Scholarlink

<p>Optimal sensor placement (OSP) is vital for effective condition monitoring concerning cost and data quality. However, previous research has not thoroughly examined the impact of missing (occurs due to failure of sensor to capture the signals) or reducing the number of sensors within an OSP to identify configurations that maintain diagnostic accuracy while reducing redundancy. This study investigates the effect of sensor number on OSP for fault diagnostics of a wind turbine gearbox using vibration and acoustic sensors. Discrete wavelet transform was employed to process raw signals from the OSP sensors, and features are extracted from wavelet coefficients for fault classification. The accuracy of five classifiers (bagging, C-support vector regression, random forest, J48, and regression tree) with varying sensor numbers is assessed. Results indicate that a two-sensor system (L<sub>4</sub>, L<sub>7</sub>) using the random forest algorithm achieves 99.797% accuracy, making it suitable for fault classification. This reduction results in approximately 60% lower data volume, leading to faster data processing, reduced computational complexity, and significant cost savings in sensor deployment and maintenance. This strategic approach highlights the practical impact of optimizing sensor placement in condition monitoring systems, ensuring efficiency in cost, time, and operational reliability while achieving robust fault classification for wind turbine gearboxes. The study’s findings underline its potential for real-world applications, particularly in scenarios with limited resources.</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

Using machine learning approach to investigate the impact of missing sensors on fault classification of multistage gear box

  • 期刊论文

  • 发表日期：

    2025年2月22日

查看全文
  • S. V. V. S. Narayana Pichika,
  • Brahmini Priya Venkata Pragada,
  • Sabareesh Geetha Rajasekharan,
  • Inturi Vamsi

摘要

Optimal sensor placement (OSP) is vital for effective condition monitoring concerning cost and data quality. However, previous research has not thoroughly examined the impact of missing (occurs due to failure of sensor to capture the signals) or reducing the number of sensors within an OSP to identify configurations that maintain diagnostic accuracy while reducing redundancy. This study investigates the effect of sensor number on OSP for fault diagnostics of a wind turbine gearbox using vibration and acoustic sensors. Discrete wavelet transform was employed to process raw signals from the OSP sensors, and features are extracted from wavelet coefficients for fault classification. The accuracy of five classifiers (bagging, C-support vector regression, random forest, J48, and regression tree) with varying sensor numbers is assessed. Results indicate that a two-sensor system (L4, L7) using the random forest algorithm achieves 99.797% accuracy, making it suitable for fault classification. This reduction results in approximately 60% lower data volume, leading to faster data processing, reduced computational complexity, and significant cost savings in sensor deployment and maintenance. This strategic approach highlights the practical impact of optimizing sensor placement in condition monitoring systems, ensuring efficiency in cost, time, and operational reliability while achieving robust fault classification for wind turbine gearboxes. The study’s findings underline its potential for real-world applications, particularly in scenarios with limited resources.