Failure Envelopes of Shallow Caisson Foundations and a Predictive Model Using a Swarm-Based Adam Optimizer

This study presents a framework for constructing failure envelopes of shallow caissons in anisotropic and homogeneous clay employing the anisotropic undrained shear strength criteria in combination with three-dimensional upper bound limit analysis using finite elements. The study examines the impact of anisotropy, embedment depth, and vertical load mobilization ratios on the variation of failure envelopes in two-dimensional and three-dimensional spaces. The development of the failure mechanism’s shape and size is discussed for different embedment depths. Another novel hybrid optimizer, adaptive moment estimation (Adam), is suggested in this paper. When training an artificial neural network (ANN) model, optimization through swarming approaches are used to the default Adam optimization technique to fine-tune the model’s architecture and hyperparameters. The proposed hybrid Adam optimizer significantly improves the prediction accuracy, creates simpler network architectures, and provides faster convergence in the training process compared to the default version. The outcomes of this study provide valuable insight into the bearing capacity of caisson foundations in anisotropic clay under general loading conditions, along with a new optimizer to improve the performance of ANN models.