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

TL;DR

This paper introduces a new method for identifying faults in micro-motors using optimized CNN techniques and signal decomposition, achieving high diagnostic accuracy.

Contribution

A novel fault diagnosis method combining JMD decomposition, GRM transformation, and an optimized CNN for micro-motors.

Findings

The proposed method achieves 99.0476% average diagnostic accuracy for multiple fault types.

The method outperforms four existing comparative fault diagnosis approaches.

The GRM transformation enhances fault feature representation for better CNN performance.

Abstract

Micro-motors are widely used in industrial applications, which require effective fault diagnosis to maintain safe equipment operation. However, fault signals from micro-motors often exhibit weak signal strength and ambiguous features. To address these challenges, this study proposes a novel fault diagnosis method. Initially, the Jump plus AM-FM Mode Decomposition (JMD) technique was utilized to decompose the measured signals into amplitude-modulated–frequency-modulated (AM-FM) oscillation components and discontinuous (jump) components. The proposed process extracts valuable fault features and integrates them into a new time-domain signal, while also suppressing modal aliasing. Subsequently, a novel Global Relationship Matrix (GRM) is employed to transform one-dimensional signals into two-dimensional images, thereby enhancing the representation of fault features. These images are then input into an Optimized Convolutional Neural Network (OCNN) with an AdamW optimizer, which effectively reduces overfitting during training. Experimental results demonstrate that the proposed method achieves an average diagnostic accuracy rate of 99.0476% for multiple fault types, outperforming four comparative methods. This approach offers a reliable solution for quality inspection of micro-motors in a manufacturing environment.

Figures