Research on the relationship between fiber dispersion and performance of engineered cementitious composites based on image recognition

To address the limitations of current ECC fiber-dispersion evaluation—namely reliance on subjective visual inspection and semi-automatic counting, and the difficulty in characterizing local clustering/void regions—this study proposes an automated fluorescence-microscopy image analysis method for multi-scale quantification of fiber dispersion. Grayscale conversion, Otsu thresholding, and morphological operations are applied to automatically identify fiber cross-sections and extract centroid coordinates, based on which a global dispersion index (FDC) and a local index, i.e., the coefficient of variation of nearest-neighbor distance (CV_NND), are calculated. Sixteen ECC mixtures were tested under uniaxial tension and compression, and correlations between the dispersion indices and mechanical properties were established. The results show that the ultimate tensile strain is strongly positively correlated with FDC (R² = 0.884) and strongly negatively correlated with CV_NND (R² = 0.94), indicating that a more uniform fiber distribution promotes distributed multiple cracking and sustains strain-hardening behavior. In contrast, the initial cracking strength and ultimate tensile strength exhibit weak correlations with the dispersion indices (R² ≤ 0.04). Leave-one-out analysis further confirms the robustness of the high-correlation relationship. Compared with conventional fluorescence observation and single-metric statistical dispersion approaches, the proposed method enables automated, reproducible, and locally interpretable quantification of fiber dispersion in ECC.