Deterministic and probabilistic prediction of maximum wall facing displacement of geosynthetic-reinforced soil segmental walls using multivariate adaptive regression splines

Abstract

A surrogate model-based response surface method is proposed to predict the maximum wall facing displacement of geosynthetic-reinforced soil (GRS) segmental walls using the multivariate adaptive regression splines (MARS). The synthetic datasets are generated using the Latin hypercube sampling (LHS) technique for both training and testing purposes. For each input dataset, the maximum wall facing displacement is evaluated using the finite difference software FLAC. An expression for the maximum wall facing displacement is proposed in terms of basis functions (piecewise linear). The performance of the proposed model is assessed in terms of the ratio of predicted and simulated maximum wall facing displacement, and the results show that the performance is satisfactory. Moreover, the study identifies that the effect of soil friction angle is most apparent in predicting the maximum wall facing displacement. The efficiency of the proposed model is compared with other soft computing techniques, and the result indicates that the proposed model gives minimum error. Further, a probabilistic study is performed in terms of the normalized maximum wall facing displacement. The results show that the probabilistic prediction of the maximum wall facing displacement using the proposed model is within the range of specified limit.