Geotechnical evaluation of embankment stability in seismic zones using Monte Carlo and subset simulations within an LRFD framework aided by machine learning
Furquan Ahmad, Pramod Kumar, Divesh Ranjan Kumar, Warit Wipulanusat, Pijush Samui
Abstract
• Integrates Monte Carlo and Subset Simulations within the LRFD framework. • Evaluates embankment stability for 11.693 m height and 2H:1 V slope under seismic loads. • Assesses failure probability under varying pore water pressure and COV conditions. • RF-YYPO model achieves R² = 0.9633 (train) and 0.9072 (test) with lowest RMSEs. • Findings enhance geotechnical risk assessment for seismic-prone embankments. This study highlights the importance of thorough risk evaluations to ensure resilient railway and roadway embankments . By integrating Monte Carlo simulation (MCS) and subset simulation (SS) methodologies within the LRFD framework, this study examines the stability and risk assessments of a typical soil slope, specifically concentrating on an embankment with a height of 11.693 m and a soil slope angle of 2H:1 V. This study utilizes the "UPSS 3.0 Add-in" for MS Excel to conduct the analysis. This evaluation includes seismic scenarios with values of 0.12 and 0.14 for Zones IV and III, respectively, in accordance with the specifications of the Indian zoning map and RDSO guidelines. The study also examines variations in pore water pressure ratios (0.0, 0.05, 0.10) and coefficients of variation (COVs), noting their significant impact on the probability of failure ( p f ) of the embankment. The results indicate a significant increase in p f with increasing COVs, r u , and k h values, with p f increasing to 40% under certain conditions. Advanced computational modeling using three random forest-based models: RF-YYPO, RF-BWOA, and RF-SMA. The RF-YYPO model outperformed the other models, showing high efficacy, with R 2 values of 0.9633 during training and 0.9072 during testing, and achieving the lowest RMSEs. These findings enhance embankment safety in seismic zones and advance geotechnical engineering practice. This study uniquely integrates probabilistic techniques with hybrid machine learning optimization algorithms (RF-YYPO, RF-BWOA, and RF-SMA) under an LRFD framework, which has not been previously implemented for seismic slope stability of railway embankments.