Liquefaction Potential Evaluation of Alluvial Soil by Neuro-Fuzzy Technique

This paper presents prediction of liquefaction potential of soils by neuro-fuzzy models evaluated using Idriss and Boulanger method. In order to address the collective knowledge built up in conventional liquefaction method, an alternative Takagi-Sugeno-Kang reliant neuro-fuzzy model has been developed. Neuro-fuzzy is one of the artificial intelligence approaches that can be classified by machine learning as it is a robust and flexible method and may easily be adopted. Idriss and Boulanger method used for evaluation of liquefaction potential of soils for its better estimation capability compared to other conventional methods. To estimate the liquefaction potential bore log data were obtained from SPT tests conducted at sites. Hundred ten datasets from fifty boreholes up to a depth of ten meters were collected for training neuro-fuzzy models whereas twenty six datasets were reserved for validating the models. The predicted results of neuro-fuzzy models compared with Idriss and Boulanger method advocate that trained neuro-fuzzy models are capable of predicting liquefaction potential adequately.