A Denoising Autoencoder for Improved Kikuchi Pattern Quality and Indexing in Electron Backscatter Diffraction

Area of Science:

• Materials Science
• Crystallography
• Data Science

Background:

• Electron backscatter diffraction (EBSD) is crucial for determining crystallographic orientation and microstructure.
• EBSD data quality is often compromised by noise, affecting indexing accuracy and analysis.
• Factors like sample preparation and data collection parameters influence pattern quality and confidence index (CI).

Purpose of the Study:

• To enhance EBSD data quality and improve orientation fit accuracy, especially with noisy datasets.
• To enable faster EBSD data collection without sacrificing analytical precision.
• To reduce errors in microstructure characterization and strain analysis.

Main Methods:

• Implementation of an image denoising autoencoder for EBSD pattern processing.
• Evaluation of denoised EBSD data for improvements in confidence index (CI) and image quality (IQ).
• Application of denoised datasets in high-resolution EBSD (HR-EBSD) for cross-correlative strain analysis.

Main Results:

• Autoencoder processing significantly improved CI, IQ, and the accuracy of orientation fits.
• Denoised EBSD data led to more reliable crystallographic and microstructural information.
• HR-EBSD strain analysis using denoised data reduced phantom strain and improved accuracy.

Conclusions:

• The autoencoder denoising method effectively enhances EBSD data quality and analytical accuracy.
• This approach facilitates higher speed EBSD data acquisition and more reliable microstructure analysis.
• Improved indexing accuracy from denoising benefits subsequent analyses like HR-EBSD strain mapping.