Interpretation of a model footing response through an adaptive neural fuzzy inference system

Abstract Modern geomechanics approaches frequently adopt collapse interaction diagrams obtained from load tests carried out on small-scale footing models in order to describe the overall behaviour of footings under the combined action of inclined and/or eccentric loads. Collapse interaction diagrams are used as yielding relationships, in a plasticity theory formulation, for the “macro-element” formed by the footing and the resisting soil. The main theme of this paper is to explore the potential of an expert system, named adaptive network-based fuzzy inference system (ANFIS), to determine such interaction diagrams for predicting foundation behaviour, subjected to vertical centred and eccentric loads, starting from results of small-scale model experiments. The inference system is trained by the results of a series of load tests both in normal and “incremented” gravity conditions, to predict load–settlement curves for a footing of given size with an assigned value of the eccentricity ratio. Cluster analysis is used to support the proposed inference system in order to learn from any simulated load–settlement curve the points corresponding to the collapse mechanism occurrence. Predicted results compare favourably with tests results. The simulation model predicts convincingly the load–settlement curve, the settlement value at which collapse takes place as well as the different types of collapse mechanisms. The results encourage the use of ANFIS in supporting the optimisation of model testing program, since the proposed method allows the identification of the minimum number of tests needed for a good performance of the expert system, particularly when this is to be used in problem formulations of the inverse type: a few careful experimental tests can thus suffice in generating new responses of the investigated geotechnical system, corresponding to conditions not yet explored from an experimental standpoint.