CEE students Hossein Kabir and Sunav Dahal, along with CEE assistant professor Nishant Garg, have developed an automated test for measuring water absorption in concrete samples. Their new, computer vision-based method will reduce time and labor required to complete testing and enhance the efficiency of researchers, technicians and practitioners who must gauge absorption regularly in their work.
Measuring concrete’s tendency to absorb water, a property called ‘sorptivity’, is standard procedure used in evaluating overall durability of concrete. Sorptivity provides insight into how concrete will react to threats like corrosion, salt-attack and freeze-thaw, making it an important marker for determining how durable the concrete will be in the long-term.
The American Society of Testing and Materials (ASTM) develops and publishes the consensus procedure for measuring sorptivity in concrete. Their C1585 test is effective in measurement, yet somewhat inefficient in practice, as it requires significant time and effort to complete. The Garg group’s computer vision-based alternative addresses these shortcomings by automating the tedious procedure of the ASTM’s current standard.
To develop the new test, the researchers trained a custom computer vision model on over 6000 images so that it is capable of automatically detecting water levels in samples every minute. Further training occurred with 1440 unique data points encompassing 15 different mixtures with diverse water/cement ratios and curing periods. The resulting model can predict initial and secondary sorptivity with high confidence, while also allowing for researchers and practitioners to carry on with other facets of their work as the test runs automatically.
While they have found success with the preliminary model, Kabir, Dahal and Garg continue searching for ways to broaden the automated test’s scope and practical application. Next steps could include training the model on larger datasets to improve accuracy across diverse conditions, enhancing its ability to take reliable measurements when viewing adjacent samples simultaneously, and comparing the model’s 2D detection of moisture flow with results recorded using advanced 3D imaging techniques.
“The entire test was built on two $30 cameras purchased from Amazon,” Garg said. “Thus, one can fully automate the sorptivity measurement in a low-cost ($60) manner - anywhere in the world. This accessibility is significant for labs which don’t have access to expensive imaging equipment. Broadly speaking, if we can automate and accelerate the performance testing of construction materials, we can significantly increase the rate of development and deployment of new materials in the field.”
The full paper, “Automated Estimation of Cementitious Sorptivity via Computer Vision” was published in Nature Communications on November 15, 2024 and can be read here.
Watch a research visualization: