Revealing geometrically necessary dislocation density from electron backscatter patterns via multi-modal deep...

Area of Science:

• Materials Science
• Crystallography
• Computational Materials Science

Background:

• Geometrically necessary dislocation (GND) characterization is crucial for understanding material plastic deformation.
• Current Electron Backscatter Diffraction (EBSD) methods face challenges in achieving fast and accurate GND density determination.

Purpose of the Study:

• To develop a novel multi-modal deep learning approach for predicting GND density.
• To enhance the speed and accuracy of GND density analysis using EBSD data.

Main Methods:

• A multi-modal deep learning architecture with two Convolutional Neural Network (CNN) streams was employed.
• One CNN stream processed electron backscatter patterns (EBSPs) for pattern shifts.
• The second CNN stream learned representations from dislocation configurations, augmented by a neighboring pairs strategy.

Main Results:

• The proposed method achieved high accuracy in predicting GND density on aluminum samples.
• The deep learning networks demonstrated robustness against various noise types.
• Prediction speed matched modern EBSD scanning rates, enabling real-time analysis.

Conclusions:

• The multi-modal deep learning approach offers a fast and accurate solution for GND density characterization.
• This method facilitates real-time analysis of plastic deformation, advancing materials science research.
• The developed technique overcomes limitations of existing EBSD-based GND analysis.