Simultaneous Burr and Cut Interruption Detection during Laser Cutting with Neural Networks
Benedikt Adelmann1, Ralf Hellmann1
1Applied Laser and Photonics Group, Faculty of Engineering, University of Applied Sciences Aschaffenburg, Wuerzburger Straße 45, 63739 Aschaffenburg, Germany.
Sensors (Basel, Switzerland)
|September 10, 2021
Summary
This study compares basic and convolutional neural networks for classifying fiber laser cutting defects. Both methods achieve high accuracy (over 92%), with CNNs showing a slight edge for real-time industrial applications.
Area of Science:
- Materials Science
- Manufacturing Engineering
- Artificial Intelligence
Background:
- Fiber laser cutting is a key industrial process.
- Monitoring cut quality in real-time is crucial for efficiency and product integrity.
- Automated defect detection systems are needed to improve quality control.
Purpose of the Study:
- To compare the effectiveness of basic neural networks (BNNs) and convolutional neural networks (CNNs) for classifying defects in fiber laser cutting.
- To evaluate the performance of these networks in detecting specific cut failures like burr formation and cut interruptions.
- To assess the feasibility of a unified system for simultaneous detection of multiple defect types.
Main Methods:
- Experiments involved cutting thin electrical sheets using a 500 W single-mode fiber laser.
- Coaxial camera images were captured during the cutting process for analysis.
- BNNs and CNNs were trained and tested for classifying cut quality into 'good cut', 'burr formation', and 'cut interruptions'.
Main Results:
- Both BNNs and CNNs achieved high classification accuracies, with a minimum of 92.8% and up to 95.8% for more complex CNNs.
- CNNs demonstrated a slight performance advantage over BNNs, despite a marginally higher computation time (under 2 ms).
- Cut interruptions were detected with significantly higher accuracy than burr formation in separate analyses.
Conclusions:
- It is possible to detect both burr formations and cut interruptions simultaneously during fiber laser cutting with high accuracy.
- The developed neural network systems are suitable for real-time monitoring in industrial applications.
- CNNs offer a promising approach for automated quality control in laser cutting processes.
