Optimization of second-derivative ATR-FTIR preprocessing for PLS quantification of calcite-dominant standard in hardened cement paste powder mixtures

This study investigated a method for quantifying calcium carbonate content using Attenuated Total Reflectance Fourier Transform Infrared (ATR-FTIR) spectroscopy and partial least squares (PLS) regression, employing samples prepared by mixing a known proportion of calcite-dominant CaCO3 standard with uncarbonated cement paste powder. Peak area ratio between the ν2-(CO3) and Si-O asymmetric stretching of Q2 in calcium silicate hydrates (C-S-H) from the second derivative spectra demonstrated the potential to simultaneously investigate calcite content and Si bonding state in C-S-H from the same spectrum. This is the motivation of developing calibration curb with second derivative spectra. The influence of zero‑order, first‑derivative, and second‑derivative spectra generated by the Savitzky–Golay (SG) method, together with window length and selected wavenumber ranges (ν2, ν3 and ν4-(CO3)), on the coefficient of determination (R²), model error (RMSE), and partition‑dependent variability (standard deviation) by iteratively varying the training–validation data split, was systematically investigated. Models using ν2-(CO3) (820-920 cm-1) range exhibited higher R2 value, and especially second derivative reduced partition-dependent variability and increase robustness with appropriate number of window length, preserving information of Q2 species of C-S-H. This study presents vibration mode-dependent preprocessing design guidelines and a procedure to evaluate robustness with practical application in mind.