pith. sign in

arxiv: 2507.01048 · v1 · submitted 2025-06-25 · 💻 cs.LG

3W Dataset 2.0.0: a realistic and public dataset with rare undesirable real events in oil wells

Ricardo Emanuel Vaz Vargas , Afr\^anio Jos\'e de Melo Junior , Celso Jos\'e Munaro , Cl\'audio Benevenuto de Campos Lima , Eduardo Toledo de Lima Junior , Felipe Muntzberg Barrocas , Fl\'avio Miguel Varej\~ao , Guilherme Fidelis Peixer
show 12 more authors
Igor de Melo Nery Oliveira Jader Riso Barbosa Jr. Jaime Andr\'es Lozano Cadena Jean Carlos Dias de Ara\'ujo Jo\~ao Neuenschwander Escosteguy Carneiro Lucas Gouveia Omena Lopes Lucas Pereira de Gouveia Mateus de Araujo Fernandes Matheus Lima Scramignon Patrick Marques Ciarelli Rodrigo Castello Branco Rog\'erio Leite Alves Pinto
This is my paper

Pith reviewed 2026-05-19 07:07 UTC · model grok-4.3

classification 💻 cs.LG
keywords 3W Datasetoil wellsundesirable eventsmultivariate time seriespublic datasetmachine learningearly detectionanomaly detection
0
0 comments X

The pith

The 3W Dataset has been updated to version 2.0.0 with structural modifications and additional labeled data for detecting rare oil well events.

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

This paper presents the current public version of the 3W Dataset as a collection of multivariate time series from real oil wells, now expanded with more expert-labeled examples of undesirable events and some changes to its structure. A sympathetic reader would care because these events can produce economic losses, environmental harm, and safety risks, and public labeled data lets machine learning models learn to spot them early enough for corrective action. The dataset began as a 2019 release and has grown through ongoing collaboration to serve as a reference for detection research.

Core claim

The 3W Dataset 2.0.0 is a publicly available set of expert-labeled multivariate time series that records both normal oil well operations and rare undesirable events, with structural updates and extra labeled instances added to support more effective machine learning for early detection.

What carries the argument

The 3W Dataset, a collection of multivariate time series labeled by experts to mark undesirable events in oil wells.

If this is right

  • Machine learning models can now be trained on a larger set of real rare-event examples to improve detection accuracy.
  • New detection methodologies can be tested and compared against prior results using the updated data.
  • Digital monitoring products can be built that give operators more advance warning before events cause damage.
  • Corrective or mitigating actions become feasible earlier in the event sequence.

Where Pith is reading between the lines

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

  • The expanded set of rare events could serve as a benchmark for testing anomaly detection methods across other industrial sensor streams.
  • Structural changes may simplify the process of feeding the data into online learning or streaming analytics systems.
  • Users could combine this dataset with synthetic generators to study how class imbalance affects model performance.
  • Cross-domain transfer from this oil-well data to similar time-series problems in other sectors becomes more practical.

Load-bearing premise

Expert labels accurately and consistently identify the rare undesirable events across the added time series data.

What would settle it

A side-by-side re-labeling of the newly added time series by independent experts that finds frequent disagreements on event timing or type would show the labels cannot be trusted for model training.

Figures

Figures reproduced from arXiv: 2507.01048 by Afr\^anio Jos\'e de Melo Junior, Celso Jos\'e Munaro, Cl\'audio Benevenuto de Campos Lima, Eduardo Toledo de Lima Junior, Felipe Muntzberg Barrocas, Fl\'avio Miguel Varej\~ao, Guilherme Fidelis Peixer, Igor de Melo Nery Oliveira, Jader Riso Barbosa Jr., Jaime Andr\'es Lozano Cadena, Jean Carlos Dias de Ara\'ujo, Jo\~ao Neuenschwander Escosteguy Carneiro, Lucas Gouveia Omena Lopes, Lucas Pereira de Gouveia, Mateus de Araujo Fernandes, Matheus Lima Scramignon, Patrick Marques Ciarelli, Ricardo Emanuel Vaz Vargas, Rodrigo Castello Branco, Rog\'erio Leite Alves Pinto.

Figure 2
Figure 2. Figure 2: Illustration of the labeling process of real instances. [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
read the original abstract

In the oil industry, undesirable events in oil wells can cause economic losses, environmental accidents, and human casualties. Solutions based on Artificial Intelligence and Machine Learning for Early Detection of such events have proven valuable for diverse applications across industries. In 2019, recognizing the importance and the lack of public datasets related to undesirable events in oil wells, Petrobras developed and publicly released the first version of the 3W Dataset, which is essentially a set of Multivariate Time Series labeled by experts. Since then, the 3W Dataset has been developed collaboratively and has become a foundational reference for numerous works in the field. This data article describes the current publicly available version of the 3W Dataset, which contains structural modifications and additional labeled data. The detailed description provided encourages and supports the 3W community and new 3W users to improve previous published results and to develop new robust methodologies, digital products and services capable of detecting undesirable events in oil wells with enough anticipation to enable corrective or mitigating actions.

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

1 major / 2 minor

Summary. The paper describes the 3W Dataset 2.0.0, an updated public collection of multivariate time series from oil wells that have been labeled by experts for rare undesirable events. It documents structural modifications relative to the 2019 release and the addition of new labeled instances, with the goal of supporting AI/ML research on early detection to prevent economic losses and safety incidents.

Significance. If the modifications and labels are faithfully documented, the release is significant because it supplies a realistic, publicly available benchmark containing rare but high-impact events. Such data are scarce in the industrial ML literature and directly enable reproducible work on anomaly detection methods that could be deployed to reduce environmental and operational risks in oil production.

major comments (1)
  1. [Dataset construction and labeling section] The description of the expert labeling process for the newly added time series is insufficiently detailed. Because the central claim of the paper is that the dataset now contains additional reliable labels for rare real events, the manuscript must specify the labeling protocol, inter-rater consistency measures, and any validation steps used for the new instances. This information is load-bearing for users who will treat the labels as ground truth.
minor comments (2)
  1. [Abstract] The abstract states that structural modifications have been made but does not enumerate them; adding a single sentence listing the main changes (e.g., new sensor channels, revised sampling rates, or file-format updates) would improve immediate clarity.
  2. [Results or Dataset Statistics] A summary table showing the number of new time series, their total duration, and the distribution of event classes in version 2.0.0 versus 1.0 would help readers quickly gauge the scale of the update.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive feedback and the recommendation for minor revision. We agree that greater detail on the labeling process will improve the manuscript and address the comment below by expanding the relevant section in the revised version.

read point-by-point responses

  1. Referee: [Dataset construction and labeling section] The description of the expert labeling process for the newly added time series is insufficiently detailed. Because the central claim of the paper is that the dataset now contains additional reliable labels for rare real events, the manuscript must specify the labeling protocol, inter-rater consistency measures, and any validation steps used for the new instances. This information is load-bearing for users who will treat the labels as ground truth.

    Authors: We agree that the current description of the expert labeling process for the newly added instances is insufficiently detailed. In the revised manuscript we will expand the Dataset construction and labeling section to include a full account of the labeling protocol (including the sequence of steps followed by domain experts at Petrobras), any inter-rater consistency measures that were applied, and the validation procedures used to confirm the new labels. These additions will allow readers to evaluate the reliability of the labels as ground truth. revision: yes

Circularity Check

0 steps flagged

No significant circularity; purely descriptive data release

full rationale

The manuscript is a data article that documents structural changes and newly added expert-labeled time series in the public 3W Dataset 2.0.0. It contains no derivations, equations, predictions, parameter fittings, or model-based claims that could reduce to self-definition or fitted inputs. The central claim is enumerative and descriptive, supported by direct description of the dataset contents rather than any internal mathematical construction or self-citation chain. Expert labeling is acknowledged as a domain limitation but does not create a circular dependency within the paper's own logic.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

This is a dataset description paper. No free parameters, mathematical axioms, or invented entities are introduced or required for the central contribution.

pith-pipeline@v0.9.0 · 5855 in / 891 out tokens · 36900 ms · 2026-05-19T07:07:41.055743+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

34 extracted references · 34 canonical work pages

  1. [1]

    An engineering look at the cause of the 2010 Macondo blowout

    Turley, J. A. (2014). “An engineering look at the cause of the 2010 Macondo blowout” inIADC/SPE Drilling Conference and Exhibition, Fort Worth, TX, United States. https://doi.org/10.2118/167970-MS

    work page doi:10.2118/167970-ms 2014

  2. [2]

    Integrated framework for abnormal event management and process hazards analysis

    Dash, S., & Venkatasubramanian, V . (2003). “Integrated framework for abnormal event management and process hazards analysis” inAIChE journal, 49(1), 124-139. https://doi.org/10.1002/aic.690490112

    work page doi:10.1002/aic.690490112 2003

  3. [3]

    Anomaly Detection Methods for Industrial Applications: A Comparative Study

    Panza, M.A., Pota, M., & Esposito, M. (2023). “Anomaly Detection Methods for Industrial Applications: A Comparative Study” inElectronics, 12, 3971. https://doi.org/10.3390/electronics12183971

    work page doi:10.3390/electronics12183971 2023

  4. [4]

    Data-driven anomaly detection and event log profiling of SCADA alarms

    Andrade, J., Rocha, C., Silva, R., Viana, J., Bessa, R., Gouveia, C., Almeida, B., Santos, R., Louro, M., Santos, P., & Ribeiro, A. (2022). “Data-driven anomaly detection and event log profiling of SCADA alarms” inIEEE Access, 10, 1–1. https://doi.org/10.1109/ACCESS.2022.3190398

    work page doi:10.1109/access.2022.3190398 2022

  5. [5]

    Two-Phase Defect Detection Using Clustering and Classification Methods

    Tran, H. M., Nguyen, T. A., Le, S. T., Huynh, G. V . T., & Lam, T. B. (2022). “Two-Phase Defect Detection Using Clustering and Classification Methods” inREV Journal on Electronics and Communications, 12(1-2). http://dx.doi.org/10.21553/rev-jec.296

    work page doi:10.21553/rev-jec.296 2022

  6. [6]

    A survey on dataset quality in machine learning , journal =

    Gong, Y ., Liu, G., Xue, Y ., Li, R., Meng, L. (2023). “A survey on dataset quality in machine learning” in Information and Software Technology, 162. https://doi.org/10.1016/j.infsof.2023.107268

    work page doi:10.1016/j.infsof.2023.107268 2023

  7. [7]

    Importance of Datasets for ML and DM

    Shark, W. (2020). “Importance of Datasets for ML and DM” inInternet of Things and Big Data Applications: Recent Advances and Challenges, 180, 122

    work page 2020

  8. [8]

    Petrobras

    Petr ´oleo Brasileiro S.A. (Petrobras) (2025). “Petrobras”. https://petrobras.com.br/en/. 20 PREPRINT

    work page 2025

  9. [9]

    A realistic and public dataset with rare undesirable real events in oil wells

    Vaz Vargas, R. E., Munaro, C. J., Marques Ciarelli, P., Gonc ¸alves Medeiros, A., Guberfain do Amaral, B., Centurion Barrionuevo, D., Dias de Ara´ujo, J. C., Lins Ribeiro, J., & Pierezan Magalh˜aes, L. (2019). “A realistic and public dataset with rare undesirable real events in oil wells” inJournal of Petroleum Science and Engineering,

    work page 2019

  10. [10]

    https://doi.org/10.1016/j.petrol.2019.106223

    work page doi:10.1016/j.petrol.2019.106223 2019

  11. [11]

    Multivariate time series analysis and its applications

    Tsay, R. S. (2010). “Multivariate time series analysis and its applications” inIn Analysis of financial time series (pp. 389–465). John Wiley & Sons, Ltd. https://doi.org/10.1002/9780470644560.ch8

    work page doi:10.1002/9780470644560.ch8 2010

  12. [12]

    Practical Time Series Analysis: Prediction with Statistics and Machine Learning

    Nielsen, A. (2019). “Practical Time Series Analysis: Prediction with Statistics and Machine Learning” inFirst Edition. O’Reilly Media, CA

    work page 2019

  13. [13]

    Maximizing information from chemical engineering data sets: Applications to machine learning

    Thebelt, A., Wiebe, J., Kronqvist, J., Tsay, C., Misener, R. (2022). “Maximizing information from chemical engineering data sets: Applications to machine learning” inChemical Engineering Science, 252, pp. 117469. https://doi.org/10.1016/j.ces.2022.117469

    work page doi:10.1016/j.ces.2022.117469 2022

  14. [14]

    Papadakis, C., Filandrianos, G., Dimitriou, A., Lymperaiou, M., Thomas, K., Stamou, G., 2025

    Pan, S.J., Yang, Q. (2010). “A Survey on Transfer Learning” inIEEE Transactions on Knowledge and Data Engineering, 22:10, pp. 1345-1359. https://doi.org/10.1109/TKDE.2009.191

    work page doi:10.1109/tkde.2009.191 2010

  15. [15]

    A Review of Deep Transfer Learning and Recent Advancements

    Iman, M., Arabnia, H. R., Rasheed, K. (2023). “A Review of Deep Transfer Learning and Recent Advancements” inTechnologies, 11(2): 40. https://doi.org/10.3390/technologies11020040

    work page doi:10.3390/technologies11020040 2023

  16. [16]

    Semantic Versioning 2.0.0

    Preston-Werner, T. (2013). “Semantic Versioning 2.0.0”. http://semver.org

    work page 2013

  17. [17]

    The 3W Community

    Petr ´oleo Brasileiro S.A. (Petrobras) (2025). “The 3W Community”. https://github.com/petrobras/3W/tree/main/ community

    work page 2025

  18. [18]

    The 3W Project

    Petr ´oleo Brasileiro S.A. (Petrobras) (2025). “The 3W Project”. https://github.com/petrobras/3W

    work page 2025

  19. [19]

    Open Lab Module of the Connections Program for Innovation

    Petr ´oleo Brasileiro S.A. (Petrobras) (2025). “Open Lab Module of the Connections Program for Innovation”. https://conexoes-inovacao.petrobras.com.br/s/openlab?language=en US

    work page 2025

  20. [20]

    “GitHub”

    Microsoft Corporation (2025). “GitHub”. https://github.com

    work page 2025

  21. [21]

    Python Language Reference

    Python Software Foundation (2025). “Python Language Reference”. http://www.python.org

    work page 2025

  22. [22]

    Plant Information Management System

    Yokogawa (2025). “Plant Information Management System”. https://www.yokogawa.com/solutions/solutions/ connected-intelligence/plant-information-management-system

    work page 2025

  23. [23]

    A VEV A PI System

    A VEV A (2025). “A VEV A PI System”. https://www.aveva.com/en/products/aveva-pi-system

    work page 2025

  24. [24]

    SLB (2025). “OLGA”. https://www.slb.com/products-and-services/delivering-digital-at-scale/software/olga

    work page 2025

  25. [25]

    Marine Petroleum (Gas) Engineering and Equipment

    Fang, H., Duan, M. (2014). “Marine Petroleum (Gas) Engineering and Equipment” inIn Offshore Operation Fa- cilities. Gulf Professional Publishing, Boston, pp. 341–536. https://doi.org/10.1016/b978-0-12-396977-4.00003- 2

    work page doi:10.1016/b978-0-12-396977-4.00003- 2014

  26. [26]

    Part IV: Artificial Lift Methods

    Guo, B., Liu, X., Tan, X. (2017). “Part IV: Artificial Lift Methods” inPetroleum Production Engineering (Sec- ond Edition). Gulf Professional Publishing, Boston, pp. 513–635. https://doi.org/10.1016/B978-0-12-809374- 0.00041-6

    work page doi:10.1016/b978-0-12-809374- 2017

  27. [27]

    On characterizations of input- to-state stability with respect to compact sets

    Sotoodeh, K. (2021). “Introduction to the Subsea Sector of the Oil and Gas Industry” inSubsea Valves and Actu- ators for the Oil and Gas Industry. Gulf Professional Publishing, Boston, pp. 1–36. https://doi.org/10.1016/b978- 0-323-90605-0.00006-2

    work page doi:10.1016/b978- 2021

  28. [28]

    CC BY 4.0

    Creative Commons (2025). “CC BY 4.0”. https://creativecommons.org/licenses/by/4.0

    work page 2025

  29. [29]

    Figshare

    Figshare LLP (2025). “Figshare”. https://info.figshare.com

    work page 2025

  30. [30]

    Apache Parquet

    Apache Software Foundation (2025) “Apache Parquet”. https://parquet.apache.org

    work page 2025

  31. [31]

    “PyArrow”

    Apache Software Foundation (2025). “PyArrow”. https://arrow.apache.org/docs/index.html

    work page 2025

  32. [32]

    Brotli Compressed Data Format

    J. Alakuijala & Z. Szabadka (2016). “Brotli Compressed Data Format” inInternet Engineering Task Force (IETF). Request for Comments: 7932. https://www.ietf.org/rfc/rfc7932.txt

    work page 2016

  33. [33]

    2020.pandas-dev/pandas: Pandas

    The pandas development team (2020). “pandas-dev/pandas: Pandas”. https://doi.org/10.5281/zenodo.3509134

    work page doi:10.5281/zenodo.3509134 2020

  34. [34]

    Apache License Version 2.0

    Apache Software Foundation (2004). “Apache License Version 2.0”. https://www.apache.org/licenses/LICENSE-2.0.txt. 21

    work page 2004