Tunnel Inspection Review: Normative Practices and Non-Destructive Method Advancements for Tunnels with Concrete Cover

Abstract

Guaranteeing tunnel integrity is an important issue for several countries worldwide. Due to the continuous increase in the number of tunnels, as well as the aging of old tunnels, several countries and companies have created manuals to standardize tunnel inspection and assessment. Most manuals still specify just visual procedures for tunnel inspection; however, because of underground conditions, the structural system of tunnels is often accessible only by one side, thus posing difficulties to the accurate evaluation of the structural conditions of the tunnel using only visual inspection. A possibility to improve the effectiveness of tunnel inspection is the use of non-destructive testing (NDT), which will assist in obtaining information about the inner condition behind the tunnel wall. The current advancements in the NDT methods allow them to be employed in all the different kinds of inspections (initial, routine, special inspection) suggested by the manuals. Therefore, in an attempt to help in the decision about the application of each method, this work provides an overview of some international practices for tunnel inspections and shows a review of different NDT methods (traditional and new methods) applied to tunnel inspections. Furthermore, this study describes their workability, advantages, and capability, and classifies the best fitting of each in the inspection procedures.

1. Introduction

A tunnel is a complex underground structure that is highly relevant to society. Its failure represents a significant social, economic, and environmental impact. Despite its importance and impact, tunnel degradation is of great concern for tunnel operators, as it can lead to structural failures. The degradation of tunnels can be influenced by various environmental factors, including ground conditions, water infiltration, and changes in temperature, added to the influence of its utilization, which has continuously increased.

Due to high financial investments and challenges involved in the design and construction of underground structures, tunnels remain in service for long periods of time, even beyond their intended service lives. According to FHWA, more than 40% of the USA’s highway tunnels are older than 50 years [1]. This raises the need for effective inspection and maintenance procedures to ensure structural integrity over their entire lifespan.

In recent years, several institutions and companies worldwide have created technical reports, standards, and manuals as reference guides for routine inspections and procedures, such as FHWA (2015) [1] and CETU (2015) [2]. The establishment of adequate inspection routines is essential to optimize the tunnel’s life cycle and ensure its safety, especially due to the aging of these structures. Also, the application of technological solutions may significantly improve the analysis of tunnel integrity.

Recent advances in inspection with Unmanned Aerial Vehicle (UAV) methods, computer vision techniques, robotic inspection, and non-destructive methods, together with machine learning and artificial intelligence, are leading to a better understanding and control of the actual tunnel condition, as well as to an automatic and autonomous tunnel inspection system [3,4,5].

Detecting qualitative and quantitative defects is essential for tunnel evaluation [6]. However, as with any underground structure, evaluating the condition of a tunnel is a challenge. Unlike other infrastructure systems, such as bridges and buildings, the tunnel structure is mostly accessible by only one side. As a result, many techniques applied to tunnel inspection are based only on superficial evaluation. Thus, the adequate evaluation of a tunnel structure depends on the application of non-destructive testing (NDT) methods. These methods should be applied to control quality during the construction phase, monitoring the integrity of existing tunnels, and to validate rehabilitation in operational tunnels.

In 1995, the International Tunnel Association (ITA) published a report from a working group describing state-of-the-art NDT methods for determining the condition of a tunnel lining [7]. The report is the result of a study conducted by ITA and STUVA (Research Association for Underground Transportation Facilities), which began in 1991. The aim of the study was to identify available techniques for non-destructive examination of tunnel linings, and figure out ways of looking through the tunnel surface. Different standards and manuals of the 1990s were listed and some possible NDT methods for application to tunnel linings were shown.

The 1995 ITA report should be considered as an interim document and should be completed with the exploration of more possibilities for non-destructive inspection for tunnel linings. Since 1995, many advances have been made in NDT methods [3,5,6,8] and many standards have been created and updated. To contribute to the development of better inspection practices, this paper explains the workability of traditional and new NDT methods and presents the advances and applications for tunnel inspection. In addition, the current international inspection practices are discussed and some possibilities for improvements are presented.

2. Current Inspection Procedures

In the last 10 years, a number of institutions and companies in the world have established guidelines and inspection manuals to obtain more consistent inspection results. Some examples are the TOMIE Manual created by the FHWA (Federal Highway Administration) in the USA [1], the Inspection Guide of CETU (Centre d’Études des Tunnels) in France [2], the standards of the New Zealand Transport Agency [9,10,11], the standard for tunnel inspection from ADIF (Administrador de Infraestructuras Ferroviarias) in Spain [12,13], and the Inspection Standards in the United Kingdom [14,15]. Additionally, in the UK, the CIRIA (Construction Industry Research and Information Association) published a guide on best practices for tunnel inspection and maintenance [16], which incorporates practices from various UK infrastructure owners, including Network Rail, Highways Agency, British Waterways, and London Underground.

These guidelines usually classify inspections into different types, which are summarized in

Table 1. Inspection frequency suggested by different guidelines.

TOMIE Manual	CETU (2015)	NZTA	ADIF	CS450/CS452
Initial inspection (before opening for operation)	Continuous inspection	Routine surveillance inspection (daily by the staff)	Basic inspection (visual)	Superficial inspection (continuous)
Routine inspection (every 2 years, possibility of 4 years)	Routine inspection (annually)	General inspection (every 2 years)	Principal inspection (after any observation in the basic inspection)	General inspection (visual, annually)
In-depth inspection (for complex tunnels; frequency defined by the manager)	Detailed inspection (every 6 years)	Principal inspection (more detailed, every 6 years)		Principal inspection and acceptance (more complete inspection, it cannot exceed a three-year interval)
Damage inspection (after special events)	Special inspection (after a specific event or defect)	Special inspection (after a specific event or defect)		Special inspection to investigate a specific defect
Special inspection to investigate specific defect	Higher-risk tunnels may have higher frequencies

. Although there are some variations between the classification of each guideline, the inspections can be grouped into four main categories, namely initial inspection (acceptance), continuous inspection (superficial or surveillance), routine inspection (basic, general, in-depth, detailed, general, or principal), and special inspection (damage). Initial inspection can be described as a detailed inspection performed before the tunnel is in operation or at the beginning of a new tunnel administration. Continuous inspections are inspections carried out continuously by the company staff. Routine inspections are periodic inspections with different complexity levels and, finally, special inspections are specific procedures in response to a special event.

Usually, these guidelines suggest procedures to evaluate the tunnel integrity, by establishing a method to quantify its condition [17]. For example, the assessment of concrete lining damage is commonly divided into local and global classifications. In the local classification, each damage is assessed separately and classified into a certain class, according to its severity. The number of classes varies from each guideline [17]. For example, FHWA [1] and ADIF [12,13] consider four classes, while [2] considers a more detailed classification, which consists of five classes for common pathologies and three more classifications when there is the presence of water. McKibbins et al. [16] recommend that a condition classification system should use simple condition grades; for example, classes ranging from A to E, with A representing the ideal condition.

In global classification, the tunnel is usually divided into sections, and its condition is assessed based on the combined effect of all local pathologies [5]. There are also practices that divide the tunnel into elements; for example, the TOMIE Manual [1] uses a classification system based on the degradation percentage of each element. This system provides information to the National Tunnel Inventory (NTIS) database, which records the condition of all tunnels in the USA.

Although these methodologies help to reduce the subjectivity involved in inspection activities, schedule, procedures, documentation, and evaluation still need improvements. For these reasons, these themes have been research topics discussed by several authors. Ai et al. [18] pointed out that the periodic inspection frequency decided subjectively can be a problem, since the deterioration rate varies for different tunnels. In [19], a methodology to predict the tunnel performance based on the tunnel degradation history is presented, as well as documentation based on a digital twin. Yuan et al. [20] suggested a method that quantifies the structural condition of the linear distance throughout the tunnel lining.

The common procedures for tunnel inspection and evaluation are mainly based on visual inspections, which are often compromised by subjectivity, due to the different levels of expertise of inspections [17]. To overcome this limitation, NDT methods can be used for structural condition assessment. In [21], a score flow chart is suggested to improve the diagnosis of the masonry tunnels of the Paris Subway System based on visual evaluation in combination with NDT methods. Dawood et al. [22] evaluated and mapped the condition of the concrete of a tunnel using Ground Penetration Radar (GPR), creating a methodology to identify voids and water inside the tunnel structure and mapping them. Thus, NDT methods should be included in inspection procedures with a view to having more detailed information that will assist in the evaluation of the integrity of tunnels. Intending to justify the need for increasing the use of NDT methods in inspection procedures, the next section presents the most common and recent NDT techniques for tunnel inspection and their applicability.

3. Non-Destructive Tests for Tunnel Inspection

There are many different non-destructive testing techniques, and each of them is more suitable for a specific problem or structure than the other. These techniques are mainly based on mechanical and acoustic readings, radiation, electric, electromagnetic, or optical methods. In general, the traditional NDT methods used for concrete investigations are the mechanical acoustic, sonic, and ultrasonic methods, infrared thermography, electromagnetic methods [23], and, recently, photogrammetry and computer vision methods. The principles of each technique are presented next.

Radiation techniques include X-rays, gamma rays, and neutron rays. They measure the effect of the target mass on a beam of sub-atomic particles [7], and are based on the detection and recording of wave radiation penetration into a medium. The radiation that passes through the medium is recorded on a film and displayed on a screen. The absorption depends on material density and thickness, as well as other characteristics [23].

The mechanical and acoustic techniques use the material’s mechanical oscillation. These methods can be used to investigate material properties, layer thickness, object, and void localization. Some examples of acoustic methods are the hammering test, impact response, Impact-Echo (IE) [24,25], sonic transmission, ultrasonic tomography, impulse echo techniques [26,27,28], and a recent technique using a Laser Vibrometer to read the local vibration by distance to locate voids in the concrete [29,30].

The sonic transmission method consists of a mechanical wave that passes through the concrete. The signal is received by an accelerometer on the opposite side of the transmission source [23]. Since this method needs access to the opposite side of the structural element, it is not very suitable for tunnel inspection. However, the pulse echo method with the tomography technique improved the mechanical wave methods allowing for the signal reading on the same surface of the excitation, using the wave reflected by the flaws in the medium.

Electrical and magnetic methods are commonly used to locate and evaluate metallic objects as in eddy current methods and electrical potential methods. However, in addition to the condition assessment of steel elements, some radar techniques that use electromagnetic waves are used for a more complete concrete tunnel lining evaluation, such as the Ground Penetration Radar (GPR) [31,32,33] or Transient Electromagnetic Radar (TER) [34,35,36].

Optical techniques can evaluate the tunnel condition superficially, or close to the surface. Some examples of these methods are visual inspection, infrared thermography [37,38], and photogrammetry [39]. The optical methods are in continuous evolution owing to the application of computer vision techniques, such as photogrammetry and thermography, along with machine learning algorithms. Thus, the inspection routine tends towards an automated inspection procedure [5].

The aim of non-destructive methods is to obtain complete, reliable, and reproducible data about the state of the entire tunnel. These methods can be applied to fast and preliminary inspections and to more detailed investigations. In the former case, the methods are intended to provide an initial survey about the state of the tunnel. Therefore, they must examine the entire tunnel as quickly as possible during short breaks in operation and establish the defective zones as accurately as possible. After defective zones have been localized, a detailed investigation is needed to obtain more precise information about the extent of the damage [7].

Some non-destructive methods are easily applied to rapid and preliminary investigation, such as the Ground Penetration Radar (GPR), thermography, and computer vision techniques. Other methods, instead, are more adequate to be used in special inspections. To define the best application of NDT methods to tunnel inspection, some of the methods will be further discussed, presenting examples of real cases taken from the literature, as well as some advancements and trends.

3.1. Visual Inspection

Visual inspections are the most traditional non-destructive method that can provide valuable information about tunnel condition. Although it is quite prone to subjectivity, inspection routines are mostly fully dependent on visual inspections. The success of the method depends on a well-trained inspector who is able to evaluate the material deterioration and the structural condition. The inspector should be able to identify and to classify different signs of anomalies, and the information generated may indicate the need for a special inspection with more sophisticated techniques.

3.2. Hammering Method

The hammering test is a traditional technique applied to tunnel inspection that involves applying a hammer impact to the wall to locate voids by the sound perception of a trained inspector [40]. Likewise, the visual inspection method is very subjective, depending on the inspector’s evaluation.

3.3. Impact-Echo

Impact-Echo (IE) is a low-frequency elastic-wave method based on the transient response of a material to mechanical impact [41]. The method consists in applying an impact force to the structure surface with a hammer. This creates a pulse of P-wave (primary or compression wave), S-wave (secondary, distortion, or shear wave), and R-wave (surface or Rayleigh wave) into the material (Figure 1). The stress waves are propagated into the material and reflected due to internal flaws, voids, and material layers, whenever there is a difference between the acoustic impedance at separating materials. When the body stress waves reach the backwall surface (Figure 1b) or a void (Figure 1c), part of the energy is dissipated and part is reflected as P-wave and S-wave. The forward P-wave, that goes from the impact point into the material, is a compression wave, and the reflected wave (PP-wave in Figure 1) is a tension wave that makes the surface close to the impact region go to an opposite displacement [42]. The reflected signal is recorded by a transducer placed on the concrete surface, close to the impact location. Due to the sensor proximity to the impact point, the signal obtained is usually dominated by P-wave reflections, since P-waves propagate faster than S-waves and Rayleigh waves [43]. Thus, the method is based on the analysis of collected P-wave signals.

Repetitive reflections between the interfaces form a periodic modal reflection, exciting local modes of vibration, corresponding to the thickness resonant frequency [44]. This thickness resonance is characterized by the transient period of the P-wave (time taken for the wave to reach the reflective interface and return to the beginning). With the resonant frequency and the wave velocity in the medium, the wave reflector position is estimated.

The Impact-Echo method is widely used for concrete evaluation. In tunnels, it has been applied to study grouting quality, tunnel shotcrete bonding state, layer thickness, material quality, and void detection. In general, there are five common situations when analyzing IE in tunnels: three of them generate a clear frequency peak in the Fourier Transform as in the case of a solid concrete (Figure 2a), a void in the concrete (Figure 2b), and void between the concrete and rock (Figure 2c). It is also possible to have multiple clear frequency peaks if there are more concrete layers causing partial reflections (Figure 2d) and a polluted signal with multiple peaks (Figure 2e). A diffuse reflection (Figure 2e) can occur because of honeycombing, which can cause a decrease in frequencies of deep reflections, and because of rough surfaces (which can present a signal amplitude lower than the level of noise, preventing thickness definition) [44].

For the three cases, the depth ($d_r$) of the reflector (which is expected to be the post prominent peaks in the frequency domain) is determined by Equation (1).

$$d_r=\beta {\displaystyle \frac{C_P}{2f_r}}$$

(1)

where $f_r$ is the recorded resonant frequency (Hz); $\beta$ is the shape factor considered equal to 0.96 for solid plate; $C_P$ is the speed of the elastic compression wave (m/s) [45].

Considering an isotropic, linear-elastic, and homogeneous material, the velocity of P-waves ($C_p$) can be estimated by Equation (2) for an infinite plate (such as a tunnel liner), where v is the Poison coefficient, E is the elastic modulus, and $\rho$ the material density. Typically, the P-waves’ velocity in concrete varies from 3000 m/s to 4000 m/s. S-waves and R-waves have lower velocities compared to P-waves: about 0.61$C_p$ and 0.56$C_p$, respectively, for v = 0.20 [42].

$$C_p=\sqrt{{\displaystyle \frac{(1-v)E}{(1+v)(1-2v)\rho }}}$$

(2)

When the element has two different layers, with surface thickness $d_1$ and wave velocity $v_{p}1$, and the base layer, with $d_2$ and wave velocity $v_{p}2$, Equation (2) assumes the form of Equation (3) [27].

$$f_{r1}={\displaystyle \frac{1}{\frac{2d_1}{0.96C_{p1}}+\frac{2d_2}{0.96C_{p2}}}}$$

(3)

When there is a void in the concrete (Figure 2b), the excited frequency by the void is higher than the thickness frequency response. However, if the void is located behind the tunnel lining (Figure 2c), the resonance frequency will be the same for concretes with and without void. In this situation, the void can be located by the impedance (Z) difference among tunnel lining soil and air, which will cause a difference in the Reflective Coefficient (R) [24]. The impedance and the Reflective Ratio are calculated by Equations (4) and (5), respectively, for a P-wave propagating from material A to B.

$$Z=\rho \times C_P$$

(4)

$$R={\displaystyle \frac{A_r}{A_i}}={\displaystyle \frac{Z_B-Z_A}{Z_B+Z_A}}$$

(5)

where $\rho$ is the material density, $A_r$ is the reflected amplitude, and $A_i$ is the incident amplitude.

If there is an interface between the tunnel lining and the air, the impedance of ($Z_A$) is much larger than the air ($Z_B$), resulting in a Reflected Ratio (R) close to −1, which means most of the signal is reflected. If there is an interface between the concrete and rock, then $Z_A$ and $Z_B$ are approximately equal, resulting in R close to zero, which means that a significant part of the energy propagates through the interface to the surrounding rock [24]. Figure 3 shows a time history of a signal at a concrete/soil and concrete/void interface. As can be seen, the reflection amplitude is higher in the concrete/void situation.

Although Impact-Echo (IE) is mainly applied to locate voids, some other applications can be found in the literature; for example, for grouting quality behind segment tunnel [46,47,48] and for shotcrete state [25]. Song and Cho [49] stated that, by analyzing the IE signal, the bonding state of shotcrete on hard rock can be classified into fully bonded, deboned, and void conditions. Aggelis et al. [46] used the IE signal to evaluate the grouting quality by locating voids using the reflection coefficient. Strong reflected signals are caused by voids, which means that there is a deficit in the grouting behind the concrete. Yao et al. [47] experimentally investigated different signal processing for grout quality with IE and found good efficiency in using Wavelet transforms and energy analysis methods. Sung et al. [48] used three different indicators (Geometric Damping Ratio, Dominant Resonance Curation, and post peak amplitude) to automate the grouting evaluation for a shield TBM tunnel using the Impact-Echo method.

The Reflective Coefficient and resonant frequency analysis are common techniques to evaluate voids with IE. However, additional signal processing approaches should be explored to improve data interpretation. Song and Cho [25,49] proposed a method to quantitatively evaluate the bonding between shotcrete and rock using the Short-time Fourier Transform of vibration waves from IE signals (Figure 4). Cao et al. [24] tested different signal processing techniques to detect voids behind tunnel linings using experimental and numerical IE signals. The results showed that a better energy distribution is obtained when using Wavelet analysis. In addition, the use of dynamic stiffness for void localization was suggested. Cheng et al. [50] proposed a method with simulated transfer function derived from the IE signal to evaluate the bond between concrete and substrate; a realistic force–time function is generated by the Rayleigh wave recorded by the transducer.

Therefore, Impact-Echo is useful for construction quality control and for inspection procedures by estimating concrete layer thickness and evaluating its quality by locating voids and detachments. Since it is a local technique able to measure only one point each time, it is more suitable for special inspections to evaluate a specific place or the cause of a pathology, or for sampling collection during the tunnel construction.

3.4. Ultrasonic Method

Ultrasonic techniques encompass all acoustic methods with frequency typically over 20 kHz. They are useful for evaluating concrete structures by estimating layer thickness, locating internal duct and reinforcements, identifying the presence of voids and cracks, and estimating the elastic constants [26]. In the ultrasonic pulse velocity (UPV) method, an ultrasonic wave is mechanically induced into the material and transferred between the transmitting and receiving transducers (probes). In the case of the ultrasonic-echo method, the stress wave is reflected by internal discontinuities, producing echoes that are received by the same probe [28].

The UPV and pulse-echo method needs a smooth concrete surface and a coupling agent to achieve its proper workability, making it a time-consuming methodology for some field applications. However, some equipment for ultrasonic tomography, such as MIRA, EyeCon, and Pundit, which also use wave propagation principles, do not require a perfectly smooth surface or the use of a coupling agent, making them an alternative for field inspection. A systematic review of ultrasound tomography, discussing advancements in techniques and technologies for concrete inspection, can be found in [51].

Ultrasonic Tomography

Ultrasonic tomography (UST), also known as ultrasonic pulse-echo or ultrasonic shear wave, is a technique based on the elastic-wave excitation of the material. The method incorporates some independent ultrasonic probes that work simultaneously as exciters and signal receivers, allowing volumetric imaging from the inside of the material (Figure 5A). The result generated can be seen in four different representations, A-Scan, B-Scan, C-Scan, and D-Scan (Figure 5B). A-Scan is the simple wave registered by one probe. B-Scan, C-Scan, and D-Scan are images that represent a matrix resulting from the results collected by the probes.

The main operation principle of the equipment is based on the coherent accumulation of reflected signals received by different elements of the antenna array, with their further processing. The implemented Synthetic Aperture Focusing Technique (SAFT) algorithm allows for the building of the 3D data of the structure’s reflected parameters in the tested area and then reviewing them in different sections and images [52]. The SAFT algorithm increases the tomography resolution and helps locate the object and voids. Figure 5C shows an example of tomography of a concrete structure.

Figure 5. (A) Tomograph workability. (B) Types of scans (adapted from [53]). (C) Tomography of a reinforced concrete wall.

Figure 5. (A) Tomograph workability. (B) Types of scans (adapted from [53]). (C) Tomography of a reinforced concrete wall.

When receiver probes detect the reflected waves (echoes), the travel time of these waves can be calculated, and the reflector can thus be located. Wave reflections are produced due to internal defects, interfaces between materials with different densities and elastic modulus, and boundaries of the solid [42]. Therefore, the method is used to determine the location of defects, objects, thickness, and material interfaces by knowing the velocity of shear waves in the material and by measuring their travel time (Figure 5C).

The travel time depends on the wave propagation in the medium. Unlike other methods based on stress wave propagation principles that use P-waves, inducing particle motion parallel to the wave propagation, UST uses shear wave propagation (Figure 6). The particle motion on shear waves is perpendicular to the wave propagation and parallel to the concrete surface. Similarly to P-wave velocity, in which $C_p$ is a function of the Young’s modulus (E), S-wave velocity $C_S$ depends on shear modulus (G), material density ($\rho$), and Poisson’s ratio ($\upsilon$), as seen in in Equation(6) [28].

$$C_S=\sqrt{{\displaystyle \frac{E}{2\rho (1+\upsilon )}}}=\sqrt{{\displaystyle \frac{G}{\rho }}}$$

(6)

In tunnels, ultrasound tomography can be applied to layer thickness estimation, rebar location, void estimation, concrete inhomogeneity, and grouting quality survey [26,52,54].

White et al. [26] evaluated the applicability of the UST method for tunnel surveys. By performing a linear analysis of 30 measurements of concrete and shotcrete specimens in a laboratory, they obtained a detection coefficient ranging from 82% to 98% in detecting surface depth and void size. The authors also tested the ultrasonic tomography in the Hanging Lake Tunnel. UST was able to perform deep inspection with anomaly location up to 50.8 cm, and possible delamination or debonding. The results also show the method’s capability to estimate the rebar position in the absence of superficial cracks. The occurrence of reflective areas near the surface indicates superficial cracks, and it hides deep information. Thus, the lack of deep details in the tomography is a good indicator of superficial cracks. In White et al. [54], the Ultrasonic Tomography System (UTS) was tested alongside Ground Penetrating Radar and Impact-Echo for the inspection of an underwater tunnel. The results demonstrated that the UTS had a higher capacity for detecting anomalies and provided a well-defined estimation of concrete thickness in this study.

The grouting quality behind the concrete panel can be investigated by the backwall signal and its intensity. In areas with good grouting quality, the backwall signal has low intensity because the main energy penetrates the grouting mixture. If there is a void, most of the energy is reflected by the air, resulting in multiple high-intensity backwall reflections [52]. An example of using UST for crack evaluation and concrete assessment can be found in [55]. The authors propose a methodology that utilizes UST to evaluate the state of concrete by analyzing the variance in wave propagation velocity during the cracking process in an experimental setup.

Tunnels often present complex signals that make data analysis challenging. Improvements in data processing and the application of machine learning approaches can be beneficial for analyzing such structures. For instance, Wu et al. [56] proposed using Variational Mode Decomposition (VMD) and bilateral filtering methods to enhance SAFT results, followed by a machine learning approach using the YOLO algorithm for defect detection. Wang et al. [57] introduced a double-ray SAFT imaging method to increase resolution and reduce signal noise, thereby improving the analysis of grouting quality using UST.

Additionally, multilayered structures can pose challenges for wave propagation methods. In the case of UST, variations in shear wave velocity across different layers can lead to errors in structural evaluation. This issue should be carefully addressed during UST inspections. Certain strategies and data analysis techniques can help mitigate this problem. For example, Lin et al. [58] combined the SAFT algorithm with the Multilayer Delay-And-Sum (MLDAS) method to improve imaging results in multilayered structures.

Therefore, UST can be applied to construction quality control and inspection procedures to estimate layer thickness, rebar covering, and to identify voids. The method has the advantage of creating 3D visualization of the inspected area. As a result of the need of contact with the surface, it is recommended to be part of special inspections to evaluate a specific place or the cause of a pathology. The surface quality influences the results; hence, for surfaces with many irregularities and cracks, UST may be difficult to apply. Therefore, in tunnels with shotcrete, adequate preparation may be needed on the surface to improve its results.

3.5. Ground Penetration Radar

Ground Penetration Radar (GPR) is one of the most recommended NDT methods for subsurface inspection, able to provide high-quality images from the inside of the structures [59]. During the last 30 years, GPR has been used for several different applications, such as the evaluation of cultural heritage buildings and modern concrete buildings, road pavement and bridge inspection, tunnel liner (plane concrete, reinforced concrete, shotcrete, and brick) inspections, geological/geotechnical applications, and underground utility location [32].

A GPR antenna comprises a signal generator (Tx) and a sensor receiver (Rx) (Figure 7). The device sends electromagnetic waves through the material, and these waves are reflected due to variations in the electromagnetic properties of the medium. Then, a receiver sensor reads the waves reflected by different materials. There are several different kinds of antennas; some antennas may send and receive the wave in the same element (monostatic antennas) or use a generator separated from the receiver (the bi-static antenna is an example).

GPR can identify the presence of objects before the antenna reaches a point above them, which generates the traditional hyperbolic transit time curve in the GPR data (Figure 7), a curve that occurs due to the transient time (period for the signal to complete the entire path from Tx to the object and return to the RX sensor), which becomes shorter when the antenna is moving towards the object and longer when it passes through the object. The lowest transient time registered indicates the reflector position [7].

The GPR results can be represented in A-Scan, B-Scan, C-Scan, and D-Scan (Figure 8). A-scan is the basic measurement in the time domain, a single waveform recorded by GPR. B-scan is formed by gathering a set of A-scans along the x-axis [60], named radargram. Merging more than one radargram allows creating a C-Scan and a 3D reconstruction of the medium [61]. The combination of these different signal representations helps to achieve a more precise data analysis and evaluation.

The method is useful for application to tunnels for quality control or tunnel evaluation surveys. It can be applied to locate and quantify rebars, estimate layer thickness, detect voids, water, corrosion, and some pathologies in the surrounding soil and rock.

The location of an object or interface is performed by the analysis of reflected signals. The deep position ($d_r$) of a reflector can be estimated by Equation (7), where t (seconds) is the wave transient time, and $V_M$ (m/s) the diffusion speed. The diffusion speed depends on the dielectric constant ($\epsilon$) of the material, and can be estimated by Equation (8), where ${\epsilon }_R$ is the relative dielectric constant, and c is the speed of light in vacuum, which is approximately 3 × 10⁸ m/s [7].

$$d_r=0.5V_{M}t$$

(7)

$$V_M={\displaystyle \frac{c}{\sqrt{{\epsilon }_R}}}$$

(8)

Equation (7) assumes a homogeneous medium with constant wave velocity. However, in the case of multiple layers with varying wave velocities, this assumption can lead to significant errors in depth estimation. Several approaches can be used to improve accuracy, such as hyperbola fitting for parameter estimation [62] and layered velocity modeling [63]. Additionally, software tools such as ReflexW, GPR-SLICE, and Radan7 offer functions to analyze velocity variations. Examples of depth estimation in tunnels with multiple layers can be found in [64], which evaluated void defects behind tunnel linings using a multilayer model, and Zeng et al. [65], that estimated grouting thickness in shield tunnels using machine learning algorithms.

The location of metallic objects is one of the main GPR applications for concrete structures. The rebar localization can be found by performing an Image-Based Analysis on radargrams. Rebars generate a hyperbolic pattern in the B-Scan image, which facilitates its location most of the time (Figure 8b). This application is very useful for tunnel quality control and evaluation. In [31], a 500 MHz frequency GPR antenna was used for this application in the Damaoshan Tunnel, in China. The findings showed that the rebar number was lower than the standard. Another example of GPR application to locate metallic objects in tunnels is presented in [66]. In that study, steel arches were located using a 300 MHz center frequency GPR with a train-mounted system for a fast periodic inspection routine.

Another important GPR application to tunnel survey is the thickness estimation of different material layers. The thickness estimation can be performed by applying different techniques, such as using the Fresnel Reflection Coefficient (FRC) [60,67] and the Time–energy Density Analysis based on Wavelet transform [68,69]. FRC is the most traditional method and it is based on the quantification of the electromagnetic wave reflection due to materials with different electromagnetic properties, calculated by Equation (9), where $\gamma$ is the reflection coefficient, and ${\epsilon }_1$ and ${\epsilon }_2$ are the relative permittivity of medium 1 and medium 2, respectively [60].

$$\gamma ={\displaystyle \frac{\sqrt{{\epsilon }_1}-\sqrt{{\epsilon }_2}}{\sqrt{{\epsilon }_1}+\sqrt{{\epsilon }_2}}}$$

(9)

When the propagated wave reaches different material interfaces, the amplitude of the reflective wave and the reflection coefficient ($\gamma$) change. If ${\epsilon }_1$ is higher than ${\epsilon }_2$, $\gamma$ is positive when the phase of the reflected wave is the same as the incident. If ${\epsilon }_1$ is lower than ${\epsilon }_2$, $\gamma$ is negative when the phase of the reflected wave is the reverse of the incident. When analyzing a single waveform in A-scan (Figure 8a), if the wave amplitude significantly increases or decreases, it is a sign that the wave may be reflecting different materials [60]. However, note that this amplitude variation hardly depends on the variation of the material properties.

For grouting quality evaluation, ming Yu et al. [70] combined GPR and image processing techniques to determine the effectiveness and quality of the grouting behind the tunnel lining in the Nanchang Metro Line 1. The tunnel lining thickness was estimated, and hidden damages were revealed. Xie et al. [71] evaluated the grouting for tunnel construction in soil areas in the Shanghai Metro, using a 200 MHz center frequency antenna, and [72] described a procedure for grouting evaluation by testing three different antennas on line 9 of the Shanghai subway. Zeng et al. [65] suggested a machine learning approach for grouting estimation using the Extreme Gradient Boosting (XGBoost) algorithm. They achieved high accuracy in estimating the backfill grouting of an experimental shield tunnel, and the method was also applied to the Tianjin Metro Line 6. For more examples of grouting evaluation, refer to [73].

GPR is an affordable solution to access the tunnel lining conditions. Several researchers have applied GPR to locate and quantify concrete lining pathologies, such as detection of voids with air and water [22,66,74], concrete delamination [59], concrete and external soil/rock disease detection [31,66], rockfall behind the concrete lining [75], and corrosion [22,76].

GPR data analysis allows for better mapping of the concrete deterioration state. For example, a method already used for bridge evaluation, as in Abouhamad et al. [76], uses signal variation and attenuation to locate anomalies such as voids, water content, and possible steel corrosion. The data obtained is used to create a concrete state mapping. Dawood et al. [22] performed a similar analysis for mapping concrete tunnel deterioration.

The possibility of using vehicle-mounted antennas brings significant improvement to tunnel inspections using GPR, since it allows a faster tunnel state survey. Zan et al. [66] built a mounted GPR multichannel system for an autonomous inspection. The solution used six GPR air-launched antennas with center frequency of 300 MHz fixed on a train, thus allowing faster layer scanning. The system was able to estimate the steel arch spacing, locate water seepage, estimate the shotcrete thickness, and locate rock diseases without traffic interruption.

Although GPR has a high potential for tunnel inspection, tunnel structures are typically composed of reinforced concrete with high concentrations of rebar, and different material layers. The presence of voids and water is also quite common, making data analysis difficult in many situations. In addition, different types of defects present similar features in B-Scan images [61], making it hard to identify a specific type of anomaly. To overcome these problems, different tools are used to help in GPR signal interpretation; for example, FDTD (Finite Discrete Time Domain) simulation and machine learning.

Zhang et al. [68] used A-Scan data from FDTD simulation to understand the application of Wavelet transformation to locate layer boundaries behind a steel mesh. Li et al. [77] used the FDTD method to better explain the signal attenuation during detection of rebar behind the second layer interface. Zhang et al. [78] used FDTD to propose a method to investigate tunnel lining cavities using the frequency band energy of the Wavelet transform packet. Xiang et al. [31] used the FDTD method to quantify and locate rebars using tunnel GPR data. Feng et al. [79] combined FDTD and Finite-element Time Domain (FETD) for GPR simulations of various defects in tunnel linings. Xie et al. [71] performed GPR simulations to define the best antenna for grouting evaluation in a soil tunnel construction.

Puntu et al. [60] applied Hierarchical Agglomerative Clustering to create a semi-automatic solution for rebar identification using GPR data from ten railway tunnels in Taiwan. Yang et al. [80] used a CNN (Convolutional Neural Network) trained with synthetic GPR data to map tunnel lining defects. Ren and Wang and Jie Xu [81] created a dataset called the MAI-GPR dataset, containing data from 18 tunnels, and tested various machine learning approaches to detect rebars, pipelines, steel arches, cavities, uncompacted defects, and voids. Li et al. [61] proposed a deep learning algorithm for subsurface defect detection, such as voids and cracks, in airport runways using multi-view GPR data, combining B-Scan and C-Scan. Liu et al. [82] used the algorithm YOLACT to simultaneously detect defects and the lining thickness of tunnels.

The high density of rebars in some tunnel structures can create electromagnetic shielding, reducing the effectiveness of Ground Penetrating Radar (GPR) in identifying defects behind reinforced concrete linings [57]. Due to the sensitivity of electromagnetic waves to metal, a high concentration of steel can cause strong reflections [83] and signal attenuation [84], which may obscure deeper defects. This represents a limitation of the method and must be taken into account.

In [83], a numerical and experimental study was conducted to investigate the influence of steel mesh on void detection, considering different GPR central frequencies, steel mesh spacings, and void depths. The study found that while rebar density does not significantly affect the estimated position of voids, it can reduce their visibility. Among the tested frequencies (400 MHz to 900 MHz), the 600 MHz antenna yielded the best results for void localization. Additionally, a method based on field strength difference was proposed to enhance void detection.

Wang et al. [84] proposed a machine learning approach to improve defect detection behind concrete linings. They employed a Residual Channel Attention Network (RCAN) to reduce rebar-related clutter that obscures reflections from defects, thereby improving the visibility of voids.

An important point to achieve efficient results in GPR applications is the selection of an antenna with adequate center frequency. For shallow concrete surveys, a high frequency (in the order of 2 GHz) should be used; for locating utilities on the soil, a 500 MHz antenna should be more adequate, and for deep geotechnical surveys, a lower frequency antenna (100 MHz or even lower) is recommended.

The antenna center frequency impacts signal resolution. The lateral resolution, expressed by Equation (10), depends on the wavelength ($\lambda$), which is a function of the wave frequency (f). Higher frequencies allow for a better resolution and more detailed results. Thus, smaller objects can be detected. However, there is a trade-off between higher- and lower-frequency antennas. The higher the frequency, the higher the signal attenuation and dissipation. The lower the frequency, the lower the resolution. However, a deep inspection is also possible with low resolution.

$${\displaystyle \frac{\lambda }{2}}={\displaystyle \frac{c}{2{\sqrt{{\epsilon }_R}}^f}}$$

(10)

As radar waves pass through a structure, their intensity is continuously reduced due to their reflections and as a result of the presence of good electric conductors, discontinuity points, and conductivity (heat loss). The wave frequency has a significant influence the heat loss. When the frequency or conductivity of the material increases, the penetration depth decreases [7].

Several researchers have investigated the best antenna option for each application. In general, frequencies varying from 300 MHz to 2.6 GHz are studied for concrete surveys; as in [32]. Although the higher the frequency, the better the resolution, in some cases, antennas with lower frequency have achieved better results in tunnel linings surveys. Zhang et al. [72] tested a 250 MHz, a 500 MHz, and a 1000 MHz antenna for grouting evaluation in the Shanghai Metro Tunnel and concluded that the 250 MHz had low resolution. The 1000 MHz had a shallow detecting depth, and the 500 MHz was the most suitable for this application. However, lower frequency does not always mean deeper inspection. In [85], GPR antennas with 400 MHz, 900 MHz, and 1500 MHz were used for inspection. The 400 MHz antenna presented limited penetration due to the rebar abundance.

To obtain better results during inspection with GPR, the combination of more than one frequency antenna is recommended. Alani and Tosti [33] applied a 900 MHz and a 2000 MHz antenna to detail the tunnel structure. The 2 GHz antenna was found to be able to evaluate the depth and position of rebars more precisely than the 900 MHz one, while the 900 MHz antenna was more useful than the 2 GHz for estimating the layers’ thickness. Baryshnikov et al. [86] detected inhomogeneities at depths of up to 400 mm using a 2.6 GHz antenna, and up to 1000 mm using a 900 MHz antenna.

In conclusion, GPR is an excellent equipment for tunnel inspection, if the appropriate frequency is used for the specific objective. GPR is suitable for all kinds of inspection (cadastral, routine, and special inspection), allowing fast and detailed structural inspections. The possibility of combining machine learning (ML), vehicle-mounted antenna, and different analysis techniques for concrete mapping can make GPR be the future technology for fast tunnel inspections, together with some image- and laser-based methods. A more detailed review about GPR is seen in [32,74,87].

3.6. Transient Electromagnetic Radar

The Transient Electromagnetic Radar (TER) is a new type of radar shown to be useful for tunnel lining inspection. The method is based on the difference between the electrical resistivity of materials [6] and can be used for void location, material thickness estimation [88], and grouting quality [34]. The TER method emerged to overcome some drawbacks of previously existing methods, such as low efficiency, impossibility of detecting shallow layer problems by applying TEM (Transient Electromagnetic Method), and the limitation of GPR when applied to complex underground reinforced concrete structures [34].

The TER method is based on the principle of the Transient Electromagnetic Method (TEM), a time domain method that detects variations in underground electrical characteristics. TER consists of two coils, a transmitter and a receiver coil (Figure 9). The transmitting coil sends a pulse magnetic field generating eddy currents in the ground. When the current in the transmitting coil is abruptly turned off, the eddy current does not disappear immediately, inducing a secondary magnetic field that is perceived by the receiver coil. The second magnetic field reduces the length of time; the magnetic field decreasing rate depends on the frequency of the transmitter current and the properties of the conductors (conductivity, volume, area, and depth of the conductor) [34].

Therefore, analyzing the received data is possible to obtain information about the distance, the position, and the velocity of the pulse. In this respect, the TER method may resemble the GPR method; however, they differ in the principles used. GPR is based on the dielectric properties of the materials, and TER is based on the difference between the resistivity of materials [34].

TEM is a well-established method for mineral investigation, hydrological investigation, and geological mapping. However, its application to structural quality surveys in engineering is relatively recent. Even so, the results from the literature are promising [34].

Ye and Ye [34] evaluated TER for grouting quality surveys by searching for cavities behind a shield tunnel segment. Their results are compared with the results from using a 900 MHz GPR antenna. The authors reported that, because of the TER principles, it was less influenced by the water and reinforcement of the concrete, demonstrating its superiority over GPR in these situations in this study.

Ye et al. [35] tested the TER method for quality evaluation of tunnel composite lining in a railway tunnel located in Sichuan in China and compared it with the results from using a 500 MHz GPR antenna. The results were validated with samples collected by drilling tests. TER showed good capacity for layer thickness estimation and void location and produced better results than GPR for primary and secondary lining thickness estimation (Figure 10a).

As a new method, TER is still under development, seeking improvements towards better results. Geng et al. [36] increased the TER depth detection, enhanced the weak signal, and reduced the signal–noise ratio (SNR). They obtained those improvements by reducing the turn-off time of the primary magnetic field and increasing the electromagnetic transmitter current to have deeper detections. The weak signal was improved by an algorithm designed based on the characteristics of TEM. Moreover, SNR was reduced by combining a bipolar superposition and multi-period sampling to denoise the signal. They reported that the technique showed clearer images and detection than the GPR method during a test for grouting evaluation (Figure 10b).

The TER is a new technique for tunnel layer inspections. As new equipment, it is still under development. The method needs a contact between the equipment and the tunnel surface, which makes its fast application difficult. However, the method showed good results for layer thickness estimation and void location, being an alternative for special inspections and quality control during tunnel construction. For more detailed information about the TER method, refer to [34,35,36,89].

3.7. Photogrammetry

Visual inspection used to be the major kind of inspection for tunnels. The advancement of AI and image processing has benefited computer vision techniques, making photogrammetry a good NDT method for tunnel inspection. In the literature, the technique is employed for surface defect detections, such as cracks and leakages, for 3D models construction, and cross-section deformation measurements [39].

Pathologies in tunnels are usually identified by photogrammetry using image processing and artificial intelligence (AI) techniques. These approaches allow extracting and classifying visual features, which may indicate the presence of defects, such as cracks and leaks. Some possibilities are the use of texture analysis, threshold techniques, morphological characteristics, and pattern recognition [39]. Texture is an intrinsic characteristic of the material surface and its variation can be used to identify defects in tunnel linings. For example, in [90], a method was proposed using conditional texture anisotropy and threshold technique in infrared images to improve crack identification in tunnel inspections.

The threshold technique uses morphological characteristics to separate regions and extract features of defects; for example, the crack and leakage areas have different reflectances when compared with their surroundings. An advantage of this technique is the low computational cost. However, it is highly dependent on the threshold definition, which may hinder the feature extraction in some cases. The threshold technique can be a good option for a first layer of a defect extraction system [39]. Some examples of threshold techniques used to complement the tunnel inspection system are provided in [91], which uses the threshold segmentation of Otsu [92] in an automated crack detection solution for subway tunnel inspections, and [93], which creates a system for crack and leakage recognition using the [92] solution.

The pattern recognition method is highly used in the literature for defect detection using photogrammetry. This method usually combines different image processing, morphological threshold, and texture techniques with ML for autonomous defect identification. The most popular ML techniques used in defect identification are the Support Vector Machine (SVM) and Artificial Neural Network (ANN); however, the use of deep learning (DL) techniques, such as the Convolutional Neural Network (CNN), are rapidly developing [94]. The deep learning (DL) method for crack identification usually performs three tasks: classification, detection, and segmentation (Figure 11). Some examples of pattern recognition methods are in [95], which uses a Fully Convolutional Network (FCN) for cracks and leakage identification in a subway tunnel.

Photogrammetry allows for the generation of “as-is” 3D models of the inspected structures. These three-dimensional models provide an accurate representation of the tunnel, enabling easier visualization of deformations and defects. The creation of these 3D models can be integrated into digital twin platforms, which are virtual replicas of physical assets, facilitating live simulations and analyses [96].

The use of photogrammetry for 3D modeling has the advantage of having more realistic color and texture (Figure 12). It may be employed to complement other methods used for 3D reconstruction, such as the Terrestrial Laser Scanner (TLS) [97,98,99], as it can be used alone for 3D reconstruction since there is an adequate overlap between the images. Some examples for tunnel applications are found in [100], which used a single camera to create a tunnel 3D model, and [101], which achieved a 3D model using an array of cameras and lights to acquire images of the tunnel surface.

Photogrammetry is a relatively cheap method, useful for cadastral and routine inspections. For cadastral inspection, it can document the tunnel condition with image mosaics and 3D models. For routine inspections, the method is relatively easy to use when mounted on vehicles for fast inspection. This approach has shown significant improvement in autonomous defect identification with machine learning techniques. More complete reviews about photogrammetry for tunnel deformation, 3D reconstruction, and pathology identification are found in [39,94].

3.8. Infrared Thermography

The thermal infrared method is an optical NDT method that uses infrared thermometry (IRT) to locate defects inside concrete. It benefits from the thermal radiation emitted by the surface with a spectral range outside the range visible to humans, also called infrared radiation, to detect voids, concrete delamination, and moisture. The infrared energy measured is a part of the electromagnetic spectrum and has a wavelength range between 0.75 mm and 800 mm [7].

All bodies emit electromagnetic radiation depending on their surface temperature and emissivity (ability of the material to emit a certain wavelength of radiation) [102]. The heat (electromagnetic radiation) on the tunnel surface can be transferred by conduction, convection, and radiation through the concrete segment, air inside the tunnel, and air inside the void. Heat transfer is influenced by some physical properties, such as thermal conductivity, specific heat, concrete density, and geometric characteristics [103,104]. The IRT method uses a temperature gradient to identify internal anomalies in the concrete using thermograms (images of temperature distribution); see Figure 13. For example, a void filled with water will vary the temperature in a different velocity than the concrete because of the difference in the thermal properties between them.

The infrared thermography can be applied using two different strategies, active or passive. In the active method, a source of heat is used to create the difference between the concrete temperature and air temperature. This method is appropriate for visualizing inhomogeneities and defects close to the surface, up to a 10 cm depth [105]. Some techniques for surface heating are the use of a halogen lamp [106,107], CO₂ laser heating [106], xenon arc lamp [108], and pulsed air-flow [109]. Other active IRT techniques are described in [110]. However, heating and cooling a tunnel lining is expensive and demands a long time [38], making the active technique difficult to apply to tunneling inspection. Conversely, the passive method does not require any heating or cooling system; the ambient temperature is measured in its natural value. This method is recommended to detect voids of less than 5 cm in depth [111].

The passive IRT method has potential for concrete inspection since its surface has a significant daily temperature variation. It has been mostly applied to building facades [112,113,114] and concrete bridge inspections [102,115]. ASTM International [116] describes a practice for applying IRT to bridge deck inspection, considering situations with debonded or delaminated concrete if the temperature differs by 0.5 °C (0.9 °F) on sunny days.

Although IRT has great potential for large concrete structure inspection [117], the need for temperature variation hinders its application to tunnel inspection. However, the recent improvements in IRT equipment optimize their detection capacity, making the method useful even when applied to tunnels with low-temperature variation. According to [23], the infrared thermography technique reached a sensitivity of 0.1 °C. Konishi et al. [38] reported the method capacity to detect the defects when the temperature between healthy concrete and unhealthy concrete is over 0.03 °C.

Several application cases of the IRT method in tunnels can be found in the literature. In [38,118], an IRT passive technique was used to detect defects in cut-and-cover and shielded tunnels. Han et al. [119] combined IRT and visual–optical color cameras to create a multispectral method for tunnel leakage detection using Convolutional Neural Networks. In [120], an infrared-based vehicle for an autonomous system for detecting seepage points in cable tunnels was used. Yu et al. [90] used infrared images to suggest a method to detect cracks on tunnel lining surfaces. The method intends to improve some issues of applying image-based methods to tunnels, such as low contrast and illumination.

Afshani et al. [121] used the 24 h variation in temperature in tunnels to simulate and to study the temperature degree in concrete with and without defects. The results showed that unhealthy concrete presented a higher temperature variation compared with the healthier one with temperature difference of less than 0.55 °C between them. Two cases were compared: case 1, the air temperature (T) higher than the concrete surface temperature (t2), a heat absorption case; case 2, T less than t2, a heat radiation case. For case 1, the detection rate of defects was approximately 40%, even for higher temperature differences. For case 2, the higher difference in temperature led to a higher detection rate, approximately 76%. Behbahani et al. [122] tested thermography for tiled tunnel inspections using equipment with a 0.1 °C resolution. They concluded that the method was effective in detecting defects during both heating and cooling cycles, but its performance was limited during the thermal equilibrium phase.

Konishi et al. [38] compared the passive IRT method with the hammering test to detect defects in a cut-and-cover and shield tunnel. They reported the need for a suitable temperature condition between the concrete and the air to create a heat radiation condition. Higher rates of void detection are achieved when the temperature variation exceeds 0.35 °C.

IRT has several advantages, such as fast application, non-contact, non-interaction, real-time measurements, automation and coverage speed over large areas [37], and being automatized. These equipment improvements make IRT an interesting methodology for tunnel inspection, suiting routine and special inspections.

3.9. Multispectral

According to [7], multispectral analysis is an optical NDT method that uses photos of the surface to identify anomalies by using a camera similar to color photography. The multispectral analysis differs from common color photography in that the light spectrum is not completely registered in the same image. The light spectrum goes through different filters that separate it into different wavelength ranges. The merge of all filters completes the entire light spectrum. The method has the advantage of being able to detect surface areas where there are changes imperceptible to the human eye.

For tunnel inspection, the method can be helpful to identify cracks and small cracks, deposits of carbonate and moist patches [7,119]. Unlike other popular methods used for tunnel inspection, such as GPR and thermography, this method can detect dry small cracks with a width of 0.3 mm [123]. Moreover, Huang et al. [124] used the hyperspectral reflectance of concrete to propose a method for classifying concrete strength using machine learning. The authors classified the concrete into strength grades C20, C30, C40, C50, and C60 based on specific wavelength ranges.

With the advances in Convolutional Neural Networks (CNN), multispectral detection is a gradually emerging technology. One promising possibility is to merge thermal cameras with common color cameras in CNN-based multispectral analysis [119]. Some advancements in that regard are presented in [125,126,127]. Although these studies are not focused on applications to structural inspection, the technique can be explored for this purpose, as in [119]. The authors use the merged thermographic and common color cameras in multispectral CNN-based analysis to improve water leakage detection in metro tunnels.

3.10. Laser Scanning

Laser scanning devices have several applications to engineering measurements. In general, these types of equipment use the LiDAR (Light Detection and Ranging) technique to obtain the distance of the object. The laser system sends a beam of light (electromagnetic radiation) that is reflected by a surface and received back. The system registers the travel time, the direction and angle of the light ($cos\left(\alpha \right)$, $cos\left(\beta \right)$, $cos\left(\gamma \right)$), and the reflectivity of the surface [128]. The distance of the reflector ($d_r$) can be calculated with Equation (7) considering $V_M=c$ (speed of light) and the relative position ($x,y,z$) of the reflective point can be estimated by Equation (11).

$$\left\{\begin{array}{c}x=d_{r}cos\left(\alpha \right)\hfill \\ y=d_{r}cos\left(\beta \right)\hfill \\ z=d_{r}cos\left(\gamma \right)\hfill \end{array}\right\}$$

(11)

Laser scanning is a well-established method in tunnel management, being useful even during the tunnel construction for quality surveys [129,130] and displacement monitoring [131,132], as well as for inspections during the operational phase.

The capacity of transmitting and collecting laser pulses reflected by physical objects in space, and recording the location of the reflection points in space, allows for precise modeling [128], being highly indicated for quality survey of tunnels under construction. As an example, [130] suggested a contour quality index to evaluate the tunnel excavation using laser profilers, whereas [129] analyzed the application of Terrestrial Laser Scanner (TLS) to drill and blast tunnels and created a precise “as-built” from the constructing tunnel, with the rock surface and final liner surface model. With the model, one can evaluate overbreaks in the excavated section, create a block model of the rock mass, help with the rock geological characterization, and estimate the shotcrete thickness of the final layer.

The laser scanning approach allows for consecutive measurements of the same location, enabling deformation estimation over time to detect geometric changes in the captured scene. There are different strategies in the literature to identify the changes in shape using laser scanning data, such as the point-to-point-based change detection, the line-to-line-based change, point-to-surface, and the surface-to-surface-based changed detection [131,133]. Some examples of application to tunnels are seen in [132], which suggested an algorithm for shield tunnel deformation monitoring based on TLS, and [134], which suggested a real 3D approach to detect deformation using Point Cloud Density (PCD).

One advantage of laser scanning for tunnel inspection is that it is an active sensing method, which means that it does not need to have an external light source to capture the data, as photogrammetry does, thus being useful for pathology mapping. For instance, if the equipment records not only the 3D geometric point cloud, but also the intensity of the reflected laser pulse, as the laser scanner operates in the near-infrared spectrum, the water absorbs the emitted light, resulting in a lower intensity value in leakage areas [135]. This characteristic allows equipment such as the TLS to be applied to routine inspections for leakage mapping as in [135,136,137,138,139].

Several studies also investigate the use of laser scanning to identify and to quantify cracks in civil engineering infrastructures [140,141]. The 3D point cloud detection depends on several parameters, such as angle of incidence, scan range, resolution of the equipment, illumination, surface roughness, and color [141]; then, the correct equipment configuration may improve the results. In the field of tunnel inspection, the LiDAR technique also had been tested for crack location. In [142], the TLS is employed together with photogrammetry techniques to identify cracks, water leakage, and deformation in tunnel structures. Xu and Yang [143] suggested an automated method for crack detection combining Gaussian filtering with signal-to-noise ratio analysis to increase the crack identification in 3D point cloud data.

Laser scanning methods may be used for several tunnel applications, from its construction for “as-is” documentation, quality survey and displacement monitoring, to the operational phase for structural health monitoring. The method may be employed for cadastral, routine, and special inspection for specific reasons.

3.11. Laser-Vibrometer Method

Most methods employed to detect detachment of concrete, such as the hammering test, Impact-Echo, and ultrasonic tomography require contact, including human risk to inspect damaged surface and hindering their automation. A promising alternative to improve detachment detection is using a Laser Vibrometer to inspect the area from a distance. This method has been tested in laboratory conditions and for applications to real structures, such as tunnels [144], bridges [145], and buildings [146].

In general, the Laser-Vibrometer technique uses the dynamic response of flawed concrete subjected to an excitation to detect concrete detachment. The dynamic response of excited flawed concrete structures is governed by stress wave propagation. Two different vibrations can be generated: the thickness vibration in sound concrete and plate-like flexural vibration in the presence of shallow defects. The difference in vibrations determines the characteristic displacement and velocity time histories on the surface, which allows for detecting defects [42]. Thickness vibration is explained in the Impact-Echo method; see Figure 1.

In the case of shallow defects (Figure 1c), the solid portion above the crack may respond as a plate-like structure, showing flexural vibrations. Considering a homogeneous, isotropic, and linear-elastic material, this case can be analyzed by Equation (12), considering a circular plate with 2a diameter and thickness $d_r$. $T_{f}x$ is the first natural period, $f_{f}x$ the corresponding frequency, and $k_1=$ 4.99 for simply supported boundary conditions [42].

$$T_{fx}={\displaystyle \frac{1}{f_{fx}}}={\displaystyle \frac{2\pi }{k_1}}·\sqrt{{\displaystyle \frac{12(1-{\nu }^2)\rho a^4}{E{d}_r^2}}}$$

(12)

To measure the vibration from a distance, a Laser Doppler Vibrometer is used to measure velocity components in an excited area. To generate the excitation from a distance, there are some different strategies in the literature, such as air shock wave, airborne sound, and pulsed laser methods.

Mori et al. [42] proposed a method using a compressed air shock tube to induce an impact on the concrete surface and a Laser Vibrometer with a sampling frequency of 500kHz to measure the velocity components, besides detecting the thickness vibration and the flexural vibration when there is a flaw in the concrete.

Akamatsu et al. [147], Akamatsu et al. [29], and Akamatsu et al. [148] proposed a method for concrete structure inspection using a high-power directional sound source (LRAD 300X) and a scanning Laser Doppler Vibrometer of Polytec PSV-400-H4. They reported a capacity of inspection at 10 m from the surface. Katakura et al. [149] evaluated the detection capacity of the method and reported a similar capacity of the hammering test. Using the same system, Sugimoto et al. [146] proposed an algorithm for non-contact acoustic inspection using spectrum entropy. The method was tested experimentally and in a concrete building. Using the same sound source, Sugimoto and Sugimoto [150] proposed using Normalized Spectral Entropy to detect internal defects and Normalized Spatial Entropy to measure the resonance frequency band of those defects. Using a scanning Laser Doppler Vibrometer (PSV-500Xtra), the author investigated voids of 200mm in diameter at a 60mm depth. Sugimoto et al. [30] applied the method to bridges using LRAD300X and PSV-500Xtra.

The method using high-power directional sound as an excitation source and Laser Doppler Vibrometer for data acquisition have been already applied to bridge inspection at a distance of 30 m, and in a tunnel with shotcrete structures from a distance of 7 m (Figure 14) [30,145,151].

Another method to induce vibration in the concrete is the application of high-power lasers. Yasuda et al. [152] suggested an inspection method using two different types of lasers. Firstly, a high-power pulsed Laser Ablation (Nd:YAG laser) was used to reproduce an impact excitation and induce vibration in the material, followed by a Laser Doppler Vibrometer (RSV-150, Polytec, Waldbronn, Germany) to measure the vibration velocity. The experimental investigation allowed for the observation that the damped oscillation caused by flexural vibration could be used as a defect indication. Yasuda et al. [144] tested the method for tunnel inspection (Figure 15). The remote laser sensing equipment was used in a region with corrosion of the reinforcement in the Shinkansen tunnel. The results showed that the method can be applied to tunnel inspection. However, the field test presented higher noise level and a damping ratio different from the one obtained in the laboratory experiment. Wakata et al. [153] suggested a methodology using a Laser-induced plasma for surface excitation and interpretation of a Rayleigh wave to detect concrete defects. They reported the advantage of this non-destructive method over the Laser Ablation method, which creates a small crater in the contact area, and also reported an advantage of detecting flaws with fewer measurement point using the Rayleigh wave.

The employment of Laser Vibrometer for concrete inspection is still under development. However, the literature reveals good results for this technique (as in [144,145]). Since it may be applied from a distance, the technique seems to be an excellent alternative for special inspections, thus reducing human risk, especially when investigating pathological manifestation at high places. Although it is a contactless technique, there is no information about the velocity of application that is important for routine inspections.

3.12. Automation Inspection Procedures

The current advancements in automation, machine learning, and computer vision aim to reduce human subjectivity in the context of inspections [154]. This trend has been evident for over a decade in tunnels, encompassing both data processing [155,156] and the development of new equipment for data acquisition [95,157]. The demand for more reliable data and effective management of very important infrastructures drive increasing investments in autonomous processes.

An innovative crack inspection system for concrete walls was proposed by [158]. It used a camera to automatically maintain a constant distance from the walls and measure the cracks. The effectiveness of this system was confirmed through laboratory and field tests, providing objective data for safety evaluation. This advancement aims to overcome the limitations of manual inspection, offering great efficiency and objectivity in assessing the integrity of concrete structures.

Victores et al. [159] employed a Robot-Aided Tunnel Inspection and Maintenance system with sensors and scripts for the inspection and maintenance of concrete surfaces in tunnels, minimizing operational interferences. A laser telemeter enhanced precision in the inspection, surface preparation, and compound application stages. An API was coded to assist the operator in crack identification, while the procedures are executed automatically by the system.

Menendez et al. [160] delineated the ROBO-SPECT project, a robotic system comprising a mobile vehicle and a high-precision mechanical arm. It utilized computer vision to detect defects in the tunnel and measured cracks using ultrasonic sensors. The automated crane is able to navigate autonomously within the tunnel, and generate collision-avoidance trajectory plans by employing 3D point cloud mapping. The data can be processed by a semi-supervised computer vision system.

Wei Huang et al. [95] employed an image acquisition tool called Moving Tunnel Inspection (MTI-200a), which conducts continuous scanning to create an information database (Figure 16). Using deep learning tools, models of Convolutional Neural Networks (Fully Convolutional Network, FCN) are trained for semantic segmentation of deteriorations related to cracks and infiltrations in tunnels.

Jiang et al. [161] proposed a system capable of extracting and evaluating defects in tunnels using image acquisition. An automated coating identification software that calculates average luminosity eliminates obstructions, detects crack contours, and removes unwanted lines is used. This software facilitates crack identification, thereby enhancing efficiency in tunnel inspections.

Automation is closely tied to the advancement of contemporary technologies, such as machine learning tools. Zhao et al. [162] introduced a method for identifying moisture spots in tunnel structures with a mask-based Convolutional Neural Network algorithm (Mask R-CNN). For algorithm training, over five thousand leakage images were utilized. The proposed method comprised three detailed steps for instance segmentation: feature extraction, region proposal generation, and moisture spot identification.

These are only a few examples of automation in tunnel inspection. There are several cases in the literature that combine robotic or vehicle-mounted NDT methods and machine learning techniques to automate the inspection. These developments are important to have more effective inspection procedures and thus reduce inspection subjectivity.

4. Summary of NDT Methods

Numerous types of NDT equipment are available, each adapted for specific applications, offering distinct advantages and presenting unique challenges in their applications. Intending to help the decision to apply NDT methods,

Table 2. NDT method applications.

Applications		GPR	Ultrasonic Tomograph	IRT	Impact-Echo	Multispectral	Laser Vibrometer	TER	Photogrammetry	Laser Scanner
Inspection Type	Cadastral	o	o	o	o	o	o	o	o	o
	Routine	o		o		o	•	o	o	o
	Special	o	o	o	o	o	o	o
	Surface			o		o			o	o
	Interior	o	o	o	o		o	o
Element Estimation	Layer thickness	o	o		o			o
	Steel rebar	o	o
	Grouting evaluation	o	o		o			o
	Shotcrete bonding	•	o		o		o	o
	Surface geometry	o							o	o
	Volume reconstruction	o	o					•
Pathologies	Concrete state	o	o	•	o	o			o
	Rebar corrosion	o		•		•			•	•
	Void localization	o	o	o	o		o	o
	Water leakage	o		o		o			o	o
	Detachment	o	o	o	o		o	o	•	•
	Cracks	•	o	o	o	o			o	•
	Small crack	•				o			o
	Displacement								o	o

∘ The method has good fitting with this application. • It is not the best alternative; however, the method workability principles indicate that it may be applied with some improvements, or in a specific situation. Note: Some methods may present some issues to be applied in tunnels with shotcrete lining.

shows the best fitting of each considering the literature review. The applications are separated into three categories (inspection type, element estimation, and pathologies). For example, although the table lists the best fitting for each method, this does not mean that a method cannot be used for a different application. For example, although TLS is not listed as good equipment for special inspection, it could be applied to special inspection in case of anomalous deformation.

The inspection types of Table 1 consider five distinct applications. Cadastral is the initial inspection and, to best represent the real condition of the structural system at the beginning of a monitoring program, the application of all methods may be useful. Although some methods have lower velocity rates, they can be used for localized investigations.

For routine inspection, it is better to use methods with fast application rates, methods easy to automate, or vehicle-based equipment. GPR and image-based techniques, such as infrared and photogrammetry, are methods already used on vehicles. TLS can be combined with VLS (Vehicle Laser Scanner), a fast option for routine inspections. Some normatives and manuals also specify detailed inspection that consists of periodic activity as the routine inspection, but with a higher scope; and thus, the recommendation remains the same as for the routine inspection.

The special inspection is for specific investigations. Photogrammetry and TLS are not considered in this group because they are superficial methods that may be used to indicate a special inspection. Finally, the surface and interior inspection types separate the methods that can collect from inner information of the structure to the method that only collects superficial data.

The element estimation class separates methods to be used to locate or estimate structural elements and investigate its properties. For the pathologies class, the method with the ability to detect those defects is listed and classified.

Finally, Figure 17 shows a flowchart that shows a decision map of each kind of inspection. In this workflow, element estimation is desirable to be performed and documented in the cadastral inspection, which is the beginning of tunnel administration, since this information may be useful for future structure evaluation in all the phases of the tunnel life cycle. All the other capacities of the NDT methods (surface and interior inspection, pathology location) are recommended for all other inspection types.