The application of metamodels based on soft computing to reproduce the behaviour of bolted lap joints in steel structures

Resumen

A promising field of research in steel structures regarding their preliminary design and optimization is the replacement of expensive computational finite element models with more efficient techniques. Without a significant loss of accuracy, new proposals should be able to consider not only the ideal load-displacement response but also relevant failure mechanisms and imprecisions in material properties. The article proposes the use of metamodels based on soft computing as an overall approximation system for structures analysis. This approach has been applied in several fields but, till nowadays, its implementation on structural analysis in early esign seems quite limited to a few theoretical cases. Taking advantage of artificial neural network as global approximation technique, the parameters for more realistic and informative loaddisplacement curve including nonlinear effects (damage mechanics) are estimated for bolted steel lap joints. Our results demonstrate the accuracy of the metamodel implemented can be close to simulations and also real experimental tests. © Springer International Publishing Switzerland 2014.