Application of finite element modeling and deep learning for displacement estimation in deep excavations

Abstract

Accurately predicting excavation-induced displacements remains a critical challenge in geotechnical engineering due to the inherent risks and complex soil-structure interactions during construction. To address this problem, this study proposes a hybrid predictive framework that integrates Finite Element Model (FEM) with Deep Neural Networks (DNN), called DNN-FEM. Initially, parametric simulations are conducted using commercial software in geotechnics, PLAXIS 2D, to simulate various deep excavation scenarios and generate a robust numerical dataset. This dataset is subsequently partitioned, allocating 80% of the data to train the DNN architecture and the remaining 20% to evaluate its predictive performance. The results demonstrate a high degree of agreement between the DNN-predicted displacements and the FEM-calculated values across both the training and independent test sets, as evidenced by R^2coefficients exceeding 0.99. Ultimately, this research demonstrates that the proposed DNN-FEM approach provides a highly accurate and computationally efficient tool for estimating Diaphragm Wall (DW) and soil displacements in deep excavation projects