A Fault Identification Method for Micro-Motors Using an Optimized CNN-Based JMD-GRM Approach

Area of Science:

• Engineering
• Signal Processing
• Machine Learning

Background:

• Micro-motors are crucial in industrial applications, necessitating robust fault diagnosis for operational safety.
• Weak signal strength and ambiguous features in micro-motor fault signals pose significant diagnostic challenges.

Purpose of the Study:

• To develop a novel and effective fault diagnosis method for micro-motors.
• To address the limitations of existing methods in handling weak and ambiguous fault signals.

Main Methods:

• Utilized Jump plus AM-FM Mode Decomposition (JMD) to decompose signals into AM-FM and jump components, extracting and integrating fault features.
• Employed a Global Relationship Matrix (GRM) to convert 1D signals into 2D images for enhanced feature representation.
• Applied an Optimized Convolutional Neural Network (OCNN) with an AdamW optimizer for accurate fault classification and reduced overfitting.

Main Results:

• The proposed method achieved an average diagnostic accuracy of 99.0476% across multiple fault types.
• Demonstrated superior performance compared to four other fault diagnosis methods.
• Successfully suppressed modal aliasing and enhanced fault feature representation.

Conclusions:

• The developed fault diagnosis method provides a reliable solution for micro-motor quality inspection in manufacturing.
• The combination of JMD, GRM, and OCNN offers a powerful approach for analyzing complex fault signals.
• This technique significantly improves the accuracy and efficiency of micro-motor fault diagnosis.