Ball bearing fault detection using an acoustic based machine learning approach

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Ball bearing failures cause significant industrial downtime and costs.
• Predictive maintenance using machine learning (ML) and deep learning (DL) offers early fault detection.
• Acoustic vibration analysis is a key indicator of bearing health.

Purpose of the Study:

• To develop a deep neural network model for predicting ball bearing faults.
• To utilize continuous wavelet transform (CWT) for analyzing vibration signals.
• To enhance the accuracy and reliability of bearing fault diagnosis.

Main Methods:

• Acoustic vibration signals from faulty bearings were collected.
• Continuous Wavelet Transform (CWT) was used to convert signals into time-frequency scalograms.
• A Convolutional Neural Network (CNN) based on the LeNet-5 architecture was trained on these scalograms.
• The model classified bearing conditions into distinct defective states.

Main Results:

• The proposed CWT-based deep neural network achieved a 99.23% prediction accuracy for ball bearing faults.
• This accuracy surpasses that of traditional numerical signal-based classification models.
• The method effectively captures the time-frequency characteristics of vibration signals for fault diagnosis.

Conclusions:

• The combination of CWT and CNN provides a highly accurate and effective method for real-time ball bearing fault detection.
• This approach represents a significant advancement in predictive maintenance for industrial machinery.
• The model offers improved performance over existing machine learning techniques for bearing fault diagnosis.