Los Alamos National Labs with logo 2021

Unsupervised phase mapping of X-ray diffraction data by nonnegative matrix factorization integrated with custom clustering


Graph showing different diagrams of results.

Unsupervised Machine Learning (ML) methods extracts sets of hidden (and often previously unknown) features from uncategorized datasets.

New machine learning method based on nonnegative matrix factorization (NMF) algorithms

The Nonnegative Matrix Factorization has shown great promise for the task of analyzing large volumes of X-ray measurements.

Its simplicity and ease-of interpretability offer great advantages, and several systems relying on this method were created and tested successfully on large datasets.

One solved key problem is determining the number of basis patterns (end members) in the data. Another is peak-shifting of X-ray patterns—a common consequence of lattice changes caused by alloying. Our new method addresses both of these problems.

Summary of full paper (pdf)