Research on Tool Wear Prediction Method Based on CNN-ResNet-CBAM-BiGRU

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Machine Learning

Background:

• Tool wear prediction is crucial for manufacturing efficiency and safety.
• Existing methods struggle with feature extraction and gradient issues.
• Accurate prediction requires robust modeling of complex sensor data.

Purpose of the Study:

• To develop an advanced deep learning model for enhanced tool wear prediction.
• To address limitations in feature extraction, vanishing gradients, and accuracy.
• To improve the stability and performance of tool wear prediction systems.

Main Methods:

• A hybrid deep neural network combining Convolutional Neural Network (CNN), Residual Network (ResNet), Convolutional Block Attention Module (CBAM), and Bidirectional Gated Recurrent Unit (BiGRU).
• Construction of a 34-dimensional multi-domain feature set (time, frequency, time-frequency) with z-score normalization.
• Integration of ResNet residual connections for deep network stability and CBAM for adaptive feature reweighting.
• Utilizing BiGRU for bidirectional temporal dependency modeling and a fully connected layer for regression.

Main Results:

• The proposed hybrid architecture demonstrated superior stability and predictive performance over baseline deep learning models on the PHM2010 dataset.
• Ablation studies showed significant improvements: Mean Absolute Error (MAE) reduced by 47.5%, Root Mean Square Error (RMSE) by 68.5%, and R-squared increased by 14.5% compared to a baseline CNN model.
• The model effectively captures temporal dependencies and extracts relevant features for accurate tool wear estimation.

Conclusions:

• The hybrid CNN-BiGRU model with ResNet and CBAM offers a robust solution for accurate tool wear prediction.
• This approach effectively mitigates vanishing gradients and enhances feature representation.
• The findings enable more reliable and precise monitoring of tool conditions in industrial applications.