Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Volvo Discovery Challenge at ECML-PKDD 2024

Mahmoud Rahat, Peyman Sheikholharam Mashhadi, Sławomir Nowaczyk, Shamik Choudhury, Leo Petrin, Thorsteinn Rognvaldsson, Andreas Voskou, Carlo Metta, Claudio Savelli

TL;DR

The Volvo Discovery Challenge at ECML-PKDD 2024 tackled predictive maintenance by forecasting a three-class failure-risk label for a Volvo truck component using real-world fleet data, with training restricted to gen1 and testing across gen1/gen2. The paper presents dataset structure, evaluation protocol, and three winning methodologies—STab-based tabular modeling with ensemble and normalization, LSTM with pseudo-labeling and boosting, and Time-Transformer–driven two-stage risk classification with rich temporal feature extraction. Key contributions include a detailed account of the competition setup, performance patterns across generations, and actionable code resources from top teams for predictive maintenance research. The work demonstrates robust generalization to a new component generation and offers practical methodologies and benchmarks for applied maintenance analytics in industrial settings.

Abstract

This paper presents an overview of the Volvo Discovery Challenge, held during the ECML-PKDD 2024 conference. The challenge's goal was to predict the failure risk of an anonymized component in Volvo trucks using a newly published dataset. The test data included observations from two generations (gen1 and gen2) of the component, while the training data was provided only for gen1. The challenge attracted 52 data scientists from around the world who submitted a total of 791 entries. We provide a brief description of the problem definition, challenge setup, and statistics about the submissions. In the section on winning methodologies, the first, second, and third-place winners of the competition briefly describe their proposed methods and provide GitHub links to their implemented code. The shared code can be interesting as an advanced methodology for researchers in the predictive maintenance domain. The competition was hosted on the Codabench platform.

Volvo Discovery Challenge at ECML-PKDD 2024

TL;DR

The Volvo Discovery Challenge at ECML-PKDD 2024 tackled predictive maintenance by forecasting a three-class failure-risk label for a Volvo truck component using real-world fleet data, with training restricted to gen1 and testing across gen1/gen2. The paper presents dataset structure, evaluation protocol, and three winning methodologies—STab-based tabular modeling with ensemble and normalization, LSTM with pseudo-labeling and boosting, and Time-Transformer–driven two-stage risk classification with rich temporal feature extraction. Key contributions include a detailed account of the competition setup, performance patterns across generations, and actionable code resources from top teams for predictive maintenance research. The work demonstrates robust generalization to a new component generation and offers practical methodologies and benchmarks for applied maintenance analytics in industrial settings.

Abstract

This paper presents an overview of the Volvo Discovery Challenge, held during the ECML-PKDD 2024 conference. The challenge's goal was to predict the failure risk of an anonymized component in Volvo trucks using a newly published dataset. The test data included observations from two generations (gen1 and gen2) of the component, while the training data was provided only for gen1. The challenge attracted 52 data scientists from around the world who submitted a total of 791 entries. We provide a brief description of the problem definition, challenge setup, and statistics about the submissions. In the section on winning methodologies, the first, second, and third-place winners of the competition briefly describe their proposed methods and provide GitHub links to their implemented code. The shared code can be interesting as an advanced methodology for researchers in the predictive maintenance domain. The competition was hosted on the Codabench platform.
Paper Structure (12 sections, 6 figures)

This paper contains 12 sections, 6 figures.

Table of Contents

  1. Introduction
  2. Challenge Setup
  3. Prizes
  4. Submission Process
  5. Submissions Statistics
  6. Description of the Dataset
  7. Evaluation
  8. Winning Methodologies
  9. Team RandomGuy
  10. Team CarloMetta
  11. Team MALTO (MAchine Learning at PoliTO)
  12. Terms and Conditions

Figures (6)

  • Figure 1: Development phase leaderboard for the top ten places.
  • Figure 2: Final phase leaderboard for the top ten places.
  • Figure 3: A histogram showing number of submissions per week.
  • Figure 4: Score improvement over time for the top three participants.
  • Figure 5: Sequence Extraction from Healthy and Failed components. Note that the length of all sequences is 10 time steps.
  • ...and 1 more figures