Application of functional networks in geotechnical engineering

Abstract

In nature spatial variability of the soil is inevitable. The analysis of such unpredictable material only on the basis of experimental, finite element method and other traditionally available methods is reliable, but overall modeling based on these methods makes it more complex and this problem necessitated the usage of statistical models to develop some empirical and semi empirical methods with the obtained input and output data. Many statistical methods came from the past outperforming one another. Since the efficiency of certain tool also depends on the data chosen, the developed models though showed good results poor generalization was observed for some of the complex problems. Functional networks introduced by Castillo as an alternative to artificial neural network (ANN), in which functions are learned instead of weights, and also the functions are random chosen, unlike ANN they are constrained to certain functions. The selection of topology depends on both domain knowledge i.e. associativity, commutatively and others, where as ANN is a black box which blindly access the data by increasing the weights (trial and error process) The objective of this study is to show how functional network can be effectively used to model certain problems in geotechnical engineering. In this thesis four examples are considered under study (1) Prediction of lateral load capacity of piles in clay, (2) Prediction of factor of safety of slope, (3) Uplift capacity of suction caisson in clay, (4) Swelling pressure in clays, and the results are analyzed based on certain criterion like correlation coefficient, root mean square error, efficiency, cumulative probability distribution function. The observed results are also compared with other statistical methods like ANN, SVM, MGGP, etc and it was observed that FN almost added a rung over all those methods and this shows that this method can be better used in every aspect of geotechnical engineering