Hybridization of data-driven and physics-based models for digital twins

Hybridization of Data-driven and Physics-based Models

Companies are forced to increase productivity in order to ensure their competitiveness. Digital twins offer needed potentials. Basic digital models of the production systems, as typically created during the development phase, are tailored to the requirements of specific use cases and enriched with data over the entire product life cycle. These expanded models enable new approaches like predictive maintenance. With regard to predictive maintenance, in addition to analysing operational data to e.g. estimate remaining useful life, digital twins based on a hybridisation of data-driven and physics-based models identify potential failure mechanisms to facilitate the diagnosis of failures. Concepts for the hybridization are needed, resulting in an up-to-date representation of the production system as well as a generation of physics-based simulation data to optimize data-driven models. In this talk, use cases and associated requirements for digital twins focusing on the development of data-driven and physics-based models of productions systems and their hybridization as well as general lessons learned of implementing digital twins in companies will be covered.