README.md

---
license: cc-by-nc-4.0
task_categories:
- tabular-classification
- tabular-regression
language:
- en
tags:
- synthetic
- oil-and-gas
- upstream
- drilling
- drill-bits
- bit-performance
- bit-wear
- dysfunction-detection
- mse-efficiency
- hydraulic-performance
- xpertsystems
pretty_name: "OIL-010 — Synthetic Drill Bit Performance Dataset (Sample)"
size_categories:
- 100K<n<1M
---

# OIL-010 — Synthetic Drill Bit Performance Dataset (Sample)

**SKU:** `OIL010-SAMPLE` · **Vertical:** Oil & Gas / Upstream Bit Performance
**License:** CC-BY-NC-4.0 (sample) · **Schema version:** `oil010.v1`
**Sample version:** `1.0.0` · **Default seed:** `42`

A free, schema-identical preview of XpertSystems.ai's enterprise drill-bit
performance dataset for bit-selection ML, dysfunction detection, wear
prediction, and bit-economics analytics. The sample covers **3,000 bits**
across **4 bit types** (PDC, Roller Cone, Hybrid, Diamond Impregnated) and
**5 formation classes**, with **15,000 drilling runs** linked across
**9 tables**.

---

## What's in the box

| File | Rows | Cols | Description |
|---|---:|---:|---|
| `bits_master.csv` | 3,000 | 6 | Bit catalog: type, manufacturer class, size, cutter & blade counts |
| `drilling_runs.csv` | 15,000 | 8 | Run spine: formation, depth interval, bit life, average ROP |
| `drilling_parameters.csv` | 142,002 | 7 | 5-15 WOB/RPM/torque/ROP samples per run + MSE efficiency |
| `bit_wear_logs.csv` | 15,000 | 6 | IADC dull grades + cutter/bearing wear % + seal failure probability |
| `vibration_measurements.csv` | 15,000 | 6 | Axial / torsional / lateral g + whirl index per run |
| `hydraulic_performance.csv` | 15,000 | 6 | Flow, pressure drop, HSI, hydraulic efficiency |
| `thermal_profiles.csv` | 15,000 | 5 | Bottomhole / cutter temperature + thermal cycles |
| `dysfunction_events.csv` | 5,204 | 5 | ~35% of runs: 5-class dysfunction (stick-slip, whirl, bounce, balling, thermal overload) |
| `drillbit_labels.csv` | 15,000 | 5 | ML labels: ROP efficiency grade (A/B/C/D), optimal-bit flag, failure risk score |

Total: **240,206 rows** across 9 CSVs, ~13.6 MB on disk.

---

## Calibration: industry-anchored, honestly reported

Validation uses a **10-metric scorecard** with targets sourced exclusively to
**named industry standards**: SPE 21943 (Pessier MSE foundational paper),
SPE 96652 (Dupriest & Koederitz MSE optimization), SPE 178850, SPE 178215,
API RP-7G drill stem design, API RP-13D drilling-fluid hydraulics, ISO
13503-5, IADC dull grading taxonomy, IADC Drilling Manual, Spears &
Associates bit market intelligence, Rystad Energy unconventional drilling
analytics.

**Sample run** (seed `42`, n_bits=3,000):

| # | Metric | Observed | Target | Tolerance | Status | Source |
|---|---|---:|---:|---:|---|---|
| 1 | avg rop fph | 72.0094 | 72.0 | ±18.0 | ✓ PASS | SPE 178850 + Rystad Energy unconventional drilling analytics — global mean ROP across mixed bit-type / formation portfolio |
| 2 | avg bit life hours | 115.3312 | 115.0 | ±30.0 | ✓ PASS | Spears & Associates bit market intelligence + IADC Drilling Manual — modern PDC/RC bit life expectancy across mixed formation portfolio |
| 3 | avg wob klbs | 38.0037 | 38.0 | ±10.0 | ✓ PASS | API RP-7G + SPE Drilling Engineering Handbook — global mean WOB across PDC/RC mixed bit portfolio |
| 4 | avg rpm | 144.9325 | 145.0 | ±30.0 | ✓ PASS | API RP-7G + SPE 178850 — global mean surface RPM across mixed top-drive / rotary-table drilling portfolio |
| 5 | avg hsi hp per in2 | 1.5986 | 1.6 | ±0.5 | ✓ PASS | API RP-13D + ISO 13503-5 drilling-fluid hydraulics — global mean HSI (hydraulic horsepower per square inch of bit area) for modern PDC bits (target 1.5-3.0 HSI) |
| 6 | avg bottomhole temp f | 244.8281 | 245.0 | ±60.0 | ✓ PASS | API + SPE thermal-drilling literature — global mean bottomhole circulating temperature across mixed onshore/offshore/HPHT drilling portfolio |
| 7 | dysfunction event rate | 0.3469 | 0.35 | ±0.1 | ✓ PASS | SPE 178215 + IADC Drilling Manual — fraction of bit runs exhibiting at least one named dysfunction event (stick-slip, whirl, bounce, balling, thermal overload) |
| 8 | avg mse efficiency | 0.8799 | 0.88 | ±0.1 | ✓ PASS | SPE 21943 (Pessier MSE) + SPE 96652 (Dupriest & Koederitz) — mean MSE drilling efficiency factor (UCS / MSE ratio) under properly-trimmed drilling parameters |
| 9 | bit type diversity entropy | 0.9997 | 0.95 | ±0.05 | ✓ PASS | Spears & Associates + IADC bit market reports — 4-class bit-type diversity benchmark (PDC, Roller Cone, Hybrid, Diamond Impregnated), normalized Shannon entropy (uniform sampling expected for ML training portfolios) |
| 10 | formation diversity entropy | 0.9999 | 0.95 | ±0.05 | ✓ PASS | Rystad Energy global drilling activity tracker + IADC — 5-class formation diversity benchmark (Permian shale, carbonate, abrasive sandstone, deepwater turbidite, geothermal hard rock), normalized Shannon entropy |

**Overall: 100.0/100 — Grade A+**
(10 PASS · 0 MARGINAL · 0 FAIL of 10 metrics)

---

## Schema highlights

**`bits_master.csv`** — the bit catalog spine, one row per bit.
4 bit types (PDC, Roller Cone, Hybrid, Diamond Impregnated), 3 manufacturer
classes (Premium, Mid-Tier, Budget), bit size 6-18 inches, blade count 4-8,
cutter count from N(45, 12).

**`drilling_runs.csv`** — 5 runs per bit on average. Each run carries
formation context (5-class: Permian shale, carbonate, abrasive sandstone,
deepwater turbidite, geothermal hard rock), depth interval, bit life in
hours, and the average ROP for that run. The `avg_rop_fph` column is
the seed for downstream `drilling_parameters` ROP samples (each per-stand
ROP sample is N(run_avg_rop, 8)).

**`drilling_parameters.csv`** — 5-15 WOB/RPM/torque/ROP/MSE-efficiency
samples per run. WOB N(38 klbs, 8), RPM N(145, 25), Torque N(18,500 ft-lb,
3,500), MSE efficiency N(0.88, 0.04).

**`bit_wear_logs.csv`** — **IADC dull grade taxonomy** (1-1-WT worn-teeth,
2-2-BT broken-teeth, 3-3-ER edge-rounded, 4-4-BH balled-up-with-heat),
plus cutter wear % (N(46, 12)) and bearing wear % (N(35, 15)).

**`vibration_measurements.csv`** — three-axis g-force measurements
matching modern downhole accelerometer ranges: axial N(1.8 g, 0.6),
torsional N(2.1 g, 0.7), lateral N(1.5 g, 0.5), whirl index N(0.18, 0.04).

**`hydraulic_performance.csv`** — flow rate N(650 gpm, 120), pressure
drop N(2,200 psi, 300), HSI (hydraulic horsepower per square inch of bit
area) N(1.6 hp/in², 0.25), hydraulic efficiency N(0.91, 0.03).

**`thermal_profiles.csv`** — bottomhole temperature N(245°F, 30), cutter
temperature N(380°F, 45), thermal cycles U(20, 400).

**`dysfunction_events.csv`** — sparse table, only ~35% of runs have a
dysfunction event (per the 0.35 trigger probability). 5-class dysfunction
taxonomy: stick-slip, bit whirl, bounce, balling, thermal overload.
Severity 0-1, downtime 5-300 minutes.

**`drillbit_labels.csv`** — three ML targets:
- `rop_efficiency_grade` (A/B/C/D) computed from `avg_rop_fph / 72` thresholds (≥1.1=A, ≥0.95=B, ≥0.8=C, else D)
- `optimal_bit_flag` (binary, 1 when efficiency > 1.0)
- `failure_risk_score` (continuous, 0-1)

---

## Suggested use cases

1. **Bit-type selection ML** — multi-class classifier on `bit_type` from
   formation + depth + manufacturer-class features, trained against
   ROP-grade target.
2. **ROP-grade classification** — train classifiers on `rop_efficiency_grade`
   (A/B/C/D) using drilling parameters, hydraulics, vibration, and bit
   spec features.
3. **Bit wear regression** — predict `cutter_wear_pct` and
   `bearing_wear_pct` from run depth, formation, drilling parameters, and
   thermal exposure.
4. **Dysfunction detection** — binary classifier on whether a run
   experiences a dysfunction event (join `dysfunction_events` to runs;
   ~35% positive rate). Then a 5-class secondary classifier on which
   dysfunction type given an event.
5. **MSE optimization** — regress `mse_efficiency` from WOB, RPM, torque,
   ROP using the 142,002-row drilling-parameters spine (5-15 samples
   per run for distribution-aware training).
6. **Hydraulic efficiency prediction** — predict `hydraulic_efficiency`
   from flow rate, pressure drop, HSI, and bit size.
7. **Thermal overload risk** — binary classifier predicting
   thermal-overload dysfunction from BH temperature, cutter temperature,
   and thermal cycles.
8. **Multi-table relational ML** — entity-resolution and graph-based
   learning across the 9 joinable tables via `bit_id` and `run_id`.

---

## Loading

```python
from datasets import load_dataset
ds = load_dataset("xpertsystems/oil010-sample", data_files="drilling_runs.csv")
print(ds["train"][0])
```

Or with pandas:

```python
import pandas as pd
bits    = pd.read_csv("hf://datasets/xpertsystems/oil010-sample/bits_master.csv")
runs    = pd.read_csv("hf://datasets/xpertsystems/oil010-sample/drilling_runs.csv")
params  = pd.read_csv("hf://datasets/xpertsystems/oil010-sample/drilling_parameters.csv")
wear    = pd.read_csv("hf://datasets/xpertsystems/oil010-sample/bit_wear_logs.csv")
joined  = runs.merge(bits, on="bit_id").merge(wear, on="run_id")
```

---

## Reproducibility

All generation is deterministic via the integer `seed` parameter (driving
`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 bit-
performance research, not for live bit-selection decisions. The
generator is a **direct-sampling design** (independent Gaussian draws
around industry-anchored target means) — fast to validate and easy to
extend, but with several limitations users should know about:

1. **No cross-feature physics coupling.** Each table is sampled
   independently of the others — WOB and torque are not correlated,
   vibration and dysfunction events are uncoupled, and thermal exposure
   does not drive wear progression. **ML models trained on this sample
   will learn marginal distributions but will not learn realistic
   cross-feature relationships.** The full OIL-010 product (v1.1
   roadmap) will introduce physics coupling via shared latent factors
   (UCS-driven ROP, WOB-driven torque, thermal-driven cutter wear,
   vibration-driven dysfunction).

2. **Formation and dysfunction are uniformly sampled, not conditioned.**
   In real drilling data, geothermal hard rock has very different
   dysfunction profiles (thermal overload dominant) than Permian shale
   (stick-slip dominant). The sample uses uniform 5-class draws for
   both; treat the joint distribution as ML-balanced rather than
   field-realistic.

3. **Three CSVs listed in the generator docstring are not generated**
   in this version: `lithology_transitions.csv`,
   `directional_performance.csv`, and `economics_metrics.csv`. These
   will ship in OIL-010 v1.1. Current product is 9 CSVs.

4. **Long-tail cutter-count outliers.** Cutter counts are drawn from
   N(45, 12), so a small fraction of bits (~0.1%) have unrealistically
   low cutter counts (<5). PDC bits in practice have 30-80 cutters;
   filter `cutter_count >= 20` if you need clean PDC training data.

5. **MSE efficiency is sampled, not computed.** The `mse_efficiency`
   column is a direct Gaussian draw N(0.88, 0.04), not derived from the
   Pessier MSE formulation (MSE = WOB/A + 120·π·N·T / (A·ROP)). For
   physically-consistent MSE labels, use OIL-007 (Drilling Parameters),
   which implements the full Teale/Pessier MSE physics with
   bit-size-aware area.

6. **All non-bit-master tables use uniform/Gaussian random IDs.** The
   `well_id` field in `drilling_runs.csv` samples from a 50,000-well
   synthetic pool independently per run, so the same well will not
   typically appear in multiple runs in the sample. For ML that
   requires well-level grouping, the full product introduces realistic
   well clustering.

---

## Full product

The **full OIL-010 dataset** ships at **5,000 bits / 25,000 runs**, with
the **v1.1 roadmap** adding cross-feature physics coupling, the three
missing tables (lithology_transitions, directional_performance,
economics_metrics), and basin-conditioned dysfunction priors — licensed
commercially. Contact XpertSystems.ai for licensing terms.

📧 **pradeep@xpertsystems.ai**
🌐 **https://xpertsystems.ai**

---

## Citation

```bibtex
@dataset{xpertsystems_oil010_sample_2026,
  title  = {OIL-010: Synthetic Drill Bit Performance Dataset (Sample)},
  author = {XpertSystems.ai},
  year   = {2026},
  url    = {https://huggingface.co/datasets/xpertsystems/oil010-sample}
}
```

## Generation details

- Sample version    : 1.0.0
- Random seed       : 42
- Generated         : 2026-05-21 23:33:12 UTC
- Bits              : 3,000
- Runs              : 15,000 (5 runs per bit on average)
- Bit types         : 4 (PDC, Roller Cone, Hybrid, Diamond Impregnated)
- Formations        : 5 (Permian shale, carbonate, abrasive sandstone,
                      deepwater turbidite, geothermal hard rock)
- Dysfunction types : 5 (stick-slip, bit whirl, bounce, balling, thermal overload)
- Calibration basis : SPE 21943 (Pessier MSE), SPE 96652 (Dupriest), SPE 178850,
                      SPE 178215, API RP-7G, API RP-13D, ISO 13503-5, IADC
                      dull grading, IADC Drilling Manual, Spears & Associates,
                      Rystad Energy
- Overall validation: 100.0/100 — Grade A+