Sustainable Approaches to Corrosion Protection in Reinforced Concrete: A Critical Review of Green Inhibitors, Surface Treatments, Self-Healing Systems, Nanomaterials, Smart Coatings, and Life-Cycle-Based Design

1. Introduction

Reinforced concrete carries most of the world’s bridges, ports, tunnels, and buildings. Its long life depends on a thin passive film that forms on the steel in the highly alkaline pore solution of concrete [1,2]. When chloride ions reach the steel in a high enough amount, or when carbonation lowers the pH, this film breaks down and corrosion starts [3,4,5]. The rust takes up more space than the steel it replaces. It cracks the cover, weakens the bond between steel and concrete, and reduces load capacity in bending and shear [6,7,8].

The cost of corrosion is not only financial. Each early repair brings new cement, new steel, traffic delays, and more emissions. Angst has argued that steel corrosion may be the weakest link for low-carbon concrete, because lowering the clinker content changes the alkaline buffer that protects the steel [9]. This creates a real tension. Designers want less clinker to cut carbon, but they also need the structure to last. A mix with a slightly higher initial footprint can still be the greener option if it doubles the service life [10,11]. We think this point is often missed when new materials are called “sustainable” based on their mix design alone.

Protection methods have moved well past calcium nitrite and epoxy-coated bars. Researchers now test plant extracts as inhibitors, silane and lithium-based surface treatments, bacteria that seal cracks, nanoparticles that refine the pore structure, and coatings that release inhibitors only when corrosion begins [12,13,14,15,16]. Design practice has also changed. Codes and researchers are moving from fixed cover and water-to-binder rules toward performance-based and probabilistic service-life design [11,17,18]. Machine learning now appears in most durability journals, often to predict chloride diffusion from mix proportions [19,20,21].

Several good reviews already exist. They cover green inhibitors [12,22,23], concrete surface treatments [24], self-healing concrete [25,26], corrosion assessment and control [27], and durable concrete in chloride environments [28]. Most of them, however, look at one family of methods at a time. What is still missing, in our view, is a review that puts the protective measure, the service-life model, and the life cycle metric in the same frame. A green inhibitor that works for five years in the field is not green if the structure needs major repair at year twenty.

This review tries to close that gap. It has five aims: (i) explain the corrosion processes that each protection method targets; (ii) assess the evidence for eco-friendly inhibitors, surface treatments, self-healing systems, nanomaterials, and smart coatings; (iii) discuss supplementary cementitious materials (SCMs) and low-carbon binders as the first line of defense; (iv) connect these materials to durability-based design, life cycle assessment (LCA), predictive modeling, and multi-scale evaluation; and (v) identify research gaps. Figure 1 shows how the review is organized.

2. Review Approach

This is a critical narrative review. We identified sources through database and web searches and through backward citation tracking of recent reviews. Search terms combined “steel corrosion”, “reinforced concrete”, “chloride”, “carbonation”, “green inhibitor”, “surface treatment”, “self-healing”, “nanomaterial”, “smart coating”, “supplementary cementitious material”, “service life”, “life cycle assessment”, and “machine learning”. We kept peer-reviewed journal papers, RILEM state-of-the-art reports, and standard reference books. We gave priority to studies in which steel was embedded in mortar or concrete. We included a few studies in acid or neutral saline media only where they explain a mechanism that also matters in concrete, and we say so where we use them. The final set has 100 sources published between 1993 and 2026.

We read each study with three questions in mind. Where does the method act: on transport, on the chemistry at the steel surface, on the steel itself, or on cracks? How long was the test, and was it done in solution, mortar, concrete, or a real structure? Did the authors report anything that can feed a service-life or life cycle model, such as a diffusion coefficient, a chloride threshold, or a corrosion rate? These questions shape the tables and the critical comments in each section.

3. Corrosion of Steel in Concrete: What Protection Must Address

3.1. Passivity and the Steel–Concrete Interface

Fresh concrete has a pore solution pH of roughly 12.5 to 13.5. At this pH, carbon steel forms a thin, protective oxide film [1]. The film is not perfect, and the zone around the bar is not uniform. Mill scale, rust from storage, air voids, and bleed water under horizontal bars all change the local conditions [29,30]. RILEM TC 262-SCI reviewed how these local features affect the start of chloride-induced corrosion. The committee showed that several of these effects are still unclear and that test results often conflict [31]. Steel surface condition alone can shift the onset of corrosion, as Boubitsas and Tang found when they compared bars with different surface states [32]. For protection design, the lesson is simple. Any method that improves or stabilizes the interface, such as a good inhibitor or a dense, well-cured cover, has value beyond what bulk tests show.

3.2. Chloride-Induced Corrosion

Chlorides come from seawater, de-icing salts, or contaminated mix materials. They move into concrete by diffusion, capillary suction, and permeation, and wet–dry cycles speed up the process near the surface [4,28]. Part of the chloride binds to the cement paste. Aluminate phases can form Friedel’s salt, and calcium silicate hydrate adsorbs chloride on its surface. Only the free chloride drives corrosion, so binding lowers the risk [33,34].

Corrosion starts when the chloride content at the steel passes a critical value, often called the chloride threshold. This value is not a material constant. A widely cited review found a very wide spread of reported thresholds, caused by different test methods, steel surfaces, binders, and exposure conditions [3]. Researchers keep proposing better ways to measure it. Examples include potential monitoring during slow chloride uptake in slag concretes [35], a short-term test for systems with corrosion-inhibiting admixtures [36], and regression models that link the threshold to the water-to-binder ratio and to the fly ash and slag contents [37,38]. We see the threshold as the single largest source of uncertainty in chloride service-life models, and it deserves more attention than it gets in green-material papers.

Exposure zone matters as much as the material. In the tidal and splash zones of marine structures, repeated wetting and drying moves chloride inward by capillary suction and then leaves it behind as the water evaporates. This creates a surface “convection zone” where Fick’s law does not hold, and the depth of this zone grows when load-induced microcracks are present [28]. Temperature and relative humidity also change the rate of transport, so the same concrete can have very different service lives in a hot, humid coast and in a cool, dry inland site [39]. For protection design, this means that surface treatments and self-healing agents are most useful exactly where the convection zone is deepest, while SCMs and low water-to-binder ratios govern the slower diffusion deeper in the cover [4,28].

3.3. Carbonation-Induced Corrosion

Carbon dioxide from the air reacts with calcium hydroxide and other hydrates. This lowers the pH near the steel to about 9, and the passive film is lost over a wide area [1]. The corrosion rate then depends mostly on moisture and oxygen. Sánchez Montero and co-workers followed carbonation-induced corrosion in real elements. They found that the average corrosion rate in elements with unacceptable section loss was about eight times higher than in cracked elements without section loss [40]. This means that the propagation phase is not steady, and a single corrosion rate in a model can mislead.

Chloride and carbonation often act together, for example in coastal cities or in car parks. Carbonation can release bound chloride and raise the free chloride at the steel [5,28]. Jiang et al. compared macrocell corrosion under carbonation, admixed chloride, and chloride penetration. At the same anodic potential drop, the macrocell current was higher for chloride-induced corrosion than for carbonation. Concretes with slag and fly ash were less affected by chloride contamination than plain Portland cement concrete, most likely because they bind more chloride [41].

3.4. Propagation, Cracking, and Structural Effects

Once corrosion starts, the rate depends on concrete resistivity, moisture, oxygen supply, temperature, and the ratio of anode to cathode area. Otieno et al. reviewed corrosion-rate models and showed that many are hard to transfer between laboratory and field conditions [42]. Cracks make things worse. They act as fast paths for water and chloride, and numerical models built from tests on cracked and coated concrete show strong local attack at the crack [43,44]. At the member level, corrosion reduces flexural capacity [6,7] and shear capacity when stirrups corrode [8]. Fibers can change both corrosion resistance and bond behavior [45].

3.5. What This Means for Protection

Service life is usually split into an initiation phase and a propagation phase (Figure 2) [1,17]. Most sustainable strategies try to lengthen initiation. They slow chloride or carbon dioxide transport, raise the threshold, or seal cracks. A smaller group works during propagation. These include migrating inhibitors, smart coatings that respond to a pH or potential change, and repair guided by monitoring. Table 1 summarizes where each family of methods acts and what limits it.

4. Corrosion Inhibitors

4.1. Conventional Inorganic and Organic Inhibitors

Corrosion inhibitors are chemicals that slow corrosion when present at a small dose. In concrete, they are either mixed into fresh concrete or applied to the surface of existing structures so that they migrate toward the steel [27,49]. They are grouped by the reaction they block: anodic, cathodic, or mixed. Calcium nitrite is the classic anodic inhibitor and the usual benchmark. It helps oxidize ferrous ions and repair the passive film. Its main drawbacks are toxicity concerns and the need to keep a high nitrite-to-chloride ratio. At too low a dose, an anodic inhibitor can make local attack worse [12,28].

Organic inhibitors work mostly by adsorbing on the steel and forming a film that blocks charge transfer or water access. Ormellese et al. screened 80 organic compounds in synthetic pore solution. They then carried the nine best substances, mainly amines, aminoalcohols, carboxylates, and amino acids, into long-term tests in concrete [50]. The same group later used atomistic simulations to study how a triethylenetetramine-based inhibitor film sits on the passive layer and competes with chloride [58]. This kind of molecular work is useful because it explains why some molecules survive in pore solution and others do not.

Migrating corrosion inhibitors are attractive for existing structures because they need no demolition. Their weak point is proof of arrival. Several reviews point out that it is hard to confirm how much inhibitor reaches the steel and how long it stays there [22,49]. Test methods matter here as well. Karuppanasamy and Pillai showed that some accelerated chloride tests are not suitable for systems with anion-containing inhibitors, and they proposed a modified short-term threshold test [36].

4.2. Green Inhibitors from Plants and Wastes

Green inhibitors are made from plant extracts, food waste, or other natural sources. Their molecules usually contain nitrogen, oxygen, sulfur, or phosphorus, along with aromatic rings or double bonds. These features let them adsorb on the steel and form a protective layer [12,23]. Table 2 lists studies that tested such extracts with steel in concrete or in conditions close to it.

The results look promising. Bambusa arundinacea leaf extract reduced corrosion of steel in concrete [59]. Olive leaf extract gave protection in concrete contaminated with seawater [62]. Maize gluten meal extract was studied with both electrochemical tests and theoretical calculations, which helped explain how its molecules bind to steel [63]. Neem (Azadirachta indica) leaf extract was compared with three commercial inorganic inhibitors under a chloride environment, and it did not change the fresh or hardened properties of the concrete in that study [64]. Other work combined plant extracts with silver nanoparticles [60] or used eco-addition inhibitors in reinforced concrete [61].

We still urge caution. Casanova et al. compared green inhibitors with nitrite and separated accelerated tests from long-term tests. They found that the evidence from long-term tests is much thinner [12]. Many studies with very high inhibition efficiencies were done in acid solution, such as fenugreek seed extract for steel in phosphoric acid [65]. Those numbers do not transfer to pH 13. Natural extracts also vary with harvest, region, and extraction method. Some contain sugars that may delay cement setting. Few papers report chemical stability of the extract in pore solution over years. Zomorodian and Behnood proposed selection criteria that weigh efficiency, toxicity, cost, and compatibility with concrete [22]. We support that approach and suggest adding one more criterion: the inhibitor’s effect on the chloride threshold measured in mortar or concrete, not only its efficiency in solution.

4.3. Encapsulated and Nano-Enabled Inhibitors

A free inhibitor can leach out, react with cement phases, or be used up before corrosion starts. Encapsulation tries to fix this. The inhibitor is loaded into micro- or nanocontainers, such as layered double hydroxides, halloysite tubes, or polymer capsules, and it is released when a trigger appears, such as a drop in pH or the arrival of chloride [16,28,54]. Koleva showed that tailored polymeric nanomaterials can change the corrosion products and the protective layer on steel in cement-based systems, which points to self-healing of the steel surface itself [67]. This is an exciting direction. However, almost all data come from laboratory specimens, and dosage, cost, and effects on hydration need more study before field use.

4.4. Testing, Dosage, and Interaction with Binders

The way an inhibitor is tested decides what its result means. Tests in synthetic pore solution are fast and good for screening, but they leave out the cement paste, the steel–concrete interface, and the slow transport of chloride through the cover [12,50]. Tests in mortar or concrete are slower but closer to practice. Accelerated tests that drive chloride with an electric field can distort the result for inhibitors that carry an electric charge, which is why modified threshold tests have been proposed [36]. We recommend that any new inhibitor be reported with at least three results: its efficiency in pore solution, its effect on the chloride threshold in mortar or concrete, and its effect on setting time and 28-day strength.

Inhibitors and SCMs are often used together, and they can help each other. The SCM slows chloride ingress, so the inhibitor is used up more slowly, and the inhibitor adds a chemical barrier where the SCM is weak, for example under carbonation. There can also be conflict. Alumina-rich SCMs such as metakaolin may adsorb part of an organic inhibitor and reduce the amount that reaches the steel [28]. Very few studies test inhibitor type and binder type in a full factorial design. Until they do, designers cannot assume that the benefits add up [22,49].

5. Surface Treatments and Coatings

5.1. Concrete Surface Treatments

Surface treatments protect the cover concrete, which is the zone that decides how fast chlorides and carbon dioxide reach the steel. They fall into three groups: film-forming coatings, pore blockers, and hydrophobic impregnations such as silanes and siloxanes [24,68]. External coatings designed for both concrete and steel protection were studied as early as the 1990s [69].

Silane-based treatments are the best documented. Coppola et al. studied a silane-based surface-applied corrosion inhibitor. It cut the non-steady-state migration coefficient by 30% to 60% in accelerated tests. In natural diffusion tests, the apparent diffusion coefficient fell by about 75%. Resistivity data showed that the main mechanism was water repellence [13]. In a marine setting, surface treatments changed both the chloride diffusion coefficient and the capillary absorption of the cover [70]. Silane can also be added to the mix as a hydrophobic admixture, and this also affects steel corrosion [71]. Newer lithium-based and nano-modified waterproof coatings on mortar have shown resistance to salt attack [14]. Polymer-modified cementitious coatings have been followed for years in chloride exposure, which gives rare long-term evidence [72].

Hydrophobic treatments have a known weak spot: cracks. Dai et al. showed that the benefit of water-repellent impregnation depends strongly on whether the concrete is cracked [51]. Treatments also age and need to be reapplied. In our view, a surface treatment should be treated in life cycle studies as a maintenance cycle, with its own reapplication interval and its own emissions, not as a one-time addition.

5.2. Coatings on the Steel and Alternative Reinforcement

Steel-level protection includes changes to steel chemistry and coatings on the bar [73]. Epoxy-coated bars work well when the coating is intact, but damage during transport and placing can create local weak spots. Bio-based coatings are being explored. For example, chitosan–gum Arabic nanocomposite coatings on mild steel in saline media raised the water contact angle from about 51° to about 101° and gave good barrier protection [66]. That test was not in concrete, so it shows promise rather than proof. Non-metallic reinforcement, such as glass, basalt, or carbon fiber-reinforced polymer bars, removes electrochemical corrosion altogether but brings new questions on bond, brittleness, and long-term resin degradation [28]. Fibers in the concrete also affect cracking and bond under corrosion [45].

5.3. Electrochemical Protection and Repair

Electrochemical methods act on the steel directly. Cathodic protection shifts the steel potential so that it no longer corrodes, either with an impressed current or with sacrificial anodes. Electrochemical chloride extraction draws chloride ions out of the cover with a temporary current, and realkalization restores a high pH in carbonated concrete [1,27]. These methods are well proven in the field and can stop active corrosion without removing sound concrete, which saves material and waste. Their sustainability depends on energy use over the years of operation, on the life of the anodes, and on side effects such as changes in pore structure near the steel. Polder et al. pointed out that such non-traditional methods often face non-technical barriers, including owner habits and procurement rules, as much as technical ones [74]. In a life cycle view, a cathodic protection system is a maintenance stream with its own energy and replacement schedule, and it should be modeled that way in Equation (3).

6. Self-Healing Systems

6.1. Why Cracks Matter for Corrosion

Concrete cracks under load, shrinkage, and temperature change. Cracks give chloride, water, and oxygen a short path to the steel [43,51]. Self-healing concrete aims to close these paths without human work. Van Tittelboom and De Belie grouped the approaches into autogenous healing, which relies on continued hydration and calcium carbonate precipitation, and autonomous healing, which uses added agents such as bacteria, capsules, or vascular networks [25]. A later review by De Belie et al. described how self-healing fits into damage management of structures [26].

6.2. Bacterial and Biogenic Healing

Jonkers et al. used alkali-resistant, spore-forming bacteria as a healing agent. The bacteria turn a calcium source into calcium carbonate that fills cracks [15]. Achal et al. showed that biogenic treatment improved durability and repaired cracks in concrete [75]. For corrosion, the most useful evidence comes from transport tests. Ling and Qian found that healed cracks in bacterial concrete reduced chloride transmission during electromigration [52]. Bacteria-based healing has also been combined with steel fibers [76].

6.3. Capsule-Based and Nanomaterial-Assisted Healing

Capsule systems release polymers or mineral agents when a crack breaks the capsule. Nanomaterials can support both healing and densification, and their sustainability depends on dosage, energy use in production, and service-life gain [53]. Koleva’s work links self-healing of the matrix with self-healing of the steel surface layer [67].

6.4. Critical View

Most self-healing studies measure crack closure, water flow, or strength recovery. Fewer measure chloride profiles at the steel depth or corrosion current. Visual closure at the surface does not prove that the crack is sealed at the bar. We recommend that future work report at least one transport measure at the steel depth and one electrochemical measure, such as corrosion potential or polarization resistance, on healed and unhealed specimens. Without this, we cannot put self-healing into a service-life model.

7. Nanomaterials

Nano-silica, nano-titania, nano-clay, carbon nanotubes, and graphene-based materials have been added to cement-based materials to refine pores, act as nucleation sites for hydration products, and block ion transport [28,53]. Nano-silica and carbon nanotubes have also been used to improve seawater sea-sand concrete, which is important for island construction where fresh water and river sand are scarce [28]. Nanoparticles can also serve as inhibitor carriers or as active agents, as in the Elaeis guineensis and silver nanoparticle system [60].

Three problems slow adoption. First, nanoparticles clump together, so mixing energy and dispersants matter as much as dosage. Second, results vary a lot between studies because of differences in particle type, surface treatment, and test method. Third, the production of some nanomaterials is energy-intensive, and their health and environmental effects are not fully known. A nanomaterial that cuts chloride diffusion by a large fraction may still be a poor choice if its production footprint cancels the service-life gain. This is exactly why we argue for life cycle evaluation per year of service life (Section 11).

8. Smart Coatings

Smart or active coatings do more than block the environment. They sense a change and respond. The main triggers are pH, redox potential, chloride, and mechanical damage [77,78]. Shchukin and Möhwald described coatings with nanoreservoirs that release inhibitors when corrosion begins [54]. Vimalanandan et al. built a redox-responsive coating that releases its inhibitor when the potential at a defect shifts, which gives protection only where and when it is needed [55]. Liu et al. reviewed self-healing coatings based on micro- and nanocarriers and discussed how loading capacity and release rate can be controlled [16].

Most smart coating research targets steel in air or water, not steel inside concrete. Transfer to concrete brings new conditions: very high pH, calcium-rich pore solution, abrasion during casting, and the need for good bond. We see two realistic paths. One is smart coatings on the bar with carriers stable at pH 13. The other is cementitious or hybrid surface coatings that carry encapsulated inhibitors near the concrete surface. Both need long-term tests in real concrete before they can enter design.

9. Supplementary Cementitious Materials and Low-Carbon Binders

9.1. How SCMs Affect Corrosion

SCMs such as fly ash, ground granulated blast-furnace slag, silica fume, metakaolin, and calcined clay replace part of the clinker. They lower the carbon footprint and, in most cases, improve resistance to chloride [46,79,80]. They work in three ways. They refine the pore structure and so lower the diffusion coefficient. Alumina-rich SCMs bind more chloride. They raise the electrical resistivity of concrete, which slows corrosion once it starts [46,81]. Abd El-Fattah et al. confirmed that the chloride binding capacity of concrete changes with the type of cementitious system [34].

The same reactions have a cost. Pozzolanic reactions consume calcium hydroxide, which lowers the alkaline reserve and can speed up carbonation [9,46]. So an SCM that is excellent in a marine splash zone may be weak in a dry, carbonation-dominated setting. Huang et al. summarized practical replacement ranges of about 5% to 15% for silica fume, 5% to 10% for metakaolin, and 20% to 30% for fly ash, while noting that no single SCM is best for all exposures [28]. Table 3 compares the main binder options.

9.2. Limestone Calcined Clay Cement

Limestone calcined clay cement (LC3) combines clinker, calcined clay, and limestone. It can cut clinker content by about half while giving similar strength [47,83]. Its dense pore structure and high alumina content give strong resistance to chloride. Pillai et al. combined service-life estimates and LCA for LC3 concrete. They reported that LC3 mixes can be designed to reach annual emissions of about 1.6 to 5 kg CO₂-eq per cubic meter per year of service life, similar to fly ash concrete and much lower than plain Portland cement concrete in their case [10]. The carbonation resistance of LC3 is lower, so cover and curing matter more.

9.3. Alkali-Activated Binders and Emerging SCMs

Alkali-activated materials and geopolymers use industrial by-products with little or no clinker [48]. Adding slag to fly ash-based geopolymer concrete improved its durability indicators in several studies [84,85], and sea sand has been tested in fly ash geopolymer mortars [86]. The corrosion of steel in these binders is less understood than in Portland systems. Pore solution chemistry, sulfide content in slag systems, and carbonation behavior all differ. Standards and long-term field data are also limited. Snellings et al. reviewed new SCM sources, such as calcined clays of lower grade and recycled concrete fines, which will matter as fly ash and slag supplies shrink [82]. Aggressive water exposures, including acid and sulfate, add further limits for all these binders [87].

9.4. Optimizing the Binder for Performance and Footprint

Choosing a binder is now a multi-objective problem. Hafez et al. built a data-driven tool that weighs functional, economic, and environmental properties of blended cement concrete [88]. Moro et al. proposed a performance-based concept that selects sustainable concretes for chloride exposure by checking both the required service life and the global warming potential [11]. Machine learning with Bayesian tuning has been used to predict the strength of eco-friendly concrete with slag and recycled aggregate, which helps screen mixes quickly [89]. We think this kind of joint screening should become routine.

9.5. Seawater, Sea Sand, and Local Materials

In many coastal and island regions, fresh water and river sand are scarce. Using seawater and sea sand saves these resources, but it adds chloride to the concrete from day one. With ordinary steel, this is a serious risk, because the initial chloride content C0 in Equation (1) is no longer zero. Two routes are being explored. One is to pair seawater sea-sand concrete with non-metallic reinforcement, which removes the steel corrosion problem but raises questions about bond and long-term resin durability [28]. The other is to use binders that bind more chloride, such as slag-rich systems, or alkali-activated systems in which sea sand has been tested [34,86]. For hot and arid coastal regions, such as the Arabian Gulf, where high temperature, high salinity, and wet–dry cycling occur together, we see this as one of the most practical research areas in the field.

10. Durability-Based Design and Service-Life Modeling

10.1. Chloride Ingress Models

Most service-life models for chloride exposure use Fick’s second law. For a semi-infinite medium with a constant surface chloride content, the chloride content C at depth x and time t is:

C(x,t) = C0 + (Cs − C0) · [1 − erf( x / (2√(D·t)) )]

(1)

where C0 is the initial chloride content, Cs is the surface chloride content, and D is the apparent diffusion coefficient. Corrosion starts when C at the steel depth reaches the threshold Ccrit. Setting C0 = 0 and solving for the initiation time ti gives:

ti = x² / (4 · D · z²), with z = erf⁻¹(1 − Ccrit/Cs)

(2)

Khan et al. reviewed these solutions and noted that the Fick-based solution tends to be conservative compared with more detailed models [18]. The water-to-cement ratio changes several inputs at once, including D, Ccrit, tensile strength, and corrosion rate [90]. Temperature and relative humidity also change chloride transport and so the predicted service life [39]. Load-induced damage matters too; bending load changes the diffusion coefficient of steel fiber-reinforced concrete [91].

10.2. Illustrative Calculation

To show how protection methods enter Equation (2), we ran a simple calculation. The values are illustrative, not measured data. We assumed Cs = 3.0% and Ccrit = 0.4% by binder mass, C0 = 0, a cover of 50 mm, and a constant D. Step 1: Ccrit/Cs = 0.4/3.0 = 0.133, so erf(z) = 0.867 and z = 1.061. Step 2: 4z² = 4 × 1.127 = 4.51. Step 3: for D = 2.0 × 10⁻¹² m²/s, ti = (0.05)² / (4.51 × 2.0 × 10⁻¹²) = 2.77 × 10⁸ s, or about 8.8 years. Table 4 and Figure 3 show the same calculation for a binder with a lower D and for that binder with a hydrophobic surface treatment, which we modeled as halving the effective D. This is a conservative choice next to the roughly 75% reduction in apparent diffusion coefficient reported by Coppola et al. [13].

The calculation shows why initiation-phase measures are so powerful. Initiation time scales with the square of the cover and inversely with D. Cutting D by a factor of four through the binder choice raises ti by a factor of four. Adding a treatment that halves D again doubles it. Real structures are less tidy. D falls with age, Cs builds up over time, and Ccrit is uncertain. Still, the example makes clear that the protection method and the design model are not separate topics.

10.3. Probabilistic and Performance-Based Design

Deterministic calculations hide uncertainty. Probabilistic design treats D, Cs, Ccrit, and cover as random variables and reports the probability that corrosion has started by a given time. Leng et al. added the maximum phenomenon, meaning that the first corrosion point along a long member governs, to a time-variant probabilistic model for marine structures [92]. Angst critically examined how well current models predict the time to corrosion initiation and argued that many of their assumptions, especially a fixed Ccrit, are weak [17]. Angst and co-authors listed present and future durability challenges, including the move toward performance-based rules [93]. Earlier research groups built linked models of corrosion that span transport, electrochemistry, and structural response [94].

For the propagation phase, models need a corrosion rate. Otieno et al. showed that rate models are still weak and that field rates vary widely [42]. The work of Sánchez Montero et al. on changing corrosion rate during carbonation-induced propagation shows that a constant rate can mislead [40]. We therefore suggest that designers treat the propagation phase as a reserve, not as a planned part of the service life, unless monitoring is in place.

11. Life Cycle Assessment and Life-Cycle Thinking

LCA of concrete often uses 1 m³ of concrete as the functional unit. This is simple, but it ignores service life. A better unit for durability studies is the emission per year of service life, which we can write as:

GWP_annual = [ GWP_initial + Σ GWP_maintenance,i ] / t_SL

(3)

where GWP_initial is the global warming potential of the materials and construction, GWP_maintenance,i is the footprint of each repair or reapplication, and t_SL is the service life. Pillai et al. used this idea for LC3 concrete and showed that ranking changes once service life is included [10]. Moro et al. applied a similar performance-based logic to chloride exposure [11]. Huang et al. reported that life cycle cost analyses from selected case studies suggest fiber-reinforced polymer bars in seawater sea-sand concrete may cut total life cycle cost by about 30% over 100 years. They also stressed that this depends on maintenance frequency, exposure, discount rate, and material prices [28].

Three points need care. First, maintenance scenarios must be realistic, including reapplication of surface treatments and inhibitors. Second, the service life in the denominator should come from a probabilistic model, and its uncertainty should flow into the LCA. Third, non-traditional assessment and maintenance methods can extend the life of aging structures, and they should be part of the comparison [74]. Angst has also warned that low-carbon binders only help if they do not shorten service life through faster corrosion [9]. We fully agree. A per-year metric with uncertainty bounds is, in our view, the fairest way to compare a green inhibitor, a surface treatment, an SCM, and a non-metallic bar on the same scale.

One more issue affects SCM-based designs. Fly ash and slag are by-products of other industries, and LCA practitioners disagree on how much of the upstream impact should be allocated to them. The choice can change the ranking of blended concretes. We suggest that durability studies report results with and without economic allocation, so that readers can see how sensitive the conclusion is. As supplies of fly ash and slag fall, emerging SCMs such as calcined clays will make this issue less important, but it has not gone away yet [82,88].

12. Predictive Modeling, Monitoring, and Multi-Scale Evaluation

12.1. Machine Learning for Chloride Transport and Corrosion

Machine learning (ML) is now widely used to predict chloride diffusion and corrosion from mix design and exposure data. Table 5 summarizes representative studies. Al Fuhaid and Alanazi compared adaptive neuro-fuzzy inference, artificial neural networks, support vector machines, and extreme learning machines for concrete with SCMs. The support vector model performed best, with R² of 0.955 in training and 0.951 in testing [19]. Hosseinzadeh et al. cleaned a dataset of 1073 experimental results. Extreme gradient boosting (XGBoost) reached R² = 0.94 and support vector regression reached R² = 0.91 for the chloride migration coefficient. They also built classification models and a web tool [20]. Liu et al. optimized several models with particle swarm optimization to predict the chloride diffusion coefficient of silica fume concrete and analyzed feature importance [21].

ML has also been used for corrosion itself. Ahmad et al. trained artificial neural networks on 72 data points with sodium chloride content (0% to 4%), inhibitor dosage (0% to 5%), and exposure time as inputs. The feedforward network reached a correlation coefficient of 0.998 in testing, but the mean absolute percentage error was about 30% [95]. That pair of numbers shows why a single fit metric is not enough. Song et al. used interpretable ML to predict maximum corrosion depth and rank influence factors [96]. Bayesian-tuned XGBoost has been applied to eco-friendly concrete with slag [89].

We see three recurring weaknesses. First, datasets are small and drawn from many labs with different test methods, so the models can learn the lab rather than the material. Second, few models respect physical limits, such as the fact that D should not rise when the water-to-binder ratio falls. Third, very few report uncertainty or test on truly new data. Physics-informed neural networks, which build Fick’s law into the loss function, and explainable methods such as SHAP are promising answers [28]. A very high R² on a random test split should not be read as proof that a model will work on a new structure.

12.2. Monitoring and Non-Destructive Testing

Monitoring closes the loop between prediction and reality. Half-cell potential mapping remains the most common field method, and RILEM guidance explains how to read it [56]. Polarization resistance measurements give the corrosion rate on site, but they need good control of the polarized area [97,98]. Electrical resistivity is a simple indicator of both transport and corrosion risk [81]. Daniyal and Akhtar reviewed these assessment methods and noted that each has limits, so a combination is best [27].

New sensors can stay inside the concrete. Embedded cement-based acoustic emission sensors have been used for both active and passive monitoring of corrosion [99]. Piezoelectric cement sensors have tracked rebar corrosion in concrete [100]. Digital twins can combine these data streams with models and update service-life estimates as new data arrive [57]. Polder et al. discussed technical and non-technical barriers to using such methods for aging structures [74]. In our view, sensor lifetime inside alkaline concrete is still the main practical limit.

Each monitoring method answers a different question. Potential mapping shows where corrosion is likely. Polarization resistance estimates how fast it is going. Resistivity shows how easily ions and current can move. Embedded sensors give a time history at a few points [56,81,97,99]. No single method gives the full picture, so data fusion is needed [27,28]. For sustainable protection, monitoring has a second role. It tells the owner whether a green inhibitor, a surface treatment, or a self-healing system is still working after years of service. This is the field evidence that the literature now lacks.

12.3. Multi-Scale Evaluation

Corrosion spans many scales. At the molecular scale, simulations explain how inhibitor films form on the passive layer [58]. At the micro scale, the steel–concrete interface controls where corrosion starts [29,31]. At the meso scale, cracks and heterogeneity create local hot spots [43,44]. At the member scale, section loss and bond loss govern capacity [6,8]. At the network scale, asset managers need probabilities and costs [74,92]. Few studies link more than two of these scales. A sustainable protection method should be judged at the scale where its benefit is claimed and at the scale where decisions are made.

13. An Integrated Framework for Sustainable Protection

Based on this review, we propose the design loop shown in Figure 4. It has seven steps. (1) Define the exposure class and the target service life. (2) Select the binder, water-to-binder ratio, and cover, using SCMs, LC3, or alkali-activated binders where suitable. (3) Add targeted protection only where the model shows it is needed, such as a surface treatment in splash zones or an inhibitor at joints. (4) Run a probabilistic service-life model, with ML surrogates where they are validated. (5) Compute the annual GWP from Equation (3), including planned maintenance. (6) Build and monitor with sensors, non-destructive testing, and, where justified, a digital twin. (7) Update the model with monitoring data and plan maintenance. If the reliability or annual GWP target is not met in step 5, the designer returns to steps 2 and 3.

The loop is not new in its parts. What we think is new is the rule that every protection choice must pass through both the service-life model and the per-year LCA before it is called sustainable. This forces green inhibitors, coatings, and self-healing agents to report the data that designers need, such as changes in D, Ccrit, or corrosion rate, and their own service life.

To make this practical, Table 7 proposes a minimum reporting checklist for papers on new protection methods. None of the items is hard to measure, but together they would let a designer place the method into Equations (2) and (3) without guessing.

14. Research Gaps and Future Directions

Table 6 lists the main gaps we found, ranked by how much they block practical use. The first three gaps are, in our view, the most urgent.

Several further directions deserve attention. Climate change will raise temperatures and change moisture and carbon dioxide levels, which will speed up both carbonation and chloride transport [28,39]. Service-life models should include climate scenarios. Seawater sea-sand concrete with non-metallic bars may help island and coastal construction, but it needs long-term bond and durability data [28,86]. Open, well-documented datasets on chloride transport, thresholds, and corrosion rates would help the ML community far more than new algorithms. Finally, collaboration between corrosion scientists, materials engineers, structural engineers, and LCA practitioners is still rare. The Special Issue that this review is written for is a good place to build that bridge.

15. Conclusions

This review examined sustainable approaches to corrosion protection in reinforced concrete and linked them to design, life cycle, and modeling tools. The main conclusions are as follows.

(1) Corrosion protection and sustainability cannot be judged apart. A method is only green if it lowers emissions per year of service life, including maintenance. Service-life-normalized LCA, as shown for LC3 concrete, should become standard.

(2) Green inhibitors from plant and waste sources show good results in concrete for several extracts, such as bamboo leaf, olive leaf, maize gluten meal, and neem. Most evidence is still short-term. Results from acid solutions should not be used to claim performance in concrete.

(3) Silane-based surface treatments have the strongest quantitative evidence among surface methods, with large cuts in chloride diffusion. Their benefit drops in cracked concrete, and reapplication must be counted in LCA.

(4) Self-healing, nanomaterials, and smart coatings are promising but are not yet ready for design. They need tests that report chloride at the steel depth and electrochemical measures, not only crack closure or strength.

(5) SCMs and LC3 give the largest and most reliable reduction in chloride diffusion per unit of carbon saved. Their lower alkaline reserve means carbonation must be checked for each exposure.

(6) The chloride threshold is the largest single uncertainty in service-life prediction. Better threshold data would improve every other part of the design loop.

(7) Machine learning models predict chloride transport well within their data. They need physics constraints, uncertainty estimates, and tests on independent data before they guide design. Monitoring and digital twins can keep these models honest over the life of a structure.

We hope the proposed design loop helps researchers report the data that designers need, and helps designers choose protection methods that make both technical and environmental sense.