Polymorphic uncertainty quantification for engineering structures via a hyperplane modelling technique

Qihan Wang

doi:10.1016/j.cma.2022.115250

2022-08-01

PRE

DOI：10.1016/j.cma.2022.115250

原文链接

原文求助

摘要

En 中文

This paper proposes a hyperplane modelling technique aided polymorphic uncertainty quantification strategy for various engineering applications. Accumulative experiences from industrial applications have continuously revealed that probability distribution characteristics for system properties cannot always be precisely determined, due to the scarcity of available information. Thus, the polymorphic uncertainty is introduced to consider fuzzy and random uncertainty simultaneously. To tackle such uncertainty quantification challenges, Multiple Target Strategy (MTS) and Single Target Strategy (STS) are proposed. MTS is a generalized polymorphic uncertainty quantification strategy, capable of providing sufficient amounts of the statistical information of the concerned structural response. On the contrary, STS considers the polymorphic uncertainty quantification in a more precise and efficient manner. Both strategies are implemented based on their respective hyperplane models. For hyperplane model construction, a newly developed ensemble meta-regression technique, namely AdaBoost Extended Support Vector Regression (Ada-X-SVR), is embedded with a novel kernel function. The proposed hyperplane modelling aided polymorphic uncertainty quantification framework provides the feasibility of both generalized and targeted estimations, autolearning and information update features. Furthermore, to demonstrate the applicability and computational efficiency of the proposed strategies, a benchmark analysis with an available analytical solution and three engineering applications with linear and nonlinear performance are fully investigated. (c) 2022 Elsevier B.V. All rights reserved.

Keyword:

Hyperplane construction technique

AdaBoost Extended Support Vector Regression (ada-X-SVR)

Polymorphic uncertainty quantification

Machine learning

Engineering application