A Dual-Stage Attention Model for Tool Wear Prediction in Dry Milling Operation

Area of Science:

• Machinery and Manufacturing
• Artificial Intelligence
• Signal Processing

Background:

• Intelligent monitoring and prediction of tool wear are crucial for the modern machinery industry.
• Existing deep learning methods for tool wear prediction face challenges like signal data instability, gradient decay, and insufficient feature importance analysis.

Purpose of the Study:

• To propose a novel dual-stage attention model for enhanced tool wear prediction.
• To address gradient decay and improve the accuracy of wear prediction by focusing on feature importance.

Main Methods:

• A CNN-BiGRU-attention network incorporating self-attention for deep feature extraction.
• IndyLSTM network to ensure layer stability and mitigate gradient decay.
• An attention mechanism applied to the output sequence to identify critical information.

Main Results:

• The proposed dual-stage attention model effectively characterizes tool wear degrees.
• Demonstrated reduction in prediction errors compared to existing methods.
• Achieved good prediction results in experimental dry milling operations.

Conclusions:

• The dual-stage attention model offers a viable solution for accurate tool wear prediction.
• The method enhances feature importance and network stability, leading to improved prediction accuracy.
• This approach contributes to the intelligent development of the machinery industry through reliable tool wear monitoring.