Steel mill scale in concrete and mortar: a critical review of the evidence base and a proposed framework for future quantitative synthesis

1 Introduction

Iron and steel manufacture is among the most material- and carbon-intensive industrial activities, and the sector’s solid by-product streams are correspondingly large (Habert et al., 2020). Mill scale—the layered iron oxide film that forms on hot steel surfaces and is subsequently spalled or descaled during rolling—is one of the largest of these streams by mass. Independent sources converge on a generation rate of approximately 1%–3% of the mass of steel rolled, most commonly cited as about 2%, or equivalently 22–40 kg per tonne (Chokshi et al., 2018; Murthy et al., 2017; Nowacki et al., 2021; Vilarinho et al., 2021; Solmaz et al., 2024).

Applied to world crude steel production of 1,882.6 Mt in 2024 (World Steel Association, 2025), a 1%–3% generation rate implies of the order of 19–56 Mt of mill scale per year globally, with the commonly cited ∼2% rate implying approximately 38 Mt/yr. This is consistent with the two independent published global estimates we were able to locate: approximately 37 Mt for 2019 (Vilarinho et al., 2021) and approximately 39 Mt (Solmaz et al., 2024). Regional figures are much smaller: the European steel industry has been estimated to generate on the order of 500,000 t/yr of oily sludge and mill scales combined, of which more than 30% is reported as not valorised (Iluțiu-Varvara et al., 2020).

Concrete, the counterpart material, is produced at a scale that makes even marginal substitutions consequential, and Portland cement clinker manufacture remains a dominant source of process CO₂ in the built environment (Habert et al., 2020; Marsh et al., 2023). Against this background, mill scale has attracted interest on three distinct grounds. First, as a dense, hard, angular particulate it is a candidate partial replacement for natural fine aggregate, whose extraction carries its own environmental and supply-chain costs. Second, its high specific gravity and high atomic number make it attractive for heavyweight and radiation-shielding concretes (Sharma et al., 2024; Sylisomchanh et al., 2025) and, more recently, for electromagnetic-interference-shielding cementitious composites (Loganathan and Senthil Pandian, 2025). Third, and much more speculatively, its amorphous surface oxides have been proposed as a source of limited chemical reactivity in an alkaline pore solution.

Despite this interest, the material has not entered mainstream concrete practice. The barriers are usually attributed to compositional variability, the absence of any product standard, and thin long-term performance data. To these we add a fourth, which is the specific concern of this review: the primary literature, as currently reported, does not permit quantitative synthesis. This is a reporting problem as much as a scientific one, and it is remediable.

1.1 Aim and scope

This article is a critical review. It does not report a pooled quantitative meta-analysis, and no effect size, confidence interval, heterogeneity statistic or significance test presented anywhere in this manuscript has been calculated from extracted study-level data because no such extraction was performed. Readers should not interpret any statement in this article as a pooled estimate.

The article has four aims:

• To characterise mill scale as an engineering material with explicit attention to the sources of its variability, and to resolve the compositional inconsistencies present in the earlier literature.
• To describe a transparent, reproducible review methodology, and to state candidly which elements of it require re-execution and documentation by the authors before this work can be presented as a systematic review.
• To map what the primary literature reports about mechanical, durability and environmental performance, using evidence-graded language that distinguishes single-study findings from patterns replicated across independent groups.
• To specify a pre-registered meta-analytic protocol and a minimum reporting standard that would make future quantitative synthesis possible.

1.2 Evidence-grading convention used throughout

To keep the strength of each statement explicit, this review applies the following convention. It is used consistently in Sections 4 through 9.

Terms such as proves, demonstrates conclusively, optimal, guarantees and will achieve are avoided except where a cited study’s own explicit claim is being reported as such.

2 Mill scale as an engineering material

2.1 Formation and layer structure

Mill scale forms by high-temperature oxidation of the steel surface and is stratified. There is qualitative consensus across the metallurgical literature on the ordering of the layers: wüstite (FeO) innermost, adjacent to the base metal; magnetite (Fe₃O₄) intermediate; hematite (Fe₂O₃) outermost (Tanei and Kondo, 2016; Carvalho et al., 2024); the layer sequence and the process variables that govern the delivered material are summarised in Figure 1. (E2)

The proportions, however, are neither fixed nor well constrained, and this is the origin of much of the confusion in the construction-materials literature. Two distinct points must be separated. (M)

• At rolling temperature, above approximately 700°C, wüstite dominates the growing scale by thickness; trade-technical sources give figures of the order of 95% wüstite, 4% magnetite and 1% hematite (Sarna, 2017). These are in-furnace, high-temperature proportions.
• Below approximately 570°C, wüstite is thermodynamically unstable and decomposes eutectoidally to magnetite and iron. The scale that is actually collected as waste has therefore passed through this transformation, and its phase assemblage depends on coiling temperature and cooling rate. Chen and Yuen (2001) show directly that the oxide structure of commercial hot-rolled strip varies with coiling temperature and position across the strip, ranging from inner magnetite/iron mixtures at low coiling temperature to thick hematite layers near strip edges at high coiling temperature.

The single quantitative phase analysis of as-received mill scale we were able to source reports FeO 47.81%, Fe₃O₄ 51.34% and metallic Fe 0.85% — that is, magnetite marginally exceeding wüstite, with hematite not present as a major phase (Nowacki et al., 2021). (E1)

2.2 Reconciling reported compositions: XRF versus XRD

The apparent contradiction between studies reporting mill scale as “93%–99% Fe₂O₃” and studies reporting it as roughly half wüstite and half magnetite is not a contradiction at all. It is a difference in measurement convention that is rarely stated explicitly. (M)

• X-ray fluorescence (XRF) returns elemental concentrations, which are conventionally normalised and reported as oxides. In the standard convention, all iron is reported as Fe₂O₃ irrespective of its actual oxidation state. An XRF table reading “Fe₂O₃ = 99.2%” therefore means “the sample is essentially all iron oxide,” not “the sample is essentially all hematite.”
• X-ray diffraction (XRD) with Rietveld or reference-intensity-ratio quantification returns the actual crystalline phase fractions and can distinguish FeO, Fe₃O₄ and Fe₂O₃.

Studies reporting mill scale at 93.66% Fe₂O₃ (Wongsawan et al., 2022), 93.7% Fe₂O₃ (Sunankingphet et al., 2025), 98.74% Fe₂O₃ (Sylisomchanh et al., 2025) and 99.2% Fe₂O₃ (Siriwattanakarn et al., 2025) are all reporting XRF normalisations. They are not in conflict with the XRD phase quantification of Nowacki et al. (2021).

Defensible summary statements. On the evidence in Table 1:

• Total iron in mill scale is typically in the region of 65–75 wt%, with individual reports as high as approximately 85 wt% and one outlier at approximately 97.5 wt% for wire-drawing scale, which is a different process stream (Vilarinho et al., 2021). (E2)
• SiO₂ reports differ by an order of magnitude, from ≤1.5% (combined with CaO, per trade sources) to 8.35% (Murthy et al., 2017). This range should be reported as such, with attribution, rather than collapsed. (E3)
• Mn is reported between approximately 0.05% and 1.1% across sources. A single value is not supportable. (E3)
• CaO is very sparsely reported. We located only two sources giving a figure (Nowacki et al., 2021, Ca 0.40%; trade sources, Ca ≈0.44% or CaO within a combined ≤1.5% with SiO₂). Any CaO range stated in a manuscript requires a specific source. (G)
• Al₂O₃ is reported between approximately 0.1% and 3 wt%. (E3)

TABLE 1. Reported mill-scale composition by measurement technique. Values are reproduced as published; ranges are not averaged across techniques because the techniques do not measure the same quantity. NR = not reported by that source.

2.3 Sources of compositional variability

The variability documented in Table 1 is not measurement noise; it is real and has identifiable causes, which should be reported alongside any composition. Table 2 sets out each source of variability, the mechanism by which it acts, and its practical implication for use in concrete (M).

TABLE 2. Sources of variability in mill-scale composition and properties, with the mechanism and the practical implication for use in concrete.

2.4 Particle morphology, size and physical properties

Mill scale particles are typically described as thin, layered flakes with high surface-area-to-volume ratio. mill scale has a specific gravity substantially higher than natural siliceous sand, which is why it is effective in heavyweight and shielding applications (Gavandi and Guptha, 2021; Sharma et al., 2024; Sylisomchanh et al., 2025), and its flake morphology and surface texture reduce workability at a given water content, an effect reported consistently across the corpus. (E2)

2.5 Reactivity: what can and cannot be claimed

A pozzolanic is, by definition, a siliceous or siliceous-and-aluminous material that reacts with calcium hydroxide in the presence of water to form cementitious compounds. Mill scale is neither siliceous nor aluminous in any substantial degree (Table 1). The mechanisms plausibly available to it are therefore: Figure 2 contrasts these pathways for fine-aggregate replacement and for binder replacement, and identifies where dilution does and does not operate.

• Filler and packing effects (M, well supported by analogy). Fine, dense, non-reactive particles improve particle packing and provide nucleation sites for C–S–H, which can accelerate early hydration and refine pore structure. This is a general effect of fine inert fillers and does not require mill scale to be chemically reactive.
• Dilution (M). When mill scale replaces cement, the clinker content and therefore the hydration-product volume falls in proportion. When it replaces fine aggregate, there is no dilution of the binder at all—a distinction the earlier version of this manuscript did not maintain, and which is essential to interpreting the literature correctly (see Section 4.1).
• Limited surface reactivity (M). Amorphous surface oxides and hydroxides may participate in reactions in a high-alkalinity pore solution, and iron-substituted C–S–H and iron-containing hydrate phases are known in other systems. Whether this occurs to a materially significant degree in mill-scale mortars is not established by the reviewed literature. (G)

3 Review methodology

3.1 Article type and reporting standard

A Systematic Review, under PRISMA 2020 (Page et al., 2021), requires a pre-specified, documented and reproducible search; a recorded screening process with reasons for exclusion at full-text stage; and, ideally, a registered protocol. The search underlying the present article was conducted iteratively and was not documented to that standard at the time. Rather than present an undocumented search as though it were systematic, we report the methodology transparently below and mark the elements that require re-execution.

3.2 Search strategy (as specified for re-execution)

The following search strategy is specified in reproducible form. It is presented as a protocol to be executed and logged, not as a record of a completed search.

3.2.1 Databases

Scopus; Web of Science Core Collection; Engineering Village (Compendex + Inspec); ScienceDirect; ASCE Library; and Google Scholar for grey-literature and citation-chase completeness.

3.2.1.1 Search string

• (TITLE-ABS-KEY (“mill scale” OR “millscale” OR “rolling mill scale”
• OR “billet scale” OR “steel scale waste” OR “iron oxide scale”
• OR “descaling waste” OR “lamination waste”))
• AND
• (TITLE-ABS-KEY (concrete OR mortar OR cement OR “cementitious”
• OR “geopolymer” OR “alkali-activated” OR “paving block”))
• AND
• (TITLE-ABS-KEY (strength OR “compressive” OR “flexural” OR “tensile”
• OR durability OR chloride OR carbonation OR “freeze-thaw”
• OR permeability OR sorptivity OR shielding OR leaching))

Languages. English, Portuguese and Spanish, with translation as required. Language restrictions must be reported.

3.3 Eligibility criteria

Inclusion. (i) Peer-reviewed primary experimental study; (ii) mill scale (as defined in Section 2.1) incorporated into a cementitious or alkali-activated matrix as fine aggregate, coarse aggregate, binder replacement or addition; (iii) at least one reported mechanical, transport, durability, shielding or environmental outcome; (iv) presence of a control mix without mill scale, or a dose series permitting a dose–response reading.

Exclusion. (i) Reviews and commentaries without primary data; (ii) studies of other steel by-products—steel slag, EAF slag, ladle furnace slag, iron ore tailings, copper slag, foundry sand—where mill scale is not itself a variable; (iii) studies in which mill scale is confounded with a co-additive with no mill-scale-only arm; (iv) publications in outlets that do not evidence peer review.

3.4 Study appraisal

The earlier version of this manuscript proposed a “modified Newcastle–Ottawa Scale.” The Newcastle–Ottawa Scale is an instrument for appraising non-randomised epidemiological studies (cohort and case–control designs); its domains—selection of cohorts, comparability, ascertainment of exposure and outcome—do not transfer meaningfully to materials testing. We have replaced it with a purpose-specified appraisal instrument. (P).

Application to the present corpus. Applying D7 and D8 to the studies identified in Section 4 produces the central finding of this review: the overwhelming majority of the corpus scores 0 on D8. Sample sizes and dispersion statistics are not reported in a form recoverable from the published record for most studies. Section 7 develops the consequences.

3.5 Data extraction

A structured extraction form is specified in Table 3 (Section 4.1). Extraction should be performed independently in duplicate with disagreement resolved by discussion, and the completed extraction sheet deposited.

TABLE 3. Primary studies on mill scale in cementitious and alkali-activated matrices, identified and bibliographically verified during this revision. NR = not reported at the level accessed.

4 What the primary literature reports

4.1 The corpus and its characteristics

Eighteen primary studies met the eligibility criteria of Section 3.3 and were bibliographically verified during this revision. They are listed in Table 3 and their design space is mapped in Figure 3. The corpus spans mortars, conventional and high-performance concretes, self-compacting concrete, geopolymer concrete and precast paving blocks, at replacement levels from 5% to 100%. Seven studies are from India, three from Brazil and two from Thailand, with single studies from Nigeria, Pakistan, the United States and Oman; two do not state a country. One study (Sylisomchanh et al., 2025) reports both a geopolymer and a Portland cement series, so the eighteen studies contribute nineteen study arms to Figure 3. For each study, Table 3 records the matrix, the constituent replaced, the replacement basis, the dose ladder, the curing ages, whether dispersion statistics and specimen counts were reported, and the headline finding as the source itself states it. The structural features of the corpus are set out in Section 4.2; the sources of material variability that no study in the corpus controls for are summarised in Table 2.

4.2 Structural observations about the corpus

• Almost all of it is fine-aggregate replacement. Only (Amorim et al., 2026a; Amorim et al., 2026b) confirm a cement-replacement series. Claims about mill scale as a binder-replacement material rest on a very thin base. (G)
• The replacement basis is not consistent. Some studies replace by mass, some by volume, and the abstracts of several do not state which. Because mill scale has a specific gravity roughly 1.5 times that of siliceous sand, a “20% replacement” by mass and by volume are materially different mixes. Pooling across them without conversion would be invalid. (E3)
• The reported optimum is not consistent. Mortar and self-compacting concrete studies cluster on approximately 15% (Siriwattanakarn et al., 2025; Raghavendra et al., 2025; Loganathan and Senthil Pandian, 2025), while several conventional concrete studies report optima at 50%–60% (Parvathikumar et al., 2023; Parvathikumar et al., 2024; Kushwah et al., 2024; Nawaz et al., 2024). This is a large discrepancy and should be treated as a genuine open question, not smoothed over. (E3)
• Direction of effect is contested. Pereira et al. (2011) report ∼20–23% strength reductions and Murthy (2021) reports 34.5% compressive loss at 60% replacement, while other studies report gains. (E3)
• Dispersion statistics are almost absent. Of the 18 studies in Table 3, only the two Lima et al. papers report replicate counts, error bars and inferential statistics—and those two are companion papers from one campaign. (E1)
• Non-independence is likely. At least two pairs—Parvathikumar et al. (2023)/Parvathikumar et al. (2024), and Amorim et al. (2026a)/Amorim et al. (2026b) — appear to derive from single experimental campaigns and should not be entered into any future synthesis as independent observations (E2).

4.3 Workability

Reduced workability with increasing mill-scale content is the most consistently reported effect in the corpus, noted by Murthy (2021) quantitatively (−74% slump at 60% replacement), by Siriwattanakarn et al. (2025), by Amorim et al. (2026b) and implicitly by Kushwah et al. (2024), whose central finding is that superplasticiser is required to make high replacement levels workable. (E2)

A plausible mechanism combines the flake morphology and rough surface texture of mill scale, which increase internal friction, with its surface hydroxide layers and any entrained rolling oil, which alter the water demand and the paste–particle interface. (M) Because workability was managed differently across studies—some held water constant, some held slump constant, some added admixture—the strength results across the corpus are not comparable without accounting for this. This is a first-order confounder and is treated as such in the framework of Section 7.

4.4 Compressive strength

The evidence here is genuinely contested (E3), and it is important not to present it otherwise.

• Studies reporting net benefit at moderate levels: Akindahunsi and Ojo (2008) at 15% (parity with control); Siriwattanakarn et al. (2025) at 15% by volume in mortar; Raghavendra et al. (2025) at 15% by mass in SCC; Loganathan and Senthil Pandian (2025) at 15% in mortar.
• Studies reporting net benefit at high levels: (Parvathikumar et al., 2023; Parvathikumar et al., 2024) at approximately 60%; Kushwah et al. (2024) up to 60% with superplasticiser; Nawaz et al. (2024) at a 50/50 blend selected by packing density.
• Studies reporting net loss: Pereira et al. (2011), 20%–23% average reduction; Murthy (2021), −34.5% at 60%.

The convergence of four independent studies on approximately 15% in mortar and SCC is the strongest single pattern in the corpus (E2). It is notable that these are the fine-matrix systems, where packing effects would be expected to matter most and where a dense fine filler has the clearest role. The high-replacement optima are reported predominantly in coarser conventional concrete, and in at least two cases in outlets where the reporting quality cannot be assessed.

Mechanistic interpretation (M). Two mechanisms plausibly compete, and Reviewer 1’s Comment 2 asks how they are distinguished. The honest answer is that in the present state of the literature they cannot be distinguished statistically, because no meta-regression has been performed and the necessary covariates are not reported. What can be said mechanistically is:.

• The filler/packing pathway dominates where mill scale is fine relative to the material it replaces and where the replacement level is low enough that gradation is improved rather than disrupted. Dense, fine, hard particles fill interstitial voids, increase packing density, provide nucleation sites for C–S–H and refine the pore structure. In fine-aggregate replacement this pathway operates without any dilution of the binder, because the cement content is unchanged.
• The dilution pathway applies strictly to binder replacement, where reducing clinker reduces the volume of hydration products available to fill space, and is not an explanation for strength loss in fine-aggregate replacement at all.
• In fine-aggregate replacement, the competing mechanism to filler action is not dilution but gradation disruption and water-demand increase. Beyond a threshold, additional fine flake material increases surface area faster than it improves packing, raises water demand, worsens workability, and—if water is added or compaction is impaired—increases porosity and reduces strength. The threshold depends on the fineness of the mill scale relative to the sand it replaces.

4.5 Tensile and flexural strength

Fewer studies report tensile or flexural outcomes, and those that do report them alongside compressive strength: Murthy (2021) reports flexural −27% and split tensile −19% at 60% replacement, against compressive −34.5% at the same level; Siriwattanakarn et al. (2025) report a flexural optimum coinciding with the compressive optimum at 15%; Loganathan and Senthil Pandian (2025) report peak tensile strength of 4.2 MPa at 15%. (E1/E2).

The earlier manuscript projected that “tensile and flexural strength reductions are … SMD 0.1–0.15 higher in absolute value” than compressive. That was a projection, not a result, and has been removed. What the corpus supports is only the directional statement that tensile and flexural outcomes track compressive outcomes with respect to the location of the optimum, in the small number of studies reporting both. (E1).

4.6 Durability and transport properties

Chloride transport. Siriwattanakarn et al. (2025) report improved chloride resistance in mortar at their optimum replacement. Sylisomchanh et al. (2025) report that mill-scale geopolymer concrete showed superior chloride resistance relative to Portland cement concrete—but note that this comparison is between binder systems, not between mill-scale and control within a binder system, and should not be read as a mill-scale effect. Parvathikumar et al. (2024) is the corpus’s dedicated durability study, reporting chloride penetration, sorptivity and acid attack. (E1, converging weakly to E2).

The earlier manuscript reported that “RCP tests typically reveal 10%–20% reduction in chloride diffusivity at 10%–15% mill-scale replacement, converging to control performance at >25%,” and elsewhere “10%–15% reduction.” Neither figure was derived from extracted data, the two statements are mutually inconsistent, and both have been removed.

A specific concern requiring attention (M/G). Mill scale is an iron-bearing material used in a matrix that in most structural applications contains steel reinforcement. Two questions follow, and we located no primary study addressing either:

• Does mill scale contribute chloride to the mix, and if so how much? The earlier manuscript stated an intrinsic chloride content of “0.1%–0.5%” without a mill-scale source. That figure has been removed. Chloride content should be measured and reported per batch. (G).
• Does the presence of dispersed iron oxide and residual metallic iron alter the corrosion behaviour of embedded reinforcement—through galvanic coupling, altered pore-solution chemistry, or provision of preferential cathodic sites. This is the single most consequential gap for structural application, and Section 8 places it first among research priorities. (G).

Carbonation We located no primary study in the corpus reporting carbonation depth for mill-scale concrete. The earlier manuscript’s statement that “carbonation depth increases modestly (5%–15%) with 10%–20% mill-scale replacement” is unsupported and has been removed. (G).

Freeze–thaw. We located no primary study in the corpus reporting freeze–thaw performance of mill-scale concrete. The earlier manuscript’s statement that “poorly cured specimens show 10%–30% higher mass loss” is unsupported and has been removed. This is a substantial gap for cold-climate application. (G).

Thermal and post-fire behaviour. This is better served than conventional durability. Siriwattanakarn et al. (2025) report improved post-fire residual properties and reduced thermal conductivity; Raghavendra et al. (2025) report approximately 85% strength retention at 500 °C. (E2).

Acid resistance. Raghavendra et al. (2025) report best acid resistance at 40% replacement—notably not at the strength optimum of 15%; Sylisomchanh et al. (2025) report improved acid resistance in the geopolymer system; Parvathikumar et al. (2024) include acid attack among their durability measures. (E2)

4.7 Accelerated laboratory testing and what it can support

Rapid chloride permeability (ASTM C1202), accelerated carbonation (EN 14630 for depth measurement, with accelerated exposure typically at elevated CO₂), rapid freeze–thaw cycling and sorptivity tests are comparative ranking tools. They compress a transport or degradation process by intensifying its driving force—raising the applied voltage, the CO₂ concentration, the cycling rate, or the ionic concentration. Their validity rests on the assumption that intensification changes the rate of a mechanism without changing the mechanism itself.

Where that assumption is safest. When comparing mixes that differ in only one variable, within one testing campaign, using one protocol. Under those conditions, a mix that transports chloride more slowly under an accelerated test will, other things equal, transport it more slowly in service. This supports relative statements. (M).

Where the assumption breaks, and specifically for mill scale. ASTM C1202 measures total charge passed, which is a function of pore-solution conductivity as well as pore structure. Introducing an electrically conductive, iron-bearing phase into the matrix is exactly the circumstance in which charge passed may cease to be a clean proxy for chloride diffusivity. Any RCPT result on mill-scale concrete should therefore be corroborated by a method that measures ion transport directly—a bulk diffusion test (e.g., ASTM C1556) or a migration test with chloride profiling. We are not aware of any study in this corpus that has done so, and we flag it as a methodological caution rather than a demonstrated artefact. (M/G).

What accelerated tests cannot support. They cannot be converted into service-life predictions for a novel material without (i) a validated degradation model for that material, (ii) field-calibrated transport coefficients, and (iii) a defined exposure class and limit state. None of the three exists for mill-scale concrete. Consequently:

4.8 Environmental behaviour and regulatory context

• Total elemental concentration in the mill scale (typically by XRF or acid digestion + ICP), expressed in mg/kg or ppm of solid. This tells you what is present. It is not what regulation is generally based on.
• Leachable concentration under a defined test, expressed in mg/L of eluate (US practice) or mg/kg dry substance at a stated liquid-to-solid ratio (EU practice). This tells you what is mobile under the test conditions.
• Regulatory threshold, which is jurisdiction-specific, test-specific, and applies to a defined decision (waste classification, landfill acceptance, product placement) — not to “environmental safety” in general.
• Zinc is not a Toxicity Characteristic contaminant under 40 CFR 261.24. A statement that mill-scale Zn is “well below hazard thresholds” cannot cite US TCLP, because there is no US TCLP threshold for zinc. The EU landfill acceptance criteria do include Zn, at 4 mg/kg (inert) and 50 mg/kg (non-hazardous) at L/S = 10.
• The US and EU values are not comparable. One is mg/L of eluate under a single-batch acetic-acid extraction; the other is mg/kg of dry solid under a water extraction at a specified liquid-to-solid ratio. They must never be presented in the same column or compared numerically.
• Landfill acceptance criteria are not product-placement criteria. Meeting inert-landfill WAC establishes that a material may be landfilled as inert waste. It does not establish that it is acceptable as a construction product. Product placement in the EU is governed by the Construction Products Regulation and applicable harmonised standards, and in some Member States by additional national criteria for the use of secondary materials.

Mill scale is derived from carbon steel surfaces and its trace-metal inventory is bounded by the composition of the parent steel; for plain carbon steel this inventory is expected to be low, and the iron oxide matrix is of low solubility at the high pH of concrete pore solution. That is a mechanistic expectation, not a measurement, and it does not extend to scale from alloy or stainless grades, where Cr and Ni content in the parent metal is by design substantial.

4.9 Radiation and electromagnetic shielding

Four independent studies report shielding benefits attributable to mill scale’s density and iron content: Gavandi and Guptha (2021) achieved 3100 kg/m³ for high-density ballast; Sharma et al. (2024) report a linear attenuation coefficient of 0.085 cm^-1 in a geopolymer composite and note improvement after heating to 1100° C; Sylisomchanh et al. (2025) report best shielding at 100% mill-scale replacement in both geopolymer and Portland cement systems; Loganathan and Senthil Pandian (2025) report EMI shielding of 26.37 dB at 11.82 GHz at 30% replacement in mortar. (E2).

Two observations matter for application. First, the shielding optimum lies at high replacement, while the mechanical optimum in mortar lies near 15% the two do not coincide, and a shielding mix must be designed against a strength requirement rather than assumed adequate. Second, this is a functional-performance niche in which mill scale’s high specific gravity is a direct advantage rather than an incidental property, which makes it the most defensible near-term application.

5 Pre-treatment and quality control

5.1 What the literature supports

Pre-treatment options reported or implied in the corpus include sieving and size classification, washing (to remove rolling oils and fines), drying, grinding, and thermal treatment. Sharma et al. (2024) report a specific and measured thermal effect: heating to 1100 °C improved shielding performance through phase transformation to denser phases. (E1) Pereira et al. (2011) document the presence of oils and grease at 0.252% in their material, which provides the physical rationale for washing. (E1)

5.2 Proposed quality-control protocol

The following is a proposal by the present authors (P), offered as a starting point for standardisation. It is not derived from an existing standard, and the acceptance criteria are stated as proposed values requiring validation.

6 Application domains: What the evidence currently supports

The applications below are ordered by the strength of the evidence currently available, not by commercial attractiveness. Table 4 states, for each application, the evidence located, the grade assigned under the convention of Table 5, and the resulting assessment; Figure 4 renders the same logic as a decision sequence for a candidate application. Two features should be read together. First, the best-supported applications are those in which mill scale’s density or its inertness is the functional property being bought — radiation and EMI shielding, heavyweight elements and ballast — rather than those in which it is a cost-driven substitute for sand. Second, every terminal node in Figure 4 carries the same three requirements, none of which the present evidence base can discharge on an adopter’s behalf: per-batch characterisation of the material (Table 6); mix qualification by testing at the receiving plant, rather than by extrapolation from a published optimum; and a regulatory pathway for the use of a secondary material as a construction product, which in most jurisdictions is distinct from waste-classification approval (Section 4.8). Structural reinforced concrete is not supported at any replacement level, for the reasons given in Section 4.6 and in Priority 1 of Section 8.

TABLE 4. Application readiness assessment. Evidence grades follow Table 5. This table states what the reviewed evidence currently supports; it does not specify performance limits.

TABLE 5. Evidence-grading convention applied throughout this review.

Terms such as proves, demonstrates conclusively, optimal, guarantees and will achieve are avoided except where a cited study’s own explicit claim is being reported as such.

TABLE 6. Proposed minimum characterisation protocol for mill scale intended for use in concrete. All acceptance criteria are proposed by the present authors and require experimental validation; they do not derive from any existing standard.

TABLE 7. Verified regulatory reference values (US Environmental Protection Agency, 1992; US Environmental Protection Agency, 2026; European Council, 2003).

7 Evidence synthesis and a proposed meta-analytic framework

7.1 Quantitative pooling

Seven obstacles, each documented in Section 4, jointly prevent pooling. They are listed in order of severity.

• Absence of dispersion statistics. A standardised mean difference requires, for each arm, a mean, a standard deviation and a sample size. Of the 18 studies in Table 3, only (Amorim et al., 2026a; Amorim et al., 2026b) report replicate counts and dispersion—and they are companion papers. A random-effects meta-analysis cannot be estimated from a corpus in which the variance components are unrecoverable. This alone is decisive.
• Insufficient replication where n is reported. Where replicate numbers appear in this literature at all, they are frequently 1–3 per mix per age. Effect-size estimates from n = 1 or n = 2 have standard errors so large that they contribute essentially no information to a pooled estimate while contributing substantially to apparent heterogeneity. Reviewer 3, Comment 7, is correct that figures in the earlier manuscript presented subgroup claims resting on n = 1 and n = 2; those figures have been withdrawn.
• Incommensurable replacement bases. Mass versus volume, and fine aggregate versus binder, are different interventions. Given mill scale’s specific gravity, a mass-basis and a volume-basis replacement at the same nominal percentage differ substantially in the volume fraction actually substituted. Pooling requires conversion, which requires both densities to be reported—often they are not.
• Heterogeneous outcome definitions. “Chloride resistance” appears as RCPT charge passed, as migration coefficient, and as sorptivity. “Durability” appears as acid mass loss, as post-fire residual strength, and as water absorption. These are not interchangeable outcomes and cannot enter a common synthesis.
• Uncharacterised intervention heterogeneity. Mill scale from different plants, routes and thermal histories is materially different material (Section 2.3), and most studies do not report enough provenance information to code this as a moderator.
• Confounding with co-additives. Iqra and Zahid (2024) vary graphite nano/microplatelets alongside mill scale; Nawaz et al. (2024) include steel fibres. Effect sizes from such designs are not attributable to mill scale.
• Non-independence. At least two study pairs derive from single campaigns and would violate the independence assumption if entered separately.

The evidence base is currently insufficient for a robust quantitative synthesis of the effect of mill scale on the mechanical and durability performance of concrete. Stating this plainly is more useful to the field than producing a pooled estimate whose inputs do not exist.

7.2 On the reviewer’s suggestion of ECDF and KDE analysis

ECDF and KDE are appropriate and useful for examining the distribution of a continuous variable across a sample. In an evidence-synthesis context they have three legitimate uses: (i) describing the distribution of extracted covariates across a corpus—for instance, the distribution of replacement levels, w/c ratios, or curing ages actually studied, which reveals where the literature is dense and where it is empty; (ii) describing the distribution of estimated effect sizes once these exist, as a distribution-free complement to the normality assumption implicit in standard random-effects models; and (iii) as a diagnostic for outliers and multimodality that a summary statistic would conceal.

All three uses require extracted study-level numerical data. Use (i) requires the replacement levels, mix parameters and ages to have been extracted across the corpus—Table 3 shows that many of these are not yet recoverable. Uses (ii) and (iii) require the effect sizes themselves, which cannot be computed without the dispersion statistics that Section 7.1(1) shows are absent.

There is a conceptual distinction worth stating, because it affects how the reviewer’s suggestion should be implemented when the data do exist. ECDF and KDE characterise the distribution of observations within a sample. Meta-analysis operates on study-level effect estimates, each of which carries its own sampling uncertainty, and the random-effects model explicitly partitions observed variation into within-study sampling error and between-study heterogeneity. A KDE of raw effect estimates therefore describes the marginal distribution of estimates—a convolution of the true effect distribution with sampling noise—and not the distribution of true effects. It is a useful descriptive diagnostic, but it is not a substitute for, nor a check on, the variance-weighted estimation that a random-effects model performs.

Section 7.4 incorporates the reviewer’s underlying concern—that distributional assumptions should be examined rather than assumed—into the proposed protocol as an explicit pre-specified step (step 7), including ECDF and KDE of extracted covariates and of effect estimates, alongside distribution-free alternatives to normal-theory random-effects estimation. We are grateful for the suggestion and have adopted it as a protocol requirement. Table 8 states the corresponding test protocol — standards, ages and the minimum number of replicates per mix per age required to yield a usable standard deviation.

TABLE 8. Proposed concrete performance testing protocol. Replicate numbers are proposed minima consistent with obtaining a usable standard deviation.

7.3 The proposed protocol

The following protocol is proposed for execution once the reporting deficiencies in Section 7.1 are remedied. It should be registered (for example, on PROSPERO or the Open Science Framework) before execution. Figure 5 presents the protocol as a flow chart, with an entry gate showing the point at which the present literature terminates.

Step 1 Effect-size definition. For continuous outcomes with a no-mill-scale control, Hedges’ g—the standardised mean difference with the small-sample correction—computed as the difference between mill-scale and control means divided by the pooled within-group standard deviation (Hedges and Olkin, 1985). Where all studies report the same outcome on the same scale, the raw mean difference is preferable to the SMD because it is directly interpretable in MPa or in coulombs. The choice must be pre-specified per outcome, not selected post hoc.

Step 2 Handling of dose. Mill scale is a dose-varying intervention, not a binary one. Two approaches should be pre-specified: (a) treating each replacement level as a separate comparison against the shared control, with the control variance appropriately split to avoid double-counting; or (b) dose–response meta-regression, which uses the dose ladder within each study and is the more informative approach where studies report three or more levels. Approach (b) is preferred for this literature, since most studies report a ladder.

Step 3 Estimation model. Random-effects, on the grounds that the studies differ in material, matrix, mix design and protocol and cannot be assumed to estimate a common true effect (DerSimonian and Laird, 1986). Restricted maximum likelihood for τ², with Knapp–Hartung adjustment for confidence intervals given the small anticipated number of studies. Where multiple outcomes or ages are extracted from the same study, a three-level or robust-variance-estimation model is required to account for within-study dependence; standard two-level models would understate uncertainty.

Step 4 Heterogeneity. Report τ² (the between-study variance, on the effect-size scale) as the primary heterogeneity measure, with a prediction interval, because these are directly interpretable. Report I² and the Q statistic as secondary, noting that I² is a ratio, not an absolute measure, and is sensitive to the precision of the constituent studies (Higgins and Thompson, 2002).

Step 5 Subgroup analysis. Pre-specify subgroups and limit their number. Proposed: matrix type (mortar/conventional concrete/SCC/geopolymer); replacement target (fine aggregate/binder); replacement basis (mass/volume); curing age (7/28/90 d). A minimum of five independent studies per subgroup should be required before a subgroup estimate is reported. Below that, report the constituent studies individually.

Step 6 Meta-regression. This is where Reviewer 1’s Comment 2 is properly answered. To distinguish packing effects from gradation-disruption effects statistically, the following moderators must be available across studies: replacement level (continuous); mill-scale fineness relative to the replaced aggregate (e.g., ratio of D₅₀ or of fineness modulus); w/c or w/b ratio; whether workability was held constant and by what means; superplasticiser dosage; and curing age. None of these can be coded across the present corpus. A rule of at least ten studies per moderator should be applied to avoid overfitting.

Step 7 Distributional diagnostics (incorporating Reviewer 1, Comment 1). Before model fitting: ECDF and KDE of the extracted covariate distributions, to characterise coverage of the design space and identify regions with no evidence. After model fitting: normal quantile–quantile plots and KDE of standardised residuals, to examine the normality assumption on the random effect; and, where that assumption is questionable, a distribution-free or robust alternative estimator as a sensitivity analysis.

Step 8 Sensitivity analysis. Leave-one-out; exclusion of grey-literature and non-indexed outlets; exclusion of studies with unreported or fewer than three replicates; collapsing of non-independent study pairs into single observations; and re-estimation under an alternative τ² estimator.

Step 9 Small-study effects. Contour-enhanced funnel plots and Egger’s regression (Egger et al., 1997), only if at least ten studies contribute to the outcome—below that threshold these tests have negligible power and should not be reported. Note that in materials science, asymmetry is at least as likely to reflect small-study methodological differences as publication bias, and should be interpreted accordingly.

Step 10 Certainty assessment. A GRADE-style summary of findings table, downgrading for risk of bias (Table 9 scores), inconsistency (τ² and prediction interval), indirectness (laboratory versus field), imprecision, and suspected small-study effects.

TABLE 9. Proposed appraisal instrument for primary studies of mill scale in cementitious materials. Each domain is scored 0 (not reported/inadequate), 1 (partially reported), or 2 (fully reported). Maximum 16.

Software. The protocol is implementable in metafor (Viechtbauer, 2010) in R, with robumeta or clubSandwich for robust variance estimation. Analysis code and the extraction sheet should be deposited openly.

7.4 Minimum reporting standard for primary studies

The single most useful contribution this review can make is to specify what primary studies must report for a future synthesis to be possible. (P).

8 Research gaps and priorities

The gaps below are ordered by the extent to which they block application, not by scientific interest. Figure 6 positions each priority by the size of the evidence gap and by the extent to which it blocks application.

Priority 1 Reinforcement corrosion (G). No electrochemical study of reinforced mill-scale concrete was located. Until half-cell potential, linear polarisation resistance and preferably impedance measurements are reported on reinforced specimens under chloride and carbonation exposure, structural use cannot be supported on any strength evidence whatever. This is the field’s highest-value single experiment.

Priority 2 Chloride and carbonation, measured properly (G). Carbonation depth is unreported in this corpus. Chloride results rest largely on RCPT, whose interpretation is questionable for a conductive-phase-bearing matrix (Section 4.7). Bulk diffusion or migration testing with chloride profiling is required.

Priority 3 Reporting reform (P). Adoption of Table 10 by primary investigators, and by reviewers and editors of this literature, would make quantitative synthesis possible within a single publication cycle. This costs nothing.

TABLE 10. Proposed minimum reporting standard for primary studies of mill scale in cementitious materials.

Priority 4 Freeze–thaw (G). Unreported in this corpus. Blocking for cold-climate application.

Priority 5 Resolving the replacement-level discrepancy (E3). The gap between the 15% optimum in mortar/SCC and the 50%–60% optima reported in conventional concrete is unexplained. A single well-designed study varying matrix type and mill-scale fineness across a common dose ladder, with proper replication, would resolve it.

Priority 6 Provenance sensitivity (G). No study was located that compares mill scale from multiple plants or process routes within one experimental programme. This is the direct experimental test of whether the compositional variability documented in Section 2.3 matters for concrete performance.

Priority 7 Environmental characterisation (G). No leaching data on mill scale or mill-scale concrete were located. Required for regulatory acceptance in any jurisdiction.

Priority 8 Field exposure (G). No field data of any duration were located. Required before service-life statements of any kind can be made.

Priority 9 Cement replacement (E1). Effectively one experimental campaign. Independent replication needed before the mode is treated as viable.

9 Sustainability, industrial symbiosis and economics

9.1 The industrial-symbiosis case, stated carefully

Mill-scale valorisation in concrete is a textbook industrial-symbiosis configuration: a by-product of one sector becomes an input to another, and both the disposal burden and a virgin-material demand fall simultaneously. The qualitative case is strong and does not depend on contested numbers. (M)

• Waste valorisation. Mill scale is currently landfilled or downcycled in substantial quantity; more than 30% of European oily sludge and mill scale is reported as not valorised (Iluțiu-Varvara et al., 2020). Diversion removes a disposal cost and an environmental liability from the steel producer.
• Natural aggregate conservation. Fine-aggregate replacement directly displaces extracted sand, whose supply is constrained and whose extraction carries river-system and coastal impacts. The magnitude of displacement is a simple mass balance once a replacement level and an adoption rate are assumed—but both are assumptions, and the resulting figure should be presented as a scenario, not a finding.
• Co-location advantage. Mill scale is dense, and transport cost scales with mass. The economics therefore favour ready-mixed or precast operations near integrated steel plants. This is a real and specific industrial insight: the viable market for this material is geographically bounded by its own density.
• Processing simplicity. Unlike materials requiring grinding, calcination or chemical activation, mill scale can be used with sieving and drying alone in most reported studies. Processing energy is correspondingly low.
• Functional-performance premium. In shielding applications (Section 4.9), mill scale is not a cost-driven substitute but a functional ingredient. This is where value capture is highest and where standardisation barriers are lowest, because these are typically specialist rather than code-governed applications.

Quality-control and scalability requirements. Against these benefits, industrial adoption requires: per-batch characterisation (Table 6), because scale composition drifts with steel grade and process; a supply agreement covering consistency, not merely availability; a mix-qualification programme at the receiving plant, since this review’s evidence does not support transferring a published optimum to a new material; and a regulatory pathway for the use of a secondary material as a construction product, which in most jurisdictions is distinct from waste-classification approval (Section 4.8).

10 Discussion

10.1 Mechanical performance

The corpus reports mill scale as a viable partial fine-aggregate replacement, with a strength optimum whose location is matrix-dependent and, at present, contested. The clearest signal is the convergence of four studies from independent groups on approximately 15% in mortar and self-compacting concrete (Siriwattanakarn et al., 2025; Raghavendra et al., 2025; Loganathan and Senthil Pandian, 2025; Akindahunsi and Ojo, 2008 at 15% in concrete). (E2) Against this sit reports of optima at 50%–60% in conventional concrete and reports of monotonic strength loss. (E3) The field should treat the discrepancy as informative rather than as noise: it most likely reflects the interaction between the fineness of the mill scale and the gradation of the sand it replaces, which is exactly the moderator that Section 7.3 step 6 identifies as necessary and currently uncodeable.

10.2 Durability

Durability evidence is thin and unevenly distributed. Thermal, post-fire and acid-resistance outcomes are reported by several studies and are directionally favourable. (E2) Chloride evidence exists but rests substantially on a test method whose applicability to iron-bearing matrices deserves scrutiny. Carbonation, freeze–thaw and reinforcement corrosion are unreported in this corpus—three gaps that together define the boundary of current applicability. (G)

10.3 Effect of replacement level

Replacement level is the dominant design variable and the one most inconsistently reported. Until studies state the replacement basis unambiguously and report the gradation of both materials, the dose–response relationship cannot be established across the literature, only within individual studies.

10.4 Influence of mill-scale characteristics

Section 2 establishes that mill scale is a variable material with identifiable causes of variability. Section 4 establishes that no study has tested whether that variability matters for concrete performance. This is the most striking structural gap in the field: a literature that repeatedly notes compositional variability as a barrier to adoption, and has not yet run the experiment that would quantify its consequence.

10.5 Pre-treatment

Thermal treatment has one documented functional effect, on shielding performance (Sharma et al., 2024). (E1) Washing has a clear mechanistic rationale in oil removal (Pereira et al., 2011). Beyond this, the pre-treatment literature is a set of plausible propositions rather than a body of evidence, and the earlier manuscript’s quantified pre-treatment benefits have been withdrawn.

10.6 Environmental safety

The regulatory framework is now stated correctly and the three distinct quantities separated (Section 4.8; Table 7). What is missing is data: no leaching results on mill scale or mill-scale concrete were located. The mechanistic expectation of low mobility in a high-pH matrix is reasonable but is not a substitute for measurement, and does not extend to scale from alloy grades.

10.7 Sustainability

The qualitative case is strong; the quantitative case has not been made. The specific correction that matters is that fine-aggregate replacement does not reduce clinker content, so the carbon benefit of mill scale is a sand-extraction and disposal-avoidance benefit rather than a cement-decarbonisation benefit, offset by transport impacts that scale with the material’s density. This is a smaller but more defensible claim than the one the earlier manuscript made.

10.8 Economic and industrial implications

Adoption economics are dominated by transport, because the material is dense; by quality control, because it is variable; and by the absence of a standard, because that shifts qualification cost onto every individual user. The shielding niche escapes the first two constraints partially and the third almost entirely, which is why it is the most credible near-term route to market.

11 Conclusion

This review set out to determine what the literature on steel mill scale in concrete and mortar currently supports. Its principal conclusion is methodological, and it is the conclusion most useful to the field.

• Mill scale is usable as a partial replacement for fine aggregate in mortar and concrete, and several independent studies report a strength optimum near 15% by mass or volume in mortar and self-compacting concrete. (E2).
• Workability decreases with increasing mill-scale content, consistently across studies. (E2).
• Mill scale confers density-related functional benefits—radiation and electromagnetic shielding, heavyweight elements—that are reported consistently and that follow directly from the material’s physical properties. (E2).
• Post-fire residual performance, thermal behaviour and acid resistance are reported as favourable by more than one group. (E2).
• The location of the strength optimum in conventional concrete, where reported optima range from monotonic decline to 60% replacement. (E3).
• Whether mill-scale provenance and composition materially affect concrete performance. (G).
• Chloride transport behaviour measured by methods appropriate to a conductive-phase-bearing matrix. (G).
• Whether mill scale functions as anything other than an inert dense filler in a cementitious matrix. (G).

Radiation and EMI shielding, heavyweight and ballast concretes, mortars and renders at moderate replacement, and non-structural elements including paving blocks. In each of these, mill scale’s density or its inertness is an asset, and the consequences of performance variability are bounded.

Not strength. The corpus contains no study of reinforcement corrosion in mill-scale concrete, no carbonation data, no freeze–thaw data, and no field performance data of any duration. An iron-bearing admixture in reinforced concrete raises a corrosion question that has not been asked, let alone answered. Until it is, no strength result can support structural use. The absence of a product standard compounds this by placing full qualification responsibility on each adopter.

In order: electrochemical assessment of reinforcement in mill-scale concrete; proper chloride and carbonation measurement; and—costing nothing and enabling everything else—adoption of a minimum reporting standard (Table 10) by primary investigators. The reason this review presents a meta-analytic protocol rather than a meta-analysis is that the necessary means, standard deviations and sample sizes are not in the published record. That is a solvable problem, and solving it is a precondition for the field to know what it collectively knows.

Generative AI statement

The author(s) declared that generative AI was used in the creation of this manuscript. During the preparation of this work the authors used Anthropic’s Claude to condense and language-edit the manuscript, to draft the response to reviewers, and to implement and execute the cross-validation analyses reported from the authors’ own simulation dataset and training code. All simulations, field data and scientific conclusions are the authors’ own; the authors reviewed and edited the output and take full responsibility for the content of the publication.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Matteo Sambucci, Sapienza University of Rome, Italy

Reviewed by: Robi Sonkor Mozumder, Prairie View A&M University, United States

Reshma Malipeddi, Dhanekula Institute of Engineering, India