An Acquisition Parameter Study for Machine-Learning-Enabled Electron Backscatter Diffraction.

Area of Science:

• Materials Science
• Artificial Intelligence
• Computer Vision

Background:

• Artificial intelligence (AI) methods, particularly deep neural networks (DNNs), are increasingly used for materials science challenges.
• Analyzing electron diffraction patterns with DNNs is a key application, but performance under varied experimental conditions is crucial for deployment.

Purpose of the Study:

• To investigate the impact of varying key experimental parameters on the performance of DNNs used for electron diffraction pattern analysis.
• To identify which parameters most significantly affect classification accuracy and guide future data collection strategies.

Main Methods:

• Electron backscatter diffraction (EBSD) patterns were collected while individually varying five parameters: frame averaging, detector tilt, sample-to-detector distance, accelerating voltage, and pattern resolution.
• A DNN, pre-trained on EBSD patterns from a fixed condition set, was used to classify patterns generated under these varied conditions.

Main Results:

• The DNN model demonstrated significant resilience to individual variations in most tested experimental parameters.
• Performance degradation was minimal across a range of altered diffraction conditions, indicating robustness.

Conclusions:

• The study suggests that DNNs for electron diffraction analysis are robust to common experimental variations.
• This resilience builds confidence in deploying AI models across diverse operating conditions in materials science research.