Multilayer-perceptron network ensemble modeling with genetic algorithms for the capacity of bolted lap joint

  1. Fernández-Ceniceros, J. 1
  2. Sanz-García, A. 1
  3. Antoñanzas-Torres, F. 1
  4. Martínez-De-Pisón-Ascacibar, F.J. 1
  1. 1 Universidad de La Rioja
    info

    Universidad de La Rioja

    Logroño, España

    ROR https://ror.org/0553yr311

Revista:
Lecture Notes in Computer Science

ISSN: 0302-9743

Año de publicación: 2012

Volumen: 7208 LNAI

Número: PART 1

Páginas: 545-556

Tipo: Artículo

DOI: 10.1007/978-3-642-28942-2_49 SCOPUS: 2-s2.0-84858777545 WoS: WOS:000309166900049 GOOGLE SCHOLAR

Otras publicaciones en: Lecture Notes in Computer Science

Resumen

The assessment of failure force in bolted lap joints is a critical parameter in the design of steel structures. This kind of bolted joint shows a highly nonlinear behaviour so traditional analytical models are not very reliable. By contrast, other classical technique like finite element analysis provides a powerful tool to solve nonlinearities but usually with a high computational cost. In this article, we propose a data-driven approach based on multilayer-perceptron network ensemble model for failure force prediction, using a data set generated via finite element simulations of different bolted lap joints. Numeric ensemble methods combine multiple predictors to obtain a single output through average. Moreover, a procedure based on genetic algorithms is used to optimize the ensemble parameters. Results show greater generalization capacity than single prediction model. The resulting ensemble includes the advantages of finite element method whereas reduces the complexity and requires less computation. © 2012 Springer-Verlag.