Location

San Diego, California

Session Start Date

3-26-2001

Session End Date

3-31-2001

Abstract

The increasing popularity of the cone penetration test (CPT) for site investigations has led to several methods for predicting liquefaction potential from CPT data. This paper describes a feed-forward neural network model trained by back-propagation for predicting liquefaction potential. The model requires the following seven input variables: cone resistance, total vertical stress, effective vertical stress, earthquake magnitude, maximum horizontal acceleration at ground surface, the mean grain size D50, and the seismic shear-stress ratio. A total of ninety-six data sets from different sites around the world were used for training, and eighty-two data sets were used for testing and validating the neural network model. The model gave an overall success rate of 96% for correctly predicting the liquefaction potential.

Department(s)

Civil, Architectural and Environmental Engineering

Appears In

International Conferences on Recent Advances in Geotechnical Earthquake Engineering and Soil Dynamics

Meeting Name

Fourth Conference

Publisher

University of Missouri--Rolla

Publication Date

3-26-2001

Document Version

Final Version

Rights

© 2001 University of Missouri--Rolla, All rights reserved.

Document Type

Article - Conference proceedings

File Type

text

Language

English

Share

COinS
 
Mar 26th, 12:00 AM Mar 31st, 12:00 AM

CPT Evaluation of Liquefaction Potential Using Neural Networks

San Diego, California

The increasing popularity of the cone penetration test (CPT) for site investigations has led to several methods for predicting liquefaction potential from CPT data. This paper describes a feed-forward neural network model trained by back-propagation for predicting liquefaction potential. The model requires the following seven input variables: cone resistance, total vertical stress, effective vertical stress, earthquake magnitude, maximum horizontal acceleration at ground surface, the mean grain size D50, and the seismic shear-stress ratio. A total of ninety-six data sets from different sites around the world were used for training, and eighty-two data sets were used for testing and validating the neural network model. The model gave an overall success rate of 96% for correctly predicting the liquefaction potential.