In this paper,we propose a novel probabilistic method for predicting the undrained bearing capacity of spatially variable soils.Our approach combines a Gaussian process regression(GPR)-based surrogate model with rando...In this paper,we propose a novel probabilistic method for predicting the undrained bearing capacity of spatially variable soils.Our approach combines a Gaussian process regression(GPR)-based surrogate model with random cell-based smoothed finite analysis.The Gaussian process emulator(GPE)serves as a statistical tool for making predictions from a data set.First,we validate the accuracy and efficiency of kinematic limit analysis using the cell-based smoothed finite element method(CS-FEM)against the standard finite element method(FEM)and edge-based smoothed FEM(ES-FEM).The numerical results demonstrate that the CS-FEM framework surpasses traditional numerical approaches,establishing its reliability in computing collapse loads.Subsequently,we conduct several hundred simulations to develop a surrogate model for predicting the undrained bearing capacity of shallow foundations.By utilizing various kernel functions,we enhance the accuracy of the GPE in these predictions.This method offers a practical and efficient solution,effectively addressing multiple uncertainties.Numerical results indicate that the GPE significantly boosts computational efficiency,achieving satisfactory outcomes within minutes compared to the days required for conventional simulations.Notably,the mean absolute percentage error(MAPE)decreases from 2.38%to 1.82%for rough foundations when employing Matérn and rational quadratic kernel functions,respectively.Additionally,combining different kernel functions further enhances the accuracy of collapse load predictions.展开更多
文摘In this paper,we propose a novel probabilistic method for predicting the undrained bearing capacity of spatially variable soils.Our approach combines a Gaussian process regression(GPR)-based surrogate model with random cell-based smoothed finite analysis.The Gaussian process emulator(GPE)serves as a statistical tool for making predictions from a data set.First,we validate the accuracy and efficiency of kinematic limit analysis using the cell-based smoothed finite element method(CS-FEM)against the standard finite element method(FEM)and edge-based smoothed FEM(ES-FEM).The numerical results demonstrate that the CS-FEM framework surpasses traditional numerical approaches,establishing its reliability in computing collapse loads.Subsequently,we conduct several hundred simulations to develop a surrogate model for predicting the undrained bearing capacity of shallow foundations.By utilizing various kernel functions,we enhance the accuracy of the GPE in these predictions.This method offers a practical and efficient solution,effectively addressing multiple uncertainties.Numerical results indicate that the GPE significantly boosts computational efficiency,achieving satisfactory outcomes within minutes compared to the days required for conventional simulations.Notably,the mean absolute percentage error(MAPE)decreases from 2.38%to 1.82%for rough foundations when employing Matérn and rational quadratic kernel functions,respectively.Additionally,combining different kernel functions further enhances the accuracy of collapse load predictions.
