Analysis of anchor uplift capacity in spatially varying soil using MARS model

Soil being a naturally formed material, the prediction and analysis of anchor uplift capacity incorporating the inherent variabilities in the soil properties is essential. This paper presents a probabilistic analysis of horizontal strip anchors in spatially varying soil using the multivariate adaptive regression splines (MARS) as a prediction model. The use of MARS in spatially varying soils is proposed, and its advantage over the conventional expensive Monte Carlo simulations is reported in this paper. The soil cohesion and friction angle have been modeled as lognormal random fields, and the influence of the spatially distributed soil properties on the anchor uplift capacity is assessed in the current study. Furthermore, a comparison between the deterministic and reliability-based design methodologies is presented, and their implications on the design of foundation are discussed. The results indicate that the code-specified factor of safety (FOS) value of 1.5 does not fulfill the target probability of failure ( \({P}_{{\text{f}}}\) $P_{\text{f}}$ ) values recommended in probabilistic design. A minimum FOS = 2 is required to satisfy the \({P}_{{\text{f}}}\le 0.001\) $P_{\text{f}}\le 0.001$ criteria, which is usually recommended for the design of foundations.