Industrial Bearing Thermography Dataset

This data product consists of a structured dataset of thermographic inspections collected from bearings under real operating conditions in industrial environments. The dataset captures thermal behavior influenced by load variations, lubrication states, and degradation effects. Each acquisition includes: - Radiometric thermographic image (with temperature per pixel) - Visible spectrum image - Temperature matrix (CSV) - Condition label (healthy or faulty) - Acquisition date and inspection batch The dataset is organized in a hierarchical structure by acquisition campaigns (YYYY_MM), with clear separation between healthy and faulty conditions. Each thermographic image has a one-to-one correspondence with its associated visible image and temperature matrix. The dataset contains 1,326 records (approximately 220 MB), with three files per record.

Data Quality

The dataset is composed of real-world industrial data acquired in situ during routine inspection activities, ensuring high realism and practical relevance. Key quality aspects include: - Consistent acquisition using a Hikmicro B20 thermal camera - Radiometric images providing accurate temperature measurements per pixel - Relevant environmental information included in temperature matrices - Direct alignment between thermal images, visible images, and temperature matrices - Clear labeling (healthy vs faulty) derived from acquisition context - Structured organization ensuring traceability and data integrity The dataset captures real operational variability, improving robustness for machine learning applications.

Use Cases

Machine learning model development for bearing fault detection - Predictive maintenance systems based on thermal analysis - Thermal anomaly detection in rotating equipment - Multimodal AI applications combining thermal and visual data - Benchmarking and validation of anomaly detection algorithms - Research in industrial condition monitoring and AI

Limitations

- The dataset is not a continuous time series (data is collected in discrete campaigns) - Fault types are not explicitly categorized beyond “faulty” label - Data is limited to specific industrial environments and equipment setups - Load variations are not explicitly annotated

Methodology

Data was collected in industrial manufacturing environments from bearings operating under normal working conditions. Thermographic inspections were performed in situ using a Hikmicro B20 thermal camera during routine inspection activities. Each acquisition includes: - A radiometric thermal image - A corresponding visible image - A temperature matrix exported as CSV Data is organized by acquisition campaigns (YYYY_MM), with manual or operational labeling of equipment condition (healthy or faulty) based on inspection context. No artificial manipulation or laboratory conditions were introduced, ensuring that the dataset reflects real-world industrial behavior.