Development and validation physical-guided data science models for finite element applications

Development and validation physical-guided data science models for finite element applications Real-time stress predictions require the continuous interaction of measurements and a complex, refined finite element model that is updated real-time based on measured conditions. Previous work has shown that this interaction can be based on simplified physical relationships that allow quick changes in the model based on actual conditions. However, even when low-fidelity models are used in conjunction with experimental measurements, the resulting computational times are still too large to be feasible with real-time predictions. This complexity has therefore restricted the ability to efficiently monitor the stresses in complex systems during operations without continuous experimental monitoring. Data science, with its ability to extract 'knowledge' from large volumes of data, has the potential to be used to predict transient stresses that a structure experiences at any given time. The objective of this thesis is to create a novel interactive framework for combining scientific knowledge of finite element methods (FEM) in mechanical systems with data science methods to predict the full-field dynamic behavior of the system. The thesis will focus on developing data-driven models based on finite element models and real-time experimental data to create mid-fidelity surrogate models that can learn the structural behavior from rich finite element simulations and predict the dynamic behavior of a system based on actual measurements. The data-driven model bridges the need of accuracy given by high-fidelity finite element simulations, of low computational cost provided by low-fidelity models and of real-time stress predictions based on actual measurements. The data-driven model will provide quick and reliable predictions of the stresses and accelerations in the presence of highly non-linear, transient response and in the presence of complex couplings. At the end of this thesis, the following goals will be attained: •??development of data-driven model to determine time-varying stresses in structural components. •??definition of a database using finite elements of simple components and complex components •??numerical validation of the approach •??experimental validation doing vibration testing of simple components and complex components.