A comprehensive multi-source uncertainty quantification method for RANS-CFD based on sparse multi-task deep active learning

doi:10.1016/j.ast.2025.110483

A comprehensive multi-source uncertainty quantification method for RANS-CFD based on sparse multi-task deep active learning

2025-06-20

0

PRE

AI

DOI：10.1016/j.ast.2025.110483

Original

Original request for help

Save

Abstract

En 中文

• The main innovations of this work include: • A novel comprehensive UQ framework for CFD is developed, integrating uncertainty propagation, model parameter estimation, and model averaging to efficiently address uncertainties in model input, parameters, selection, predictions, and experimental data. • To reduce the high computational cost of comprehensive UQ in CFD, a sparse multi-task Bayesian deep active learning method is proposed. This approach integrates a multi-task Bayesian neural network (BNN) metamodel for simulations with multiple turbulence models and uses a clustering-based active learning strategy to improve efficiency. • The proposed UQ method is applied to aerodynamic analysis, showing improved CFD prediction accuracy with confidence intervals that encompass all experimental data, demonstrating the method's effectiveness.

A comprehensive multi-source uncertainty quantification method for RANS-CFD based on sparse multi-task deep active learning

Abstract

Journal

Organization