HyOpt
In the development of new types of products, there are usually many requirements for highly stressed components. These can occur, for example, in terms of structural mechanics, aerodynamics or coupling and therefore represent major hurdles in engineering. Since the evaluation can usually only be provided by hardware-intensive simulations, the design process is usually iterative with the help of optimization methods.
Within the HyOpt project, highly integrative digital development processes in aeronautics, energy and transport are to be strengthened by means of a combination of physics-based models and machine learning (ML) approaches. For this purpose, in an interdisciplinary approach from aerodynamics, thermodynamics and structural mechanics, both physical process simulation and ML-based approaches in the sense of Reduced Order Models (ROMs) will be further developed in order to use them in combination both in the manual digital design process and within automated optimizations.
For this purpose, the physical process simulations are improved and reduced order models based on deep neural networks are developed, which approximate the high-dimensional aerodynamic and structural mechanical behavior of the considered components. Subsequently, these models are trained with the multi-fidelity simulation methods and/or experimental sample data and made available via a client-server architecture in order to significantly accelerate the respective designs under consideration of propagated uncertainties.