Transfer Learning-Based Condition Monitoring of Single Point Cutting Tool.

Area of Science:

• Manufacturing Engineering
• Artificial Intelligence
• Machine Learning

Background:

• Cutting tool wear is critical in manufacturing, impacting dimensional accuracy and productivity.
• Conventional tool condition monitoring requires significant human expertise.
• Intelligent, automated diagnostic tools are needed to overcome these limitations.

Purpose of the Study:

• To propose and evaluate transfer learning technology for single point cutting tool condition monitoring.
• To leverage deep learning algorithms for automated classification of tool states.
• To identify optimal hyperparameters for pretrained deep learning networks.

Main Methods:

• Collected vibration signals from cutting tools and generated plots as input for deep learning.
• Employed pretrained deep learning networks (VGG-16, AlexNet, ResNet-50, GoogLeNet).
• Investigated the impact of hyperparameters like batch size, solver, learning rate, and train-test split ratio.

Main Results:

• Deep learning algorithms successfully learned from vibration signal plots to classify tool conditions.
• Pretrained networks demonstrated capability in identifying the state of single point cutting tools.
• The study identified the best-performing network and optimal hyperparameters for tool condition monitoring.

Conclusions:

• Transfer learning offers an effective approach for automated tool condition monitoring.
• Deep learning models can accurately assess cutting tool health using vibration signal data.
• This method enhances manufacturing productivity by enabling proactive tool maintenance.