A virtual sensor for online fault detection of multitooth-tools.

Area of Science:

• Manufacturing Engineering
• Machine Tool Monitoring
• Artificial Intelligence in Manufacturing

Background:

• Direct sensor installation near tool tips in industrial milling centers is often infeasible.
• Online fault detection is crucial for many industrial tasks, necessitating alternative sensing solutions.
• Multitooth tools present unique challenges for signal reliability due to intermittent cutting insert engagement.

Purpose of the Study:

• To develop a robust virtual sensor for online fault detection of multitooth tools in milling operations.
• To create a system that minimizes the need for frequent recalibration, even after maintenance.
• To validate the virtual sensor's performance under real-world industrial conditions.

Main Methods:

• Implementation of a virtual sensor utilizing a Bayesian classifier.
• Integration of mathematical models with physical sensor data (electrical power consumption and machining time).
• Validation using k-fold cross-validation for performance assessment.

Main Results:

• The virtual sensor demonstrated high recognition accuracy, with an average of 0.957 true positives and 0.986 true negatives.
• Measured accuracy reached 98%, indicating a strong capability to correctly identify new fault cases.
• Successful validation on a milling center used for mass production of automobile engine crankshafts.

Conclusions:

• The developed virtual sensor is a viable and effective solution for online fault detection in industrial milling processes.
• The system's robustness and high accuracy make it suitable for demanding production environments.
• This approach offers a practical alternative to physical sensors where direct installation is not possible.