Fault Detection in Induction Machines Using Learning Models and Fourier Spectrum Image Analysis

Area of Science:

• Electrical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Induction motors are vital industrial components. Current fault detection methods can be limited.
• Early fault detection is crucial for preventing costly downtime and ensuring operational safety.

Purpose of the Study:

• To develop an automated fault detection system for induction motors using deep learning and Fourier spectra analysis.
• To enhance spectral feature learning with a novel preprocessing technique.
• To improve the accuracy and interpretability of fault diagnosis.

Main Methods:

• Generating images from the Fourier spectra of induction motor current signals.
• Employing a deep learning model based on the 19-layer Visual Geometry Group (VGG) architecture.
• Utilizing a new preprocessing technique with a distinctive background for enhanced feature learning.
• Applying explainability techniques to interpret model behavior and feature identification.

Main Results:

• The VGG-based model achieved an overall accuracy of 98% in detecting four fault types: healthy motor coupled to a generator with a broken bar (HGB), broken rotor bar (BRB), race bearing fault (RBF), and bearing ball fault (BBF).
• Specific fault detection accuracies were 99% for HGB, 100% for BRB, 100% for RBF, and 95% for BBF.
• Explainability analysis revealed that different convolutional blocks capture distinct features (signal shape, background) and specific layers correlate with individual fault types.

Conclusions:

• The proposed methodology offers a scalable and accurate solution for the predictive maintenance of induction motors.
• Combining signal processing, computer vision, and explainability techniques provides a robust framework for automated fault diagnosis.
• The interpretability of the model enhances trust and understanding in AI-driven industrial monitoring systems.