LIQUEFACTION SUSCEPTIBILITY BASED ON AN ARTIFICIAL NEURAL NETWORK

Abstract

The traces of liquefaction were recognized in the area of Zagreb in the Sava valley in previous earthquakes and liquefaction can be expected in future earthquakes as well similar to the many cases which occurred in the Petrinja earthquake. Therefore, it is useful to have a tool allowing quick identification of susceptibility to liquefaction in larger areas. CPTU testing covers many aspects of soil behaviour and enables the estimation of parameters needed in liquefaction susceptibility analysis. During the 2010-2011 series of earthquakes in Christchurch and Canterbury, New Zealand, a very rich dataset was collected that links soil data obtained by the CPTU, earthquake data, and on-site liquefaction manifestations – or lack of it. An artificial neural network was developed from these data. In addition to the description of location and time, the data contains CPTU measurements, earthquake magnitude, medial peak ground acceleration, its standard deviation, groundwater depth and classification of the manifestation of liquefaction on the ground surface. The data collected after the Petrinja earthquake – obtained from CPTU tests and from analysis of the manifestations of liquefaction and the available data on the earthquake – are used in the developed artificial neural network.