Computational & Technology Resources
an online resource for computational,
engineering & technology publications
PROCEEDINGS OF THE FOURTH INTERNATIONAL CONFERENCE ON ENGINEERING COMPUTATIONAL TECHNOLOGY
Edited by: B.H.V. Topping and C.A. Mota Soares
Application of General Regression Neural Networks (GRNNs) in Assessing Liquefaction Susceptibility
H.K. Amin, K.M. ElZahaby and A.E. Abdel-Salam
Soil Mechanics and Foundation Engineering Department, Housing and Building Research Center, Giza, Egypt
H.K. Amin, K.M. ElZahaby, A.E. Abdel-Salam, "Application of General Regression Neural Networks (GRNNs) in Assessing Liquefaction Susceptibility", in B.H.V. Topping, C.A. Mota Soares, (Editors), "Proceedings of the Fourth International Conference on Engineering Computational Technology", Civil-Comp Press, Stirlingshire, UK, Paper 142, 2004. doi:10.4203/ccp.80.142
Keywords: artificial neural networks, general regression, seismic, liquefaction, susceptibility, mitigation.
Liquefaction is considered one of the most important hazards associated with earthquakes. The damage resulting from seismic liquefaction may be huge; thus, there is always a need to mitigate the damage associated with such risks. In order to assess the expected damage, and consequently, the possible means to alleviate the risks resulting from seismic liquefaction, it is crucial first to determine and specify the locations of possible seismic liquefaction and their degrees of severity. Since it is impossible to avoid the occurrence of earthquakes, and consequently, liquefaction, special considerations need to be undertaken. Among which, is the improvement of soil susceptible to liquefaction. Thus, after specifying the degree of severity, appropriate means of mitigation to specified site locations may be adopted.
One of the main problems challenging geotechnical engineers is how to assess the seismic liquefaction hazard. Researchers have adopted several approaches to tackle such problems. Statistical and probabilistic approaches for the seismic liquefaction are currently available. The selection of the appropriate type is debatable. This is due to the fact that the hazards associated with seismic liquefaction include more than one source. More specifically, several variables, some representing the site conditions, other depicting the earthquake properties, together with other significant factors, have to be integrated together so as to assess the susceptibility to seismic liquefaction. The more the quality data is available, the better and more precise predictions can be obtained.
Based on the vast amount of data, one of the most famous soft computing techniques, namely, artificial neural networks, is a perfect tool to be used in such analysis. In this paper, an artificial neural networks (ANNs) tool is used to assess the liquefaction hazard, based on geotechnical data gathered from several locations in Egypt. The collected data has been analyzed to assess the seismic liquefaction hazard. the general regression neural networks approach (GRNNs) has been used to generalize the obtained results. In other words, data available from new locations under study, other than the collected ones, can be analyzed using GRNNs to obtain the seismic liquefaction risk associated with this new site. The well-known computer package termed "Neuroshell 2R" has been extensively used to build up the GRNNs models. The obtained results proved to be promising and the quality of the data has been enhanced for possible advancement in the field of seismic liquefaction mitigation. The gain attained using such a method is that the built models can be easily improved via the incorporation of additional data points obtained from new locations. In addition, the models built are quite simple and can be straightforwardly used by ordinary users.
purchase the full-text of this paper (price £20)