--- license: cc-by-nc-4.0 task_categories: - tabular-classification - tabular-regression language: - en tags: - synthetic - oil-and-gas - upstream - core-analysis - petrophysics - rock-properties - SCAL - mineralogy - geomechanics - xpertsystems pretty_name: "OIL-006 — Synthetic Core Sample Dataset (Sample)" size_categories: - 100K log(k) ≈ log(k_basin_mean) + 6.0·(φ − φ_basin_mean) + ε with basin priors calibrated to industry-typical values: Permian Wolfcamp (φ̄ ≈ 8.5%, k̄ ≈ 0.08 mD), Marcellus (φ̄ ≈ 6.5%, k̄ ≈ 0.0003 mD), North Sea Sandstone (φ̄ ≈ 22%, k̄ ≈ 350 mD), GoM Deepwater (φ̄ ≈ 26%, k̄ ≈ 800 mD), Canadian Oil Sands (φ̄ ≈ 32%, k̄ ≈ 2500 mD), etc. **`special_core_analysis.csv`** — Archie's law parameters per plug: > F = a / φᵐ (formation resistivity factor) with a/m/n drawn from industry-typical ranges (a ≈ 1.0, m ≈ 1.95, n ≈ 2.0) matching the SPWLA conventions and the original Archie (1942) JPT paper. **`xrd_xrf_analysis.csv`** — Dirichlet-sampled mineralogy guaranteeing mass balance (quartz + feldspar + clay + carbonate = 100% per row), plus clay sub-fractions (illite/smectite/kaolinite/chlorite), TOC, kerogen type (I/II/II-S/III/IV), and vitrinite reflectance (oil window ~0.6-1.3%, gas window >1.3%). **`geomechanical_tests.csv`** — porosity-modulated elastic properties: > E_static ≈ 8e6 · (1 − 2.5·φ) + ε (psi) > UCS ≈ E_static / 250 + ε (psi) matching the Chang et al. (2006) empirical correlation for sedimentary rocks. --- ## Suggested use cases 1. **Porosity-permeability regression** — train ML estimators of permeability from porosity + lithology + grain density using the 37,398-plug spine. 2. **Reservoir quality classification** — multi-class classifier on `reservoir_quality_grade` (A/B/C/D) from petrophysical features. 3. **Pay zone identification** — binary classification on `pay_zone_flag` from RCA + lithology + mineralogy features. 4. **SCAL surrogate models** — predict Archie m/n, wettability index, and relperm endpoints from petrophysical and mineralogical inputs (multi- output regression). 5. **Hydrocarbon type prediction** — 7-class classifier on `hydrocarbon_type` from basin, depth, and rock properties. 6. **Multi-table relational ML** — entity-resolution and graph-based learning across the 11 joinable tables via `core_id` / `plug_id`. 7. **Mineralogy → petrophysics ML** — predict porosity and permeability from XRD/XRF mineralogy (quartz/clay/carbonate/feldspar fractions). 8. **Geomechanical surrogates** — predict Young's modulus, UCS, and brittleness from porosity + lithology for unconventional completion design. --- ## Loading ```python from datasets import load_dataset ds = load_dataset("xpertsystems/oil006-sample", data_files="plug_measurements.csv") print(ds["train"][0]) ``` Or with pandas: ```python import pandas as pd cores = pd.read_csv("hf://datasets/xpertsystems/oil006-sample/cores_master.csv") plugs = pd.read_csv("hf://datasets/xpertsystems/oil006-sample/plug_measurements.csv") rca = pd.read_csv("hf://datasets/xpertsystems/oil006-sample/routine_core_analysis.csv") scal = pd.read_csv("hf://datasets/xpertsystems/oil006-sample/special_core_analysis.csv") joined = plugs.merge(rca, on="plug_id").merge(cores, on="core_id") ``` --- ## Reproducibility All generation is deterministic via the integer `seed` parameter. The ID conventions (`CORE_{i:06d}`, `PLUG_{i:08d}`, `RCA_{i:08d}`, etc.) guarantee schema-stable joins across runs. 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 ML prototyping and core-analysis research, not for live drilling or completion decisions. A few notes: 1. **Permeability is heavy-tailed.** The lognormal Kozeny-Carman model produces realistic but right-skewed permeability distributions (sample p90 ≈ 1100 mD, median ≈ 0.3 mD). Use log-transformed permeability for statistical work and `np.log10(permeability_md + 1e-5)` for correlation analyses to match the φ-k coefficient reported in the scorecard. 2. **Basin / lithology coverage at sample scale** — at 500 cores, each basin has 29-79 cores. All 6 lithologies are present but tight_ss and dolomite are under-represented (~10% and ~6% of plugs respectively). Full product (25,000 cores) gives 2,000-4,000 cores per basin and converges all lithology distributions. 3. **2.8% controlled anomaly injection** is present in `plug_measurements` (`anomaly_flag` column) and `routine_core_analysis` (`anomaly_flag` column). This simulates stress-relief microfractures inflating permeability (plug level, 2-10× multipliers) and measurement repeatability artifacts (RCA helium porosity, ±1.5% noise). Use these flags as QC training targets or filter them out for clean regression baselines. 4. **Wettability index** is sampled with a global mean of -0.10 (mixed-to- slightly-oil-wet), not stratified by basin wettability prior. The v1.1 generator will introduce basin-stratified wettability sampling for tighter calibration. --- ## Full product The **full OIL-006 dataset** ships at **25,000 cores** with **~3.5M plug measurements**, full per-basin wettability stratification, basin-conditioned TOC sampling, and complete petrophysics-SCAL-mineralogy-geomechanics relational schema — licensed commercially. Contact XpertSystems.ai for licensing terms. 📧 **pradeep@xpertsystems.ai** 🌐 **https://xpertsystems.ai** --- ## Citation ```bibtex @dataset{xpertsystems_oil006_sample_2026, title = {OIL-006: Synthetic Core Sample Dataset (Sample)}, author = {XpertSystems.ai}, year = {2026}, url = {https://huggingface.co/datasets/xpertsystems/oil006-sample} } ``` ## Generation details - Generator version : 1.0.0 - Sample version : 1.0.0 - Random seed : 42 - Generated : 2026-05-21 22:43:24 UTC - Cores : 500 - Plugs : 37,398 - Basins : 10 (Permian Wolfcamp, Eagle Ford, Bakken, Marcellus, North Sea Sandstone, GoM Deepwater, Middle East Carbonate, Canadian Oil Sands, Pre-Salt Brazil, North Africa Carbonate) - Lithologies : 6 (clean SS, shaly SS, tight SS, shale, carbonate, dolomite) - Calibration basis : API RP-40, SCA, SPWLA, Archie (1942), Anderson (1986), Kozeny-Carman, Chang et al. (2006), SPE PEH - Overall validation: 100.0/100 — Grade A+