An interpretable hybrid FELA–boosting framework for predicting pullout capacity of rectangular buried pipelines in dense sand

Accurately predicting the pullout capacity of buried rectangular pipelines in dense sand remains a significant challenge because of the complex interaction between pipeline geometry, stress-dependent soil behavior, and failure mechanism development. This study combines Finite Element Limit Analysis (FELA), the Bolton model, and gradient-boosting machine learning algorithms to evaluate the normalized pullout capacity factor (Nγ). The parametric study varied the embedment ratio (H/B), aspect ratio (L/B), soil unit weight (γ), relative density (DR), particle crushing strength (Q), and critical-state friction angle (ϕcv). The results show that geometric parameters dominate the uplift response, with higher H/B increasing confinement and larger L/B promoting wider plastic failure zones. Soil parameters influence the response through stress-dependent dilatancy, where higher DR, Q, and ϕcv increase Nγ, whereas higher γ slightly reduces the normalized pullout capacity factor. The 2880-case FELA database generated using Optum G2 was divided into four unit-weight-specific subsets of 720 cases, with separate XGBoost, CatBoost, and LightGBM models developed for each subset. For γ = 20 kN/m3, LightGBM achieved the best performance on the adopted testing split (R2 = 0.9934), while CatBoost achieved the highest training accuracy. SHAP analysis of the XGBoost models and first-order Sobol analysis of the FELA results both identified H/B and L/B as the dominant parameters, although their ordering differed. Overall, the framework provides efficient surrogate estimates of FELA-derived pullout capacity and interpretable parameter trends within the investigated ranges.

Authors

Institutions

Publication Details

Journal
Ships and Offshore Structures
Published
2026-11-20
DOI
https://doi.org/10.1080/17445302.2026.2764191
Primary Topic
Geotechnical Engineering and Underground Structures
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

An interpretable hybrid FELA–boosting framework for predicting pullout capacity of rectangular buried pipelines in dense sand

Wittaya Jitchaijaroen, Suraparb Keawsawasvong, Gaoqiao Wu, Syaifulloh Qoimuddin Ali Basyah et al.
6 links
Ships and Offshore Structures
Geotechnical Engineering and Underground Structures
article

An interpretable hybrid FELA–boosting framework for predicting pullout capacity of rectangular buried pipelines in dense sand

Wittaya Jitchaijaroen, Suraparb Keawsawasvong, Gaoqiao Wu, Syaifulloh Qoimuddin Ali Basyah, Rachma Auliya Marifah
article en

Abstract

Accurately predicting the pullout capacity of buried rectangular pipelines in dense sand remains a significant challenge because of the complex interaction between pipeline geometry, stress-dependent soil behavior, and failure mechanism development. This study combines Finite Element Limit Analysis (FELA), the Bolton model, and gradient-boosting machine learning algorithms to evaluate the normalized pullout capacity factor (Nγ). The parametric study varied the embedment ratio (H/B), aspect ratio (L/B), soil unit weight (γ), relative density (DR), particle crushing strength (Q), and critical-state friction angle (ϕcv). The results show that geometric parameters dominate the uplift response, with higher H/B increasing confinement and larger L/B promoting wider plastic failure zones. Soil parameters influence the response through stress-dependent dilatancy, where higher DR, Q, and ϕcv increase Nγ, whereas higher γ slightly reduces the normalized pullout capacity factor. The 2880-case FELA database generated using Optum G2 was divided into four unit-weight-specific subsets of 720 cases, with separate XGBoost, CatBoost, and LightGBM models developed for each subset. For γ = 20 kN/m3, LightGBM achieved the best performance on the adopted testing split (R2 = 0.9934), while CatBoost achieved the highest training accuracy. SHAP analysis of the XGBoost models and first-order Sobol analysis of the FELA results both identified H/B and L/B as the dominant parameters, although their ordering differed. Overall, the framework provides efficient surrogate estimates of FELA-derived pullout capacity and interpretable parameter trends within the investigated ranges.

Ships and Offshore Structures
Thammasat University (TH), Changsha University of Science and Technology (CN)
National Natural Science Foundation of China, Thailand Science Research and Innovation
Industry, innovation and infrastructure
Geotechnical Engineering and Underground Structures
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