The use of machine learning models for predicting maximum bridge pile lateral displacements caused by excavation of adjacent foundation pit

Foundation pit excavation triggers soil displacement and stress redistribution, posing a risk of lateral displacement to adjacent infrastructure. Traditional methods for predicting the risk of lateral displacement of adjacent pile foundations are inefficient and inadequate to meet the urgent tasks. This paper developed a numerical model using FLAC 3D, based on real-world excavation of foundation pits and adjacent bridge engineering. It investigated the impact of key parameters on the lateral displacement of the pile foundation of an existing bridge. Various machine learning (ML) algorithms were employed to create predictive models for the lateral displacement of existing pile foundations. The optimal ML algorithm were validated for its rationality through interpretable algorithms. Results indicate that the FLAC3D model based on the actual soil conditions at the site accurately reflects the deformation behavior by foundation pit excavation. The developed GPR model achieves over 10,000 times higher prediction efficiency than traditional simulations for assessing foundation pit excavation effects on adjacent bridge pile displacement. Analysis confirms the necessity of maintaining a foundation pit-to-bridge pile distance exceeding the excavation depth. Increasing the stiffness of the retaining structure and internal supports will improve the construction displacement issue, while changing the length and stiffness of the pile foundation has limited effect on suppressing deformation.