CNC Tool Condition Image Dataset (ELTE-TCM-46k)

This dataset contains approximately 46,000 high-resolution TIFF images of 116 unique CNC cutting tools (including drills, end mills, and chamfer tools) under various conditions. The images were captured to support research in automated, direct Tool Condition Monitoring (TCM).

The primary purpose of this dataset is to facilitate the novel methodology presented in Falah, Andó, & Szekeres (2025), which transforms a sequence of 2D images of a rotating tool into a 1D signal of its projected area for periodicity-based fracture detection.

However, the dataset is also a valuable resource for general machine vision and deep learning applications in smart manufacturing, such as image classification, segmentation, and anomaly detection for tool defects.

IMPORTANT NOTE: This is a preliminary release of the dataset. It will be gradually updated with more tools in various conditions. The final version will be available by [January, 2026].

Supported Tasks

This dataset is designed to support a range of tasks, from signal processing to deep learning.

• Primary Task (Signal Processing): Replicating the methodology of Falah, Andó, & Szekeres (2025). This involves:

  • Image processing to segment the tool silhouette from each frame.
  • Quantifying the projected area (pixel count) for each rotational degree.
  • Generating a 1D signal and performing statistical analysis on its periodicity to detect fractures, attached chips, and Built-Up Edge (BUE).

• Secondary Tasks (Machine Vision / Deep Learning):

  • Image Classification: Training classifiers to distinguish between tool states (e.g., intact, fractured, bue).
  • Image Segmentation: Training models to precisely segment the tool, the cutting edge, or the defect area.
  • Anomaly Detection: Developing models that can identify a defective tool image when trained primarily on healthy examples.

Data Instances

Each instance is a single .tiff image of a CNC tool at a specific angle of rotation. A full 360-degree rotation of a single tool is captured across approximately 360-400 frames.

Curation Rationale

This dataset was created to address the challenges of direct Tool Condition Monitoring (TCM) in industrial environments.

• Curated by: Alireza Falah, Soninkhuu Baatarchuluun, with invaluable guidance and supervision from Dr. Mátyás Andó and Dr. Béla Szekeres at the Faculty of Informatics, Eötvös Loránd University (ELTE), Budapest, Hungary.
• License: cc-by-4.0

Source Data & Acquisition

The images were captured using the following professional-grade experimental setup:

• Camera: Basler a2A2600-64ucBAS industrial color camera (5-megapixel).
• Lens: VS Technology VS-LDV75 fixed focal length lens.
• Acquisition Protocol: The tool was rotated at a constant 5 RPM, and images were captured at 30 FPS. This results in each frame corresponding to approximately one degree of rotation.

Citing This Work

If you use this dataset or our methodology in your research, please cite both the dataset and our paper.

Citing the Dataset

@misc{falah_2025_elte_tcm_46k,
  author    = {Falah, Alireza and Baatarchuluun, Soninkhuu and Andó, Mátyás and Szekeres, Béla},
  title     = {{{ELTE-TCM-46k}}: A CNC Tool Condition Image Dataset},
  year      = {2025},
  publisher = {Hugging Face},
  journal   = {Hugging Face repository},
  doi       = {10.57967/hf/6145},
  url       = {https://huggingface.co/datasets/alirezafalah/ELTE-TCM-46k}
}

Licensing Information

This dataset is licensed under the Creative Commons Attribution 4.0 International License (CC BY 4.0).

Dataset Card Contact

[email protected]

alirezafalah.github.io