Detail publikace

# Surrogate modelling in the stochastic analysis of concrete girders failing in shear

Originální název

Surrogate modelling in the stochastic analysis of concrete girders failing in shear

Anglický název

Surrogate modelling in the stochastic analysis of concrete girders failing in shear

Jazyk

en

Originální abstrakt

The paper is focused on the reliability analysis of a real structure - a prestressed concrete girder failing in shear. This was represented using the non-linear finite element model (NLFEM), which was validated via a destructive laboratory experiment. First, the basic theoretical background of polynomial chaos expansion (PCE) and the employed algorithm is presented, and then a description of the computational finite element model of the structure is given. The full stochastic model is based on laboratory experiments on concrete specimens and JCSS probabilistic model code, though it was reduced according to previous research. The experimental design was created using Latin Hypercube Sampling (LHS) in combination with Nataf transformation (Gaussian copula). The use of the reduced stochastic model and LHS lead to small-size experimental design: in this case only 100 simulations were performed. Once the surrogate model based on PCE had been created, it was possible to perform a million simulations within a negligible amount of time, thus allowing the design value of resistance and Spearman coefficients to be easily obtained. Moreover, Sobol indices can be derived from PCE without any additional computational effort. The obtained results are discussed in the last part of the paper, special attention is given to sensitivity analysis and the role of statistical correlation among random variables.

Anglický abstrakt

The paper is focused on the reliability analysis of a real structure - a prestressed concrete girder failing in shear. This was represented using the non-linear finite element model (NLFEM), which was validated via a destructive laboratory experiment. First, the basic theoretical background of polynomial chaos expansion (PCE) and the employed algorithm is presented, and then a description of the computational finite element model of the structure is given. The full stochastic model is based on laboratory experiments on concrete specimens and JCSS probabilistic model code, though it was reduced according to previous research. The experimental design was created using Latin Hypercube Sampling (LHS) in combination with Nataf transformation (Gaussian copula). The use of the reduced stochastic model and LHS lead to small-size experimental design: in this case only 100 simulations were performed. Once the surrogate model based on PCE had been created, it was possible to perform a million simulations within a negligible amount of time, thus allowing the design value of resistance and Spearman coefficients to be easily obtained. Moreover, Sobol indices can be derived from PCE without any additional computational effort. The obtained results are discussed in the last part of the paper, special attention is given to sensitivity analysis and the role of statistical correlation among random variables.

Dokumenty

BibTex

``````
@inproceedings{BUT157213,
author="Lukáš {Novák} and Drahomír {Novák}",
title="Surrogate modelling in the stochastic analysis of concrete girders failing in shear",
annote="The paper is focused on the reliability analysis of a real structure - a prestressed concrete girder failing in shear. This was represented using the non-linear finite element model (NLFEM), which was validated via a destructive laboratory experiment. First, the basic theoretical background of polynomial chaos expansion (PCE) and the employed algorithm is presented, and then a description of the computational finite element model of the structure is given. The full stochastic model is based on laboratory experiments on concrete specimens and JCSS probabilistic model code, though it was reduced according to previous research. The experimental design was created using Latin Hypercube Sampling (LHS) in combination with Nataf transformation (Gaussian copula). The use of the reduced stochastic model and LHS lead to small-size experimental design: in this case only 100 simulations were performed. Once the surrogate model based on PCE had been created, it was possible to perform a million simulations within a negligible amount of time, thus allowing the design value of resistance and Spearman coefficients to be easily obtained. Moreover, Sobol indices can be derived from PCE without any additional computational effort. The obtained results are discussed in the last part of the paper, special attention is given to sensitivity analysis and the role of statistical correlation among random variables.",