Local-feature and global-dependency based tool wear prediction using deep learning.

Area of Science:

• Manufacturing Engineering
• Artificial Intelligence
• Machine Learning

Background:

• Tool wear evaluation is critical for ensuring workpiece quality and machining efficiency in manufacturing systems.
• Traditional tool wear prediction methods are labor-intensive and rely on expert experience.
• Deep learning offers automated feature extraction and predictive modeling capabilities.

Purpose of the Study:

• To propose a novel hybrid deep learning method for tool wear prediction.
• To combine manual and automatic features for improved prediction accuracy.
• To enhance the reliability and efficiency of tool wear monitoring in intelligent manufacturing.

Main Methods:

• A hybrid approach integrating manual and automatic features for tool wear prediction.
• Utilizing an enhanced Convolutional Neural Network (CNN) on wavelet scalograms to learn local and multi-scale features.
• Employing a multi-layer Long Short-Term Memory (LSTM) network to capture global dependencies from sequential feature vectors.
• Training a fully connected layer for the final tool wear prediction.

Main Results:

• The proposed method effectively learns local single-scale and multi-scale correlation features.
• Global dependencies are successfully captured by the sequential processing of feature vectors in LSTM.
• The hybrid model demonstrates strong prediction performance and generalization ability across multiple working conditions.
• Experimental validation confirms the effectiveness of the proposed tool wear prediction technique.

Conclusions:

• The developed hybrid deep learning model offers an effective solution for tool wear prediction.
• This approach enhances intelligent manufacturing by automating feature extraction and improving prediction accuracy.
• The method provides a robust framework for ensuring workpiece quality and optimizing machining efficiency through early tool wear detection.