Skip to content

Literature Review

A complete synthesis of 1,952 papers covering the dissertation topic — scour assessment of offshore wind turbine tripod suction-bucket foundations — from 1902 foundational mechanics through 2026 state of the art.

Date: 2026-04-17 (synthesis session). Source corpus: PHD/papers/literature_review/ — 1,952 Markdown files from the internal digested-literature store. Output root: PHD/papers/_shared/literature_summaries/ — 45 batch summaries, 5 domain maps, 1 master map.

The dissertation topic in one mindmap

mindmap
  root((Scour assessment<br/>OWT tripod<br/>suction buckets))
    Geotechnics
      V-H-M envelopes
      Suction caisson design
      Scour mechanics
      p-y curves and PISA
      Cyclic loading effects
    Offshore wind
      IEC 61400 / DNV
      Metocean loading
      Aeroelastic coupling
      Wake effects
      Monopile vs tripod
    SHM
      Vibration monitoring
      OMA SSI FDD
      Frequency vs mode shape
      EOV compensation
      Feature ranking
    Machine learning
      Digital twins
      Encoder models
      Physics-informed ML
      Reliability updating
      Bayesian decision
    Centrifuge
      Scaling laws
      Sand characterisation
      Tripod bucket gap
      KAIST / UWA / Cambridge
      Cyclic test protocols

No single person can read 2,000 papers across those five vocabularies in the time available. Hierarchical divide-and-conquer with parallel agents was the only viable approach.

The four-round synthesis pipeline

Each round reduces information by roughly an order of magnitude while keeping only claims that survive cross-referencing.

%%{init: {"theme":"base","themeVariables":{"fontSize":"16px","fontFamily":"Inter, system-ui, -apple-system, sans-serif","primaryTextColor":"#1a1a1a","lineColor":"#555","clusterBkg":"#fcfcfc","clusterBorder":"#cfd8dc"},"flowchart":{"nodeSpacing":30,"rankSpacing":45,"padding":16,"useMaxWidth":true}}}%%
flowchart TB
    %% ====== LAYER 0 — CORPUS ======
    subgraph L0 ["&nbsp;<b>🗃️ CORPUS · raw literature store</b>&nbsp;"]
        CORPUS[("<b>1,952 papers</b><br/><span style='font-size:13px'>literature_review/*.md<br/>1902 → 2026 · 5 communities<br/>alphabetically partitioned</span>")]:::raw
    end

    %% ====== ROUND 1 ======
    subgraph L1 ["&nbsp;<b>⚙️ ROUND 1 · parallel reading · 45 agents</b>&nbsp;"]
        direction TB
        R1Meta["<b>Prompt pattern</b> · read ~40 papers · tag domain · cross-reference · 1-page summary<br/><b>Parallel execution</b> · 45 agents · single session wall-clock"]:::meta
        R1Out["<b>📄 45 batch summaries</b><br/><span style='font-size:13px'>9 batches × 5 agents · <code>batchNN_agentM.md</code></span>"]:::r1out
        R1Meta ==> R1Out
    end

    %% ====== ROUND 2 ======
    subgraph L2 ["&nbsp;<b>🔬 ROUND 2 · domain compression · 5 agents</b>&nbsp;"]
        direction TB
        R2Meta["<b>Every agent reads all 45 summaries</b> · writes only its domain slice<br/><b>Five communities</b> · geotech · SHM · centrifuge · ML · offshore wind"]:::meta
        R2Out["<b>🗺️ 5 domain maps</b><br/><span style='font-size:13px'>established · frontiers · debates · methods · PhD relevance</span>"]:::r2out
        R2Meta ==> R2Out
    end

    %% ====== ROUND 3 ======
    subgraph L3 ["&nbsp;<b>🧭 ROUND 3 · master synthesis · 1 agent</b>&nbsp;"]
        direction TB
        R3Meta["<b>Claims kept only if they survive ≥3 of 5 domain maps</b><br/>Redundancy-verified · cross-referenced · traceable"]:::meta
        R3Out["<b>📜 1 master knowledge map</b><br/><span style='font-size:13px'>~3 200 words · 10 consensus · 5 debates · 10 gaps<br/>11×10 coverage matrix · 11 offensive framings</span>"]:::r3out
        R3Meta ==> R3Out
    end

    %% ====== ROUND 4 ======
    subgraph L4 ["&nbsp;<b>🎯 ROUND 4 · per-paper gap claims · 11 agents</b>&nbsp;"]
        direction TB
        R4Meta["<b>Each paper agent reads</b> master map + own manuscript intro<br/><b>Writes</b> one-paragraph enhanced gap claim + top-5 reading list"]:::meta
        R4Out["<b>🏹 11 enhanced gap claims</b><br/><span style='font-size:13px'>drops into each paper's intro opener<br/>provenance traceable back to original papers</span>"]:::r4out
        R4Meta ==> R4Out
    end

    %% ====== FLOW ======
    CORPUS ==>|"<b>÷ 9 batches × 5 agents</b><br/>~40 papers per slot"| R1Meta
    R1Out ==>|"<b>domain-filter + synthesise</b>"| R2Meta
    R2Out ==>|"<b>cross-reference + rank</b>"| R3Meta
    R3Out ==>|"<b>apply to each paper</b>"| R4Meta

    %% ====== METRICS RAIL ======
    subgraph Metrics ["&nbsp;📊 <b>Compression ratio</b>&nbsp;"]
        direction LR
        M0["1 952"]:::m0
        M1["45"]:::m1
        M2["5"]:::m2
        M3["1"]:::m3
        M4["11"]:::m4
        M0 -- "÷43" --> M1 -- "÷9" --> M2 -- "÷5" --> M3 -- "×11" --> M4
    end

    L0 -.-> Metrics

    click CORPUS "../../PHD/papers/literature_review/" _blank
    click R1Out "batch-summaries.md"
    click R2Out "#domain-maps"
    click R3Out "master-map.md"

    %% ====== CLASSES ======
    classDef raw fill:#eceff1,stroke:#455a64,stroke-width:2.5px,color:#263238
    classDef meta fill:#fff,stroke:#90a4ae,stroke-width:1px,stroke-dasharray:3 3,color:#455a64,font-size:13px
    classDef r1out fill:#e3f2fd,stroke:#1565c0,stroke-width:2.5px,color:#0d47a1
    classDef r2out fill:#e8f5e9,stroke:#2e7d32,stroke-width:2.5px,color:#1b5e20
    classDef r3out fill:#fff3e0,stroke:#e65100,stroke-width:3px,color:#bf360c
    classDef r4out fill:#f3e5f5,stroke:#6a1b9a,stroke-width:2.5px,color:#4a148c
    classDef m0 fill:#eceff1,stroke:#455a64,color:#263238
    classDef m1 fill:#e3f2fd,stroke:#1565c0,color:#0d47a1
    classDef m2 fill:#e8f5e9,stroke:#2e7d32,color:#1b5e20
    classDef m3 fill:#fff3e0,stroke:#e65100,color:#bf360c,stroke-width:3px
    classDef m4 fill:#f3e5f5,stroke:#6a1b9a,color:#4a148c

    %% ====== SUBGRAPH STYLES ======
    style L0 fill:#fafafa,stroke:#b0bec5,stroke-width:2px
    style L1 fill:#e3f2fd22,stroke:#64b5f6,stroke-width:2px
    style L2 fill:#e8f5e922,stroke:#81c784,stroke-width:2px
    style L3 fill:#fff3e022,stroke:#ffb74d,stroke-width:3px
    style L4 fill:#f3e5f522,stroke:#ba68c8,stroke-width:2px
    style Metrics fill:#fcfcfc,stroke:#cfd8dc,stroke-width:1px,stroke-dasharray:4 4

How to read the pipeline

  • Input scale

    1,952 papers · every Markdown file in literature_review/. Alphabetical partition into 9 batches of ~200 each.

  • Compression ratio

    1,952 → 45 → 5 → 1 master map. Each arrow is roughly an order-of-magnitude reduction, with redundancy across independent agents.

  • Verification principle

    A claim reaches the master map only if it appears in ≥3 of 5 domain maps. No single-agent assertions make it through.

  • Provenance

    Every master-map claim → domain map section → batch summary → original paper. The chain is bidirectional.

See the methodology page for the agent briefs, what each round produced, and the plan-vs-reality deltas.

What was produced

Three stacked layers, top to bottom. Every claim in a layer above is backed by the full layer below.

%%{init: {"theme":"base","themeVariables":{"fontSize":"17px","fontFamily":"Inter, system-ui, -apple-system, sans-serif","primaryTextColor":"#1a1a1a","lineColor":"#666"},"flowchart":{"nodeSpacing":35,"rankSpacing":55,"padding":18,"useMaxWidth":true}}}%%
flowchart TB
    subgraph Layer3 ["&nbsp;<b>Layer 3 · Master synthesis (1 file)</b>&nbsp;"]
        direction TB
        M["<b>Master Knowledge Map</b><br/><span style='font-size:14px'>MASTER_KNOWLEDGE_MAP.md · ~3 200 words<br/>10 consensus claims · 5 debates · 10 gaps<br/>11 × 10 coverage matrix</span>"]:::master
    end

    subgraph Layer2 ["&nbsp;<b>Layer 2 · Domain maps (5 files)</b>&nbsp;"]
        direction TB
        D1["<b>D1 · Geotechnical</b><br/><span style='font-size:14px'>170 lines</span>"]:::domain
        D2["<b>D2 · SHM</b><br/><span style='font-size:14px'>194 lines</span>"]:::domain
        D3["<b>D3 · Centrifuge</b><br/><span style='font-size:14px'>199 lines</span>"]:::domain
        D4["<b>D4 · ML + decision</b><br/><span style='font-size:14px'>181 lines</span>"]:::domain
        D5["<b>D5 · Offshore scour</b><br/><span style='font-size:14px'>204 lines</span>"]:::domain
    end

    subgraph Layer1 ["&nbsp;<b>Layer 1 · Batch summaries (45 files)</b>&nbsp;"]
        direction TB
        B["<b>9 batches × 5 agents = 45 summaries</b><br/><span style='font-size:14px'>~40 papers per agent · alphabetical partition<br/>files 1–200 · 201–400 · 401–600 · 601–800<br/>801–1000 · 1001–1200 · 1201–1400<br/>1401–1600 · 1601–1952</span>"]:::batch
    end

    Layer3 ==> Layer2 ==> Layer1

    click M "master-map.md"
    click D1 "domain-1-geotechnical.md"
    click D2 "domain-2-shm-monitoring.md"
    click D3 "domain-3-centrifuge.md"
    click D4 "domain-4-ml-decision.md"
    click D5 "domain-5-offshore-wind-scour.md"
    click B "batch-summaries.md"

    classDef master fill:#fff3e0,stroke:#e65100,stroke-width:3px,color:#e65100
    classDef domain fill:#e8f5e9,stroke:#2e7d32,stroke-width:2px,color:#1b5e20
    classDef batch fill:#e3f2fd,stroke:#1565c0,stroke-width:2px,color:#0d47a1

    style Layer3 fill:#fff8f0,stroke:#ffb74d,stroke-dasharray:5 5
    style Layer2 fill:#f1f8e9,stroke:#aed581,stroke-dasharray:5 5
    style Layer1 fill:#e1f5fe,stroke:#81d4fa,stroke-dasharray:5 5

Every domain map is backed by all 45 batch summaries (they were written with full access, not nine-per-domain); every master-map claim is cross-referenced against ≥3 of 5 domain maps. Each box above is a clickable link into the published synthesis.

One-click entry points

  • Master Knowledge Map

    10 field-consensus claims, 5 open debates, 10 verified gaps, 11×10 coverage matrix, one-sentence offensive framing per paper.

    Open master map

  • D1 — Geotechnical foundations

    V-H-M envelopes, suction caissons, scour mechanics, API p-y inadequacy, cyclic sand stiffening, centrifuge as gold standard.

    Open D1

  • D2 — SHM & scour detection

    Vibration-based SHM, frequency-shift scour detection, EOV compensation, 15+ turbine-year OMA practice.

    Open D2

  • D3 — Centrifuge experimental

    Scaling laws, model sand characterisation, tripod-bucket experimental gap, KAIST/UWA/Cambridge protocols.

    Open D3

  • D4 — ML & decision theory

    Digital twin architectures, autoencoder representation learning, Bayesian decision for maintenance, reliability updating.

    Open D4

  • D5 — Offshore wind scour

    Metocean coupling, wake-scour interaction, DNV 1.3D controversy, countermeasures, monitoring practice.

    Open D5

  • Round 1 batch summaries (45)

    Raw reading log — one page per agent per batch. Use for tracing a master-map claim back to original papers.

    Open index

  • Methodology

    Agent briefs, round-by-round structure, plan-vs-reality deltas, provenance chain from claim to paper.

    Open methodology

The master map at a glance

mindmap
  root((Master<br/>Knowledge<br/>Map))
    10 consensus claims
      V-H-M envelopes govern
      SSI controls OWT frequency
      Monitored f 5–15% > design
      Scour degrades stiffness
      Global scour > local
      Centrifuge is gold standard
      Cyclic stiffens sand
      PISA > API p-y
      ±180% fatigue variation
      OMA SSI/FDD dominant
    5 open debates
      Frequency vs mode shape
      Local vs general scour
      Physics vs data DT
      Constitutive for cyclic SSI
      10-min DLC for fatigue
    10 verified gaps
      G1 scour + cyclic critical
      G2 bucket scour critical
      G3 field OWT scour critical
      G4 closed-loop DT high
      G5 probabilistic capacity
      G6 silty mixed soils
      G7 prototype validation
      G8 EOV benchmark
      G9 encoder for SHM
      G10 unified framework
    6 dissertation firsts
      Centrifuge tripod scour
      Field tripod scour
      Probabilistic tripod capacity
      EOV benchmark on OWT
      Cross-soil encoder
      Bayesian decision loop

See the Master Knowledge Map for each node elaborated with anchor citations.

Portfolio coverage at a glance

Every one of the 11 papers addresses at least one verified gap. Gaps are grouped by severity; each gap lists the papers that contribute directly to it. G2 (scour on suction buckets) is the dissertation's centre of gravity — five of the eleven papers contribute.

%%{init: {"theme":"base","themeVariables":{"fontSize":"16px","fontFamily":"Inter, system-ui, -apple-system, sans-serif","primaryTextColor":"#1a1a1a","lineColor":"#666"},"flowchart":{"nodeSpacing":30,"rankSpacing":45,"padding":14,"useMaxWidth":true}}}%%
flowchart TB
    subgraph Crit ["&nbsp;🔴 CRITICAL gaps&nbsp;"]
        direction TB
        G1["<b>G1 · scour + cyclic loading</b><br/><span style='font-size:14px'>single-paper coverage</span>"]:::crit
        G1 --> J3c(["J3"]):::paper

        G2["<b>G2 · scour on suction buckets</b><br/><span style='font-size:14px'>centre of gravity · 5 papers</span>"]:::crit
        G2 --> J1c(["J1"]):::paperPub & J2c(["J2"]):::paper & J3c2(["J3"]):::paper & J5c(["J5"]):::paper & J11c(["J11"]):::paper

        G3["<b>G3 · long-term field OWT SHM</b><br/><span style='font-size:14px'>field-evidence gap</span>"]:::crit
        G3 --> V1c(["V1"]):::paper & V2c(["V2"]):::paper
    end

    subgraph High ["&nbsp;🟠 HIGH-severity gaps&nbsp;"]
        direction TB
        G4["<b>G4 · closed-loop digital twin</b>"]:::high --> Ah(["A"]):::paper
        G5["<b>G5 · probabilistic capacity</b>"]:::high --> J5h(["J5"]):::paper & Ah2(["A"]):::paper
        G6["<b>G6 · silty and mixed soils</b>"]:::high --> J1h(["J1"]):::paperPub & J3h(["J3"]):::paper
        G7["<b>G7 · prototype-scale validation</b>"]:::high --> V1h(["V1"]):::paper
    end

    subgraph Med ["&nbsp;🟣 MEDIUM-severity gaps&nbsp;"]
        direction TB
        G8["<b>G8 · EOV compensation benchmark</b>"]:::med --> V2m(["V2"]):::paper & Bm(["B"]):::paper
        G9["<b>G9 · SHM encoder representation</b>"]:::med --> Em(["E"]):::paper
        G10["<b>G10 · unified software framework</b>"]:::med --> Op3m(["Op3"]):::paper
    end

    Crit ==> High ==> Med

    classDef crit fill:#ffebee,stroke:#c62828,stroke-width:3px,color:#b71c1c
    classDef high fill:#fff3e0,stroke:#ef6c00,stroke-width:2px,color:#e65100
    classDef med fill:#f3e5f5,stroke:#7b1fa2,stroke-width:2px,color:#4a148c
    classDef paper fill:#fff,stroke:#333,stroke-width:1.5px,color:#222
    classDef paperPub fill:#fff,stroke:#2e7d32,stroke-width:2.5px,color:#1b5e20

    style Crit fill:#fff5f5,stroke:#ffcdd2,stroke-dasharray:5 5
    style High fill:#fffaf0,stroke:#ffe0b2,stroke-dasharray:5 5
    style Med fill:#faf5ff,stroke:#e1bee7,stroke-dasharray:5 5

Reading the chart: bold outlined paper circles are published (J1). Everything else is in submission or in preparation. Stack order (top → bottom) reflects severity. The full text of each gap appears in the Master Knowledge Map §3.

Six "firsts" the portfolio contributes

First Paper Gap it closes
Centrifuge scour-frequency dataset for tripod buckets J1 G2
Field scour-frequency dataset on operational OWT tripod V1 G3
Probabilistic capacity ensemble for scoured tripods J5 G5
Benchmark of EOV compensation methods on OWT data V2 G8
Physics-informed encoder for cross-soil state E G9
Closed-loop Bayesian decision for scour maintenance A G4

Offensive gap framings — one sentence per paper

Round 4 produces a sentence per paper that positions it against the field rather than against whatever the author happened to have read.

Paper One-sentence indictment
J1 Three decades perfecting monopile scour prediction, zero centrifuge datasets on tripod buckets.
J2 Scour-stiffness literature treats foundations as isolated columns; tripods redistribute through three footings.
J3 Every scour experiment prepares pristine beds and excavates once; reality backfills with loose sediment.
J5 Probabilistic reliability is 40 years old, yet tripod-bucket scour literature is deterministic point estimates.
J11 Vesic 1972 cavity expansion used unchanged, despite tripod geometry bearing no resemblance to a sphere.
V1 Dozens of numerical demonstrations, zero field records linking scour to frequency on an operational OWT.
V2 Four EOV methods advocated in parallel literatures with no shared benchmark.
E Every SHM paper treats each turbine as unique; a physics-informed encoder can generalise across soils.
A Monitoring and capacity built in parallel but never closed into the loop that answers "inspect today or next year?".
B Hundreds of feature proposals; no systematic ranking of which ones detect scour vs react to environment.
Op3 Three tool chains, three computers, manual CSV hand-offs — no open-source pipeline connects them.

Source access

Everything is traceable back to the originals:

  • Raw literature corpus: F:\TREE_OF_THOUGHT\PHD\papers\literature_review\ (1,952 Markdown files).
  • Synthesis outputs: F:\TREE_OF_THOUGHT\PHD\papers\_shared\literature_summaries\ and F:\TREE_OF_THOUGHT\PHD\papers\_shared\MASTER_KNOWLEDGE_MAP.md. (Published here as the Master Knowledge Map and the five domain maps.)
  • Reading log per batch: Batch summaries index.