Seismic Performance of Gravity Retaining Walls Under Quasi-static Approach Using Probabilistic Analysis: Fid-Bau Portal

Seismic Performance of Gravity Retaining Walls Under Quasi-static Approach Using Probabilistic Analysis

Mustafa, Rashid / Samui, Pijush / Kumari, Sunita

This paper presents a study on the probabilistic-based design of a gravity retaining wall under seismic condition. Mononobe-Okabe equations are used to compute dynamic earth pressures on the wall. The conventional methods are used to find factor of safety (FOS) against sliding, overturning, and bearing failure. Three machine learning (ML) models, viz. group method of data handling (GMDH), Gaussian process regression (GPR), and minimax probability machine regression (MPMR), are used to predict FOS against all the three failure modes under seismic condition. To form homogeneity and distribution of datasets, Anderson–Darling and Mann–Whitney U tests are carried out, respectively. The three machine learning models are applied to 100 datasets by considering different numbers of influential input parameters for predicting FOS against sliding, overturning, and bearing failure, respectively. The performance of ML models is measured by numerous statistical parameters. The obtained results from computational approach indicated that MPMR attained the best predictive performance against sliding and overturning failure but GMDH gave best predictive performance against bearing failure. The results of the model were also analyzed by using performance curve and rank analysis. Sensitivity analysis was also done to reveal the influence of input variables.