The use of neural networks in geotechnical engineering

Liquefaction is an external manifestation of decrease in shear strength, due to the cyclic pore-pressure generation mechanism. During a seismic event agitation of saturated fine grained cohesionless soils can result transient, but total, loss of intergranular stress due to a sudden rise in the water filling the intergranular voids of the soil mass (ASCE, Kartam, Flood, 1997).In this study, Seed?s method proposed by Seed and Idriss used to estimate liquefaction potential. This SPT-N based method using factor of safety for separate liquefaction and non-liquefaction situations, compute the cyclic stress ratio (CSR) due to the cyclic resistance ratio (CRR) of soils. The effects of fines content (FC), earthquake magnitude (M), Horizontal peak ground acceleration, (amax), saturated unit weight (?d) and natural unit weight (?n)of soil with ground water level incorporated in the CRR estimation. For the stress reduction factor ( rd ) the relationship suggested by Seed and Idriss used in different depth limits ( 9.15m, 9.15 m < z 23m ).The concept of factor of safety (FS) against damage was commonly employed in the design in civil engineering. Then the factor of safety (FS) can be calculated as FS = CRR/CSR. Liquefaction is said to occur if FS?1, and no liquefaction occurs if FS>1.The limitations of various numerical modeling techniques, for highly non-linear behavior of soils is also considered to be complex and time-consuming for geotechnical approaches. ANN is a powerful data modeling tool that is able to capture and represent complex input/output relationships. ANNs resemble the human brain in two aspects; firstly an ANN acquires knowledge through learning and secondly an ANN?s knowledge is stored within inter-neuron connection strengths known as synaptic weights (Baykasoğlu at al., 2009). İn this study, ANN were used to predict the liquefaction factor of safety. a feed forward with Levenberg?Marquardt algorithm, back propagation neural network model is developed to predict the liquefaction potential due to factor of safety calculated by cyclic stress ratio (CSR). The sigmoid function is used as the transfer function. Moreover, several performance indices such as determination of coefficient (R2), variance account for (VAF), mean absolute error (MAE), and root mean square error (RMSE) were calculated to check the prediction capacity of the ANN model. The obtained indices make it clear that the developed ANN model has shown high prediction performance. Sensitivity analyses were also carried out to examine the relative importance of the factors affecting liquefaction potential prediction.Key Words: Artificial neural networks, Liquefaction potential, Standard penetration test.