Adaptive surrogate models with partially observed information

Surrogate models have been developed to replace expensive physical and reduce the computational cost in various engineering applications, such as reliability analysis uncertainty quantification. Gaussian process (GP) model exhibits superior performance among surrogate with a distinguishing feature of estimating uncertainty. However, fully observed datasets are generally required establish GP model, which is often scarce obtain complex systems. Partially overserved information available relatively plentiful collected datasets, contain data from different sources that multi-fidelity or dimensionality missing values. Therefore, correctly accounting for partially important order take advantage all increase prediction be developed. This paper presents new method modeling system information, integrates Bayesian latent variable (BGPLVM) adaptive sampling iteratively select observable training sample points improve efficiency. A novel approach considering frame proposed refine model. To best authors' knowledge, this first work designing approaches dataset containing The numerical experiments demonstrated can effectively use including both points. methodology provides an accurate cost-effective solution extra developing models.