An Adaptive Meta-Modelling Approach for Multi-Dimensional Correlated Flow Field Responses

ABSTRACT

It has become common practice to build inexpensive surrogate models to supplant time-consuming numerical simulations. By gradually increasing the number of training samples, the efficiency of model construction can be significantly improved. This study proposes an adaptive meta-modelling approach for multi-dimensional correlated responses within the framework of proper orthogonal decomposition (POD) and the Kriging model. The algorithm begins with the adaptive sampling algorithm for each reduced-dimension response, which integrates the prediction variance, distances between samples, and sensitivity indicator of each parameter. The adaptive sampling criterion for each reduced-dimension response is weighted by the energy of the modal, forming the adaptive sampling algorithm for multi-dimensional correlated responses. Tests on an analytical function and M6 wing simulation show that, under the same number of training samples, the proposed adaptive algorithm results in a model with lower prediction error than the random sampling algorithm, offering a more efficient model for flow field prediction.

KEYWORDS:

• Adaptive sampling
• surrogate model
• multi-dimensional correlated responses
• flow field reduction
• machine learning

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This work was supported by NSAF Joint Fund: [Grant Number U2230208] and the National Numerical Wind Tunnel Project.