Vibration-based gearbox fault diagnosis using a multi-scale convolutional neural network with depth-wise feature...

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Gearbox failures are critical in industrial settings, causing downtime and safety risks.
• Accurate and timely fault diagnosis is essential for operational efficiency and cost reduction.
• Existing methods may lack robustness or require invasive monitoring.

Purpose of the Study:

• To propose a novel deep learning approach for vibration-based gearbox fault diagnosis.
• To enhance diagnostic accuracy and robustness in industrial environments.
• To enable non-invasive, automated, and early fault detection.

Main Methods:

• Utilized a multi-scale convolutional neural network with depth-wise feature concatenation (MixNet).
• Employed Short-time Fourier Transform (STFT) to generate spectrograms from vibration signals.
• Extracted discriminative features using multi-scale convolutional layers and feature concatenation.

Main Results:

• MixNet achieved a diagnostic accuracy of 99.32% on the Gearbox fault diagnosis dataset.
• The model demonstrated superior performance compared to existing deep learning models.
• Training time was efficient at 4 minutes and 29 seconds.

Conclusions:

• The proposed MixNet model offers a highly accurate and computationally efficient solution for gearbox fault diagnosis.
• This non-invasive, automated approach supports condition-based maintenance and reduces costs.
• MixNet is suitable for real-time monitoring in manufacturing, energy, and automotive applications.