Dissertation Gap Analysis — Ten Papers, One Spine¶

Date: 2026-04-17 Status: Working notes, merged into dissertation planning. Defence: 2026-09-03.

Personal notes taken while merging ten separate first-author papers into a single dissertation. The point of the exercise: make the shared spine explicit, sharpen each paper's gap claim from defensive ("nobody has done X") to offensive ("everybody assumed Y, and Y is wrong"), and audit which papers are ready to stand that harder test.

Context¶

The dissertation is about scour assessment of offshore wind turbine tripod suction bucket foundations. Ten papers, one foundation type, one instrumented turbine, a spine of field data and a parallel spine of numerical and experimental work. Each paper answers a distinct question; together they close the loop from cavity-expansion theory to cost-optimal maintenance decision.

One foundation, one site, one question, five research communities that normally don't talk to each other: geotechnics, structural dynamics, aeroelastics, statistical SHM, decision theory under uncertainty.

The Portfolio Arc¶

THEORY → NUMERICAL → EXPERIMENTAL → SOFTWARE → PROBABILISTIC → FIELD → FEATURES → STATISTICS → DECISION → TRANSFER
  J11       J2           J3             Op3        J5            V1       B           V2          A          E

Each paper produces something the next consumes:

J11 justifies the spring shape → J2 uses it.
J2 produces power-law coefficients (PL-1) → V1, A, Op3 consume them.
J3 validates J2 in the centrifuge → cross-checks PL-1.
Op3 automates the computation → J5 runs 1,794 Monte Carlo realisations on it.
J5 produces the capacity distribution → A prices the decision.
V1 produces the field detection threshold → A fuses it.
B ranks the features → V2 and A pick from the ranking.
V2 picks the compensation method → V1 and B use it.
A closes the loop from measurement to maintenance action.
E compresses J5's database into a transferable latent representation → enables A to deploy at new sites without re-running Monte Carlo.

The Ten Ultimate Research Questions¶

Each paper is organised around a single question. No two overlap.

J2 (Winkler Model). Can the scour–frequency relationship for tripod suction bucket foundations be predicted from first principles (3D FE → 1D Winkler) with sufficient accuracy to replace underwater inspection?
J3 (Centrifuge Saturation). Does the soil saturation state fundamentally change the scour–frequency sensitivity of offshore wind foundations, and if so, by how much and through what mechanism?
V1 (Field Monitoring). Can irreversible scour be distinguished from reversible environmental variability on an operational offshore wind turbine using physics-informed regression, and for how long must monitoring continue to establish a reliable baseline?
V2 (Cointegration Diagnosis). Why does cointegration — the most widely recommended statistical method for SHM — fail categorically on offshore wind monitoring data, and what does this failure reveal about the requirements for successful environmental compensation?
B (Buckingham Pi Features). Among all physically meaningful vibration-derived features, which one carries the scour signal through the tripod's load-sharing geometry, and can it transfer from centrifuge to field via dimensional analysis?
A (Bayesian Decision). How should an engineer translate ambiguous multi-channel monitoring evidence into a cost-optimal maintenance action, and how much is that monitoring system worth in avoided lifecycle cost?
Op3 (Software Framework). Can the three-stage foundation assessment chain (geotechnical → structural → aero-elastic) be automated into a single pipeline that enables population-scale probabilistic analysis?
J5 (Monte Carlo Capacity). How does soil-parameter uncertainty interact with scour to amplify the capacity degradation of tripod suction bucket foundations, and where does the Gunsan site sit within the plausible population?
J11 (Vesic Dissipation). Can the classical Vesic cavity-expansion framework be generalised through a spatially varying dissipation weight to explain why suction bucket Winkler springs have their characteristic depth-varying shape?
E (Physics-Informed Encoder). Can a simulation-trained encoder predict foundation capacity across unseen soil types through CPT conditioning, without retraining, while satisfying physics constraints that the training data alone does not enforce?

Gap Verdict Table¶

Every paper's claim to novelty, audited against the current literature:

Paper	Gap Claim	Verdict
J2	No BNWF for suction buckets under scour	Confirmed
J3	First saturated centrifuge for TSB scour–frequency	Confirmed
V1	Longest scour-detection field record for any OWT	Confirmed
V2	Cointegration never tested on parked-state OWT data	Confirmed
B	No systematic dimensionless feature evaluation	Confirmed
A	No multi-channel monitoring-to-decision framework for OWT	Confirmed
Op3	No open-source pipeline coupling 3D LA + dynamics + aero-elastic	Confirmed
J5	No 3D MC limit analysis for multi-legged scoured foundations	Confirmed
J11	No dissipation-weighted Vesic for depth-varying springs	Confirmed
E	No cross-soil encoder with physics constraints	Confirmed (strongest gap)

All ten gaps are defensible — they pass the "nobody has done X" test. But that's the minimum bar. The point of the next section is to convert each defensive gap into an offensive one.

Per-Paper Framing¶

For each paper: defensive gap (what hasn't been done), offensive gap (what's been done wrong), constructive gap (what unlocks downstream), plus the opening sentence drafted for the rewritten introduction.

Paper J2 — FE-Calibrated Winkler Model for Scour-Frequency Degradation¶

Question. Can a 3D-FE-calibrated 1D Winkler model reproduce the scour-induced frequency degradation of a tripod suction bucket foundation with accuracy sufficient to replace full 3D analysis in design practice?

Defensive. No published Winkler model has been calibrated against 3D FE analysis for the scour–frequency relationship of tripod suction bucket foundations.
Offensive. The field treats every foundation as a monopile in disguise — applying single-column Winkler models derived for L/D > 10 to a three-footed system with L/D ≈ 1, whose load redistribution makes those models fundamentally non-conservative under asymmetric scour.
Constructive. A validated 1D surrogate enables rapid parametric sweeps (J5), real-time digital-twin updating (A), and field-deployable monitoring thresholds (V1) that are computationally prohibitive with full FE.

Opening. Every published method for predicting scour-induced frequency loss in offshore wind foundations assumes a single embedded column, yet the tripod suction bucket — already deployed at gigawatt scale — distributes load through three footings whose interaction no Winkler model has ever captured.

Anchors. Arany et al. (2016), Prendergast et al. (2015), Qi et al. (2016), Jalbi & Bhattacharya (2020), Li et al. (2020).

Paper J3 — Centrifuge: Saturation and Backfill Effects¶

Question. Does natural backfill after a scour event recover the dynamic stiffness of a tripod suction bucket foundation, or does the loose backfill material conceal ongoing structural vulnerability?

Defensive. No centrifuge study has quantified how natural backfill after scour affects the dynamic stiffness of suction bucket foundations under saturated conditions.
Offensive. The field's scour experiments universally excavate a pristine sand bed once and declare the problem solved, ignoring that every storm cycle backfills the scour hole with loose sediment whose near-zero stiffness creates a false sense of geometric recovery.
Constructive. Quantifying stiffness recovery (or its absence) determines whether bathymetric surveys alone suffice for monitoring, or whether vibration-based methods (V1, B) are the only reliable indicators.

Opening. The offshore wind industry treats scour as a permanent excavation, yet field surveys consistently show that storm-scoured seabeds partially refill within weeks — and no experiment has ever measured whether that backfill restores any structural stiffness to the foundation.

Anchors. Sumer et al. (2013), Ciancimino et al. (2022), Kim et al. (2025, J1), Qi & Gao (2014), Pan et al. (2019).

Paper V1 — 32-Month Field Vibration Monitoring¶

Question. Can vibration-based monitoring detect scour on an operational offshore wind turbine tripod, and what is the minimum detectable scour depth under real environmental noise?

Defensive. No operational offshore wind turbine has been continuously monitored for scour-induced frequency changes with co-located ground-truth data over multiple years.
Offensive. The community has published dozens of numerical proofs that natural frequency tracks scour depth, yet the entire evidence base for this claim rests on laboratory tanks and finite-element models — not one study has recorded both measurements simultaneously on a turbine exposed to real wind, waves, and tides.
Constructive. A validated field detection threshold converts vibration-based scour monitoring from an academic curiosity into a deployable O&M tool, providing the detection channel that feeds the Bayesian decision framework (A).

Opening. Three decades of research have established that an offshore wind turbine's natural frequency falls when its foundation is scoured, yet until this study, not a single operational turbine had been continuously monitored to verify that claim under the full complexity of wind, waves, temperature, and tidal loading.

Anchors. Weijtjens et al. (2017), Prendergast et al. (2015), Doebling et al. (1998), Zaaijer (2006), Stuyts et al. (2022).

Headline number. 32 months, zero false alarms, 0.24D detection threshold.

Paper V2 — Cointegration Diagnosis and State-Function Decomposition¶

Question. Why does Johansen cointegration — widely recommended for SHM — fail categorically on parked-state OWT data, and which compensator survives the same dataset?

Defensive. No study has benchmarked competing EOV compensation methods on the same operational OWT scour-monitoring dataset.
Offensive. Four research communities independently advocate regression, cointegration, PCA, and Gaussian-process methods for environmental compensation, each validated only on its own dataset — leaving practitioners to choose by academic reputation rather than comparative evidence. When they're put on the same data, cointegration's I(1) assumption collapses (full-rank Johansen, 99.94% false-alarm rate) because parked-state OWT features are stationary around a reversible environmental manifold, not a drifting common trend.
Constructive. Identifying the optimal compensator improves the signal-to-noise ratio of V1's detector and reduces the minimum detectable scour depth, which directly lowers the inspection cost priced in A.

Opening. The offshore structural health monitoring community has spent fifteen years developing environmental compensation methods in parallel silos, yet no study has ever applied them to the same turbine, the same sensors, and the same scour event to determine which one actually works.

Anchors. Cross et al. (2012), Sohn (2007), Weijtjens et al. (2016), Deraemaeker & Worden (2018), Hu et al. (2015).

Headline numbers. Cointegration FAR 99.94%. State-function FAR 0.24%, 18.9% lower residual σ than PCA.

Paper B — Buckingham Pi Feature Evaluation¶

Question. Among the combinatorial space of vibration-derived features, which subset is simultaneously sensitive to scour and insensitive to environmental variability across different soil conditions?

Defensive. No systematic evaluation of vibration-derived scour-detection features has been conducted across multiple soil types.
Offensive. Hundreds of SHM papers propose novel vibration features for damage detection, but the community has never asked the prior question: which of these features actually respond to scour rather than to wind speed, tidal height, or seasonal temperature? Most of the literature's "sensitive" features are merely sensitive to whatever environmental variable happens to co-vary with the damage in that particular dataset.
Constructive. A ranked feature set eliminates the combinatorial explosion in monitoring system design, directly informing what V2 should compensate, what V1 should track, and what A should fuse.

Opening. The structural health monitoring literature contains hundreds of proposed vibration features, yet no study has ever asked which of them detect scour specifically — as opposed to merely responding to the wind, waves, and temperature that dominate every offshore measurement.

Anchors. Farrar & Worden (2012), Buckingham (1914), Prendergast & Gavin (2016), Sohn et al. (2004), Li et al. (2022).

Headline. 64 features evaluated. Strain–acceleration coherence wins. Fixity ratio fails under channel isolation. 50% LOSO.

Paper A — Bayesian Scour Assessment Decision Framework¶

Question. What is the cost-optimal inspection interval for a scoured tripod foundation when capacity uncertainty from a physics-informed encoder is fused with field-monitored frequency data in a Bayesian decision framework?

Defensive. No decision framework fuses probabilistic capacity estimates with field-monitored vibration data for scour-related maintenance scheduling.
Offensive. The field builds increasingly sophisticated monitoring systems and capacity models in complete isolation, yet the only question an asset owner needs answered — inspect now or next year? — falls in the gap between these two communities. Asset owners don't lack data; they lack a probabilistic language that converts data into a decision.
Constructive. Closing the monitor-to-decision loop converts every upstream contribution (J2 power-law, J5 capacity distribution, V1 detection, B feature ranking) from an academic result into an actionable O&M policy.

Opening. The offshore wind industry invests millions in scour monitoring systems and capacity analyses that exist on separate servers, in separate departments, answering separate questions — while the only question that determines whether a turbine stays online remains unanswered: is this foundation safe enough to skip next year's inspection?

Anchors. Straub & Faber (2005), Nielsen & Sørensen (2011), Bull et al. (2025), Ritto & Rochinha (2021), Zhu & Frangopol (2016).

Headline. 3-channel Bayesian fusion, encoder prior, 97% lifecycle cost reduction vs. fixed-interval inspection.

Paper Op3 — Open-Source Coupled Analysis Framework¶

Question. Can an open-source pipeline coupling 3D limit analysis, nonlinear structural FEM, and aeroelastic simulation produce consistent scour assessments without manual data translation?

Defensive. No open-source pipeline integrates 3D geotechnical limit analysis, nonlinear structural FEM, and aeroelastic simulation for scour assessment.
Offensive. Geotechnical engineers run OptumGX, structural engineers run OpenSees, and wind engineers run OpenFAST — on three separate computers with manual CSV hand-offs — because the industry has normalised a workflow that would be considered engineering malpractice in any other discipline. A 1,794-realisation Monte Carlo under that workflow requires 5,382 manual file transfers; of course nobody does it.
Constructive. Automated per-realisation calibration makes J5 feasible at all, and guarantees cross-paper numerical consistency throughout the dissertation.

Opening. The offshore wind industry's scour assessment workflow — geotechnical, structural, and aeroelastic models running on separate machines, joined by copy-paste — is an open invitation to inconsistency that no other engineering discipline would tolerate.

Anchors. McKenna (2011), Jonkman et al. (2009), Krabbenhøft et al. (2015), Branlard et al. (2020), Dutta & Hawlader (2019).

Headline. First integrated pipeline. Per-realisation calibration at 0.2 s. 39-benchmark V&V, 92% pass rate.

Paper J5 — Probabilistic 3D Limit Analysis (1,794 Realisations)¶

Question. What is the full probability distribution of horizontal capacity and natural frequency for a tripod suction bucket foundation across the plausible range of soil conditions and scour depths?

Defensive. No probabilistic capacity assessment exists for tripod suction bucket foundations under scour, at any scale.
Offensive. Probabilistic methods have been the standard of care in structural reliability since the 1980s, yet the entire published literature on scoured tripod bucket capacity consists of deterministic point estimates that conceal a 30% capacity spread hidden in the soil variability every site investigation measures but nobody propagates.
Constructive. A population-scale capacity distribution replaces the single safety factor with a probability of failure — the input currency A requires to price inspection intervals.

Opening. Forty years after probabilistic methods became standard in structural engineering, the offshore wind industry still evaluates whether a scoured tripod foundation will fail using a single soil profile, a single scour depth, and a single safety factor — ignoring the 30% capacity spread that site investigation data already contains.

Anchors. Ukritchon et al. (1998), Fu et al. (2017), Phoon & Kulhawy (1999), DNV-RP-C212 (2019), Lacasse & Nadim (1996).

Headline. 1,794 realisations. 5th-percentile capacity 30% below deterministic estimate at 1D scour.

Paper J11 — Generalised Vesic Cavity Expansion¶

Question. How must Vesic's rigidity index be modified to account for the dissipation-weighted, three-dimensional stress field around a tripod suction bucket foundation?

Defensive. No modification of Vesic's cavity expansion theory accounts for the 3D, dissipation-weighted stress state around a multi-footed suction bucket foundation.
Offensive. Vesic's rigidity index has been applied unchanged since 1972 to geometries that violate its fundamental assumption of spherical or cylindrical symmetry — and the offshore geotechnics community has accepted this without correction for half a century.
Constructive. A corrected rigidity index provides the theoretical justification for the Winkler spring shapes in J2, grounding the entire numerical modelling chain in mechanics rather than curve-fitting.

Opening. Vesic's cavity expansion theory has underpinned bearing capacity calculations since 1972, yet its assumption of radial symmetry has never been questioned for the tripod suction bucket — a foundation whose three-footed stress field is as far from a single expanding sphere as any geometry in offshore engineering.

Anchors. Vesic (1972), Yu (2000), Houlsby & Byrne (2005), Randolph & Gourvenec (2011), Salgado & Prezzi (2007).

Paper E — Physics-Informed Encoder for Structural State¶

Question. Can a single physics-informed encoder represent the structural state of suction bucket foundations across multiple soil types without site-specific retraining?

Defensive. No encoder-based representation learning has been applied to geotechnical or OWT structural health monitoring data.
Offensive. Every SHM study treats each turbine as a unique snowflake requiring site-specific model calibration, when the structural state is governed by a small number of physics-constrained latent variables (CPT-derived stiffness, scour depth ratio, embedment ratio) that should generalise across sites without retraining. The field has confused data scarcity per site with problem-intrinsic complexity.
Constructive. A cross-soil encoder compresses J5's 1,794-realisation capacity database into a compact, transferable representation that A can query in real time rather than re-running Monte Carlo for each new site. This is the strongest confirmed gap in the portfolio.

Opening. The structural health monitoring community has spent two decades developing turbine-specific damage detection models that cannot transfer to the next wind farm, while the physics governing scour-induced degradation — soil stiffness, embedment geometry, and load path — is universal enough to encode once and deploy everywhere.

Anchors. Worden & Manson (2007), Raissi et al. (2019), Bull et al. (2021), Zhang et al. (2022), Phoon & Ching (2022).

Meta-Insight: Defensive vs Offensive Gaps¶

The current gap claims are defensive — they say "nobody has done X." That's sufficient for publication but insufficient for excellence. A reviewer accepts "nobody has done X" but doesn't get excited. What gets a reviewer excited is: "everybody assumed Y, and Y is WRONG — here's the proof."

The difference:

Defensive gap (boring-but-publishable). "No BNWF exists for suction buckets." — J2, literal.
Offensive gap (same paper, indictment framing). "The entire field uses API p-y curves calibrated for L/D > 10 on foundations with L/D ≈ 1, and here's quantitative proof of the 20% error this causes."

Same paper, same contribution, different rhetoric. The offensive version doesn't require any new experiments — it requires naming the assumption the field has been making and showing the cost of it.

The rule. Every paper should have both a defensive gap (what hasn't been done) AND an offensive gap (what's been done wrong). Three-layer framing satisfies three audiences:

Layer	Audience	Question they ask
Defensive	Journal reviewer	What is new?
Offensive	External examiner	Why should I care?
Constructive	Thesis committee	How does this fit the dissertation?

The three-layer structure reveals the dissertation's meta-narrative: the field has developed sophisticated components (scour prediction, capacity analysis, frequency monitoring, EOV compensation, decision theory) but has never connected them. The contribution is not merely 10 individual gap-fills — it is the integration architecture that makes each component useful.

What's Next¶

For each paper, the tasks are now concrete:

Lift the offensive gap into the opening paragraph. Current intros open with the defensive claim. Rewrite to lead with the indictment, then pivot to "here is the proof."
Name the wrong assumption explicitly. J2: L/D > 10 vs. L/D ≈ 1. J3: pristine-bed vs. backfilled-bed. V1: lab-only vs. operational. V2: own-dataset validation vs. common-dataset benchmark. B: "sensitive" vs. "scour-sensitive." A: separate-servers vs. closed-loop. Op3: CSV hand-off vs. integrated pipeline. J5: deterministic vs. probabilistic. J11: radial-symmetry vs. 3D stress field. E: site-specific vs. cross-site.
Quantify the cost of the wrong assumption. If I can put a number on what the field gets wrong, the opening writes itself. J5's 30% capacity spread, V2's 99.94% FAR, V1's 0.24D threshold, A's 97% cost reduction — each of these is an offensive-gap currency.
Use the logic maps as intro skeletons. Each paper's three-paragraph introduction should trace: field assumption → empirical contradiction → this paper's resolution → downstream enabler.

The next note in this series will walk through one intro rewrite end-to-end, probably starting with J2 because it's under R2 revision and has the nearest deadline.