Built by an IoT practitioner with real hardware deployment experience. Data covers 6 sensor types across 50 machines and 4 production lines, with realistic normal behavior and 7 anomaly patterns.

📊 Statistics

Split	Examples	Normal	Anomalies
Train	40,000	35,200	4,800
Val	5,000	4,400	600
Test	5,000	4,400	600
Total	50,000	44,000	6,000

Anomaly rate: 12%
Format: JSONL (one example per line)

🔧 Sensor Types

Sensor	Unit	Normal Range
`temperature`	celsius	15–35°C
`humidity`	percent	30–70%
`pressure`	bar	0.8–1.2 bar
`vibration`	mm/s	0.1–4.5 mm/s
`current`	ampere	2–18 A
`flow_rate`	liter/min	5–50 L/min

🚨 Anomaly Types

Type	Description	Severity
`spike`	Sudden brief value peak outside normal range	Medium
`dropout`	Complete signal loss from sensor	High
`drift`	Progressive out-of-range deviation	Low
`stuck`	Sensor value frozen	Medium
`noise_burst`	Excessive signal noise	Low
`critical_high`	Critical upper threshold reached	Critical
`critical_low`	Critical lower threshold reached	Critical

📋 Schema

Each example contains the following fields:

{
  "id": "iot_00000001",
  "timestamp": "2025-01-01T00:02:50Z",
  "device_id": "DEV_4821",
  "machine_id": "MACHINE_012",
  "production_line": "LINE_B",
  "sensor_type": "temperature",
  "location": "machine_room",
  "value": 23.4,
  "unit": "celsius",
  "normal_range_low": 15.0,
  "normal_range_high": 35.0,
  "firmware_version": "v2.3.1",
  "battery_percent": 87.4,
  "signal_strength": -62.3,
  "is_anomaly": false,
  "anomaly_type": null,
  "severity": null,
  "description": null,
  "label": "normal_temperature"
}

🏗️ Infrastructure Context

Data simulates a French manufacturing environment:

50 machines across 4 production lines (LINE_A, LINE_B, LINE_C, LINE_D)
Locations: machine_room, server_room, warehouse, outdoor, production_floor, cleanroom, hydraulic_system, motor, pump, conveyor...
Firmware versions: realistic versioning per device
Signal strength & battery: included for wireless sensor realism

🎯 Use Cases

Predictive maintenance models
Industrial IoT anomaly detection
Edge ML model training
Manufacturing quality control
Time-series anomaly benchmarking

📦 Access

This dataset is gated — request access using the button above.

Access is free. Once approved, you will receive a download link with:

✅ Full 50,000 examples (train / val / test splits)
✅ Python generation script
✅ Commercial use license (CC BY 4.0)

📜 License

Creative Commons Attribution 4.0 International (CC BY 4.0)

Commercial use allowed with attribution.

📬 Contact & Custom Datasets

Need a custom dataset with specific sensor types, volume, or format?

📧 Contact: synthlab.datasets@gmail.com

Downloads last month: -

Total file size:

19.5 MB