Deep Learning Tool Wear State Identification Method Based on Cutting Force Signal.

Area of Science:

• Manufacturing Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Accurate identification of milling cutter wear is crucial for optimizing machining processes and preventing tool failure.
• Existing methods may lack the efficiency and accuracy required for real-time industrial monitoring.

Purpose of the Study:

• To develop an accurate, expeditious, and efficient method for identifying the wear state of milling cutters.
• To improve the performance of lightweight neural networks for tool wear monitoring.

Main Methods:

• A novel method combining Continuous Wavelet Transform (CWT) and an improved MobileViT lightweight network was proposed.
• CWT transformed cutting force signals into time-frequency images.
• The MobileViT network was enhanced with a Contextual Transformer module and a Global Attention Mechanism module.

Main Results:

• The enhanced MobileViT model demonstrated significant improvements in accuracy (1.58%), precision (1.23%), recall (1.92%), and F1 score (1.57%) compared to the original MobileViT.
• The proposed model showed advantages in memory occupation and prediction accuracy over VGG16, ResNet18, and Pool Former.
• The method enables near real-time identification of tool wear states.

Conclusions:

• The proposed method offers an efficient and accurate approach for milling cutter wear state identification.
• The enhanced MobileViT network provides superior performance for tool wear monitoring.
• This technique has strong potential for application in industrial production environments.