Data-driven approach for water-to-cement ratio prediction in fresh cement paste from raw EIS measurements

초록

Maintaining a consistent water-to-cement (w/c) ratio is critical for the strength development and long-term durability of cementitious materials; however, reliable on-site assessment remains challenging due to environmental variability and uncontrolled water addition. This study proposes a non-destructive, data-driven approach for directly estimating the w/c ratio of fresh cement paste by integrating electrochemical impedance spectroscopy (EIS) with a Gradient Boosting model. A total of 538 impedance spectra were collected under controlled laboratory conditions across a w/c range of 0.30-0.45 at early hydration stages. Raw impedance features measured within the 250 kHz-1 Hz frequency range were analyzed without relying on equivalent circuit fitting, and the proposed model achieved a prediction accuracy of up to R2 = 0.85. Statistical preprocessing using median absolute deviation (MAD) filtering improved spectral stability, while frequency-window specification was shown to be critical for robust w/c estimation. SHapley Additive exPlanations (SHAP) analysis further revealed that the imaginary impedance component (Z(y)) and the frequency region near 1 kHz dominate the model predictions, reflecting sensitivity to interfacial polarization and ionic relaxation processes associated with early-age microstructural conditions. The proposed EIS-machine learning framework enables a rapid and physically interpretable estimation of the w/c ratio at the paste scale and provides a foundation for future extension to mortar and concrete for practical quality control applications.

키워드