Data Models for Expression of Uncertainty in Materials Data

Summary

One of the major goals of the Materials Genome Initiative (MGI) is to facilitate the exchange of materials data to speed material discovery and development. The Materials Data Curation System (MDCS) is a comprehensive system developed under the MGI designed to enable investigators to easily share diverse datasets with domain-specific metadata that can be “automatically understood” by different software tools used for analysis and display of data.

As a component of the larger MGI project Facilitating the Development of Modular Data Models in Materials Science, the goal of this project is to develop and disseminate broadly applicable data models to describe the expression of uncertainty in materials measurements. Assessment of uncertainty is critical to all types of measurements and as such is a critical part of any exchange of measurement-based data.

When completed, the data models developed for the expression of uncertainty in this project can be assembled, at the user’s discretion, into the Template for any particular data set that includes measurement uncertainty assessed using one or more of the supported methods. This approach both decreases the burden on the user and increases the interoperability and discoverability, while still providing flexibility for users to define a unique Template. 

Description

The two main parts of this work are

1. definition of appropriate data structures for different types of uncertainty analyses, and
2. implementation of these data structures into analysis and visualization software for demonstration, testing, and use.

The different types of uncertainty analyses to be covered initially include analyses using the methods described in the JCGM Guide to the Expression of Uncertainty in Measurement (GUM), in Supplement 1 to the GUM (GS1), and Bayesian uncertainty assessments. All of these methods are based on the use of probabilistic models for the data and result in probability-based assessments of uncertainty such as confidence or credible intervals. After further evaluation, analogous data structures for methods based extensions of probability or on non-probabilistic methods also may be developed.

As data structures are developed, our goal is to add the capability for reading and writing these structures to one or more software packages for testing and demonstration of their use, as appropriate. The first analysis package into which we are incorporating the GUM and GS1 data structures is the NIST Uncertainty Machine, a free, web-based tool for the general assessment of measurement uncertainty using the GUM and GS1 methods. The figure below shows the NIST Uncertainty Machine’s input page.

The software currently only accepts configuration files using its own input format, but will be able to read and write MDCS structures soon as well.