Deep learning-driven interval uncertainty propagation for aeronautical structures

Adaptive combination of line sampling for imprecise time-variant reliability analysis

verfasst von

Yan Shi, Michael BEER

Abstract

Interval Uncertainty Propagation (IUP) holds significant importance in quantifying uncertainties in structural outputs when confronted with interval input parameters. In the aviation field, the precise determination of probability models for input parameters of aeronautical structures entails substantial costs in both time and finances. As an alternative, the use of interval variables to describe input parameter uncertainty becomes a pragmatic approach. The complex task of solving the IUP for aeronautical structures, particularly in scenarios marked by pronounced nonlinearity and multiple outputs, necessitates innovative methodologies. This study introduces an efficient deep learning-driven approach to address the challenges associated with IUP. The proposed approach combines the Deep Neural Network (DNN) with intelligent optimization algorithms for dealing with the IUP in aeronautical structures. An inventive extremal value-oriented weighting technique is presented, assigning varying weights to different training samples within the loss function, thereby enhancing the computational accuracy of the DNN in predicting extremal values of structural outputs. Moreover, an adaptive framework is established to strategically balance the global exploration and local exploitation capabilities of the DNN, resulting in a predictive model that is both robust and accurate. To illustrate the effectiveness of the developed approach, various applications are explored, including a high-dimensional numerical example and two aeronautical structures. The obtained results highlight the high computational accuracy and efficiency achieved by the proposed approach, showcasing its potential for addressing complex IUP challenges in aeronautical engineering.

Organisationseinheit(en)

Institut für Risiko und Zuverlässigkeit

Externe Organisation(en)

The University of Liverpool
Tongji University

Typ

Artikel

Journal

Chinese journal of aeronautics

Band

37

Seiten

71-86

Anzahl der Seiten

16

ISSN

1000-9361

Publikationsdatum

12.2024

Publikationsstatus

Veröffentlicht

Peer-reviewed

Ja

ASJC Scopus Sachgebiete

Luft- und Raumfahrttechnik, Maschinenbau

Elektronische Version(en)

https://doi.org/10.1016/j.cja.2024.05.009 (Zugang: Offen)