AI-Twin: Retrieving digital twins using Physics-Informed AI for in-situ and operando imaging reconstructions

AI-Twin seeks to streamline the development of digital twins from complex in-situ and operando experiments carried out at large-scale facilities such as MAX IV in Lund, Sweden.

Current data analyses of in-situ and operando X-ray imaging experiments at large-scale facilities are complex, restricted to experts, and require years to process. This hinders its applicability and the integration of the acquired knowledge into technological developments.

Thus, simplifying such processes is crucial to extend the accessibility and impact of large-scale facilities beyond the expert community.

To simplify this procedure, we propose to reduce such complex processes to a single reconstruction step to retrieve digital twins.

This involves the development of: i) new AI-based Physics-Informed Neural Networks (PINNs) reconstructions that retrieve the physical laws behind in-situ and operando experiments and ii) optimal structure-preserving Finite Element Methods (FEMs) to test and transfer those laws into actual digital twins.

Such developments are timely due to the AI research in PINNs and GPU hardware infrastructures.

As a proof-of-concept demonstration of our method, we will focus on the study of fluid dynamic processes, which are of interest to academia as well as the pharmaceutical, transport, and packaging industries.