Intelligent vibration signal denoising method based on non-local fully convolutional neural network for rolling...

Area of Science:

• Engineering
• Computer Science
• Artificial Intelligence

Background:

• Machinery health management increasingly relies on data-driven approaches.
• Limited research exists on data-driven denoising methods for complex machinery signals.
• Convolutional neural networks (CNNs) show promise but require advanced architectures for effective denoising.

Purpose of the Study:

• To propose a robust denoising method for machinery health management using a novel neural network architecture.
• To enhance the signal-to-noise ratio in machinery data for improved fault diagnosis.
• To evaluate the effectiveness of the proposed method against conventional techniques.

Main Methods:

• Development of a non-local fully convolutional neural network (NL-FCNN).
• Integration of Leaky-ReLU activation for signal information preservation.
• Utilization of wide kernels to expand the receptive field.
• Incorporation of non-local means (NLM) to create non-local blocks (NLBs) for enhanced long-range dependency learning.

Main Results:

• The NL-FCNN demonstrated superior denoising performance compared to three conventional methods.
• The method was validated on the Case Western Reserve University (CWRU) motor bearing dataset under various noise levels.
• The proposed denoising technique showed significant improvements in learning long-range dependencies within the signals.

Conclusions:

• The proposed NL-FCNN offers a robust and effective solution for data-driven signal denoising in machinery health management.
• The method shows strong potential for improving the accuracy of rolling bearing fault diagnosis.
• This work contributes to advancing deep learning applications in industrial condition monitoring.