A Multimodal Railway Vibration and Vision Dataset (Rail-VIVID).

Area of Science:

• Engineering
• Data Science
• Materials Science

Background:

• Indirect, onboard monitoring of railway infrastructure is crucial for safety and efficiency.
• Current methods like vibration analysis (subsurface sensitivity) and vision-based inspection (surface detection) have limitations.
• Lack of a benchmark dataset hinders progress in multimodal sensor integration.

Purpose of the Study:

• Introduce the novel, temporally synchronized railroad vibration and vision dataset (Rail-VIVID).
• Provide a benchmark for analyzing the relationship between track anomalies and their signatures across modalities.
• Establish a foundation for developing integrated sensor fusion approaches for robust detection and predictive maintenance.

Main Methods:

• Collected synchronized vibration and vision data from repeated runs over a track segment with known anomalies.
• Ensured controlled conditions for data acquisition.
• Dataset includes multimodal cases for comprehensive analysis.

Main Results:

• The Rail-VIVID dataset enables investigation into anomaly signatures in both vibration and vision data.
• It serves as a benchmark for evaluating anomaly detection algorithms.
• Facilitates research into multimodal sensor fusion for enhanced railway monitoring.

Conclusions:

• The Rail-VIVID dataset addresses the need for a controlled, multimodal benchmark for railway track monitoring.
• It supports the development of advanced, integrated systems for improved safety and predictive maintenance.
• This resource is expected to accelerate research in railway infrastructure health assessment.