Multi-modal data-driven microstructure characterization

TL;DR

The paper tackles the challenge of extracting microstructural information from complex EBSD Kikuchi patterns without user input. It introduces an automated workflow that integrates PCA/IPCA for dimensionality reduction, Gaussian mixture modeling for clustering, constrained non-negative matrix factorization with automatically selected reference components, and a variational autoencoder for latent-space representation. The approach yields autonomous hyperparameter decisions, robust grain/phase segmentation, and boundary detection, with an automatically determined ROI size roughly twice the characteristic grain diameter. Experimentally, the workflow demonstrates consistent alignment with conventional EBSD results and reveals intra-grain heterogeneities via anomalies and latent features, offering a transferable, data-driven framework for multi-modal microstructure analysis with practical implications for materials characterisation.

Abstract

Electron backscatter diffraction is one of the most prevalent techniques used for microstructural characterization. In recent years, there has been an increase in the use of data-driven methods to analyze raw Kikuchi patterns. However, most of these require user input and the interpretation of the data-derived features is often challenging and subject to \textit{informed interpretation}. By using a combination of principal component analysis, constrained non-negative matrix factorization, and a variational autoencoder along with information-theoretical considerations on a multimodal dataset, it is shown that a) automated decision on method-specific hyperparameters, here the number of components in principal component analysis, the number of components for constrained non-negative matrix factorization, and the selection of reference constraints; and b) latent space features can be mapped to physically-meaningful quantities. In addition, the recommended region-of-interest (ROI) size for optimal model performance is approximated automatically to be twice the characteristic grain size based on information content of the dataset. Implemented in a workflow, this allows for a transferable, dataset-specific autonomous data-driven phase and grain segmentation including grain boundary detection and the analysis of very-small-angle intra-grain variations to complement conventional electron backscatter analysis.

Figures (14)

• Figure 1: Workflow overview with indication of which methods are used: (a) Preprocessing, (b) Selection of the region of interest (ROI), (c) clustering and constrained non-negative matrix factorization, (d) Kikuchi pattern decomposition and segmentation, and (e) characterization outputs.
• Figure 2: Normalized cumulative variance as a function of the normalized index of the number of principal components. Red marker shows the detected knee-point.
• Figure 3: Mahalanobis vs. Euclidean contours (equal distance to centroid) aligned with principle components 1 and 2.
• Figure 4: Variational autoencoder architecture framework for Kikuchi pattern latent space creation.
• Figure 5: Conventional analysis phase map with selected ROIs (R1-R3; X1-X3; X regions within R regions)