Image Analysis of Air Voids in Air-Entrained Concrete

Abstract

The application of image analysis techniques to characterize the air-void system in hardened concrete is demonstrated. Both lineal and areal feature analyses are investigated. Feature size distributions and total air contents are obtained using s both types of analysis. The areal analyses also include the measurement of individual feature perimeters for use in comparing void shapes. A two phase standard specimen ip developed to insure the consistency of measurements and repeatability of results. Correction methods, based on geometric probability, are developed to remove the distortions in the image analysis data resulting from frame edge effects. Separate r:nethods are presented for lineal and areal analyses. Using discrete class sizes, both correction procedures are expressed in a matrix format. The corrected areal feature distributions are used to obtain volume distributions of spherical air voids using standard stereological procedures. The procedures are applied to ten concrete specimens, at magnifications of 12x and 30x. The specimens represent concretes made using three different airentraining admixtures, as well as non-air entrained concrete. Air-void parameters calculated from corrected image analysis results for the ten specimens are compared to results obtained using the modified point count method and to freeze-thaw results obtained from surface scaling tests of companion specimens. The differences in the air-void systems created by the various air entraining agents are studied by comparing different characteristics including: the Powers spacing factor, the Philleo factor, profile shape, average feature size, numerical density of features, and the cumulative percent of total air versus feature size. The study demonstrates that image analysis provides a viable alternative to traditional lineal traverse and modified point count methods for characterization of air-void systems in hardened concrete and, in the process, provides significant detail not available with the traditional methods. The study indicates that airentraining agents produce characteristic air void distributions. Comparisons made in the study show that lineal and areal image analysis techniques provide similar determinations of total air content that are, on average, 1. i 5 volume percent lower than those obtained from a modified point count analysis. Application of the frame edge effect correction procedures to the lineal data results in an average decrease in the total chord density of 2.3% and 5.0% for magnifications of i 2x and 30x, respectively. Application of the frame edge effect correction proced.ures tq the areal data results in an average decrease in the total profile density of 3.3% and 5.9% for magnifications of i 2x and 30x, respectively. The accuracy of the analysis decreases if size classes are much greater than 30 Jlm. Accurate lineal analyses require the class size to be an exact increment of pixel length. A similar requirement does not apply to areal analyses.