Prediction of uplift capacity of Suction caisson in clay using Functional Network and Multivariate Adaptive Regression Spline

Abstract

Suction caissons are extensively used as anchors for offshore foundation structures. The uplift capacity of suction caisson is an important factor with respect to effective design. In this paper, two recently developed AI techniques, i.e. Functional Network (FN) and Multivariate Adaptive Regression Spline (MARS), have been used to predict the uplift capacity of suction caisson in clay. The performances of the developed models are compared with those of other AI techniques: artificial neural network, support vector machine, relevance vector machine, genetic programming, extreme learning machine, and Group Method of Data Handling with Harmony Search (GMDH-HS). The model's inputs include the aspect ratio of the caisson, undrained shear strength of soil at the depth of the caisson tip, relative depth of the lug to which the caisson force is applied, load inclination angle, and load rate parameter. The results of the above AI techniques are comparatively analysed via different statistical performance criteria: correlation coefficient (R), root mean square error, Nash-Sutcliffe coefficient of efficiency, and log-normal distribution of ratio of the predicted load capacity to observed load capacity, with a ranking system to determine the best predictive model. The FN and MARS models are found to be comparably efficient which can outperform other AI techniques.