Predicting Micropile Group Capacity in Soft Cohesive Soil by Artificial Neural Network

Small diameter grouted micropiles have the capability to resolve certain kind of foundation related problems such as building of low to medium height, lightweight houses or other establishments on thick soft cohesive soil layers. In the present study, advantages of ANN are utilized for evaluation of micropile group capacity for particular value of settlement in deposit of cohesive soil with soft consistency. Experimental investigation was completed on prototypical micropile groups casted inside a soft cohesive soil deposit prepared in a test pit dug in the ground. Length, diameter, spacing between micropiles and number were different variables associated in the study. An extensive dataset was compiled with the help of load–settlement diagrams plotted for 32 numbers of micropile load test and that dataset were employed to construct different artificial neural network models. Four different algorithms with four different validation models for each were used and the best performed model was compared with multivariable regression model. Bayesian Regularization algorithm of artificial neural network with 90%-10% validation model was observed to be most effective one. To estimate the importance of various input variable parameters upon micropile group capacity, sensitivity analysis was conducted. A graphical representation known as neural interpretation diagram showing the inter-relation between input variables–hidden neurons output variable was plotted. The best fitted artificial neural network model was utilized to formulate an empirical equation to forecast the micropile group capacity. Calculation procedure for predicting the micropile group capacity on soft cohesive soil was also demonstrated with an example.