HYB - Hybrid AI Analytics
Hybrid AI Analytics
- Develop robust, stable and explainable data-driven models for physical systems
- Constrain models to enforce meaningful predictions
- Transfer data-driven models from simulations to reality
- Characterise and quantify uncertainty of data-driven models
This work package will develop methods to predict and reduce the uncertainty of data-driven models. The models will be constrained by existing knowledge, allowing to interpret the model (explainable AI) and reducing the amount of required training data. Applying this methods on real world applications will allow the industry partners to better predict the behavior of their facilities and improve their simulations, e.g. for condition monitoring, predictive maintenance, optimal utilization.
Short description: The aim of this project is to develop methodology for transfer learning for quantifying uncertainty from non-representative training data , i.e. how we can learn from simulated data and utilize this to quantify uncertainty in physical systems. Developed methodology will be tested in two usecases.
Time perspective: 2020-2022
Usecase 1:Virtual flow meter
Involved partners:
Usecase 2: Predictive maintenance for wind turbines
Involved partners:
Research stay at Brown university
Katarzyna Michalowska, one of the NorwAI PhD students and researcher at SINTEF Digital is halfway into a one-year research stay at Brown University in the United States as a part of the NorwAI project. Katarzyna shares her experiences so far from working with the CRUNCH group at Brown.

2023-02-09