Predictive models for bolted T-stub connections combining FEM with Intelligent Artificial techniques

  1. Fernández-Ceniceros, J. 1
  2. Lostado-Lorza, R. 1
  3. Fernández-Martínez, R. 1
  4. Sanz-García, A. 1
  5. Martínez-De-Pisión, J. 1
  1. 1 Universidad de La Rioja
    info

    Universidad de La Rioja

    Logroño, España

    ROR https://ror.org/0553yr311

Libro:
Computational Plasticity XI - Fundamentals and Applications, COMPLAS XI

ISBN: 978-848992573-1

Ano de publicación: 2011

Páxinas: 1467-1478

Tipo: Capítulo de libro

beta Ver similares en nube de resultados

Resumo

The behavior of bolted connections is inherently nonlinear because of geometric discontinuities, stress concentrations, contacts or local yielding. In structural field, component method adopted by Eurocode 3 separates the joint in individual elements that are called T- stub. Each component should be defined at least by three parameters: initial stiffness, strength and deformation capacity. In this paper, a new methodology based on a combination of Finite Element Method (FEM) and Artificial Intelligence (AI) techniques is presented in order to predict force-deformation response of bolted T-stub connection. An advanced finite element model is generated and validated by comparison with experimental tests in the literature. Parametric study combining bolt diameters, hot-rolled profile geometry and steel class is carried out for generating training database with output variables of T-stub characterization. Several AI algorithms as regression trees, neural artificial networks or radial basis function networks are training so as to find the best generalizing model of the problem. The results show a high correlation between FEM and the predictive model, which replaces the first one. Test errors for output variables of the T-stub model prediction are lower than 5%. Finally, this methodology provides an alternative to analytical models which includes the Eurocode 3 for the determination of T-stub parameters. This alternative includes the advantages of FEM (realistic simulation validated against experimental tests or the ability to obtain stress and strain values) but minimizes the complexity and computing time by using AI techniques.