AURSAD: Robot Screwdriving Anomaly Detection Measurements

A dataset from Aarhus University containing 2,045 samples of normal and anomalous operation measurements from a UR-3e robot and OnRobot screwdriver, sampled at 100 Hz. Błażej Leporowski created this collection for research into industrial robot anomaly detection, with a technical report available on arXiv. The dataset includes 1,420 normal operation samples and 625 anomalous samples across five specific fault categories.

Use Cases

• Train anomaly detection models based on labeled time-series data for robotic screwdriving.
• Benchmark classification algorithms for specific fault types like damaged screws or missing screws.
• Study temporal patterns in robot sensor data during normal and anomalous assembly operations.

Strengths

• Contains 2,045 total samples with clear class distribution: 1,420 normal (69.44%) and 625 anomalous samples.
• Anomalies are categorized into five specific, labeled types including damaged screws (221 samples) and missing screws (218 samples).
• Includes an accompanying Python library (aursad) for easier data loading and manipulation.
• Data is continuous with events grouped in samples, and a technical report provides detailed methodology.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Last update date is unknown; freshness unverified.

Provenance

Source

Błażej Leporowski, Aarhus University

Collection Method

Measurements collected from a UR-3e robot and OnRobot screwdriver during operation.