Uncertainty analysis of 3D post-failure behavior in landslide and reinforced slope based on the SPH method and the random field theory

At present, the three-dimensional (3D) landslide post-failure behaviors probabilistic model has been limited due to many technical challenges, especially computational efficiency. This also restrains the exploration of the impact of geometries and geo-conditions in the direction perpendicular to the 2D plane. This study proposes a novel 3D stochastic numerical simulation model combined the high-performance GPU-accelerated SPH method with geotechnical random field theory. Utilizing GPUs for deterministic calculation of landslide large deformation post-behavior achieves computational speeds exceeding those of CPUs by approximately 53 to 100 times. Furtherly, the computational cost as low as 4.3 min per deterministic sample, thus markedly enhancing computational efficiency. Moreover, it considers the number of Karhunen-Loe`ve expansion terms, the fluctuation scale of anisotropy in the direction perpendicular to the 2D plane, and the cross-correlation of the internal friction angle and cohesion to illustrate their influence on probability distribution and variability of landslide behavior indexes. Additionally, slope model with retaining wall is conduct for risk assessment, which suggests that strengthening reinforcement of slope may not only restrains the post-failure behavior of landslide, but also optimizes the probability distribution of its evaluation indexes, reducing the difficulty of prediction. The 3D stochastic simulation framework excels in characterizing complex slope geometries and geo-conditions, providing more accurate risk assessment and mechanism analysis of landslides. This study advances the understanding of 3D landslide large deformation risk analysis, offering practical insights for real-region slope engineering application.