NEURAL-NETS FOR EVALUATION OF LIQUEFACTION POTENTIAL

Abstract

This study examines the feasibility of using neural networks for accessing the liquefaction potential of a probable site. The neural network was developed and tested using data from sites which have or have not liquefied during past earthquakes. The trained network was then used to analyse the Solani aqueduct site near Roorkee and the results compared with those obtained from the conventional methods, namely (i) Seed and Idriss (1971), (ii) Chinese Building Code (1974), (iii) Seed et al (1983), and (iv) Japanese Highway Bridge Code (1986). The results obtained from the neural network were found to be in very good agreement with those obtained from the conventional methods. A parametric study was carried out to study the influence of the principal factors affecting the liquefaction potential of a probable site. In this study the parameters considered were, the earthquake magnitude M, the effective normal stress o, the normalized SPT value N and the mean grain size D5 of the soil. The relative interdependence of these factors have also been evaluated. The parametric study indicates that at a larger effective normal stress and SPT N-value, the soil with more fines content does not liquefy at a smaller value of earthquake magnitude; while at a lesser effective normal stress and SPT N-value, the soil with a larger mean grain size liquefies at a smaller value of earthquake magnitude.