Vibration analysis of structure with uncertainty is computationally costly, especially when the finite element model involved has high dimensionality. In this research a combination of two-level Gaussian processes and Bayesian inference is employed to facilitate the development of an efficient and accurate probabilistic order-reduced model. We first employ the two-level Gaussian processes emulator to integrate together small amount of high-fidelity data from full-scale finite element analysis and large amount of low-fidelity data from order-reduced component mode synthesis (CMS) model to improve the response variation prediction. We then utilize the improved response variation prediction on modal characteristics to update the CMS model in the probabilistic sense. The effectiveness of this method is demonstrated through a case study.
展开▼
