Automated estimation of cementitious sorptivity via computer vision

Area of Science:

• Materials Science
• Civil Engineering
• Computer Vision

Background:

• Monitoring water uptake is vital for assessing cementitious material durability against environmental factors.
• Current methods like ASTM C1585 rely on manual, infrequent weight measurements, limiting real-time analysis.
• Developing automated, accurate methods for water absorption monitoring is essential for construction material science.

Purpose of the Study:

• To develop and validate a computer vision model for automated, real-time monitoring of water uptake in cementitious systems.
• To replace labor-intensive traditional methods with an efficient, automated approach for durability assessment.
• To enable accurate prediction of sorptivity values for enhanced material performance evaluation.

Main Methods:

• A custom computer vision model was trained using 6234 images (4000 real, 2234 synthetic).
• The model was trained on 1440 data points from 15 paste mixtures (w/c ratios 0.4-0.8, curing 1-7 days).
• The model automatically detects water levels in prismatic samples, estimating water penetration every minute.

Main Results:

• The computer vision model accurately predicts initial and secondary sorptivities in real time (R² > 0.9).
• Achieved high confidence in predictions after training on diverse paste mixtures.
• Demonstrated successful application on mortar and concrete systems.

Conclusions:

• The developed computer vision model offers a low-cost, automated solution for monitoring water uptake.
• This technology facilitates real-time durability assessment of construction materials.
• Paves the way for improved quality control and material development in the construction industry.