Data-centric materials science and engineering

This event will be a two-day workshop with roughly 10 short talks per day with the following breakdown. It is expected that many people may be interested in attending only one of the two days, although all participants will be welcome on both days.

Day 1: Digitization of images.

Given the importance of microstructure in determining material properties, in order to take advantage of computational science and machine learning approaches in materials discovery and manufacturing we need to find ways to digitise microstructural information.  In other words we need concise ways to capture the essence of a material’s microstructure so that we can relate it to the resulting materials properties. A salient theme of day one will be the pre-processing and digitization of images — including techniques for data handling, compression, feature extraction and dimension reduction.

Day 2: Optimising materials manufacturing and discovery via data-centric engineering.

Talks on the second day will have a distinctly more data-centric engineering focus, centered around materials informatics—i.e., techniques for post-processing and leveraging the value embedded in materials databases. Existing and novel approaches within the fields of uncertainty quantification, digital twinning, design of experiments and prediction will be covered. There is particular interest in Bayesian methods, as well as methods to combine known physical models and expert knowledge together with data driven approaches for extrapolation and predictive capabilities beyond just explanatory models.

Further details

Overall there are two key application areas. The first application area pertains to the use of artificial intelligence, machine learning and traditional mathematical techniques to extract information from images. Known as microstructure informatics, this new subject area seeks to use computational methods for extracting mathematical descriptors / features to capture the essence of 2D and 3D images. While materials microstructures present particular challenges, this topic extends into may other areas of science where data extraction from images is important (e.g., PET/medical X-ray imaging, histology, mineralogy, etc.). As a result those with an interest in characterising and quantifying images in other fields are welcome.

The ability to digitise microstructural data is a necessary precursor to the second application area, which is the use of machine learning and data science in accelerating the design and development of new materials. This topic, sometimes called materials informatics, is a core application area of data centric engineering, and has a strong manufacturing and industry 4.0 focus, in terms of refining processes to optimise materials performance.

A range of computational statistics, applied mathematics, and machine learning technology is required to achieve these goals, including but not limited to probabilistic Bayesian methods, data assimilation, numerical analysis, surrogate modeling, non-parametric statistics, and supervised, active, and reinforcement learning. Accelerated computational design and manufacture of materials requires:

• Effective computational methods/algorithms and tools;
• High-throughput make, test, and characterisation tools;
• Databases and materials informatics tools including a widely accepted taxonomy for describing materials microstructure.

These workshops will bring together those with interests in image digitisation, materials science and materials modelling, digital twining, machine learning approaches, materials databases and high performance computing, and high throughput testing. We will discuss the state of the art and strategically prioritise future research initiatives at this fertile intersection