Application of ANN for prediction of settlement of ring foundation

In recent years, artificial neural networks (ANNs) have gained significant popularity in a variety of engineering fields, with major breakthroughs in addressing geotechnical difficulties such as structure settlement, soil compaction, slope stability, and pile settlement. The goal of this research is to use MATLAB to create multiple ANN models based on different feedforward techniques, including Levenberg-Marquardt (LM), Bayesian regularization (BR), and Scaled Conjugate Gradient (SCG). In addition, the Adam Optimizer will be employed using Python and to compare it with backward progress ANN algorithm (BP-ANN) and prediction of the best suited ANN model for the present data set of settlement of ring footing. The input parameters for the development of present prediction model have been taken as inner diameter of footing, outer diameter of footing, load pressure, thickness, ring radii ratio (n). Experimental validation using FEM (ANSYS model) on a dataset of 200 ring footings demonstrates the correctness of the BP-ANN model, which has an R2 value of 0.98 when compared to alternative feed forward techniques. The results show that ANN model utilizing back propagation ANN algorithm (BP-ANN) predicted settlement with exceptional accuracy, generating R2 of 0.98 compared to other feed forward techniques. Notably, this study goes beyond typical assessments by performing SOBOL and SHAP sensitivity analyses, which give light on the impact of the input parameter on settlement prediction. Furthermore, parametric research improve our understanding of complicated systems, allowing for more informed decision-making. The entire methodology, which incorporates various ANN methodologies, validation methods, and sensitivity assessments, is innovative and original, increasing the accuracy and usefulness of settlement predictions in geotechnical engineering.