@

초록

Accurate estimation of ultimate bearing capacity of reinforced soil foundations is essential for the safety of geotechnical structures, but remains challenging due to the difficulty in quantifying the contribution of geosynthetic reinforcement. Accordingly, a physics-informed deep learning model was developed, incorporating a modified theoretical solution that accounts for both the unreinforced soil contribution and the geosynthetic reinforcement effect. A multi-branch architecture was adopted. Within this framework, the geosynthetic reinforcement effect was quantitatively represented by a mobilization factor for reinforcement tensile force (λ), whose value was inferred via a sub-branch of the developed model based on eight selected influencing factors. The mobilization factor λ was then integrated by the main branch with ten influencing factors characterizing soil properties, foundation geometry, and the configuration and mechanical properties of the reinforcement to estimate the total bearing capacity. Model hyperparameters optimization, using 90% of the literature-derived dataset, identified five optimal configurations with different configuration sizes, all using the ReLU (Rectified Linear Unit) activation function and a learning rate of 0.001. Evaluation on the remaining 10% dataset showed strong agreement between estimated and actual values, with low errors of MAE (Mean Absolute Error), RMSE (Root Mean Square Error), and high R2 (coefficient of determination).

키워드