Simultaneous-Fault Diagnosis of Gearboxes Using Probabilistic Committee Machine

Area of Science:

• Mechanical Engineering
• Signal Processing
• Machine Learning

Background:

• Gearbox fault detection is crucial for industrial machinery maintenance.
• Existing methods often struggle with complex fault signatures and noise.
• Accurate fault diagnosis enhances operational reliability and reduces downtime.

Purpose of the Study:

• To develop an intelligent diagnostic framework for enhanced gearbox fault detection.
• To improve diagnostic accuracy and the number of detectable faults.
• To create a robust system capable of handling single and simultaneous faults.

Main Methods:

• Signal de-noising using wavelet threshold method.
• Feature extraction via Hilbert-Huang Transform (HHT) and energy pattern calculation.
• Ensemble learning with two Pairwise-Coupled Relevance Vector Machines (PCRVMs) optimized by Particle Swarm Optimization (PSO).

Main Results:

• The proposed framework achieved superior diagnostic accuracies compared to individual classifiers.
• The ensemble method effectively combined vibration and sound signal analyses.
• The system demonstrated effectiveness in detecting both single and simultaneous gearbox faults.

Conclusions:

• The intelligent diagnostic framework offers a significant advancement in gearbox fault detection.
• The combination of signal processing and ensemble machine learning provides robust fault identification.
• This approach enhances the reliability and maintainability of gearbox systems.