An Improved Recursive Feed Forward Neural Network Based Sand Constitutive Modelling

Data-driven predictive modeling offers a powerful alternative to traditional physics-based models for understanding the complex constitutive behavior of sands. This work introduces a novel deep learning framework based on recursive feedforward neural networks (R-FFNNs) to capture the inherent temporal dependencies and nonlinear relationships within experimental and simulated data. The proposed R-FFNN model is rigorously evaluated against other state-of-the-art deep learning architectures, including feedforward neural networks (FFNNs), long short-term memory (LSTM) networks, bidirectional LSTMs (Bi-LSTMs), and gated recurrent units (GRUs). Through comprehensive training, validation, and testing, it is shown that the R-FFNN model demonstrates superior performance in predicting the evolution of stress and strain states in sand.

Keywords: Soil constitutive modeling, Recursive feedforward neural network, Deep learning, Granular soils.