Two-stage framework for lateral-torsional buckling resistance prediction of cellular steel beams under fire conditions

Autor(es)

Ben Seghier, Mohamed El Amine ; Elshaboury, Nehal ; Abdelkader, Eslam Mohammed ; Carvalho, Hermes ; de Faria, Caroline Correa ; Miguel, Leandro Fadel

• Tipo de documento: Artigo Científico
• Fonte: STRUCTURES; v. 68, p. 17-pg., 2024-09-10.

Resumo

In this study, the behavior of lateral-torsional buckling of cellular steel beams under fire circumstances is investigated and modeled using a novel two-stage framework. In the first stage, the investigation of the lateral-torsional buckling behavior is carried out based on extensive numerical simulations using finite element analysis under various elevated temperature settings, which generates over 10,771 numerical simulations. During the second stage, the collected information is thoroughly analyzed and processed, then a hybrid machine learning using self-organizing fuzzy neural network (SOFNN) with an adaptive quantum particle swarm optimization algorithm (AQPSO) is used to predict the lateral-torsional buckling taking into account various fire (i.e. high-temperature) scenarios. Furthermore, the proposed AQPSO-SOFNN's performance is compared to two other developed hybrid machine learning models, namely: Hybrid Gustafson-Kessel adaptive recursive fuzzy neural network (GK-ARFNN) and Hybrid hierarchical pruning scheme self-organizing fuzzy neural network (SOFNN-HPS). Several statistical and graphical metrics are implemented to evaluate and compare the performance of the proposed hybrid machine learning models. Overall findings show that the developed AQPSO-SOFNN model can effectively capture the lateral-torsional buckling behavior. Additionally, the results demonstrate that this model outperforms the other hybrid machine learning models, indicating a high level of agreement with the results of the numerical simulation with an R-2 value of 0.991 during the testing phase with a high level of predictive accuracy. (AU)

• Processo FAPESP: 23/09449-2 - Desenvolvimento de modelo probabilístico para análise à fadiga de pontes ferroviárias com utilização de redes bayesianas dinâmicas
• Beneficiário: Caroline Correa de Faria
• Modalidade de apoio: Bolsas no Brasil - Doutorado