Comparative Evaluation of Machine Learning Algorithms for Fault Diagnosis in Automotive Press Lines


Creative Commons License

Öner A. E., Bayraktar M.

Sensors, vol.26, no.16, pp.1-31, 2026 (Scopus)

  • Publication Type: Article / Article
  • Volume: 26 Issue: 16
  • Publication Date: 2026
  • Doi Number: 10.3390/s26165058
  • Journal Name: Sensors
  • Journal Indexes: Scopus
  • Page Numbers: pp.1-31
  • Open Archive Collection: AVESIS Open Access Collection
  • Yıldız Technical University Affiliated: Yes

Abstract

Modern industrial operations heavily rely on minimizing unplanned downtime to safeguard productivity boundaries. Within this context, Predictive Maintenance (PdM) offers a viable track to intercept mechanical failures early by marrying sensor networks with machine learning tools. In this study, we introduce a field-tested PdM framework deployed in an automotive components plant. Over a twelve-month observation window, we captured raw vibration and process parameters from two operational transfer presses to build a comparative dataset containing both localized gear damage and intact baseline dynamics. Following a data-purging phase to isolate signal anomalies, we systematically tested the diagnostic limits of six algorithms, including SVM, Random Forest, Naive Bayes, k-NN, Decision Trees, and Logistic Regression. Ultimately, Random Forest and k-Nearest Neighbor frameworks stood out, yielding the most robust detection accuracy under real-world factory constraints. These findings demonstrate that the proposed PdM approach significantly reduces resource waste and prevents unnecessary downtime, offering a scalable model for industrial asset management. Furthermore, angle-based position information was acquired via an integrated high-resolution encoder, enabling the developed software to detect specific teeth exhibiting defects.