Setting Design Ground Motion Reflecting Information of a Set of Observed and Simulated Ground Motions Based on Data-Driven Approach


Abstract eng:
Number of ground motions available for the design purposes is increasing, because strong motion observation stations are increasing and strong motion simulation techniques have been improved. In order to utilize information possessed by a suite of those ground motions, we proposed a concept to use a set of design ground motions for seismic design. It is impossible, however, to utilize huge number of such ground motion data for the evaluation of seismic performance. Selection or synthesis of a design input motion among the available ground motion records is required. Conventionally, design ground motions are selected based on the response spectra or intensity measures (IMs) such as PGA. But they cannot fully consider complicated nonlinear behavior of real structures under strong motions. This paper proposes to utilize a data-driven approach, a general scheme for information extraction from big amount of data as a tool to generate the design input motions which represent ground motions included in a design set of ground motions. First, for the purpose of evaluation of ground motion characteristics in terms of their influence on the structures, we set up indices that are associated with expected damage mechanism of nonlinear structures. When a ground motion set consists of wide variety of ground motion records, it is not efficient to integrate all the information into a single input motion. Therefore, we first divide the original set of ground motions into several subgroups according to their characteristics. Response values of simplified structural models can be used as indices. For the purpose of robust evaluation, fluctuations are given to the models, and response values are obtained as stochastic variables. It allows us to evaluate the characteristics of ground motions including their sensitivity to fluctuation of structural parameters. Based on the evaluated characteristics, we perform cluster analysis of ground motions set. Jensen-Shannon divergence is used as the distance between stochastic variables, considering its advantage over the conventional Kullback-Leibler divergence. In the stage of representative ground motion synthesis, information of ground motions are extracted and representative waves are synthesized in one subgroup. Applying wavelet transform, we construct a scheme to extract information from ground motions. Important characteristics relevant to target structure’s response, which is difficult to be expressed by forms of indices, are extracted as wavelet functions. Extracted information is superposed to one artificial wave, and its performance as a representative of one subgroup is evaluated by numerical experiments.

Contributors:
Conference Title:
Conference Title:
16th World Conference on Earthquake Engineering
Conference Venue:
Santiago (CL)
Conference Dates:
2017-01-09 / 2017-01-13
Rights:
Text je chráněný podle autorského zákona č. 121/2000 Sb.



Record appears in:



 Record created 2017-01-18, last modified 2017-01-18


Original version of the author's contribution as presented on USB, paper 1748.:
Download fulltext
PDF

Rate this document:

Rate this document:
1
2
3
 
(Not yet reviewed)