pith. sign in

arxiv: 2603.13343 · v2 · submitted 2026-03-07 · 💻 cs.LG · cs.AI· cs.SY· eess.SY

AI-Driven Predictive Maintenance with Environmental Context Integration for Connected Vehicles: Simulation, Benchmarking, and Field Validation

Kushal Khemani (Independent Researcher , India) , Anjum Nazir Qureshi (Rajiv Gandhi College of Engineering Research , Technology) This is my paper

Pith reviewed 2026-05-15 15:11 UTC · model grok-4.3

classification 💻 cs.LG cs.AIcs.SYeess.SY
keywords predictive maintenanceconnected vehiclescontextual data fusionedge inferencewear predictionfield validationenvironmental signals
0
0 comments X

The pith

Integrating vehicle sensors with environmental context detects all wear events with 12.2-day mean error.

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

This paper tries to show that adding environmental context to vehicle sensor data improves predictive maintenance for connected vehicles. The authors build a fusion framework and test it on synthetic data, a standard benchmark, and real trips from vehicles in India, Germany, and Brazil. In the field validation, the model spots every wear-related service event and predicts timing within about 12 days on average. A reader would care because this could lead to more reliable fleets with fewer unexpected repairs by using data from outside the vehicle itself.

Core claim

The central claim is that the contextual data fusion framework, which combines internal vehicle signals with external factors like road quality, weather, traffic density, and driver behaviour, delivers 100% detection of wear-driven maintenance events in real-world conditions. Across six such events from four vehicles, the mean absolute error is 12.2 days. The base model from synthetic training already detects all events, and vehicle-specific fine-tuning improves the timing accuracy.

What carries the argument

The contextual data fusion framework that integrates vehicle-internal sensor streams with external environmental signals acquired via V2X communication and third-party APIs for edge inference.

If this is right

  • Contextual features raise macro F1 by 2.6 points on the physics-informed synthetic dataset.
  • LightGBM reaches AUC-ROC 0.973 on the AI4I 2020 benchmark under stratified cross-validation.
  • Real telemetry from four vehicles yields 100% detection on six wear events with 12.2-day mean MAE.
  • Edge inference reduces latency from 3.5 seconds to under 1 second compared to cloud-only processing.
  • SHAP analysis places contextual and interaction features among the top 15 predictors.

Where Pith is reading between the lines

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

  • The approach could extend to larger fleets to identify region-specific wear patterns not captured in the current three-country sample.
  • V2X integration opens the possibility of dynamic routing that avoids routes likely to accelerate predicted wear.
  • Similar fusion of external context may improve maintenance prediction for other mobile systems such as heavy machinery or rail fleets.
  • Further tests on longer observation windows or additional failure modes would clarify how far the 100% detection result generalizes.

Load-bearing premise

The 11 service events identified from component wear resets in 992 trips across five vehicles are representative and sufficient to support 100% detection and generalizability claims.

What would settle it

A new set of vehicle trips containing a wear-driven service event that the model either misses entirely or predicts with timing error well above 12 days would disprove the detection performance.

Figures

Figures reproduced from arXiv: 2603.13343 by Anjum Nazir Qureshi (Rajiv Gandhi College of Engineering Research, India), Kushal Khemani (Independent Researcher, Technology).

Figure 1
Figure 1. Figure 1: Feature group ablation study. (a) Absolute macro F1-score per configuration with 95% error bars. (b) F1 drop relative to the full model when each [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: AI4I 2020 real-world benchmark results. (a) Binary failure classification F1-score across all models with 5-fold CV error bars. (b) AUC-ROC comparison. [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Multi-model classification benchmark on the synthetic contextual [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: SHAP feature importance for the LightGBM classifier. (a) Top 15 features by mean [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Model robustness to increasing sensor noise. F1 remains above 0.88 [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Probability calibration analysis. Left: uncalibrated LightGBM (Brier [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Service time regression — LightGBM. (a) Actual vs. predicted days until service; near-diagonal scatter indicates high accuracy within the simulation. [PITH_FULL_IMAGE:figures/full_fig_p010_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Field validation results across the five-vehicle fleet pilot. (a) Pre-service window MAE by event category: wear-driven events (red) achieve mean MAE [PITH_FULL_IMAGE:figures/full_fig_p012_8.png] view at source ↗
read the original abstract

Predictive maintenance for connected vehicles offers the potential to reduce unexpected breakdowns and improve fleet reliability, but most existing systems rely exclusively on internal diagnostic signals and are validated on simulated or industrial benchmark data. This paper presents a contextual data fusion framework integrating vehicle-internal sensor streams with external environmental signals -- road quality, weather, traffic density, and driver behaviour -- acquired via V2X communication and third-party APIs, with inference at the vehicle edge. The framework is evaluated across four layers. A feature group ablation study on a physics-informed synthetic dataset shows contextual features contribute a 2.6-point F1 improvement; removing all context reduces macro F1 from 0.855 to 0.807. On the AI4I 2020 benchmark (10,000 samples), LightGBM achieves AUC-ROC 0.973 under 5-fold stratified cross-validation with SMOTE confined to training folds. A noise sensitivity analysis shows macro F1 remains above 0.88 at low noise and degrades to 0.74 at high noise. Most critically, the pipeline is validated on real-world telemetry from five vehicles across three countries (India, Germany, Brazil), comprising 992 trips and 11 evaluable service events identified from component wear resets in the trip logs. Across six wear-driven events spanning four vehicles, the model achieves 100% detection with mean MAE of 12.2 days. A fine-tuning ablation shows the base synthetic model already achieves 6/6 binary detection; per-vehicle adaptation reduces wear-driven MAE from 25.9 to 12.2 days. SHAP analysis confirms contextual and interaction features rank among the top 15 predictors. Edge-based inference reduces estimated latency from 3.5 seconds to under 1.0 second relative to cloud-only processing.

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

3 major / 2 minor

Summary. The paper proposes a contextual data fusion framework for predictive maintenance in connected vehicles that integrates internal sensor streams with external signals (road quality, weather, traffic, driver behaviour) acquired via V2X and third-party APIs, with edge inference. It reports a 2.6-point macro-F1 gain from context on a physics-informed synthetic dataset, AUC-ROC 0.973 on the AI4I 2020 benchmark with LightGBM under 5-fold CV and SMOTE, noise robustness results, and real-world validation on 992 trips from five vehicles yielding 100% detection and 12.2-day mean MAE on six wear-driven events out of 11 total service events.

Significance. If the real-world results can be strengthened, the work would usefully demonstrate the incremental value of environmental context for vehicle-edge predictive maintenance, supported by consistent synthetic ablations, benchmark performance, and SHAP rankings that place contextual features among the top predictors. The synthetic and benchmark layers provide reproducible evidence for the framework's design choices.

major comments (3)
  1. [Real-world telemetry validation] Real-world validation (992 trips, 11 service events): the central claim of 100% detection and 12.2-day MAE on the six wear-driven events is reported without error bars, statistical significance tests, per-trip false-positive counts on the full corpus, or bootstrap/cross-validation, leaving the result vulnerable to overfitting on the specific vehicles, routes, and conditions observed.
  2. [Real-world telemetry validation] Event identification: service events are derived post-hoc from component wear resets in the trip logs, yet no information is given on the total number of candidate resets examined, any discarding criteria, blinding of labeling, or the rationale for retaining only the six wear-driven subset versus the remaining five events.
  3. [Fine-tuning ablation] Fine-tuning ablation: while the base synthetic model already achieves 6/6 binary detection and per-vehicle adaptation lowers MAE from 25.9 to 12.2 days, the manuscript provides no variance estimates, per-event breakdown, or comparison against a non-contextual baseline on the same real-world splits, weakening the attribution of the MAE reduction to contextual features.
minor comments (2)
  1. [Noise sensitivity analysis] The noise-sensitivity analysis states macro F1 remains above 0.88 at low noise and falls to 0.74 at high noise, but the exact noise amplitudes, injection method, and number of trials should be tabulated or plotted for reproducibility.
  2. [Abstract and results] The abstract and results sections use 'mean MAE of 12.2 days' without explicitly defining the prediction horizon or whether MAE is computed only on true-positive detections.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive comments on our manuscript. We address each major comment point-by-point below, proposing revisions to improve statistical rigor and transparency in the real-world validation while honestly noting constraints imposed by the limited number of observed events.

read point-by-point responses

  1. Referee: [Real-world telemetry validation] Real-world validation (992 trips, 11 service events): the central claim of 100% detection and 12.2-day MAE on the six wear-driven events is reported without error bars, statistical significance tests, per-trip false-positive counts on the full corpus, or bootstrap/cross-validation, leaving the result vulnerable to overfitting on the specific vehicles, routes, and conditions observed.

    Authors: We acknowledge the small sample of only six wear-driven events limits traditional statistical power. The model is pre-trained on a large physics-informed synthetic dataset and only lightly fine-tuned per vehicle, which reduces overfitting risk to the specific real-world conditions. We will add bootstrap resampling (1000 iterations) to report 95% confidence intervals around the 12.2-day MAE and 100% detection rate. We will also report false-positive counts across the full 992 trips (currently zero, as all detections matched logged resets). Leave-one-vehicle-out results will be included as a form of cross-validation. These additions address the concern without overstating the data scale. revision: partial

  2. Referee: [Real-world telemetry validation] Event identification: service events are derived post-hoc from component wear resets in the trip logs, yet no information is given on the total number of candidate resets examined, any discarding criteria, blinding of labeling, or the rationale for retaining only the six wear-driven subset versus the remaining five events.

    Authors: The 11 service events represent all wear-metric resets observed in the 992-trip corpus; no additional candidate resets existed and none were discarded. Event identification was performed automatically via predefined sensor-threshold crossings in the trip logs, so no manual labeling occurred and blinding was not applicable. The six wear-driven events were retained because they map directly to the physics-based degradation components in our framework (brakes, tires, suspension, engine). The remaining five events were non-wear actions (e.g., software updates, fluid top-ups) outside the predictive-maintenance scope. We will add this full description, including the total count and selection rationale, to the revised manuscript. revision: yes

  3. Referee: [Fine-tuning ablation] Fine-tuning ablation: while the base synthetic model already achieves 6/6 binary detection and per-vehicle adaptation lowers MAE from 25.9 to 12.2 days, the manuscript provides no variance estimates, per-event breakdown, or comparison against a non-contextual baseline on the same real-world splits, weakening the attribution of the MAE reduction to contextual features.

    Authors: We agree that these details are necessary for clear attribution. We will add a per-event MAE table with bootstrap-derived variance estimates. We will also evaluate and report a non-contextual baseline model trained and tested on the identical real-world splits, allowing direct comparison of the MAE reduction attributable to contextual features. These results will be incorporated into the fine-tuning ablation section. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation relies on independent benchmarks and telemetry

full rationale

The paper evaluates its contextual fusion framework through standard ablation studies and cross-validation on three distinct data sources: a physics-informed synthetic dataset, the public AI4I 2020 benchmark (with 5-fold stratified CV and SMOTE restricted to training folds), and separate real-world telemetry comprising 992 trips from five vehicles. Reported metrics such as F1 improvements, AUC-ROC, and MAE on the six wear-driven events are computed directly from these external inputs without any equations that reduce outputs to quantities defined by the same fitted parameters. No self-citation load-bearing premises, uniqueness theorems from prior author work, or ansatz smuggling appear in the derivation chain. The central performance claims remain statistically independent of the model's own training outputs.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 0 invented entities

The framework rests on standard supervised-learning assumptions plus domain assumptions about data availability and labeling accuracy rather than new physical entities or large numbers of ad-hoc fitted constants.

free parameters (2)
  • LightGBM hyperparameters
    Tuned on synthetic and benchmark data; exact values not stated in abstract.
  • SMOTE oversampling parameters
    Applied only to training folds to address imbalance; specific ratios not reported.
axioms (2)
  • domain assumption External environmental signals from V2X and third-party APIs are timely and accurate
    Required for the data-fusion layer to function as described.
  • domain assumption Component wear resets recorded in trip logs correctly mark service events
    Used to label the 11 evaluable events for real-world validation.

pith-pipeline@v0.9.0 · 5660 in / 1476 out tokens · 60263 ms · 2026-05-15T15:11:09.673955+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

49 extracted references · 49 canonical work pages

  1. [1]

    A review on machinery diagnostics and prognostics implement- ing condition-based maintenance,

    A. K. S. Jardine, D. Lin, and D. Banjevic, “A review on machinery diagnostics and prognostics implement- ing condition-based maintenance,”Mech. Syst. Signal Process., vol. 20, no. 7, pp. 1483–1510, 2006. https: //doi.org/10.1016/j.ymssp.2005.09.012

    work page doi:10.1016/j.ymssp.2005.09.012 2006

  2. [2]

    R. K. Mobley,An Introduction to Predictive Mainte- nance, 2nd ed. Oxford: Butterworth-Heinemann, 2002

    work page 2002

  3. [3]

    A systematic literature review of machine learning methods applied to predictive mainte- nance,

    T. P. Carvalhoet al., “A systematic literature review of machine learning methods applied to predictive mainte- nance,”Comput. Ind. Eng., vol. 137, p. 106024, 2019. https://doi.org/10.1016/j.cie.2019.106024

    work page doi:10.1016/j.cie.2019.106024 2019

  4. [4]

    State-of-the-art predictive mainte- nance techniques,

    H. M. Hashemian, “State-of-the-art predictive mainte- nance techniques,”IEEE Trans. Instrum. Meas., vol. 60, no. 1, pp. 226–236, 2011. https://doi.org/10.1109/TIM. 2010.2047662

    work page doi:10.1109/tim 2011

  5. [5]

    Prognostics and health management design for rotary machinery systems,

    J. Lee, F. Wu, W. Zhao, M. Ghaffari, L. Liao, and D. Siegel, “Prognostics and health management design for rotary machinery systems,”Mech. Syst. Signal Pro- cess., vol. 42, no. 1–2, pp. 314–334, 2014. https://doi. org/10.1016/j.ymssp.2013.06.035

    work page doi:10.1016/j.ymssp.2013.06.035 2014

  6. [6]

    Machine learning in manufacturing: Advantages, chal- lenges, and applications,

    T. Wuest, D. Weimer, C. Irgens, and K. D. Thoben, “Machine learning in manufacturing: Advantages, chal- lenges, and applications,”Prod. Manuf. Res., vol. 4, no. 1, pp. 23–45, 2016

    work page 2016

  7. [7]

    Machine learning for predictive maintenance: A multiple classifier approach,

    G. A. Susto, A. Schirru, S. Pampuri, S. McLoone, and A. Beghi, “Machine learning for predictive maintenance: A multiple classifier approach,”IEEE Trans. Ind. Infor- mat., vol. 11, no. 3, pp. 812–820, 2015. https://doi.org/ 10.1109/TII.2014.2349359

    work page doi:10.1109/tii.2014.2349359 2015

  8. [8]

    Explainable artificial intelligence for predic- tive maintenance applications,

    S. Matzka, “Explainable artificial intelligence for predic- tive maintenance applications,” inProc. 3rd Int. Conf. Artif. Intell. Ind. (AI4I 2020), 2020. UCI ML Repository. https://archive.ics.uci.edu/dataset/601/

    work page 2020

  9. [9]

    En- semble approach for automotive predictive maintenance,

    S. Ferreiro, A. Arnaiz, B. Sierra, and I. Irigoien, “En- semble approach for automotive predictive maintenance,” IEEE Access, vol. 8, pp. 180507–180518, 2020. https: //doi.org/10.1109/ACCESS.2020.3031394

    work page doi:10.1109/access.2020.3031394 2020

  10. [10]

    Data-driven methods for predictive maintenance of industrial equipment: A survey,

    W. Zhang, D. Yang, and H. Wang, “Data-driven methods for predictive maintenance of industrial equipment: A survey,”IEEE Syst. J., vol. 13, no. 3, pp. 2213–2227,

    work page

  11. [11]

    https://doi.org/10.1109/JSYST.2018.2813800

    work page doi:10.1109/jsyst.2018.2813800 2018

  12. [12]

    A review of machine learning techniques for automotive maintenance prediction,

    A. Gupta and M. S. Raval, “A review of machine learning techniques for automotive maintenance prediction,”Int. J. Eng. Res. Technol., vol. 5, no. 10, pp. 111–117, 2016

    work page 2016

  13. [13]

    Multisensor data fusion: A review of the state-of-the- art,

    B. Khaleghi, A. Khamis, F. O. Karray, and S. N. Razavi, “Multisensor data fusion: A review of the state-of-the- art,”Inf. Fusion, vol. 14, no. 1, pp. 28–44, 2013. https: //doi.org/10.1016/j.inffus.2011.08.001

    work page doi:10.1016/j.inffus.2011.08.001 2013

  14. [14]

    An introduction to multisensor data fusion,

    D. L. Hall and J. Llinas, “An introduction to multisensor data fusion,”Proc. IEEE, vol. 85, no. 1, pp. 6–23, 1997. https://doi.org/10.1109/5.554205

    work page doi:10.1109/5.554205 1997

  15. [15]

    Dedicated short-range communications (DSRC) standards in the United States,

    J. B. Kenney, “Dedicated short-range communications (DSRC) standards in the United States,”Proc. IEEE, vol. 99, no. 7, pp. 1162–1182, 2011. https://doi.org/10. 1109/JPROC.2011.2132790

    work page arXiv 2011

  16. [16]

    LTE evolution for V2X services,

    H. Seoet al., “LTE evolution for V2X services,”IEEE Commun. Mag., vol. 54, no. 6, pp. 22–28, 2016. https: //doi.org/10.1109/MCOM.2016.7452272

    work page doi:10.1109/mcom.2016.7452272 2016

  17. [17]

    Edge computing: Vision and challenges,

    W. Shi, J. Cao, Q. Zhang, Y . Li, and L. Xu, “Edge computing: Vision and challenges,”IEEE Internet Things J., vol. 3, no. 5, pp. 637–646, 2016. https://doi.org/10. 1109/JIOT.2016.2579198

    work page arXiv 2016

  18. [18]

    A survey on the edge computing for the Internet of Things,

    W. Yuet al., “A survey on the edge computing for the Internet of Things,”IEEE Access, vol. 6, pp. 6900–6919,

    work page

  19. [19]

    https://doi.org/10.1109/ACCESS.2017.2778504

    work page doi:10.1109/access.2017.2778504 2017

  20. [20]

    Adams, and Nando de Freitas

    J. Chen, X. Ran, and X. Ma, “Deep learning with edge computing: A review,”Proc. IEEE, vol. 107, no. 8, 15 pp. 1655–1674, 2019. https://doi.org/10.1109/JPROC. 2019.2921977

    work page doi:10.1109/jproc 2019

  21. [21]

    DeepX: A software accelerator for low-power deep learning inference on mobile devices,

    N. D. Laneet al., “DeepX: A software accelerator for low-power deep learning inference on mobile devices,” inProc. ACM/IEEE IPSN, 2016

    work page 2016

  22. [22]

    LightGBM: A highly efficient gradient boosting decision tree,

    G. Keet al., “LightGBM: A highly efficient gradient boosting decision tree,” inAdv. Neural Inf. Process. Syst., vol. 30, 2017. https://papers.nips.cc/paper files/paper/2017/hash/ 6449f44a102fde848669bdd9eb6b76fa-Abstract.html

    work page 2017

  23. [23]

    Vachtsevanos, F

    G. Vachtsevanos, F. Lewis, M. Roemer, A. Hess, and B. Wu,Intelligent Fault Diagnosis and Prognosis for Engineering Systems. Hoboken, NJ: Wiley, 2006

    work page 2006

  24. [24]

    Main- tenance in digitalised manufacturing: Delphi-based sce- narios for 2030,

    J. Bokrantz, A. Skoogh, C. Berlin, and J. Stahre, “Main- tenance in digitalised manufacturing: Delphi-based sce- narios for 2030,”Int. J. Prod. Econ., vol. 191, pp. 154– 169, 2017. https://doi.org/10.1016/j.ijpe.2017.06.012

    work page doi:10.1016/j.ijpe.2017.06.012 2030

  25. [25]

    Linking maintenance strategies to perfor- mance,

    L. Swanson, “Linking maintenance strategies to perfor- mance,”Int. J. Prod. Econ., vol. 70, no. 3, pp. 237–244,

    work page

  26. [26]

    https://doi.org/10.1016/S0925-5273(00)00128-8

    work page doi:10.1016/s0925-5273(00)00128-8

  27. [27]

    A tutorial survey on vehicular ad hoc networks,

    H. Hartenstein and K. Laberteaux, “A tutorial survey on vehicular ad hoc networks,”IEEE Commun. Mag., vol. 46, no. 6, pp. 164–171, 2008. https://doi.org/10. 1109/MCOM.2008.4539481

    work page arXiv 2008

  28. [28]

    Remain- ing useful life estimation — A review on the statistical data driven approaches,

    X. S. Si, W. Wang, C. H. Hu, and D. H. Zhou, “Remain- ing useful life estimation — A review on the statistical data driven approaches,”Eur. J. Oper. Res., vol. 213, no. 1, pp. 1–14, 2011. https://doi.org/10.1016/j.ejor.2011. 02.021

    work page doi:10.1016/j.ejor.2011 2011

  29. [29]

    Goodfellow, Y

    I. Goodfellow, Y . Bengio, and A. Courville,Deep Learn- ing. Cambridge, MA: MIT Press, 2016. https://www. deeplearningbook.org/

    work page 2016

  30. [30]

    Hastie, R

    T. Hastie, R. Tibshirani, and J. Friedman,The Elements of Statistical Learning, 2nd ed. New York, NY: Springer,

    work page

  31. [31]

    https://hastie.su.domains/ElemStatLearn/

    work page

  32. [32]

    S. I. Nikolenko,Synthetic Data for Deep Learning. Cham: Springer, 2021

    work page 2021

  33. [33]

    Influence of con- nected and autonomous vehicles on traffic flow stability,

    A. Talebpour and H. S. Mahmassani, “Influence of con- nected and autonomous vehicles on traffic flow stability,” Transp. Res. C Emerg. Technol., vol. 71, pp. 143–163,

    work page

  34. [34]

    https://doi.org/10.1016/j.trc.2016.07.009

    work page doi:10.1016/j.trc.2016.07.009 2016

  35. [35]

    Provost and T

    F. Provost and T. Fawcett,Data Science for Business. Sebastopol, CA: O’Reilly Media, 2013

    work page 2013

  36. [36]

    SMOTE: Synthetic minority over-sampling technique,

    N. V . Chawla, K. W. Bowyer, L. O. Hall, and W. P. Kegelmeyer, “SMOTE: Synthetic minority over-sampling technique,”J. Artif. Intell. Res., vol. 16, pp. 321–357,

    work page

  37. [37]

    https://doi.org/10.1613/jair.953

    work page doi:10.1613/jair.953

  38. [38]

    A unified approach to interpreting model predictions,

    S. M. Lundberg and S.-I. Lee, “A unified approach to interpreting model predictions,” in Adv. Neural Inf. Process. Syst., vol. 30, 2017. https://proceedings.neurips.cc/paper/2017/hash/ 8a20a8621978632d76c43dfd28b67767-Abstract.html

    work page 2017

  39. [39]

    Probabilistic outputs for support vector ma- chines and comparisons to regularized likelihood meth- ods,

    J. C. Platt, “Probabilistic outputs for support vector ma- chines and comparisons to regularized likelihood meth- ods,” inAdvances in Large Margin Classifiers, A. Smola, P. Bartlett, B. Sch ¨olkopf, and D. Schuurmans, Eds. Cambridge, MA: MIT Press, 1999, pp. 61–74

    work page 1999

  40. [40]

    A survey on transfer learning,

    S. J. Pan and Q. Yang, “A survey on transfer learning,” IEEE Trans. Knowl. Data Eng., vol. 22, no. 10, pp. 1345– 1359, 2010. https://doi.org/10.1109/TKDE.2009.191

    work page doi:10.1109/tkde.2009.191 2010

  41. [41]

    2021, IEEE Proceedings, 109, 43, doi: 10.1109/JPROC.2020.3004555

    F. Zhuanget al., “A comprehensive survey on transfer learning,”Proc. IEEE, vol. 109, no. 1, pp. 43–76, 2021. https://doi.org/10.1109/JPROC.2020.3004555

    work page doi:10.1109/jproc.2020.3004555 2021

  42. [42]

    A convolutional neural network based feature learning and fault diagnosis method for the condition monitoring of gearbox,

    L. Jing, M. Zhao, P. Li, and X. Xu, “A convolutional neural network based feature learning and fault diagnosis method for the condition monitoring of gearbox,”Mea- surement, vol. 111, pp. 1–10, 2017. https://doi.org/10. 1016/j.measurement.2017.07.017

    work page 2017

  43. [43]

    Industrial big data in an industry 4.0 environment: Challenges, schemes and applications for predictive maintenance,

    J. Yan, Y . Meng, L. Lu, and L. Li, “Industrial big data in an industry 4.0 environment: Challenges, schemes and applications for predictive maintenance,”IEEE Access, vol. 5, pp. 23484–23491, 2017. https://doi.org/10.1109/ ACCESS.2017.2765544

    work page arXiv 2017

  44. [44]

    Artificial in- telligence for fault diagnosis of rotating machinery: A review,

    R. Liu, B. Yang, E. Zio, and X. Chen, “Artificial in- telligence for fault diagnosis of rotating machinery: A review,”Mech. Syst. Signal Process., vol. 108, pp. 33– 47, 2018. https://doi.org/10.1016/j.ymssp.2018.02.016

    work page doi:10.1016/j.ymssp.2018.02.016 2018

  45. [45]

    Gradient boosting machines, a tutorial,

    A. Natekin and A. Knoll, “Gradient boosting machines, a tutorial,”Front. Neurorobot., vol. 7, p. 21, 2013. https: //doi.org/10.3389/fnbot.2013.00021

    work page doi:10.3389/fnbot.2013.00021 2013

  46. [46]

    Support vector machine in machine condition monitoring and fault diagnosis,

    A. Widodo and B.-S. Yang, “Support vector machine in machine condition monitoring and fault diagnosis,” Mech. Syst. Signal Process., vol. 21, no. 6, pp. 2560– 2574, 2007. https://doi.org/10.1016/j.ymssp.2006.12.007

    work page doi:10.1016/j.ymssp.2006.12.007 2007

  47. [47]

    Automotive OBD-II Dataset,

    M. Weber, “Automotive OBD-II Dataset,” Karlsruhe Institute of Technology, RADAR4KIT Repository,

    work page

  48. [48]

    https://radar.kit.edu/radar/en/ dataset/bCtGxdTklQlfQcAq

    DOI: 10.35097/1130. https://radar.kit.edu/radar/en/ dataset/bCtGxdTklQlfQcAq

    work page doi:10.35097/1130

  49. [49]

    carOBD: An OBD-II database for Toyota Etios 2014 vehicle,

    E. C. Ferreira, “carOBD: An OBD-II database for Toyota Etios 2014 vehicle,” GitHub Repository, 2019. https:// github.com/eron93br/carOBD

    work page 2014