Probabilistic uncertainty analysis of an FRF of a structure using a Gaussian process emulator

This paper introduces methods for probabilistic uncertainty analysis of a frequency response function (FRF) of a structure obtained via a finite element (FE) model. The methods are applicable to computationally expensive FE models, making use of a Bayesian metamodel known as an emulator. The emulator produces fast predictions of the FE model output, but also accounts for the additional uncertainty induced by only having a limited number of model evaluations. Two approaches to the probabilistic uncertainty analysis of FRFs are developed. The first considers the uncertainty in the response at discrete frequencies, giving pointwise uncertainty intervals. The second considers the uncertainty in an entire FRF across a frequency range, giving an uncertainty envelope function. The methods are demonstrated and compared to alternative approaches in a practical case study.

► A Bayesian metamodel (an emulator) is built for a finite element model.
► The multivariate nature of finite element models is taken into account.
► This facilitates probabilistic uncertainty analysis of an expensive model.
► Our method produces an uncertainty envelope for the frequency response function.
► This has distinct advantages over pointwise uncertainty intervals.