Scholarlink

<p>Gear fault identification is critical for maintaining operational efficiency and reliability in industrial systems. This study presents an industry-ready model that integrates multi-sensor data fusion with ensemble machine learning techniques to enhance fault detection accuracy under diverse operating conditions. Leveraging the MCC5-THU gearbox fault diagnosis dataset, which encompasses 12 working conditions, faults were introduced at speeds of 1000, 2000, and 3000 RPM with loads of 10 Nm and 20 Nm. Key fault types such as pitting, wear, and cracks were classified across three severity levels with a model accuracy of 95.98%. Signal preprocessing included noise removal and feature extraction from time and frequency domains. Dimensionality reduction through PCA reduced computational load by 30%, ensuring real-time implementation feasibility. The ensemble model outperformed traditional methods, achieving a sensitivity increase of 15% and maintaining robust accuracy even in noisy environments. Synthetic data testing with 1500 combinations demonstrated adaptability to unseen scenarios, with accuracy ranging from 91%. This research addresses limitations of traditional diagnostic systems by enhancing fault sensitivity, minimizing computational overhead, and ensuring real-time application readiness. It provides a scalable solution for industries like manufacturing and aerospace, improving maintenance strategies, reducing downtime, and optimizing operational costs effectively.</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

Gear Fault Identification Using Sensor Fusion and Ensemble Machine Learning: Development and Validation of an Industry Ready Pre-Trained Model

  • 期刊论文

  • 发表日期：

    2025年12月3日

查看全文
  • B. Geeta Chandra Sekhar,
  • T. V. Hanumantha Rao,
  • Tadela Naveen Kumar,
  • Sireesha Veeramachaneni

摘要

Gear fault identification is critical for maintaining operational efficiency and reliability in industrial systems. This study presents an industry-ready model that integrates multi-sensor data fusion with ensemble machine learning techniques to enhance fault detection accuracy under diverse operating conditions. Leveraging the MCC5-THU gearbox fault diagnosis dataset, which encompasses 12 working conditions, faults were introduced at speeds of 1000, 2000, and 3000 RPM with loads of 10 Nm and 20 Nm. Key fault types such as pitting, wear, and cracks were classified across three severity levels with a model accuracy of 95.98%. Signal preprocessing included noise removal and feature extraction from time and frequency domains. Dimensionality reduction through PCA reduced computational load by 30%, ensuring real-time implementation feasibility. The ensemble model outperformed traditional methods, achieving a sensitivity increase of 15% and maintaining robust accuracy even in noisy environments. Synthetic data testing with 1500 combinations demonstrated adaptability to unseen scenarios, with accuracy ranging from 91%. This research addresses limitations of traditional diagnostic systems by enhancing fault sensitivity, minimizing computational overhead, and ensuring real-time application readiness. It provides a scalable solution for industries like manufacturing and aerospace, improving maintenance strategies, reducing downtime, and optimizing operational costs effectively.