A deep learning framework to estimate water-to-cement ratio in mortar exploiting frequency domain reflectometry sensors

Water-to-cement ratio (WCR) is a crucial factor that directly affects the strength and durability of cementitious materials, such as mortar and concrete. Existing methods for estimating WCR often take a considerable amount of time or require expensive equipment, limiting their practicality on actual construction sites. In this work, we propose a deep learning framework to estimate WCR using a cost-effective Frequency Domain Reflectometry (FDR) sensor and a deep model, WCRnet, which leverages residual connections. The proposed method was evaluated on mortar samples with varying WCRs, and the results demonstrated that WCRnet significantly outperforms machine learning models and other conventional methods in both accuracy and speed, achieving an R${}^2$ of 0.9627, root mean square error (RMSE) of 1.2677% and a computation time of 1.9158ms. This approach offers a practical, user-friendly, and reliable solution for on-site WCR estimation, highlighting its potential applicability in the construction industry for enhanced quality control and safety. The code used in our research is publicly available at https://github.com/Hanyang-Robot/WCRnet.