Artificial neural networks in computational earthquake engineering

Abstract eng:
The main scope of the present work is to propose new adaptive soft computing techniques for reducing the computational cost required for estimating the seismic response of real world structural systems in multiple hazard levels. In particular Artificial Neural Networks are incorporated into a methodology which aims to reduce significantly the computational cost of nonlinear dynamic analysis procedure. This reduction of the computational cost could have a serious and positive impact in many fields of structural dynamics. In this work a number of prediction schemes are examined in three hazard levels of increased intensity, namely 50% 10% and 2% in 50 years. Three reinforced concrete one steel test examples and with nine cases in total are used. A.N.N. are incorporated into the proposed procedure for predicting the seismic response of the structure with respect to the time. The predicting results, which refer to the top displacement of the models, are compared with the “real” results that derive from the conventional nonlinear dynamic procedure. The conclusions are very promising since it can be said that A.N.N. work remarkably well. Cases with strong motions, which lead to inelastic behavior, or big structures, in terms of degrees of freedom, can be simulated more difficult with A.N.N. but still the prediction is very close to the results of conventional nonlinear dynamic analysis when the proper scheme is used.

• Export as: BibTeX | MARC | MARCXML | DC | EndNote | NLM | RefWorks
• Add to your basket