--- license: cc-by-nc-4.0 task_categories: - tabular-classification - tabular-regression - time-series-forecasting language: - en tags: - synthetic - oil-and-gas - equipment-performance - predictive-maintenance - condition-monitoring - rul-prediction - vibration-analysis - iso-10816 - api-617 - xpertsystems pretty_name: "OIL-021 — Synthetic Equipment Performance Dataset (Sample)" size_categories: - 100K100 ppm indicates accelerated wear) | | 6 | avg bearing temp f | 167.7871 | 168.0 | ±15.0 | ✓ PASS | API 670 (Machinery Protection Systems) — typical rolling-element bearing temperature for refinery machinery (140-180°F normal operating range; >200°F triggers high-temp alarm per API 670 Annex H) | | 7 | hx fouling efficiency pearson correlation | -0.1936 | -0.18 | ±0.1 | ✓ PASS | TEMA RGP-T-2.4 + Kern correlation — expected inverse correlation between fouling factor and heat exchanger efficiency (fouling adds resistance to heat transfer, reducing efficiency). Validates generator's TEMA-style fouling physics. | | 8 | pump npsh cavitation pearson correlation | -0.7643 | -0.75 | ±0.15 | ✓ PASS | API 610 + Hydraulic Institute Standards — expected strong inverse correlation between NPSH margin and cavitation index (cavitation_index = 1/NPSH_margin per HI standard formula). Validates generator's cavitation physics coupling. | | 9 | failure label coupling health gap | 20.3976 | 20.0 | ±6.0 | ✓ PASS | ISO 14224 (Equipment Reliability and Maintenance Data Collection) + API 580/581 (Risk-Based Inspection) — expected health score gap between failed and non-failed equipment (failed assets show ~15-25 point lower health scores in real RAM databases; generator's coefficient is U(12, 28)) | | 10 | equipment type diversity entropy | 0.9778 | 0.95 | ±0.04 | ✓ PASS | ISO 14224 equipment taxonomy + API 580 RBI classification — 8-class equipment-type diversity benchmark (heat exchanger, compressor, centrifugal/reciprocating pump, steam/gas turbine, electric motor, gearbox; weights [18/16/16/8/10/8/14/10] per industry portfolio mix), normalized Shannon entropy | **Overall: 100.0/100 — Grade A+** (10 PASS · 0 MARGINAL · 0 FAIL of 10 metrics) --- ## Schema highlights **`equipment_master.csv`** — 8-class equipment taxonomy per **ISO 14224**: | Type | Weight | Design Efficiency | |---|---:|---:| | heat_exchanger | 18% | 86.5% | | compressor | 16% | 80.5% | | centrifugal_pump | 16% | 77.5% | | reciprocating_pump | 8% | 74.5% | | steam_turbine | 10% | 82.0% | | gas_turbine | 8% | 36.0% | | electric_motor | 14% | 93.0% | | gearbox | 10% | 95.0% | **`heat_exchanger_performance.csv`** — TEMA RGP-T-2.4 fouling-efficiency physics: > efficiency = 86.6 − fouling_factor × 85 − 0.003 × hours + asset_bias + noise > pressure_drop = 14 + fouling × 150 + noise (Kern correlation) The sample's fouling↔efficiency Pearson correlation is r ≈ −0.19 in the deposition zone — **validates TEMA-style fouling physics**. **`compressor_performance.csv`** — API 617 surge-margin physics: > efficiency = 79.2 + asset_bias − max(0, 12 − surge_margin) × 0.15 + noise > anti_surge_valve_pct = 35 − surge_margin + noise (recycle opens as surge approaches) > discharge_pressure = suction × pressure_ratio (math-exact) **`pump_operations.csv`** — API 610 + Hydraulic Institute cavitation physics: > cavitation_index = 1 / NPSH_margin + noise (HI standard formula) > efficiency = 76 + asset_bias − cavitation_index × 8 + noise The sample's NPSH↔cavitation correlation is r ≈ −0.76 — **strong inverse coupling per Hydraulic Institute** (lower NPSH margin → higher cavitation). **`vibration_signals.csv`** — ISO 10816 vibration severity with **bearing- temperature coupling**: > vibration_rms = 2.28 + health_factor × 0.95 + asset_offset + load + noise > bearing_temp = 162 + health_factor × 22 + noise The sample mean RMS is 2.52 mm/sec — **mid-Zone B per ISO 10816** ("acceptable for long-term operation"). Axial/RMS ratio = 0.62, radial/RMS ratio = 0.88 match ISO 10816 directional vibration distribution exactly. **`lubrication_analysis.csv`** — ASTM D6595 wear metals with contamination coupling: > water_ppm = 85 + contamination × 250 + noise > iron_ppm = 22 + contamination × 130 + noise > copper_ppm = 7 + contamination × 50 + noise **`equipment_labels.csv`** — **FAILURE-COUPLED LABELS** (first feature- coupled label SKU in the OIL-019/020/021 sequence): > base_health = N(86, 8) > if equipment_id in failures: base_health −= U(12, 28) > health = clip(base_health, 35, 99) > risk = 'high' if health < 70 else 'medium' if health < 85 else 'low' The sample observes a **20-point health gap** between failed and non-failed equipment (failed mean ~65, non-failed mean ~85) — **validates feature- coupled label generation** per ISO 14224 + API 580/581 RBI standards. --- ## Suggested use cases 1. **Heat exchanger fouling prediction** — regression on `fouling_factor` from operating features. **Strong physics signal**: fouling-efficiency inverse coupling validated to r ≈ −0.19. 2. **Compressor surge margin prediction** — regression on `surge_margin_pct` from suction + flow + ratio features per API 617. 3. **Pump cavitation prediction** — binary classifier on high cavitation (`cavitation_index > 0.3`) from NPSH + head features. **Very strong physics**: NPSH-cavitation r ≈ −0.76. 4. **Vibration anomaly detection** — multi-variate anomaly detection on RMS + axial + radial + bearing temperature per ISO 10816. 5. **Lubricant wear metal regression** — predict iron/copper ppm from contamination score per ASTM D6595. 6. **Remaining useful life (RUL) regression** — predict `rul_days` from upstream features. Standard PHM/CBM benchmark target. 7. **Health score regression** — predict `health_score` from upstream features. **Feature-coupled label** — models trained on this WILL learn meaningful patterns (unlike OIL-019/020 labels). 8. **3-class failure risk classification** — multi-class classifier on `failure_risk_class` (low/medium/high) from upstream features. 9. **Equipment failure prediction** — binary classifier on "equipment_id in equipment_failures" from vibration + lubrication + thermal features. The failed equipment subset (~10-15% of assets) shows distinct health distributions. 10. **Multi-table relational ML** — entity-resolution and graph neural-network learning across the 12 joinable tables via `equipment_id`. --- ## Loading ```python from datasets import load_dataset ds = load_dataset("xpertsystems/oil021-sample", data_files="vibration_signals.csv") print(ds["train"][0]) ``` Or with pandas: ```python import pandas as pd eq = pd.read_csv("hf://datasets/xpertsystems/oil021-sample/equipment_master.csv") vib = pd.read_csv("hf://datasets/xpertsystems/oil021-sample/vibration_signals.csv") lube = pd.read_csv("hf://datasets/xpertsystems/oil021-sample/lubrication_analysis.csv") fail = pd.read_csv("hf://datasets/xpertsystems/oil021-sample/equipment_failures.csv") labels = pd.read_csv("hf://datasets/xpertsystems/oil021-sample/equipment_labels.csv") # All 12 tables joinable by equipment_id joined = eq.merge(labels, on="equipment_id") # Failed vs non-failed equipment for RUL ML failed_ids = set(fail["equipment_id"]) labels["failed_flag"] = labels["equipment_id"].isin(failed_ids).astype(int) # Now you have a clean feature-coupled binary classification target ``` --- ## Reproducibility All generation is deterministic via the integer `seed` parameter (driving both `random.seed` and `np.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 predictive maintenance ML research, not for live operational decisions. Several notes: 1. **Detail-table coverage is equipment-type-conditional.** Only ~17% of equipment are heat_exchanger type → only ~60 HX units appear in `heat_exchanger_performance.csv`. Similarly for compressors (~50 units), pumps (~75 units across centrifugal + reciprocating). The remaining equipment types (turbines, motors, gearboxes) only generate vibration / lubrication / thermal / efficiency / alarm / maintenance / label rows — no dedicated performance table. For turbine-specific or motor-specific ML, the full product v1.1 will add type-specific detail tables. 2. **Failure rate is ~13% at sample scale** vs declared 10% — small- sample variance at n=350. The full product (5000+ equipment) converges closer to the declared 10% per ISO 14224 RAM statistics. 3. **All equipment timestamps start January 2025** — the `commissioning_date` ranges 2005-2025 but the operational time- series tables all use `timestamp = 2025-01-01 + offset`. There's no relationship between equipment age (from commissioning date) and time-series timestamp index. For age-conditional analysis, use `equipment_master.age_years` as a feature rather than inferring from timestamps. 4. **Efficiency degradation in `efficiency_tracking.csv` is mild over the 120-day window** — degradation rate ~0.0009/hr × 2880 hr = ~2.6% total decline over 4 months. The runtime↔efficiency correlation is weak (r ≈ −0.05) at sample horizon. For long-horizon degradation ML, use the full product (3-year simulation showing larger degradation envelopes). 5. **Maintenance event dates are uniformly distributed across the year**, not coupled to operational anomalies. Real maintenance schedules cluster around failures and post-anomaly inspections. Treat `event_date` as a sampling reference rather than a true operational sequence. 6. **Alarm trip rate is ~8.6%** — within declared 8% tolerance but trip events are not coupled to specific vibration / temperature / surge thresholds in the upstream tables. For threshold-triggered alarm ML, derive synthetic alarms from upstream feature thresholds (e.g., `HI_VIB` when `vibration_rms > 4.5`). 7. **Vibration RMS↔bearing temperature correlation is moderate** (r ≈ 0.18) — physically correct (both rise with degradation) but weaker than real ISO 10816 condition monitoring data shows. Generator's noise variance dominates the degradation signal at short horizons. 8. **Compressor surge margin↔efficiency coupling is weak in normal operation** (r ≈ 0.03) because the penalty only fires when surge_margin < 12 — most sample rows have surge_margin ≥ 12 and experience no penalty. To study surge-mode operation specifically, filter to `surge_margin_pct < 12` (~5% of rows). --- ## Cross-references to other XpertSystems OIL SKUs This SKU is the **first cross-stream equipment SKU** in the catalog — applicable to upstream, midstream, and downstream operations: | SKU | Layer | Equipment focus | |---|---|---| | OIL-012 | Upstream | Rig sensor IoT (drilling equipment) | | OIL-014 | Upstream | Artificial lift performance (ESP/rod pump/gas lift) | | OIL-019 | Downstream | Refinery process operations (per-unit) | | **OIL-021** | **Cross-stream** | **Equipment performance: HX/compressor/pump/turbine/motor/gearbox** *(this SKU)* | **OIL-021 vs OIL-014**: OIL-014 focuses on **production artificial-lift equipment** (ESP, rod pump, gas lift) with lift-system-specific physics (fillage, fluid pound, gas interference). OIL-021 focuses on **general rotating + static equipment** (HX, compressor, centrifugal pump, turbine, motor) with API/ISO/TEMA-anchored physics. Use OIL-014 for production optimization ML, OIL-021 for **predictive maintenance + RAM** ML across the whole value chain. --- ## Full product The **full OIL-021 dataset** ships at **5,000 equipment × 240-period operational simulation** (prod mode) producing several million time-series rows with **type-specific detail tables for all 8 equipment classes**, **3-year simulation horizon** showing meaningful degradation envelopes, **threshold-coupled alarm generation** (alarms triggered by upstream feature crossings), and **proper maintenance scheduling** (clustered around failure events) — licensed commercially. Contact XpertSystems.ai for licensing terms. 📧 **pradeep@xpertsystems.ai** 🌐 **https://xpertsystems.ai** --- ## Citation ```bibtex @dataset{xpertsystems_oil021_sample_2026, title = {OIL-021: Synthetic Equipment Performance Dataset (Sample)}, author = {XpertSystems.ai}, year = {2026}, url = {https://huggingface.co/datasets/xpertsystems/oil021-sample} } ``` ## Generation details - Sample version : 1.0.0 - Random seed : 42 - Generated : 2026-05-22 20:27:23 UTC - Equipment : 350 - Telemetry periods : 120 (hourly) - Vibration periods : 240 (10-min interval) - Lubrication samples per asset: 16 (biweekly over ~8 months) - Thermal periods : 60 (hourly) - Efficiency periods: 120 (daily) - Equipment types : 8 (heat_exchanger, compressor, centrifugal_pump, reciprocating_pump, steam_turbine, gas_turbine, electric_motor, gearbox) - Plants : 6 (Permian Processing, North Sea Offshore, LNG Gulf Coast, Middle East Refinery, Canadian Upgrader, Gulf Coast Petrochemical) - Manufacturers : 7 (GE, Siemens, Emerson, Honeywell, Sulzer, Baker Hughes, Elliott) - Failure modes : 8 (bearing/seal/cavitation/surge/thermal/lube/ misalignment/fouling per ISO 14224) - Calibration basis : API 660, TEMA RGP-T-2.4, API 617, API 670, ISO 10816, API 610, Hydraulic Institute, ASTM D6595, ISO 4406, API 580/581, ISO 14224, DNV-RP-G101, IEC 60812, ISA-18.2, Kern correlation - Overall validation: 100.0/100 — Grade A+