Hugging Face's logo

Datasets:
xpertsystems
/
mgg001-sample

Tasks:
Tabular Classification
Tabular Regression
Time Series Forecasting
Languages:
English
Size:
10K<n<100K
Tags:
Synthetic
manufacturing
industrial-iot
predictive-maintenance
condition-monitoring
iso-10816
License:
Dataset card Files
xet
 Community
mgg001-sample / README.md
pradeep-xpert's picture
pradeep-xpert
Upload folder using huggingface_hub
fa42c6c verified
preview code
|
raw
history blame contribute delete
18.8 kB
metadata 
license: cc-by-nc-4.0
task_categories:
  - tabular-classification
  - tabular-regression
  - time-series-forecasting
language:
  - en
tags:
  - synthetic
  - manufacturing
  - industrial-iot
  - predictive-maintenance
  - condition-monitoring
  - iso-10816
  - iso-13373
  - iso-14224
  - iso-17359
  - oreda
  - vibration-analysis
  - bearing-fault-detection
  - bpfo
  - bpfi
  - mcsa
  - motor-current-signature-analysis
  - rul
  - remaining-useful-life
  - oee
  - sensor-fusion
  - anomaly-detection
  - smart-factory
  - industrie-40
  - centrifugal-pump
  - induction-motor
  - cnc
pretty_name: MGG-001 — Factory Sensor Dataset (Sample)
size_categories:
  - 10K<n<100K

MGG-001 — Factory Sensor Dataset (Sample)

A schema-identical preview of MGG-001, the XpertSystems.ai synthetic factory IoT sensor cohort dataset for predictive maintenance ML, condition monitoring research, bearing fault detection, motor current signature analysis, RUL (Remaining Useful Life) prediction, and Industrie 4.0 manufacturing analytics. The full product covers 200 assets × 90 days × 15-min cadence (~17M sensor observations). This sample is HF-sized at 12 assets × 45 days × 1-hour cadence.

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 MGG-001 does — and how it opens a new XpertSystems vertical

MGG-001 is the first Manufacturing & Industrial Systems SKU in the XpertSystems catalog, complementing our existing Oil & Gas vertical (17 SKUs) and Healthcare/Neurology vertical (10 SKUs). Where Oil & Gas targets upstream/midstream operators and Healthcare targets pharma R&D, Manufacturing targets a different industrial buyer category: discrete manufacturers, plant maintenance teams, industrial IoT platforms, and the AI-for-manufacturing ecosystem.

Vertical SKUs Primary Audience
Oil & Gas 17 Upstream/midstream operators, ISO 14224 / API 689 / OREDA users
Healthcare/Neurology 10 Pharma R&D, clinical trial design, biomarker validation
Manufacturing 1+ (this is MGG-001) Plant maintenance, MES/CMMS vendors, AI-for-manufacturing, Industrie 4.0

The dataset captures physics-based, temporally correlated sensor streams from 8 industrial equipment types covering the full Industrial IoT (IIoT) sensor stack: temperature, vibration (with ISO 10816 zone classification + bearing fault frequencies BPFO/BPFI/BSF/FTF), pressure, flow, electrical (3-phase + power quality), process, oil/lubricant, and health/RUL indicators.

Buyer Persona Use Case
Predictive Maintenance Platform Sensor fusion + anomaly detection ML
CMMS / EAM Vendors Failure mode + maintenance recommendation ML
Industrial IoT Platforms Multi-protocol (MQTT/OPC-UA/Modbus) data modeling
Industrie 4.0 Researchers Digital twin training data
Vibration Analysis Specialists ISO 10816 zone classification + BPFO/BPFI ML
Motor Current Signature Analysis MCSA sideband + broken rotor bar detection
RUL Prediction Researchers Weibull-degradation + sensor trajectory ML
Bearing Manufacturers (SKF, Schaeffler) Bearing fault progression simulation
Smart Manufacturing Analytics OEE + availability + performance + quality ML

What's inside — three related CSV files

MGG-001 is a multi-table relational dataset (similar pattern to HC-NEU-004 Multiple Sclerosis). Three CSV files share asset_id as join key.

File Rows (sample) Columns Size
mgg001_equipment_registry.csv 12 25 ~3 KB
mgg001_sensor_data.csv 12,960 79 ~6.7 MB
mgg001_failure_events.csv 0–3 7 ~250 B

Schemas are provided in three matching JSON files:

  • MGG_001_registry_schema.json
  • MGG_001_sensor_schema.json
  • MGG_001_failure_schema.json

Sensor schema module structure (79 columns total)

Module Cols Sensors
Identification 8 asset_id, plant_id, line_id, cell_id, asset_type, timestamp, shift_id, frequency_hz
Temperature 8 bearing DE, bearing NDE, motor winding, coolant in/out, ambient, delta, thermal gradient
Vibration 13 overall RMS, peak, crest factor, kurtosis, 1× RPM, 2× RPM, BPFO, BPFI, BSF, FTF, high-freq dB, axial RMS, ISO 10816 zone
Pressure 5 inlet, outlet, differential, lube oil, pulsation pp
Flow 3 flow rate, NPSH, cavitation index
Electrical 12 3 phase currents, imbalance %, line voltage, PF, P/Q/S, motor load, THD V + I, insulation, MCSA
Process 6 operating RPM, motor load, efficiency, OEE A/P/Q + overall
Oil 4 viscosity, contamination, ISO 4406 code, hrs since lube
Health & RUL 6 health index, RUL hours, RUL CI%, fault mode, severity, anomaly label
Alarms 6 fault probability, alarm level, alert/alarm/shutdown flags, maintenance rec, sensor quality
Network 4 protocol, gateway_id, facility temp, facility humidity

Calibration sources

Every distribution is anchored to named international standards or industry benchmarks. The headline anchors are ISO 10816 (mechanical vibration evaluation), ISO 14224 (reliability/maintenance data collection), and OREDA-2015 (Offshore Reliability Data — MTBF distributions). Other anchors:

  • ISO 10816-3 / ISO 10816-7 — vibration severity zones A/B/C/D for Group 1-4 machines.
  • ISO 13373 — condition monitoring and diagnostics of machines.
  • ISO 17359 — condition monitoring general guidelines.
  • ISO 14224 — petroleum, petrochemical, and natural gas industries: collection and exchange of reliability and maintenance data for equipment.
  • OREDA-2015 — Offshore and Onshore Reliability Data Handbook (5th edition); MTBF, failure mode, and severity distributions for industrial rotating equipment.
  • NEMA MG-1 — Motors and Generators standard; 3-phase current imbalance limits, motor insulation classes (B/F/H), de-rating.
  • IEEE 519 — Recommended Practices and Requirements for Harmonic Control in Electrical Power Systems.
  • IEEE 117 + IEC 60034-1 — motor electrical insulation system temperature classifications.
  • API 610 — Centrifugal Pumps for Petroleum, Petrochemical and Natural Gas Industries.
  • API 619 — Rotary-Type Positive-Displacement Compressors.
  • ISO 4406 — hydraulic fluid power cleanliness code (three-number particle counts).
  • ISA-18.2 + EEMUA 191 + IEC 62682 — Management of Alarm Systems for the Process Industries.
  • Nakajima 1988 + SME Industry Benchmarks — OEE (Overall Equipment Effectiveness) framework and real-world benchmarks.
  • SKF Bearing Manual + ISO 281 — bearing geometry constants (BPFO, BPFI, BSF, FTF) for 6205, 6305, 6206, NU205 bearings.
  • Pump Affinity Laws — Q ∝ N, H ∝ N², P ∝ N³ scaling relationships.

Validation scorecard

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

ID Metric Target Observed Source
M01 ISO 10816 Zone A Share 0.47–0.97 0.707 ISO 10816-3/7
M02 Bearing DE Temp Mean (°C) 50–80 63.0 ISO 14224 / SKF / API 610
M03 Motor Winding Temp (°C, CEILING ≤180) ≤180 96.1 IEEE 117 / NEMA MG-1 Class F
M04 Current Imbalance (%, CEILING ≤10) ≤10 2.87 NEMA MG-1
M05 Power Factor Mean 0.79–0.95 0.873 IEEE 519 / utility tariffs
M06 Motor Load (% of rated) 55–85 65.4 NEMA MG-1 / IEEE 739
M07 OEE Overall (FLOOR ≥40%) ≥40 67.8 Nakajima 1988 / SME
M08 Alarm Normal Share (FLOOR ≥65%) ≥65 0.806 ISA-18.2 / EEMUA 191
M09 Critical Fault Tail 0.00–0.06 0.014 ISO 13373 / Weibull
M10 Shutdown Alarm Rate (CEILING ≤5.5%) ≤5.5 1.36 EEMUA 191 / IEC 62682

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

Standout calibration: M05 power factor lands at 0.873 vs target 0.87 — 0.003 deviation 🎯. M08 alarm normal share at 80.6% directly matches ISA-18.2's "≥80% normal-state" alarm engineering best practice. M04 current imbalance at 2.87% is comfortably under NEMA MG-1's 10% de-rating threshold (well-balanced plant). M09 critical fault tail at 1.4% reflects effective predictive maintenance catching incipient/minor faults before critical.

Suggested use cases

  • Bearing fault detection ML — BPFO/BPFI/BSF/FTF amplitude bands + envelope analysis × ground-truth fault mode for bearing diagnostic ML training.
  • ISO 10816 zone classification ML — vib RMS + crest factor + kurtosis × A/B/C/D zone prediction.
  • RUL (Remaining Useful Life) prediction — sensor trajectory × Weibull degradation × health index for prognostic ML.
  • Motor Current Signature Analysis (MCSA) — MCSA sideband + THD + current imbalance × broken rotor bar / electrical stator fault detection.
  • Cavitation detection — flow + NPSH + cavitation index + pressure pulsation × cavitation event classification for pump diagnostics.
  • Anomaly detection benchmarks — labeled anomaly flag + multi-sensor fusion for unsupervised + supervised anomaly detection comparison.
  • OEE analytics — A/P/Q decomposition × equipment type × shift/criticality for manufacturing efficiency ML.
  • Multi-protocol IIoT data modeling — MQTT/OPC-UA/Modbus/Profibus/ HART/IO-Link/EtherNet-IP × protocol-specific data quality patterns.
  • Maintenance recommendation engine — sensor state + fault severity × maintenance action prediction (none / monitor / inspection / replacement / shutdown).
  • Digital twin training data — physics-based generation matches thermal response (first-order exponential), pump affinity laws (Q∝N, H∝N², P∝N³), and bearing fault frequencies for digital twin validation.
  • Bearing manufacturer R&D — SKF / Schaeffler / NSK / Timken bearing fault progression modeling for product development.

Loading 

from datasets import load_dataset

registry = load_dataset(
    "xpertsystems/mgg001-sample",
    data_files="mgg001_equipment_registry.csv",
    split="train",
)
sensor = load_dataset(
    "xpertsystems/mgg001-sample",
    data_files="mgg001_sensor_data.csv",
    split="train",
)
failures = load_dataset(
    "xpertsystems/mgg001-sample",
    data_files="mgg001_failure_events.csv",
    split="train",
)

Or with pandas directly:

import pandas as pd
from huggingface_hub import hf_hub_download

reg_path = hf_hub_download(
    repo_id="xpertsystems/mgg001-sample",
    filename="mgg001_equipment_registry.csv",
    repo_type="dataset",
)
sensor_path = hf_hub_download(
    repo_id="xpertsystems/mgg001-sample",
    filename="mgg001_sensor_data.csv",
    repo_type="dataset",
)
registry = pd.read_csv(reg_path)
sensor = pd.read_csv(sensor_path)

# Join sensor data with asset master data
full = sensor.merge(registry, on="asset_id", suffixes=("", "_registry"))

# Per-asset trajectory analysis
for asset_id, sub in sensor.groupby("asset_id"):
    sub = sub.sort_values("observation_timestamp")
    # ... fit degradation trajectory, RUL forecast

Three schema JSON files are bundled for pipeline integration:

import json
schema_sensor = json.load(open("MGG_001_sensor_schema.json"))
schema_registry = json.load(open("MGG_001_registry_schema.json"))
schema_failure = json.load(open("MGG_001_failure_schema.json"))

This dataset is multi-table relational — different from most other XpertSystems HC/OIL SKUs which use single-table architecture. The sensor stream is longitudinal time-series (12,960 hourly observations across 12 assets × 45 days) while the registry is cross-sectional master data (one row per asset) and the failure event log is event-stream (sparse, one row per critical failure).

Schema highlights

mgg001_equipment_registry.csv (25 columns)

Identification & location — asset_id, plant_id, line_id, cell_id, asset_type ∈ {pump_centrifugal, motor_induction, compressor_screw, compressor_reciprocating, cnc_machining_center, conveyor_belt, gearbox, fan_industrial}.

Manufacturer & model — manufacturer ∈ {Siemens, ABB, Grundfos, Atlas Copco, Fanuc, SKF, Emerson, Schneider Electric, WEG, Mitsubishi Electric, Sulzer, KSB, Bosch Rexroth, Parker Hannifin, Nidec}, model_number, serial_number, installation_date, asset_age_years, design_life_years, life_consumed_pct.

Operational metadata — criticality_class ∈ {critical_production, important_production, general_purpose, auxiliary, standby}, iso_protection_class (IP rating), atex_zone (ATEX hazard zone).

Maintenance — last_maintenance_date, maintenance_interval_days, days_since_maintenance, cumulative_operating_hours.

Specifications — nominal_speed_rpm, rated_power_kw, bearing_model ∈ {6205, 6305, 6206, NU205}, lubricant_type.

Networking — protocol ∈ {MQTT, OPC_UA, Modbus_TCP, Profibus, HART, IO_Link, EtherNet_IP}, gateway_id.

mgg001_sensor_data.csv (79 columns)

See README "Sensor schema module structure" table above for all 79 columns organized by module. Key columns:

  • anomaly_label (binary 0/1, ground truth)
  • failure_mode_active ∈ 16 fault modes + None
  • failure_mode_severity ∈ {none, incipient, minor, moderate, severe, critical}
  • vib_iso10816_severity_zone ∈ {A_new, B_acceptable, C_alarm, D_danger}
  • alarm_level ∈ {normal, alert, alarm, danger, shutdown}
  • maintenance_recommendation ∈ {none, monitor_increase_frequency, schedule_inspection, schedule_replacement, immediate_shutdown}
  • oil_particle_count_iso4406 (3-number ISO 4406 cleanliness code, e.g., "17/15/12")

mgg001_failure_events.csv (7 columns)

asset_id, failure_timestamp, fault_mode, fault_onset_timestamp, days_from_onset_to_failure, asset_type, criticality_class.

Calibration notes & limitations

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

  1. Failure event count is intentionally sparse in the HF preview sample. At seed 42, the 12-asset × 45-day window produces 1 failure event. This reflects real-world MTBF (OREDA-2015) — even with elevated failure_rate=0.20 parameter, the time between asset onset and critical failure is typically 30-60 days. At full scale (200 assets × 90 days × 15-min cadence), the product produces ~10-30 failure events per generation. For demonstration purposes, the sparse event count is consistent with real-world predictive maintenance datasets where critical failures are rare events.

  2. Anomaly rate 18.5% (seed 42) is above the target failure_rate=0.20 parameter. The failure_rate parameter controls per-asset probability of having a fault state during the observation window; the observed anomaly_label rate aggregates faulty-asset observation counts. For larger-window simulation, the two converge.

  3. Asset type weights deviate from configured targets at n=12 (small-sample variance). At full scale (200 assets), the distribution closely matches CONFIG weights (pump_centrifugal 25%, motor_induction 20%, compressor_screw 15%, etc.).

  4. The sensor stream uses 1-hour cadence in this sample. The full product supports 1-min / 5-min / 15-min / 1-hr sampling — and for high-frequency vibration FFT analysis, 1-min cadence is required. For ML training, 15-min cadence is typically sufficient.

  5. Bearing fault frequencies (BPFO, BPFI, BSF, FTF) are calculated correctly for the 4 supported bearing models (6205, 6305, 6206, NU205) using ISO 281 geometry constants × shaft frequency. For bearings outside these 4 types, the full product supports custom bearing geometry input.

  6. OEE quality component 90.94% is elevated above world-class >99% benchmarks. The generator's quality model is conservative; for process-industry quality modeling, the full product calibrates per industry-specific yield rates.

  7. Pump affinity law scaling (Q ∝ N, H ∝ N², P ∝ N³) is applied to pump_centrifugal asset types only. For positive-displacement compressors, different scaling applies (constant volumetric output below cavitation).

  8. Thermal response model uses first-order exponential (single time constant τ ~2 hr). More complex thermal models with multiple time constants (bearing + winding + frame) are available in the full product.

  9. Multi-protocol IIoT is represented as a metadata field (protocol); protocol-specific data quality patterns (packet loss, latency, dropouts) are not simulated in this preview. For protocol-aware data quality modeling, the full product includes protocol-specific error injection.

  10. Deterministic seeding. Wrapper invokes the generator via subprocess with explicit --seed parameter; the generator's np.random.default_rng(seed) and random.seed(seed) ensure full reproducibility. Seed sweep verifies Grade A+ across {42, 7, 123, 2024, 99, 1}.

Commercial / full product

The full MGG-001 product covers 200 assets × 90 days × configurable sampling cadence (1-min to 1-hr) producing ~17M sensor observations with refined failure event density (10-30 events per generation), configurable cohort enrichment (high-failure / low-failure / balanced), protocol-specific data quality patterns (MQTT broker disconnects, OPC-UA session timeouts, Modbus polling drops), multi-bearing configurations beyond the 4 default types, custom asset_type extensions (compressors, agitators, mixers, crushers, robots), refined OEE quality model per industry vertical (discrete vs process vs hybrid), and pre-built feature engineering pipelines for time-series ML (rolling statistics, FFT decomposition, envelope analysis, MCSA spectral features). 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) and Healthcare/Neurology (10 SKUs, ENROLL-HD/PRO-ACT/TRACK-HD/CLARITY-AD clinical trial calibration). Catalog: huggingface.co/xpertsystems.