A novel intelligent fault diagnosis method for gearbox based on multi-dimensional attention denoising convolution

Area of Science:

• Artificial Intelligence
• Mechanical Engineering
• Signal Processing

Background:

• Convolutional Neural Networks (CNNs) struggle with noisy, multi-scale vibration data in industrial rotating machinery fault diagnosis.
• Effective noise suppression and multi-scale feature extraction are critical for accurate intelligent fault diagnosis.
• Existing methods face challenges in handling the complexities of real-world industrial vibration signals.

Purpose of the Study:

• To propose a novel deep neural network framework, the Multidimensional Fusion Residual Attention Network (MFRANet), for gearbox fault diagnosis.
• To address the limitations of CNNs in processing noisy and multi-scale industrial vibration data.
• To enhance the accuracy and robustness of fault diagnosis in rotating machinery.

Main Methods:

• Developed the Multidimensional Fusion Residual Attention Network (MFRANet) incorporating a multi-scale deep separable convolution module.
• Integrated a residual channel attention module for feature map denoising and weighting.
• Employed an external attention module to capture implicit correlations within denoised multi-scale features.

Main Results:

• The MFRANet demonstrated superior diagnostic performance compared to benchmark and state-of-the-art methods on a gearbox fault dataset.
• The proposed method exhibited robust noise resilience across various noise levels, indicating enhanced reliability.
• Experimental evaluations confirmed the effectiveness of MFRANet in accurately extracting multi-scale fault features.

Conclusions:

• MFRANet offers an innovative and efficient solution for gearbox fault diagnosis in rotating machinery.
• The study highlights the potential of advanced deep learning architectures in addressing practical industrial challenges.
• The proposed framework provides a reliable method for intelligent fault diagnosis in demanding industrial environments.