Material Data Types
Five distinctive data types fall out of a survey of fifty common materials characterization and modeling techniques: scalar, series, spectral, categorical and image data. This post, which is expected to be updated with some degree of regularity, is dedicated to looking at examples of these data types and how those data types are processed to extract critical features, i.e. featurized, from the raw data. By subdividing these data into categories, the goal is to identify featurization techniques that may be readily transferrable to other measurement modalities that produce similar classes of materials data.
This is meant to be an evolving list, so please do feel free to recommend examples.
Scalar data is comprised of both numerical data and vectors of numerical data, e.g. phase transition temperatures upon repeated cycling.
A 1D line scan is the archetype for series data; data collected while an independent variable sweeps through a range of values. Time series data, stress-strain curves, magnetic hysteresis, etc.
Spectral data is a series data, but with distinctive features that make featurization of this data type distinctive. For example, X-ray diffraction, IR-spectroscopy, or NEXAFS measurements.
Scalar, series, and spectral data are generally continuous. Categorical data, such as thermodynamically stable phases as a function of composition and temperature, take on discrete values over a range of input conditions.
Much of the data collected in materials and manufacturing is image data. Electron microscopy, optical microscopy, and digital image correlation, to name a few.