Scholarlink

This study introduces a modern solution to predictive maintenance by integrating a Modified Overall Equipment Effectiveness (OEE) framework and comparing the performance of various Machine Learning (ML) and Deep Learning (DL) models. Traditional OEE, based only on Availability (A), Performance (P), and Quality (Q), often provides limited insights, especially in real-time industrial applications. This research incorporates additional parameters. Utilization (U) and Condition (C) to form a more comprehensive Modified OEE, which better reflects machine health and operational efficiency. To support predictive maintenance, models such as Random Forest (RF), XGBoost, Artificial Neural Networks (ANN), and Long Short-Term Memory (LSTM) were evaluated on both sample and real-time datasets. Results indicate that LSTM consistently outperforms other models with the lowest False Negative Rate (FNR = 0) and highest F1 Score (100%) on real-time data. Modified OEE values highlight a more accurate degradation trend compared to traditional OEE, especially during suboptimal machine performance days. This integrated approach not only improves maintenance scheduling but also enhances production reliability and decision-making in Industry 4.0 environments.

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

Data-Driven Predictive Maintenance: Evaluating ML and DL Models with Enhanced OEE Metrics

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Pranita Bhosale,
  • Sangeeta Jadhav

摘要

This study introduces a modern solution to predictive maintenance by integrating a Modified Overall Equipment Effectiveness (OEE) framework and comparing the performance of various Machine Learning (ML) and Deep Learning (DL) models. Traditional OEE, based only on Availability (A), Performance (P), and Quality (Q), often provides limited insights, especially in real-time industrial applications. This research incorporates additional parameters. Utilization (U) and Condition (C) to form a more comprehensive Modified OEE, which better reflects machine health and operational efficiency. To support predictive maintenance, models such as Random Forest (RF), XGBoost, Artificial Neural Networks (ANN), and Long Short-Term Memory (LSTM) were evaluated on both sample and real-time datasets. Results indicate that LSTM consistently outperforms other models with the lowest False Negative Rate (FNR = 0) and highest F1 Score (100%) on real-time data. Modified OEE values highlight a more accurate degradation trend compared to traditional OEE, especially during suboptimal machine performance days. This integrated approach not only improves maintenance scheduling but also enhances production reliability and decision-making in Industry 4.0 environments.