Considerations on the Use of Natural and Artificial Time Histories for Seismic Transient Non- Linear Analyses of Structures and Variability Assessment

摘要

Due to increased potentialities of computers, it is nowadays possible to perform dynamic non-linear computation of structures to evaluate their ultimate behavior under seismic loads using refined finite element models. Nevertheless, one key parameter for such complex computations is the input motion, which may lead to important discrepancies in the results and therefore difficulties to deal with for engineering purpose. In this situation, the opportunity to use artificial time histories or not for such analyses is a still open question.The objective of this study is (i) to give some elements concerning the potential effectiveness of artificial time histories to reproduce structural damages led by real earthquakes, (ii) to quantify variability associated with transient non-linear seismic analyses and (iii) to propose a method to account for this variability for design and verification of structures.For this purpose, some simplified non-linear models are used (reinforced concrete structure with structural NL behavior and PWR core with impact NL behavior). The study is divided in two phases.- Phase 1 : Comparison between natural and artificial time histories for non-linear analyses,- Phase 2 : Quantification of variability associated with non-linear calculations.Results of phase 1 show a very good agreement between natural and artificial time histories computations. Therefore, it is possible to conclude that artificial time histories may be representative of real earthquakes if they are generated base on ground response spectrum and other ground motion parameters (ground velocity, ground displacement, CAV, duration ...), which is a key factor according to our experience. In that context, this paper gives some attenuation relationships to evaluate Ground Motion Parameters based on characteristics of the earthquake under consideration (i.e. magnitude and focal distance).In addition, we observed that the variability on the non-linear parameter is significant and may be different for artificial and natural accelerograms. We were able here to generate artificial time histories with an induced damage variability close to the expected one. The key factor is to allow variability around ground motion parameters using an appropriate generation procedure (which may not be the standard engineering practice).In another hand, the variability induced by natural time histories is significantly higher than the expected one. The main reason comes from the fact that it is not possible to get a relatively high number of natural time histories with characteristics (magnitude and distance) close to the target ones. Consequently, the resulting variability is a combination of the variability coming from the seismic motion itself and the variability coming from earthquake characteristics.Results of phase 2 show that for low number of calculations (typically less than 10), the variability on the average value of the non-linear parameter may be very important depending on the different sets of accelerograms used. In addition, the variability decreases significantly for high number of calculations but the convergence to the average value of the population is low, even for numbers of calculations higher than 50.Finally, a method to account for variability is proposed, based on Student-Fisher statistical estimator and allows to estimate the design parameter with an appropriate confidence level. This method is conservative and leaves a relative freedom to the designer for engineering purpose (more calculations can be performed to get an optimized estimation of the design parameter or a lower number of calculations is possible but with a more conservative estimation of the design parameter).