A novel feature extraction method for bearing fault classification with one dimensional ternary patterns.

Area of Science:

• Mechanical Engineering
• Signal Processing
• Machine Condition Monitoring

Background:

• Bearing failures in rotary machines can lead to catastrophic system malfunctions.
• Effective fault diagnosis relies on precise feature extraction from vibration signals.
• Traditional methods for bearing fault diagnosis can be complex and time-consuming.

Purpose of the Study:

• To introduce a novel feature extraction method, one-dimensional ternary pattern (1D-TP), for bearing fault diagnosis.
• To identify the fault size (mm) and bearing part (inner ring, outer ring, ball) using vibration signals.
• To evaluate the effectiveness of 1D-TP combined with various classification models.

Main Methods:

• Acquired vibration signals from a bearing test setup with artificial faults.
• Applied the proposed one-dimensional ternary pattern (1D-TP) statistical method for feature extraction.
• Utilized Random Forest (RF), k-nearest neighbor (k-NN), Support Vector Machine (SVM), BayesNet, and Artificial Neural Networks (ANN) for classification.

Main Results:

• The 1D-TP method achieved high success rates across different datasets.
• Dataset_1 (varying speed) achieved 91.25% success rate.
• Dataset_2 (fault type) and Dataset_3 (fault size) achieved 100% success rates.

Conclusions:

• The one-dimensional ternary pattern (1D-TP) is a highly effective method for bearing fault diagnosis.
• 1D-TP enables accurate identification of fault type and size in bearings.
• The proposed method demonstrates significant potential for improving machine condition monitoring.