--- license: cc-by-nc-4.0 task_categories: - tabular-classification - tabular-regression language: - en tags: - synthetic - oil-and-gas - upstream - directional-drilling - wellbore-trajectory - geosteering - survey-qc - anti-collision - minimum-curvature - iscwsa - xpertsystems pretty_name: "OIL-008 — Synthetic Wellbore Trajectory Dataset (Sample)" size_categories: - 100K3.0, alarm <1.5) | | 7 | avg uncertainty ellipse ft | 11.4819 | 11.5 | ±4.0 | ✓ PASS | ISCWSA MWD error model + SPE 90408 (Williamson 2000) — characteristic survey uncertainty ellipse major axis for MWD-surveyed horizontal wells at TD | | 8 | planned vs actual inc mae deg | 0.3182 | 0.4 | ±0.3 | ✓ PASS | SPE 178215 (geosteering delivery accuracy) + Halliburton Sperry directional engineering benchmarks — mean absolute inclination delivery error vs plan | | 9 | trajectory curvature realism | 0.9287 | 0.93 | ±0.05 | ✓ PASS | SPE 67616 + IADC — composite curvature realism index (1 − σ(DLS)/10), benchmarking dogleg-severity dispersion vs field-data envelopes | | 10 | basin diversity entropy | 0.9885 | 0.92 | ±0.08 | ✓ PASS | Rystad Energy + IHS Markit global rig fleet — 10-class basin diversity benchmark (Permian, Eagle Ford, Bakken, Marcellus, North Sea, Gulf of Mexico, Middle East, Canadian Oil Sands, Brazil Pre-Salt, North Africa), normalized Shannon entropy | **Overall: 100.0/100 — Grade A+** (10 PASS · 0 MARGINAL · 0 FAIL of 10 metrics) --- ## Schema highlights **`actual_trajectory.csv`** — the surveyed trajectory spine, one row per station per well. Computed via the **minimum-curvature method** (Bourgoyne et al., 1986; API/SPE industry standard): > Δnorth = ΔMD/2 · (sin(I₁)·cos(A₁) + sin(I₂)·cos(A₂)) · RF > Δeast = ΔMD/2 · (sin(I₁)·sin(A₁) + sin(I₂)·sin(A₂)) · RF > Δtvd = ΔMD/2 · (cos(I₁) + cos(I₂)) · RF where RF is the dogleg ratio factor `RF = (2/β)·tan(β/2)` and β is the dogleg angle between consecutive station vectors. This is the same math used by Compass, Landmark, SLB DDS, and every commercial survey-calculation package. **`drilling_sections.csv`** classifies each station as **Vertical** (MD < kickoff), **Build** (kickoff ≤ MD < build-end), or **Lateral** (MD ≥ build-end). DLS distributions are section-aware: | Section | DLS μ | DLS σ | |---|---:|---:| | Vertical | 2.7 | 0.55 | | Build | 3.9 | 0.65 | | Lateral | 3.05 | 0.55 | **`collision_monitoring.csv`** uses the **IADC separation factor** convention (target SF > 3.0, alarm SF < 1.5) with a mean ~4.7 — typical for mature basins with established offset-well drilling history. **`survey_uncertainty.csv`** ellipse axes follow **ISCWSA error model** conventions for MWD-surveyed wells (Williamson 2000, SPE 90408): major axis 5–18 ft, minor axis 2–9 ft, covariance index 0.88–0.98. **`bha_directional_data.csv`** distinguishes **rotary-steerable systems (RSS, ~58%)** from positive-displacement-motor (PDM) BHAs via the `rss_flag` column, matching the modern industry mix where RSS dominates long-lateral and ERD wells. --- ## Suggested use cases 1. **Trajectory anomaly detection** — flag stations where DLS exceeds section-specific envelopes using ML on the 30,605-row station- resolution spine. 2. **Geosteering target-hit prediction** — binary classifier on `target_hit_flag` (whether the lateral landed in the target zone) from BHA + trajectory + geosteering features. 3. **Anti-collision risk scoring** — regress `collision_risk_flag` and `separation_factor` from trajectory and offset-well features. 4. **Survey QC ML** — predict `qc_score`, `magnetic_interference_flag`, and `gyro_discrepancy_flag` from station-resolution trajectory data to triage surveys for human review. 5. **Planned-vs-actual delivery analytics** — quantify drilling delivery accuracy by regressing the inclination/azimuth/TVD delta between planned and actual at each station. 6. **Section classification** — multi-class classifier on `section_type` (Vertical/Build/Lateral) from trajectory shape features for automated well section segmentation. 7. **Torque-drag prediction** — regress torque and drag from trajectory complexity (DLS, inclination profile) and BHA features. 8. **Multi-table relational ML** — entity-resolution and graph-based learning across the 11 joinable tables via `well_id` and `survey_id`. --- ## Loading ```python from datasets import load_dataset ds = load_dataset("xpertsystems/oil008-sample", data_files="actual_trajectory.csv") print(ds["train"][0]) ``` Or with pandas: ```python import pandas as pd wells = pd.read_csv("hf://datasets/xpertsystems/oil008-sample/wells_master.csv") actual = pd.read_csv("hf://datasets/xpertsystems/oil008-sample/actual_trajectory.csv") planned = pd.read_csv("hf://datasets/xpertsystems/oil008-sample/planned_trajectory.csv") sections = pd.read_csv("hf://datasets/xpertsystems/oil008-sample/drilling_sections.csv") joined = actual.merge(planned, on=["well_id","md_ft"], suffixes=("_act","_plan")) ``` --- ## Reproducibility All generation is deterministic via the integer `seed` parameter (seeds 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 ML prototyping and trajectory research, not for live well planning. A few notes: 1. **Global-mean inclination is structurally lower than the generator's 72° target.** The generator's section composition (~19% Vertical + ~21% Build + ~60% Lateral) mathematically averages to ~64° — Vertical at 4°, Build at 47°, Lateral at 88.5° — even though each individual section is correctly modeled. The scorecard validates the **lateral- section inclination** (88.5°, on target) and **lateral section fraction** (60%, on target) directly, which are the operationally meaningful quantities. Future generator v1.2 will rebalance section weights to bring the global mean closer to 72° per the file header intent. 2. **Each station has an aligned row across all 11 tables** — the per-station tables (planned/actual/geosteering/collision/uncertainty/ sections/BHA/torque/QC/labels) are joinable by both `well_id` and station index. This is convenient for ML but slightly over-coupled relative to real-world data where uncertainty, BHA, and QC are typically sparser than the trajectory itself. 3. **Offset-well IDs in `collision_monitoring.csv` are synthetic** — the `offset_well_id` field samples from a 10,000-well synthetic pool independently per station, so the same offset well will not appear in multiple collision rows. For graph-based anti-collision ML, treat each row as an independent (well, offset_well) pair rather than as evidence of shared offset structure. 4. **Section spacing is uniform at 100 ft** in the sample. Real surveys are sparser in vertical sections (200-500 ft) and denser through build (50-100 ft). Future generator v1.2 will introduce non-uniform station spacing. 5. **Anomaly rate is 1.5%** (`anomaly_rate=0.015`) injected as randomly-elevated DLS values. This is a controlled noise channel for QC model training; filter `qc_score < 0.95` to remove the noisy stations. --- ## Full product The **full OIL-008 dataset** ships at **1,000 wells** with full ISCWSA error model error-band stratification per survey tool type (MWD/gyro/ inertial), per-basin offset-well graph structure with realistic neighborhood density, and non-uniform station spacing matching field survey practice — licensed commercially. Contact XpertSystems.ai for licensing terms. 📧 **pradeep@xpertsystems.ai** 🌐 **https://xpertsystems.ai** --- ## Citation ```bibtex @dataset{xpertsystems_oil008_sample_2026, title = {OIL-008: Synthetic Wellbore Trajectory Dataset (Sample)}, author = {XpertSystems.ai}, year = {2026}, url = {https://huggingface.co/datasets/xpertsystems/oil008-sample} } ``` ## Generation details - Generator version : 1.1-fixed - Sample version : 1.0.0 - Random seed : 42 - Generated : 2026-05-21 23:11:22 UTC - Wells : 200 - Station spacing : 100 ft - Anomaly rate : 1.5% - Basins : 10 (Permian, Eagle Ford, Bakken, Marcellus, North Sea, Gulf of Mexico, Middle East Carbonates, Canadian Oil Sands, Brazil Pre-Salt, North Africa) - Well types : 4 (Horizontal, Extended Reach, J-Well, S-Well) - Survey method : Minimum curvature (Bourgoyne et al. 1986) - Calibration basis : SPE 67616, SPE 90408 (Williamson 2000), SPE 178215, ISCWSA error model, API SPEC 7, IADC Directional Drilling Manual, OWSG, Rystad Energy, Spears & Associates, Halliburton/SLB directional handbooks - Overall validation: 100.0/100 — Grade A+