Abstract
As the construction industry advances toward more efficient methods for assessing durability, the need for automated sorptivity evaluation has become increasingly critical. Consequently, this study introduces SorpVision, a dataset of 7,384 images (5,000 real and 2,384 synthetic) designed to support our custom computer vision-based framework for automated sorptivity evaluation in cementitious materials. Traditional methods, such as ASTM C1585, depend on manual weighing, which is time-consuming and limits measurement intervals. SorpVision, combined with a cost-effective USB camera setup and a robust vision algorithm, facilitates real-time water level detection in cementitious systems. The framework, trained using 1,440 data points from pastes with water-to-cement (w/c) ratios of 0.4-0.8 and curing durations of 1-7 days, achieves high predictive accuracy for initial and secondary sorptivities (R(2) > 0.9 for cement pastes). Moreover, it generalizes well to mortar and concrete, yielding R(2) values of 0.96 and 0.87 for initial sorptivity and 0.74 and 0.65 for secondary sorptivity, respectively. SorpVision offers an accurate, data-driven foundation for scalable, automated durability evaluations, supporting sustainable infrastructure development.