Publication detail

Utilization of artificial neural networks for global sensitivity analysis of model output

KALA, Z. LEHKÝ, D. NOVÁK, D.

Original Title

Utilization of artificial neural networks for global sensitivity analysis of model output

Type

conference paper

Language

English

Original Abstract

The paper deals with the application of artificial neural networks to sensitivity measurement of the output quantity to the variability of input quantities. The original nonlinear FEM model calculates ultimate load-bearing capacity of a T-shaped prestressed concrete roof girder. Latin hypercube sampling algorithm is used to generate samples of input variables. The global Sobol sensitivity analysis is proposed to understand the effect of the input variability on the quantity of interest. The outputs of the Sobol sensitivity analysis are verified by subsequent two sensitivity analyses. The first studies show that artificial neural networks are very promising for effective evaluation of global sensitivity analysis. Artificial neural networks do not eliminate mutual interaction among input quantities; it is a very important piece of knowledge connected with maintaining the satisfactory accurateness of the reliability computation.

Keywords

artificial neural networks, Sobol, global sensitivity analysis, interactions, reliability

Authors

KALA, Z.; LEHKÝ, D.; NOVÁK, D.

Released

24. 7. 2019

Publisher

American Institute of Physics Inc.

Location

MELVILLE, USA

ISBN

978-0-7354-1854-7

Book

AIP Conference Proceedings

ISBN

0094-243X

Periodical

AIP conference proceedings

Year of study

2116

State

United States of America

Pages from

120005-1

Pages to

120005-4

Pages count

4

URL