Jvachier/industrial-maintenance-synthetic

Industrial Maintenance Synthetic Dataset

Synthetic dataset of 1M sensor tag records and 1M maintenance work orders across 37 industrial equipment types, generated for domain-specific NLP models in industrial maintenance.

Dataset Details

Dataset Structure

Sensor Tags (sensor_tags)

Tag_name ; Server_name ; Description ; Units ; Building ; Production_line ; Equipment_ID ; Equipment_type ; Variable_type

Maintenance Records (maintenance)

Location (building number) ; Region ; WorkOrder ; Equipment_ID ; OrderType ; WorkOrderDescription ; OperationDescription ; MaintenanceWorkCenter ; OperationWorkCenter ; Maintenance_activity_type ; CreatedDate ; ScheduledStartDate ; ScheduledFinishDate ; ActualStartDate ; ActualFinishDate ; TechnicalCompletionDate ; CloseData ; ActualWork ; PlannedWork ; OrderStatus

Work orders cover order types: CM (Corrective), PM (Preventive), PDM (Predictive), EM (Emergency), STAT (Statutory).

Equipment Types Covered

Centrifugal Pump, Compressor, Heat Exchanger, Reactor, Distillation Column, Storage Tank, Boiler, Centrifuge, Evaporator, Crystalliser, Filter, Agitator, Belt Conveyor, Motor, Separator, Dryer, Extruder, Cooling Tower, HVAC Unit, Pressure Vessel, Gas Turbine, Ball Mill, Screw Conveyor, Rotary Dryer, Hydraulic Press, Rotary Lobe Pump, Diaphragm Pump, Dosing Pump, Gas Scrubber, Vacuum Pump, Gearbox, Air Compressor, Steam Turbine, Spray Dryer, Ultrafiltration Unit, Fermentor, Fluid Catalytic Cracker.

Data Quality

This dataset includes intentional impurities (10% rate) to simulate real-world industrial data:

• —Missing values (3% of cells)
• —Typographical errors in text descriptions
• —Unit format variants (degC / °C / Celsius)
• —Equipment abbreviations (CP, HX, COMP)
• —Date anomalies (actual finish before actual start)

This makes the dataset suitable for testing data cleaning pipelines and training robust NLP models.

Usage

from datasets import load_dataset

# Load sensor tags
tags = load_dataset("jvachier/industrial-maintenance-synthetic", data_files="sensor_tags.parquet")

# Load maintenance records
maintenance = load_dataset("jvachier/industrial-maintenance-synthetic", data_files="maintenance.parquet")

Or with pandas:

import pandas as pd

tags_df = pd.read_parquet("sensor_tags.parquet")
maintenance_df = pd.read_parquet("maintenance.parquet")

Generation

This dataset was generated using the open-source pipeline at: [github.com/jvachier/industrial-maintenance-synthetic-data](https://github.com/jvachier/industrial-maintenance-synthetic-data)

Also available on Kaggle (CSV format): [kaggle.com/datasets/jvachier/industrial-maintenance-synthetic-dataset](https://www.kaggle.com/datasets/jvachier/industrial-maintenance-synthetic-dataset)

The generation pipeline uses config-driven templates which can be freely modified to generate custom variants of this dataset.

Intended Use

• —Pre-training domain-specific encoder transformers (MLM)
• —Building and evaluating RAG retrieval pipelines
• —Predictive maintenance NLP research
• —Benchmarking embedding models on industrial text
• —Testing ETL and data quality frameworks

Citation

@dataset{vachier2026industrial,
  author    = {Vachier, Jeremy},
  title     = {Industrial Maintenance Synthetic Dataset},
  year      = {2026},
  publisher = {HuggingFace},
  url       = {https://huggingface.co/datasets/jvachier/industrial-maintenance-synthetic}
}

Author

Jeremy Vachier — jvachier.github.io