xpertsystems/oil041-sample

OIL-041 — Synthetic Spare Parts Demand Dataset (Sample)

A schema-identical preview of OIL-041, the XpertSystems.ai synthetic spare-parts demand and inventory-optimization dataset for upstream, midstream, and downstream oil & gas operations. The full product covers ~6,000 facilities and ~450,000 parts across a 5-year horizon. This sample is the generator's sample mode (75 facilities × 1,500 parts × 45 warehouses × 365 days) covering all 13 product tables.

Built by XpertSystems.ai — Synthetic Data Platform Contact pradeep@xpertsystems.ai · xpertsystems.ai License CC-BY-NC-4.0 (sample); commercial license available for the full product.

---

How OIL-041 complements the PdM triptych (OIL-038/039/040)

The PdM triptych predicts equipment failure. OIL-041 models the supply-chain response to that failure: inventory positions, supplier lead-time variability, stockout root causes, emergency procurement, and ML-ready stockout/failure probability labels. Together the four SKUs form a complete asset-to-supply-chain stack:

Layer	SKU	Workload
Vibration / FFT signal processing	OIL-040	Anomaly detection, fault classification
Failure-event analytics + reliability KPIs	OIL-038	MTBF, MTTR, criticality modeling
RUL prognostics	OIL-039	Per-timestamp RUL + 7d/30d failure probability
Spare parts + procurement response	OIL-041	Stockout prediction, demand forecasting, lead-time variability

Buy or download all four for full upstream-asset coverage from sensor → failure → supply-chain response.

---

What's inside

13 CSV tables covering the complete MRO supply-chain plane: parts master → inventory levels → warehouse movements → maintenance consumption → supplier lead times → procurement orders → stockouts → emergency orders → reorder recommendations → ML-ready labels, plus 3 dimensional master tables.

Table	Rows (sample)	What it represents
spare_parts_master.csv	1,500	20-category × 15-equipment-type part master with criticality, lead-time, ABC class
inventory_levels.csv	~4,400	Sparse part-warehouse matrix with stock + reorder point + safety stock + stockout risk
warehouse_movements.csv	35,000	7-class movement taxonomy (receipt/issue/transfer/adjustment/reservation/return)
maintenance_consumption.csv	25,000	6-class maintenance × 8-class failure mode × downtime + labor hours
emergency_orders.csv	~450	4-class emergency level with freight multiplier + approval override
supplier_lead_times.csv	15,000	Quoted vs actual lead time + delivery status + logistics disruption
procurement_orders.csv	8,000	4-class approval status + emergency flag + budget exception
stockout_events.csv	~800	7-class root cause × 6-class mitigation + lost-production USD
reorder_recommendations.csv	~4,400	3-class policy (EOQ / service-level / risk-buffer) + optimization action
reliability_labels.csv	~4,400	Stockout probability + failure probability + 4-class priority + ML training label
facility_master.csv	75	9-class facility dim with complexity + maintenance intensity
warehouse_master.csv	45	5-class warehouse dim with capacity + utilization + hazmat flag
supplier_master.csv	90	4-tier supplier dim with reliability + contracted + single-source flag

Total: ~99,000 rows, ~9.5 MB. The full OIL-041 product is ~18 million rows.

---

Calibration sources

Every distribution and ratio is anchored to named public references. Highlights:

• APICS / ASCM Pareto-ABC inventory classification — critical part share and ABC class distribution.
• ISO 14224:2016 + XpertSystems schema — 20-class part taxonomy.
• ISO 55001 Asset management + APQC Process Classification — below-safety-stock fill-rate benchmarks.
• MRO Magazine + Plant Engineering MRO annual benchmarks — rotating- equipment lead times, emergency procurement rates.
• IPA Institute / IACCM Contract management — single-source supplier reliance baselines.
• Reliability Web Maintenance Strategy Survey — preventive + predictive consumption share.
• ARC Advisory Group / Logistics Bureau — expedited freight cost multipliers.
• APQC Supply-Chain Root-Cause Taxonomy — 7-class stockout root causes.
• SAP MM / Oracle EBS Inventory Management — 7-class warehouse movement 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	Critical Part Share (Pareto/ABC)	0.05–0.10	0.081	APICS / ASCM
M02	Median Lead Time (days)	15–45	26.4	MRO Magazine
M03	Preventive + Predictive Share (floor)	≥ 0.45	0.572	Reliability Web
M04	Single-Source Supplier Share	0.10–0.26	0.189	IPA / IACCM
M05	Below-Safety-Stock Share (ceiling)	≤ 0.07	0.046	ISO 55001 / APQC
M06	Procurement Emergency Share (ceiling)	≤ 0.09	0.071	MRO Magazine
M07	Part Category Taxonomy (floor)	≥ 20	20	ISO 14224 / XpertSystems
M08	Stockout Root Cause Taxonomy (floor)	≥ 7	7	APQC Supply Chain
M09	Emergency Freight Multiplier (mean)	2–4×	2.54×	ARC Advisory
M10	Warehouse Movement Taxonomy (floor)	≥ 7	7	SAP MM / Oracle EBS

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

---

Suggested use cases

• Stockout-probability prediction — pre-built stockout_probability labels in reliability_labels.csv calibrated from stockout-risk score, lead-time, and criticality. Train binary classifiers or regressors with ~30% positive class density.
• Demand forecasting from maintenance consumption — maintenance_consumption.csv has 25,000 work-orders × failure modes × downtime, enabling consumption-driven demand forecasting models.
• Supplier lead-time variability modeling — supplier_lead_times.csv contains quoted vs actual delays + reliability scores + logistics disruption flags. Train delay-prediction regressors for procurement planning systems.
• Stockout root-cause classification — 7-class root cause × 6-class mitigation action in stockout_events.csv enables process-mining and root-cause attribution models.
• Inventory optimization policy modeling — reorder_recommendations.csv carries 3-class policy (EOQ / service-level-optimized / risk-buffer) with optimization action labels. Compare policy ROI under stockout-driven loss models.
• Emergency procurement cost modeling — emergency_orders.csv has per-record freight multiplier × priority × approval override. Train cost-optimization models for expedited-shipment workflows.
• ABC-classification supervised learning — spare_parts_master.csv ships pre-computed ABC class labels (A: 16% / B: 32% / C: 51%) for classification training on demand × cost features.
• Pareto-critical part identification — criticality (4-class) × failure_linkage_score × obsolescence_risk_score enables criticality re-classification model training without proprietary CMMS data.

---

Loading

from datasets import load_dataset

parts = load_dataset(
    "xpertsystems/oil041-sample",
    data_files="spare_parts_master.csv",
    split="train",
)
inventory = load_dataset(
    "xpertsystems/oil041-sample",
    data_files="inventory_levels.csv",
    split="train",
)
labels = load_dataset(
    "xpertsystems/oil041-sample",
    data_files="reliability_labels.csv",
    split="train",
)

Or with pandas directly:

import pandas as pd
from huggingface_hub import hf_hub_download

path = hf_hub_download(
    repo_id="xpertsystems/oil041-sample",
    filename="stockout_events.csv",
    repo_type="dataset",
)
df = pd.read_csv(path)

All 13 tables join on:

• part_id → spare_parts_master ↔ inventory ↔ movements ↔ consumption ↔ leadtimes ↔ procurement ↔ stockouts ↔ recommendations ↔ labels
• facility_id → facility_master ↔ warehouse_master ↔ inventory ↔ movements ↔ consumption ↔ procurement ↔ stockouts ↔ emergencies ↔ recommendations ↔ labels
• warehouse_id → warehouse_master ↔ inventory ↔ movements ↔ stockouts ↔ recommendations ↔ labels
• supplier_id → supplier_master ↔ leadtimes ↔ procurement ↔ emergencies
• stockout_id → stockouts ↔ emergencies

---

Schema highlights

spare_parts_master.csv — part_id, part_category (20-class: bearing, seal, gasket, valve, pump_impeller, filter, motor, gearbox, compressor_blade, instrumentation, pressure_transducer, thermocouple, flange, hose, belt, lubrication_kit, control_module, actuator, cable, fastener), equipment_type (15-class), criticality ∈ {low, medium, high, critical}, unit_cost_usd (lognormal), standard_lead_time_days (gamma), mean_monthly_demand, failure_linkage_score ∈ [0, 1], obsolescence_risk_score ∈ [0, 1], repairable_flag, hazmat_flag, serialized_flag, abc_class ∈ {A, B, C}.

inventory_levels.csv — Sparse part × warehouse matrix: stock_qty, reserved_qty, safety_stock_qty, reorder_point_qty (safety-stock + lead-time-demand), max_stock_qty, inventory_value_usd, stockout_risk_score (logistic on stock vs reorder point).

supplier_lead_times.csv — quoted_lead_time_days, actual_lead_time_days, delay_days, supplier_reliability_score, logistics_disruption_flag, delivery_status ∈ {on_time, late, severely_late}.

stockout_events.csv — stockout_duration_hours (gamma), unplanned_downtime_hours, lost_production_usd (lognormal), severity_score ∈ [0, 1], root_cause (7-class: supplier_delay / forecast_error / emergency_failure / inventory_miscount / logistics_delay / turnaround_spike / obsolete_part), mitigation_action (6-class).

reliability_labels.csv — pre-built ML labels per part-warehouse pair: stockout_probability, failure_probability, critical_spare_gap_qty, inventory_health_score, recommended_priority ∈ {normal, medium, high, immediate}, ai_training_label ∈ {normal_replenishment, stockout_risk, failure_driven_demand}.

---

Calibration notes & limitations

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

1. Supplier on-time delivery rate is ~44%, not the industry-mature 85%+. The generator's actual-lead-time formula base × lognormal(σ=0.25) × (1 + (1 - reliability) × 1.6) is intentionally noisy — the goal is training delay-prediction models with real positive-class density, not reflecting best-in-class delivery performance. For "what does on-time look like" analysis, filter supplier_lead_times.csv to supplier_reliability_score > 0.92; that subset recovers the ≥85% on-time rate. The scorecard validates the more useful logistics_disruption_flag rate (~0.024, generator target 0.025) instead.

2. AI training label class imbalance. ai_training_label in reliability_labels.csv is ~87% normal_replenishment / 8% stockout_risk / 5% failure_driven_demand at sample scale — class-imbalanced by design (real MRO operations are imbalanced). For balanced training, stratified-sample by ai_training_label or threshold stockout_probability directly with custom cutoffs.

3. Inventory-levels table is a sparse sampling. With 1,500 parts × 45 warehouses = 67,500 possible pairs, the generator builds ~4,400 (≈6.5% density) representing realistic MRO sparsity (not every warehouse stocks every part). For "all part × all warehouse" studies, use the full product or override n_facilities / n_parts knobs on the underlying generator.

4. Recommendation policy distribution is dominated by EOQ (~48%). The 3-class policy logic prefers EOQ for non-critical-non-risky pairs. For service-level-optimized policy-specific studies (which only fires on criticality ∈ {high, critical} ≈ 28%), filter on criticality before training.

5. Stockout lost-production figures are lognormal-heavy-tailed. Median lost-production $379K is realistic, but the upper tail extends well into $10M+ at sample scale (rare catastrophe-class events). For tail-aware modeling, the sample's ~800 stockout events have limited high-tail density; the full product (200K stockouts) recovers tail statistics.

6. Single-source supplier share has small-sample variance. With only 90 suppliers and a target rate of 18%, the 6-seed sweep observed range 12–24%. Scorecard M04 tolerance is widened to ±0.08 to account for this; the full product (~9,000 suppliers) recovers tight 18% ± 0.4%.

7. Deterministic seeding. All 13 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-041 product covers 6,000 facilities × ~450,000 parts × ~9,000 suppliers × ~4,800 warehouses across a 5-year horizon (18 million rows total), with full part × warehouse matrix density, calibrated supplier-on-time performance distributions, balanced AI training label variants, and configurable demand-shock and disruption mode-packs. Available under commercial license — contact 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.