Publication: Motor fault diagnosis based on vibration signals analysis
Loading...
Date
2024-08
Authors
Meow, Ying Yan
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Motor failure can be caused by a variety of reasons, and early motor failure detection systems can help avoid service interruptions and financial losses. However,
current methods of manual troubleshooting and preventive maintenance are time consuming and can be ineffective. Therefore, the use of vibration signals to diagnose motor faults has increased in recent years. Machine Learning (ML) can be used to automate the task of motor fault detection based on vibration signals, and thus improve the effectiveness of motor preventive maintenance. In this project, a motor fault diagnosis system based on vibration signals and machine learning (ML) technology was proposed. Feature extraction is one of the crucial components for an intelligent fault diagnosis system. Therefore, several data preprocessing methods were investigated to identify the most effective way to extract the features and enhance the accuracy of motor fault diagnosis. These methods encompass two categories: feature learning, involving direct input of raw time-domain and frequency data into an intelligent model, and manual feature extraction, necessitating the computation of statistical features before model input, including time domain and wavelet domain features. The extracted features were trained on the Convolutional Neural Network (CNN) and Artificial Neural Network (ANN) models to assess their performance. The comparison results show that wavelet domain features greatly improve ANN performance, achieving an overall accuracy of 99.78%. CNN, on the other hand, achieves remarkable accuracy of 100%, demonstrating that CNN can learn features adaptively from frequency data. This project advances the field of motor fault diagnosis, emphasizing the important role of preprocessing in optimizing ML model performance.