Keywords: seismic design spectra, competitive neural network, ground motion, classification, soil type.

Accelerograms are considered to contain the most direct engineering information of the earthquake and are capable of measuring strong motion excitations of major earthquakes. Peak ground velocities are recognized to be related strongly to structural damages. The ultimate displacement demand of a structure has also been recognized as a key parameter governing the seismic behavior of structures. Time history analyses use these records for their purposes. However, the most practical means to evaluate earthquake effects on a structure are spectra.

Earthquake response spectra are often used in the analysis and design of structures. Spectrum analyses use acceleration, velocity and displacement spectra that are obtained using earthquake records. The combined deformation-velocity-acceleration spectra which, benefit physical meaning of all three quantities, may also be used in obtaining the earthquake response of a structure. Seismic codes purpose design spectra for engineering design. The design spectra are based on statistical analysis of the response spectra for an ensemble of ground motions. The design spectrum is obtained for different sites or soil characteristics, near field or far field ground motions and moderate or high intensity earthquakes. Thus, the primary step to obtain a design spectrum is to classify existing ground motion records according to their characteristics. This job has been performed here using the application of artificial neural networks. Different researchers have applied the artificial intelligence phenomena in order to construct artificial accelerographs, estimating the fault motions and process earthquake data. But in the knowledge of the authors it is the first time that the neural networks are being used for generating design spectra.

In this paper, the application of Artificial Neural Networks (ANN) in classification has been used to obtain appropriate design spectra for different site conditions. More than 2000 earthquake records of Iran have been classified based on their spectral characteristics to construct design spectra. These ground motion records are primarily selected and have been classified through a multi steps procedure to obtain the final categories for the calculation of the design spectra. The results obtained from classification method have been verified comparing to the available experimental and theoretical methods. The classification has been performed using linear acceleration, velocity and displacement response spectra of the records. The spectra have been prepared for a long period range in order to take into account the higher mode effects. The competitive learning neural networks have been employed for this problem, which benefits from the advantages of the unsupervised learning algorithm. The results of the classifications have been used to prepare the design spectra that have been finally compared to the existing code spectra and the sources of differences have been investigated.

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £82 +P&P)