Helical Gearbox Defect Detection with Machine Learning Using Regular Mesh Components and Sidebands

Area of Science:

• Mechanical Engineering
• Condition Monitoring
• Vibration Analysis

Background:

• Helical gearboxes are critical components in many industrial applications.
• Gearbox failures can lead to significant downtime and economic losses.
• Effective condition monitoring is essential for predictive maintenance.

Purpose of the Study:

• To develop and validate models for detecting helical gearbox defects using raw vibration signals.
• To investigate the relationship between gear faults and spectral characteristics.
• To identify optimal features for accurate defect classification.

Main Methods:

• Vibration signals were acquired using a triaxial accelerometer under various operating conditions (rotational velocities and load levels).
• Gear faults including pitting and wear were simulated and analyzed.
• Feature extraction involved analyzing power in specific frequency bands, peak amplitudes, and statistical measures in time and frequency domains.
• Support Vector Machine (SVM) with a cubic kernel was employed for defect detection.

Main Results:

• A strong correlation was observed between gear faults and the fundamental meshing frequency (GMF), its harmonics, and sidebands in the vibration spectrum.
• The best performing model utilized band powers of six GMF harmonics and two sideband pairs across all accelerometer axes.
• This SVM-based model achieved a testing accuracy of 99.73% for defect detection.

Conclusions:

• Vibration analysis, particularly focusing on spectral features around the GMF, is a highly effective method for helical gearbox defect detection.
• The proposed feature set and SVM model demonstrate excellent performance, offering a robust solution for condition monitoring.
• The developed models are capable of detecting faults irrespective of operational speeds and load conditions.