 (2025), 18775–18791

    2025

  55. [60]

    Chong Zhou and Randy C Paffenroth. 2017. Anomaly detection with robust deep autoencoders. InProceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. 665–674

    2017

  56. [61]

    Zhuangdi Zhu, Kaixiang Lin, Bo Dai, and Jiayu Zhou. 2020. Off-policy imitation learning from observations.Advances in neural information processing systems Label-free Industrial Fault Detection via Adversarial Inverse Reinforcement Learning: A System for Run-to-Failure Prognostics 33 (2020), 12402–12413

    2020

  57. [2019]

    Intelligent fault diagnosis for rotating machinery using deep Q-network based health state classification: A deep reinforcement learning approach.Ad- vanced Engineering Informatics42 (2019), 100977

    2019

  58. [2024]

    Multi-Sensor data fusion in intelligent fault diagnosis of rotating machines: A comprehensive review.Measurement(2024), 114658

    2024

  59. [2025]

    Data-driven machinery fault diagnosis: A comprehensive review.Neuro- computing(2025), 129588

    2025