Performance of Bearing Ball Defect Classification Based on the Fusion of Selected Statistical Features

Area of Science:

• Mechanical Engineering
• Signal Processing
• Machine Learning

Background:

• Ball bearing faults are challenging to detect and classify.
• Early fault detection is crucial for preventing equipment failure.

Purpose of the Study:

• To develop a robust methodology for classifying incipient ball bearing faults.
• To enhance the accuracy and efficiency of bearing fault diagnosis.

Main Methods:

• Vibration signals decomposed using Empirical Mode Decomposition (EMD).
• Relevant Intrinsic Mode Functions (IMFs) selected based on energy and sensitivity.
• Feature extraction using statistical moments and Kullback-Leibler Divergence (KLD).
• Classification using various statistical clustering techniques (linear, kernel, tree-based, probabilistic).

Main Results:

• EMD decomposition reduced 18 IMFs to 4 relevant ones.
• Feature fusion using IMFs' variance and KLD achieved 100% classification rates with kernel/data-mining classifiers.
• Linear classifiers were found to be inefficient.
• The proposed method outperformed multiscale permutation entropy.
• Training and testing times were compatible with real-time applications (<2s and <0.2s respectively).

Conclusions:

• The proposed methodology effectively classifies incipient ball bearing faults.
• Feature fusion and advanced classifiers are key to achieving high diagnostic accuracy.
• The method's efficiency and speed make it suitable for real-time industrial monitoring.