ABSTRACT
Small deviations from the Gaussian distribution can result in significantly different acceleration of the rate of accumulation of fatigue damage in the system.
The simulation of Gaussian processes has been explored for several decades, while non-Gaussian simulation has not been as widely addressed. Shinozuka and Jan (1972) proposed a multivariatemultidimensional random process with a known spectral density. Many studies have been undertaken to generate non-Gaussian time series using AutoRegressive Moving Average (ARMA) models but replacing Gaussian with non-Gaussian white noise residuals (Lawrence and Lewis 1985; Stathopoulos and Mohammadian 1991). Mengali and Micheli (1994) used a bank of linear filters driven by nonGaussian white noise inputs to generate non-Gaussian process. A summary of several classes of nonGaussian processes and their simulation details have been described by Grigoriu in 1995, including filtered Poisson processes, ARMA models driven by nonGaussian noise, and alpha stable processes. Another widely recommended method of simulating nonGaussian time series is to generate Gaussian time series using either ARMA or FFT model followed by a
nonlinear static transformation from Gaussian to nonGaussian (Yamazaki and Shinozuka 1988; Janacek and Swift 1990; Iyengar and Jaiswal 1993). Seong and Peterka (1993) proposed a promising approach which simulates wind pressure fluctuations of nonGaussian nature with the help of FFT and AR models, and so did Kumar and Stathopoulos (1999). Hermite moment models have been also used to represent nonGaussian processes resulting from the transformation of a standardized Gaussian process (Winterstein 1988; Winterstein and Lange 1995). Spectral Correction method based on modified Hermite polynomial transformation has been presented by Gurley (1996). A. Steinwolf, Neil S. Ferguson and Robert G. White (2000) generated a non-Gaussian excitation in the form of a Fourier expansion with a special procedure of harmonic phase adjustment. George Deodatis and Raymond C. Micaletti (2001) proposed a modified version of the Yamazaki and Shinozuka iterative algorithm to simulate highly skewed nonGaussian stochastic processes. In a recent paper by Jiunn-Jong Wu (2004), Johnson translator system is used to obtain the phase part and FFT is used to simulate non-Gaussian surfaces.
