In recent years, variance-based global sensitivity analysis (GSA) methods have been increasingly used in a variety of engineering applications, especially in spacecraft thermal design. However, this kind of GSA estimation requires large number of samples to ensure sufficient accuracy, which makes GSA expensive to perform in case of modeling difficulties. In addition, multimode or highly skewed output distributions can lead to conflicting results using variance as uncertainties. Therefore, it is necessary to develop a simple and effective theoretical design principle and method for GSA with optimized space thermophysical modelling and accelerated GSA convergence to overcome the above problems.

In a study published in Aerospace Science and Technology, a research group led by Prof. GUO Liang from Changchun Institute of Optics, Fine Mechanics and Physics (CIOMP), Chinese Academy of Sciences (CAS) proposed an intelligent density-based GSA framework based on machine learning and multi-fidelity metamodels, called IDGSA-3M (IDGSA: intelligent density-based GSA framework; 3M: machine learning and MF metamodels).

This is the first application of density-based GSA in thermal design of a spacecraft. Moreover, IDGSA-3M is bundled with an intelligent batch processing system (IBPS) for real-time data exchange between MATLAB and NX/TMG, along with the use of many inexpensive low-fidelity sampling points to reduce the model evaluation cost, and few expensive high-fidelity sampling points to maintain high accuracy. The feature of the system is the automatic evaluation of models in their variability space based on sample input combinations, without supervision. Crucially, this system is better than traditional manual Monte Carlo estimation by at least five times.

An optimized radial basis function (RBF) neural network using improved mind evolutionary algorithms was applied to approximate the multi-fidelity metamodel of a spacecraft thermophysical model computed with IBPS, which is faster than traditional thermophysical models by more than 1000 times and has a computational accuracy of more than 99%. Furthermore, the density-based sensitivity index is obtained from its cumulative distribution function characterized by the output distribution of RBF.

In summary, both theoretical and experimental results demonstrate that the IDGSA-3M-based GSA evaluation works better than the traditional Monte Carlo-based and manually tuned variance-based GSA methods, providing better model evaluation accuracy and higher computational efficiency, together with a better statement of difference in importance between parameters. The most crucial advantage is that the whole process is automated, which helps to improve the efficiency of the spacecraft thermal design.

Contact:

Author: Prof. GUO Liang

Changchun Institute of Optics, Fine Mechanics and Physics, Chinese Academy of Sciences

Changchun, Jilin 130033. China

E-mail: guoliang@ciomp.ac.cn

Article links: https://www.sciencedirect.com/science/article/pii/S1270963821004375