Concrete Aggregate PSD Imaging Dataset

Dataset Summary

This repository contains the data related to the paper

"Coenen, M., Beyer, D., Mohammadi, S., Meyer, M., Heipke, C., and Haist, M. (2026): Automating Concrete Production Control with Computer Vision-based Aggregate Characterisation. In: Automation in Construction."

This dataset provides image data of concrete aggregate for the task of estimating particle size distributions (PSD) using computer vision. The images were captured using a controlled camera setup installed above a conveyor belt in a concrete mixing research facility. Each image has a corresponding text file containing the ground-truth PSD obtained through mechanical sieving.

Data Acquisition

Data was recorded at a medium-scale concrete mixing plant equipped with:

• Two Allied Vision Alvium 1800 C-508 cameras
• 25 mm focal length → fine aggregate (0–2 mm)
• 12 mm focal length → coarse aggregate (>2 mm)
• Global shutter, 1 ms exposure time
• LED panel illumination for motion-blur-free imaging
• A sensor mount above the conveyor belt transporting the aggregate

This setup enabled consistent imaging conditions with sufficient resolution for particle analysis.

Datasets

Two datasets were created to cover common aggregate size ranges used in concrete production:

𝑀ᶠⁱⁿᵉ — Fine Material (< 2 mm)

• 16 material samples
• Natural river sand
• PSDs synthetically varied by mixing pre-fractionated material

𝑀ᶜᵒᵃʳˢᵉ — Coarse Material (2–16 mm)

• 26 material samples
• 16 natural river gravel
• 10 recycled concrete aggregate (RCA)

Each material sample weighs 150 kg, and its PSD was systematically varied to cover a broad range of grading curves.

Note:
This repository contains only a subset of the data set that was used in the paper. In order to receive the full data set, please reach out to the authors. The dataset represents controlled variability. While this is ideal for benchmarking and model development, real industrial plants may exhibit additional stochastic variability.

Reference PSD Measurement

A 10 kg subsample from each material batch was mechanically sieved to obtain the reference PSD.

Each PSD is represented using six particle size intervals (B = 6):

• Fine dataset:
  0.063, 0.125, 0.25, 0.5, 1.0, 2.0 mm
• Coarse dataset:
  0, 2, 4, 8, 11.2, 16 mm

Each .txt reference file contains six percentile values that sum to 1.0.

Use Cases

This dataset is intended for:

• PSD estimation using deep learning or classical CV
• Regression and distribution prediction tasks
• Material characterization and granulometry research
• Benchmarking computer vision methods on granular material datasets

Citation

If you use this dataset in academic or industrial research, please cite the corresponding paper:

Coenen, M., Beyer, D., Mohammadi, S., Meyer, M., Heipke, C., and Haist, M. (2026): Automating Concrete Production Control with Computer Vision-based Aggregate Characterisation. In: Automation in Construction.