Literature Synthesis: Batch 03 Agent 2 (Positions 441-480)

Scope: 40 papers from 2021-2022, covering offshore wind, geotechnics, SHM, ML applications, slope stability, foundations, and sensing.

---

Individual Paper Summaries

#	Author(s)	Year	Title	Core Finding	Method	Tags
1	Savvides & Papadrakakis	2021	Uncertainty quantification of failure of clays with modified Cam-Clay	Failure load and displacements follow Gaussian distribution despite nonlinearity; failure load decreases with soil depth	Stochastic FEM, Monte Carlo, Latin hypercube sampling	uncertainty clay bearing-capacity SFEM
2	Shi et al.	2021	DFOS Applications to Geo-Engineering Monitoring	DFOS provides continuous spatial monitoring superior to point-wise sensors for geo-engineering hazards	Review of distributed fiber optic sensing (DFOS) technologies	DFOS monitoring geohazard review
3	Shi et al.	2021	Small-strain shear modulus of calcareous sand under anisotropic consolidation	Characterized G_max of calcareous sand under anisotropic stress states	Bender element tests, anisotropic consolidation	calcareous-sand shear-modulus bender-element lab-test
4	Shittu et al.	2021	S-N vs fracture mechanics reliability of OWT jacket foundations	FM reliability is highly sensitive to initial crack size; S-N recommended at design stage, FM near end of life	FEA + ANN response surface + FORM reliability	OWT fatigue reliability jacket ANN
5	Si et al.	2021	Power-take-off control on combined semi-submersible FOWT	(Insufficient text extracted -- quality warning: too short)	AQWA numerical simulation	FOWT PTO-control semi-submersible
6	Somoano et al.	2021	Uncertainties in real-time hybrid model testing of FOWT	Assessed accuracy of ReaTHM testing for a 15 MW FOWT on concrete semi-submersible	Scaled ocean basin experiment with multi-fan system	FOWT hybrid-testing uncertainty experimental
7	Srivastava et al.	2021	GRS wall subjected to static footing loading	Settlement reduced 36%, bearing capacity increased 42% vs conventional retaining walls	FEM (Optum G2), parametric study	GRS-wall bearing-capacity FEM retaining-wall
8	Stavi et al.	2021	Soil salinity and sodicity in drylands	Review of causes, effects, monitoring, and restoration of saline/sodic soils	Literature review	soil-salinity drylands environmental review
9	Sun et al.	2021	Clogging of dredge slurry under vacuum preloading	Visualized clogging formation mechanism using PIV/PTV digital image technology	PIV, PTV, vacuum preloading experiment	soft-ground vacuum-preloading clogging PIV
10	Sunday & Brennan	2021	Review of OWT monopile structural design	Current design codes from O&G are insufficient for larger monopiles; damping, SSI, corrosion, and grouted connections are key challenges	Literature review of DNVGL standards	monopile OWT design-review SSI corrosion
11	Svendsen et al.	2021	Data-based SHM for damage detection in steel bridges	Unsupervised ML performs almost as well as supervised ML for detecting fatigue damage in steel bridges	Experimental study, ML (supervised + unsupervised), ROC analysis	SHM bridge ML damage-detection fatigue
12	Sychterz et al.	2021	Nonlinear soil-structure behavior of deployable anchor system	Characterized nonlinear SSI behavior of a novel deployable compliant anchor	Limit equilibrium, soil-structure interaction analysis	anchor SSI deployable-structures nonlinear
13	Sysala et al.	2021	Optimization variant of shear strength reduction method	New OPT-SSR computes FoS without elasto-plastic analysis; duality between static/kinematic principles derived	Convex optimization, FEM, mesh adaptivity, damped Newton	slope-stability SSR optimization FEM
14	Tarpo et al.	2021	Virtual sensing and strain estimation on OWT using supervised learning	PCA-based data-driven model enables versatile strain estimation across wind scenarios	Supervised learning, PCA, virtual sensing	OWT virtual-sensing strain ML PCA
15	Tran et al.	2021	Fast single-cell data analysis using hierarchical autoencoder (scDHA)	scDHA outperforms state-of-the-art in cell segregation, visualization, classification, and pseudo-time inference	Non-negative kernel AE + stacked Bayesian AE with perturbation	autoencoder scRNA-seq deep-learning dimensionality-reduction
16	Tuck et al.	2021	Sediment supply dampens erosive effects of sea-level rise on reef islands	Sediment supply promotes island elevation increase and dampens migration/volume loss under SLR	Physical modelling experiments	reef-islands sea-level-rise sediment coastal morphodynamics
17	Varkonyi et al.	2021	Rigid impacts of 3D rocking structures	3D impacts exhibit extreme sensitivity to geometric imperfections; spontaneous symmetry breaking observed	Experimental + analytical 3D impact model (Housner generalization)	rocking earthquake 3D-dynamics impact-model
18	Wang et al.	2021	Macro-element model for spudcan foundations in clay overlying sand	Macro-element predicts combined VHM loading on spudcans in layered soil	Macro-element modelling	spudcan foundation clay-sand macro-element
19	Wang et al.	2021	Progressive failure in layered clayey slopes under seismic loading (PFEM)	Layered slopes can exhibit both shallow and deep failure modes sequentially; strain-softening controls slip surface	Particle FEM (PFEM), strain-weakening clay model	slope-failure seismic PFEM clay landslide
20	Wang et al.	2021	Coral sand mechanical characteristics under repeated loading	Characterized particle breakage of coral sand under 1D cyclic loading	Laboratory testing, 1D repeated loading	coral-sand particle-breakage cyclic-loading lab-test
21	Wang et al.	2021	Review of bridge SHM based on GNSS	GNSS technology enables displacement monitoring and dynamic characteristic identification for bridges	Review of GNSS positioning, multi-frequency/multi-system solutions	bridge SHM GNSS displacement review
22	Xie et al.	2021	Loess landslide mechanism and numerical simulation stabilization	Anti-slide piles effectively stabilized the Zhonglou Mountain landslide; maximum shear strain shifts under rainfall	Strength reduction FEM, field monitoring	loess landslide anti-slide-pile FEM stabilization
23	Xing et al.	2021	Large-scale landslide susceptibility mapping with integrated ML	Integrated ML model for regional landslide susceptibility in Lvliang Mountains	Machine learning, landslide susceptibility mapping	landslide susceptibility ML GIS mapping
24	Yadav et al.	2021	Soil analysis and crop fertility prediction using ML	ANN deep learning achieves best accuracy for soil fertility classification and crop yield prediction	SVM, RF, Naive Bayes, ANN, MLP	agriculture soil-classification ML crop-yield
25	Yang et al.	2021	Centrifuge modelling for seismic retrofit of bridge abutment pile foundation in liquefied ground	Steel pipe sheet pile wall (front separation type) effectively reduces lateral flow forces on existing piles	Dynamic centrifuge model tests	centrifuge liquefaction bridge-abutment seismic-retrofit pile
26	Yang & Yin	2021	Soil-structure interface modelling with nonlinear incremental approach	Nonlinear incremental approach for SSI interface modelling	Numerical method, interface constitutive model	SSI interface nonlinear constitutive-model
27	Yavari et al.	2021	ML models for autonomous downhole inflow control devices	SVM achieves best prediction (RMSE 0.14 MPa, R2=0.98) for AICD differential pressure	ANFIS, ANN, SVM, regression models	petroleum ML SVM downhole-control
28	Yin & Wang	2021	Micro-mechanical analysis of caisson foundation: particle shape effect	Angular particles increase horizontal/rotational bearing capacity and force chain anisotropy vs spheres	DEM with spherical vs tetrahedral particles	DEM caisson particle-shape micromechanics
29	Yun & Han	2021	Dynamic behavior of pile-supported wharves by slope failure during earthquake	Slope failure increased deck acceleration by 24% and pile moment by 31% via kinematic forces	Dynamic centrifuge model tests	centrifuge wharf pile slope-failure seismic
30	Zerrouki et al.	2021	Desertification detection using improved VAE on Landsat data	VAE consistently outperformed RBM, deep learning, and clustering for land cover change detection	Variational autoencoder, Landsat multitemporal imagery	VAE remote-sensing desertification deep-learning
31	Zhang et al.	2021	Seismic analysis of 10-MW OWT with monopile considering seabed liquefaction	Large-diameter monopile motion causes more extensive liquefaction zone; combined loading critical	Multi-surface elastoplastic model, dynamic FEA	OWT monopile liquefaction seismic 10MW
32	Zhang et al.	2021	State-of-the-art review: ML in constitutive modeling of soils	Comprehensive review of ML replacing or augmenting traditional constitutive soil models	Review paper	ML constitutive-model soil review
33	Zhao et al.	2021	SCDRHA: scRNA-seq dimensionality reduction via hierarchical autoencoder	Hierarchical autoencoder achieves effective dimensionality reduction for single-cell RNA-seq data	Hierarchical autoencoder	autoencoder scRNA-seq dimensionality-reduction
34	Zhu et al.	2021	Sensing Earth dynamics by telecom fiber-optic sensors (FORESEE)	DAS on existing telecom fiber can discriminate earthquakes, storms, vehicles, blasts for continuous urban monitoring	Distributed acoustic sensing (DAS), field experiment	DAS fiber-optic seismic-monitoring urban
35	Zhuang et al.	2021	Pullout behaviour of inclined shallow plate anchors in sand	Characterized effects of anchor inclination and interface conditions on pullout capacity and failure mechanism	1g model tests in sand	anchor pullout sand model-test inclination
36	x (Ritik)	2021	Diabetes prediction using machine learning	XGBoost achieves 85.24% accuracy on PIMA diabetes dataset	Random Forest, XGBoost	ML diabetes prediction medical
37	(Unnamed)	2021	Scour- and erosion-related issues (Lecture Notes in Civil Engineering)	Book/proceedings on scour and erosion topics in civil engineering	Conference proceedings	scour erosion offshore proceedings
38	Abbas et al.	2022	ROSCO: Reference open-source controller for fixed and floating OWT	ROSCO reduces max rotor thrust >10% and platform pitch ~30%; modular and auto-tunable	Controller design, OpenFAST simulation	OWT control ROSCO floating open-source
39	Abdalrada et al.	2022	ML models for co-occurrence of diabetes and CVD	Two-stage ML model achieves 94.09% accuracy for DM+CVD co-occurrence prediction	Logistic regression, MARS, classification/regression tree	ML diabetes cardiovascular prediction medical
40	Andersen et al.	2022	Validation of RC pile caps using NLFEM and FELA	FELA provides safe, automated capacity estimates; NLFEM gives richer but more complex behavioral predictions	NLFEM (DIANA), FELA, comparison with experiments	pile-cap NLFEM FELA concrete validation

---

SYNTHESIS

CONSENSUS

1. ML is a viable replacement/complement to physics-based models across domains. Papers on SHM (Svendsen, Tarpo), soil constitutive modeling (Zhang review), slope stability (Xing), and foundation analysis (Shittu) all converge on the finding that ML models -- particularly SVM, ANN, and ensemble methods -- match or exceed classical analytical approaches when sufficient training data exists. The Zhang (2021) review explicitly maps the landscape of ML-for-constitutive-models, confirming this as an established trajectory.

2. Soil-structure interaction remains critical for offshore wind foundations. Sunday's monopile review, Zhang's liquefaction study, and the broader centrifuge work by Yang and Yun all affirm that SSI -- especially under combined loading, cyclic degradation, and seismic conditions -- is the dominant uncertainty source in OWT foundation design. The inadequacy of oil-and-gas-era codes for next-generation monopiles (10 MW+) is repeatedly flagged.

3. Centrifuge and physical modelling continue to anchor experimental validation. Yun (wharf piles), Yang (bridge abutment retrofit in liquefied ground), Zhuang (inclined anchors), and Somoano (FOWT hybrid testing) all rely on centrifuge or scaled physical models. These remain the gold standard for validating numerical predictions of dynamic soil-foundation-structure response.

4. Distributed and remote sensing technologies are maturing for infrastructure monitoring. DFOS (Shi), DAS on telecom fiber (Zhu), and GNSS-based SHM (Wang) collectively demonstrate that spatially continuous monitoring is moving from research to deployment, offering orders-of-magnitude improvement over point sensors.

DEBATES

1. S-N curves vs fracture mechanics for fatigue life assessment. Shittu et al. argue that S-N is appropriate for design and FM for end-of-life assessment, but this division remains contested. The sensitivity of FM results to initial crack size (which is poorly known in practice) undermines confidence in FM predictions, while S-N curves lack mechanistic resolution.

2. Traditional constitutive models vs ML surrogates. Zhang's review notes growing ML adoption for soil modeling, but the lack of physical interpretability, extrapolation reliability, and training-data requirements remain unresolved. The community has not reached consensus on when ML surrogates are "safe" to deploy in design-critical geotechnical applications.

3. Supervised vs unsupervised ML for SHM damage detection. Svendsen shows unsupervised ML performing nearly as well as supervised ML for bridge damage detection, which challenges the conventional assumption that labeled damage data is essential. This has significant practical implications since damage-state data is rare for real structures.

GAPS

1. Combined multi-hazard loading on OWT foundations. While Zhang studies earthquake + wind + wave on a 10 MW OWT, and Sunday reviews monopile design, no paper addresses the full coupled problem of scour + liquefaction + fatigue + corrosion under long-term operation. Each hazard is studied in isolation.

2. Transfer learning and domain adaptation for geotechnical ML. The autoencoder papers (Tran, Zhao, Zerrouki) demonstrate hierarchical/variational architectures for dimensionality reduction in biology and remote sensing, but analogous architectures have not been applied to geotechnical data (e.g., CPT profiles, load-displacement curves, sensor streams). Cross-domain transfer of representation learning methods is unexplored.

3. Full-scale validation of virtual sensing on operating OWTs. Tarpo's PCA-based strain estimation is promising but demonstrated only under controlled conditions. No paper validates virtual sensing against long-term field measurements on operating turbines including sensor degradation and environmental non-stationarity.

4. 3D effects in rocking and dynamic foundation response. Varkonyi shows that 3D impacts exhibit extreme sensitivity and spontaneous symmetry breaking, but current earthquake assessment methods remain predominantly 2D. Bridging the gap between 3D lab observations and design-level models is an open problem.

5. Sediment supply in coastal erosion models. Tuck et al. demonstrate that sediment supply substantially changes reef island response to SLR, yet most flood-risk and erosion models treat islands as static. This gap applies broadly to any coast where sediment transport is coupled with morphological change.

METHODS

• Numerical: FEM dominates (Savvides SFEM, Sysala OPT-SSR, Xie strength reduction, Andersen NLFEM/FELA). PFEM (Wang) addresses large-deformation problems that standard FEM cannot handle. DEM (Yin) provides particle-scale insight for foundations.
• Physical modelling: Centrifuge tests (Yang, Yun) and 1g model tests (Zhuang, Tuck) remain primary experimental tools. Scaled ocean basin testing with real-time hybrid methods (Somoano) extends physical modelling to aero-hydro-structural coupled systems.
• Machine learning: SVM, ANN, RF, and ensemble methods (XGBoost) appear across geotechnics, SHM, petroleum, and medical applications. PCA-based virtual sensing (Tarpo) and autoencoder-based feature extraction (Tran, Zhao, Zerrouki) represent the deep learning frontier.
• Sensing: DFOS, DAS, GNSS, and bender elements span macro-infrastructure monitoring to lab-scale soil characterization.

BENCHMARKS

• Bearing capacity: Savvides benchmarks stochastic FEM against Meyerhoff analytical solutions for clay.
• Slope stability FoS: Sysala compares OPT-SSR against Plaxis and Comsol Multiphysics solutions.
• Pile cap capacity: Andersen validates NLFEM (DIANA) and FELA against four-pile-cap experiments.
• FOWT controller: Abbas benchmarks ROSCO against NREL 5 MW reference controller (>10% thrust reduction, ~30% platform pitch reduction).
• ML accuracy baselines: Shittu (ANN for fatigue), Svendsen (ML for bridge SHM), Yavari (SVM R2=0.98 for AICD), Abdalrada (94% for DM+CVD). These provide reproducible accuracy targets for future ML-in-engineering studies.
• Centrifuge scaling: Yang (bridge abutment in liquefied ground) and Yun (wharf piles) provide centrifuge-scale benchmark datasets for seismic SSI validation.

---

Generated: 2026-04-17 | 40 papers reviewed | Positions 441-480