Robotic Welding Defect Data With Video And Sensor Streams

Over 4000 annotated samples capture the welding process through video, audio, sensor time-series, and post-weld images. IntelLabs collected this data in an automotive production floor setting with an industry supplier. The dataset was published in September 2025 to support multimodal defect detection research.

Use Cases

• Train video-based anomaly detection models to identify welding defects from the recorded video streams.
• Develop audio signal processing classifiers to detect anomalies in the welding process using the associated audio data.
• Build time-series forecasting models to predict weld quality from the sensor data streams.
• Create multimodal fusion models that combine video, audio, sensor, and post-weld image features for improved defect classification.

Strengths

• Over 4000 annotated samples provide a substantial base for model training.
• Each sample includes five synchronized data modalities: video, audio, sensor time-series, and five post-weld images.
• Data was collected in a real-world automotive production environment, enhancing practical relevance.

Limitations

• The exact number of rows, columns, and total dataset size are unspecified.
• Geographic coverage is limited to a single automotive production facility, which may introduce bias.
• A portion of the data requires a separate license, adding complexity for potential users.

Provenance

Source

IntelLabs, collected in collaboration with an automotive industry supplier.

Collection Method

Gathered on an automotive production floor during robotic welding operations.

Freshness

Last updated in September 2025.

Geography

United States (based on 'Regionus' platform tag).