Spalling

In addition to the effect of corrosion development on the GPR amplitude, the effect of crack propagation and the occurrence of wide cracks on the amplitude of the GPR signal was also noted by Lai et al. [56]. A decrease in amplitude was observed after the occurrence of a wide longitudinal crack. This was explained by the scattering of signal energy caused by additional irregularities when a wide crack propagates through the concrete cover. The extension of the experimental setup was presented in Reference [20] to obtain a better representation of the crack influence on the signal amplitude.

#### 3.1.2. Conclusions from Laboratory Simulated Corrosion Inspection

The corrosion process can be divided into the initiation and propagation phases, with each phase having specific effects on concrete microstructures. From the long-term corrosion monitoring experiments, the influence of corrosion promotion on GPR attributes was summarized, as in Figure 3.

**Figure 3.** Changes in the GPR signal during corrosion process.

The effects shown in Figure 3 are determined by the accelerated corrosion processes, in particular the formation of corrosion products on the surface of the reinforcing bar, their diffusion into the concrete cover and thus, crack propagation. These effects modify the amplitude in terms of different reflection coefficient and different dielectric properties of the concrete cover. The ability of corrosion products to migrate depends on the moisture content in the concrete cover since their movement is favored in the presence of moisture [59]. Similarly, the ability of their migration depends on the duration of the acceleration process. When accelerated corrosion with high current density is established in a short timeframe, it leads to increased crack width due to the sudden accumulation of corrosion products and increased pressure around the reinforcement [50,60]. Therefore, an appropriate current density should be selected to ensure the best possible simulation of natural conditions still within a reasonable timeframe [50,61].

The results of these studies also indicate that laboratory simulated corrosion studies mentioned above do not provide relevant results unless the level of corrosion is properly described. In these studies, results were recorded before and after corrosion acceleration and conflicting results were reported. Some authors reported higher amplitudes at the end of the experiments, while others reported lower amplitudes. Since these studies differ in terms of experiments setup, corrosion level and induced damage, it is possible that the observed changes in the GPR signal were recorded during different stages of the corrosion processes.

#### *3.2. On-Site Corrosion Inspection*

Most of the published research focuses on the application of GPR to the assessment of bridge decks, while other structures are represented to a lesser extent (tunnels, buildings, wharves, etc.). In terms of geographic location, most studies using GPR are conducted in

the United States of America. The authors are sure that this is also a consequence of the existence of relevant standards [62].

The use of ground-penetrating radar data for condition assessment of concrete bridges dates back to the early 1980s [63]. The amplitude of the ground-penetrating radar signal during an inspection is affected by the presence of structural elements, variations in cover depth, moisture, chlorides [22,64,65] and other variables. Therefore, a simple approach to evaluating changes in the GPR signal is not practical. Coexisting influences with other phenomena, such as variations in moisture and chlorides, are unavoidable and prevent the development of methods for direct location of corroded rebar. Therefore, indirect methods are used in which the localization of corroded areas is correlated with the areas of high signal attenuation. Attenuation has been found to be strongly related to the increased conductivity around the rebar [22] caused by the accumulated chloride ions and corrosion products [25].

The inspection of concrete piers and wharves is very similar to the inspection of bridge decks where moisture and chlorides are the main causes of deterioration. The use of ground-penetrating radar has also been reported in the inspection of tunnels. It was noted that the complicated design and compound deterioration mechanisms of these structures made a simple corrosion assessment impossible. Instead, the data was used to assess the overall condition.

Quantification of attenuation can be determined by numerical analysis of the signal or by visual analysis of B-scans. These are discussed in more detail in the following sections.

#### 3.2.1. Numerical Analysis of GPR Attributes

The ASTM standard [62] for the evaluation of concrete bridge decks using groundpenetrating radar proposes two procedures for numerical analysis using GPR data. The first procedure is based on considering the reflection amplitude from the bridge deck bottom and the bridge deck surface. The second procedure considers the reflection amplitudes from the top reinforcement layer. In most cases, the reflection amplitudes from the top reinforcement are considered for corrosion evaluation. The amplitude is derived from the A-scan.

Numerical analysis is usually performed by normalizing the amplitude, which represents the deterioration rate, and is calculated as follows [66]:

$$\text{Normalized amplitude} [\text{dB}] = -20 \log \frac{\text{signal amplitude}}{\text{reference signal amplitude}} \tag{1}$$

The signal amplitudes are compared to the reference signal amplitude which is usually the amplitude with the lowest degree of attenuation and represents sound concrete [66]. The GSSI [67] suggests 32,767 for 16-bit data and 2,147,483,648 for 32-bit data as the reference signal amplitude. This approach may be inconvenient when a concrete structure is in an advanced stage of deterioration and high attenuation is primarily detected. The differences between amplitudes are then smaller and the deterioration could be underestimated [68,69]. On the other hand, if the structure is in a relatively good condition, the attenuation may be misinterpreted. In this case, the attenuation could come from a different source, namely the variation of the concrete cover thickness. In such cases, it is recommended to consider the whole amplitude and not only the attenuation zones [70]. Other approaches have also been used, with Dinh et al. [25] using the average direct coupling wave as the reference amplitude. According to Pashoutani et al. [71], the use of a constant value of a reference amplitude does not take into account the contribution of concrete surface quality to the signal amplitude, even to the normalized amplitude. Therefore, a normalization procedure was proposed in which each signal amplitude is normalized to its own direct coupling amplitude. In order to eliminate the influence of the cover depth variation on the signal amplitude, Barnes et al. [21] demonstrated an amplitude correction method. It was shown that subtracting the depth-dependent amplitude loss gives a better correlation of groundpenetrating radar amplitude maps with ground truth results than maps without correction.

The method is based on determining the linear-dependent function of signal loss from the two-way travel time (TWTT) for the 90th percentile value of normalized amplitude. The 90th percentile value of the normalized amplitude is supposed to represent the sound concrete where the attenuation is mainly caused by the propagation of the signal through the dielectric material, i.e., the dielectric loss [25]. After correction of the amplitude, the attenuation should represent the signal loss due to chloride and moisture, i.e., the conductive loss. The method was improved after it was found that the conductive loss was also depth-dependent, so that an additional correction was necessary [25]. Two automated methods for depth correction have also been proposed [72]. In these studies, the correction was performed at the two-way travel time level. A more accurate correction could be performed if the linear function is determined using the real reinforcement depth instead of the two-way travel time [71]. This procedure requires the determination of the real velocities of the signal.

Obviously, it is of interest to establish a threshold for attenuation that is suitable for identifying the area of deterioration. However, a universally applicable threshold has not been established. It is usually based on the experience of the analyst and is related to a specific case [21]. There have been several attempts to relate the attenuation, mostly in comparison with thresholds of other methods. In References [73–76], ground-penetrating radar data were correlated with half-cell potential data, with the aim of determining the threshold value of attenuation. In Reference [75], the ROC (Receiver Operating Characteristic) curve was used, and in Reference [76], the author used statistical parameters to obtain the threshold value. In the second paper, the relationship between the percentage of corroded area, based on the results obtained with the half-cell potential at several bridges, and the product of the mean and skewness of the amplitude of the ground-penetrating radar, was established. The relationship can be used to predict the corroded area based on the analysis of the statistical parameters of the amplitudes. In some studies [77,78], the k-means clustering method was used to determine thresholds values.

Numerical analysis is generally used to obtain the deterioration map, which in most cases is the spatial distribution of normalized amplitudes. The main steps of the numerical approach in the condition assessment of concrete bridge decks are shown in Figure 4. The ability of the numerical approach to provide autonomous assessment of concrete structures using GPR is one of the reasons for its predominant use, while the algorithms for automatic reinforcement selection can be found in References [79,80].

**Figure 4.** Numerical approach in condition assessment of concrete bridge decks.

The results obtained by periodical inspections can be collected in databases, so that the correlation of successive data allows continuous monitoring of the progress of deterioration. Dinh et al. [65] also proposed a method based on comparing the complete waveform (amplitudes and shapes of the electromagnetic wave) at a point with baseline data. The advantage of this method is that it excludes non-corrosion attenuation causes. However, baseline data is required for proper detection of deterioration, which is often not available. The method was improved in Reference [81] and the waveform was compared with the simulated waveform. The simulated waveform has the original direct wave, but it has no reflected wave, so it simulates full attenuation. The higher similarity with the simulated wave correlates with a higher degree of deterioration. Hong et al. [82] proposed a method to monitor the corrosion process by comparing different GPR data using the image registration technique.

Despite the deterioration maps, the statistical distribution of amplitudes had shown the relationship with the condition of the structure. In Reference [83], where several bridges with different environmental conditions were observed, it was found that after correcting the amplitude depth, the distribution of the amplitude of the sound deck was symmetrical with high kurtosis. In contrast, severely damaged concrete exhibited higher dispersion of amplitude distribution with lower value of kurtosis. This statistical dependence has been previously confirmed [74]. An automated crack tracking method based on the analysis of the processed amplitude of the ground-penetrating radar was presented in Reference [84]. The model considers the amplitude compared with the threshold value. The final result of the model is a three-dimensional (3D) visualization of the cracks, which provides the possibility to evaluate their geometry. However, the reliability of the model depends on the threshold value, which is difficult to determine accurately.

#### 3.2.2. Visual-Based or Combined Analysis of GPR Attributes

In addition to the numerical approach, a visual or combined visual and numerical approach has been supported by a number of authors [22,77,85]. The visual approach implies the visual analysis of B-scans. This method is highly dependent on the expertise of the analyst, especially in the case of severely damaged structures [86], so the final conclusion is prone to error. As noted by some authors [22], numerical analysis of amplitudes misinterprets most anomalies that alter the signal and are not causes of deterioration (surface anomalies, reinforcement spacing, reinforcement depth, structural variations). Due to of these drawbacks, a method is proposed in which an analyst reviews the ground-penetrating radar profiles (B-scans), considers the reflections of the reinforcement and concrete surfaces and marks the boundaries of deteriorated areas. The profiles are processed, and the final output is the corrosion map. The detailed procedure is described in Reference [87]. This method was improved to overcome the subjective opinion of analysts in visual-based interpretation [85]. A set of if/then rules was created to locate anomalies that alter the signal but do not indicate deterioration. Dinh et al. [77] used visual analysis of ground-penetrating radar profiles as a tool to determine the number of condition categories as input to the k-means clustering method. It is a combined method: after determining the number of condition categories, the amplitudes of the signal are grouped and thresholds between the groups are determined. The corrosion map obtained in this way was used for the deterioration modelling of concrete bridge decks presented in Reference [88]. Dawood et al. [89] presented an improved visual-based analysis of ground-penetrating radar data for the detection of air and water voids in tunnels. Moreover, an evaluation flowchart based on inspection of pier structure considering B-scans and GPR signal energy was proposed in Reference [90].

#### 3.2.3. Condition Assessment by Combination of Multiple NDT

Ground-penetrating radar has a number of advantages over other non-destructive techniques (NDT), and it is not surprising that it has shown much interest in replacing other techniques. It is completely non-destructive, and it is rational to give it precedence over other techniques that make surveying slow and less efficient. In the next sections, a brief overview is shown on current research results obtained by comparing GPR data with other test methods, such as electrical resistivity (ER), half-cell potential (HCP), chain drag (CD), hammer sounding (HS), infrared thermography (IRT), acoustic emission (AE) and impact-echo (IE). These studies are summarized in Table 2.

Electrical resistivity and half-cell potential are fundamental tools for determining the probability of corrosion in the condition assessment of concrete structures. A good correlation has been found in the analysis of electrical resistivity and GPR data [81,91–93]. However, such behavior is to be expected as both techniques are affected by the conductivity of the concrete [91].

The comparison between HCP and GPR data can be found in References [21,24,69,73,74,81,92,93]. All observations were obtained by superimposing the signal attenuation and potential maps. In most studies, a good correlation was found since the attenuation is indicative of a corrosive environment and coincides with the areas of extremely negative half-cell potentials [92]. However, when the degree of deterioration is low, the ground-penetrating radar could overestimate corroded areas [69].

Other techniques can also serve for condition assessment and correctly predict potential deterioration due to corrosion propagation. These techniques include chain drag (CD), hammer sounding (HS), infrared thermography (IRT), acoustic emission (AE) and impact-echo (IE). Compared to the chain-drag method, the ground-penetrating radar is effective while the deterioration level ranged between 10% and 50% [69]. However, the divergence between the result of the ground-penetrating radar and the acoustic scanning system was observed in Reference [94], where the authors investigated the suitability of these techniques for delamination detection. The area of high attenuation was larger than the delaminated area detected by the acoustic system because the ground-penetrating radar generally detects the deterioration earlier than the acoustic system. The GPR can detect deterioration before delamination occurs. Also, the comparative feasibility study on delamination detection using ground-penetrating radar (GPR) and infrared thermography (IRT) based on ROC (Receiver Operating Characteristic) analysis showed that IRT is more reliable than GPR in detecting delamination [95]. However, the contribution of IRT is limited to a shallow cover depth, while GPR can provide a deeper insight. Also, the usefulness of GPR in predicting repair quantities was presented in Reference [96], where the results of ground-penetrating radar matched the depth of removal measured by LiDAR (Light Detection and Ranging) method after hydro-demolition.


**Table 2.** Review of studies that combined GPR with other techniques.


**Table 2.** *Cont.*


**Table 2.** *Cont.*

In a very detailed study, Omar et al. [109] presented the weaknesses and advantages of the most commonly used methods for condition assessment of concrete bridges. The conclusion was that none of the commonly used techniques are able to detect active corrosion, delamination and vertical cracking simultaneously, so that the most reliable condition assessment lies in a combination of multiple techniques. Such an approach ensures accurate condition assessment as deterioration can be detected from its onset to an advanced stage [23].

The simultaneously used non-destructive techniques usually consider methods such as ground-penetrating radar, electrical resistivity, half-cell potential, ultrasonic surface waves, impact-echo, etc. In References [14,100,110], an example of integration of different non-destructive testing methods in robotic systems, RABIT (Robotics Assisted Bridge Inspection Tool), was presented, which ensures real-time visualization of the concrete deck condition. Here, the evaluation is supported by a Jensen–Shannon probability method that focuses on the determination of the condition index [108]. Additional support in the interpretation of GPR data for delamination detection can be provided by infrared thermography (IRT) [99,103,104,107]. Solla et al. [106] demonstrated the technique to inspect a military base in an advanced stage of corrosion with visible signs of damage such as cracking and spalling. The results obtained with GPR were combined with the IRT technique. The corrosion assessment was based on the observation of GPR signal attenuation, changes in signal velocity and amplitude polarity. Overall, high signal attenuation was declared to indicate the presence of mineral salts and moisture, while reverse reflection polarity could be a sign of voids. The same parameters have been used in the assessment of wastewater plants [111], although the corrosion process is different in this case.

Deterioration modelling was part of the study in Reference [102], in which deterioration curves were developed based on the condition assessment of 10 bridges. Similar assessments were carried out by Alani et al. [101], where finite element models were constructed based on inputs from ground-penetrating radar and the deflection and vibration sensor system. In Reference [105], a data fusion model for bridge deck evaluation was developed based on the combination of the results from the ground-penetrating radar, half-cell potential method, electrical resistivity method and impact-echo method. In Reference [112], the ground-penetrating radar data combined with the capacitive technique and the impact-echo method were correlated with durability indicators for the overall assessment of the wharf.

#### 3.2.4. Conclusions from the On-Site Corrosion Inspection

The previous section has shown that corrosion assessment in on-site corrosion testing is mostly based on the assessment of the attenuated areas identified by signal amplitude analysis. Most of the studies are carried out on the bridge decks. In terms of comparison with other NDT, GPR has been compared with various techniques used for the service life condition assessment of the structures, Figure 5.

The high correlation between the electrical resistivity and the attenuation maps obtained with ground-penetrating radar is to be expected, as the signal depends on the material properties, so the conductive medium generated by moisture and chlorides changes its properties. In general, the GPR has shown good agreemen<sup>t</sup> with the HCP. However, there are certain situations where the GPR does not agree very well with the HCP. In cases where moisture and chlorides provide a favorable environment for corrosion, but their concentration is not sufficient to start corrosion, the GPR and HCP maps may differ. The applicability of ground-penetrating radar in detecting delamination is also uncertain [113]. In many cases, it does not detect delamination directly, and the assessment is based on the localization of deteriorated areas [114]. Moreover, the visual signs of delamination are not always visible on B-scans [87]. If the delamination is too thin to be detected by the antenna, it will not show any detectable change on the scan.

In summary, additional information, such as the age of the structure or the environmental conditions, may be helpful in analyzing GPR results. Moreover, this additional

information can be obtained with other NDT, so a suitable combination of NDT can be a very powerful tool for the condition assessment of concrete structures.

**Figure 5.** Multiple NDT used for the condition assessment.

### **4. Conclusions**

This paper investigated the evaluation of corrosion probability in concrete using ground-penetrating radar. The study analyzed laboratory and on-site investigations and the results were related to the evolution of the corrosion process. Advantages and recommendations for future research are presented below.

GPR is a completely non-destructive method, which gives it an advantage over other techniques for corrosion assessment of reinforced concrete. Its ability to examine large areas in a short time, together with providing information on the depth and spacing of reinforcement, makes it a multifunctional NDT. The literature review identified certain challenges in the use of GPR for corrosion assessment, one of the main being the understanding of the influence of concrete conditions on GPR parameters. In fact, in most laboratory studies, moisture and chloride content were controlled after depassivation of the reinforcement. On-site in real conditions, variations of moisture and chloride content are inevitable, which makes the detection of corroded areas based only on the observation of amplitude potentially ambiguous. Since opposing data have been reported in the literature, further laboratory studies are needed to show the influence of the change in dielectric properties of the concrete cover on the GPR amplitude and the change in reflection coefficient due to the formation of corrosion products and their migration. Since an absolute comparison of studies is difficult due to the variance in experimental design and the degree of damage induced by the accelerated corrosion process, further studies should correlate the degree of damage with the change in GPR amplitude. Obtaining concluding results from the proposed research topics could enable the use of GPR as a stand-alone tool for detecting corroded areas, moving from its use for the detection of corrosive environment towards detection of corrosion itself.

In conclusion, as the knowledge of the effect of corrosion on the GPR signal increases, GPR will be a very valuable tool for condition assessment of reinforced concrete structures. This method will certainly be improved, leading to an upgrade of the construction management system and making the assessment more reliable with reduced maintenance costs.

**Author Contributions:** Conceptualization, K.T., A.B. and M.S.; methodology, K.T. and A.B.; investigation, K.T. and A.B.; data curation, K.T.; writing—original draft preparation, K.T.; writing—review and editing, A.B. and M.S.; visualization, K.T.; supervision, A.B.; project administration, M.S.; funding acquisition, M.S. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by the European Union through the European Regional Development Fund's Competitiveness and Cohesion Operational Program, gran<sup>t</sup> number KK.01.1.1.04.0041, project "Autonomous System for Assessment and Prediction of infrastructure integrity (ASAP)".

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** Not applicable.

**Conflicts of Interest:** Authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.
