On-machine surface defect detection using light scattering and deep learning.

Area of Science:

• Materials Science
• Optical Engineering
• Artificial Intelligence

Background:

• Surface defects significantly impact product quality and performance in precision manufacturing.
• Current defect detection methods can be time-consuming, labor-intensive, or lack robustness.
• Light scattering patterns contain rich information about surface topography and defects.

Purpose of the Study:

• To develop and validate an efficient on-machine system for surface defect detection.
• To leverage deep learning for analyzing light scattering patterns to identify defects.
• To create a robust and accurate defect detection method applicable to homogeneous materials.

Main Methods:

• Utilized a supervised deep learning approach, specifically a convolutional neural network (CNN).
• Trained the CNN on a comprehensive dataset of light scattering patterns generated by a rigorous forward scattering model.
• Validated the system's performance against experimental data and on microstructured surfaces.

Main Results:

• The deep learning model successfully mined defect-related information from light scattering patterns.
• The trained neural network demonstrated fast, accurate, and robust defect detection capabilities.
• The system's effectiveness was confirmed on surfaces created via ultraprecision diamond machining.

Conclusions:

• The proposed on-machine system offers a powerful solution for real-time surface defect analysis.
• Deep learning applied to light scattering provides a versatile and reliable method for defect detection.
• This approach is suitable for various surface topographies with homogeneous materials, enhancing manufacturing quality control.