شبکه عصبی چند جمله ای در برآورد سرعت امواج برشی نهشته های خاکی: کاربردی برای گدازش

ABSTRACT Estimating shear wave velocity of soil deposits using polynomial neural networks: Application to liquefaction

Small strain shear modulus (G_max) and shear wave velocity (V s) are essential parameters for the dynamic analysis of soil systems. Shear wave velocity can be directly used for some geotechnical problems such as evaluating liquefaction potential (e.g., Dobry et al., 1981; Seed et al., 1983; Tokimatsu and Uchida, 1990; Andrus and Stokoe, 2000), soil classification (e.g., Dobry et al., 2000), site response analysis (e.g., Choi and Stewart, 2005), and earthquake-induced ground motions (e.g., Jafarian et al., 2010).

These parameters may be determined via small strain laboratory tests on undisturbed soil samples. Resonant column and bender element tests are the most common devices to estimate small-strain parameters in laboratory; however, cyclic triaxial apparatus with the precise measurement of axial strain has been also used for this purpose. The effect of sample disturbance on small-strain stiffness is significant in laboratory-based measurements since the weak bonds between the soil particles are broken during sampling process. Furthermore, undisturbed sampling in granular deposits is not possible without expensive freezing techniques. Therefore, small-strain shear modulus (G_max) is more rational to be estimated from shear wave velocity (Vs). In-situ shear wave velocity can be measured by a number of geophysical techniques such as cross-hole test (CHT), down-hole test (DHT), seismic cone penetration tests (SCPTs), refraction microtremor (ReMi), multi-station analysis of surface waves (MASW), and spectral analysis of surface waves (SASW). The accuracy of these techniques depends on implementation details, soil conditions, and interpretation methods (Andrus and Stokoe, 1997). Moreover, field measurement techniques involve clear advantages compared with laboratory methods since they examine the soil in its natural condition with minimum disturbance. On the other hand, although in-situ geophysical measurements are the most appropriate methods for this purpose, several limitations such as space constraints, cost considerations, and high noise levels arising from some of these tests, especially in urban areas, make them impractical or restricted in many situations. Hence, empirical equations are essential tools for preliminary assessment of shear wave velocity based on the relevant geotechnical parameters. Researchers have carried out several studies during the last four decades to develop relationships between shear wave velocity and geotechnical parameters of soils. Majority of these studies have tried to correlate ð Þ V s with SPT blow count, SPT-N, and a few studies have employed other parameters such as effective overburden stress, fine content, depth, cone penetration tip resistance (Kanai et al., 1966; Hamilton, 1976; Ohsaki and Iwasaki, 1973; Campbell and Duke, 1976; Ohta and Goto, 1978; Seed and Idriss, 1981; Jinan, 1987; Lee, 1990; Lodge, 1994; Athanasopoulos, 1995; Sisman, 1995; Iyisan, 1996; Jafari et al., 1997; Kiku et al., 2001; Hanumantharao and Ramana, 2008; Hasancebi and Ulusay, 2006; Dikmen, 2009; Brandenberg et al., 2010; Akin et al., 2011).

As discussed above, numerous data were measured in the field condition during the previous studies. These data provide possibility for the development of a predictive correlation for the insitu shear wave velocity of soils. It is clear that a precise correlation is easier to be used in the routine geotechnical projects compared with the mechanics-based measurement techniques. The current study aims to present such correlation using a large database containing soil parameters and the corresponding values of shear wave velocity. Available correlations of shear wave velocity are reviewed first. Performance of the reviewed correlations is subsequently examined through a comprehensive database of measured shear wave velocity, which was gathered herein. Then, a new correlation is developed via an advanced polynomial neural network algorithm. Reasonable precision and applications of the developed correlation are demonstrated in the final part of the paper.