--- license: cc-by-nc-4.0 task_categories: - tabular-classification - tabular-regression - time-series-forecasting language: - en tags: - synthetic - oil-and-gas - midstream - pipeline - corrosion - cathodic-protection - nace-sp0169 - rbi - integrity-management - xpertsystems pretty_name: "OIL-027 — Synthetic Pipeline Corrosion Dataset (Sample)" size_categories: - 100K15 mpy triggers RBI high-risk classification per API 581) | | 4 | avg cp voltage v | -0.9440 | -0.95 | ±0.2 | ✓ PASS | NACE SP0169 (External Corrosion Control) — typical cathodic protection potential for buried pipelines (-0.85V vs Cu/CuSO4 minimum protection criterion; -1.0 to -1.2V for fully-protected pipelines) | | 5 | avg soil resistivity ohm cm | 2567.9367 | 2500.0 | ±1000.0 | ✓ PASS | NACE TM0497 (Soil Resistivity Measurement) + NACE SP0169 — typical soil resistivity for mixed onshore/offshore/subsea/arctic portfolio (100-5000 ohm-cm range; <1000 ohm-cm is highly corrosive per NACE classification) | | 6 | avg coating health pct | 79.8284 | 80.0 | ±8.0 | ✓ PASS | API 1163 (In-Line Inspection Systems) + NACE SP0502 (External Corrosion Direct Assessment) — typical coating health for mid-life transmission pipelines (70-90% typical; FBE coating degrades 2-4% per decade) | | 7 | water cut corrosion pearson correlation | 0.3275 | 0.3 | ±0.12 | ✓ PASS | de Waard & Milliams (1991) CO2 corrosion prediction model — expected positive correlation between water cut and corrosion rate (generator formula: rate = base × temp_factor × (1 + water_cut/100); within-pipeline coupling deterministic, cross-pipeline base-rate variation dilutes global correlation). Validates water-conditioned corrosion physics. | | 8 | cp voltage external risk pearson correlation | 0.6422 | 0.55 | ±0.15 | ✓ PASS | NACE SP0169 -0.85V cathodic protection criterion — expected strong positive correlation between CP voltage (less negative) and external corrosion risk numeric (LOW=0/MEDIUM=1/HIGH=2). Validates generator's physics-gated risk classification per NACE. | | 9 | soil resistivity external risk pearson correlation | -0.3809 | -0.3 | ±0.15 | ✓ PASS | NACE SP0169 + NACE TM0497 soil corrosivity classification — expected inverse correlation between soil resistivity and external corrosion risk numeric (low resistivity soil drives high corrosion). Validates generator's NACE-anchored gating. | | 10 | material grade diversity entropy | 0.9998 | 0.96 | ±0.04 | ✓ PASS | API 5L Line Pipe material grade taxonomy (X52, X65, X70) — 3-class diversity benchmark for mixed transmission portfolio per API 5L specification + PHMSA pipeline inventory, normalized Shannon entropy | **Overall: 100.0/100 — Grade A+** (10 PASS · 0 MARGINAL · 0 FAIL of 10 metrics) --- ## Schema highlights **`pipeline_assets.csv`** — 3 × 4 portfolio matrix per **API 5L + PHMSA**: | Material Grade | Use Case | Specified Min Yield Strength (psi) | |---|---|---:| | API 5L X52 | Older transmission / gathering | 52,000 | | API 5L X65 | Modern transmission mainline | 65,000 | | API 5L X70 | High-pressure transmission | 70,000 | 4 environment classes per **NACE SP0169 classification**: | Environment | Corrosion Drivers | |---|---| | Onshore | Soil chemistry, stray current, AC interference | | Offshore | Splash zone, marine atmosphere, oxygen ingress | | Subsea | Cathodic protection mandatory, biofouling | | Arctic | Permafrost cycling, ice gouging, low temperature stress | **`internal_corrosion.csv`** — **de Waard-Milliams (1991) water-cut coupling**: > corrosion_rate_mpy = base_rate × temp_factor × (1 + water_cut/100) > wall_loss_pct = corrosion_rate × day/365 × 0.1 (deterministic accumulation) The sample's water cut ↔ corrosion rate Pearson correlation is r ≈ **+0.33** — **moderate positive coupling validates de Waard-Milliams water-cut physics** (within-pipeline coupling is deterministic; cross-pipeline base- rate variability dilutes the global correlation). **`external_corrosion.csv`** — **NACE SP0169 -0.85V cathodic protection threshold gating**: > external_corrosion_risk = HIGH if (soil_resistivity < 800 OR cp_voltage > -0.85) > external_corrosion_risk = MEDIUM if soil_resistivity < 1500 > external_corrosion_risk = LOW otherwise The sample's CP voltage ↔ external risk Pearson correlation is r ≈ **+0.64** — **strong positive coupling validates NACE SP0169 -0.85V cathodic protection criterion physics**. Risk class distribution: **49% LOW / 10% MEDIUM / 41% HIGH** — meaningful 3-class diversity for ML class-balancing (better than degenerate single-class outcomes). **`pitting_profiles.csv`** — per-pipeline pit catalog per **ASTM G46 pit density standard**: > n_pits_per_pipeline = U(2, 9) > pit_depth_mm = U(0.5, 12) > pit_width_mm = U(1, 25) > growth_rate_mm_year = U(0.05, 1.5) --- ## Suggested use cases 1. **Internal corrosion rate regression** — predict `corrosion_rate_mpy` from water cut + environment features per de Waard-Milliams (1991). **Strong physics signal** within-pipeline; moderate global. 2. **External corrosion risk classification** — 3-class classifier on `external_corrosion_risk` from soil resistivity + CP voltage + coating features per NACE SP0169. **Strong physics**: CP↔risk r ≈ +0.64. 3. **Wall loss time-series forecasting** — predict accumulated `wall_loss_pct` over 180-day horizon per API 510 remaining life calculations. 4. **Pit growth rate regression** — predict `growth_rate_mm_year` from pit depth + width features per ASTM G46. 5. **Cathodic protection optimization** — predict CP voltage thresholds from soil resistivity + coating health features per NACE SP0169. 6. **Integrity grade classification** — 4-class classifier on `integrity_grade` (LOW/MEDIUM/HIGH/CRITICAL). **Note**: integrity labels in this sample are not feature-coupled (see Honest Disclosure §1). 7. **Multi-table relational ML** — entity-resolution learning across the 5 tables via `pipeline_id`. Join asset metadata with corrosion time-series for feature-rich ML pipelines. --- ## Loading ```python from datasets import load_dataset ds = load_dataset("xpertsystems/oil027-sample", data_files="internal_corrosion.csv") print(ds["train"][0]) ``` Or with pandas: ```python import pandas as pd assets = pd.read_csv("hf://datasets/xpertsystems/oil027-sample/pipeline_assets.csv") internal = pd.read_csv("hf://datasets/xpertsystems/oil027-sample/internal_corrosion.csv") external = pd.read_csv("hf://datasets/xpertsystems/oil027-sample/external_corrosion.csv") pits = pd.read_csv("hf://datasets/xpertsystems/oil027-sample/pitting_profiles.csv") labels = pd.read_csv("hf://datasets/xpertsystems/oil027-sample/integrity_labels.csv") # Multi-table corrosion feature engineering: corr_avg = internal.groupby("pipeline_id")["corrosion_rate_mpy"].mean().reset_index() joined = (assets .merge(corr_avg, on="pipeline_id") .merge(external, on="pipeline_id") .merge(labels, on="pipeline_id")) # Now you have water_cut + corrosion_rate + CP voltage + integrity_grade in one frame ``` --- ## Reproducibility All generation is deterministic via the integer `seed` parameter (driving `np.random.seed` and `random.seed`). A seed sweep across `[42, 7, 123, 2024, 99, 1]` confirms Grade A+ on every seed in this sample. --- ## Honest disclosure of sample-scale limitations This is a **sample** product calibrated for pipeline corrosion ML research, not for live operational decisions. **Several important limitations should be understood before use:** 1. **Integrity grade is NOT feature-coupled.** The 4-class `integrity_grade` label is sampled from `np.random.uniform(0, 1)` with threshold gates, without any coupling to corrosion_rate, wall_loss, pitting depth, or external_corrosion_risk. **The sample's integrity_grade↔corrosion_rate correlation is r ≈ +0.04 — essentially noise.** For integrity ML, train on labels you derive yourself from the physics features (e.g., `wall_loss_pct > 30% → CRITICAL`) rather than the provided integrity_grade. The full product (v1.1) will implement feature-coupled integrity grading per API 580 RBI methodology. 2. **Remaining life is NOT physics-computed.** The `remaining_life_days` field is sampled from `np.random.uniform(180, 7200)` and is **not computed from corrosion rate or wall loss per API 510 RBI**. Real remaining life = (current_wall_thickness - retirement_limit) / corrosion_rate. For RUL ML, **derive remaining life from internal_ corrosion features** rather than using the provided field. 3. **CO2 and H2S do not drive corrosion rate.** The pipeline_assets table includes `co2_pct` and `h2s_ppm` fields, but **neither is used in the corrosion rate calculation**. Real CO2 corrosion follows de Waard 1991 `rate ∝ partial_pressure_CO2^0.67 × temp^1.41` and real H2S service triggers NACE MR0175 / ISO 15156 sulfide stress cracking thresholds. **CO2 ↔ corrosion rate r ≈ +0.08 in sample (not de Waard physics)**. Full product v1.1 will implement de Waard 1991 CO2 corrosion model and MR0175 H2S service classifications. 4. **Pitting growth rate is NOT coupled to environment.** Pit growth rates are uniformly sampled `U(0.05, 1.5) mm/year` without coupling to internal corrosion rate, environment class, or external_corrosion_risk. Real pit growth follows ASTM G46 + environment-specific kinetics. For pit-growth ML, **filter to environment subsets and treat growth rate as residual variance** rather than predictable from physics. 5. **Temperature varies only by noise factor.** The `temperature_f` field in internal_corrosion is sampled per-row `U(70, 180)` without conditioning on pipeline service or seasonality. Real pipeline temperature tracks ambient + fluid-source temperature with strong seasonal cycles. **For temperature-conditioned corrosion ML, use temperature as a noisy random feature**, not a true operational signal. 6. **Pressure is independent of MAOP.** The `pressure_psi` field is sampled `U(500, 2200)` without coupling to material grade or design MAOP. Real operating pressures are typically 60-80% of MAOP. For pressure-conditioned corrosion ML, **filter to realistic MAOP-conditioned operating ranges**. 7. **Coating health is not age-coupled.** The `coating_health_pct` field is sampled `U(60, 100)` without conditioning on pipeline age. Real FBE coating health degrades 2-4% per decade per NACE SP0502 direct assessment. **For coating-degradation ML, the sample is uniform-prior over coating quality** rather than age-conditioned. 8. **Pipeline age is independent of all features.** The `pipeline_age_years` field is sampled `U(1, 40)` without coupling to material grade (older pipelines were X42/X52, modern are X65/X70 per API 5L history). For age-conditioned ML, **expect age to be uncoupled from material choice** at sample scale. --- ## Where physics IS strong (use these for ML) Three coupling signals in this sample are **physically valid and ML-useful**: | Coupling | Pearson r | Physics | Use For | |---|---:|---|---| | **Water cut → corrosion rate** | +0.33 | de Waard-Milliams (1991) water-cut formula | Within-pipeline corrosion ML | | **CP voltage → external risk** | +0.64 | NACE SP0169 -0.85V criterion | External corrosion risk classification | | **Soil resistivity → external risk** | -0.38 | NACE TM0497 + SP0169 soil corrosivity | External corrosion risk classification | | **Wall loss accumulation** | deterministic | Time-integrated corrosion rate | RUL forecasting | --- ## Cross-references to other XpertSystems OIL SKUs This SKU is the **fourth midstream SKU** in the catalog — specializing in **corrosion physics** complementing the leak detection trilogy: | SKU | Layer | Focus | |---|---|---| | OIL-015 | Midstream | Pipeline flow assurance (wax / hydrate / asphaltene threshold gating) | | OIL-024 | Midstream | Full pipeline hydraulics + SCADA + 15 transient events | | OIL-025 | Midstream | Leak detection + rupture prediction + acoustic + CPM | | **OIL-027** | **Midstream** | **Corrosion progression + cathodic protection + pitting + integrity** *(this SKU)* | **OIL-027 vs OIL-025**: OIL-025 simulates **event-centric leak/rupture detection** with acoustic + pressure wave physics. **OIL-027 simulates continuous corrosion progression** (180-day time series) driving the underlying integrity degradation. Use OIL-025 for **leak detection ML**, OIL-027 for **corrosion progression + cathodic protection ML**. **OIL-027 vs OIL-022**: OIL-022 simulates **refinery vessel/piping corrosion** (RBI + inspection findings). **OIL-027 simulates pipeline external + internal corrosion** with cathodic protection physics specific to buried/subsea pipelines. Use OIL-022 for **vessel inspection planning**, OIL-027 for **pipeline corrosion ML**. --- ## Full product The **full OIL-027 dataset** ships at **15,000 pipelines × 730-day daily corrosion progression** (prod mode) producing tens of millions of rows with **feature-coupled integrity grades per API 580 RBI methodology**, **physics-computed remaining life per API 510**, **de Waard 1991 CO2 corrosion model with partial-pressure conditioning**, **NACE MR0175 / ISO 15156 H2S service classifications**, **environment-conditioned pit growth kinetics per ASTM G46**, **age-coupled coating degradation per NACE SP0502**, and **MAOP-conditioned operating pressures** — licensed commercially. Contact XpertSystems.ai for licensing terms. 📧 **pradeep@xpertsystems.ai** 🌐 **https://xpertsystems.ai** --- ## Citation ```bibtex @dataset{xpertsystems_oil027_sample_2026, title = {OIL-027: Synthetic Pipeline Corrosion Dataset (Sample)}, author = {XpertSystems.ai}, year = {2026}, url = {https://huggingface.co/datasets/xpertsystems/oil027-sample} } ``` ## Generation details - Sample version : 1.0.0 - Random seed : 42 - Generated : 2026-05-23 00:15:44 UTC - Pipelines : 1300 - Simulation days : 180 (daily corrosion progression per pipeline) - Material grades : 3 (API 5L X52, X65, X70) - Environments : 4 (Onshore, Offshore, Subsea, Arctic) - Risk classes : 3 (LOW, MEDIUM, HIGH) per NACE SP0169 - Integrity grades : 4 (LOW, MEDIUM, HIGH, CRITICAL) — sample-scale random - Calibration basis : NACE SP0169, NACE MR0175 / ISO 15156, NACE TM0497, de Waard & Milliams (1991), API 510, API 570, API 580/581, API 5L, ASME B31.4/B31.8, PHMSA 49 CFR 195, NACE SP0502, API 1163, ASTM G1, ASTM G46 - Overall validation: 100.0/100 — Grade A+