Predicting of liquefaction potential in soils using artificial neural networks

The phenomenon of liquefaction is usually caused by dynamic factors where there is a mass of saturated soil sand. In order to prevent probablE destruction of structures in such areas, prediction of liquefaction potential seems to be necessary. For the purpose of data collection, we need to sound boreholes and carry out many experiments, each of which requires expenditure of time and money. Therefore, prediction of liquefaction by using the existing data in sounding leads us to decreasing costs and efficiency pro gramming and choosing sufficient area for our construction. Neural networks are intelligent systems that use specific processing characteristics of the brain such as: learning examples, ignoring errors in the data, and their parallel processing -these are not possible through current programming methods. The present study attempts to predict the potential of liquefaction through neural network approach by using data from sounding in the southeast part of Tehran. It is an area with 30 km2 and a high level of underground water. The neural network in question, having one hidden layer, is trained and tested by some new data, based on standard penetration test, in order to ensure the efficiency operation of the network. After all, the result of neural network method can be compared with the result of Seed method for predicting liquefaction and was shown that the neural network method could predict with 92 percent accuracy in the southeast area of Tehran.