Recycled Concrete Aggregate: Insights

• RCA is crushed concrete from demolition waste that serves as a sustainable substitute for natural aggregates in pavement and concrete, though its grading and composition can vary significantly.
• Studies employing techniques like nonlinear regression for slump prediction and mechanical testing show that RCA integrated in mixtures influences workability and strength through complex interactions with other materials.
• Environmental and digital monitoring research highlights RCA’s evolving leachate chemistry and grading variability, emphasizing the need for integrated material characterization and adaptive mix design.

Recycled concrete aggregate (RCA) is crushed concrete derived from construction and demolition waste and reused as aggregate. In the literature summarized here, RCA appears both as a constituent of recycled concrete mixtures and as an aged pavement base-course material. Its engineering significance follows from two linked facts: it is recognized as a readily available, mechanically sufficient substitute for natural aggregate in pavement construction, and its variability in composition, grading, and interfacial behavior complicates prediction of workability, mechanical response, and leachate chemistry (Sanger et al., 2 Oct 2025, Xu et al., 2017, Coenen et al., 2022).

1. Scope, terminology, and material identity

Domain	RCA role	Representative evidence
Fresh concrete	Explicit input variable “recycled aggregate” in slump prediction	34 samples; 8 inputs; $R = 0.9765$ (Xu et al., 2017)
Structural concrete	Coarse recycled demolition-concrete aggregate replacing granite	Specific gravity 2.70; water absorption 0.45%; 40% and 60% replacement (M. et al., 2021)
Pavement base course	Aged recovered RCA from MnROAD	pH 10.2–10.6 declining to 9.4–9.9 over 24 h (Sanger et al., 2 Oct 2025)
Digital quality control	Recycled aggregate variability motivates image-based grading prediction	95.5% ± 0.7 overall accuracy at 2 px/mm (Coenen et al., 2022)
Mesoscale mechanics	Indirect recycled-aggregate analogue via crushed brick	$R_{M/A} \in [0.1,10]$; most favorable behavior near $R_{M/A}=1$ (Mašek et al., 19 Jun 2025)

RCA is not represented uniformly across the cited work. In the slump-prediction study, the mixture includes an explicit input variable called “recycled aggregate,” alongside cement, fly ash, water, sand, stone, water reducer, and total mass. That formulation establishes RCA dosage as a distinct explanatory variable, but the material is not physically characterized in detail, and the paper does not state whether it is coarse RCA, fine recycled concrete aggregate, or a blend (Xu et al., 2017).

In the eco-concrete study, the coarse recycled material is described as “recycled demolition waste aggregate,” but the materials section narrows the actual test material to mechanically crushed demolished concrete for which “mortar waste was discarded/eliminated and only coarse aggregates of demolished concrete were considered for recycling.” On that basis, the aggregate is best interpreted as RCA-like material rather than mixed recycled aggregate. The reported properties are particle size 20 mm and down, fineness modulus 6.19, specific gravity 2.70, water absorption 0.45%, and bulk density 1504 kg/m$^3$, compared with 2.74, 0.33%, and 1545 kg/m$^3$ for natural crushed granite gravel (M. et al., 2021).

The MnROAD leachate study treats RCA as an aged, field-deployed base-course material recovered after eight years of service. The recovered materials had no significant differences in physical and hydraulic properties relative to the original MnROAD RCA, but their carbonate content was higher, indicating carbonation during the eight-year field deployment. This distinction matters because the paper directly links carbonation to later leachate behavior and to the preservation of a calcium carbonate surface coating (Sanger et al., 2 Oct 2025).

A persistent terminological issue across the literature is that RCA is often central to the application while remaining only partially characterized. The slump study does not report water absorption, density, particle size distribution, crushing index, adhered mortar content, source concrete strength, impurity content, old paste content, or shape/angularity. The eco-concrete study likewise does not report adhered mortar content, abrasion indices, contamination fraction, or source concrete quality. This suggests that many data-driven or performance-oriented RCA studies remain stronger on system-level response than on aggregate-level materials science (Xu et al., 2017, M. et al., 2021).

2. Fresh-state behavior and slump prediction

The cited slump study formulates recycled-concrete workability as a nonlinear regression problem. The dataset contains 34 samples, of which the first 28 samples are used as the training set and the remaining 6 samples as the test set. The implemented model uses eight inputs, $\{x_1,x_2,\ldots,x_8\}$, corresponding to cement, fly ash, water, sand, stone, recycled aggregate, water reducer, and total mass, with slump as the output variable. The paper notes that recycled concrete has a more complex composition than conventional concrete, and argues that conventional linear methods scarcely obtain satisfactory results (Xu et al., 2017).

The adopted method is Geometric Semantic Genetic Programming (GSGP). Its function set is $\{+,-,\times,\div\}$, the initial population size is 500, the maximum iteration epoch is 50, and the genetic mutation step is 0.1. The paper defines the original fitness by

$e(i,t)=\sum_{j=1}^{n}\left|s(i,j)-c(j)\right|,$

and uses geometric semantic crossover and mutation operators,

$T_c=(T_1 \cdot T_R)+(1-T_R)\cdot T_2,$

$T_M=T + ms\cdot(T_{R1}-T_{R2}).$

In this formulation, RCA is not treated as a fixed replacement ratio class but as one variable in an eight-dimensional compositional space (Xu et al., 2017).

The reported predictive performance is strong within the reported dataset. The correlation coefficient between experimental and computational slump values is

$R_{M/A} \in [0.1,10]$0

and the relative prediction error is reported as less than 5%. In comparison with repeated runs of LSSVM and STGP, the interquartile ranges of RMSE are 0.0892 for GSGP, 1.9933 for LSSVM, and 3.8651 for STGP, with Wilcoxon rank-sum test $R_{M/A} \in [0.1,10]$1-values $R_{M/A} \in [0.1,10]$2, $R_{M/A} \in [0.1,10]$3, and $R_{M/A} \in [0.1,10]$4. The paper therefore concludes that GSGP has higher accuracy and reliability than conventional methods on this dataset (Xu et al., 2017).

The contribution is predictive rather than mechanistic. No final explicit symbolic slump equation is reproduced, and the paper does not provide sensitivity analysis, feature-importance ranking, partial dependence analysis, or controlled trend analysis isolating the effect of RCA content. It therefore does not establish whether increasing RCA increases or decreases slump under the studied conditions. What is supported is narrower but still important: recycled aggregate is one of the core predictor variables, and slump is modeled as a joint nonlinear function of RCA and seven other mixture descriptors (Xu et al., 2017).

This has a clear practical interpretation. Since slump testing requires batching and testing actual mixtures, and therefore consumes raw materials, labor, and time, a nonlinear predictor can serve as a mix-design aid. A plausible implication is that RCA-containing mixtures can be screened computationally before laboratory validation, provided that the limitations of the small 34-sample dataset and simple hold-out validation are kept in view (Xu et al., 2017).

3. Mechanical performance in concrete and mesoscale interpretation

The eco-concrete study provides the most direct structural-concrete evidence in the set. It produced five M20 mixes with OPC 53 grade cement, $R_{M/A} \in [0.1,10]$5, designed slump 75 mm, no reported admixtures, and coarse recycled demolition-concrete aggregate replacing natural coarse granite at 40% and 60% by weight of the coarse aggregate. The recycled aggregate was always combined with treated incinerator bottom ash replacing fine aggregate at 5% or 10%, so the RCA effect is not isolated. The best-performing mixture was M2, containing 5% incinerator bottom ash and 40% recycled demolition waste aggregate (M. et al., 2021).

The reported strength values show a consistent pattern. At 28 days, the control mix M1 had compressive strength 25.8 MPa and M2 had 27.5 MPa. At 90 days, M1 had 26.2 MPa and M2 had 29.0 MPa, corresponding to +10.69% relative to the control. The 90-day increases for M3, M4, and M5 were +8.02%, +9.16%, and +6.68%, respectively. At both ash levels, 40% recycled coarse aggregate outperformed 60% recycled coarse aggregate: M2 exceeded M3, and M4 exceeded M5. The paper attributes the reduction at higher recycled aggregate levels to greater matrix heterogeneity and weaker interfacial bonding between cement paste and RDW aggregate (M. et al., 2021).

The same trend appears in tensile, flexural, and nondestructive indicators. At 28 days, split tensile strength was 2.40 MPa for M1 and 2.62 MPa for M2, while flexural strength was 3.63 MPa for M1 and 4.19 MPa for M2. M5, with 10% ash and 60% recycled coarse aggregate, dropped below the control in both split tensile strength and flexural strength. Ultrasonic pulse velocity placed all mixes in the “Good” quality category, and the paper reports a strong correlation between compressive strength and pulse velocity, $R_{M/A} \in [0.1,10]$6. Volume of permeable voids was lower for M2 than for the control by 2.8% at 28 days and 4.2% at 90 days, whereas higher recycled aggregate contents tended to increase voids (M. et al., 2021).

The microstructural interpretation remains cautious because the SEM discussion is more centered on bottom ash than on RCA-specific interfaces. The paper reports dense aggregate paste matrix interfaces with less micro-pores and insignificant micro-cracks due to the incorporation of incinerator bottom ash as a partial replacement to the fine aggregate, but it also describes a non-uniform compact pore structure system with C-S-H, calcium alumino-sulfate hydrates, and microcracks typically less than 10 $R_{M/A} \in [0.1,10]$7m at the interfacial transition zone. The absence of dedicated RCA-versus-natural-aggregate ITZ imaging limits the conclusiveness of the microstructural evidence for RCA itself (M. et al., 2021).

A more general mechanics rationale is provided indirectly by the mesoscale finite element study on recycled crushed brick aggregate. That paper is not a direct RCA study, but it isolates a mechanism relevant to RCA whenever the recycled aggregate is more compliant than the surrounding matrix. The governing heterogeneity parameter is

$R_{M/A} \in [0.1,10]$8

with investigated values 0.1, 0.2, 0.25, 0.6, 0.8, 1, 2, 5, and 10. The authors find minimum amplification of normalized maximum stress near $R_{M/A} \in [0.1,10]$9, and increasing stress concentration as the stiffness ratio moves away from 1 in either direction. The most consequential regime for recycled-aggregate concrete is $R_{M/A}=1$0, where softer aggregates in a stiffer matrix shift load to the matrix, intensify local stress peaks at sharp polyhedral corners and interfaces, and can make the composite approach the behavior of a cement matrix containing voids (Mašek et al., 19 Jun 2025).

This does not quantify RCA strength loss directly, because the model is purely elastic, does not simulate cracking or damage evolution, and does not include an explicit ITZ phase. Still, it provides a plausible mechanics explanation for the empirical observation that higher recycled aggregate contents in the eco-concrete mixes were associated with lower strength, lower UPV, and higher void tendency. The shared principle is stiffness and interface compatibility rather than a specific recycled-aggregate chemistry (Mašek et al., 19 Jun 2025, M. et al., 2021).

4. Environmental geochemistry, carbonation, and leachate evolution

The environmental study focuses on aged RCA used as pavement base course at the Minnesota Road Research (MnROAD) test facility. The material came from Experimental Cell 16, constructed in September 2008, and the recovered RCA samples were taken in July 2016 from beneath the passing lane (16P-1), driving lane (16D-1), and centerline (16C-1). Field observations from the same site had shown lysimeter leachate pH of 7.5 to 8.5 in the initial time period after construction and 7.2 to 7.4 about eight years later. These field values contrast strongly with conventional laboratory batch-test pH values around 11.1 to 11.4 for the recovered RCA (Sanger et al., 2 Oct 2025).

To investigate that discrepancy, the study used a modified open-system batch methodology with continuous gentle shaking rather than end-over-end tumbling. Each test used 50 g RCA with 500 mL Milli-Q ultrapure water, corresponding to a liquid-to-solid ratio of 10, and monitored pH, alkalinity, and Ca$R_{M/A}=1$1 for 24 hours. The reactors were open to atmospheric CO$R_{M/A}=1$2, and the sampling schedule had especially high temporal resolution in the first hour, with measurements at 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 15, 20, 25, 30, 40, 50, 60, 90, 120, 180, 240, 300, 360, and 1440 minutes (Sanger et al., 2 Oct 2025).

The core result is that RCA leachate chemistry is strongly time-dependent. Leachate pH was high upon initial contact with water and then declined steadily over 24 hours. Reported values were pH 10.6 declining to 9.7 for 16P-1, 10.5 declining to 9.9 for 16D-1, and 10.2 declining to 9.4 for 16C-1, with the maximum pH value measured within the first hour for all three samples. Calcium ion concentration increased rapidly in the initial contact period and then more gradually: 16P-1 rose from 0.0002 M to 0.0017 M, 16D-1 from 0.0003 M to 0.0017 M, and 16C-1 from 0.0001 M to 0.0018 M. Most of the Ca$R_{M/A}=1$3 ions were released within the first 4 to 6 hours. Alkalinity rose rapidly and then stabilized, with the abstract and conclusions summarizing the stabilization range as 50–65 mg CaCO$R_{M/A}=1$4/L (Sanger et al., 2 Oct 2025).

The paper interprets these trends through dissolution and carbonation reactions. Portlandite dissolution is given by

$R_{M/A}=1$5

which explains the initial high pH. Atmospheric CO$R_{M/A}=1$6 then dissolves and reacts through the carbonate system,

$R_{M/A}=1$7

$R_{M/A}=1$8

$R_{M/A}=1$9

$^3$0

The paper defines alkalinity as

$^3$1

and uses this to explain why alkalinity can remain elevated even as pH declines: hydroxide is converted into bicarbonate and carbonate rather than simply disappearing (Sanger et al., 2 Oct 2025).

The study also emphasizes carbonation as a protective process. Calcium carbonate precipitation can form a surface coating that inhibits further dissolution of alkaline cement phases. This is why the choice of test method matters. Conventional end-over-end tumbling abrades particles, strips or damages carbonated surface coatings, and yields higher pH values than the modified low-abrasion shaker-plate method. The reported 24-hour values show this clearly: for 16P, conventional 11.3 versus modified 9.9; for 16D, 11.4 versus 10.1; for 16C, 11.1 versus 9.5 (Sanger et al., 2 Oct 2025).

This section of the literature directly addresses a common misconception. Field and laboratory measurements are not necessarily contradictory; rather, contact time, atmospheric exposure, and preservation of carbonated surface coatings strongly influence the observed chemistry. The environmental concern remains real—high alkalinity, high pH leachate, heavy metal leaching risks, and drain clogging or permeability loss due to calcium carbonate precipitation are all identified in the literature—but the paper argues that environmentally relevant assessment of RCA base-course use requires open-system, low-abrasion, time-resolved testing rather than a single aggressive batch endpoint (Sanger et al., 2 Oct 2025).

5. Grading variability, digital monitoring, and process control

Aggregate grading substantially affects the properties and quality characteristics of concrete in both the fresh and hardened states, including workability, consistency, mix stability, strength, and durability. The image-based grading paper stresses that unknown variations in size distribution can be large especially when using recycled aggregate materials, and that these variations are usually compensated by a distinct increase of the cement proportion, which is neither economically nor ecologically justifiable. This motivation is highly relevant to RCA because it treats variability itself as the principal production problem (Coenen et al., 2022).

The proposed solution is a deep-learning method that predicts aggregate grading curves from images. Formally, the mapping is written as

$^3$2

or, for a CNN parameterized by $^3$3,

$^3$4

In the reported implementation, the task is discretized into nine predefined grading-curve classes,

$^3$5

with

$^3$6

The samples are built from four particle-size fractions: 0–2 mm, 2–8 mm, 8–16 mm, and 16–32 mm (Coenen et al., 2022).

The network, AggNet, uses an initial $^3$7 convolution with $^3$8 filters, followed by four multi-scale residual feature encoder modules, a $^3$9 convolution, global average pooling, and softmax. The multi-scale branch uses dilated convolutions with dilation rates $^3$0, 2, and 4. The dataset contains 900 images in total: two physical samples per class, 50 images per sample, over nine classes. Ground truth was obtained by mechanical sieving. Training used 396 images and validation 54 images from S1 samples, while all S2 images were used only for testing (Coenen et al., 2022).

On this controlled dataset, the best reported model is AggNet:MS with augmentation at a ground sampling distance of 2 px/mm, achieving $^3$1 overall accuracy. The corresponding values for AggNet:Base with augmentation, AggNet:MS without augmentation, and AggNet:Base without augmentation are $^3$2, $^3$3, and $^3$4, respectively, while human experts achieve $^3$5. Increasing GSD from 0.4 to 2.0 px/mm improved overall accuracy by more than 20%, whereas increasing further from 2.0 to 2.8 px/mm reduced mean overall accuracy by 5.4% (Coenen et al., 2022).

The paper does not explicitly test RCA as a separate dataset, and its output is limited to nine DIN-based classes rather than a continuous grading curve. Even so, its relevance to RCA is direct at the level of quality control. A plausible implication is that real-time image-based grading on conveyor belts could support adaptive mix design when RCA feed variability is high, potentially reducing uncertainty-driven cement overcompensation. The paper itself points toward this direction by proposing future development of an online concrete control scheme that adapts the mix composition in real time to fluctuations in raw materials (Coenen et al., 2022).

6. Limitations, controversies, and research directions

Across the cited work, the main limitation is uneven material characterization. The slump study uses recycled aggregate as an explicit input variable but reports no standard RCA descriptors such as water absorption, density, particle size distribution, adhered mortar content, or source concrete quality. The eco-concrete study reports basic physical properties but omits adhered mortar percentage, abrasion indices, contamination, and source concrete strength. The image-based grading study is motivated by recycled aggregates but does not explicitly include RCA as a separate experimental domain. This means that RCA often enters the analysis as a variable or application context rather than as a fully resolved material system (Xu et al., 2017, M. et al., 2021, Coenen et al., 2022).

A second limitation is model scope. The slump study uses only 34 samples with a simple hold-out split of 28 training and 6 testing samples, and provides no cross-validation, no explicit final predictive equation, and no feature-importance analysis. The mesoscale FEM study is elastic only, does not include cracking, plasticity, or nonlinear damage, does not explicitly model the ITZ, and treats each aggregate particle as a homogeneous inclusion rather than a multilayer RCA particle with original aggregate core, adhered old mortar, and new mortar interaction. The leachate study examines recovered, aged RCA for 24 hours but does not provide a complete long-term transport simulation or new time-resolved trace-metal data (Xu et al., 2017, Mašek et al., 19 Jun 2025, Sanger et al., 2 Oct 2025).

The most explicit controversy concerns environmental assessment. Conventional 24-hour batch “material pH” values can suggest highly alkaline RCA, while field leachate at MnROAD was near neutral to mildly alkaline. The modified open-system batch results show that this discrepancy is at least partly methodological, because abrasion, atmospheric exposure, and contact time strongly affect the measured chemistry. The issue is therefore not whether RCA can generate alkaline leachate, but how laboratory tests should represent field conditions (Sanger et al., 2 Oct 2025).

The literature also identifies several concrete research directions. For environmental performance, recommended next steps include characterizing freshly crushed RCA leachate chemistry, using geochemical modeling to connect grain size distribution, solid-phase composition, and degree of carbonation to leachate behavior, monitoring chemistry after phase separation to simulate drainage into subgrade soils and aquifers, and developing industry guidelines for prescribed aging and stockpiling criteria (Sanger et al., 2 Oct 2025). For digital control, the image-based grading study proposes predicting percentiles of the size distribution directly rather than classifying into nine predefined grading curves, and using the method as a basis for online mix adaptation (Coenen et al., 2022). For fresh-state prediction, a reasonable extension is richer RCA descriptor sets, larger datasets, and stronger validation, because the present GSGP model is accurate on its reported dataset but materially underdetermined from an RCA-characterization standpoint (Xu et al., 2017).

Taken together, these studies define RCA less as a single standardized material than as a variable recycled aggregate class whose performance is governed by dosage, grading, aging and carbonation state, stiffness compatibility with the matrix, and the fidelity of the test or model used to interrogate it. This suggests that progress in RCA research will depend on integrating three strands that are often separated: detailed aggregate characterization, field-representative environmental and mechanical testing, and data-driven process control that can respond to source-to-source variability.