Anomaly Detection Techniques in the Service of Data Labeling for Fault Diagnosis in Manufaturing

Aldonso Martins de O. Junior, Aldonso Martins de O. Junior, Emmanuel A. de B. Santos, Denis Leite, Denis Leite, Alexandre M. A. Maciel

2025

Abstract

The lack of labeled fault data in industrial environments presents a major challenge for developing effective fault detection and diagnosis models. This study investigates the application of unsupervised anomaly detection techniques to identify abnormal machine behavior without relying on labeled data. By enabling the early detection of anomalous conditions, these techniques assist in distinguishing normal from faulty instances, supporting the labeling process for improved fault diagnosis. Ten different techniques are evaluated across multiple performance metrics to determine their effectiveness in industrial fault detection. Experimental results demonstrate that Angle-Based Outlier Detection (ABOD) outperformed other methods, achieving a higher F1-score and improved accuracy in recognizing unseen normal data. These findings highlight the potential of unsupervised learning for enhancing industrial fault detection, facilitating the transition to data-driven maintenance strategies, and optimizing data collection processes. The study provides valuable insights into model selection, dataset structuring, and cost-efficient implementation strategies for industrial applications, contributing to the broader adoption of anomaly detection in manufacturing environments.

Paper Citation

in Harvard Style

O. Junior A., Santos E., Leite D. and Maciel A. (2025). Anomaly Detection Techniques in the Service of Data Labeling for Fault Diagnosis in Manufaturing. In Proceedings of the 27th International Conference on Enterprise Information Systems - Volume 1: ICEIS; ISBN 978-989-758-749-8, SciTePress, pages 921-928. DOI: 10.5220/0013437100003929

in Bibtex Style

@conference{iceis25,
author={Aldonso O. Junior and Emmanuel Santos and Denis Leite and Alexandre Maciel},
title={Anomaly Detection Techniques in the Service of Data Labeling for Fault Diagnosis in Manufaturing},
booktitle={Proceedings of the 27th International Conference on Enterprise Information Systems - Volume 1: ICEIS},
year={2025},
pages={921-928},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0013437100003929},
isbn={978-989-758-749-8},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 27th International Conference on Enterprise Information Systems - Volume 1: ICEIS
TI - Anomaly Detection Techniques in the Service of Data Labeling for Fault Diagnosis in Manufaturing
SN - 978-989-758-749-8
AU - O. Junior A.
AU - Santos E.
AU - Leite D.
AU - Maciel A.
PY - 2025
SP - 921
EP - 928
DO - 10.5220/0013437100003929
PB - SciTePress