Physics-Informed Machine Learning for the Sensor Data Integration into Digital Twins
Nowadays, the powerful tools of machine learning are beginning to radically change computational science and engineering, allowing for the revolutionary development of new numerical techniques of a data-driven nature. For the effectiveness of real-time data integration into digital twins, it is also of critical importance to develop efficient, scalable, machine-learning-based algorithmic schemes for the incorporation of prior physical knowledge and observational data.
This project aims at developing principled physics-informed machine learning techniques to enable real-time sensor data integration throughout the lifecycle of engineering assets. Innovative physics-constrained machine learning models will be proposed to effectively encode heterogeneous data from both simulations and sensing systems, and robust numerical methods will be developed for the quantification of uncertainties introduced by sensor data and the modeling procedure. The developed numerical tools will guarantee both wide applicability to industrial modeling and enhanced efficiency in computational analysis, enabling real-time decision support via digital twins.