Scholarlink

<p>This study explores the use of online machine learning (ML) techniques to improve predictive maintenance (PdM) in variable and energy-intensive industrial settings. We focus on a real-world case involving an EVS fan system at a cement manufacturing plant. Our research compares the performance of online ML models with traditional offline models across four key areas: predictive accuracy, adaptability to concept drift, reduction in power consumption, and training energy efficiency. The results demonstrate that the online model outperforms the offline counterpart in detecting early anomalies and adapting to evolving process conditions with an accuracy score of 90%. It enabled proactive interventions that led to a quantifiable reduction of 9221 kWh of energy over a short period. Furthermore, the online model achieved this with 260 × less training time compared to the offline model, validating its suitability for edge deployment in Industry 5.0 contexts.</p>

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

Online learning for sustainable predictive maintenance in dynamic industrial environments: A case study in cement manufacturing

  • 期刊论文

  • 发表日期：

    2025年7月12日

查看全文
  • Hassana Mahfoud,
  • Mohammed Toum Benchekroun,
  • Zineb Elaboudi,
  • Fatima Ezzahra Bkirich

摘要

This study explores the use of online machine learning (ML) techniques to improve predictive maintenance (PdM) in variable and energy-intensive industrial settings. We focus on a real-world case involving an EVS fan system at a cement manufacturing plant. Our research compares the performance of online ML models with traditional offline models across four key areas: predictive accuracy, adaptability to concept drift, reduction in power consumption, and training energy efficiency. The results demonstrate that the online model outperforms the offline counterpart in detecting early anomalies and adapting to evolving process conditions with an accuracy score of 90%. It enabled proactive interventions that led to a quantifiable reduction of 9221 kWh of energy over a short period. Furthermore, the online model achieved this with 260 × less training time compared to the offline model, validating its suitability for edge deployment in Industry 5.0 contexts.