Nonstationary modeling and simulation of earthquake acceleratio

The objective of this research is to develop a methodology for analyzing a single component earthquake acceleration record and to develop a method for synthesizing time histories which have characteristics similar to the original record. The acceleration record is characterized as a zero mean, non-stationary covariance process. The data is modeled as the output of a time-varying parametric time series model excited by white noise. The types of models investigated are linear in both the data and white noise terms. The parameters of the model are changing with time and procedures for identifying the parameters at each time index are discussed. Both on-line and off-line methods are described for identifying the auto-regressive, $AR$, the auto-regressive moving-average, $ARMA$, and the constrained ARMA models.;The time-dependent auto-covariance and power spectrum of the earthquake record are defined in terms of the changing parameters of the model. In addition, the estimated model is used in generating synthetic acceleration time histories with similar time-varying characteristics. A set of criteria is developed to evaluate the synthetic time histories.;Several earthquake records are analyzed and an ensemble of artificial acceleration time histories are generated. The synthetic time histories are checked using the simulation criteria. This type of test and the model validity tests are used in evaluating the capability of time-dependent parametric time domain models to characterize the time-varying nature of the acceleration record. Based on the tests carried out on the synthetics the constrained ARMA model is seen to generate synthetics which best preserve the characteristics of the original record. Finally, recommendations regarding the use of time-varying models for modeling and synthesizing acceleration records are made.