Developing new Adaptive Neuro-Fuzzy Inference System models to predict granular soil groutability

Three Neuro-Fuzzy Inference Systems (ANFIS) including Grid Partitioning (GP), Subtractive Clustering (SCM) and Fuzzy C-means clustering Methods (FCM) have been used to predict the groutability of granular soil samples with cement-based grouts. Laboratory data from related available in litterature was used for the tests. Several parameters were taken into account in the proposed models: water:cement ratio of the grout, relative density of the soil, grouting pressure, soil and grout particle size dimenstions namely D15 soil , D10 soil, d85 grout and d95 grout and percentage of the soil to pass through a 0.6 mm sieve. A accuracy of the ANFIS models was examined by comparing these models with the results of the experimental grout-ability tests. Sensitivity analysis showed that ratios of D15 soil / d85 grout and D10 soil / d95 grout were the most effective parameters on groutability of granular soil samples with cement-based grouts and the grouet water:cement ratio of the grout was determined as the least effective parameter.

Keywords:

Groutability , ANFIS , Clustering Algorithm , Granular soil