Hybrid models of production processes & digital twins
In AI-PROFICIENT, digital twins are linked to a range of AI services, such as predictive production quality assurance and process optimization. In the report D3.1: AI-PROFICIENT hybrid models and digital twins first version (state of the art, designing and specification), Sirpa Kallio (VTT), Kerman Lopez de Calle (TEK), Eider Garate (TEK) and Aitor Arnaiz (TEK) present the state of the art of hybrid modelling and digital twins, covering the techniques from first principles modelling, to fully data based surrogate models and digital twins based on these approaches.
As the authors explain, ‘depending on the availability of data and first principles models and theories, different modeling approaches are preferred in the derivation of a digital twin.’ In specific:
- Surrogate modelling requires either a large data set that covers the considered conditions, or a possibility to carry out designed experiments
- First principles modeling requires that the relevant phenomena can be described in sufficient details and with acceptable accuracy by mathematical formulations. Here in most cases, the models need to be reformulated and simplified to achieve sufficient accuracy required for on-line use in the digital twin
- In many cases, hybrid modelling approaches that combine data-driven and first principles modelling can be the optimal method