Loading...

Fault Diagnosis in Rolling Element Bearings Using Machine Learning Based on Vibration Analysis Results

Ershadi, Mohammad | 2025

107 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 57978 (08)
  4. University: Sharif University of Technology
  5. Department: Mechanical Engineering
  6. Advisor(s): Pasharavesh, Abdolreza; Ahmadian, Mohammad Taghi
  7. Abstract:
  8. Rotating machinery plays a significant role in various industries. One of the most important and influential parts in rotating machinery is rolling bearings, which play a vital role in the proper functioning of this equipment. Therefore, bearing failures can lead to unintended equipment downtime, financial losses, and reduced productivity. As a result, accurate diagnosis of bearing faults is crucial for improving the reliability and performance of industrial machinery. Despite significant advances in the field of condition monitoring and intelligent fault diagnosis, challenges such as the lack of labeled data and the impact of noise on the accuracy of fault diagnosis models remain major obstacles. The present research consists of two main parts. In the first part, a hybrid method based on the finite element method (FEM), machine learning (ML), and deep learning (DL) models is presented for diagnosing rolling bearing faults. In fact, the advantage of the FEM in this research is that various faults, which cannot be collected from rolling bearings operating in the real world, can be simulated using FEM models. For this purpose, in the first step, the initial finite element model of the healthy bearing is created. Then, the data obtained from the simulation is compared with the experimental data (CWRU dataset), and the model parameters are updated within the specified limits to achieve the highest cosine similarity. In the second step, various faults (inner ring fault, outer ring fault, and rolling element fault) are introduced into the model, and faulty models are created. In the third step, the simulated and experimental data in both healthy and faulty states are used as training and test data for artificial intelligence (AI) models, respectively. The AI models used include the support vector machine (SVM) algorithm, stochastic gradient descent (SGD) classifier, and transformer network. The results show that the classification accuracy of these models on the experimental data is 95.00%, 96.25%, and 83.20%, respectively. Additionally, a transformer network was used to diagnose faults based on experimental data, where some samples were used for training and others for testing. This model achieved an accuracy of 99.60%. In the second part, to enhance the model's robustness against noise, a hybrid method based on the wavelet packet transform (WPT) and the SVM algorithm is presented. In this part, both training and test data are obtained from experimental data (half for training and half for testing) and include white, pink, and purple noise. The results show that the proposed method performs better for noisy data compared to models that rely solely on time signals and AI models. Specifically, as the noise power in the data increases, the accuracy of the proposed model remains higher compared to those models. The results of this research show that combining FEM simulation with ML and DL models is an effective approach to addressing the issue of insufficient and incomplete labeled training samples. Additionally, integrating WPT with the SVM algorithm is an efficient method for noisy data
  9. Keywords:
  10. Fault Diagnosis ; Bearing ; Transformer Network ; Transformer Network Optimization ; Support Vector Machine (SVM) ; Finite Element Method ; Noise ; Stochastic Gradient Descent Classifier (SGD-Classifier) ; Rolling Bearing

 Digital Object List

 Bookmark

No TOC