xpertsystems/mfg005-sample

MFG-005 — Manufacturing Line Performance Dataset (Sample)

A schema-identical preview of MFG-005, the XpertSystems.ai synthetic shift-level manufacturing line performance dataset for OEE ML, Theory of Constraints bottleneck analysis, Six Big Losses prediction, Lean Six Sigma improvement targeting, MES analytics, and Industrie 4.0 production research. The full product covers 10,000-100,000 records. This sample is HF-sized at 3,000 records.

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.

---

What MFG-005 does — completing the 5-SKU Manufacturing vertical

MFG-005 is the fifth Manufacturing & Industrial Systems SKU in the XpertSystems catalog, completing a comprehensive Manufacturing vertical covering BOTH reliability engineering AND quality + operations management:

SKU	Domain	Granularity	Primary Audience
MGG-001	Reliability — sensor streams	1-min to 1-hr	IIoT, anomaly detection
MFG-002	Reliability — failure events	One row per event	CMMS, reliability engineering
MFG-003	Reliability — RUL training	Multi-obs per asset	PdM ML, PHM Society
MFG-004	Quality — inspection records	One row per inspection	QMS, SPC, MSA, 6 Sigma
MFG-005	Operations — line performance	One row per shift × line	MES, OEE, TPM, Lean

Where MFG-004 captures per-part quality (inspection-record granularity), MFG-005 captures per-shift line performance (production-system granularity). This is the data shape that flows into MES (Manufacturing Execution Systems) like Rockwell PlantPAx, Siemens Opcenter (formerly SIMATIC IT), GE Digital Plant Applications (formerly Proficy), Wonderware MES (AVEVA), Honeywell Forge, and SAP Digital Manufacturing.

Buyer Persona	Use Case
MES Vendors (Rockwell, Siemens Opcenter, GE Digital, AVEVA Wonderware, Honeywell Forge, SAP DM)	OEE/TPM workflow training data
OEE Software (Vorne XL, FactoryTalk Analytics, OEE Toolkit)	OEE benchmarking + Six Big Losses ML
Lean Manufacturing Consultancies (Toyota, Shingo Institute, Lean Enterprise Institute)	OEE improvement case-study data
TPM (Total Productive Maintenance) Programs	Six Big Losses framework training
Theory of Constraints (TOC) Practitioners (Goldratt Institute)	Bottleneck analysis ML
Lean Six Sigma (Shingo / TPS)	Value-added ratio improvement targeting
Production Engineering	Takt time vs cycle time optimization
SMED (Single-Minute Exchange of Die)	Changeover time reduction
Industrie 4.0 / Smart Factory	Digital twin training data
Energy Management (ISO 50001)	Energy per good unit ML

This is the substrate MES vendors, OEE software companies, Lean Six Sigma consultancies, Theory of Constraints practitioners, TPM programs, and Industrie 4.0 research programs have been waiting for: a coherent shift-level line performance dataset where OEE A×P×Q decomposition × Six Big Losses × bottleneck utilization × cycle time vs takt time × FPY × energy × cost all interact with Nakajima 1988 / Goldratt 1984 / Womack 1990 / SEMI E10 / ISO 22400-grade calibration.

---

What's inside — five related CSV files

MFG-005 is a multi-output relational dataset with five CSVs sharing line_id as join key.

File	Rows (sample)	Columns	Size
mfg005_synthetic_line_performance.csv	3,000	118	~2.3 MB
throughput_vs_takt.csv	3,000	11	~316 KB
bottleneck_analysis_report.csv	235	5	~14 KB
oee_summary_by_line.csv	36	9	~5 KB
downtime_pareto.csv	6	3	~360 B

Plus mfg005_metadata.json with run configuration.

Schemas are provided in five matching JSON files.

Main schema module structure (118 columns total)

Module	Cols	Coverage
Line identity	16	line_id, plant_id, work_center_id, shift_id, date, shift number, duration, product_id, family, line_type, automation level, stations, configuration, plant location, sector, production order
Cycle time	14	takt time, designed CT, actual CT avg/std/min/max, P50/P95, CoV, bottleneck station + CT + util, secondary bottleneck, CT loss
Production volume	14	planned/actual quantity, good/defective/scrap/rework units, throughput actual/design UPH, ratio, RTY, total parts, WIP queue avg/max, output variance, UPH per operator
OEE (Nakajima 1988)	13	planned production time, available time, A/P/Q components, OEE overall, TEEP, Loading, A/P/Q losses, Six Big Losses breakdown, OEE benchmark class, primary loss driver
Changeover & downtime	17	changeover time + count, unplanned/planned DT, DT events, MTBF, MTTR, largest DT, equipment failure flag, 6 downtime categories (mech/elec/tool/material/operator/quality), planned maintenance, last maintenance type
Equipment	2	age years, condition score
Operator	10	headcount planned/actual, utilization, skill level avg, absenteeism, overtime, ergonomic incidents, variance flag, cross-training ratio, supervisor
Quality	11	defect rate PPM, sigma level, Cpk + Cp primary, primary + secondary defect type, inspection method, SPC chart flag, OOC flag, quality alert, customer complaint, FPY, RFT
Energy & sustainability	7	energy kWh + per good unit, peak demand, compressed air, coolant, energy efficiency, carbon footprint
Cost	12	direct labour, overhead, material per unit, scrap, rework, downtime cost per min + total, COPQ, value-added ratio, production cost per unit

---

Calibration sources

Every distribution is anchored to named manufacturing engineering standards or canonical frameworks. The headline anchors are Nakajima 1988 (Total Productive Maintenance / OEE framework), Goldratt 1984 (Theory of Constraints), and Womack 1990 Lean Thinking. Other anchors:

• Nakajima 1988 Introduction to TPM — Six Big Losses framework (equipment failure, setup/adjustment, idling/minor stops, reduced speed, process defects, startup yield loss); OEE = A × P × Q decomposition.
• Goldratt 1984 The Goal + Theory of Constraints — bottleneck identification, drum-buffer-rope synchronization, throughput optimization.
• Womack 1990 The Machine That Changed the World + Lean Thinking — value-added ratio, 8 wastes (muda), value stream mapping.
• Ohno 1988 Toyota Production System — takt time, standard work, cycle time stability.
• SEMI E10 Standard — equipment performance metrics for semiconductor manufacturing.
• ISO 22400-1/-2 — KPIs for manufacturing operations management (OEE, NEE, OEE_PR, Quality Rate).
• ANSI/ISA-95 — enterprise-control system integration.
• SAE J4000 — Identification and Measurement of Best Practice in Implementation of Lean Operation.
• ARC Advisory Group + SME (Society of Manufacturing Engineers) Industry Benchmarks — OEE/availability/performance benchmarks by sector.
• Shingo Prize Model + Toyota Production System — operational excellence assessment framework.
• AIAG / VDA standards — automotive manufacturing benchmarks.
• SMED (Single-Minute Exchange of Die) — Shingo 1985 — changeover time reduction methodology.
• ISO 50001 — energy management systems for manufacturing.
• Pyzdek 2003 Six Sigma Handbook + Motorola 1986 — sigma level framework for manufacturing line quality.

---

Validation scorecard

The wrapper ships a 10-metric Nakajima/TOC/Lean/ISO 22400-anchored scorecard (validation_scorecard.json) that re-scores the dataset on every generation. Default seed 42 result:

ID	Metric	Target	Observed	Source
M01	OEE Overall Mean	0.53–0.77	0.653	Nakajima 1988 / ARC / SME
M02	OEE Availability	0.75–0.95	0.858	Nakajima 1988 / SMRP
M03	OEE Quality (FLOOR ≥95%)	≥0.95	0.994	Nakajima 1988 / Six Sigma
M04	Bottleneck Utilization %	82–102	95.01	Goldratt 1984 TOC
M05	World-Class OEE Share (≥0.85)	0.00–0.20	0.065	Nakajima 1988 / ARC
M06	Performance Loss Primary Driver	0.40–0.90	0.752	Six Big Losses / TPM
M07	Sigma Level Mean (FLOOR ≥2.5σ)	≥2.5σ	4.67σ	Motorola 1986 / ASQ
M08	Throughput Ratio	0.60–1.00	0.819	ISO 22400-2 / SEMI E10
M09	Cycle Time CoV (CEILING ≤0.23)	≤0.23	0.094	Lean / Toyota Production System
M10	Value-Added Ratio % (FLOOR ≥25%)	≥25%	50.10	Womack 1990 Lean Thinking

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

Standout calibration depth — this is the most precisely-centered Manufacturing SKU:

• M01 OEE 65.28% vs target 65% — 0.28pp deviation 🎯
• M02 Availability 85.84% vs target 85% — 0.84pp deviation 🎯
• M10 Value-added ratio 50.10% vs Womack target 50% — 0.10pp deviation 🎯
• M08 Throughput ratio 0.819 vs target 0.80 — 1.9pp deviation

OEE A×P×Q math verifies: 0.858 × 0.763 × 0.994 = 0.651 (matches observed OEE 0.653 within rounding).

OEE by sector reproduces published Nakajima benchmarks:

• Medical 0.69, Pharma 0.67, Industrial 0.65, Electronics 0.66, Automotive 0.64, Food/Bev 0.65

Six Big Losses primary driver matches TPM textbook: performance_loss 75% / availability_loss 25% / quality_loss <1% (mature plants are performance-bound, not availability-bound or quality-bound).

---

Suggested use cases

• OEE prediction ML — line characteristics + shift features × OEE overall prediction (regression).
• Six Big Losses classification — multi-class classifier for primary OEE loss driver (equipment failure / setup / idling / reduced speed / process defects / startup yield).
• Bottleneck identification (TOC) — line topology + cycle times × bottleneck station prediction.
• Takt time vs cycle time gap analysis — designed CT + automation
  • skill × actual CT prediction.
• Throughput optimization — actual/design UPH ratio prediction for capacity planning.
• Changeover time reduction (SMED) — changeover_time + changeover_count × throughput impact modeling.
• Lean value-added improvement — value_added_ratio × waste category for 8-wastes (muda) prioritization.
• Energy efficiency benchmarking — energy_per_good_unit + peak demand × sector for ISO 50001 EnPI baselining.
• Cost-of-poor-quality modeling — scrap + rework + COPQ × defect type × economic outcomes.
• Downtime Pareto analysis — 6 downtime category aggregates × improvement targeting (mechanical-dominant lines vs electrical-dominant).
• Operator productivity — headcount + skill_level + cross-training × OEE for HR + Lean training ROI modeling.

---

Loading

from datasets import load_dataset

main = load_dataset(
    "xpertsystems/mfg005-sample",
    data_files="mfg005_synthetic_line_performance.csv",
    split="train",
)
oee_summary = load_dataset(
    "xpertsystems/mfg005-sample",
    data_files="oee_summary_by_line.csv",
    split="train",
)

Or with pandas directly:

import pandas as pd
from huggingface_hub import hf_hub_download

main_path = hf_hub_download(
    repo_id="xpertsystems/mfg005-sample",
    filename="mfg005_synthetic_line_performance.csv",
    repo_type="dataset",
)
df = pd.read_csv(main_path)

# OEE A×P×Q decomposition by sector
for sector, sub in df.groupby("industry_sector"):
    a = sub["oee_availability"].mean()
    p = sub["oee_performance"].mean()
    q = sub["oee_quality"].mean()
    oee = sub["oee_overall"].mean()
    print(f"{sector:14s}: OEE={oee:.3f} = A({a:.3f}) × P({p:.3f}) × Q({q:.3f})")

# Six Big Losses primary driver distribution
print(df["oee_loss_primary_driver"].value_counts(normalize=True))

# Bottleneck utilization by line type (TOC)
bn_by_type = df.groupby("line_type")["bottleneck_utilisation_pct"].mean()
print(bn_by_type.sort_values(ascending=False))

Five schema JSON files are bundled for pipeline integration:

import json
schema_main = json.load(open("MFG_005_main_schema.json"))
schema_oee = json.load(open("MFG_005_oee_summary_schema.json"))
schema_bn = json.load(open("MFG_005_bottleneck_schema.json"))
schema_takt = json.load(open("MFG_005_throughput_schema.json"))
schema_dt = json.load(open("MFG_005_downtime_pareto_schema.json"))

This dataset is cross-sectional with shift-level granularity — one row per shift × line, ordered by shift_date but not strictly longitudinal per asset. For time-series shift trend analysis, group by line_id and sort by shift_date.

---

Schema highlights

Line identity — line_id, plant_id, work_center_id, shift_id, shift_date, shift_number ∈ {1, 2, 3}, shift_duration_minutes, product_id, product_family, production_order_id, line_type ∈ {assembly_line, machining_cell, packaging_line, fabrication, chemical_process, discrete_manufacturing, batch_process, continuous_process, hybrid}, automation_level ∈ {manual, semi_automated, highly_automated, lights_out, cobotic}, number_of_stations, line_configuration ∈ {serial, parallel, u_shaped, flexible, transfer_line, job_shop}, plant_location (12 global locations), industry_sector ∈ {automotive, electronics, pharma, food_bev, aerospace, industrial, consumer, medical, chemical, packaging}.

Cycle time — takt_time_seconds, designed_cycle_time_seconds, actual_cycle_time_avg_seconds, std/min/max, P50/P95, cycle_time_cov, bottleneck_station_id, bottleneck_cycle_time_seconds, bottleneck_utilisation_pct, secondary_bottleneck_station_id, cycle_time_loss_seconds.

Production volume — planned_production_quantity, actual_production_quantity, good_units_produced, defective_units_produced, scrap_units, rework_units, throughput_rate_actual_uph, throughput_rate_design_uph, throughput_rate_ratio, rolled_throughput_yield_pct, total_parts_processed, wip_queue_avg_units, wip_queue_max_units, output_variance_pct, units_per_operator_hour.

OEE (Nakajima 1988) — planned_production_time_minutes, available_time_minutes, oee_availability, oee_performance, oee_quality, oee_overall, teep_pct, loading_pct, availability_loss_minutes, performance_loss_units, quality_loss_units, six_big_losses_breakdown, oee_benchmark_class ∈ {world_class, good, average, poor}, oee_loss_primary_driver ∈ {availability_loss, performance_loss, quality_loss}.

Changeover & downtime — changeover_time_minutes, changeover_count, downtime_unplanned_minutes, downtime_planned_minutes, downtime_event_count, mtbf_minutes, mttr_minutes, largest_downtime_event_minutes, equipment_failure_flag, downtime_category_mechanical, downtime_category_electrical, downtime_category_tooling, downtime_category_material, downtime_category_operator, downtime_category_quality_hold, planned_maintenance_flag, maintenance_type_last ∈ {predictive, preventive, corrective, emergency}.

Equipment — equipment_age_years, equipment_condition_score.

Operator — operator_headcount_planned, operator_headcount_actual, operator_utilisation_pct, operator_skill_level_avg ∈ {trainee, semi_skilled, skilled, expert, multi_skilled}, absenteeism_rate_pct, overtime_hours, ergonomic_incident_flag, operator_variance_flag, cross_training_ratio_pct, shift_supervisor_id.

Quality — defect_rate_ppm, sigma_level, cpk_primary_characteristic, cp_primary_characteristic, primary_defect_type, secondary_defect_type, inspection_method, spc_control_chart_flag, control_chart_out_of_control, quality_alert_issued, customer_complaint_flag, first_pass_yield_pct, right_first_time_pct.

Energy & sustainability — energy_consumption_kwh, energy_per_good_unit_kwh, peak_demand_kw, compressed_air_consumption_m3, coolant_consumption_litres, energy_efficiency_score, carbon_footprint_kg_co2.

Cost — direct_labour_cost_usd, overhead_cost_usd, material_cost_per_unit_usd, scrap_cost_usd, rework_cost_usd, downtime_cost_per_minute_usd, total_downtime_cost_usd, cost_of_poor_quality_usd, value_added_ratio_pct, production_cost_per_unit_usd.

---

Calibration notes & limitations

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

1. Industry sector mix is skewed at n=3,000. Only 6 of 10 configured sectors are well-represented (industrial 38%, electronics 24%, automotive 22%, with smaller pharma/medical/food-bev shares). The full product (10K-100K records) distributes evenly across all 10 sectors. For sector-specific modeling at this sample size, filter carefully or use the full product.

2. TEEP averages 20.3% and Loading 31.1% — both significantly below typical 24/7 plant values. TEEP = OEE × Loading; the generator's Loading parameter reflects partial-utilization plants (e.g., 1-2 shifts/day rather than continuous 3-shift). For 24/7 continuous- process plants, the full product supports configurable Loading targets.

3. Quality component of OEE is 99.36% (very high). Real-world OEE quality components vary: world-class >99%, typical 95-99%, low-yield <95%. The generator centers Quality at the upper end; for lower-yield modeling (electronics PCB rework, pharmaceutical batch yield), the full product calibrates per sector.

4. Absenteeism averages 8.5% — above typical industrial 3-5%. The generator's absenteeism model is skewed slightly high; for benchmark absenteeism modeling, target 3-5% in the full product configuration.

5. Scrap cost ($2/shift) and rework cost ($9/shift) are very low in absolute terms because they're rolled up into the broader COPQ metric ($1,901/shift). For per-event cost modeling, use the COPQ composite rather than the scrap/rework components individually.

6. OEE by automation level is non-monotonic: cobotic 0.69 > lights_out 0.68 ≈ manual 0.68 > semi-auto 0.65 > highly_auto 0.62. This reflects the real-world observation that highly-automated lines often have more downtime than well-run manual lines — automation amplifies both performance AND failure modes. For automation-ROI modeling, this non-monotonicity is realistic.

7. Cycle time exceeds takt time on 81% of shifts (CT > takt). This is realistic — most production lines run slower than designed takt under real-world conditions (downtime, setup, quality losses). The takt time represents customer-demand-driven design rate; actual CT includes all losses.

8. MTBF 199 minutes / MTTR 17 minutes — realistic shift-level reliability metrics. Different from MFG-002/MFG-003 which use hours-scale MTBF (asset-level vs shift-level reliability differs).

9. 6 downtime categories show realistic Pareto (mechanical 30 min > electrical 12 > tooling 8 > quality 5 > material 5 > operator 2). Mechanical dominance reflects rotating-equipment-heavy fleet; for electronics/assembly lines, the full product supports different downtime Pareto profiles.

10. Deterministic seeding. Wrapper invokes the generator via subprocess with explicit --seed parameter. Seed sweep verifies Grade A+ across {42, 7, 123, 2024, 99, 1}.

---

Commercial / full product

The full MFG-005 product covers 10,000-100,000 shift records with configurable --automation_profile (modern_greenfield / brownfield_mixed / manual_intensive / lights_out), --oee_target_class (world_class / good / average / poor / mixed), refined sector-specific OEE benchmarks per Nakajima 1988 published targets, configurable absenteeism profiles per region (US/EU/APAC industrial benchmarks), 24/7 vs 2-shift vs 1-shift Loading configurations, pre-built feature engineering pipelines for OEE prediction ML (shift lag features, rolling MTBF/MTTR, seasonal patterns), Industrie 4.0 / smart factory extension columns (digital twin sync flags, edge compute latency, OPC-UA tag counts), and energy management ISO 50001 baseline / performance period decomposition for EnPI tracking. Available under commercial license — contact pradeep@xpertsystems.ai.

XpertSystems.ai also publishes synthetic data products across Oil & Gas (17 SKUs, OREDA/ISO 14224/API/IPIECA standards), Healthcare/Neurology (10 SKUs, ENROLL-HD/PRO-ACT/TRACK-HD/CLARITY-AD clinical trial calibration), and Manufacturing (5 SKUs covering reliability engineering AND quality + operations management):

• MGG-001: Factory Sensor Dataset (IIoT sensor streams)
• MFG-002: Machine Failure Event Records (CMMS, ISO 14224)
• MFG-003: Predictive Maintenance Dataset (RUL ML training)
• MFG-004: Quality Control Dataset (SPC, MSA, 6 Sigma)
• MFG-005: Manufacturing Line Performance (OEE, TPM, Lean) — this SKU

Catalog: huggingface.co/xpertsystems.