Machine Learning-Assisted Reliability and Risk Assessment of Rock Slopes under Geotechnical Uncertainty

Rock-slope stability is strongly controlled by discontinuities and uncertain geotechnical parameters, while conventional deterministic factor-of-safety (FOS) calculations do not explicitly quantify the probability of failure. This study develops an integrated machine-learning (ML)-assisted framework for reliability and risk assessment of rock slopes under geotechnical uncertainty using random-field modelling and surrogate learning. Five stochastic input variables, discontinuity cohesion c^{\prime }, friction angle {\phi }^{\prime }, unit weight γ, slope height H, and face angle β, are considered, with spatial variability explicitly represented for the discontinuity-strength parameters. The deterministic planar-sliding model is calibrated to an FOS of approximately 1.20 under the mean input conditions, while 10,000 Monte Carlo simulations (MCS) yield a stochastic mean FOS of approximately 1.22 and a reference failure probability of P_{f}=0.3034. A multilayer perceptron (MLP) surrogate developed from 300 direct model evaluations is used to approximate the relationship between the stochastic inputs and FOS. The MLP achieves R^{2}=0.925, RMSE = 0.101, and MAE = 0.079, and estimates P_{f}=0.3155, corresponding to a relative difference of approximately 4.0% from the direct MCS estimate. Aleatory uncertainty is represented through the variability of the stochastic geotechnical inputs, while surrogate-related epistemic uncertainty is evaluated using bootstrap resampling. The surrogate-assisted approach reduces the number of required direct model evaluations from 10,000 to 300, substantially improving computational efficiency for reliability analysis. A consequence-based risk-screening framework further demonstrates how probabilistic reliability estimates can support monitoring and risk-mitigation decisions. The proposed framework provides a reproducible basis for extending ML-assisted rock-slope reliability analysis to site-specific numerical models and monitoring datasets.

Authors

Institutions

Publication Details

Journal
Rock mechanics letters.
Published
2026-11-18
DOI
https://doi.org/10.70425/rml.202604.53
Primary Topic
Geotechnical Engineering and Analysis
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Machine Learning-Assisted Reliability and Risk Assessment of Rock Slopes under Geotechnical Uncertainty

Mir Mujtaba Laghari, Khuzaima Saeed, Nirandoal Cheng, Ahmed Khalid et al.
1 links
Rock mechanics letters.
Geotechnical Engineering and Analysis
article

Machine Learning-Assisted Reliability and Risk Assessment of Rock Slopes under Geotechnical Uncertainty

Mir Mujtaba Laghari, Khuzaima Saeed, Nirandoal Cheng, Ahmed Khalid, Ahmed Mansoor
article en

Abstract

Rock-slope stability is strongly controlled by discontinuities and uncertain geotechnical parameters, while conventional deterministic factor-of-safety (FOS) calculations do not explicitly quantify the probability of failure. This study develops an integrated machine-learning (ML)-assisted framework for reliability and risk assessment of rock slopes under geotechnical uncertainty using random-field modelling and surrogate learning. Five stochastic input variables, discontinuity cohesion c^{\prime }, friction angle {\phi }^{\prime }, unit weight γ, slope height H, and face angle β, are considered, with spatial variability explicitly represented for the discontinuity-strength parameters. The deterministic planar-sliding model is calibrated to an FOS of approximately 1.20 under the mean input conditions, while 10,000 Monte Carlo simulations (MCS) yield a stochastic mean FOS of approximately 1.22 and a reference failure probability of P_{f}=0.3034. A multilayer perceptron (MLP) surrogate developed from 300 direct model evaluations is used to approximate the relationship between the stochastic inputs and FOS. The MLP achieves R^{2}=0.925, RMSE = 0.101, and MAE = 0.079, and estimates P_{f}=0.3155, corresponding to a relative difference of approximately 4.0% from the direct MCS estimate. Aleatory uncertainty is represented through the variability of the stochastic geotechnical inputs, while surrogate-related epistemic uncertainty is evaluated using bootstrap resampling. The surrogate-assisted approach reduces the number of required direct model evaluations from 10,000 to 300, substantially improving computational efficiency for reliability analysis. A consequence-based risk-screening framework further demonstrates how probabilistic reliability estimates can support monitoring and risk-mitigation decisions. The proposed framework provides a reproducible basis for extending ML-assisted rock-slope reliability analysis to site-specific numerical models and monitoring datasets.

Rock mechanics letters.Vol. 3(4)
Chinese Academy of Sciences (CN), Chengdu University of Technology (CN), UNSW Sydney (AU), State Key Laboratory of Oil and Gas Reservoir Geology and Exploitation (CN), Institute of Mountain Hazards and Environment (CN), University of Chinese Academy of Sciences (CN), State Key Laboratory of Geohazard Prevention and Geoenvironment Protection (CN)
Geotechnical Engineering and Analysis
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon