Hier finden Sie wissenschaftliche Publikationen aus den Fraunhofer-Instituten.

Assessing Model Form Uncertainty for a Suspension Strut using Gaussian Processes

: Feldmann, Robert; Platz, Roland

Papadrakakis, M. ; European Community on Computational Methods in Applied Science -ECCOMAS-; International Association for Structural Safety and Reliability -IASSAR-; National Technical University of Athens -NTUA-, Institute of Structural Analysis and Antiseismic Research, School of Civil Engineering:
UNCECOMP 2019, 3rd International Conference on Uncertainty Quantification in Computational Sciences and Engineering. Proceedings : Crete, Greece, 24-26 June 2019
Athens: NTUA, 2019
ISBN: 978-618-82844-9-4
International Conference on Uncertainty in Computational Sciences and Engineering (UNCECOMP) <3, 2019, Hersonissos>
Fraunhofer LBF ()
uncertainty quantification; structural dynamics; Model Form Uncertainty

In this paper, a modular spring-damper system that is integrated into a spacetruss structure is considered that was developed in the collaborative research centre SFB 805“Control of Uncertainty in Load-Carrying Structures in Mechanical Engineering” at the Technische Universität Darmstadt. An idealized two degree of freedom (2DOF) model serves as a mathematical model to describe the dynamical system behaviour, yielding a system of two coupled ordinary differential equations (ODE) of second order. Previous own research already addressed the dynamic behaviour of the suspension system as regression curves from experiments for both stiffness and damping behaviour. Combining the regression models with the system equations of a 2DOF model of the modular spring-damper system yielded several model candidates to describe the dynamic behaviour. The resulting model form uncertainty is addressed in the framework of a model selection process. The approach employed in this paper uses a simplified form of the Kennedy and O’Hagan framework. Assuming that all models incorporate a model error, measurements of a system can be expressed as a the sum of the simulation model output, a discrepancy function and measurement noise. The discrepancy function gives information about the accuracy of the simulation model. It can therefore be used to compare model candidates and thus assess model form uncertainty. Among the approaches to model the discrepancy function, Gaussian processes (GP)have proofed to be suitable due to their versatility. Hence, for each model candidate, a GP representation for the discrepancy function can be determined based on experimental data. This paper shows the comparison and evaluation of the model candidates’ discrepancy functions. Characterization of the underlying GP with regard to its confidence intervals is employed as a measure to select models that represent the dynamic behaviour of the modular spring-dampersystem most adequately.