Integrating Electromagnetic NDT and IoT for Enhanced Structural Health Monitoring of Corrosion in Reinforced Concrete as a Key to Sustainable Smart Cities

Abstract

The paper addresses a critical gap in early-stage corrosion detection in reinforced concrete, a leading cause of structural failures with significant impacts on humans, the economy, and the environment. It presents the M5 (Magnetic Force-Induced Vibration Evaluation) method, an innovative Structural Health Monitoring (SHM) approach that avoids damping in concrete by using electromagnetic excitation and transferring rebar vibrations through magnetic coupling over the sample. By inducing and analyzing natural vibrations directly in reinforcement, M5 enables sensitive, non-destructive evaluation (NDE) of corrosion before deterioration occurs. The study follows a systematic literature review based on PRISMA standards and utilizes EmbedSLR v1.0 free software. The methodology combines NDE with IoT deployment using Low-Power Wide Area Networks (LPWANs) and advanced machine learning (ARA) to detect frequency changes caused by corrosion, ensuring continuous monitoring. Findings suggest that M5 has the potential to enhance sustainable asset management by extending infrastructure lifespan, optimizing maintenance, and reducing waste. Its practical implications are significant for urban planners and engineers aiming to align infrastructure management with smart city strategies. The originality of this work lies in integrating electromagnetic NDT with IoT and data-driven decision-making, offering new insights at the intersection of engineering and sustainable smart city management.

1. Introduction

1.1. Motivation

Corrosion of rebar in reinforced concrete (RC) structures remains a significant and widespread issue in the construction industry, resulting in major safety hazards, substantial economic losses, and environmental concerns. Despite years of research, there is currently no method capable of directly detecting early-stage corrosion on a large scale with high sensitivity beneath typical concrete cover thickness. This early detection is crucial for enabling timely maintenance, which can prevent structural failures and reduce costly repairs. The existing detection methods are briefly summarized in Section 1.3 of this work.

According to the Federal Highway Administration and the National Association of Corrosion Engineers—International, about 30% of buildings and structures are currently damaged by corrosion. The costs associated with this issue make up roughly 15% of the total operating expenses for residential buildings. Worldwide, the problem results in annual losses of up to $2.5 trillion, equivalent to approximately 3.2% of global GDP [1,2,3,4]. Additionally, many lives are lost each year in various construction accidents caused by rebar corrosion. This issue not only endangers human safety and incurs substantial financial costs but also depletes natural resources and contributes to increased greenhouse gas emissions linked to repairs, rebuilding, and the production of new construction materials. Presently, the construction and operation of our built environment account for 39% of global greenhouse gas emissions [5].

Most of the data on the corrosion of RC structures comes from the USA. However, similar studies in other higher-income countries show a comparable level of the problem. The cost of reinforced concrete corrosion in higher-income countries typically ranges from less than one to several percent of a country’s GDP, highlighting the global scope of the issue. For example, in the USA, the estimate is around $276 billion, which is approximately 3.1% of the country’s GDP [6,7]. In Switzerland, the annual cost of just repairs to RC structures can amount to between $6.6 billion and $26.3 billion CAD [8]. In Japan, where lighter materials are used due to seismic conditions, the costs amount to roughly 1.02% of GDP, or 5258 billion yen [9].

Failure to quickly address corrosion has serious implications for public safety, as shown by numerous construction failures. A prominent example is the collapse of the bridge in Laval, Canada, which resulted in five fatalities [10]. Even more tragic cases have happened recently, including the 2018 collapse of the Genoa viaduct in Italy, which claimed 43 lives and left 566 people homeless, making it the largest disaster linked to rebar corrosion in recent history [11,12]. Such disasters occur repeatedly throughout the year. In 2024 alone, numerous incidents of this nature occurred. For instance, in Brazil, near the border between Tocantins and Maranhão, a bridge fell, killing at least four people. Many vehicles, including tankers, plunged into the river, spilling 70 tons of sulfuric acid and 22,000 L of pesticides, causing severe environmental damage [13]. These issues also affect more higher-income countries. In the same year, the Carola Bridge in Dresden also collapsed [14]. Similar corrosion problems impact reinforced concrete structures worldwide, especially in countries with seismic activity [15]. Another significant problem is concrete spalling, which in Broomfield led to the death of a man after being struck by a falling piece of concrete [16].

An additional negative consequence of rebar corrosion in reinforced concrete structures is the need for repairing and rebuilding infrastructure, which entails a significant expenditure of materials and energy. On a global scale, this means enormous resource consumption and greenhouse gas emissions, which contradict the principles of a circular economy and sustainable development. Therefore, particular emphasis should be placed on effective methods of detecting and preventing corrosion, which can significantly extend the lifespan of existing structures, thereby reducing the need for replacement and limiting the negative environmental impact of the construction industry [5].

The enormous costs associated with corrosion of reinforcement are due to the scale effect. Reinforced concrete has been the dominant construction material for over a century. Most reinforced concrete structures are designed to last 50–100 years [17,18]. However, older structures, particularly those built in the 1950s and 1960s, often lacked proper anti-corrosion protection, resulting in a greater need for regular inspections and repairs [19]. Effective monitoring of these structures is crucial not only for public safety but also to minimize the significant investments and resources needed if repairs, reconstruction, or replacement become necessary. As a result, there has been a substantial increase in interest in the effects of corrosion on reinforced concrete structures and related topics. As shown in Figure 1, the number of publications on this topic has increased by almost seven times since 1999 (dashed line indicating the current year).

Decisions about actions for aging reinforced concrete structures must be made quickly to prevent major financial losses from early facility shutdowns and to protect public safety by lowering the risk of structural failure. Extending the lifespan of these structures through timely maintenance helps reduce natural resource use and CO₂ emissions, supporting circular economy principles by cutting waste and decreasing the need for new construction materials.

Determining the best time to end operations is challenging due to the limitations of current corrosion detection methods. Most reinforced concrete structures are inspected every five years, typically in a targeted and indirect manner, primarily using the Schmidt hammer. This non-destructive technique measures carbonation levels, which indirectly indicate the potential for corrosion but do not provide direct, reliable information about the actual corrosion state of the rebar.

The Schmidt hammer test, although quick and cost-effective, has several limitations: it tests concrete at only one spot, requires smooth surfaces, is affected by moisture and the concrete’s age, and only measures surface hardness, which indirectly correlates with corrosion risk. These limitations can result in inaccurate estimates and delay the detection of critical corrosion stages, ultimately leading to untimely maintenance decisions.

Advanced detection technologies, like the M5 system, directly assess rebar corrosion beneath typical concrete cover using electromagnetic excitation and modal analysis. These tools can extend the lifespan of structures by enabling earlier and more accurate interventions. Early detection reduces the risk of catastrophic failure and costly repairs, potentially extending service life by years or even decades. By providing precise, direct detection of corrosion, M5 supports optimized maintenance planning, which significantly cuts material use and CO₂ emissions from premature replacements or extensive repairs. This aligns with the Sustainable Development Goals by improving resource efficiency and lowering environmental impact. Additionally, continuous monitoring through IoT integration allows real-time condition assessment, decreases inspection frequency and labor costs, and promotes economic efficiency and circular economy principles through better asset management and less waste. Techniques like M5 can shift corrosion management from reactive to proactive, offering notable benefits such as longer lifecycles, reduced emissions, and more sustainable infrastructure. As the scale of the problem grows and available technologies remain limited, developing non-destructive diagnostic methods to monitor reinforced concrete is increasingly important for effective urban infrastructure management. Implementing these solutions helps boost structural durability, decrease material consumption, and reduce greenhouse gas emissions from the construction industry—supporting global sustainable development and climate protection efforts.

1.2. Research Objectives and Novelty

The innovation in this paper is in seamlessly combining the M5 method with Internet of Things (IoT) technology, creating a continuous, real-time Structural Health Monitoring (SHM) system specifically for detecting corrosion in reinforced concrete. Unlike earlier studies that mainly focused on periodic inspections and simple adaptations of the M5 technique, this research transforms the method into an intelligent, interconnected monitoring solution suitable for large-scale deployment in smart city environments.

Previous studies used electromagnetic excitation to directly induce vibrations in reinforcement bars. This work proposes integrating this NDT technique with advanced data analysis through Association Rules Analysis (ARA). This combination aims to enhance early-stage corrosion detection beneath concrete cover, addressing the damping issues often associated with mechanical vibration methods.

Furthermore, implementing wireless sensor networks allows for continuous data transmission, supporting real-time monitoring and predictive maintenance. This proactive strategy greatly improves infrastructure management by lowering inspection frequency and costs while also enhancing safety and sustainability.

Additionally, the paper employs EmbedSLR software to conduct an AI-driven, systematic literature review, emphasizing the vital role of non-destructive testing methods like M5 in the construction sector. In the end, this research elevates the M5 method from a simple periodic testing tool to a comprehensive, IoT-enabled SHM platform that meets modern demands for smart and sustainable infrastructure management in urban settings.

1.3. Process of Corrosion Degradation in Reinforced Concrete Structures

Understanding the process of concrete carbonation is crucial for developing effective strategies to minimize its effects and extend the lifespan of reinforced concrete. By identifying early signs of carbonation, engineers and maintenance teams can take preventive actions, such as applying protective coatings or using carbonation-resistant materials, to improve the durability and strength of concrete structures [4,17].

Reinforced concrete deterioration typically begins with a process known as concrete carbonation. Concrete carbonation happens when carbon dioxide (CO₂) from the air penetrates the surface of the concrete and reacts chemically with calcium hydroxide (Ca(OH)₂) in the concrete structure. This reaction produces calcium carbonate (CaCO₃) and water (H₂O). Similar reactions occur with other ions and acidic substances in the atmosphere [4,21].

As carbonation progresses, the concrete’s pH decreases, weakening the alkalinity needed to protect embedded steel reinforcement bars (rebar) from corrosion, making the rebar more susceptible to corrosion when exposed to moisture and other environmental factors [17,21]. The process of loss of the rebars’ alkaline protection of concrete is presented in Figure 2.

Concrete cover, with its high pH, serves as a crucial protective barrier for reinforcement bars against harmful compounds like sulfur oxides, chloride ions, and carbon dioxide. This alkaline environment preserves the integrity of embedded steel by preventing corrosion through a passive layer [4,17]. However, over time, aggressive substances decrease the pH of the concrete, weakening this protection (Figure 2). As carbonation progresses, it affects the surface layers and penetrates deeper, stripping the protective layer from the reinforcement bars and increasing their susceptibility to corrosion. Once carbonation fully penetrates the concrete cover, water and airborne substances can react directly with the steel, causing rust on the reinforcement [17,18].

Various factors, including environmental conditions, concrete mix quality, and exposure to pollutants, can accelerate carbonation. For example, areas with high industrial emissions or urban pollution may experience higher carbonation rates due to increased CO₂ levels [18,21].

Corrosion greatly endangers the structural integrity of reinforced concrete by weakening the reinforcement, decreasing its diameter, and causing debonding, which can compromise load-bearing capacity and potentially result in catastrophic failures [4,17].

When iron oxides form during corrosion, their volume is about 10 times that of the original steel. As rust builds up, it applies pressure to the surrounding concrete, causing corrosion cracks that can spread and lead to dangerous spalling, where pieces of concrete break off. This exposes the reinforcement to environmental elements, making it easier for moisture and corrosive agents to penetrate, which accelerates degradation and creates a vicious cycle that shortens the lifespan of reinforced concrete structures [4,18].

1.4. Systematic Literature Review of NDT Methods Used in Reinforced Concrete Structure Evaluation

The systematic literature review (SLR) was conducted using the free EmbedSLR software, which helps identify articles that closely match the researcher’s natural-language description. The SLR preparation process is swift, with detailed procedural steps outlined in Section 2 of the Methodology and referenced in [22].

This review aims to quantify and critically synthesize current knowledge on NDT techniques for diagnosing early RC reinforcement corrosion. It also evaluates the potential of next-generation methods, particularly M5, and explores ways to integrate these with IoT. Following PRISMA guidelines, our systematic literature review ensures transparent and reproducible reporting [23]. The article selection process is detailed in the PRISMA flow diagram in Figure 3.

Initially, we searched the Scopus database for articles on NDT techniques used in corrosion detection, yielding 5709 records (more than sufficient for a comprehensive review). The search was conducted using the following query:

(“reinforced concrete” OR “RC structure*” OR (concrete W/2 (reinforc* OR rebar* OR “steel bar*” OR “steel reinforcement”))) AND (corrosion OR “rebar corrosion” OR “steel corrosion” OR “corrosion-induced” OR debond* OR delaminat*) AND (“non-destructive” OR nondestructive OR “non destructive” OR NDT OR “structural health monitoring” OR SHM) AND (“eddy current” OR “pulsed eddy current” OR PEC OR “remote field eddy current” OR RFEC OR “magnetic flux leakage” OR MFL OR magnetometr* OR “residual magnetization” OR “infrared thermograph*” OR IRT OR “active thermograph*” OR “microwave infrared thermography” OR MIRT OR “ground penetrating radar” OR GPR OR terahertz OR THz OR radiograph* OR “X-ray” OR “X ray” OR gamma OR “magnetic force induced vibration” OR “M5 method”)

Next, during the initial selection phase (title and abstract screening), irrelevant records were excluded, such as studies focusing solely on plain concrete without reinforcement or those related to steel corrosion but lacking the reinforced concrete (RC) context.

To improve our selection, we used the EmbedSLR method to identify the most relevant articles [22]. This search was conducted on 16 September 2025, focusing on RC structures, corrosion, NDT/SHM, and various electromagnetic and mechanical techniques, such as EC, GPR, IRT, UT, AE, radiography, and M5 (as detailed in the query).

Next, we used embedded models and algorithms from EmbedSLR for semantic selection. Each record, which includes a title and abstract, was transformed into a vector and compared to the research problem vector using cosine similarity. The problem was defined as:

“A wide range of NDT methods are used in reinforced concrete diagnostics, from electromagnetic (radiography, eddy currents, GPR, IR thermography) to mechanical (UT, AE, low-frequency modal tests) and hybrid solutions (e.g., FOCS). Yet most either assess the concrete environment (carbonation, resistivity) and only estimate corrosion risk, or detect it only at advanced stages. High-resolution techniques (radiography, EC) face safety, logistics, or depth-sensitivity limits, while UT/GPR lack sensitivity and vibration methods are strongly damped by concrete; AE, and FOCS require costly, continuous SHM. The key gap lies in non-invasive, sensitive detection of early rebar corrosion under typical cover and integration with IoT/ML. Against this backdrop, the M5 method combining electromagnetic excitation with analysis of bar eigen-vibrations emerges as a promising alternative for early, direct steel–concrete interface assessment in smart, sustainable cities”.

The description not only highlights the problem but also discusses known methods that can be used as a solution to SHM. Additionally, it provides a brief explanation of the M5 method to aid in identifying techniques similar to M5. Preliminary tests indicate that providing a more detailed description, enriched with relevant keywords for the specific topic, greatly improves the chances of obtaining representative results. Selection for synthesis: 80 publications with the smallest cosine distance (highest semantic similarity) were chosen as the corpus for review and in-depth analysis.

This innovative AI-based approach used in EmbedSLR replaces less intuitive and effective manual keyword filtering with embedding-driven semantic matching, significantly enhancing thematic coherence, bibliometric consistency, and the reliability of SLR article selection [24,25]. The following questions were applied during this process:

• RQ1: What is the current state of knowledge on NDT techniques that provide direct sensitivity to early corrosion processes under typical concrete cover thickness?
• RQ2: To what extent can these methods be integrated with SHM/IoT solutions and ML processing?
• RQ3: How does the M5 method compare to other approaches in terms of early detection and implementation feasibility?

Next, we used cosine distance to measure the similarity between the problem vector and record embeddings. Based on this ranking, we selected 80 publications with the smallest cosine distance (indicating the highest semantic similarity) as the corpus for our review and in-depth analysis.

The summary of the SLR is presented in Figure 4. The papers range from 1998 to 2026, including 80 documents, 599 authors, and 40 sources. The average number of co-authors per article is 11.9, indicating a high level of collaborative effort among teams. The absence of single-author publications and a relatively high rate of international co-authorship (18.75%) highlight the global nature of the research. The citation count, averaging 18.51 per document, suggests strong scientific interest. However, the lack of systematic growth (0% annual growth rate) may indicate either stagnation of the topic or its niche character.

Figure 5 displays the map of author collaboration, highlighting the regions most closely connected to the topic. It shows three main clusters: Chinese (blue), European (green), and Indian-French (red). Although international collaboration is visible, it remains limited, with local clusters being more established. Because of the larger font, the biggest nodes are shown only by the first letter of the author’s surname (figure generated automatically; the origin is important, not the authors’ surnames).

Figure 6 illustrates the trends in scientific publications from 1998 to 2024 related to non-destructive testing (NDT) methods in reinforced concrete diagnostics. In the first decade (1998–2010), publication numbers were low and irregular, with most years producing 0 to 2 articles. Starting in 2011, a gradual uptick in research activity was observed, with some fluctuations. A notable increase occurred after 2020, reaching a peak in 2024 with over 13 articles.

Most current and widely used methods for assessing carbonation are indirect and focus on testing carbonation rather than actual corrosion, such as using Schmidt’s hammer. However, the scientific literature on non-destructive testing (NDT) provides various approaches for diagnosing reinforced concrete. Unfortunately, so far, no NDT method can be considered fully adequate for everyday engineering practice due to different limitations. Standard techniques discussed in scholarly articles include Ground Penetrating Radar (GPR), Infrared Thermography (IRT), Acoustic Emission (AE), and Ultrasonic Testing (UT), which are major research areas in non-invasive structural evaluation. Less frequently, studies examine carbonation levels through electrical properties like resistivity testing and Half-Cell Potential (El). Other methods explored include Eddy Current Testing (ECT) and Modal Analysis (MA). High-frequency techniques, such as Terahertz (THz) imaging and radiographic methods (X-ray), face many limitations and are rarely used. Fiber-optic techniques (FOCS) and magnetic flux leakage (MFL) methods are even less commonly reported. The distribution of NDT methods among the selected articles is shown in Figure 7.

Modal analysis methods reveal significant challenges related to damping mechanical waves by concrete, which limits their application. Embedded approaches, such as fiber-optic sensors, are less commonly used because of their high cost. Additionally, studies emphasize the complementary nature of these techniques and their limitations, especially in early reinforcement corrosion detection [26,27,28].

Common environmental and electrical measurements, such as resistivity and half-cell potential, mainly evaluate the risk of corrosion rather than detecting early deterioration at the steel-concrete interface. Wave-based methods, like ultrasonic testing (UT) and ground-penetrating radar (GPR), often lack sensitivity to initial debonding and are limited by the concrete’s absorption. Furthermore, these techniques typically offer low resolution, with GPR designed to detect objects farther from the antenna [26,27,28,29].

High-resolution techniques, such as X-ray and eddy current methods, face operational limitations. Radiography is limited by the use of ionizing radiation and measurement logistics, while eddy currents have a limited penetration depth. The same issue applies to IRT, as the sensitivity of various thermography versions is too low to detect corrosion [30,31,32].

Vibroacoustic methods and embedded sensors are sensitive but typically require dense sensor networks or long-term monitoring, which increases the cost of large-scale SHM deployment. Experiences with AE show its potential for degradation detection but also its dependence on long-term observations and event processing [33,34,35].

Within GPR research, two parallel streams are observed: matrix imaging for early corrosion detection [36] and temperature compensation, which reduces false alarms and stabilizes long-term trends [37]. Field applications, including diagnostics of prestressed bridges, demonstrate GPR’s practical scalability but also highlight the need to combine it with other methods for complete interpretation [38]. Comparisons on RC slabs show that GPR/IR alone rarely detects debonding initiation without support from other channels [39].

Within the lower frequency electromagnetic spectrum, there is a transition from conventional ECT to more sophisticated imaging methods. Feature learning of pulsed ECT signals enhances the ability to differentiate between corrosion severities [40], while emerging techniques, such as magnetic resonance eddy current penetrating imaging, are promising for visualizing reinforcement corrosion beneath concrete cover [41]. At the same time, comparative reviews of electromagnetic methods highlight the trade-off between penetration depth and the need for calibration, considering moisture and chlorides [42].

The map shown in Figure 8 demonstrates the relationships between articles through shared cited sources.

The nodes shown in Figure 8 reveal clusters of authors and studies that share similar theoretical bases, indicating a common research approach. The largest nodes [33,43] represent the most frequently co-cited publications, serving as key references in the field. These nodes are related to the AE method, highlighting its significant potential as an SHM technique. However, the effectiveness of Acoustic Emission in detecting corrosion in rebars may be limited, as the method is designed to capture active damage events, such as cracking, and might miss early or dormant corrosion stages that do not produce acoustic signals. Additionally, acoustic waves attenuate within concrete, reducing detection range and signal clarity. Environmental noise can mask weak signals, and accurately interpreting them requires expertise. Locating emission sources and distinguishing corrosion-related activity in reinforced concrete structures remains a challenging task, which limits the reliability of these structures.

Sensor hybridization consistently enhances RC diagnostics, showing great potential for the M5 method. Combining vibration and AE detects corrosion-related changes early during progressive beam loading, offering clear signs of steel–concrete interface deterioration [50]. The idea of merging multiple modalities for lifecycle management is supported by reviews and case studies of “multimodal monitoring” [51], as well as practical multiphysics scanning of bridge decks using thermography and related techniques [52].

From an implementation perspective, there is an increasing emphasis on automation and IoT, with semi-contactless, fully automated SHM stations standardizing data collection and enabling consistent defect detection [52]. The integration of embedded sensors with multivariate analysis provides a framework for assessing corrosion in RC [53], while wireless AE networks with deep learning (WAESN + GADF AlexNet MHA) demonstrate the readiness of edge models for corrosion event classification [4]. Bridge studies utilizing NDT packages connect this analytical layer with practical maintenance strategies [54].

From the perspective of smart cities, IoT/ML integrations are just beginning to expand. Wireless sensor networks and deep learning methods for corrosion signal detection are emerging but are still isolated and small-scale for now [4,28].

The M5 method, which uses electromagnetic excitation of the bar and examines its eigen-vibrations, stands out for its ability to quickly and directly assess the steel–concrete interface’s integrity. It can detect issues such as frequency shifts and increased damping caused by debonding or corrosion, all without disturbing the cover. Additionally, it can be combined with methods like AE and used to develop effective SHM systems within IoT nodes, as noted in reference [54].

The SLR indicates that few methods are effective for corrosion detection outside labs and on a large scale. Currently, only AE and M5 are potential options for SHM systems. However, AE, which depends on monitoring mechanical waves in concrete, may lack sufficient sensitivity and noise resistance in real-world environments. This presents significant opportunities to develop M5 for SHM applications further.

2. Materials and Methods

2.1. Magnetic Force-Induced Vibration Evaluation (M5) Measurement System

The resonance frequencies of each object are unique, much like fingerprints and iris patterns are distinctive to individual humans. Defects, degradation, and structural changes influence these frequencies. Usually, specific issues affect particular frequencies in unique ways. This paper focuses on detecting debonding caused by rust, which results from the corrosion of rebars in reinforced concrete (RC) structures. However, analyzing natural vibrations can also help identify other types of faults. As a result, the M5 system could be adapted for comprehensive testing of reinforced concrete in the future. The proposed version focuses on detecting corrosion, which is a major problem. It can be implemented in two setups: an active version for periodic inspections of RC structures, and a passive version suitable for structural health monitoring (SHM), which can also be used for active testing when necessary.

2.1.1. M5 System

Active M5 is an advanced modal analysis technique that enables direct evaluation of rebars in RC structures. Its main advantage is its ability to avoid concrete damping and detect corrosion in any RC structure without needing prior knowledge of its dynamic or frequency characteristics. Therefore, this method can be used for the required periodic inspections of any RC structure, as mandated by law in most countries. The M5 system can be implemented in several different ways. The method is explained in detail in [4,54]. The block diagram of the system is presented in Figure 9.

To induce vibrations in rebar, the sweep frequency technique is used. A control computer creates a signal that is sent to the excitation subsystem. This digital data is converted into an analog signal by a D/A converter, amplified, and then transmitted to an electromagnet. The resulting electromagnetic wave, characterized by its sweeping frequency, directly interacts with the rebar, penetrating the concrete cover and producing vibrations.

Induced vibrations are relatively weak, so to minimize damping, they are transmitted through magnetic coupling to a magnet placed above the sample. The magnet’s vibrations are then detected by an accelerometer, which converts the signals via an A/D converter and sends them back to the control computer.

Technically, the main advantages of the M5 method stem from its ability to directly excite vibrations in the rebars, thereby effectively minimizing the damping effects of the surrounding concrete cover. The damping effect is especially crucial in mechanical methods, making it nearly impossible to inspect rebars just a few centimeters beneath the surface (typical concrete cover thickness ranging from 2 to 5 cm). In contrast to mechanical waves, electromagnetic waves completely bypass concrete’s damping effects, as the magnetic permeability of concrete is nearly the same as that of air [4,54].

This direct excitation not only improves measurement accuracy but also provides clearer insights into the structural integrity of the concrete. Furthermore, in the M5 technique, both the excitation signal and the pick-up signal are transmitted through the concrete via magnetic waves. In this case, coupling with a magnet placed above the surface is used. This approach significantly enhances the quality and repeatability of the results [4,54].

The M5 process starts with modeling and generating an excitation signal in the control computer. This signal is then sent to the excitation subsystem, which can be set up in different ways, such as using an electromagnet or a rotating wheel with permanent magnets [4,54].

In the electromagnet setup, the sinusoidal excitation signal with a sweep frequency causes a current to flow through the coils of the electromagnet. This current produces an alternating magnetic field with a sweeping frequency. On the other hand, the wheel setup adjusts the spinning frequency of the wheel with magnets, creating a quasi-sinusoidal magnetic field with a sweeping frequency. Detailed descriptions of both setups can be found in [4,54].

The generated alternating magnetic field penetrates the tested reinforced concrete (RC) sample, causing the rebar beneath the excitation subsystem to interact with the magnetic field, which results in vibrations of the rebar [4].

The vibrations of the rebar are transmitted to the magnet positioned directly above it through magnetic coupling, significantly enhancing both the sensitivity and repeatability of the measurements. This configuration effectively mitigates the effects of concrete damping, allowing for the detection of even subtle vibrations. Additionally, the use of magnetic coupling eliminates the need for direct contact with the sample, simplifying the testing procedure and opening up new possibilities for assessment [4].

The neodymium magnet is securely attached to a highly sensitive seismic accelerometer, which is powered by a conditioning system. The accelerometer detects vibrations and converts them into an electrical signal, which is then sent through an analog-to-digital (A/D) converter to the control computer for analysis [4].

2.1.2. Modal Analysis in the Evaluation of Reinforced Concrete Structures

Modal analysis is a technique used to study the dynamic behavior of RC structures. It identifies natural frequencies, mode shapes, and damping characteristics, which describe how structures vibrate in response to external forces. The pattern of vibrations for each complex structure is unique, like a fingerprint for each human. Each characteristic frequency or set of frequencies can be associated with a specific deformation pattern. Modal analysis reveals these frequencies and shapes, enabling the prediction of dynamic responses. This analysis can be performed experimentally by applying a known force, such as a pulse (e.g., using an impact hammer), or by analyzing the response to a known waveform (e.g., sweep frequencies) with sensors like accelerometers or laser vibrometers.

Currently, modal analysis allows for the identification of various changes in concrete, such as detecting cracks and damages [55,56], which decrease the stiffness of the RC structure and lower its natural frequencies. The method can also be used to identify general stiffness changes [57] or those related to load-carrying capacity [58]. Additionally, it can detect changes related to temperature variations [59]. However, because concrete strongly damps mechanical waves, modal analysis has not been very effective in testing reinforced mesh. Previous methods only worked if the concrete cover was thin. The M5 method was developed to address this problem and fill this critical research gap [4].

2.1.3. Internet of Things as a Crucial Part of Structural Health Monitoring Systems

Recent advancements in low-power radio frequency (RF) chip transceiver technology and related structural health monitoring (SHM) platforms have significantly enhanced the capability for high-rate, lossless transmission of measurement data across extensive sensor networks. These advanced features enable high-quality, rapid operational modal analysis of in-service structures by utilizing distributed accelerometers to experimentally characterize their dynamic responses. The insights gained from these dynamic data sets allow for the application of structural identification techniques, facilitating a comprehensive analytical evaluation of reinforced concrete RC structures [60]. The fundamental concept of the combination of the Internet of Things (IoT) with SHM is illustrated in Figure 10.

The sensors used in the system, as shown in Figure 10, are integrated with the tested object and linked to the structural integrity monitoring function of that object. This monitoring is tied to a specific type of physical phenomenon closely related to damage. Based on the changing condition of the tested object, reflected in the measurements, the sensor produces a signal and sends it to the acquisition and storage subsystem [61].

The system’s gateway connects to both the sensor and the network. It converts sensor data into data packets for transmission to a remote control and service room. The gateway manages and optimizes data requests, event notifications, checks node connectivity, and performs system integrity tests using IoT protocols. Additionally, it should include an embedded local database to store large amounts of data, which helps resolve remote connection issues and prevents data loss [61].

The remote control and service room (RCSR) allows on-demand queries to specific sensor nodes to check their status and management parameters, such as battery percentage and estimated transmission latency. It also stores all collected data in a database for big data analysis and connects to an open platform communications (OPC) server for interoperability with standard industrial systems. As the final component of the system, the RCSR is responsible for storing all data gathered from the monitoring segments. Access to the RCSR and OPC is available via computers or mobile devices [61].

2.1.4. M5 Structural Health Monitoring

For an SHM system using M5 technology, the sensor network should be placed near the rebars but not directly on them. The overall concept of the system is shown in Figure 11.

Embedded accelerometers measure rebar vibrations and transmit the data through a gateway to the network, then to the cloud. A control computer manages the data in the cloud.

This setup enables sensors to monitor both the vibrations of the surrounding concrete and the rebar, allowing comprehensive structural monitoring. Ideally, the SHM system should be incorporated during the structural design phase. Adding it to existing buildings can be expensive because it requires drilling at multiple locations for installation. It can also lead to unexpected issues at the connection between the original concrete and the new concrete with the sensors. Such installations distinguish buildings in smart cities from modern structures. The structural health monitoring (SHM) network of accelerometers provides thorough monitoring of the entire structure’s condition. Different patterns of changes in characteristic frequencies are observed for various defects.

Sensors used in the grid can be significantly less sensitive (and therefore more cost-effective) than seismic elements used for periodic inspections, as they are strategically placed near the rebar (lower damping) and provide a stronger connection to the structure’s concrete (lower noise).

The grid of sensors should be installed during the construction phase of the structure. Placing sensors just 5–10 mm from the rebars makes the system highly sensitive to vibrations in the rebars. Still, it can also make it less responsive to changes in the concrete itself. Earlier research suggests that if sensors are positioned more than 20–30 mm from the rebar (depending on the sensor type), detecting minor corrosion changes may be impossible [4,54].

Structural health monitoring (SHM) systems can detect corrosion much more accurately than traditional periodic inspection methods. This improved accuracy results from the well-known frequency response of the structure before any corrosion, allowing for the detection of even minor changes (deviations from known characteristics).

Previous studies [4] have shown that corrosion significantly lowers resonant frequencies within a specific range. These changes can be detected through long-term observation and quantitative analysis of the structure’s vibrations. A consistent decrease in vibration amplitude at these frequencies may (but does not necessarily) indicate the start of corrosion (environmental factors could also cause the change). In such cases, it is recommended to speed up periodic inspections. Continuous monitoring can not only allow early detection of corrosion but also provide warnings before the structure reaches a critical condition (for any reason), which is indicated by major changes in frequency responses, thus preventing structural failure or construction disasters.

Integrating sensors into the structure allows for faster inspections (eliminating the need to install sensors before each measurement), increased sensitivity (since the frequency response of a healthy structure is known and sensors keep better contact), improved accuracy, and greater reliability (measurements can be performed more quickly and easily at multiple points, providing dependable results). Additionally, with built-in sensors, tests can be performed in two stages: first, mechanical excitation to test the concrete, and second, electromagnetic excitation to test the reinforcement and potential delamination from the concrete. This approach ensures a comprehensive investigation and helps protect the structure against damage and potential construction failures.

2.1.5. Comparison Between Structural Health Monitoring and Periodic Inspection Versions of the M5 System

The comparison between the Structural Health Monitoring and periodic inspection versions of the M5 system is shown in

Table 1. Comparison between Structural Health Monitoring and periodic inspection versions of the M5 system.

	SHM Version	Periodic Inspection Version
Purpose	Continuous and automated monitoring of structural health + facilitating periodic inspections	Scheduled, periodic inspections of reinforced concrete
Implementation	Embedded sensor network near rebars (5–10 mm)	Sensors or devices used temporarily during inspection
Data Collection	Real-time, continuous acquisition	Intermittent data collection during inspection visits
Sensitivity	High sensitivity due to close placement and monitoring	Effective but limited by inspection frequency and coverage
Cost and Installation	Higher upfront cost and complexity, planned installation	Lower immediate cost, but repeated regularly over time
Data Analysis	Advanced data integration—usually carried out locally on microcontrollers, in the cloud, or at dedicated processing centers	Semi-automated data analysis after data collection
Detection Capability	Early detection of corrosion and structural changes	Detect corrosion or damage present at inspection times
Integration	IoT-enabled, with remote data transmission and edge processing	Standalone system requiring manual data handling
Maintenance	Enables predictive, condition-based maintenance	Supports preventive maintenance triggered by inspections
Operational Impact	Minimal disruption; sensors continuously deployed	Temporary disruption during inspection activities

.

2.2. Samples

Researchers identified frequencies that change in distinct patterns during rebar corrosion by analyzing two significantly different types of samples. The first group consisted of three standardized concrete cubes, each with a single ribbed, hard steel rebar (Class A–III, Type B500 SP, 20 mm diameter) positioned symmetrically for a 20 mm concrete cover. Two of these samples had wax-coated rebar: sample C01 had half of its rebar covered, while sample C02 had full insulation. In contrast, the reference sample C00 remained unchanged. The wax coating was meant to simulate corrosion and potential bond loss between the rebar and concrete. A more detailed description of these samples is available in references [4,54].

Another category consisted of samples taken from a disassembled bridge slab in Świerkocin on the Warta River, Poland. In this study, three distinct groups of specimens were identified: the first included samples with corroded rebar (C11), which showed significant corrosion; the second consisted of samples with non-corroded rebar (C10), where the rebar remained intact and free of corrosion; and the third featured debonded rebar (C12), with rebar that experienced severe corrosion and lost adhesion to the concrete [4,22]. The rebars had a diameter of 10 mm, half the size of the laboratory sample. The steel alloy used was also different, being highly elastic (class A–I), non-ribbed soft steel. This combination of low hardness and high elasticity results in a lower yield strength. The sample categories vary significantly.

Table 2. Samples used in the experiments.

	Laboratory Sample	Field Sample
No corrosion	C00	C10
Partially corroded	C01	C11
Fully corroded	C02	C12
Rebar diameter	Large (D = 20 mm)	Small (D = 10 mm)
Steel yield strength	Very high (class A-III)	Very low (class A-I)

provides an overview of all the samples.

Additionally, samples C10 were split into two categories: C10_L, with a low concrete cover thickness (20–30 mm), and C10H, with a high concrete cover thickness (30–50 mm).

Since the magnetic permeability of concrete is approximately the same as that of air, and steel’s permeability is about 1000 times higher, the concrete parameters can usually be disregarded. However, a more detailed description of the samples and system, including the concrete composition, is provided in [4].

2.3. Methodology

2.3.1. Methodology of Systematic Literature Review

The new AI-based systematic literature review (SLR) method, EmbedSLR, was used for this review. This method allows users to select publications that best meet their requirements using natural language descriptions, making the SLR process much easier and more convenient [22]. The software is available in [22].

The process begins with the user querying the Scopus database to identify articles that may meet their needs. It’s important that the query remains broad enough to prevent missing important publications; in this case, we found 5709 articles related to NDT and corrosion. The user then provides a natural language description of their problem, and the software works similarly to modern generative AI. This description is compared against the combined abstract and title of each article. Using cosine distance, the software ranks the articles based on their vector representations, highlighting the most relevant ones. It supports both free and paid embedding models [22].

Next, the user decides how many of the top-ranked records they wish to analyze. They can make informed decisions by examining the ranking and the cosine similarity matrix calculated for each article, often revealing a clear distinction between well-fitting articles and those that are not. The analysis incorporates eleven complementary bibliometric indicators. By integrating embedding models with bibliometric metrics, systematic reviews can be conducted more objectively, accurately, and efficiently through automation while also minimizing reliance on team evaluations, as various models can effectively substitute for human judgment. The software allows for the rapid selection of the most relevant papers based on the provided description and is user-friendly [22].

A systematic literature review (SLR) is essential before conducting empirical research, as it enables scholars to assess existing knowledge, identify gaps, and inform future studies with robust theories and methods. It synthesizes evidence transparently, guiding robust strategies and emphasizing innovative approaches [62,63,64].

An SLR reduces selection bias, ensures transparent decision making, and synthesizes evidence to highlight established knowledge and research gaps. It provides empirical studies with a strong theoretical basis, clear definitions, and suitable methods for the research problem [65,66,67,68].

Beyond semantic screening and corpus selection, we conducted advanced bibliometric and text-mining analyses. Using VOSviewer 1.0 (1.8.0_431), we visualized bibliometric coupling, co-citation, and collaboration networks, which helped identify the intellectual foundations, main publication clusters, and patterns of international cooperation. Additional analyses using the Bibliometrix 4.3.0 module in R-Studio 4.4.2 provided quantitative metrics, including annual growth rate, international collaboration index, most cited sources and authors, and keyword co-occurrence data, thereby strengthening the bibliometric profile.

During the preparation of the SLR, the all-distilroberta-v1 embedding model was used, and most graphics were generated with VOSviewer.

2.3.2. Feature Extraction from Frequency Characteristic

Methods for extracting features from signals to identify association rules between ongoing structural changes and their corresponding frequency changes differ from those used for identification. During the identification process, the goal is to strike a balance between a detailed description of the characteristic and the fewest possible descriptive features. Minimizing these features helps prevent the curse of dimensionality and removes highly correlated features that don’t add useful information. The extraction methods are discussed in [69].

For association rule extraction, the only selection criterion in feature definition is to represent the signal’s shape accurately. In [70], four methods for feature extraction were explicitly proposed for determining association rules:

• Feature extraction through equal division in the domain of the independent variable.
• Feature extraction through equal division in the domain of the amplitude.
• Feature extraction through equal division with normalization.
• ACO decomposition.

One of the simplest techniques is to divide the domain of the independent variable into equal parts to extract features. This approach typically works well for analyzing the basic frequency characteristics and frequency spectrum of signals. The other techniques are more advanced and designed to be effective with time signals, waveforms, and more complex features [69].

For SHM systems where the object’s characteristics are known and the goal is just to detect changes, the most straightforward approach (feature extraction through equal division in the domain of the independent variable) is highly effective. In this method, the amplitude of each frequency is treated as a separate attribute, and the data aren’t processed any further [69,70].

2.3.3. Association Rules Analysis (ARA)

To connect corrosion with variations in frequency characteristics, the association rule analysis (ARA) technique is employed. It is based on the classic Apriori algorithm, originally designed to discover non-trivial sales patterns, such as how often a customer who buys X also tends to buy Y. This analysis helps sellers position products X and Y far apart to increase the likelihood of a buyer purchasing different items [71]. The same approach can be applied for quantitative analysis of signals, indicating that if corrosion increases, a specific feature of the characteristic changes more frequently. However, unlike the original Apriori, ARA is explicitly adapted for signal analysis and has several significant differences from the original Apriori. In classic algorithm rules are presented in the form of (1):

If (BODY) Then (HEAD) [support, confidence]

(1)

With Apriori, the same element (product) is considered both as BODY (A) and HEAD (B). For ARA, BODY indicates the progress of corrosion in rebars within an RC structure, while HEAD (B) shows the change in a specific frequency within the frequency characteristic. The traditional Apriori algorithm assesses a rule’s quality using two percentage metrics: support and confidence. Support (supp) is the ratio of records where the rule appears to the total number of records in the database (D), as defined by (2):

$$\mathrm{supp} (\mathrm{A})={\displaystyle \frac{\#\left\{{\mathrm{d}}_{\mathrm{i}}\in \mathrm{D}:\mathrm{A}\subset {\mathrm{d}}_{\mathrm{i}}\right\}}{\#\mathrm{D}}}\approx \mathrm{P}(\mathrm{A})$$

(2)

For SHM, where all elements are well understood, the database is specifically designed to monitor the onset and progression of corrosion, ensuring continuous support remains at 100%. As a result, this parameter isn’t used in the ARA technique.

Creating the database involves several steps. First, signals from non-corroded samples (C00, C10) are compared with those from partially corroded samples (C01, C11). Additionally, partially corroded samples are compared with fully corroded ones (C02, C12) to demonstrate how corrosion progresses. We test all possible combinations. Then, depending on the magnitude of the change, we describe changes in specific parameters as an increase (↑), a decrease (↓), or no change (-).

ARA introduces a new parameter, sensitivity, for extracting association rules. This parameter serves as a threshold, determining whether a specified change should be recognized as an increase (↑), decrease (↓), or no change (-). Any changes smaller than the sensitivity are marked as no change (-).

Confidence is the key parameter. It represents the likelihood that the entire rule will occur, assuming the BODY (A) is present. This can be mathematically expressed as follows (3):

$$\mathrm{conf}\left(\mathrm{A}\to \mathrm{B}\right)={\displaystyle \frac{\mathrm{supp} (\mathrm{A}\cup \mathrm{B})}{\mathrm{supp} (\mathrm{A})}}\approx \mathrm{P}(\mathrm{B}|\mathrm{A})$$

(3)

The ARA method provides a comprehensive evaluation by integrating both quantitative and qualitative assessments through a sensitivity parameter analysis. The rules governing the ARA method are outlined in Equation (4), where the confidence interval is responsible for the quantitative evaluation, while the sensitivity parameter contributes to the qualitative assessment.

If (BODY) Then (HEAD) [confidence, sensitivity]

(4)

Both a detailed explanation of the algorithm and an example of its use can be found in [72,73,74]. The first example describes a pilot study on identifying corrosion in RC structures, while the second example outlines a survey of using NDT to determine basic RC structure parameters.

3. Results

The research involved several studies focused on identifying association rules related explicitly to corrosion. These patterns were derived from analyzing the frequency characteristics of samples C00 and C01, as well as C10 and C11, which indicate the onset of the corrosion process. Additionally, the study examined patterns associated with the progression of corrosion by comparing samples C01 and C02, as well as C11 and C12.

The investigation was conducted in several stages, starting with theoretical studies. Theoretically, low-frequency mechanical waves propagate more effectively because of lower damping, whereas high-frequency waves experience stronger damping. To confirm this, a hammer impact test was performed. The results are presented in Figure 12.

The hammer impact test is not highly precise and requires access to the rebar to be effective. However, the obtained results clearly demonstrate that higher frequencies are significantly damped in the case of a debonded sample. Additionally, it was found that the main resonance frequencies ranged from 10 to 200 Hz, a result consistent across all sample parameters.

Next, different versions of the M5 system were tested, all yielding similar results. The analysis primarily relied on attributes derived from the ACO decomposition [30]. This method allowed a quantitative evaluation of how often corrosion in the database affects the Amplitude (A), Correlation (C), or Offset (O). Previous studies have demonstrated that these three features usually suffice to describe the signal [73,74] accurately. Samples were tested within the 20–220 Hz range, where strong resonance frequencies were observed. The quantitative analysis did not show a clear effect of corrosion on the signal offset. The impact on amplitude was only apparent when the corroded sample differed from the non-corroded one in other parameters.

Experiments showed a weaker correlation between the frequency characteristics of the two types of corroded samples (C01 and C11) and the two types exhibiting significant debonding (C02 and C12). In contrast, a stronger correlation was observed among samples that only differ in their corrosion levels (C00, C01, and C02). This difference occurs because corroded or debonded samples vary significantly in other parameters, such as size, concrete cover thickness, rebar class, and diameter. In comparison, all laboratory samples have the same parameters except for corrosion level. The average correlation across different sample types is summarized in

Table 3. Correlation matrix of all samples used in the experiments (in percent).

	C00	C01	C02	C10L	C10H	C11	C12
C00	100	85	74	69	21	19	16
C01	85	100	92	22	16	11	7
C02	74	92	100	28	13	1	6
C10L	69	22	28	100	60	39	38
C10H	21	16	13	60	100	5	5
C11	19	11	1	39	5	100	89
C12	16	7	6	38	5	89	100

.

Table 3 illustrates that significant corrosion progression can result in changes in the correlation level of approximately 10–15% among the laboratory samples C00, C01, and C02. In contrast, much larger variations are observed due to differences in concrete cover thickness, particularly in samples C10_L and C10_H. Figure 13 presents the differences in frequency characteristics between samples with low and high concrete cover thickness. It also highlights variations in attributes within the same category for samples that exhibit slight differences in concrete cover thickness and size. Furthermore, Table 3 indicates that there is virtually no correlation between the corroded samples (C01 and C11) and between the debonded samples (C02 and C12).

Figure 13 illustrates that even small variations in concrete cover thickness can greatly affect the correlation and consistently influence the amplitude. Numerous parameters can impact the frequency characteristics, including factors such as sample size, the diameter of the reinforcing bars, and the type of steel used in their production. This highlights the importance of precisely measuring the concrete cover thickness of the tested sample, underscoring a key advantage of Structural Health Monitoring (SHM) systems over periodic inspection methods. Additionally,

Table 4. Correlation matrix of samples from the same category but different parameters.

	C10H1	C10H2	C10H3	C10H4	C10H5	C10H6
C10H1	100	67	65	42	44	46
C10H2	67	100	97	78	68	63
C10H3	65	97	100	85	75	65
C10H4	42	78	85	100	74	55
C10H5	44	68	75	74	100	68
C10H6	46	63	65	55	68	100

reveals notable differences among samples within the same category, C10H. However, specific defects generally affect the characteristic frequency in a particular way.

The results presented so far indicate that detecting corrosion in unknown samples can be difficult, if not impossible, due to factors that may influence the results more than the corrosion itself. These findings highlight the vital role of SHM in quickly and effectively detecting corrosion at its early stages.

For further analysis, feature extraction is performed through equal division in the domain of the independent variable. This method allows focusing on smaller parts of the frequency characteristic than ACO decomposition.

To identify patterns specific to corrosion progression, association rule analysis (ARA) was developed and applied. Some examples of the analyzed signals are shown in Figure 14.

The analysis steps are thoroughly outlined in references [53,72]. In [72], RAW signals were examined. This version of the ARA method is especially effective at identifying groups of characteristic frequencies associated with corrosion. Conversely, the analysis in [53] uses the derivative of the characteristic, making it more suitable for evaluating individual frequencies. Both studies show that the most important frequency range, where the rules are most evident, is between 130 and 146 Hz. The confidence levels of the rules are shown in Figure 15.

The general rule for comparing frequency characteristics across the 132 to 146 Hz range, derived from a sensitivity level of 10% of the maximum magnitude of a less corroded sample, can be summarized as follows (5). The 10% threshold was selected experimentally to ensure a high level of confidence. This sensitivity parameter provides a qualitative analysis showing how significant changes in the specific feature corrosion can cause, for example, transitioning from class C00 to class C01. The value of sensitivity is consequently increased until confidence does not decrease. The highest value where confidence remains unchanged is a parameter that qualitatively describes rules.

If (corrosion ↑) Then (frequencies from 132 Hz to 146 Hz ↓)
[confidence = 100%, sensitivity = 10%]

(5)

Rule (5) can be interpreted as: ‘If corrosion increases, then features corresponding to frequencies between 132 and 146 Hz decrease.’ All tested cases validate this rule, and it applies to changes exceeding 10% of the amplitude (qualitative analysis). Figure 16 displays the rules derived for identifying early corrosion stages (samples C00, C01, and C11). Examples of selected ARA feature values are also shown in Figure 16.

Changes in frequency characteristics, as shown in Figure 16, are more pronounced in the early stages of corrosion than in advanced stages. This applies to all sample parameters and is observed in both laboratory and field samples, with a greater difference noted in field samples due to the thinner concrete cover.

The ARA method can verify complex rules and reveal an interesting relationship: with the algorithm’s sensitivity set to 10%, significant differences in corrosion effects on amplitude were noted between frequencies above and below 120 Hz. Higher frequencies showed a 100% confidence coefficient for corrosion damping amplitude, particularly between 140 and 160 Hz, while lower frequencies had less than 65% confidence. Furthermore, for corroded samples, the amplitude of signals in the 20–119 Hz range consistently exceeded that of the 120–220 Hz range, even at a 40% sensitivity, indicating that corrosion can be detected based on these findings.

4. Discussion

The SLR mapped the knowledge state and showed limited support in the literature for early, direct reinforcement corrosion detection. This justifies the development and validation of the M5 method, as well as the design of comparative studies. Bibliometric analysis identified thematic clusters and key publications, aligning with good SLR practices [67,69].

Linking SLR findings with IoT and Smart City highlights early corrosion diagnostics as key to resilient urban infrastructure. Integrating NDT with IoT and machine learning can enable intelligent monitoring and efficient asset management in future smart cities.

As highlighted in the motivation, assessing corrosion in reinforced concrete (RC) structures is a critical and urgent task. Currently, there is no non-destructive testing (NDT) method that allows for easy, direct, and rapid evaluation of corrosion in RC structures. The M5 method is the first method to demonstrate significant potential for widespread use in this area, supporting sustainable infrastructure management by reducing the need for destructive inspections and extending the lifespan of concrete structures.

Utilizing the M5 method for structural health monitoring (SHM) can significantly enhance its effectiveness. When combined with electromagnetic excitation, this method proves highly effective in assessing corrosion in the rebars of reinforced concrete (RC) structures, while mechanical excitation aids in evaluating the concrete itself. Furthermore, the M5 method can be employed passively to detect notable changes in frequency characteristics, which may indicate potential structural issues and highlight the need for ongoing inspection. This proactive monitoring enables timely maintenance, reducing waste and the environmental impact associated with large-scale repairs or premature replacement.

By integrating SHM with the Internet of Things (IoT), the M5 method can establish a foundational technique for smart buildings and play a vital role in the advancement of smart cities.

In this study, we introduced a novel approach for identifying valuable patterns through Association Rules Analysis (ARA). This method enables both quantitative and qualitative evaluations of association rules, thereby enhancing the detection process. It is particularly well-suited for non-destructive testing (NDT) investigations, effectively uncovering all relevant patterns that can aid in detecting corrosion. The utilization of multiple patterns is crucial, as it increases the reliability of the results, especially in scenarios where the sample size is limited, which is a common challenge in corrosion testing of reinforced concrete.

The association rules obtained at a 10% sensitivity have confidence levels of nearly 100% in all cases, indicating that these patterns are consistently observed across all tests. As sensitivity exceeds 10%, confidence levels start to decline. Moreover, differences between the characteristics of non-corroded samples and those of partially corroded samples are much larger than the differences between partially corroded and fully debonded samples (Figure 8). This indicates that the method may be more effective in the early stages of corrosion than in the later stages, which is a positive outcome. Early detection of corrosion supports sustainable infrastructure management by allowing targeted interventions before severe damage occurs.

Conversely, Figure 6 demonstrates that the thickness of the concrete cover has a significant effect on the energy of the resulting characteristics. This finding suggests that Structural Health Monitoring (SHM) systems, which compare the frequency characteristics of a healthy structure to its current state, may offer greater effectiveness than traditional periodic inspections. Moreover, it underscores the importance of analyzing patterns beyond merely the amplitude of specific frequencies. Notably, the ability to detect early signs of corrosion by comparing the maximum magnitudes of frequency characteristics both below and above 120 Hz is particularly noteworthy. Additionally, in SHM systems, the energy of the signal can play a vital role in corrosion identification. This continuous monitoring reduces unnecessary inspection efforts and resource usage, enhancing sustainable asset management.

In the experiments, two distinct types of samples were utilized: irregular field samples with thin elastic rebars and fully normalized laboratory samples featuring thick rebars made from hard, non-elastic steel. Initial findings indicated that the varying parameters of these two groups had a significantly greater impact on the results than corrosion itself. However, by employing ARA, we were able to discern patterns in how corrosion affects the resonance frequencies in both sample groups. Furthermore, the results were validated through an alternative method—the hammer impact test (HIT).

Although the research was conducted on a relatively small number of samples, the findings can still be considered reliable. Further investigations with more samples are warranted. Developing reliable and efficient non-destructive evaluation methods is essential for promoting sustainable maintenance practices and reducing the environmental footprint associated with infrastructure degradation. While the detailed numerical quantification of early corrosion detection and its impact on extending the lifespan of structures varies by context, early detection technologies consistently align with the principles of a circular economy and climate protection. They achieve this by extending asset lifetimes, optimizing resource use, and lowering greenhouse gas emissions related to repairs and rebuilds [75,76].

5. Conclusions

The SLR identified a research gap in sensitive, non-invasive early corrosion diagnostics beneath concrete. It found the M5 method promising for NDT/SHM, especially in Smart City contexts. Miniaturized M5 systems integrated with IoT can enable early risk warnings and infrastructure maintenance. These findings support the further development of intelligent monitoring systems for enhancing city safety, sustainability, and digital transformation.

The results presented indicate that the M5 method has the potential to become a foundational technique for Structural Health Monitoring (SHM) of buildings in smart cities. The findings confirm established engineering knowledge regarding the differential impacts of various defects on resonance frequencies. While modal analysis methods have proven their ability to detect defects in concrete (a well-studied area), M5 also marks a significant step forward in identifying corrosion of rebars, demonstrating high versatility. By enabling early and accurate detection of structural deterioration, the M5 method supports sustainable infrastructure management by minimizing resource-intensive repairs and extending the life cycle of buildings.

However, even if promising, these results are preliminary, and further laboratory experiments are necessary, followed by evaluations in real-world constructions. Such comprehensive validation is essential to ensure that the method can contribute effectively to the sustainable maintenance of existing structures. Additionally, integrating SHM with the Internet of Things (IoT) remains a critical area for future development. This integration will drive the development of more innovative and sustainable urban environments, playing a vital role in the ongoing evolution of resilient smart cities. There is still significant work to be done to fully realize this potential. Several challenges must be addressed during the integration process, with the following key issues being fundamental:

• Communication: Ensuring robust, low-latency communication for the M5’s sensor network is critical. The electromagnetic excitation and vibration data require reliable transmission, especially when deployed in complex urban environments with potential interference.
• Signal Damping: Variability in concrete cover thickness and material heterogeneity significantly affects signal quality and response, making accurate corrosion detection more challenging and necessitating careful sensor placement and calibration. The results obtained are promising, but further research is necessary.
• Implementation: M5 sensors should be placed close to rebars (within 5–10 mm) for sensitivity, but retrofitting existing structures can be costly and invasive. Deploying dense sensor networks across many smart city assets demands scalable, cost-effective solutions.
• Signal processing: Handling and analyzing high-frequency signal data from many sensors requires advanced feature extraction, noise reduction, and robust machine learning (ML) algorithms to identify corrosion progression stages accurately.
• Power Management: Long-term SHM requires sensors and communication nodes optimized for low power consumption or supported by energy harvesting to minimize maintenance needs.

After addressing these issues, it is essential to implement real-world pilot studies and collaborate with industry partners for practical testing.