Scholarlink

<p>Condition monitoring of gears has become popular in recent years, as the gear fault may cause machine downtime and additional costs. This paper presents vibration-based gear fault detection using random forest classifier based on mean frequency and phase modulation (PM) features of vibration signals. Gear failures are artificially seeded on a pinion gear with different fault intensities. The mean frequency of the resulting vibration signal is obtained using the short-time Fourier transform (STFT). In addition to mean frequency, the phase deviation caused by the fault is also revealed. Moreover, feature vectors are constructed using mean frequency and phase modulation which are both used as input to a random forest (RF) classifier. Other classifiers such as support vector machines (SVM) and neural networks (NN) have also been employed to compare the classification performance of RF. The results have shown that the RF classification method outperforms other approaches and provides a classification accuracy of more than 98%, which is satisfactory for classification applications.</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

An Improved Fault Detection and Classification in Helical Gears Based on Vibration Signal Analysis

  • 期刊论文

  • 发表日期：

    2025年12月16日

查看全文
  • Gurkan Kucukyildiz

摘要

Condition monitoring of gears has become popular in recent years, as the gear fault may cause machine downtime and additional costs. This paper presents vibration-based gear fault detection using random forest classifier based on mean frequency and phase modulation (PM) features of vibration signals. Gear failures are artificially seeded on a pinion gear with different fault intensities. The mean frequency of the resulting vibration signal is obtained using the short-time Fourier transform (STFT). In addition to mean frequency, the phase deviation caused by the fault is also revealed. Moreover, feature vectors are constructed using mean frequency and phase modulation which are both used as input to a random forest (RF) classifier. Other classifiers such as support vector machines (SVM) and neural networks (NN) have also been employed to compare the classification performance of RF. The results have shown that the RF classification method outperforms other approaches and provides a classification accuracy of more than 98%, which is satisfactory for classification applications.