Sequential sampling method for surrogate modelling of time-consuming finite element analysis

Abstract

Finite element (FE) model is widely used to assess and predict the structural performance under the uncertainty. Generally, this technique demands a large number of FE analysis (FEA). Despite a substantial improvement of computing performance, a challenging issue related to the computational cost still exists, so that a surrogate model has been gaining a considerable attention as a cost-effective substitute for a time-consuming FEA. Surrogate model is an approximate model of FEA to map inputs on outputs. The efficiency can be improved enormously through an accurate surrogate model. Conventionally with user-intervention, the trial-and-error method with different design of experiments has been carried out for the surrogate model construction process. To minimize the subjective user intervention, this study presents the sequential sampling strategy based on Kriging model for surrogate modelling of FEA. The sequential sampling method is based on the probabilistic criteria to guide sampling toward on a region which is likely to improve the accuracy of Kriging model or reduce an high uncertainty in prediction. The feasibility of the proposed method is first validated by using a mathematical test function. Then, it is applied as a substitute of iterative FEA to the Monte Carlo simulation with random variables such as loading, material and sectional properties. Finally, the performance and computational efficiency of the proposed method is discussed with the comparison of the result from FEA.