README.md

---
license: cc-by-nc-4.0
task_categories:
  - tabular-classification
  - tabular-regression
  - time-series-forecasting
language:
  - en
tags:
  - synthetic
  - equipment-failure
  - predictive-maintenance
  - reliability
  - mtbf
  - mttr
  - oil-and-gas
  - rotating-equipment
  - pump
  - valve
  - compressor
  - vibration
  - thermal-monitoring
  - lubrication
  - iso-14224
  - iso-10816
  - iso-4406
  - api-rp-580
  - api-rp-691
  - isa-18-2
  - eemua-191
  - sae-arp4761
  - condition-monitoring
pretty_name: "OIL-038 — Synthetic Equipment Failure Dataset (Sample)"
size_categories:
  - 100K<n<1M
---

# OIL-038 — Synthetic Equipment Failure Dataset (Sample)

A schema-identical preview of **OIL-038**, the XpertSystems.ai synthetic
equipment-failure and predictive-maintenance dataset for oil & gas rotating
and stationary assets (pumps, valves, compressors). The full product covers
~6,500 facilities across a 3-year horizon with ~6 million telemetry rows.
This sample is the generator's `demo` mode (20 facilities × ~18 assets ×
90 days) covering all 16 product tables.

> **Built by** XpertSystems.ai — Synthetic Data Platform
> **Contact** [pradeep@xpertsystems.ai](mailto:pradeep@xpertsystems.ai) · [xpertsystems.ai](https://xpertsystems.ai)
> **License** CC-BY-NC-4.0 (sample); commercial license available for the full product.

---

## What's inside

16 CSV tables covering the complete equipment-reliability lifecycle: facility
master → asset master → failure events (3 equipment groups) → telemetry
(vibration / thermal / lubrication / environmental / alarms) → maintenance
work orders → root cause analysis → downtime → reliability KPIs → sensor
drift → pre-built ML labels.

| Table | Rows (sample) | What it represents |
|---|---:|---|
| `facility_master.csv` | 20 | 10-type facility master with environment, corrosivity, maturity |
| `equipment_master.csv` | ~340 | 6-type asset master with baseline MTBF, criticality, maintenance strategy |
| `pump_failures.csv` | ~130 | Centrifugal + PD pump failures with cavitation, seal, bearing signals |
| `compressor_failures.csv` | ~90 | Reciprocating + centrifugal compressor failures with surge, vibration |
| `valve_failures.csv` | ~175 | Control + isolation valve failures with leakage, actuator response |
| `vibration_signals.csv` | ~30,500 | Daily vibration telemetry: RMS velocity, FFT peak, kurtosis, crest factor |
| `thermal_profiles.csv` | ~30,500 | Daily thermal telemetry: temperature, stress index, cooling, cycle counts |
| `lubrication_analysis.csv` | ~4,400 | Weekly oil samples: quality, contamination, water ppm, ISO 4406 codes |
| `maintenance_work_orders.csv` | ~400 | 8-type repair categories with labor hours, parts delays |
| `root_cause_analysis.csv` | ~400 | RCA method, immediate/systemic causes, contributing factors |
| `downtime_events.csv` | ~400 | Shutdown type, downtime hours, production loss, repair cost |
| `reliability_metrics.csv` | ~340 | Per-asset MTBF, MTTR, availability, reliability score, risk rank |
| `sensor_drift.csv` | ~1,000 | 6 sensor types with drift %, calibration status |
| `environmental_conditions.csv` | ~30,500 | Daily environmental telemetry: ambient, humidity, salt fog, H₂S |
| `alarm_events.csv` | ~15,500 | 7-class alarm taxonomy with severity, response delay, flood flag |
| `equipment_failure_labels.csv` | ~340 | Pre-built ML labels: 30d/90d failure probability, priority, action |

Total: ~125,000 rows, ~9.4 MB. The full OIL-038 product is ~6 million rows.

---

## Calibration sources

Every distribution and ratio is anchored to **named public references**. The
validation scorecard (see below) re-scores observed vs. target for 10
industry-anchored metrics, every one citing its source. Highlights:

- **SAE ARP4761 / API RP 691** — rotating equipment baseline MTBF benchmarks.
- **SAE ARP** rotating equipment maintenance standards — typical MTTR ranges.
- **PSAM / SAE** rotating equipment performance — availability benchmarks.
- **ISO 4406:2021** Hydraulic fluid cleanliness code + **Noria Lubrication
  Practices** — lubricant water content thresholds.
- **ARC Advisory Group** Predictive Maintenance Maturity Survey — sensor-based
  detection share benchmarks.
- **Reliability Web Maintenance Strategy Survey** — proactive maintenance
  strategy share.
- **API RP 580** Risk-Based Inspection — criticality-tier distribution.
- **ISO 14224:2016** Reliability and Maintenance Data — work-classification
  taxonomy.
- **ISO 10816** Mechanical vibration evaluation — vibration severity classes.
- **ISA 18.2 / EEMUA 191** — alarm management taxonomy.

---

## Validation scorecard

The wrapper ships a 10-metric scorecard (`validation_scorecard.json`) that
re-scores the dataset on every generation. Default seed 42 result:

| ID | Metric | Target | Observed | Source |
|---|---|---|---:|---|
| M01 | Median baseline MTBF (hours) | 4,000–15,000 | **4,119** | SAE ARP4761 / API RP 691 |
| M02 | Median MTTR (hours) | 8–24 | **17.4** | SAE ARP rotating equipment |
| M03 | Median availability (floor) | ≥ 0.92 | **0.992** | PSAM / SAE |
| M04 | Lubrication water ppm median (ceiling) | ≤ 250 | **77** | ISO 4406 / Noria |
| M05 | Sensor-based detection share (floor) | ≥ 0.40 | **0.578** | ARC Advisory PdM |
| M06 | Proactive maintenance share (floor) | ≥ 0.40 | **0.516** | Reliability Web survey |
| M07 | Critical+High criticality share | 0.40–0.60 | **0.516** | API RP 580 RBI |
| M08 | Failure-mode taxonomy coverage | 17–25 | **25** | ISO 14224 / API RP 691 |
| M09 | Repair-type taxonomy coverage (floor) | ≥ 8 | **8** | ISO 14224:2016 |
| M10 | Alarm-code taxonomy coverage (floor) | ≥ 7 | **7** | ISA 18.2 / EEMUA 191 |

**Grade: A+ (100/100). Verified across seeds 42, 7, 123, 2024, 99, 1.**

---

## Suggested use cases

- **Predictive maintenance modeling** — train classifiers that predict
  `failure_probability_30d` or `failure_probability_90d` from vibration,
  thermal, lubrication, and environmental telemetry. Pre-built labels in
  `equipment_failure_labels.csv`.
- **Remaining useful life (RUL) regression** — use `health_index` time
  series in `vibration_signals.csv` joined with failure events for
  prognostics models. Each asset has a degradation trajectory across the
  90-day sample window.
- **Fault classification across 25 fault modes** — pump (cavitation,
  bearing wear, seal leak, …), compressor (surge, rotor imbalance, …),
  and valve (actuator failure, stiction, …) failure modes are decomposable
  by equipment type for multi-task learning.
- **Alarm-flood and ISA 18.2 alarm-management benchmarking** — 15K alarm
  events × 7 codes × 5-class severity supports nuisance-alarm filtering,
  rationalization, and prioritization research.
- **MTBF / MTTR / availability benchmarking** — per-asset reliability
  metrics in `reliability_metrics.csv` enable Weibull / exponential
  reliability function fitting and what-if maintenance-strategy ROI
  modeling.
- **Sensor-fusion and anomaly detection** — vibration + thermal + lubrication
  + environmental telemetry are pre-aligned per asset per timestamp,
  enabling multi-modal anomaly detection benchmarks.
- **Maintenance strategy ROI** — `maintenance_strategy` × `availability`
  cross-tab enables direct quantification of reactive vs. time-based vs.
  condition-based vs. predictive program ROI.

---

## Loading

```python
from datasets import load_dataset

# Load failure events
pumps = load_dataset(
    "xpertsystems/oil038-sample",
    data_files="pump_failures.csv",
    split="train",
)
# Load telemetry (largest tables; consider streaming)
vibration = load_dataset(
    "xpertsystems/oil038-sample",
    data_files="vibration_signals.csv",
    split="train",
)
labels = load_dataset(
    "xpertsystems/oil038-sample",
    data_files="equipment_failure_labels.csv",
    split="train",
)
```

Or with pandas directly:

```python
import pandas as pd
from huggingface_hub import hf_hub_download

path = hf_hub_download(
    repo_id="xpertsystems/oil038-sample",
    filename="reliability_metrics.csv",
    repo_type="dataset",
)
df = pd.read_csv(path)
```

All 16 tables join on:

- `facility_id` → facility ↔ equipment ↔ all telemetry/failure/alarm tables
- `equipment_id` → equipment ↔ failures ↔ telemetry ↔ reliability ↔ labels
- `failure_id` → failure events ↔ work orders ↔ RCA ↔ downtime

---

## Schema highlights

**`equipment_master.csv`** — `equipment_id`, `facility_id`, `equipment_type`
(6-class: centrifugal_pump / positive_displacement_pump / control_valve /
isolation_valve / reciprocating_compressor / centrifugal_compressor),
`manufacturer_family` (6-class), `installed_date`, `equipment_age_years`,
`criticality` ∈ {low, medium, high, critical}, `rated_capacity`,
`design_margin`, `normal_operating_pressure_psi`, `normal_operating_temp_f`,
`maintenance_strategy` ∈ {reactive, time_based, condition_based, predictive},
`sensor_coverage_score`, `baseline_mtbf_hours`.

**Failure events** — 18 fault modes split by equipment family:
- *Pumps* (centrifugal + PD): cavitation, bearing_wear, seal_leak,
  shaft_misalignment, impeller_erosion, motor_overload, valve_plate_wear,
  overpressure, lubrication_failure, drive_failure
- *Valves* (control + isolation): actuator_failure, stiction, seat_leakage,
  positioner_fault, stem_corrosion, packing_leak, seal_degradation,
  blocked_operation
- *Compressors* (reciprocating + centrifugal): surge, valve_failure,
  rod_load_exceedance, lubrication_failure, cylinder_overheat,
  rotor_imbalance, fouling, high_vibration

Plus 14-class `root_cause` (normal_wear, poor_lubrication, process_upset,
corrosion, maintenance_error, operator_error, design_limit_exceeded,
contamination, thermal_cycling, misalignment, foreign_object_damage,
sensor_drift, high_humidity, H2S_exposure).

**`vibration_signals.csv`** — `rms_velocity_mm_s` (ISO 10816 unit),
`fft_peak_hz`, `kurtosis`, `crest_factor`, `health_index` (per-asset
degradation trajectory), `load_factor`. Daily samples per asset.

**`lubrication_analysis.csv`** — `oil_quality_score`, `contamination_index`,
`water_ppm`, `particle_count_iso_code` ∈ {16/14/11, 17/15/12, 18/16/13,
19/17/14, 20/18/15, 21/19/16} (ISO 4406 cleanliness codes), `viscosity_change_pct`.

**`alarm_events.csv`** — `alarm_code` ∈ {HIGH_VIB, HIGH_TEMP,
LOW_OIL_QUALITY, PRESSURE_DEVIATION, SENSOR_DRIFT, SURGE_RISK,
CAVITATION_RISK}, `severity` ∈ {info, low, medium, high, critical} (ISA 18.2),
`response_delay_minutes`, `acknowledged_flag`, `alarm_flood_flag`.

**`equipment_failure_labels.csv` — pre-built ML labels per asset**:
`failure_probability_30d`, `failure_probability_90d`, `criticality_score`,
`risk_rank_score`, `predictive_maintenance_priority` ∈ {low, medium, high,
urgent}, `recommended_action` ∈ {normal_monitoring, increase_monitoring,
schedule_maintenance, immediate_inspection}.

---

## Calibration notes & limitations

In the spirit of honest synthetic data, a few things buyers of the sample
should know:

1. **Telemetry health-index trajectory.** The vibration and thermal
   telemetry use an aggressive degradation drift model — across the 90-day
   sample window, the mean `health_index` drops to ~0.40 and ~80% of
   vibration RMS values exceed ISO 10816 Class 4 alarm thresholds (≥ 4.5
   mm/s). **This is intentional**: the sample compresses a longer
   degradation horizon into 90 days so ML utility is high (positive-class
   density for failure classifiers). For studies that require steady-state
   healthy operations, filter to `health_index > 0.65` or use the early
   telemetry window (first 14 days). The full product uses a slower
   degradation drift over 1,095 days.

2. **Alarm acknowledgment rate.** Sample acknowledgment rate (~0.78) is
   below the ISA 18.2 mature target of ≥ 0.85. This is a function of the
   stress-coupled acknowledgment probability in the generator and is
   intentional (sample window is degradation-heavy, so operators are
   modeled as overloaded). Filter to `severity ∈ {high, critical}` to
   recover the mature acknowledgment rate.

3. **Alarm flood rate.** The 12.5% flood rate in `alarm_events.csv` is
   slightly above EEMUA 191's ≤ 10% mature target — same root cause as
   acknowledgment rate. Useful for *training* alarm-flood detection
   models (positive-class density). The full product targets 6–9% flood
   rate at production scale.

4. **Pre-built ML label priority skew.** `predictive_maintenance_priority`
   concentrates on `low` (44%) and `medium` (56%) at sample scale, with
   `high` and `urgent` near zero. This is because risk_rank_score
   distributes between 0.25–0.50 for most assets at the 90-day demo
   window. For balanced multi-class training, **use `failure_probability_90d`
   directly with custom quantile thresholds**, or use the full product's
   3-year window which exhibits the full priority distribution.

5. **No-failure asset share.** ~29% of assets have zero failures across the
   90-day sample (Poisson zero-events). This is correct probabilistic
   behavior, but means median `observed_mtbf_hours` is the time-horizon
   itself (2,160 hours) for those assets. Filter to `failure_count > 0`
   when computing reliability KPIs.

6. **Deterministic seeding.** All 16 tables are deterministic on `--seed`.
   Catalog default is seed 42. Seed sweep verifies Grade A+ across
   {42, 7, 123, 2024, 99, 1}.

---

## Commercial / full product

The full **OIL-038** product covers ~6,500 facilities × 38 assets across a
1,095-day horizon (~6 million telemetry rows total), with calibrated
degradation drift, ISA 18.2 / EEMUA 191-aligned alarm acknowledgment and
flood rates, and a balanced priority-label distribution. Available under
commercial license — contact
[pradeep@xpertsystems.ai](mailto:pradeep@xpertsystems.ai).

XpertSystems.ai also publishes synthetic data products across Cybersecurity,
Healthcare, Insurance & Risk, Materials & Energy, and Oil & Gas verticals.
Catalog: [huggingface.co/xpertsystems](https://huggingface.co/xpertsystems).