Numerical-Informational methodology for characterising steel bolted components coupling finite element simulations and soft computing techniques
Over the last few decades, the characterisation of steel joints has been a highly active research topic thanks to its inherent complexity and utmost importance in the behaviour of a whole structure. The emergence of the semi-rigid concept provided significant benefits from both the structural and ec...
Guardado en:
Autor principal: | |
---|---|
Otros Autores: | |
Formato: | text (thesis) |
Lenguaje: | eng |
Publicado: |
Universidad de La Rioja (España)
2015
|
Acceso en línea: | https://dialnet.unirioja.es/servlet/oaites?codigo=45992 |
Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
Sumario: | Over the last few decades, the characterisation of steel joints has been a highly active research topic thanks to its inherent complexity and utmost importance in the behaviour of a whole structure. The emergence of the semi-rigid concept provided significant benefits from both the structural and economic perspectives, in exchange for more advanced and sophisticated calculation procedures. An approach that has gained popularity is the component-based method, in which the overall behaviour of the joint can be determined from the force-displacement responses of its individual components. Although this method is very versatile for modelling any joint configuration, a detailed characterisation of components is necessary to ensure accuracy.
In this context, this thesis presents a hybrid methodology to determine the comprehensive force-displacement curve of bolted components: from initial stiffness up to the fracture point. This methodology couples numerical and informational models to predict key parameters of curves, such as initial stiffness, maximum resistance and displacement at failure. To this end, numerical models based on the finite element method (FEM) are first developed to reproduce the real response of bolted components. These models incorporate progressive damage mechanisms and failure criteria to accurately estimate the displacement at fracture. In order to minimise the computational burden of the FEM, the results of a set of simulations are then utilised to train informational models based on soft computing (SC). A genetic algorithm (GA) optimisation is included to set up model parameters and select the most relevant input variables for predicting the force-displacement response. Taken together, the proposed methodology is capable of providing accurate and parsimonious informational models.
The applicability of the hybrid methodology is demonstrated for the characterisation of two fundamental bolted components: the lap and the T-stub. The results obtained highlight the superior accuracy of this methodology as compared to current regulatory codes and traditional analytical models. Once trained and validated, the informational models are able to replace costly FE simulations without a significant decrease in accuracy, and at a negligible computational cost. Therefore, the hybrid methodology could represent an effective tool to be implemented in structural analysis software for designers and practitioners.
Overall, the contributions presented in this thesis provide evidence of the great potential of combining FEM and SC to predict the behaviour of structural components. |
---|