As part of the digital transformation, more and more industries realize the potential of digital twins for system optimization and behavior prediction. Research and development at Fraunhofer SCAI address several aspects of this promising but highly complex topic.
A digital twin is a detailed digital representation of a physical system that describes the current and potentially past states of the system. Early versions of this concept from around the year 2000 implied a comprehensive database of digital information spanning the entire life cycle of the system. Today, it is clear that the full potential of the concept can only be exploited if a digital twin entails integrated multi-physics-multi-scale simulations that are matched against sensor data and provide high-quality, real-time predictions of system behavior.
Apart from few quite specific applications, high-fidelity digital twins are currently more vision than reality. The National Aeronautics and Space Administration and the Air Force in the United States, which define the digital twin as a key technology in their technical roadmaps, estimate a 25 year development frame to realize full-fledged digital twins. The general challenges are heterogeneity of data, complexity in the interaction of models, and flexibility in the event of unforeseen changes in requirements. However, with digital sensor technology and computational resources constantly improving, the concept is becoming more and more interesting for various branches of industry. Fraunhofer SCAI – with its strong background in machine learning, optimization, and numerical simulation – contributes in many ways to progress in the development of digital twins. Application areas at SCAI include new materials, manufacturing, and life sciences, while digital twins for energy networks are starting to emerge as a component for the energy transition.
Simulation, coupling, and standardized interfaces
A new ingredient for the realization of digital twins are numerical simulations. Here, SCAI has developed numerous software packages for the simulation and optimization of industrial applications. Recent developments are the flexible simulation environments PUMA and MESHFREE, which are particularly suited for large changes in geometry and topology.
With the MpCCI CouplingEnvironment, SCAI offers a vendor-neutral solution for the coupling of simulation programs for different physical disciplines. MpCCI thus allows a more comprehensive digital simulation than conventional mono-disciplinary codes and has established itself as the de facto standard for multiphysics simulations.
Complex engineering workflows represent a further step in the direction of digital twins. The lack of software standards in virtual engineering workflows and incompatible interfaces for the transfer of virtual material information not only causes additional costs and complex manual adjustments, but also leads to inflexible IT solutions, loss of information, and considerable delays in the entire design process. The standardization of material interfaces in computer-aided engineering (CAE) is therefore of decisive importance whenever material behavior is at the center of product and process design. The European VMAP project, coordinated by Fraunhofer SCAI, aims at a common understanding and interoperable definitions for virtual material modeling in CAE.
Model generation and validation
High-fidelity simulations continue to be too expensive for many applications such that meta modeling remains a highly relevant topic. In the context of digital twins, model validation based on local quantities of interest reaches its limits. It requires more complex similarity and quality measures that take complete geometries into account, e.g., to gain a deeper understanding of car crashs. In several projects, compact data-driven representations for large simulation bundles have been developed. The SCAI software DesParO, a toolbox for the intuitive exploration, automatic analysis, and optimization of parameterized problems in production processes, supports the generation of multi-level meta models. The new Fraunhofer flagship project EVOLOPRO investigates methods of multitask and transfer learning to enable more efficient surrogate modeling across multiple stages in process chains. In the MADESI project, SCAI investigates how the interaction of sensor data and numerical simulation can be used to increase the capabilities of machine learning methods for the predictive maintenance of wind turbines.
Digital twins down to the material level
Experiments and simulations are used on many scales to better understand processes and improve their usage. In addition to macroscopic simulations as mentioned above, SCAI is addressing high-resolution sensor data in the millimeter range in the Fraunhofer DigitalTPC project. This requires the acquisition and processing of high-resolution sensor data. This information is used to detect defects in the fiber structures of thermoplastic composite tapes at an early stage and to optimize subsequent work steps (e.g. laying processes) accordingly. The heterogeneous microstructure of the composite material and its influence on the manufacturing process poses a challenge for process control and quality assurance and requires continuous digitization of the production process. In this context, SCAI is involved in numerical simulations for the development of new materials with improved properties, in particular with the software Tremolo-X for material models on the atomistic scale.
Digital twins for thermoplast lightweight construction
The potential of digital twins is still largely untapped for cross-value-added chain and material-triggered process control. This also applies to plastic-based composite structures. The digitalTPC project is intended to demonstrate this potential by means of the hybrid injection moulding technology, which is currently establishing itself on the market and is capable of large series production, in which continuously fibre-reinforced tape-laminate semi-finished products are formed and back-injected. digitalTPC aims at the comprehensive and holistic consideration of all sub-process steps from semi-finished product to component production. Relevant material, process and component characteristics (e.g. fibre orientation, pore content, degree of consolidation, temperature, pressure, etc.) are to be measured, recorded and virtually modelled and analyzed in a digital twin in local resolution across the entire real value chain. The challenge of the project is the material- and process-related intelligent acquisition of sensor data and their linkage with the integrated simulation chain within the framework of the digital twin by SCAI.
Virtual patients in dementia research
The concept of a digital twin has recently also been adopted by the Innovative Medicine Initiative (IMI) project AETIONOMY in the form of a virtual dementia cohort (VDC). A complete VDC is a derivative of a real-world clinical study; it represents all relevant variables and the dependencies between these variables that have been measured in the clinical study. Such VDCs allow to simulate huge numbers of virtual patients that do not differ from real-world patients. VDCs open completely new perspectives for Alzheimer research – ranging from sharing of patient-level data without compromising patient-data privacy to clinical trial simulation.