Research Projects

The industry heavily relies on large-scale computer-based finite-element (FE) models for asset management decisions in infrastructural systems like buildings, bridges, and wind turbines. These models are often idealized representations of the physical systems, leading to misspecification errors. However, uncertainties in inferences and predictions directly affect risk and health management decisions. This project aims to create a robust and efficient virtual proxy of an infrastructural system called digiTwin by fusing measurement data and FE models. The digital twin paradigm will create predictive models that add robustness to the current design methodology, introduce robust, scalable, and efficient statistical inference tools, and transform uncertainty management of key industrial assets. The main research works include designing appropriate error models for digiTwin, developing scalable, likelihood-free, robust variational Bayesian inference algorithms, and verifying and validating these procedures. The project will also create a software package for robust variational inference techniques and illustrate procedures developed on broad-ranging mechanical and structural engineering problems. The project will deliver transformative tools to integrate confidence in digital twin technology for infrastructures like bridges, buildings, and wind turbines.