TESTS TO DETECT NONLINEAR SITE AMPLIFICATION

摘要

When dealing with nonlinear site amplification, an important problem is the detection and identification of nonlinear soil behavior from records. There are two methods commonly used, forward modeling and spectral ratios. Forward modeling involves fitting nonlinear soil models to data, whereas spectral ratio approach involves calculating soil-to-rock ratios of Fourier amplitude spectra from multiple sets of records. Both methods assume that the behavior of rock site is linear. This paper presents four alternative methods to detect soil nonlinearity. The methods require only one record, and are based on the detection of nonlinearity in serially dependent time series. Three of the methods are based on the premise that once the linear serial dependence is removed from the data, any remaining serial dependence is due to a nonlinear generating mechanism. The procedure involves fitting an auto-regressive model to the series, and testing the residuals of the fit for serial independence by using various techniques. The fourth method uses the bispectrum of the series to test nonlinearity. When compared to the existing methods, the methods presented here are simpler, require only one record, and give a quantitative measure of nonlinearity in site amplification.