Three-dimensional laser damage positioning by a deep-learning method.

Area of Science:

• Optics and Photonics
• Artificial Intelligence
• Materials Science

Background:

• Accurate localization of laser-induced damage is critical for material analysis and quality control.
• Existing methods for 3D laser damage location face challenges in precision and distinguishing damage from various surfaces.

Purpose of the Study:

• To develop and validate a holographic and deep learning-based method for precise three-dimensional laser damage location.
• To improve the accuracy and efficiency of identifying laser damage sites on material surfaces.

Main Methods:

• A holographic approach combined with a deep learning neural network (Diffraction-Net) was employed for damage localization.
• Numerical focusing of diffraction rings determined axial position; Diffraction-Net identified lateral position and distinguished rings from different surfaces/positions.
• The neural network was trained exclusively on simulative data.

Main Results:

• The proposed method successfully achieved damage pointing on cascade slabs, identifying the smallest inspectable damage size at 8µm.
• Diffraction-Net demonstrated the ability to distinguish diffraction rings with an overlap rate greater than 61%, surpassing previous reports.
• Experimental validation showed high precision with lateral positioning error < 38.5µm and axial positioning error < 2.85mm.

Conclusions:

• The developed holographic and deep learning method offers a practical and highly precise solution for 3D laser damage site localization.
• The findings highlight the potential for online damage inspection with significantly improved accuracy and efficiency.
• Diffraction-Net's performance, trained on simulative data, demonstrates its robustness in complex scenarios.