Fig. 3: Real-time prediction of penetration/ sorptivity values via our setup.

a Predicted penetration-time values obtained using our dual-camera setup (Camera 1 = blue points/masks, Camera 2 = green points/masks) for a denser matrix, (b) alongside their corresponding time-lapse images. c Predicted penetration-time values for a more porous matrix, (d) along with their corresponding time-lapse images. The use of a single-camera system may result in sporadic disparities between the weighing scale readings and the computed sorptivity values. This is due to the variability in sorptivity values, which are influenced by factors such as tortuosity and capillary pore distribution in different areas of the paste specimens, thus reducing reliability. The dual-camera system thus enhances the reliability of absorption measurements through computer vision. In this subplot, ‘Pre’ and ‘Post’ refer to the original and FPN-analyzed images, respectively. e Time-lapse of paste specimens with varying porosities, showing real-time sorptivity analyses. f In more porous matrices (higher initial sorptivity), there is a rapid evolution in the water level (∆ area/√min) over time, while in denser matrices, water levels increase more gradually. Our machine learning successfully combines water level data (predicted by machine vision) and absorption time to accurately estimate cementitious absorption rates. Source data are provided in the Source Data file.