In production system for tool wear prediction using multi-sensor time series and machine learning models.

Area of Science:

• Manufacturing Engineering
• Materials Science
• Machine Learning

Background:

• Optimizing micro-machining milling is crucial for reducing production costs.
• Monitoring tool wear and machining quality allows for process adjustments.
• Predictive maintenance through tool wear detection enhances efficiency.

Purpose of the Study:

• To explore machine learning models for predicting machining tool wear.
• To evaluate the performance of different classifiers in tool wear prediction.
• To assess the feasibility of an in-production system for real-time tool wear monitoring.

Main Methods:

• A dataset was created from a face milling operation on stainless steel (AISI 303) using a tungsten carbide tool.
• Acoustic emission, accelerometers, and axis currents sensors were used to measure tool wear.
• Four machine learning approaches were evaluated using F1-score and a weighted expected value.

Main Results:

• The extra-trees classifier achieved the optimal F1-score of 73% across five classes.
• This classifier demonstrated superior performance compared to other evaluated models.
• The study evaluated the implementation feasibility of the best model for production systems.

Conclusions:

• Machine learning, particularly the extra-trees classifier, shows significant potential for accurate tool wear prediction in micro-machining.
• Accurate tool wear prediction can lead to optimized milling processes and reduced production costs.
• The findings support the development of in-production systems for real-time machining monitoring and control.