Industrial Bearing Thermography Dataset

Data Product Features

• Thermographic Image (Radiometric): Infrared image containing full temperature information per pixel.
• Visible Spectrum Image: RGB image captured simultaneously for contextual understanding.
• Temperature Matrix (CSV): Pixel-wise temperature values corresponding to the thermographic image.
• Condition Label: Classification of the bearing state (healthy or faulty), derived from the folder structure.
• Acquisition Date: Timestamp or date of data capture.
• Inspection Batch: Identifier corresponding to each acquisition campaign.

Distribution

• YYYY_MM/
  • Healthy/
    • Infrared Images/
    • Temperature Matrices/
    • Visible Spectrum Images/
  • Faulty/
    • Infrared Images/
    • Temperature Matrices/
    • Visible Spectrum Images/

• Healthy: Bearings operating under normal conditions.
• Faulty: Bearings exhibiting known defects or abnormal behavior.

• Infrared Images: Radiometric thermographic files captured with the thermal camera.
• Temperature Matrixes: CSV files containing pixel-wise temperature values.
• Visible Spectrum Images: Standard RGB images captured alongside thermographic data.

• One visible image
• One temperature matrix CSV

• Radiometric images: JPEG (Radiometric info included)
• Visible images: JPEG
• Temperature data: CSV (2D matrix of float values in °C)

• Data Volume: 220 MB
• Number of records: 1326; 3 files per record (thermal + visible + CSV)

Usage

• Development of machine learning models for fault detection in industrial bearings using thermography.
• Predictive maintenance systems based on thermal pattern recognition.
• Multimodal AI combining visual and thermal data for industrial diagnostics.
• Benchmarking and validation of anomaly detection algorithms.
• Research in condition monitoring and industrial AI.

License

AI Training Rights

• Use the Data Product to train, fine-tune, and evaluate machine learning models, including large language models.
• Incorporate Data Product content into models and commercialize resulting model outputs.
• Create derivative works (model weights, embeddings, etc.) for any lawful purpose.

• The Data Product itself may not be sold, redistributed, or shared outside of licensed usage.
• Licensee must comply with all applicable laws, including data protection and privacy regulations.

Who Can Use It

• Data Scientists: For training machine learning models for anomaly detection and classification.
• Researchers: For studies in thermography, industrial monitoring, and AI.
• Industrial Companies: For developing predictive maintenance solutions.
• Technology Providers: For building diagnostic tools or integrating into monitoring platforms.

Additional notes

• The data has been acquired using a Hikmicro B20 thermal camera, ensuring consistent acquisition parameters across the dataset.
• The availability of both radiometric data and raw temperature matrices allows for flexible processing pipelines, from image-based deep learning to physics-based analysis.
• The dataset supports both supervised and unsupervised learning approaches, including anomaly detection, classification, and segmentation tasks.
• Real industrial variability (ambient conditions, load changes, installation differences) is inherently captured, increasing model robustness in real-world deployments.
• The dataset’s folder-based labeling strategy ensures traceability and minimizes annotation ambiguity, as condition labels are assigned at acquisition time rather than post-processed.