Explainable Remaining Tool Life Prediction for Individualized Production Using Automated Machine Learning

Area of Science:

• Manufacturing Automation
• Industry 4.0
• Machine Learning

Background:

• Individualized manufacturing of complex parts faces challenges due to varying process conditions and limited automation.
• Online monitoring and tool life prediction are crucial for automation but complicated by unclear sensor-to-tool condition correlations.
• Limited machine learning (ML) expertise on the shop floor hinders model adaptation to changing conditions.

Purpose of the Study:

• To develop a novel method for remaining cutting tool life prediction in individualized production using automated machine learning (AutoML).
• To enable the integration of machining expert knowledge into ML models for better adaptability.
• To enhance the degree of automation and optimize resource utilization in complex manufacturing environments.

Main Methods:

• Development of an AutoML method for end-to-end ML pipeline creation, utilizing optimized ensembles of regression and forecasting models.
• Incorporation of machining expert knowledge through model inputs and outputs.
• Implementation of an explainability algorithm to visualize input relevance for decision-making.

Main Results:

• The proposed AutoML method significantly outperforms the state-of-the-art approach for series production when applied to variable process conditions.
• A new milling dataset capturing gradual tool wear under changing parameters was utilized for evaluation.
• The study highlights the difficulty of directly transferring series production methods to variable manufacturing scenarios.

Conclusions:

• The developed AutoML method effectively optimizes individualized production economically and resource-wise.
• Machining experts with limited ML knowledge can utilize their domain expertise to develop, validate, and adapt tool life prediction models.
• The approach facilitates increased automation and improved decision-making in complex, customized manufacturing settings.