--- license: cc-by-nc-4.0 task_categories: - tabular-classification - tabular-regression - time-series-forecasting language: - en tags: - synthetic - oil-and-gas - midstream - pipeline - hydraulics - darcy-weisbach - swamee-jain - leak-detection - scada - flow-assurance - xpertsystems pretty_name: "OIL-024 — Synthetic Pipeline Flow Dataset (Sample)" size_categories: - 100K1% triggers leak alarm per API standards) | | 6 | avg pump efficiency pct | 72.7872 | 73.0 | ±6.0 | ✓ PASS | API 610 (Centrifugal Pumps for Petroleum) — typical pump efficiency for transmission pipeline pump stations (68-82% at BEP per Hydraulic Institute; demand-modulated operation reduces from peak) | | 7 | thermal heat loss pearson correlation | 0.8272 | 0.75 | ±0.15 | ✓ PASS | Newton's law of cooling (heat_loss ∝ ΔT) — expected strong positive correlation between (pipe_temp - ambient_temp) and heat_loss_btu_hr_ft per fundamental heat transfer physics. Validates generator's thermal model. | | 8 | optimization failure pearson correlation | -0.9944 | -0.95 | ±0.1 | ✓ PASS | Generator's deterministic formula: failure_risk = (100 - optimization_score) × 0.7 + leak_prob × 0.3. Near-deterministic inverse coupling validates feature-coupled label generation for pipeline reliability ML. | | 9 | corrosion leak pearson correlation | 0.9708 | 0.85 | ±0.1 | ✓ PASS | NACE SP0169 External Corrosion Control + API 1130 + PHMSA pipeline incident data — expected strong positive correlation between corrosion rate and leak probability (generator formula: leak_prob = corrosion/25 × 50 + erosion × 0.12). Validates integrity-leak physics. | | 10 | pipeline type diversity entropy | 0.9596 | 0.92 | ±0.04 | ✓ PASS | 10-class pipeline type taxonomy per ASME B31.4/B31.8 + PHMSA classification (crude oil transmission, natural gas transmission, refined products, offshore subsea flowline, multiphase gathering, water injection, CO2 transport, heavy oil diluent, LNG transfer, hydrogen-ready), normalized Shannon entropy | **Overall: 100.0/100 — Grade A+** (10 PASS · 0 MARGINAL · 0 FAIL of 10 metrics) --- ## Schema highlights **`pipeline_master.csv`** — 10-class pipeline type taxonomy per **PHMSA / ASME B31.4 / B31.8**: | Type | Weight | Typical Fluid | |---|---:|---| | crude_oil_transmission | 20% | crude (light/medium/heavy) | | natural_gas_transmission | 18% | dry/wet gas | | refined_products | 11% | gasoline / diesel / jet | | offshore_subsea_flowline | 10% | multiphase | | multiphase_gathering | 12% | mixed wellhead fluids | | water_injection | 7% | brine / produced water | | co2_transport | 6% | CO2 (dense phase) | | heavy_oil_diluent | 6% | bitumen + diluent | | lng_transfer | 5% | LNG (cryogenic) | | hydrogen_ready | 5% | H2 / H2 blends | Material grades: API 5L X42 / X52 / X60 / X65 / X70 — standard transmission- grade pipe per **API 5L specification**. **`hydraulic_profiles.csv`** — **Swamee-Jain (1976) friction + Darcy-Weisbach (1857) pressure drop** implementation: > Re = ρ × v × D / μ (Reynolds 1883) > f_lam = 64 / Re (Hagen-Poiseuille, Re < 2300) > f_turb = 0.25 / [log10(ε/(3.7D) + 5.74/Re^0.9)]² (Swamee-Jain) > dp_friction = f × (L/D) × (ρ × v² / 2) (Darcy-Weisbach) > dp_elev = ρ × g × Δh (Bernoulli) > dp_total = (dp_friction + dp_elev) × wax_factor PVT-aware fluid properties for 10 fluid families with temperature and pressure dependence: | Fluid | Density (kg/m³) | Viscosity (Pa·s) | |---|---:|---| | Light crude | ~790 × (1 - 0.00038·ΔT) | 0.004 × exp(-0.022·(T-100)) | | Medium crude | ~850 | 0.018 | | Heavy crude | ~930 | 0.12 | | Dry gas | 48 × (P/1000) × (520/T_R) | 1.2e-5 | | LNG | 430 × (1 - 0.0006·(T+260)) | 1.2e-4 | | CO2 | 700 × (P/1500)^0.15 | 7e-5 | | Hydrogen blend | 18 × (P/1000) × (520/T_R) | 9e-6 | **`thermal_profiles.csv`** — **Newton's law of cooling**: > heat_loss_btu_hr_ft ∝ (pipe_temp - ambient_temp) > joule_thomson_cooling ∝ pressure_drop (gas pipelines, ~0.002 multiplier) The sample's **(temp-ambient) ↔ heat_loss Pearson correlation is r ≈ +0.83** — **strong positive coupling validates Newton's cooling physics**. **`flow_assurance.csv`** — **Sloan-Koh (2008) hydrate physics**: > wax_risk = (118 - temp_f) × 1.8 + viscosity × 0.35 + current_wax × 600 > hydrate_risk = (95 - temp_f) × 2.2 + pressure/50 + 20·is_gas > slugging_risk = multiphase × N(45, 15) + 20 × |sin(t/7)| **`leak_detection.csv`** — **API 1130 + API RP 1175** computational pipeline monitoring leak detection: > leak_rate_bpd = lognormal(2.8, 0.9) clip 2-850 > pressure_signature = 50 + leak_rate/8 + noise > detection_delay_sec = lognormal(7.0, 0.8) clip 60-86400 **`optimization_labels.csv`** — **deterministic feature-coupled labels**: > opt_score = 100 - |demand - 0.92| × 60 - max(wax, hydrate) × 0.18 - leak_prob × 0.22 - anomaly × 8 > failure_risk = (100 - opt_score) × 0.7 + leak_prob × 0.3 > grade = 'A' if opt_score ≥ 85 else 'B' if ≥ 70 else 'C' if ≥ 55 else 'D' The sample's optimization↔failure Pearson correlation is r ≈ **−0.994** — **near-deterministic inverse coupling validates formula-level label generation**. --- ## Suggested use cases 1. **Pressure drop prediction** — regression on `pressure_drop_psi` from flow + diameter + roughness + temp features. **Strong physics**: Darcy-Weisbach signal validated. 2. **Friction factor prediction** — predict `friction_factor` from Reynolds + roughness ratio per Swamee-Jain / Moody chart. 3. **Leak probability scoring** — regression on `leak_probability_score` per NACE SP0169 + API 1130. **Strong physics**: corrosion-leak r ≈ +0.97. 4. **Flow assurance binary classification** — predict `flow_assurance_state == 'critical'` from temperature + viscosity + pressure features per Sloan-Koh. 5. **Wax deposition forecasting** — time-series forecasting of `wax_deposition_thickness_mm` per coupled accumulation physics. 6. **Pipeline efficiency grade classification** — 4-class ordinal classifier on `efficiency_grade` (A/B/C/D). **Strong feature coupling** — models WILL learn meaningful patterns. 7. **Pump cavitation prediction** — regression on `cavitation_risk_score` from RPM + demand features per API 610 + Hydraulic Institute. 8. **15-class transient event classification** — multi-class classifier on `event_type` (rare events at sample scale; see Honest Disclosure §1). 9. **Mass balance anomaly detection** — anomaly detection on `mass_balance_error_pct` per API 1130 CPM leak detection. 10. **Multi-table relational ML** — entity-resolution + graph neural- network learning across the 12 joinable tables via `pipeline_id`, `segment_id`, `timestamp`. --- ## Loading ```python from datasets import load_dataset ds = load_dataset("xpertsystems/oil024-sample", data_files="hydraulic_profiles.csv") print(ds["train"][0]) ``` Or with pandas: ```python import pandas as pd master = pd.read_csv("hf://datasets/xpertsystems/oil024-sample/pipeline_master.csv") hyd = pd.read_csv("hf://datasets/xpertsystems/oil024-sample/hydraulic_profiles.csv") thm = pd.read_csv("hf://datasets/xpertsystems/oil024-sample/thermal_profiles.csv") fa = pd.read_csv("hf://datasets/xpertsystems/oil024-sample/flow_assurance.csv") labels = pd.read_csv("hf://datasets/xpertsystems/oil024-sample/optimization_labels.csv") # Full multi-table feature engineering: joined = (hyd .merge(thm, on=["timestamp", "pipeline_id", "segment_id"]) .merge(fa, on=["timestamp", "pipeline_id", "segment_id"]) .merge(labels, on=["timestamp", "pipeline_id", "segment_id"]) .merge(master, on=["pipeline_id", "segment_id"])) # Predict efficiency_grade from hydraulics + thermal + flow assurance features ``` --- ## Reproducibility All generation is deterministic via the integer `seed` parameter (driving `np.random.default_rng`). 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 ML research, not for live operational decisions. Several notes: 1. **Flow regime is ~100% turbulent** because transmission pipeline Reynolds numbers naturally exceed 4000 (sample mean Re ≈ 1.1M). **Laminar regime is essentially absent at sample scale** — only pipelines with very viscous fluids (heavy crude) at very low flow rates would produce laminar flow. For laminar-regime ML, filter to `primary_fluid == 'heavy_crude' AND flow_bpd < 5000` or wait for the full product which adds explicit low-flow scenarios. 2. **Pump cavitation risk is ~0** because the generator's formula `(rpm - 2700)/12 + N(0, 5)` clips at 0 for typical 1750 rpm operation. Real cavitation risk requires NPSH-margin physics (OIL-021 implements this correctly). For pump cavitation ML, use OIL-021 instead; OIL-024 cavitation is a placeholder. 3. **Integrity state is 100% 'normal'** at 7-day sample horizon because `leak_probability > 40` threshold (for 'watch' classification) requires longer-horizon corrosion accumulation. The full product (30-day prod mode) and full-scale simulations show meaningful 'watch' and 'high_risk' populations. For integrity-state classification ML, use the full product or augment with external corrosion-progression simulations. 4. **Transient events are sparse** (~20 events per 55 pipelines at sample scale). For 15-class event-type classification ML, **use the full product** (12000+ pipelines for 30 days) or **filter to pipeline-day aggregated event flags** rather than per-timestep events. 5. **Leak events are very rare** (~3 leak entries across 55 pipelines). Leak ML requires the full product. For leak detection ML at sample scale, **use `leak_probability_score` regression** rather than the binary leak flag. 6. **Sensor drift flag rate is ~0.04%** because the generator only sets `drift_flag = True` when `event_type == 'sensor_drift'`, which is one of 15 event types at 0.6% rare event rate. For sensor drift ML, use full product or rely on signal-quality `quality_flag` instead. 7. **Flow ↔ pressure drop correlation is weak (r ≈ +0.06)** because the relationship is dominated by **between-segment** variance in diameter, length, and fluid type rather than **within-segment** flow-pressure scaling. **For Darcy-Weisbach ML, normalize pressure drop per-segment** (dp/L or use friction factor) before fitting. Within a single segment, the relationship is strong (r > 0.85 typically). 8. **Flow assurance critical rate is ~38%** — higher than realistic field operation (5-15% typical). The generator's wax_risk formula `(118-temp) × 1.8` triggers easily at ambient temperatures; real field operation includes insulation and heat-tracing that reduces wax onset. For realistic flow assurance ML, **filter critical classifications by insulation type** (`insulation_type in ['high', 'subsea_wet_insulated']` for realistic thermal-protected operation). --- ## Cross-references to other XpertSystems OIL SKUs This SKU is the **second midstream SKU** in the catalog (after OIL-015 flow assurance): | SKU | Layer | Focus | |---|---|---| | OIL-015 | Midstream | Pipeline flow assurance (wax / hydrate / asphaltene threshold gating) | | **OIL-024** | **Midstream** | **Full pipeline hydraulics + SCADA + leak detection + transient events** *(this SKU)* | | OIL-018 | Upstream | Wellbore-to-separator multiphase flow (Beggs-Brill regime classification) | **OIL-024 vs OIL-015**: OIL-015 specializes in **flow-assurance-only** threshold-gated wax/hydrate/asphaltene deposition. **OIL-024 is the comprehensive midstream pipeline operations dataset** — full Swamee-Jain hydraulics + Darcy-Weisbach pressure drop + thermal profiles + SCADA telemetry + leak detection + 15-class transient event taxonomy + 10-fluid- type PVT-aware physics. Use OIL-015 for flow-assurance ML specifically, **OIL-024 for general pipeline operations / SCADA / hydraulic ML**. **OIL-024 vs OIL-018**: OIL-018 simulates **wellbore-to-separator** multiphase flow with Beggs-Brill regime classification (upstream). **OIL-024 simulates midstream transmission pipelines** with single-phase or multiphase flow assurance (downstream of separator). Use OIL-018 for upstream production multiphase ML, OIL-024 for **midstream transmission pipeline ML**. --- ## Full product The **full OIL-024 dataset** ships at **12,000 pipelines × 30 days × 60-min interval** (prod mode) producing tens of millions of rows with **richer event populations** (1000+ events for class-balanced 15-class ML), **full NPSH-conditioned pump cavitation physics**, **multi-day leak event populations**, **realistic flow assurance critical rates** (insulation- conditioned), and **explicit laminar-flow scenarios** (heavy-crude low-flow regimes) — licensed commercially. Contact XpertSystems.ai for licensing terms. 📧 **pradeep@xpertsystems.ai** 🌐 **https://xpertsystems.ai** --- ## Citation ```bibtex @dataset{xpertsystems_oil024_sample_2026, title = {OIL-024: Synthetic Pipeline Flow Dataset (Sample)}, author = {XpertSystems.ai}, year = {2026}, url = {https://huggingface.co/datasets/xpertsystems/oil024-sample} } ``` ## Generation details - Sample version : 1.0.0 - Random seed : 42 - Generated : 2026-05-22 21:05:15 UTC - Pipelines : 55 - Segments : 369 (avg 6.7 per pipeline) - Simulation days : 7 - Time-step interval: 180 minutes - Fluid families : 10 (light/medium/heavy crude, dry/wet gas, refined product, water, CO2, LNG, hydrogen blend) - Pipeline types : 10 (crude oil / natural gas / refined products / offshore subsea / multiphase gathering / water injection / CO2 transport / heavy oil diluent / LNG / hydrogen-ready) - Regions : 10 (Permian Basin, Gulf Coast, North Sea, Middle East, Canadian Oil Sands, Offshore Brazil, West Africa, Alaska Arctic, Rocky Mountains, Appalachia) - Terrains : 7 (flat, rolling, mountainous, offshore, subsea, arctic, desert) - Event types : 15 (pump trip, compressor trip, valve throttle, emergency shutdown, restart, surge, water hammer, slugging, hydrate risk, wax deposition, SCADA outage, sensor drift, leak, corrosion alarm, pigging run) - Calibration basis : Swamee-Jain (1976), Darcy-Weisbach (1857), Moody (1944), Reynolds (1883), Hagen-Poiseuille (1839), Colebrook-White (1939), API 5L, ASME B31.4/B31.8, API 1130, API RP 1175, NACE SP0169, Sloan-Koh (2008), PHMSA, CSA Z662, API 610, API 617 - Overall validation: 100.0/100 — Grade A+