Literature Synthesis -- Batch 04 Agent 5¶
Covers 40 papers at file positions 761--800 (Simpson 2024 through Kang 2025).
Individual Paper Summaries¶
| # | Author | Year | Title | Core Finding | Method | Tags |
|---|---|---|---|---|---|---|
| 1 | Simpson et al. | 2024 | Estimating foundation parameters of OWTs through Bayesian model updating | Bayesian updating reduces uncertainty in OWT foundation stiffness estimation for fatigue assessment | Bayesian model updating, modal analysis | OWT, SHM, Bayesian, foundation-stiffness |
| 2 | Sindi (Paik) et al. | 2024 | Advancing digital healthcare engineering for aging ships and offshore structures | Five-module DHE framework integrating sensors, digital twins, and AI for lifetime structural healthcare | Review, digital twin, ML/DL diagnosis | digital-twin, SHM, offshore, corrosion, fatigue |
| 3 | Sk & Potnis | 2024 | Enhancing urban tunnelling efficiency through real-time geotechnical parameter analysis | Numerical regression links TBM operational parameters to geotechnical conditions in real time | Numerical regression, EPB-TBM field data | tunnelling, real-time, geotechnical-parameter |
| 4 | Srivastava & Singh | 2024 | Effect of seismic loading on V-H-M capacity of well foundation | Seismic inertial forces significantly reduce combined V-H-M capacity envelopes of well foundations | FELA, pseudo-static analysis | foundation, seismic, V-H-M, well-foundation |
| 5 | Tian, Wang & Phoon | 2024 | Real-time fusion of multi-source monitoring data with geotechnical numerical model results | Sparse dictionary learning fuses monitoring data with FE predictions for real-time geotechnical digital twins | Physics-informed sparse dictionary learning | digital-twin, data-fusion, monitoring, geotechnics |
| 6 | Tom, O'Loughlin & White | 2024 | Reliability of dynamically embedded anchors in soft clay | Numerical reliability framework quantifies probability of anchor capacity failure for floating offshore infrastructure | Reliability analysis, numerical installation/capacity model | anchor, reliability, soft-clay, offshore |
| 7 | Tsang et al. | 2024 | Geotechnical seismic isolation based on high-damping polyurethane: centrifuge modelling | HDPU-based GSI achieves 35--80% structural demand reduction across eight earthquake events | Centrifuge modelling | seismic-isolation, centrifuge, polyurethane, SFSI |
| 8 | Isnaniati et al. | 2024 | Effectiveness of micro-pile installation configuration for shallow foundation bearing capacity on soft clay | Micro-piles beneath/within foundation edge increase bearing capacity most effectively; extending beyond perimeter is not beneficial | Lab-scale model tests | micro-pile, soft-clay, shallow-foundation |
| 9 | He et al. | 2024 | Energy harvesting control of current transformers based on dynamic impedance regulation | Dynamic impedance VSC control achieves stable energy harvesting under wide CT current range and suppresses core saturation | Experimental, VSC impedance control | energy-harvesting, power-line, CT (off-topic) |
| 10 | Valdez-Yepez et al. | 2024 | Bolt-loosening detection in OWT jacket-type supports | (File unreadable -- extraction based on title) Bolt-loosening detected in jacket supports via vibration-based methods | SHM, vibration analysis | OWT, jacket, SHM, bolt-loosening |
| 11 | Waheed & Asmael | 2024 | Using soft soil models in geotechnical engineering: a review | Soft Soil model gives stiffer stress-strain than Hardening model; satisfactory for settlement of foundations in clay compared to Mohr-Coulomb | Review, FEM constitutive model comparison | constitutive-model, soft-soil, FEM, settlement |
| 12 | Walia et al. | 2024 | Spatially resolved multi-omic data integration using graph-attention variational autoencoder | Graph-attention VAE integrates spatial multi-omic data for tissue domain characterization | Graph attention, VAE, deep learning | bioinformatics, VAE (off-topic) |
| 13 | Wang et al. | 2024 | p-y curves for piles in uniform sand and normally consolidated clay | (File unreadable -- extraction based on title) Develops refined p-y curves for laterally loaded piles in sand and NC clay | Analytical/numerical p-y derivation | pile, p-y curve, lateral-loading, sand, clay |
| 14 | Xie et al. | 2024 | Investigation of local scour depth of pile foundation on migrating sand waves seabed | Dominant flow directions crucially shape scour under sand wave migration; proposes risk assessment and safety zoning method (RASZ) | Multibeam echo sounding, CFD simulation | scour, pile-foundation, sand-wave, OWT |
| 15 | Xin (Zhang) et al. | 2024 | Design calculation and construction control of suction drum foundation for guided frame platform | Proposes sinking feasibility analysis and parameter value method; verified via six installed four-barrel suction drum foundations | Field measurement, analytical design | suction-foundation, offshore-wind, design |
| 16 | Xu, Madabhushi et al. | 2024 | Comparison of rock-scour protection berm with shallow foundation under seismic liquefaction | For large earthquakes, rock berm and plate foundation settlements are comparable; some seabed ingress into rock after liquefaction | Dynamic saturated centrifuge tests | scour-protection, liquefaction, centrifuge, OWT |
| 17 | Xu, Soga et al. | 2024 | SHM of offshore wind turbines using distributed acoustic sensing (DAS) | Phi-OTDR DAS captures both local (loose bolt) and global tower strain over km-scale distances; first full-scale OWT fiber optic validation | Shake table test, DFOS (OFDR + phi-OTDR) | SHM, DAS, fiber-optic, OWT |
| 18 | Yaghoubi et al. | 2024 | Systematic review and meta-analysis of ANN, ML, DL, and ensemble learning in geotechnical engineering | Ensemble learning outperforms ANN, standalone ML, and DL in geotechnical prediction tasks; ANN still most widely used | Systematic review, meta-analysis, bibliometric | ML-review, geotechnics, ANN, ensemble |
| 19 | Yan et al. | 2024 | Experimental investigation on coupling of principal and shear stress of wind turbine under dynamic wind direction change | Wind direction misalignment causes abnormal aerodynamic distribution and stress coupling in turbine towers | Experimental (wind tunnel analogue) | wind-turbine, stress-coupling, aerodynamics |
| 20 | Yang et al. | 2024 | Rockburst prevention by microwave destressing: a numerical investigation | Microwave preconditioned zone closer to free face improves rockburst prevention but reduces static stability | Numerical simulation (impact-induced rockburst model) | rockburst, microwave, mining (off-topic) |
| 21 | Yao et al. | 2024 | Influence of local scour on dynamic responses of OWTs under wind-wave loads | Local scour alters OWT natural frequency and increases dynamic responses under combined wind-wave loading | Numerical (coupled wind-wave-structure-scour) | scour, OWT, dynamic-response, wind-wave |
| 22 | Yi et al. | 2024 | Design and analysis of novel offshore gravity energy storage support structure based on wind power jacket foundation | Jacket-integrated gravity energy storage structure passes static and modal feasibility checks (SACS analysis) | FEM (SACS), static and modal analysis | jacket, energy-storage, OWT, structural-design |
| 23 | Yoo et al. | 2024 | Numerical simulation of centrifuge tests for seismic response of liquefied slopes | FLAC 2D reproduces centrifuge-measured excess pore pressure, acceleration, and settlement for 15-deg and 27-deg liquefied slopes | Numerical (FLAC 2D) validated against centrifuge | liquefaction, slope, centrifuge, numerical |
| 24 | Zar et al. | 2024 | Towards vibration-based damage detection of civil engineering structures: overview, challenges, and future prospects | ML/DL (CNN, SVM, ANN) expanding SDD scope; unsupervised methods and physics-data blending are key future directions | Review (vibration-based SDD + ML/DL) | SHM, vibration, damage-detection, ML, review |
| 25 | Zeng et al. | 2024 | Soil slope landslide failure -- parameter analysis via strength reduction and orthogonal experimental design | Orthogonal design identifies dominant geotechnical factors for slope safety factor using strength reduction FEM | FEM (strength reduction), orthogonal design | slope-stability, landslide, parametric-study |
| 26 | Zhang X et al. | 2024 | Comparative analytical study of suction drum foundation penetration characteristics with real measurements | Compares multiple sinking resistance formulae (API, DNV, Houlsby) against field data from guide frame platforms | Analytical comparison, field measurement | suction-foundation, penetration-resistance |
| 27 | Zhang Z et al. | 2024 | Strain response prediction of OWT tower under free vibration | Modal superposition predicts full-field strain from limited measurement points; displacement-based modal coordinates outperform strain-based | FEM, modal superposition method | OWT, strain-prediction, virtual-sensing, modal |
| 28 | Zhu et al. | 2024 | Numerical model of MICP multi-process considering scale size | Scale-dependent MICP model accounts for multiple bio-cementation processes in soil improvement | Numerical modelling | MICP, bio-cementation, soil-improvement |
| 29 | Arbi et al. | 2025 | Optimized ML-based modeling of pile bearing capacity in layered soils using RS and GS | XGBoost with Grid Search achieves highest accuracy (R2 > 0.9) for pile bearing capacity in layered soils | ML (XGBoost, RF, SVM) + hyperparameter optimization | pile, bearing-capacity, ML, XGBoost |
| 30 | Byun et al. | 2025 | ML-based pattern recognition of bender element signals for predicting sand particle-size | 1D CNN classifies four sand types from bender element signals at given stress and cutoff frequency | CNN, bender element testing | bender-element, CNN, sand, particle-size |
| 31 | Chen et al. | 2025 | Sediment scouring at foundation of coupled wind-wave device: a numerical study | Wind-wave coupling device foundation experiences distinctive scour patterns requiring separate assessment from monopile-only cases | CFD numerical simulation | scour, wind-wave, coupled-device, OWT |
| 32 | Cheng et al. | 2025 | Lateral cyclic responses of OWT tripod suction bucket in clays | Contact condition, loading direction, and cyclic mode (one-way vs. two-way) significantly affect permanent rotation and failure mechanism | 3D FEM, bounding surface clay model | suction-bucket, tripod, cyclic-loading, clay |
| 33 | Dudzik & Szelag | 2025 | Analysis of odometric localization uncertainty using motion capture and Monte Carlo simulation | Monte Carlo simulation quantifies mobile robot localization uncertainty growing with time; normal distribution approximation valid | Monte Carlo, motion capture, kinematics | robotics, localization (off-topic) |
| 34 | Frech et al. | 2025 | New gridded offshore wind profile product for US coasts using ML and satellite observations | Random forest regression extrapolates satellite 10 m winds to 200 m hub-height profiles; outperforms ERA5 and conventional methods | Random forest, satellite data, triple collocation | offshore-wind, wind-resource, ML, satellite |
| 35 | Gendy | 2025 | Reliable prediction of bored pile load-settlement response using ML and Monte Carlo | GPR is top-performing model for pile load-settlement; Monte Carlo provides 95% confidence intervals on predictions | ML (GPR, XGBoost, GBM, RF, KNN, SVR), Monte Carlo, SHAP | pile, settlement, ML, GPR, uncertainty |
| 36 | Ghiasi & Malekjafarian | 2025 | Monitoring railway tracks maintenance needs using dynamic responses from in-service train | CFNN achieves 95% accuracy classifying tamping and surfacing needs from train acceleration data | Data-driven SHM, CFNN, ANOVA feature selection | railway-SHM, data-driven, maintenance |
| 37 | Gueye et al. | 2025 | RTCNet: robust hybrid DL model for soil property prediction under noisy conditions | Hybrid RNN-Transformer-CNN model maintains stable performance under increasing noise levels for SOC and soil fertility prediction | Hybrid DL (RNN + Transformer + CNN) | soil-property, deep-learning, noise-robustness |
| 38 | Harle & Wankhade | 2025 | ML techniques for predictive modelling in geotechnical engineering: a succinct review | Hybrid ML algorithms and comprehensive ground assessment are essential for advancing geotechnical prediction accuracy | Review (ML in geotechnics) | ML-review, geotechnics, seismic, settlement |
| 39 | Huynh et al. | 2025 | CNN deep learning approach to scour depth estimation around complex bridge piers | 1D CNN with Buckingham Pi non-dimensionalization outperforms FDOT, HEC-18, and Coleman empirical equations (R2 = 0.85) | 1D CNN, LSTM, Buckingham Pi | scour, bridge-pier, CNN, deep-learning |
| 40 | Kang & Kwon | 2025 | Scour reduction technique for offshore wind suction foundations using artificial seaweed mats | Artificial seaweed mats reduce flow velocity and suppress vortex around suction buckets; field-verified eco-friendly scour mitigation | Physical model test, field verification | scour, suction-foundation, eco-friendly, OWT |
Synthesis¶
CONSENSUS¶
-
ML/DL is transforming geotechnical prediction. Multiple reviews (Yaghoubi 2024, Harle 2025, Zar 2024) and applied studies (Arbi 2025, Gendy 2025, Huynh 2025) converge on the finding that ML models -- particularly ensemble methods (XGBoost, RF) and deep learning (CNN) -- outperform traditional empirical and deterministic approaches for predicting bearing capacity, settlement, scour depth, and damage states in geotechnical and structural systems.
-
Scour is a critical and multi-faceted threat to OWT foundations. Six papers (Xie 2024, Yao 2024, Xu-Madabhushi 2024, Chen 2025, Huynh 2025, Kang 2025) consistently show that scour modifies natural frequency, increases dynamic response, and can threaten structural stability. The consensus is that scour must be assessed alongside environmental loads rather than treated in isolation.
-
Suction bucket foundations are viable and increasingly adopted for OWTs, but design still relies on semi-empirical methods that need field validation. Three papers (Xin 2024, Zhang X 2024, Cheng 2025) confirm that penetration resistance formulae from API/DNV remain the standard, yet discrepancies with field data persist.
-
Distributed and data-driven SHM enables condition-based maintenance. Xu-Soga 2024 (DAS), Simpson 2024 (Bayesian updating), Valdez-Yepez 2024 (bolt loosening), Zhang Z 2024 (virtual sensing), and Ghiasi 2025 (railway) all demonstrate that combining physics-based models with measured data can detect local damage and track global structural changes in real time.
DEBATES¶
-
ANN vs. ensemble vs. deep learning for geotechnical prediction. Yaghoubi 2024 finds ensemble learning superior overall, yet ANN remains most widely used due to simplicity and smaller data requirements. Gueye 2025 argues hybrid DL architectures (RNN+Transformer+CNN) are needed for noisy real-world conditions. No consensus on when simpler models suffice vs. when deep architectures are justified.
-
Constitutive model choice for soft soils. Waheed 2024 reviews Soft Soil vs. Hardening Soil vs. Mohr-Coulomb but notes each has context-dependent strengths. The field lacks a universal recommendation for when advanced models justify their computational cost.
-
Scour protection strategy. Conventional rock-dump (Xu-Madabhushi 2024) vs. eco-friendly artificial seaweed mats (Kang 2025) represent competing paradigms. Rock-dump is well-proven but ecologically disruptive; seaweed mats are sustainable but lack long-term performance data.
-
Physics-informed vs. purely data-driven approaches. Tian 2024 advocates physics-informed sparse dictionary learning; Ghiasi 2025 and others use purely data-driven classification. The optimal balance between physics constraints and data flexibility for geotechnical digital twins remains unresolved.
GAPS¶
-
Long-term cyclic and combined loading on suction bucket foundations in clay is under-studied for tripod configurations (Cheng 2025 notes literature is sparse). Field-scale cyclic data are essentially absent.
-
Scour under seismic liquefaction is investigated by only one paper (Xu-Madabhushi 2024). Interaction between earthquake-induced liquefaction and scour around OWT foundations in seismically active regions needs systematic study.
-
Transfer learning and generalization of ML models across geotechnical sites. Most ML studies (Arbi 2025, Gendy 2025, Huynh 2025) train and validate on single datasets. Cross-site generalization and domain adaptation are not addressed.
-
Noise robustness of geotechnical ML models. Gueye 2025 is the only paper explicitly testing model degradation under noisy inputs. Real sensor data are inherently noisy, yet most studies assume clean training data.
-
Integration of scour monitoring into OWT digital twin frameworks. Although digital twins (Sindi 2024, Tian 2024) and scour studies (Yao 2024, Xie 2024) exist independently, no paper integrates real-time scour evolution into a full structural digital twin for OWT.
METHODS¶
| Method Category | Papers | Notes |
|---|---|---|
| Centrifuge modelling | Tsang 2024, Xu-Madabhushi 2024, Yoo 2024 | Dynamic saturated tests for seismic and liquefaction |
| FEM / numerical | Srivastava 2024, Yao 2024, Yi 2024, Yoo 2024, Cheng 2025, Chen 2025 | FELA, FLAC 2D, SACS, OpenFAST |
| ML -- ensemble (RF, XGBoost, GBM) | Arbi 2025, Gendy 2025, Frech 2025 | XGBoost consistently top performer for tabular geotechnical data |
| ML -- deep learning (CNN, LSTM, Transformer) | Byun 2025, Huynh 2025, Gueye 2025, Zar 2024 | 1D CNN prominent for time-series and signal data |
| Bayesian / probabilistic | Simpson 2024, Tom 2024, Gendy 2025 | Bayesian updating and Monte Carlo for uncertainty quantification |
| Fiber optic / DAS sensing | Xu-Soga 2024 | First full-scale OWT DAS validation |
| Modal superposition / virtual sensing | Zhang Z 2024 | Strain prediction from limited sensors |
| Physical model (lab-scale) | Isnaniati 2024, Kang 2025 | Scaled foundation and scour tests |
| Field measurement / validation | Xin 2024, Zhang X 2024, Kang 2025 | Suction foundation installation records |
| Systematic review / meta-analysis | Yaghoubi 2024, Harle 2025, Waheed 2024, Zar 2024 | Bibliometric and qualitative synthesis |
BENCHMARKS¶
- Pile bearing capacity ML: R2 > 0.9 (Arbi 2025, XGBoost+GS); GPR top for pile settlement (Gendy 2025).
- Scour depth prediction: 1D CNN R2 = 0.85, RMSE = 0.1125 (Huynh 2025), outperforming HEC-18 and FDOT.
- SHM bolt-loosening detection: DAS phi-OTDR captures flange bolt loosening at full-scale (Xu-Soga 2024).
- Seismic demand reduction via GSI: 35--80% across eight earthquake motions (Tsang 2024).
- Railway maintenance classification: 95% accuracy with CFNN using time-domain features (Ghiasi 2025).
- Soil property prediction under noise: RTCNet MSE = 0.1032 (noise-free), stable under increasing noise (Gueye 2025).
- Wind profile extrapolation: RF model outperforms ERA5 at 100 m hub height across US coasts (Frech 2025).
Off-topic papers excluded from synthesis: He 2024 (energy harvesting CT), Walia 2024 (bioinformatics VAE), Yang 2024 (rockburst microwave), Dudzik 2025 (robot localization). Metadata retained in the table for completeness.