Identification and synthesis of components for uncertainty propagation

Doctoral thesis, 2020

For automotive structures, built-up of hundreds of components with property spread, knowing the effects of component variability and its propagation through the system assembly is important in order to mitigate noise and vibration problems. To increase the understanding of how the spread propagates into variability in built-up structures, both experimental and computational aspects are considered in this thesis.

In the first part of the thesis, methods to identify models from experimental data are developed. Physical insight is often required for accurate experimental models. To this end, two-phase state-space system identification algorithms are developed where physically motivated residual states are included and physically motivated constraints are enforced. The developed identification algorithms are used together with finite element model updating to investigate the variability in dynamical properties between nominally identical components. Furthermore, the accurate and physical experimental models are used in synthesis with the updated finite element models. It is shown that experimental-analytical synthesis of complex and modally dense structures is possible, and that the component variability can be predicted in such assemblies.

In the second part of the thesis, methods to reduce the computational cost of variability analysis are developed. An efficient multifidelity interface reduction method is developed for component synthesis. It is also shown that modal truncation augmentation vectors can be computed efficiently from the multifidelity interface reduction basis. Lastly, an efficient uncertainty propagation method is developed, based on a second-order modal model. Utilising several approximations, it is shown that industrial-sized models can be handled with small loss in accuracy compared to a purely Monte Carlo based approach.