Physics-informed ensemble learning for robust tunnel-induced building settlement with sparse field measurement

Settlement induced by tunnel boring machine (TBM) excavation is a serious threat to building safety, for which traditional data-driven methods often lack physical interpretability. In this study, a physics-informed machine learning (PIML) is proposed to estimate the building settlement induced by TBM excavation. The settlement formula is added to the training process in the form of a physics constraint as a loss function. A simulation model of Wuhan Metro Line 19 in China is built using the finite element method (FEM) to complement the dataset. The model is pre-trained using simulation data together with sparse field measurement and validated using real measurement data. The proposed PIML model can effectively predict building settlements with R2, RMSE, and MAE of 0.9440, 0.00031, and 0.00022, respectively, showing improvements of 3.8 %, 20.7 %, and 10.6 % in R2, RMSE, and MAE, respectively. Combining the fitting of physical laws enhances the rationality and credibility of the model and improves the stability when dealing with small-size datasets. Moreover, the robustness of the PIML model against noises is significantly improved by up to 47 %. The proposed method has proved to be effective in predicting building settlement and improving the interpretability and robustness of the model.