Scholarlink

Predictive maintenance leverages critical machinery's data generated from sensor in real-time and past data to identify potential failures and mitigate their impact. While machine and deep learning techniques are increasingly used for accurate fault detection, the manual selection of models and hyperparameter tuning remains time-consuming and requires significant domain expertise, affecting model generalization. To overcome these challenges, in the paper an automated machine learning (AutoML) framework is being introduced that autonomously designs an effective pipeline, minimizing human intervention. In this paper, time-series data generated from sensors integrated on the Metro Do Porto train network's Air Production Unit (APU) is utilized to create a versatile pipeline for the rail transportation sector. Through the work, we are achieving an accuracy of 99.98% and an F1-score of 0.999, the H2O AutoML technique outperforms other models.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Automatic Machine Learning-Driven Anomaly Detection for Predictive Maintenance and Fault Diagnosis in Urban Metro Systems

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Ankit Kumar,
  • Anurag Choubey,
  • Shivendu Mishra,
  • Maheshkumar H. Kolekar,
  • Rajiv Misra,
  • Gaddi Suresh Kumar

摘要

Predictive maintenance leverages critical machinery's data generated from sensor in real-time and past data to identify potential failures and mitigate their impact. While machine and deep learning techniques are increasingly used for accurate fault detection, the manual selection of models and hyperparameter tuning remains time-consuming and requires significant domain expertise, affecting model generalization. To overcome these challenges, in the paper an automated machine learning (AutoML) framework is being introduced that autonomously designs an effective pipeline, minimizing human intervention. In this paper, time-series data generated from sensors integrated on the Metro Do Porto train network's Air Production Unit (APU) is utilized to create a versatile pipeline for the rail transportation sector. Through the work, we are achieving an accuracy of 99.98% and an F1-score of 0.999, the H2O AutoML technique outperforms other models.