Main project

AI for materials characterisation

Physics-informed machine learning for the prediction of material properties

• Physics-informed ML
• Industry and engineering
• Ongoing

Illustration. Physics-informed models describe how the properties of a material evolve, consistently with the underlying physical and chemical processes.

Challenge

The development of new materials relies on extensive laboratory testing to characterise how composition and processing conditions affect the final properties. Models learned only from data need large data sets and may produce predictions that are inconsistent with the underlying physics and chemistry.

Approach

We develop physics-informed machine learning models that combine knowledge of the physical and chemical processes involved with experimental measurements, so that reliable predictions can be obtained from a limited number of tests.

Expected impact

A more efficient characterisation of new materials, with fewer experiments and with models that materials scientists can interpret.