Wet Carbonation of Industrial Recycled Concrete Fines: Experimental Study and Reaction Kinetic Modeling

Wet Carbonation of Industrial Recycled Concrete Fines: Experimental Study and Reaction Kinetic Modeling Z. Tabrizi, C. Rodriguez, E. Barbera, W. R. Leal da Silva, and F. Bezzo* Cite This: Ind. Eng. Chem. Res. 2025, 64, 21412−21425 Read Online ACCESS Metrics & More Article Recommendations * sı Supporting Information ABSTRACT: Carbon dioxide mineralization via wet carbonation of industrial Recycled Concrete Fines (RCFs) offers a promising pathway for mitigating emissions in the cement industry, necessitating reliable kinetic models for technology scale-up. This work proposes a validated diffusion-based Shrinking Core Model describing the wet carbonation kinetics of RCFs. The model, based on parabolic diffusion law, is rigorously selected and calibrated among mineralization models in wet systems. Experimental results demonstrate a maximum carbonation efficiency of 0.81, corresponding to 95 kg CO2uptake per tonne of RCFs, and acceptable compressive strength development when incorporating RCFs up to 10% in blended cement. Reaction rates showed a minimal temperature impact due to the offset between the CO2solubility and diffusion through the product layer. Compared to Recycled Cement Paste (RCP) carbonation, higher diffusion coefficients are predicted, likely caused by looser product layer. Analysis highlights the importance of particle size and the CO2partial pressure, providing insights for efficient scale-up. 1. INTRODUCTION The cement industry, often cited as contributing ∼5−8% of global anthropogenic CO2emissions, has become a major focus of research into carbon capture, storage, and utilization (CCUS). 1−3 Among these, CO2mineralization has gained significant interest due to its dual role in decarbonization and valorization of industrial byproducts. 4 By converting CO2into calcium or magnesium carbonates (i.e., CaCO3and MgCO3), this process offers permanent storage with minimal risk of rerelease. 5,6 Moreover, CO2mineralization enhances the value of the supplementary cementitious materials (SCMs) by converting them into carbonated SCMs (cSCMs). 6 Incorporating cSCMs into cement not only reduces the demand for clinker but also contributes to carbon sequestration. Among a wide range of SCMs, RCFs are of particular interest due to their enhanced pozzolanic reactivity upon carbonation, which improves the mechanical properties of blended cements and enables CO2sequestration, contributing to more sustainable concrete materials. 7,8 RCFs are considered emerging SCMs crucial for supporting the decarbonization of the cement industry. 9 They are obtained during the recycling of construction and demolition of concrete and accounted for approximately 35% of the global generation of carbonatable solid materials in 2020, equivalent to 1.4 Gt/y. 10 The recycling process involves collecting, crushing, sieving, and separating to produce aggregates, sands, and fines. Despite the challenge of complete separation of the fine fraction (i.e., cement paste) from aggregates and sands, 11 the RCFs hold significant potential for CO2sequestration, with ca. 70 to 300 kg CO2per tonne of RCFs reported in the literature. 12,13 The upper bound values are typically associated with laboratory-produced RCFs, which mainly represent pure hydrated cement paste; whereas the lower bound values relate to industrial RCFs due to their inherited natural carbonation during service life. 14 If the entire volume of RCFs were carbonated and used as clinker replacement in cement, the carbonated RCFs (cRCFs) could offer a decarbonization potential of up to 0.39 Gt CO2/y, i.e., 15% of that from cement production. 10 Various approaches have been proposed to produce cRCFs, including slurry grinding to refresh reactive surfaces, 15 semiwet processes with additive salts to enhance CO2dissolution, 16 and hyper-gravity carbonation, which accelerates mass transfer and promotes rapid calcite and silica gel formation; 17 nevertheless, wet carbonation remains the baseline for its simplicity, fast kinetics, and high efficiency, with emerging methods offering further improvements. 18 Producing cRCFs and their subsequent use as SCM require optimally designed reactors, supported by kinetic modeling, Received: July 11, 2025 Revised: October 10, 2025 Accepted: October 14, 2025 Published: October 30, 2025 Articlepubs.acs.org/IECR © 2025 The Authors. Published by American Chemical Society 21412 https://doi.org/10.1021/acs.iecr.5c02835 Ind. Eng. Chem. Res. 2025, 64, 21412−21425 This article is licensed under CC-BY 4.0 Downloaded via UNIV OF PADOVA on November 13, 2025 at 13:37:25 (UTC). See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles. lab-to-industrial scale-up, and precisely calculated residence time of species. 19 Mineral carbonation kinetic models generally fall into two groups: surface activity models such as Surface Coverage Model, 20−24 and diffusion-reaction resistance models such as the Basic and Modified Shrinking Core Models (BSCMs, MSCMs). 25 The latter considers more physical variability and incorporates mechanisms such as product layer and gas−liquid film diffusivity, 26 clogging, and diffusivity changes with reaction progress, 26−28 and are distinguished by assumed particle geometry. While these approaches have described the carbonation of SCMs, studies on the carbonation of materials derived from concrete demolition sites remains limited. Existing studies have focused mostly on CaO-rich materials such as RCP or laboratory-produced RCFs, which predominantly consist of hydrated cement paste. These materials serve as a proxy for RCFs but do not account for the long-term environmental exposure of concrete. Thus, there exists a research gap in proposing accurate kinetic model(s) for the carbonation of industrial RCFs. Carbonation modeling of laboratory-produced RCFs together with the effects of water-to-solid ratio has been carried out by Mehdizadeh et al. in a dry-pressurized autoclave using the surface coverage model. 29 This model describes reaction progress as a result of surface deactivation, simplifying the material properties by treating the reaction contact area as the only variable. 22 In another study, Mao et al. investigated RCP carbonation and used the Jander’s model to predict carbonation efficiency; 30 which consists in a simpler diffusionbased model of Shrinking Core Models dominated by product layer diffusion. These models neither explicitly account for material properties beyond specific surface area or particle size nor include key process driving forces such as CO2partial pressure and reaction temperature within the model structure as part of a first-principles approach which limits their applicability. A more generalized form of Shrinking Core Models can account for some of the missed variabilities that are critical for application in realistic reactors. It considers the reaction as a quasi-steady process dominated by either diffusion through the liquid film around the particle, reaction on the surface, or, more relevantly, diffusion through the product layer. 25 The latter can be interpreted as involving a time-varying diffusion coefficient 28 or a parabolic diffusion law 27 rather than the Fick’s law. All these formulations vary based on the initial particle geometry; thermal dependency is typically described using the Arrhenius’ law. Various carbonation systems, dry or wet, can influence the model by accounting for the solubility of CO2in the solvent, rather than directly considering the CO2 partial pressure in diffusion and reaction. 26 Applying such models to describe the carbonation of RCFs could offer a more comprehensive understanding of the influence of the raw material properties and gas phase composition. Additionally, the process of discriminating between various predictive models, selecting the most representative one, and calibrating it to an acceptable level of accuracy�commonly referred to as model identification�has not been adequately addressed in the field of mineral carbonation. This gap is particularly critical given the limited availability of data and the deviations from idealized model assumptions often observed in such systems, which contribute to uncertainties in modeling. A rigorous approach to model identification can help address and reduce these uncertainties by applying the appropriate statistical metrics and numerical techniques. In that light, this study aims to address the identified modeling gaps by proposing a predictive kinetic model for the wet carbonation of industrial RCFs, supported by experiments conducted in a semibatch reactor. The reacting RCFs and its carbonated counterpart (cRCFs) are characterized to confirm their suitability as SCMs and support the formulation of physically representative models. A rigorous approach has been employed to discriminate between candidate models and to calibrate the selected model under low-information conditions. The consistency between model-based parameters and experimental measurements serves as a basis for confirming the validity of modeling assumptions and discussing trade-offs between the influencing mechanisms considering the observed data. Furthermore, these insights are used to propose optimization pathways for the reaction process and guide the design of more informative kinetic experiments. Figure 1. Schematic diagram of the wet semicontinuous carbonation setup. Industrial & Engineering Chemistry Research pubs.acs.org/IECR Article https://doi.org/10.1021/acs.iecr.5c02835 Ind. Eng. Chem. Res. 2025, 64, 21412−21425 21413 2. MATERIAL AND METHODS 2.1. Materials. Waste concrete was obtained after demolition of a structure with 100+ years of service in Denmark. In the field, the debris were screened and classified in coarse aggregate (4−20 mm) and sand (0−4 mm). Subsequently, the sand fraction was sieved in the laboratory to obtain industrial RCFs with a narrow particle size between 45 and 75 μm. For the carbonation experiments, demineralised water and commercial-grade CO2(>99.9%) and N2(99.6%) gases were used in all experiments. 2.2. Experimental Setup. A wet carbonation setup (Figure 1) was commissioned and used to perform the RCFs carbonation experiments. This apparatus was designed to allow for adjustments on gas composition and temperature control of both gas and liquid phases in the reactions. Experiments were carried out at the operational conditions of 25, 45, 65, and 85 °C, ambient pressure, and a solid-to-liquid mass ratio of 10% (corresponding to 20 g of RCFs) and samples were taken at 0, 10, 20, 40, 80, 120, and 180 min during the reaction time. A simulated gas mixture, comprising 15 vol % CO2and 85 vol % N2to represent emissions from clinker production in cement plants or from limestone calcination in lime kilns, was injected into a mixed slurry reactor at a flow rate of 2 L·min−1. Temperature was selected as the controlled variable because it exerts a dual influence on the process by affecting both CO2 solubility and reaction kinetics. 31 Capturing sufficient temperature variation was necessary to enable the precise estimation of kinetic model parameters, particularly the activation energy. Pure water was used without additives that could influence ionic activity (e.g., salts 32 ) or mass transfer (e.g., nanoparticles 33 ), in order to avoid altering diffusion coefficients and to maintain the validity of the underlying modeling assumptions. The solid-to-liquid ratio was chosen based on values frequently reported as optimal in mineral carbonation literature. 27,28,31 The gas flow rate was adjusted to ensure sufficient CO2supply according to Henry’s law, while avoiding channelling, thereby preserving effective gas−liquid contact. 20 The apparatus contains CO2and N2gas capsules equipped with ball valves (V01 and V02 in Figure 1) to direct gas to the control valves (V03 and V04 in Figure 1). These valves, triggered by a flow controller (FIC 01), operate after measuring the gas stream flows using FIT 01 and FIT 02. The user-controlled gas streams are then directed to a mixer and furnace (M01 and E01), regulated by TIC 01, to adjust the temperature to the reaction set point before being injected into the reactor. These controllers are part of a larger Distributed Control System (DCS), which allows for adjusting process variables via a Human−Machine Interface (HMI). The reactor is equipped with a thermal water bath and thermometers for monitoring the bath and reactor temperatures during the reaction time to ensure the local uniformity and isothermal behavior through the batch process. This reactor is an agitated and bubbled system, fitted with a condenser (E02) to return the vapor stream, and a pH meter probe. It operates semicontinuously by holding the slurry while passing gas through and venting. In each experiment, the slurry was prepared by dispersing the RCFs in water, stabilizing the mixture at the target batch temperature, and injecting the gas mixture at matching temperature conditions. After each experiment, cRCFs were collected and oven-dried at 105 °C for subsequent analysis. 2.3. Characterization Tests and Metrics for CO2 Mineralization. The chemical composition (oxides in % weight and normalized to 100%) of the material was obtained by X-ray fluorescence, PANalytical AXIOS. The quantitative crystalline and amorphous contents of the RCFs were determined by X-ray diffraction using an Empyrean diffractometer (Malvern-PANalytical) equipped with a multistrip PIXcel3D detector, Cu Kαradiation, and Bragg−Brentano optics. Measurements were conducted with an internal standard (20 wt % α-Al2O3) in the range of 5°to 120 2θ and sample spinning. Analyses of the samples were performed with the Rietveld method following the fundamental parameters approach implemented in TOPAS V4.2 (BrukerAXS). Mass losses and CO2content were quantified by thermogravimetric analysis (TGA) using a Jupiter STA 449 F3 (NETZSCH) with a heating rate of 10 °C/min from 25 to 1075 °C and a nitrogen gas flow (70 mL/min). Density of RCFs and cRCFs were measured using a helium pycnometer AccuPyc II 1340. Particle morphology and geometry were investigated using a scanning electron microscope (SEM, Zeiss Supra VP55) with a field emission gun operating up to 25 kV accelerating voltage and 30 μm of aperture size. Elemental mapping of the indentation areas is performed using a Quantax Energy-Dispersive X-ray Spectroscopy (EDS) system (XFlash 5010 with 10 mm2detection area and 127 eV fwhm/MnKα) coupled to the SEM. Attenuated Total Reflection (ATR) spectroscopy measurements were obtained on a Tensor II spectrometer (Bruker Optics, Ettlingen, Germany) with a deuterated triglycine sulfate (DTGS) detector and a golden gate ATR cell with diamond crystal (Specac LTD, Orpington, UK). Each spectrum is the average of 64 scans acquired over 2 min in the range 400− 4000 cm−2with a spectral resolution of 2 cm−1. The evaluation of the spectra was perfomed by OPUS (Bruker Optics). Particle size distributions were obtained with a Malvern Mastersizer 2000 with hydro 2000S using ethanol as dispersant. Specific surface area (SSA) was measured using a Quantachrome Nova 4000e (Quantachrome Instruments, Boynton Beach, FL, USA) with Nitrogen as probe and liquid nitrogen as coolant. All samples were dried under vacuum at 105 °C prior to analysis. Blended cements were obtained by mixing neat CEM I 52.5N with RCFs and cRCFs at replacement ratios of 5%, 10%, and 15%. Compressive strengths tests were conducted using EN mini prism with dimensions of 19 ×10 ×44 mm (width ×height ×length). Each mortar was composed of blended cement, sand, and water at a constant mass ratio set at 1:3:0.5 (% in relation to the mass of binder). At the given curing age, each blend was tested in triplicate. The CO2uptake, denoted as α, is a metric that indicates the net CO2sequestration in the mass fraction of the RCFs during carbonation. It is evaluated at a set of discrete sampling times tsd p∈{tsd 0,tsd 1, ..., tsd N}, where Nis the number of sampling points after starting the reaction, and tsd 0= 0 denotes the precarbonation baseline. To isolate the CO2fixed by carbonation, the uptake at time tsd pis calculated by correcting for the CO2content already present in the material at tsd 0. It is calculated using eq 1, which is based on the gravimetric changes between 500 to 900 °C associated with CO2content and the total mass loss up to 975 °C serving as normalization basis�as described in eqs 2 and (3), respectively. To note, outside the parentheses, the Industrial & Engineering Chemistry Research pubs.acs.org/IECR Article https://doi.org/10.1021/acs.iecr.5c02835 Ind. Eng. Chem. Res. 2025, 64, 21412−21425 21414 difference in CO2is rescaled by the total mass loss of the RCFs ensuring that the results are meaningful and comparable to the starting feed of the process. = × Ä Ç Å Å Å Å Å Å Å Å Å Å Å Å Å É Ö Ñ Ñ Ñ Ñ Ñ Ñ Ñ Ñ Ñ Ñ Ñ Ñ Ñ t m m m m m ( ) 100 100 100 100 s t t t t t ,500 900 ,30 975 ,500 900 ,30 975 ,30 975 p sp sp s s s 0 0 0 (1) where: =° ° m m m , and t t t,500 900 ,500 C ,900 C spspsp (2) =° ° m m m t t t,30 975 ,30 C ,975 C spspsp (3) The carbonation efficiency, X(tsd p), see eq 4, is the metric that normalizes the CO2uptake by considering the material theoretical carbonation potential. The theoretical potential, ThCOd 2, is calculated using the modified Steinour equation, see eq 5, which uses the chemical composition from XRF as input. 34 =X t t Th ( ) ( ) s s CO p p 2 (4) = · · · + · i k j j jy { z z z Th CaO CaCO SO MgO 44 56 56 100 %56 80 1.091 CO 3 3 2 (5) 2.4. Candidate Kinetic Models and Key Modifications. To describe the mineral carbonation process under wet conditions, different product layer diffusion-limited Shrinking Core Models were selected as candidate models. These models are reported to be more predictive and relevant to mineral carbonation processes when the reaction extends toward equilibrium and becomes limited due to mass transport through the product layer. 27,28 The selected models are summarized in Table 1, including two models that are developed based on the approach proposed by Miao et al. 27 to complement the geometric dependency of product layer growth regime by adding 1-D and 3-D growth systems in plateshaped and spherical particles. Two diffusion-based Shrinking Core Models are formulated to address the oneand three-dimensional growth due to carbonation by leveraging the parabolic diffusion law, which is reported to have a better description of the diffusion through solid phases (product layer). 35 The models are formulated based on the following assumptions: 27 a. CO2acts as an excess reactant, and the slurry remains saturated with it at all times. b. The particle geometry and size stay constant throughout the reaction. c. CO2(aq) and mineral particles are evenly distributed within the slurry. d. Temperature is uniform and remains constant during the reaction process. e. CO2diffuses molecularly, and chemical reactions are restricted to carbonation of CaO on the solid interface. f. The effective diffusion coefficient changes over time, and the reaction is primarily diffusion-controlled. g. The reaction follows the quasi-steady-state approximation. The model derivation procedure is represented in Supporting Information, which results in a 1-D model, i.e., MSCMp,p and a 3-D model, i.e., MSCMp,s. The structure of the two models is included in Table 1 together with other candidate particle-fluid Shrinking Core Models used to predict carbonation kinetics. It includes model name as abbreviation, description based on the diffusion law, intrinsic diffusion Table 1. Structure of Candidate Product Layer Diffusion Limited Shrinking Core Models Model name Remarks Structure BSCMf,c D: constant = · · · =kC D d X t k X 16 ,d d ln(1 ) p a a 0 0 0 2 Diffusion: Fick Geometry: cylindrical BSCMf,s D: constant = · · · = + · kC D d X t k X 24 ,d d2 2 (1 ) p a 0 0 0 2 a 1/3 Diffusion: Fick Geometry: spherical MSCMf,p D: dynamic = · · · · · = · · · · kM C D f d X t k k t X 8,d d exp( ) 2 p a aCaO 0 0 CaO 02 a Diffusion: Fick Geometry: plate MSCMf,c D: dynamic = · · · · · = · · · · kM C D f d X t k k t X 8,d d 2 exp( ) ln(1 ) p aCaO 0 0 CaO 02 a a Diffusion: Fick Geometry: cylindrical MSCMf,s D: dynamic = · · · · · = · · · · kM C D f d X t k k t X 8,d d exp( ) ((1 ) 1) p a aCaO 0 0 CaO 02 3 21 1/3 Diffusion: Fick Geometry: spherical MSCMp,p D: dynamic = · · · = · · · kC D d X t k k t X 4,d d exp( ) p a0 0 CaCO 02 a a 3 Diffusion: Parabolic Geometry: plate MSCMp,c D: dynamic = · · · = · · · kC D d X t k k t8,d d exp( ) ((1 X) 1) p a0 0 CaCO 02 a a 0.5 3 Diffusion: Parabolic Geometry: cylindrical MSCMp,s D: dynamic = · · · = · · · · kC D d X t k k t X X 12 ,d d exp( ) (1 (1 ) ) (1 ) p a0 0 CaCO 0 2 a a 1/3 2/3 3 Diffusion: spherical Geometry: plate Industrial & Engineering Chemistry Research pubs.acs.org/IECR Article https://doi.org/10.1021/acs.iecr.5c02835 Ind. Eng. Chem. Res. 2025, 64, 21412−21425 21415 coefficient, particle geometry, and structure of the kinetic rate expression. The models are categorized as follows: •Basic Shrinking Core models (BSCMs): Assuming a constant diffusion coefficient and a cylindrical (2-D) or spherical (3-D) particle geometries. •Modified Shrinking Core models (MSCM): Accounting for the time-dependency of the diffusion coefficient and divided into: •Fickian diffusion-based models: Developed for plate, cylindrical, and spherical geometries. •Parabolic diffusion-based models: Developed for plate, cylindrical, and spherical geometries. The model nomenclature follows a structural convention, where the prefix BSCM or MSCM identifies the model type. This prefix is followed by two subscripts: the first subscript denotes the diffusion mode (ffor Fickian and pfor parabolic), while the second subscript indicates the particle geometry (p for the plate, cfor the cylinder, and sfor the sphere). These models consider dependency of apparent reaction coefficient (ka) on variables like: Dissolved CO2concentration in water (C0), intrinsic diffusion coefficient through the product layer (D0), density of RCFs (ρ), initial diameter of particles (dp0), molar mass of CaO (MCaO), mass fraction of CaO in raw material available for carbonation (fCaO), and molar density of the reaction product (ρCaCOd 3). This arrangement lets the carbonation efficiency (X) be dependent only on time (t), apparent reaction coefficient (ka), and decay ratio (θ). When applied to wet carbonation systems, these models require two key modifications to account for the concentration of diffusing CO2and the system temperature dependency: 1. Correlation of dissolved CO2concentration (C0) with partial pressure (PCOd 2): The concentration of the diffusive species (CO2) must be correlated with its partial pressure in the injected gas. This modification accounts for the effective concentration of dissolved CO2at the mineral particle surface by considering the liquid−gas equilibrium between the reactor liquid phase and the CO2-containing injected stream, which is treated as the excess reactant. Assuming the RCFs-containing liquid phase is saturated with CO2, Henry’s law can be applied to calculate the dissolved CO2concentration in the aqueous (water) phase, denoted as C0.eq 6 is a temperature-dependent correlation for the CO2-water equilibrium and reads as follows: = · · · Ä Ç Å Å Å Å Å Å Å Å Å Å i k j j j j j y { z z z z z É Ö Ñ Ñ Ñ Ñ Ñ Ñ Ñ Ñ Ñ Ñ C P T T 35 exp 2400 1 1 0 CO ref 2 (6) 2. Decomposition of the intrinsic diffusion coefficient (D0): The second modification introduces temperature dependency in the reaction kinetics by decomposing the intrinsic diffusion coefficient (D0) rather than the apparent reaction coefficient (ka). In this approach, D0 is expressed as the product of the pre-exponential factor (A) and an exponential term involving the activation energy (Ea) according to the Arrhenius eq (eq 7). This decomposition incorporates thermal dependency into the model structure, improving its predictive capability for systems where temperature varies. 36 = · · i k j j jy { z z z D A E R T exp 0 a (7) 2.5. Model Identification Approach. A systematic approach was employed for model identification, targeting two main objectives: discrimination between model candidates and the calibration of the selected model. Figure 2 illustrates this approach, which comprises four main steps: (1) Preliminary calibration, (2) Estimability analysis, (3) CalibraFigure 2. Schematic diagram of the model identification approach with the following steps: (1) Preliminary Calibration, (2) Estimability Analysis, (3) Calibration and Discrimination, and (4) Validation. Industrial & Engineering Chemistry Research pubs.acs.org/IECR Article https://doi.org/10.1021/acs.iecr.5c02835 Ind. Eng. Chem. Res. 2025, 64, 21412−21425 21416 tion and discrimination, and (4) Validation. Each of these steps is described in more detail as follows. 1. Preliminary calibration: This initialization step aims at estimating the parameters of all model candidates and at quantifying the associated modeling uncertainty. It is formulated as a nonlinear regression problem, employing the Least Squares (eq S1) as the loss function and R2(eq S2) as a metric for model predictability. 37 To evaluate the precision of the estimation, a ttest is applied by calculating the t-value for each parameter (eq S5) and comparing it against the reference t-values. Higher tvalues indicate greater precision in parameter estimation. 38 2. Estimability analysis: Following the preliminary calibration of model candidates, parameters are ranked based on their impact on model responses employing Estimability Analysis (EA) with Orthogonalisation. 39 EA-Orthogonalisation removes dependencies among parameters to ensure their independent contribution. It ranks parameters based on significance, calculates the corrected critical ratio, and selects the most important ones from the ranked list, prioritizing those with the lowest corrected critical ratio. 3. Calibration and discrimination: Once the models are analyzed and insignificant parameters fixed, the retained parameters are re-estimated, and their precision is reevaluated using the same criteria described in Step 1. Models with parameters exhibiting low t-values are excluded from further consideration, because they cannot be calibrated with available experimental data set. Model discrimination is then conducted by calculating P-values (eq S4), which serve as a metric for the likelihood that a given model is the best descriptor of the system. 40 In cases where P-values are similar�indicating comparable predictive behavior among models�the model with the fewest fixed parameters was selected. 4. Validation: Once a representative model with acceptable precision is identified, a validation test is conducted. The test follows the leave-one-out (LOO) method, 41 where one entire reaction isotherm is iteratively removed, followed model recalibration using the remaining data, and by validating it against the excluded data. This iterative approach assesses the model robustness and ensures its reliability when exposed to unseen data. All of the aforementioned numerical procedures were carried out using the Python library MIDDoE. 42 2.6. Model Sensitivity Analysis for Inputs. The contribution of process controls was evaluated using Global Sensitivity Analysis (GSA) with Sobol’s method. 43 GSA-Sobol captures nonlinear interactions among process controls by computing the total Sobol index, which serves as a metric for their influence on carbonation efficiency over reaction time for the identified model. This method is applied to rank the impact of process controls. 3. RESULTS AND DISCUSSION 3.1. Characterization of RCFs and cRCFs. Table 2 lists the chemical and mineral compositions of the RCFs employed in the study. SiO2and CaO constitute the highest proportion of the sample, corresponding with the presence of quartz, calcite, and C−S−H (CaxSiHyO(x+2+y/2)), the latter represented in the amorphous content. XRD results (Figure S3) reveal the presence of clinker minerals such as C3S (3CaO·SiO2), C2S (2CaO·SiO2), and C3A (3CaO·Al2O3), which is likely due to the coarser grinding practices typical of past decades, in contrast to the higher fineness targeted in more recent years. Among the hydrated cement phases, ettringite (3CaO·Al2O3·3CaSO4·32H2O), monocarboaluminate (3CaO·Al2O3·CaCO3·11H2O), portlandite (Ca(OH)2), and hydrogarnet (3CaO· (Al2O3)0.8(Fe2O3)1.2·0.84SiO2·4.32H2O) were identified. Additionally, numerous mineral phases, including alkali-feldspars, biotite, kaolinite, and cordierite, were also found. These latter minerals are associated with sand and aggregates, highlighting the challenge of addressing an industrial material. For reliable quantification using the Rietveld method, the optimal structure files of the phases were verified and crosschecked with the mass composition. Mass loss between 500 to 900 °C and between 30 to 975 °C is also included in Table 2, as these values serve to define inherited natural carbonation of RCFs and, as the basis of normalization calculations as carbonation progresses. The CaCO3content in the material prior to carbonation was estimated from TGA results as 10.78/0.44 = 24.5%, indicating that nearly 48% of the CaO is already carbonated, thereof the mass fraction of CaO available for carbonation (fCaO) is estimated around 14.6%. Mass balance calculations further suggested the presence of approximately 23% C−S−H and 8.6% hydrogarnet. After carbonation, combined XRD and mass balance analyses indicate a significant increase in the relative amount of CaCO3and the total consumption of portlandite; the content of clinker phases decreases while feldspars and other minerals remain roughly constant. Figure 3 illustrates the particle size distribution of the RCFs and two carbonated samples at 65 °C after 10 and 180 min (see Figure S4 for the full distribution set). The sieving (pretreatment) on the fines resulted in RCFs with a narrow size distribution, which progressively shifts toward finer particles as carbonation progresses. For instance, at start 10% Table 2. Chemical and Mineralogical Compositions of the Studied RCFs Chemical composition Mineralogical composition XRF wt % XRD wt % SiO250.96 Amorphous 42.4 Al2O35.61 Quartz 21.3 Fe2O32.41 Calcite (CaCO3) 19.1 CaO 36.06 Feldspars (Na, K, Ca) c 5.5 MgO 0.89 Clinker phases d 1.1 SO31.27 Portlandite (Ca(OH)2) 2.1 K2O 1.36 Other hydrated phases e 6.3 Na2O 0.80 Other minerals f 2.2 Others a 0.64 LOI b , 975 °C 21.67 TGA wt % Δmtd s0,500−900 10.78 Δmtd s0,30−975 19.76 a Mn2O3, TiO2, P2O5, SrO. b Loss on ignition. c Microcline, Labradorite, Lazurite. d C3S, C2S, C3A. e Ettringite, Monocarboaluminate, Hydrogarnet. f Biotite, Kaolinite, Cordierite. Industrial & Engineering Chemistry Research pubs.acs.org/IECR Article https://doi.org/10.1021/acs.iecr.5c02835 Ind. Eng. Chem. Res. 2025, 64, 21412−21425 21417 of RCF particles were smaller than 45 μm, while upon carbonation at 65 °C this fraction increased to 38.4% after 10 min and further rose to 64% after 180 min of reaction. Figure 4 displays SEM images of the industrial RCFs (a−c) and of the carbonated counterpart at 65 °C for 180 min (d−g). Results under this condition are presented as no significant variations in the shape or size of the particles were observed at any of the evaluated temperatures based on SEM images. The numbered circles mark the location where EDS point spectra were acquired (1−4). The RCFs consist of irregular, angular particles with rough surfaces (Figure 4a−c). Two representative surface types are apparent in (Figure 4b): a relatively smooth, blocky particle (point 1) and a particle coated by a microcrystalline layer (point 2). EDS of point 1 is consistent with a feldspar-rich particle, whereas the coating at point 2 likely contains C−S−H mixed with a minor content of carbonate-bearing mineral. A definitive phase assignment cannot be made from EDS alone, because carbonate may occur either as CaCO3or as carboaluminate phases, as previously presented by the XRD results above. After carbonation (Figure 4d−f), the coarse particles are covered by abundant fine polyhedral (rhombohedral-like) microcrystals. The fine particles form a discontinuous surface layer, at least partially composed of CaCO3as indicated by increased C in EDS results at points 3−4, consistent with XRD/TGA. Additionally, decreased Si and Al relative to Ca also suggest the presence of decalcified C−S−H and Si(Al)-gel among the carbonation products. Sieving the carbonated sample through a 22 μm mesh isolates a fine fraction composed of agglomerates of plate-like particles (Figure 4g).TGA measurements (Figure S5) confirmed that this finer fraction contains approximately 70% CaCO3. This observation is particularly interesting, as particles smaller than 45 μm had been mostly removed from the original RCFs prior to carbonation. This suggests that the conditions provided by wet carbonation�e.g., turbulences by stirring and CO2flow�favor the detachment and agglomeration of CaCO3 formed on the particle surface. 44 Regarding the formation of calcium carbonate, the higher proportion of these minerals observed on the surface indicates that the crystals do not grow in a confined space but rather result from a transport-limited process. This ultimately passivates the surface and slows down the reaction, as reported elsewhere. 16,30,45 Regarding the specific surface area (BET), cRCFs have a value of 18.0 m2/g, which is nearly four times higher than that of the RCFs at 4.7 m2/g, reflecting significant microstructural changes induced by carbonation. The increase in surface area Figure 3. Particle size distribution for the RCFs and cRCFs at 65 °C after 10 and 180 min of reaction. Figure 4. SEM images of (a−c) RCFs, (d−f) cRCFs, and (g) cRCFs < 22 μm at 65 °C, after 180 min. Red circles indicate location of EDS point analysis (1−4). Industrial & Engineering Chemistry Research pubs.acs.org/IECR Article https://doi.org/10.1021/acs.iecr.5c02835 Ind. Eng. Chem. Res. 2025, 64, 21412−21425 21418 may result from the disaggregation of inert particles caused by the carbonation of the binder. Note that the carbonate fines formed below 22 μm exhibit an even greater specific surface area 22.6 m2/g, yet this fraction only accounts for about 20 wt % of the carbonated product. This suggests that the remaining 80 wt % also experienced a considerable increase in surface area, likely attributed to the carbonation of C−S−H and formation of the Si(Al)-gel, which is known to have a specific surface area of approximately 100 m2/g and to contain more loosely bound water. 18,46 ATR spectra in Figure S6 indicate the formation of a Si(Al)- gel, as there is a shift of the asymmetric stretching vibration ν3Si−O of C−S−H from 967 to higher wavenumbers in the range of 1025−1032. 47,48 This shift is consistent with increased silicate polymerization and approaches the band position of amorphous SiO2(Aerosil). For comparison, a C−S−H with Ca/Si = 5/6 (Calcium to silica ratio) is included as a reference to infer the calcium content in the RCFs. The spectra also display bands attributable to carbonate stretching (ν4C−O∼ 713 cm−1,ν2C−O∼873 cm−1,ν3C−O∼1390−1405 cm−1). 49 Notably, the spectra are very similar for all four temperatures tested, consistent with the SEM description presented above and the carbonation efficiency, as discussed in section 3.2. Furthermore, the observed increase in density from 2.46 g/cm3to 2.57 g/cm3after carbonation is consistent with the formation of CaCO3. 50 Figure 5 presents the compressive strength of blended cement mortars prepared by using RCFs and cRCFs (carbonated at 25 °C) at replacement ratios of 5%, 10%, and 15%. The results suggest that up to 10% substitution with cRCFs is feasible without significantly compromising the mechanical performance of the blended cement. In terms of the influence of carbonation on strength development, both early and late-age responses are comparable between RCFs and cRCFs samples, although a slight improvement is observed for cRCFs at early ages. Previous studies have attributed such improvements to the presence of Si(Al)-gel and the increase in specific surface area, which enhance the reactivity and rapid synergy between cRCFs and cement. 18,51 However, it is important to note that most of the positive outcomes reported in the literature involve either the carbonation of pure cement paste or synthetic RCFs with a higher content of carbonatable phases, which provide a greater source of reactivity. In contrast, the RCFs used in this study exhibit a more limited availability of carbonatable materials, which constrains their overall performance enhancement. 3.2. Carbonation Efficiency. Figure 6a illustrates the carbonation efficiency used as a metric to evaluate the reaction progress of wet carbonated samples over time. Carbonation efficiency after 180 min of reaction ranges from 0.74 to 0.81, corresponding to a CO2uptake between 87 and 95 kg per tonne of RCFs. As a side note, the final carbonation might also include natural carbonation occurring during service life. Therefore, considering only the CaO content in the RCFs, the total CO2uptake is estimated at 199 kg per tonne of RCFs, with an overall efficiency of 90%. Limited to the tested material and carbonation conditions, no significant changes were observed throughout the analyzed period, with carbonation efficiency variations remaining below 0.029 across all tested conditions. To ensure confidence in the reproducibility of the data, the experiment at 25 °C was repeated. The replication yielded a negligible average standard deviation of 0.01 for pH and carbonation efficiency measurements, reinforcing the reliability of the observed trends. Figure 6b presents the pH values recorded during the experiments. Higher pH values were observed at the onset of the reaction, likely due to the dissolution of CaO from Ca(OH)2during the initial minutes. Since the RCFs contained only a limited amount of Ca(OH)2, this phase was fully consumed within the first 10 min of carbonation (Figure S5). Consequently, the subsequent carbonation of the RCFs was primarily driven by the decalcification of C−S−H. Interestingly, despite temperature variations, the overall extent of carbonation remained similar across all of the tested conditions. This outcome suggests a counterbalancing effect, wherein the influence of temperature on the diffusion coefficient (which according to model formulations defines the rate of the reaction) and CO2dissolution offset each other. This aspect is further discussed in section 3.3.2. 3.3. RCFs Carbonation Kinetic Modeling. In this section, two kinetic models are formulated based on Shrinking Core Models and the parabolic diffusion law in product layer − for cases where diffusivity coefficient changes occur. Using the set of rival models from Table 1, the proposed workflow for model screening, discrimination, calibration, and validation (Figure 2) is applied, alongside the experimental data from RCFs wet carbonation. 3.3.1. Step 1: Preliminary Calibration of Models. Table 3 presents the preliminary estimates for parameters (Figure 2, Step 1) of different models before any screening is applied to models. The 95% t-values of these estimations are also calculated and compared with the reference t-value of this system, which is 2.04. It is observed that for all models estimations of Aare not precise enough, as indicated by lower t-values. Additionally, the decay ratio θwas not estimated with an acceptable precision for model MSCMp,c. While all models demonstrate acceptable predictive performance, MSCMf,p and MSCMp,p are the most predictive, exhibiting the same R2value. 3.3.2. Step 2: Estimability Analysis of Models. To select the parameters that can be estimated based on the available experimental data set, estimability analysis was carried out. Figure 5. Compressive strength of blended cement mortars prepared with 5, 10, and 15% RCFs and cRCFs at 25 °C as partial replacements of CEM I. Industrial & Engineering Chemistry Research pubs.acs.org/IECR Article https://doi.org/10.1021/acs.iecr.5c02835 Ind. Eng. Chem. Res. 2025, 64, 21412−21425 21419 Table 4 (derived from the corrected critical ratios in Figure S7) presents the results of the estimability analysis for parameter ranking and subset selection. Across all models, Ea is identified as the most appropriate parameter to retain and pre-exponential factor as the most insignificant. Interestingly, although the experimental data indicated a limited direct influence of temperature on observable outcomes, still Ea�the parameter most strongly tied to temperature dependence�is identified as the dominant contributor. This implies that, while temperature may not appear to affect the process significantly, the underlying kinetics must compensate for temperature-related effects such as those arising from Henry’s law. This is a key insight that illustrates how the model can validate mechanistic assumptions even when direct trends are not evident in the measurements. The persistent insignificance of the pre-exponential factor can be attributed to the well-known high correlation between Eaand Ain Arrhenius-type formulations, eq 7, particularly within the narrow temperature range considered in this study. 52 In the case of MSCMf,p and MSCMp,p�identified as the most predictive models�this method suggests retaining Eafor both models, along with θonly for the formulated model, MSCMp,p, while fixing the remaining parameters. For the remaining models, only Eais retained as a parameter to be estimated. 3.3.3. Step 3: Model Calibration and Discrimination. Another round of parameter estimation and uncertainty analysis (Figure 2, Step 3) was conducted to reestimate only the selected parameters. Table 5 presents the variations in the t-values and calculated P-values for all identified models. These Figure 6. Measurements of (a) carbonation efficiency of the RCFs at temperatures of 25, 45, 65, and 85 °C and (b) pH at 25 °C during reaction time. Table 3. Parameter Estimates and t-values for the RCFs Carbonation Candidate Models A Eaθ Models Estimation [m2·s−1]t-Value Estimation [J mol−1]t-Value Estimation t-Value R2 BSCMf, c 6.0938 ×10−07 1.852 20211 14.075 - - 0.9722 BSCMf, s 3.0709 ×10−07 1.905 20209 14.459 - - 0.9774 MSCMf, p 3.1837 ×10−07 1.904 20253 14.468 1.2140 12.086 0.9841 MSCMf, c 8.7542 ×10−08 1.808 20216 13.734 2.1995 4.406 0.9814 MSCMf, s 3.7633 ×10−08 1.837 20222 13.936 6.9512 6.716 0.9823 MSCMp, p 1.3414 ×10−06 1.904 20252 14.467 2.4279 12.084 0.9841 MSCMp, c 3.8531 ×10−07 1.780 20197 13.515 0.8469 1.475 0.9802 MSCMp, s 1.8455 ×10−07 1.707 20187 12.941 1.8627 ×10−07 13.936 0.9787 Table 4. Estimability Analysis Results for Different Candidate Models A Eaθ Models Rank Selected Rank Selected Rank Selected BSCMf,c 2 no 1 yes � � BSCMf,s 2 no 1 yes � � MSCMf,p 3 no 1 yes 2 no MSCMf,c 3 no 1 yes 2 no MSCMf,s 3 no 1 yes 2 no MSCMp,p 3 no 1 yes 2 yes MSCMp,c 2 no 1 yes 3 no MSCMp,s 3 no 1 yes 2 no Table 5. t-Values of the Parameter Estimates and P-Values for RCFs Carbonation Candidate Models (after Fixing Parameters) Models Ea(t-value) θ(t-value) P-value BSCMf,c 202.598 - 9.8083 BSCMf,s 208.033 - 12.2518 MSCMf,p 212.44 - 16.2518 MSCMf,c 201.537 - 12.1775 MSCMf,s 204.465 - 13.2978 MSCMp,p 133.650 12.3087 16.2525 MSCMp,c 197.996 - 10.8931 MSCMp,s 189.606 - 9.0672 Industrial & Engineering Chemistry Research pubs.acs.org/IECR Article https://doi.org/10.1021/acs.iecr.5c02835 Ind. Eng. Chem. Res. 2025, 64, 21412−21425 21420