Surrogate-Based Uncertainty Propagation in Aerostructural Analysis of a Wing in Subsonic Flow

Designing aerospace vehicles is a process that requires multidisciplinary analysis (MDA) that involves several disciplines. Such aerospace vehicles are subject to uncertainties in their operational parameters and geometries. Methods for analyzing the uncertainty of such coupled systems are prohibitive, given that Monte Carlo simulations require a large number of evaluations of a high-fidelity model of a coupled system. The computational cost can be alleviated by using cheap-to-evaluate surrogate models that are created from data generated using high-fidelity models. Typically, such surrogate-based approaches have focused on uncertainty analysis of scalar quantities in MDA while neglecting vector-valued quantities. Additionally, such surrogate models do not provide a simultaneous prediction of the vector-valued and scalar quantities, such that the uncertainties in both can be analyzed. This paper proposes a new and unique surrogate-based framework to propagate uncertainties from operating flow conditions to both vector-valued and scalar quantities of interest simultaneously in the aerostructural analysis of a wing operating at a Mach number of 0.5 and an altitude of 10,000 m. The uncertainty analysis is conducted using quasi-Monte Carlo simulations (QMCS). The results show that the proposed surrogate-based approach can accurately predict both vector-valued and scalar quantities for the aerodynamics and structural disciplines. The normalized root mean squared error for each quantity is on the order of 0.0001 to 0.001, indicating a very high accuracy achieved by the surrogate modeling approach. The results of the uncertainty analysis show that the absolute difference between the summary statistics obtained from the proposed surrogate-based approach and by directly running QMCS on the high-fidelity model is on the order of 0.001 to 0.1. The summary statistics of the vector-valued quantities also bear a large similarity to those obtained from a direct QMCS of the high-fidelity model. This illustrates that the proposed approach is a promising framework for efficient uncertainty propagation in coupled multidisciplinary systems and is able to quantify uncertainties in both scalar and vector-valued quantities of interest.