Determining the carbonation depth is of paramount importance in assessing the durability of cementitious materials as the carbonation process alters their near-surface physicochemical properties. This paper introduces a novel quantitative ultrasound (QUS) technique capable of nondestructively identifying the carbonation depth, regardless of the tortuosity of the carbonation front. A 2.5 MHz phased array transducer was employed to construct beamformed radio frequency images and measure two microstructure-dependent parameters – spectral slope (SS) and spectral intercept (SI) – based on backscatter and attenuation coefficients. The change in obtained SS and SI were then visualized as QUS images, which displayed the geometry of the carbonation front, correlating with images constructed using a phenolphthalein solution (1.6 mm difference in depth). Mineral compositions and porosities were analyzed to account for microstructural changes in each layer. This study opens an opportunity for nondestructive detection of carbonation depth, overcoming the limitations in conventional methods or regression-based approaches.

Introduction

In the cement sector, the carbonation process has been considered as one of the strategies for mitigating CO2 emissions owing to its ability to capture atmospheric CO2 [1]. The main mechanism underlying the carbonation process is the reaction between hydration products (e.g., Ca(OH)2) and atmospheric CO2, trapping CO2 into concrete structures. Despite a long-term durability issue caused by the drop in pH (<9), which initiates the corrosion of reinforcements, the carbonation process holds the potential to induce a synergistic effect on improving the durability performance via the carbonation products (e.g., CaCO3), such as densification of porosities [[2], [3], [4]], crack healing [5,6], and occasionally strength enhancement [[7], [8], [9]]. For this reason, extensive research has been conducted to elucidate the essential features of carbonation regarding pH level [10,11], mineral composition [2,12,13], microstructure [14,15], density [[16], [17], [18]], portlandite depletion [19], and carbonation shrinkage [20,21] in cementitious materials (near the surface).

In particular, understanding the spatial distribution of the carbonation front also called carbonation depth, is of importance because the physicochemical properties of the carbonated area may be different from those of the uncarbonated area, similar to a layered material. To estimate the carbonation front, many prediction methods have been developed based on analytical and numerical approaches considering gas diffusion and hydration kinetics. Some examples include a mathematical model for analyzing the carbonation process in concrete [22,23] and a spatial profile of the carbonation front using thermogravimetry (TG) or Quantitative X-ray diffraction (QXRD) analysis with a carbonation model [24]. The evidence in these references well accounts for the chemical reactions as well as the flow of gas and moisture. Recently, a new experimental method that visualizes the diffusion of alkalis and sulfur has been introduced, taking into account the sharpness and tortuosity of the carbonation front [25]. Nevertheless, most of the approaches introduced here have been conducted in a destructive manner and rely on the use of additional samples, only suitable for limited sample dimensions in the laboratory. Recently, prediction models based on machine-learning (ML) algorithms that consider key features of the carbonation process, e.g., exposure time and concentration of CO2, binder amount, aggregates, and water-to-binder ratio, have been developed for determining the carbonation depth [[26], [27], [28], [29], [30]]. While the ML-based approaches show some promising opportunities for linking the change in the mixture and properties with the carbonation depth, there remain important unresolved issues. For example, ML-based approaches lack the ability to consider the roughness at the carbonation front and the differences caused by the spatial distribution of aggregates and pores in the vicinity of carbonation front. Moreover, the prediction accuracy can vary depending on the quality/quantity of training data, causing a huge difficulty in collecting the input data.

Read the full article at Science Direct.