Fault Diagnosis Method for Rotating Machinery Based on Threshold-Free Recurrence Distance Visualization Convolutional...

Area of Science:

• Mechanical Engineering
• Signal Processing
• Machine Learning

Background:

• Recursive Plots (RPs) offer valuable signal analysis but are limited by manual threshold selection, impacting fault diagnosis accuracy.
• Existing no-threshold methods for RPs have limitations and are not fully comprehensive.

Purpose of the Study:

• To propose a novel Threshold-Free Recurrence Distance (TFRD) method to overcome the limitations of manual threshold selection in RPs.
• To develop and validate a TFRD-CNN model for accurate rotating machinery fault diagnosis.

Main Methods:

• Development of the Threshold-Free Recurrence Distance (TFRD) metric based on Recursive Plots.
• Integration of TFRD with a Convolutional Neural Network (CNN) to create the TFRD-CNN fault diagnosis model.
• Validation using bearing vibration data and a planetary gearbox gear fault dataset.

Main Results:

• The TFRD-CNN model achieved high accuracy in diagnosing rotating machinery faults.
• Comparative analysis showed TFRD-CNN outperforms traditional RPs, Markov Transition Fields (MTF), Gramian Angular Fields (GAF), and RP/URP combined with CNN methods.
• The proposed method effectively addresses the threshold selection issue inherent in RPs.

Conclusions:

• The TFRD-CNN model presents a significant advancement in rotating machinery fault diagnosis by eliminating manual threshold dependency.
• This threshold-free approach enhances the reliability and accuracy of fault diagnosis systems.
• The TFRD-CNN method offers a robust and effective solution for analyzing complex machinery vibration data.