Loading...

Intelligent Fault Diagnosis of Rotating Machines by Convolutional Neural Network

Farahani, Ali | 2024

0 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 57139 (08)
  4. University: Sharif University of Technology
  5. Department: Mechanical Engineering
  6. Advisor(s): Behzad, Mehdi
  7. Abstract:
  8. Early and confident detection of faults in rolling element bearings plays a significant role in enhancing the reliability of equipment and optimizing future decisions within a set. The purpose of this research is to investigate and compare raw signals and various signal processing transformations, including Fast Fourier Transform, Frequency Domain Envelope Analysis, Time Domain Envelope Analysis, Wavelet Transform, Short-Time Fourier Transform, and Cepstrum Transform, in order to find the best representation of vibration signals for more accurate equipment fault diagnosis. Additionally, examining and comparing simple convolutional neural network, wide-kernel convolutional neural network, and transfer learning technique based on convolutional neural network for finding the best model for fast and reliable fault diagnosis of these equipment. In this regard, four experiments of rolling element bearing faults were conducted, including three types of artificially induced faults and one healthy condition under variable operating conditions, on an experimental platform. Vibration values of bearings were measured and recorded in the horizontal direction (direction of applied load) using an accelerometer sensor at 5 different loadings and with 9 different speeds for each loading. In the first chapter, the introduction of the conducted research and its importance has been discussed. In the second chapter, the vibrational characteristics of various faults in rolling element bearings and how to identify these faults are compared and discussed. Following that, signal processing techniques and various transformations used in the detection and diagnosis of rolling element bearings are discussed and examined. Then, the advantages, limitations, and characteristics of convolutional neural networks and transfer learning techniques are detailed, followed by a review of the work done. In the third chapter, the experimental platform and conducted experiments are discussed, and all aspects related to the practical part of this research are examined and discussed. In the fourth chapter, the results obtained from different models are presented, and in the fifth chapter, these results are analyzed, showing that the convolutional neural network model with transfer learning technique and input data with Short-Time Fourier Transform is the best model for diagnosing rolling element bearing faults, achieving a 99% accuracy in industrial test data. Finally, a general conclusion is drawn
  9. Keywords:
  10. Rolling Bearing ; Convolutional Neural Network ; Transfer Learning ; Condition Monitoring ; Vibrations Signal Processing ; Variable Working Conditions ; Artificial Faults

 Digital Object List

 Bookmark

...see more