Scholarlink

Purpose <p>The objective of this research is to integrate wavelet analysis and machine learning (ML) algorithms to improve the accuracy and dependability of roller element bearing defect diagnostics. Roller element bearings are vital parts of many industrial machinery, and identifying problems early on is crucial to avoiding expensive maintenance and downtime.</p> Method <p>The process entails taking vibration signals from rolling element bearings and using wavelet analysis to extract pertinent features. Wavelet transformations are a good choice for defect diagnostic jobs because of their advantages in capturing both the time and frequency domain features of signals. Also, this project incorporates a variety of machine learning methods to classify faults according to the features that are extracted.</p> Result <p>It is investigated how wavelets can be used to denoise vibrational data. The vibrational data is subjected to a variety of machine learning algorithms, and their performance is compared in terms of accuracy and training time. The efficacy of machine learning (ML) methods including Support Vector Machines (SVM), Decision Tree, KNN, and Neural Networks in precisely diagnosing various kinds of bearing defects is investigated. Although the neural network’s training period varies according to the system parameters, it provides us with a clear understanding of its intended use. The decision tree model demonstrated a balance between speed and accuracy, with a training time of 8.583&#xa0;s and an accuracy of 93.3%. On the other hand, the ANN model obtained a better accuracy of 93.8% and needed 10&#xa0;s to train. With a training time of 72&#xa0;s, the DNN model showed the best accuracy of all the models, at 98.5%, highlighting its potential for accurate fault classification at the expense of more computing effort. The KNN model demonstrated quick training in 6.59&#xa0;s with a 93.5% accuracy rate, which makes it a good option in situations where training speed is prioritised without requiring a large trade-off in accuracy. The trade-offs between model complexity, training time, and accuracy were highlighted by the SVM model, which demonstrated a high accuracy of 94.1% but needed a longer training period of 44.8&#xa0;s.</p> Conclusion <p>We used the ANN and DNN neural network models, together with three machine learning methods. It is discovered that the DNN classifies the data the most precisely, while the ANN strikes a balance between accuracy and training time.</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

Rolling Element Bearing Fault Diagnosis Using Discrete Mayer Wavelet and Fault Classification Using Machine Learning Algorithms

  • 期刊论文

  • 发表日期：

    2025年1月13日

查看全文
  • T. Narendiranath Babu,
  • Jalaj Saraya,
  • Kaushlesh Singh,
  • D. Rama Prabha

摘要

Purpose

The objective of this research is to integrate wavelet analysis and machine learning (ML) algorithms to improve the accuracy and dependability of roller element bearing defect diagnostics. Roller element bearings are vital parts of many industrial machinery, and identifying problems early on is crucial to avoiding expensive maintenance and downtime.

Method

The process entails taking vibration signals from rolling element bearings and using wavelet analysis to extract pertinent features. Wavelet transformations are a good choice for defect diagnostic jobs because of their advantages in capturing both the time and frequency domain features of signals. Also, this project incorporates a variety of machine learning methods to classify faults according to the features that are extracted.

Result

It is investigated how wavelets can be used to denoise vibrational data. The vibrational data is subjected to a variety of machine learning algorithms, and their performance is compared in terms of accuracy and training time. The efficacy of machine learning (ML) methods including Support Vector Machines (SVM), Decision Tree, KNN, and Neural Networks in precisely diagnosing various kinds of bearing defects is investigated. Although the neural network’s training period varies according to the system parameters, it provides us with a clear understanding of its intended use. The decision tree model demonstrated a balance between speed and accuracy, with a training time of 8.583 s and an accuracy of 93.3%. On the other hand, the ANN model obtained a better accuracy of 93.8% and needed 10 s to train. With a training time of 72 s, the DNN model showed the best accuracy of all the models, at 98.5%, highlighting its potential for accurate fault classification at the expense of more computing effort. The KNN model demonstrated quick training in 6.59 s with a 93.5% accuracy rate, which makes it a good option in situations where training speed is prioritised without requiring a large trade-off in accuracy. The trade-offs between model complexity, training time, and accuracy were highlighted by the SVM model, which demonstrated a high accuracy of 94.1% but needed a longer training period of 44.8 s.

Conclusion

We used the ANN and DNN neural network models, together with three machine learning methods. It is discovered that the DNN classifies the data the most precisely, while the ANN strikes a balance between accuracy and training time.