Tool Health Monitoring of a Milling Process Using Acoustic Emissions and a ResNet Deep Learning Model.

Area of Science:

• Industrial Engineering
• Machine Learning
• Acoustics

Background:

• Tool health monitoring is crucial for reducing costs and waste in industrial settings.
• Traditional methods may not be efficient or accurate enough for real-time monitoring.
• Acoustic emission analysis offers a non-invasive approach to assess tool condition.

Purpose of the Study:

• To develop and evaluate a machine learning model for monitoring the health of end-milling tools.
• To utilize spectrograms of acoustic emission data for tool condition assessment.
• To differentiate between new, moderately used, and worn-out cutting tools.

Main Methods:

• Airborne acoustic emission signals were recorded from end-milling operations.
• Spectrograms of the acoustic data were generated.
• A Residual Network (a type of convolutional neural network) was trained on the spectrograms.
• Experiments were conducted with varying cut depths (1-3 mm) and wood types (hardwood and softwood).

Main Results:

• The trained model achieved an overall classification accuracy of 99.7% on 710 test samples.
• The model demonstrated high accuracy in classifying wood types: 100% for hardwood and 99.5% for softwood.
• The system effectively distinguished between new, moderately used, and worn-out tools.

Conclusions:

• The proposed method using acoustic emission spectrograms and a Residual Network is highly effective for industrial tool health monitoring.
• This approach offers a precise and efficient way to assess tool condition, leading to improved industrial operations.
• The model's high accuracy supports its potential for real-time implementation in manufacturing environments.