Coupling multi-fidelity kriging and model-order reduction for the construction of virtual charts

Abstract

This article presents the coupling between multi-fidelity kriging and a database generated on-thefly by model reduction to accelerate the generation of a surrogate model. The two-level multi-fidelity kriging method Evofusion is used for data fusion. The remarkable point is the generation of low-fidelity and highfidelity observations from the same solver using the Proper Generalized Decomposition, a model-order reduction method. A 17× speedup is obtained here on an elasto-viscoplastic test case.

Model-data-driven constitutive responses: application to a multiscale computational framework

“…In this regard, since the strategy developed in this work for the definition of the training dataset is based on an incremental approach, it is suitable to be employed for incorporating a strain input outside the current training domain. The framework could be also extended to incorporate multi-fidelity kriging, where different simulation tools (converged and nonconverged results, or different level of simulations) are combined in a unique kriging metamodel, see for instance [35]. These improvements will allow to extend the proposed implementation to three-dimensional problems, to extend the proposed approach to time-dependent processes, and to consider more than two length-scales.…”

Multifidelity adaptive kriging metamodel based on discretization error bounds

“…In the context of optimization, References 24 and 25 adapt the stopping criterion of the iterative contact solver to use multifidelity kriging. In Reference 26, a multifidelity approach is used on a elasto‐viscoplasticity problem with different proper generalized decompositions. It is to be noted that usually, the choice of levels of discretization or grids is done a priori and the construction of the design of experiments is based on heuristics.…”