Prediction of Pile Capacity Parameters using Functional Networks and Multivariate Adaptive Regression Splines

Suman, Shakti (2015) Prediction of Pile Capacity Parameters using Functional Networks and Multivariate Adaptive Regression Splines. MTech thesis.

[img]PDF
798Kb

Abstract

The soil is found to vary spatially everywhere in nature. As such, it’s generally a difficult task to predict the nature of soil for any particular application with traditional methods like experimental, empirical, finite element or finite difference analysis. Analysis with traditional methods taking into factor all the varying inputs makes it a complex problem, which is difficult to solve and comprehend. This necessitates the use of statistical modelling tool for the solution to problems concerning soil. Pile foundations are widely used in civil engineering construction. However, owing to the variable behavior of soil and the dependence of vertical pile load capacity on numerous factors, there does not exist a definite equation which can estimate the pile load accurately and include all the factors comprehensively. Artificial intelligence techniques are known to successfully develop accurate prediction models with the obtained input and output data form laboratory experiments or field data. Therefore, in the present study, Functional Network (FN) and multivariate adaptive regression splines (MARS) were used to develop prediction models for the lateral load capacity of piles, vertical capacity of driven piles in cohesionless soil, friction capacity of piles in clay, axial capacity of piles and pullout capacity of ground anchors. In all the cases, prediction equations were provided for the developed models which were found to be simple and can be easily used by practicing geotechnical engineers. A standalone application was also developed to facilitate the calculation of required pile capacity parameters based on the prediction equations. The prediction models built by FN and MARS were compared with different artificial intelligence (AI) techniques and empirical models available in the literature and FN and MARS were found to invariably outperform other AI techniques and empirical methods.

Item Type:Thesis (MTech)
Uncontrolled Keywords:pile capacity, artificial intelligence, functional networks, multivariate adaptive regression splines
Subjects:Engineering and Technology > Civil Engineering > Geotechnical Engineering
Divisions: Engineering and Technology > Department of Civil Engineering
ID Code:6860
Deposited By:Mr. Sanat Kumar Behera
Deposited On:06 Jan 2016 17:53
Last Modified:06 Jan 2016 17:53
Supervisor(s):Das, S K

Repository Staff Only: item control page