Intelligent GPR Detection of Backfill Grouting Quality and Adjacent Soil Cavities in Shield Tunnels: A Case Study

Abstract

During the construction of shield tunnels, backfill grouting behind segmental linings is a crucial technique for filling the shield tail gap resulting from over-excavation by the cutter head. Direct observation of grouting thickness distribution is challenging, necessitating nondestructive testing (NDT) technologies like ground-penetrating radar (GPR). This study introduces a loaded-to-frame (LTF) device designed to automate the collection and intelligent analysis of GPR data, enabling rapid and intelligent detection of grouting thickness distribution and adjacent soil cavities in the excavation area. Focusing on the complex geological and environmental conditions of Xiamen Metro Line 3, the research highlights the critical role of grouting quality and cavity detection in ensuring tunnel construction safety and surface stability. Time domain reflectometry was employed to assess the electrical properties of the grouting materials, revealing a 12-h relative dielectric constant of 24.08 and conductivity of 4.29 mS/m. These properties significantly differ from those of the surrounding soil, confirming the suitability of GPR for detection. The LTF device, combined with its intelligent analysis system, can complete automated and efficient detection of single-ring shield tunnels within 5 min, providing dynamic feedback on grouting quality control. Furthermore, the study validates the feasibility of using the LTF device to detect adjacent soil cavities during tunnel excavation. This research advances the automation and intelligence of backfill grouting and cavity detection, supporting the safe and efficient progress of shield tunnel construction.

This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.

1. Introduction

In recent years, with the development of urbanization in China, the urban population has been increasing steadily, leading to a rapid growth in demand for rail transit such as metros [1]. Because metro construction often traverses urban underground areas, the requirements for controlling construction disturbances are higher compared to mountain tunnels using the drill and blast method [2, 3]. The shield tunneling method is commonly employed in metro construction [4, 5], significantly reducing strata disturbance while traversing complex urban surface structures and underground infrastructure. The excavation by the shield machine’s cutter head often results in an actual excavation diameter more significant than the tunnel’s designed diameter, creating a shield tail gap between the segment and soil. In actual engineering practice, backfill grouting behind the shield tunnel segment is typically carried out by injecting grout into the grouting pipes at the shield tail [6]. The assembly of prefabricated tunnel segments and backfill grouting constitutes the structural support of shield tunnels, ultimately forming a stable tri-layer composite structure of segment, grout, and soil. The quality of backfill grouting is crucial for controlling strata losses and segment misalignment [7–9]. Insufficient backfill grouting may lead to uncontrolled strata disturbance, resulting in excessive ground loss rates. Additionally, if cavities are present near the excavation area, the disturbances from the excavation are likely to accelerate their expansion, leading to rapid ground settlement.

Insufficient backfill grouting and adjacent cavities in the soil body can lead to engineering problems, including adverse effects on the tunnel under construction and its crossing environment. Figure 1a and b illustrates that for the tunnel structure itself, issues with grouting quality and adjacent soil cavities may lead to uneven stress distribution on the segment structure, causing misalignments, fractures, and other issues, which in turn can lead to problems such as leakage. For the safety of the crossing environment, if strata loss is not controlled promptly and effectively, it may lead to excessive subsidence of the strata, potentially triggering surface collapse. Figure 1c illustrates some reported incidents of surface collapse posing a serious threat to the safety of surface structures and roads, resulting in significant economic losses and social impacts. It is crucial to understand the presence and location of natural cavities in the strata near the tunneling area and to take timely monitoring. This can effectively prevent the development of these cavities due to construction disturbances, thereby avoiding secondary disasters.

[figure(s) omitted; refer to PDF]

Controlling strata losses [10–12] and ensuring structural safety during shield tunnel construction is crucial. Understanding the distribution of backfill grouting and the adjacent soil cavities is of significant engineering importance. However, since the backfill grouting process occurs behind the shield and tunnel segments, direct observation of the grouting formation and its final quality is challenging, complicating the control of grouting quality. Traditional core sampling methods risk damaging the tunnel lining structure and fail to meet the efficiency requirements for detecting insufficient grouting and adjacent soil cavities. Over the past years, with the continuous development of nondestructive testing (NDT) technology [10], the use of ground-penetrating radar (GPR) and other NDT techniques for detecting defects in tunnel linings has been widely welcomed in the engineering field [11–13]. GPR, as a NDT technology based on electromagnetic waves, offers advantages such as ease of detection and the absence of the need for sensor installation. It has found widespread application in tunnel engineering [14–17]. For example, Ma et al. [18] optimized detection parameters based on the electrical parameters of different media, such as segments and grout, under rich water-rounded gravel geological conditions, effectively solving the problem of overlapping multiple waves of segments and reflected waves from grouting layers. Jianying [19] researched imaging laws using different antenna polarization methods and concealed defect aspects in shield tunneling. Li et al. [20] analyzed the imaging effects and mechanisms of dual-liquid grouting behind super-large-diameter shield tunnels.

However, traditional GPR applications typically rely on manual detection data processing, which is time-consuming in terms of data collection and processing, resulting in low detection efficiency. Moreover, data processing often depends on the skills and engineering experience of the operator, making it challenging to achieve quantitative, accurate, and objective analysis results. With the continuous development of automation and artificial intelligence [12, 21, 22], the development of automated data acquisition equipment and intelligent rapid data analysis systems for backfill grouting detection has become possible. Although artificial intelligence has been widely studied in GPR data analysis, reports on its mature applications remain scarce. Additionally, there is limited research on using GPR to detect adjacent soil cavities within tunnels. Few studies have addressed detection equipment and solutions integrating backfill grouting quality inspection and adjacent soil cavity detection.

To ensure the structural integrity of tunnels and minimize environmental impact, this study introduces an automated inspection system, the loaded-to-frame (LTF) system, designed for intelligent evaluation of backfill grouting quality and detection of adjacent cavities during shield tunneling operations. The proposed system addresses significant limitations inherent in conventional GPR tunnel inspection methods, particularly their operational inefficiency, substantial subjectivity, and limited intelligent application capabilities. This research employs a case study from the Xiamen Metro Line 3 project, specifically focusing on the engineering complexities encountered in the Xiamen University South Gate Station to Xiamen University Baicheng Station section (X–X section). The study demonstrates the LTF system’s efficacy in assessing grouting quality and identifying surrounding soil cavities through meticulous analysis of the section’s intricate geological conditions and underground environment, providing critical technical insights for analogous underground engineering projects. This investigation has developed and implemented a dynamic feedback mechanism that effectively integrates intelligent inspection data with practical engineering applications. This integration significantly enhances operational efficiency and decision-making processes in shield tunnel construction. The research outcomes substantiate the system’s practical applicability and establish a robust framework for achieving automated, safe, and multifunctional shield tunneling operations.

2. Research Background

2.1. Project Overview

As depicted in Figure 2, the southern extension of Xiamen Metro Line 3 spans 8.5 km from Shapowei Station in the south to Xiamen Railway Station in the north, encompassing five underground stations and six subterranean sections. Xiamen University Baicheng Station is an interchange hub with the planned Line 7. The inter-station distances vary significantly, with a maximum of 3.733 km, a minimum of 0.765 km, and an average of 1.702 km. This study concentrates on the X–X section of the southern extension, which aligns in a northwest-southeast orientation. This section commences at Xiamen University Nanmen Station, proceeds along Daxue Road, traverses beneath the Baicheng Viaduct, extends along Huandao South Road, and culminates at Xiamen University Baicheng Station, located north of Hulishan Battery. The tunnel depth within the X–X section fluctuates between 7.78 and 18.81 m, navigating through a diverse geological profile that includes silty clay, silty clay interspersed with silt, silty clay mixed with mud, completely weathered granite, highly weathered granite in both loose and fragmented states, and moderately weathered granite. Constructing both the left and right tunnels employs the shield tunneling method, with respective lengths of 1738.182 and 1748.057 m.

[figure(s) omitted; refer to PDF]

The study area primarily comprises two dominant geomorphic units: coastal plains and residual tablelands, with localized occurrences of low hills and ridges. Extensive urban development and land reclamation activities have significantly modified the original topography, resulting in subdued terrain variations across the region. The area exhibits generally gentle to moderate slopes, with ground surface elevations between 3.80 and 12.41 m above sea level.

The X–X section exhibits a complex upper stratigraphic profile with significant soil and rock composition heterogeneity, spatial distribution, and engineering properties. This section traverses multiple geological formations with varying burial depths (7.97–18.81 m), thicknesses, and material characteristics. As shown in Figure 3, the stratigraphic sequence consists of silty clay and its variants (with silt and mud interbeds) and a weathering profile of granitic and dioritic rocks. Specifically, the geological formations include: (1) silty clay; (2) silty clay with silt intercalations; (3) silty clay with mud interbeds; (4) utterly weathered granite; (5) highly weathered granite in both loose and fragmented states; (6) moderately weathered granite; and (7) moderately weathered diorite. This diverse geological composition, coupled with the substantial variation in overburden depth (exceeding 10 m differential), presents significant engineering challenges for tunnel construction and ground stability management.

[figure(s) omitted; refer to PDF]

2.2. Complex Transverse Environments and the Necessity of Stratum Settlement Control

The tunnel alignment traverses a complex urban environment comprising residential buildings, recreational facilities, and critical infrastructure. As depicted in Figure 4a, the tunnel passes beneath the Shuyou Restaurant, a two-storey brick–concrete composite structure, maintaining a minimum vertical clearance of 17.25 m. The geological profile in this section consists of (from top to bottom): (1) loose-fill soil, (2) medium-coarse sand, (3) fragmented highly weathered granite, and (4) moderately weathered granite. During the excavation process beneath the Shuyou Restaurant, the shield tunneling may induce ground disturbances, resulting in subsidence of the building foundation, and potential cracking of the hotel walls. Consequently, there is a significant construction safety risk.

[figure(s) omitted; refer to PDF]

As illustrated in Figure 4b, the tunnel alignment passes beneath Yanwu Bridge, a double-span reinforced concrete structure supported by bored pile foundations with a diameter of 2.2 m. The tunnel maintains a critical minimum horizontal clearance of ~1.39 m from the bridge pile foundations. The geological profile in this section comprises a complex stratigraphic sequence, including (from top to bottom): (1) loose fill soil, (2) residual sandy clay, (3) fragmented highly weathered granite, (4) moderately weathered diorite, and (5) moderately weathered granite. The shield tunneling operations in this critical zone pose substantial engineering challenges due to the potential for ground disturbance-induced pile displacement. This risk is particularly acute given the shallow overburden and proximity to the bridge foundations, which could compromise traffic safety and construction integrity. The interaction between tunneling-induced ground movements and the existing bridge infrastructure necessitates careful monitoring and mitigation strategies to ensure structural stability throughout the excavation process.

Furthermore, the section is characterized by bedrock protrusions predominantly composed of weathered granite, exhibiting substantial variations in uniaxial compressive strength ranging from 44 to 141 MPa. These geotechnical conditions present significant challenges for shield tunneling operations, as the increasing strength coefficient of rock layers in larger geological formations accelerates disc cutter wear rates. The operational risks are particularly pronounced when cutter wear exceeds design thresholds or when the shield machine’s excavation diameter becomes insufficient relative to the geological conditions. Under such circumstances, disc cutters may become lodged within rock formations due to spatial constraints, leading to potentially hazardous situations. Additionally, the high-frequency vibrations generated during excavation can induce the loosening of cutter mounting bolts, compromising cutter stability. In severe cases, dislodged cutters may fall onto the rapidly rotating screw conveyor, significantly elevating excavation risks and potentially causing extensive damage to tunneling equipment.

The complex geological and structural environment encountered during tunneling operations imposes stringent requirements on grout injection quality throughout the excavation process. This necessity stems from multiple critical factors: (1) the need to maintain ground stability in varied geological conditions, (2) the requirement to minimize surface settlement in urban areas, and (3) the imperative to ensure structural integrity when tunneling in proximity to existing buildings and infrastructure.

2.3. Grouting Process and Parameters

2.3.1. Grout Mixture Ratio and Grouting Parameters

Synchronous backfill grouting is an essential measure to prevent ground subsidence. Control of synchronous grouting involves regulating both the grout volume and grouting pressure. The grout mixture comprises commonly available commercial mortar, categorized as single-component grout. Its composition parameters are detailed in Table 1, primarily consisting of lime, sand, fly ash, water, bentonite, and additives. Level II sand refers to sand with a nonuniform particle size distribution but relatively high strength. It is commonly used as an abrasive and filter material in industrial production processes. The particle size range of Level II sand is broad, typically between 0.15 and 4.75 mm, with a certain proportion of finer and coarser particles. The grout possesses high density and shear strength, and its properties have been determined in laboratory tests: water loss rate ≤12%, slump value between 12 and 14 cm, and specific gravity of the grout mixture ≥1.9.

Table 1

Grout mixture ratio.

Materials	Composition	Standard
Lime	100	—
Sand	900	Fine sand
Coal ash	400	Level II
Water	340	Piped water
Bentonite	50	Sodium Bentonite
Additives	Add on demand	—

The additives in the backfill grouting of shield tunnels usually include polymer thickeners, cementing agents, anti-seepage agents, waterproofing agents, etc. The functions of these additives include increasing the viscosity and cementing properties of the grouting fluid, improving the effect of the grouting, preventing the loosening and collapsing of the geotechnical body, reinforcing the geotechnical structure around the tunnel, and protecting the safety and stability of the tunnel.

Synchronous backfill grouting is conducted concurrently with tunnel excavation. Considering that the grouting pressure should be 1.3–1.5 times the static earth pressure, the pressure should be controlled within 0.25–0.3 MPa. The single-casing grouting volume can be calculated using Equations (1) and (2), ~6.68 m³, with dynamic adjustments based on specific geological conditions.$$\begin{array}{}\text{(1)}& Q=V\cdot \lambda ,\end{array}$$$$\begin{array}{}\text{(2)}& V=\frac{\pi D^2-d^{2}L}{4}.\end{array}$$

$Q$ represents the volume of injected grout, measured in cubic meters (m³); $\lambda$ denotes the grout filling ratio, with a value of 1.6 used for the initial formation and a range of 1.3–1.6 for regular formations; $V$ signifies the theoretical volume of shield annular clearance, where D and d represent the diameters of the cutterhead and the outer diameter of the segment respectively, with values of 6.48 and 6.2 m, respectively in this context; L stands for the width of a single ring segment, set at 1.5 m.

The secondary grouting volume is analyzed based on geological conditions and grouting records to assess the grouting effectiveness. Monitoring data is integrated, and grouting pressure is controlled accordingly.

2.3.2. Physical Parameters of the Grout

The physical parameters of the grout mainly determine its cohesion, flow, and strength characteristics, which influence the filling capacity and strength of the grouting body. Some of the physical parameters of the grout were tested, as shown in Figure 5.

[figure(s) omitted; refer to PDF]

The key physical and mechanical performance parameters of the grout used in the X–X section are shown in Table 2. The gel time of the grout should be less than 4 min, the 1-day compressive strength should reach 0.3 MPa, and the 28-day strength should reach 4.5 MPa. In practical applications, different batches of grout materials exhibit certain variations, making it difficult to quantify these parameters with a single precise value. Therefore, the table specifies the important parameter ranges, which are determined based on engineering requirements.

Table 2

Grouting performance parameters.

Consistency (cm)	Specific gravity (g) cm−3	Stone rate (%)	Gelation time (min)	1-d compressive strength (MPa)	28-d compressive strength (MPa)
12.5–13	1.43–1.55	>97	<4	>0.3	>4.5

2.3.3. Electrical Parameters of the Grout

The main electrical parameters of the materials include relative dielectric constant ($\epsilon$), conductivity, and relative magnetic permeability ($\mu$). Since grout is a non-magnetic material, its relative magnetic permeability is 1. The grout’s relative dielectric constant and conductivity are measured through time domain reflectometry (TDR) [24–27]. The basic principle of TDR involves transmitting a signal along a transmission path, whereupon encountering an impedance change, a portion of the signal is reflected. At the same time, the remainder continues along the transmission path. TDR calculates the impedance variation by measuring the voltage amplitude of the reflected wave, thereby determining information such as the moisture content of the medium. The relative dielectric constant of the medium is subsequently calculated through the TOPP formula, as displayed in Equation (3).$$\begin{array}{}\text{(3)}& \text{VWC}=-5.3\times {10}^{-2}+2.92\times {10}^{-2}\epsilon -5.5\times {10}^{-4}\times {\epsilon }^2+4.3\times {10}^{-6}\times {\epsilon }^3,\end{array}$$where $\epsilon$ represents the relative dielectric constant and VWC represents the volumetric water content.

As depicted in Figure 6a, this study employed TDR probes to conduct electrical parameter measurements on the grout used in the X–X section. The measurement equipment comprises TDR probes, a host machine, and a data reading module. The data reading module is a smartphone equipped with Bluetooth capability, enabling rapid retrieval and recording of data, greatly facilitating parameter acquisition.

[figure(s) omitted; refer to PDF]

This study tests the relative dielectric and conductivity of the single-liquid grout 12 h after mixing, with a testing interval of 1 h. The test results, as shown in Figure 6b, indicate that the relative permittivity of the serum exhibits some fluctuations within the first 2 h post-mixing, followed by stabilization around an average value of ~24.08. The conductivity of the grout stabilizes throughout the 12 h, with an average value of 4.29 mS/m. The observed patterns suggest no significant variation in the electrical parameters of the grout within the first 12 h after injection, indicating relative stability in these parameters.

Building upon this, the study integrates existing research and engineering experience to determine the electrical parameters of media closely adjacent to the grout material, such as tunnel segments, grout, and air, as shown in Table 3. From the table, it is evident that there are significant differences in the electrical parameters of the four media (air, tunnel segments, grout, and soil) in the application scenario of this study. This suggests the feasibility of using GPR to detect the interfaces of the backfill grouting layer.

Table 3

Electrical parameters of the detection materials.

Materials	Relative dielectric constant (–)	Conductivity mS/m−1	Relative permeability (–)
Grout	24.08	4.29	1
Soil	7.23	3.75	1
Concrete [28]	6	(—)	1
Rebar [29]	1.45	9.93 × 10−6	0
Air [30]	1	0	1

3. Development of Intelligent LTF Detection Equipment and Data Intelligence Analysis

3.1. Principle of GPR

The main principle of GPR for grouting detection behind shield tunnel segments is illustrated in Figure 7. GPR consists of a transmitting antenna (T), a receiving antenna (R), and a data acquisition system. The transmitting antenna (T) generates electromagnetic waves with specific frequencies. These waves propagate into deeper layers, such as tunnel segments, grouting material, and surrounding soil, undergoing reflection, transmission, and attenuation during propagation. Reflections of electromagnetic waves occur at interfaces with different electrical parameters, such as the interfaces between segments and the grouting layer, the grouting layer and soil, and the cavity and grouting layer. When the incident waves pass through the detection medium and are captured by the receiving antenna (R), the positions of different reflection waveforms in the time-domain waveform are analyzed to determine the location and size of the detected targets.

[figure(s) omitted; refer to PDF]

The propagation of GPR waveforms in a medium follows Maxwell’s equations [32] and its capability to detect targets is determined by the electrical parameters of the two media at the interface of the detection medium. Specifically, it is mainly determined by the reflection coefficient, as shown in Equation (4). This indicates that the more significant the difference in relative dielectric constants between adjacent media, the greater the value of the reflection coefficient and the stronger the reflected waveform on the reflection surface of the GPR. Due to significant differences in the electrical parameters between tunnel segments, grout, and soil, it is feasible to use GPR to detect the interface between grouting behind shield tunnel segments and tunnel segments or soil.$$\begin{array}{}\text{(4)}& C_{\mathrm{r}}=\frac{\sqrt{{\epsilon }_1}-\sqrt{{\epsilon }_2}}{\sqrt{{\epsilon }_1}+\sqrt{{\epsilon }_2}},\end{array}$$where $C_{\mathrm{r}}$ is the reflected coefficient and ${\epsilon }_1$ and ${\epsilon }_2$ represent the relative dielectric constant at the interface boundaries. According to Equation (4), it is evident that when the relative dielectric constants of two mediums differ more significantly, the reflection coefficient increases, resulting in a greater reflection intensity of the GPR wave at that point.

3.2. Development of LTF Detection Equipment

3.2.1. Hardware Composition

To achieve automation and rapid data analysis of backfill grouting detection behind the shield, this study employs an LTF detection system [33], as illustrated in Figure 8, to inspect the grouting quality within the X–X section. The detection system primarily consists of a control system and LTF detection equipment. The control system encompasses signal display, operational status indication, motion parameter configuration, data acquisition, and presentation. It serves for human–machine interaction and control of the detection equipment, enabling automated backfill grouting inspection.

[figure(s) omitted; refer to PDF]

The LTF detection equipment is primarily composed of and functions as follows:

1. 400/600 MHz dual-frequency GPR antenna: The GPR antenna emits and receives electromagnetic waves during detection. High-frequency GPR offers higher detection accuracy but with a reduced depth, while low-frequency GPR provides lower detection accuracy but greater detection depth. Both high- and low-frequency GPR can better balance the detection accuracy and range for assessing the backfill grouting quality and adjacent soil cavities. Regarding frequency selection, the 400–900 MHz range is commonly used for tunnel lining detection with GPR. Within this range, the specific frequency choice varies based on tunnel characteristics, and there is already a wealth of research reports [13, 35–37]. Therefore, the 400 and 600 MHz used for backfill grouting quality and adjacent soil cavity detection in this paper are rational. The primary detection parameters for the two frequencies under conditions of high attenuation and low attenuation are shown in Table 4. Backfill grouting is detected behind the shield tunnel lining and adjacent soil cavities in a high attenuation environment. It is necessary to ensure the detection of grouting thickness with accuracy while also detecting cavities within a specific range.

2. Elevator mechanism: This adjusts the air-coupled distance between the segment and the GPR antenna. This distance is set to ensure the safety of the GPR antenna during the automated detection process. The specific value can be pre-set, considering factors such as an ultrasonic rangefinder at the top of the GPR antenna, which can maintain a constant air-coupling distance. This ensures the stability of the GPR data collection by keeping the coupling distance consistent.

3. Track: Driven by stepper motors, the GPR antenna moves upward along the track in the tunnel’s circumferential direction. Theoretically, the LTF allows for the detection of tunnels over a 360-degree range. Due to the limited space inside the tunnel, the tracks are assembled from individual standard blocks for ease of installation.

4. Support frame: The support frame, a steel rigid structure, securely fastens the detection equipment to the shield machine. It must withstand complex engineering environments without experiencing bending or deformation to ensure the LTF equipment’s safety and data collection stability.

5. Servo system: The servo system controls the operational movements of the detection equipment and ensures safety. The servo system facilitates the operation of the control system and hardware equipment.

Due to gravitational force, the grouting quality within the top range of the shield tunnel is prone to areas of low density. The LTF detection system enables automated and rapid scanning of a specific range at the top of the shield tunnel. When combined with intelligent real-time data interpretation, it facilitates real-time detection of backfill grouting quality and dynamic feedback to grouting parameters, allowing sufficient time for manual intervention. LTF greatly improves the efficiency of data collection; the single-ring collection time is usually less than 5 min, the collected data is transmitted to the shield machine operation room (or any data processing location) through the cable, and the operator only needs to click on the “one-button processing,” then the data can be automatically analyzed.

Table 4

Theoretical detection parameters at 400 and 600 MHz.

GPR frequency (MHz)	Maximum time window (ns)	Penetration thickness at low attenuation (m)	Penetration thickness at high attenuation (m)	Theoretical resolution (cm)
400	160	8	2.5	6
600	128	6.4	1.5	4

3.2.2. Detection and Intelligent Analysis of Backfill Grouting

To achieve intelligent analysis of GPR data, the LTF detection equipment is equipped with a dedicated data rapid analysis system. Intelligent analysis relies on the bi-frequency back projection (BBP) [38] algorithm to fuse the GPR data of two different frequencies and then machine learning algorithms to identify the grout thickness [31]. The intelligent recognition of grouting thickness relies primarily on supervised learning algorithms based on A-scans. The main principle is to use A-scans corresponding to various monitoring points as samples, with the thickness values at the corresponding points serving as labels. A regression model is then established to analyze engineering monitoring data [31, 39]. The schematic diagram of the intelligent detection results, as shown in Figure 9, comprises three parts:

[figure(s) omitted; refer to PDF]

1. Thickness distribution of grout within the monitoring line range (top left side): This section provides a quantitative representation of grout thickness within the monitoring line range, allowing for direct observation of the thickness distribution of grout within the specified area.

2. Actual detection positions of grout thickness distribution within the tunnel (right side): This part visually depicts the actual detection positions within the tunnel, providing an intuitive understanding of the detection area.

3. Evaluation results of backfill grouting quality (bottom): The evaluation results allow for assessing grout thickness within different ranges along the monitoring line based on engineering experience. The evaluation consists of three tiers: “Excellent,” “Qualified,” and “Insufficient,” correspondingly denoted by green, yellow, and red, aiding in the qualitative appraisal of backfill grouting quality.

3.2.3. Process of GPR Detection

The support frame fixed the LTF detection equipment onto the shield machine. As the shield machine advances during tunneling, to avoid interference with adjacent operations and ensure regular operation when the equipment is not in use, the default docking point for the GPR is set at 90° from the top of the track.

At least one measuring line is arranged for detection for each tunnel ring. The detection process is as follows:

1. After clicking the “Start” button, the radar antenna adjusts the segment-antenna spacing to 10 cm under the elevator’s guidance. Then, driven by the stepper motor, it gradually moves along the guide rail to the starting point of the detection.

2. The GPR automatically triggers data collection at the starting point and proceeds to collect data along the monitoring line until reaching the endpoint.

3. After reaching the endpoint, the GPR stops automatically. Clicking the “Reset” button returns it to the docking point, completing one data acquisition cycle.

After each data collection, the data for each ring is named “Segment Number-Date-Time” and saved in a specific directory.

3.3. Dynamic Feedback Mechanism for Engineering Application

Traditional grouting quality control controls parameters such as grout volume and pressure. However, this approach is relatively blind to geological conditions and environments with cavities and leaking channels. To leverage GPR rapid detection results for grouting quality control, this paper proposes a dynamic feedback system based on GPR detection results, as illustrated in Figure 10. As depicted in the figure, personnel inject grout into the shield tail gap according to the designed grouting parameters for any tunnel section. Before the grout completely solidifies, the LTF mechanism accompanying the shield is activated to inspect the grouting quality of the section, providing rapid detection results.

[figure(s) omitted; refer to PDF]

Technically, the implementation of the dynamic feedback mechanism involves several processes, including data acquisition, data transmission, data analysis, result output, result distribution, and parameter adjustment. After collecting GPR data on the LTF site, it is wirelessly transmitted to a local or cloud-based server, where intelligent processing and analysis are conducted. The results, such as grouting thickness and nearby cavities, are then displayed and can be shared in real-time through mobile apps or other platforms to management teams and construction personnel, aiding them in making timely adjustments to tunneling technical parameters. In fact, with the advancements in cloud computing, wireless transmission, and artificial intelligence technologies, the implementation of a dynamic feedback mechanism is now feasible. For example, we developed the GPR-AI Master [40, 41] intelligent platform for the specific scenario of backfill grouting detection, which enables intelligent detection and dynamic feedback for back-grouting in shield tunnels. In the future, with the continued advancement of engineering applications and cost optimization, the dynamic feedback mechanism is expected to be further applied in large-scale engineering projects, potentially enabling widespread intelligent detection and management services.

Construction proceeds according to the original parameters if the grouting quality meets the criteria. However, suppose insufficient grouting thickness or anomalies are detected. In that case, manual secondary grouting is implemented for the current detection section, and grouting parameters are adjusted for the next section to ensure its grouting quality meets the standards. The generated intelligent interpretation results can be shared via social media and archived as detection outcomes, facilitating convenient retrieval and review of the detection results at any time.

The entire tunneling process employs this dynamic feedback detection mechanism, enabling fine-grained control of grouting quality ring by ring, ensuring a tailored approach for each section, and achieving safe tunneling operations.

4. Engineering Application and Validation

4.1. Installation of LTF Equipment on the Shield Machine

As illustrated in Figure 11a,b, the LTF was installed at the location of the second ring segment behind the excavation face, securely affixed to the tunnel boring machine via supports. The positioning of its installation considers three factors: (1) The operation of LTF should not interfere with the ongoing work and safety of the shield machine on either side. (2) Its installation should not disrupt the regular operation of the segment assembly machine. (3) The installation position should enable timely detection of backfill grouting and adjacent soil cavities, allowing opportunities for remediation at insufficiently grouted locations. The on-site equipment installation is depicted as shown in Figure 11c. The institution utilizes support structures affixed to the shield machine, installed at the rear of the excavation face at Ring 2, to monitor grouting conditions while ensuring noninterference with the assembly of tunnel segments and the operational space of the side channels.

[figure(s) omitted; refer to PDF]

During the inspection process, point 1 serves as the starting point of the detection, while point 3 marks the endpoint of the detection range, covering a total range of 87.5° around the tunnel crown. Point 2 is designated as the resting position for the GPR when the LTF equipment is not in operation, ensuring the safety of the equipment.

4.2. Detection of Backfilling Grouting Distribution

The grouting thickness distribution of the grout is complex and nonuniform. For extensive survey results, detailed analysis of each monitoring point is challenging. Therefore, this paper adopts a statistical analysis of the grouting thickness values of all monitoring points on each GPR survey line as a unit, enabling comprehensive control over the grouting thickness of multiple rings.

The theoretical shield tail gap value of the shield tunnel in the X–X section of the southern extension of Xiamen Metro Line 3 is 0.14 m (half of the difference between the cutter diameter and the outer diameter of the segment). The statistical results based on intelligent detection results of the 20th–50th rings on the left track of the X–X section are depicted in Figure 12. Each box in the figure represents the 25%–75% confidence interval of the grouting thickness distribution for each ring, and the interquartile range (IQR) can be used to describe the degree of grouting thickness dispersion, assisting in the identification of outliers in the data. Additionally, each monitoring point analyses the thickness values’ mean, median, and outliers statistically. The mean and medium values of the grouting thickness of the tunnel are ~16 cm, with the filling rate of grouting generally exceeding 1.3. The confidence intervals of the grouting thickness, ranging from 25% to 75%, are above 14 cm, suggesting that the grout’s average thickness effectively fills the shield tail gap.

[figure(s) omitted; refer to PDF]

4.3. Detection of Adjacent Soil Cavities

4.3.1. GPR Data Processing System for Cavity Detection

Intelligent detection of adjacent soil cavities supported by LTF equipment lacks corresponding intelligent algorithms and is still under research. In this study, a self-developed dual-frequency GPR antenna was employed. To accommodate the specific hardware characteristics and meet the corresponding data analysis needs, our team developed a proprietary software platform called TGIS. As illustrated in Figure 13, TGIS is designed for the analysis and processing of engineering GPR data. It supports the import of raw GPR data paired with LTF files and offers both manual analysis and one-click automatic processing based on predefined parameters.

[figure(s) omitted; refer to PDF]

TGIS software includes traditional standard functions such as gain adjustment, filtering, and background removal, enabling rapid data analysis.

This paper provides a recommended processing flow and method for analyzing LTF’s original GPR detection data, focusing on cavity identification, as shown in Table 5. Based on the parameters in this table, the detection signals of anomalous bodies near the excavation area can be highlighted. This process involves:

Table 5

Recommended parameters of GPR data processing using TGIS software.

Parameters	Values	Parameters	Values
Dewow/step	0	Direct current removal/step	0
Background removal/step and coefficient	4/1	Inter-trace equalization: step and window size (ns)	0/0.5
Automatic gain control: step and window size (ns)	2/3	FIR filter: step and frequency range	3/250–550

1. Calibration and preprocessing of raw data. During this stage, filtering and processing are conducted on detected data to distinguish between erroneous and accurate detections.

2. Signal enhancement and noise reduction using gain and filtering functions. The primary principle of signal processing is to amplify sound signals as much as possible while minimizing noise, thereby enhancing the signal-to-noise ratio.

3. Environmental noise is consistently present; thus, it will reasonably exist in all temporal and spatial GPR data. Background removal to eliminate environmental interference.

4. Further data analysis and validation to determine the location and nature of anomalous bodies. This stage aims to identify abnormal GPR waveform features from processed results and, in combination with the electrical parameters of the materials and the temporal distribution range of abnormal waveforms, calculate the depth of the anomalous body (cavity).

4.3.2. Engineering Field Data Analysis and Results

The analyzed data is the same as the data for backfill grouting detection. Taking the data from the 49th ring as an example, after data analysis using TGIS, the result under 400 MHz is shown in Figure 14. Position and depth information are extracted from the GPR time sequence waveform, as shown in Equation (5). By using the velocity of electromagnetic waves in the medium and the time-domain data, the depth of the anomalous waveform can be calculated. According to the electrical parameters as shown in Table 3 and the calculation formula represented by Equation (5), the propagation speed of the 400 MHz electromagnetic wave in the segment layer is determined to be 9.49 × 10⁷ m/s and it takes ~3.69 ns to penetrate the segment with a 35 cm thickness. The velocity through the grout material is 5.67 × 10⁷ m/s, and the round-trip travel time is ~2.65 ns; The velocity through soil material is ~1.12 ×10⁷ m/s.$$\begin{array}{}\text{(5)}& \begin{array}{c}v=\frac{c}{\sqrt{\mu \epsilon }}\\ D=2vt\end{array},\end{array}$$where c represents the speed of electromagnetic waves in a vacuum, which is the same as the speed of light, taken as 3 × 10⁸ m/s; $\epsilon$ is the relative dielectric constant of the medium being traversed; $\mu$ is the relative dielectric constant of the medium being traversed; v represents the speed of electromagnetic waves in a medium; t represents the one-way travel time of electromagnetic waves in a medium. D is the depth of the target detected, calculated based on the travel time of the waveform.

[figure(s) omitted; refer to PDF]

The calculations indicate that the interface between the grout and soil should be ~10 ns in the GPR time domain, with the range beyond 10 ns representing the soil. As the detection of grout thickness is obtained through artificial intelligence methods, waveform analysis is not further explored here. Instead, emphasis is placed on analyzing the waveforms and positions of defects within the soil. Based on Figure 14, it can be seen that there are two cavities in the data of Ring 49 within the 25–35 ns, corresponding to a depth of ~1 m. The cavity is located at about 2.3 m along the monitoring line. Regarding waveforms, regular circular or square cavities exhibit characteristic parabolic features in GPR waveforms. However, in practical engineering data, the shapes of cavities are often irregular and may contain different filling materials. Thus, they frequently appear as discontinuities in the phase axes, among other characteristics.

The LTF system was originally designed to assess the quality of backfill grouting, and as a result, its use in data intelligence analysis for grouting detection has become relatively well-developed. However, its application for detecting cavities in adjacent soil remains in the research phase and has not yet matured into a reliable engineering solution. This represents a critical area for future investigation, warranting particular attention. A primary challenge lies in identifying and extracting waveform characteristics associated with cavities, given their variable shapes, diverse distribution distances, and unknown fill conditions.

The use of GPR for detecting both backfill grouting and adjacent cavities depends on balancing detection distance and resolution. A single frequency range is often inadequate to simultaneously evaluate near-field grouting quality and identify voids at greater distances. To detect more distant voids, a lower GPR frequency is necessary, which increases the size of the GPR antenna. This, in turn, complicates equipment installation and compromises safe operation within the constrained environment of tunnel construction.

4.4. Engineering Construction Quality Control

To assess the construction quality during the tunnel construction, comprehensive monitoring of key indicators, as shown in Table 6, was conducted for the X–X section. The monitoring includes environmental safety factors such as strata settlement and building settlement, and convergence values of the tunnel structure itself. Specifically, these indicators include vertical displacement of the ground, vertical displacement of pipelines, vertical displacement of buildings, and horizontal displacement of buildings.

Table 6

Key monitoring indicators and control values during tunnel construction.

Indicators	Control values		Maximum monitoring values
	Total (mm)	Rate (mm/d)	Total (mm)	Rate (mm/d)
Vertical displacement of ground	−20, +10	±3	−13.01	−0.62
Vertical displacement of pipelines	−10, +5	±2	−0.58	−0.58
Vertical displacement of buildings	−20, +5	±2	−3.14	0.92
Horizontal displacement of buildings	±3	±1	−1.89	−0.53
Convergence of segments	±12	±3	−5.18	0.27
Vertical displacement of the segment crown	±10	±2	−6.04	0.48

Taking the left line of the X–X section as an example, the surface settlement values during the tunnel’s passage beneath the Shuyou Hotel and the Yanwu Bridge are illustrated in Figure 15a and b, respectively. Specifically, when traversing the Yanwu Bridge, the cumulative settlement does not exceed 4.9 mm, with a daily settlement rate not surpassing 2.28 mm. In contrast, when passing under the Shuyou Hotel, the cumulative settlement remains below 2.35 mm, and the daily settlement rate does not exceed 0.88 mm. All monitoring values met the control values shown in Table 6 during construction. The results indicate that all monitoring indicators during the crossing were within the required control range, ensuring the standard tunnel’s structural safety and the crossing’s safety were well safeguarded.

[figure(s) omitted; refer to PDF]

5. Conclusions

To achieve the automated detection of the backfill grouting thickness distribution and adjacent soil cavities behind the shield tunnel lining, this paper proposed the LTF automated detection device and intelligent analysis system. The system is applied and validated in the X–X section of the southern extension of Xiamen Metro Line 3. The research ensures the safe construction of shield tunnels in complex geological and underground conditions, providing valuable insights for tunnel safety and intelligent construction. The main research conclusions are as follows:

1. TDR measurements demonstrated distinct electrical properties of the grout. These values exhibited significant differences compared to the surrounding soils and concrete segments, providing a substantial electromagnetic contrast. This contrast ensures the effectiveness of GPR for detecting grout distribution and potential voids behind tunnel linings, confirming the feasibility of using GPR for backfill grout detection in tunnel applications.

2. The proposed LTF system enables automated, ring-by-ring tunnel segment detection, completing each ring in under 5 min. This significantly advances traditional manual methods, eliminating inherent inefficiencies, and subjective data interpretation limitations.

3. Detection results from the X–X section demonstrate effective grout filling in the left line’s 30th–50th rings, with 25%–75% confidence intervals exceeding 14 cm, confirming satisfactory grouting quality, and complete shield tail gap filling.

4. This study investigated soil cavity detection using LTF equipment, demonstrating its effectiveness in identifying adjacent cavities within a defined excavation range through optimized GPR frequency selection and data analysis.

References

[1] H. Guo, "A Review of Metro Tunnel Construction Methods," IOP Conference Series: Earth and Environmental Science, 2019.

[2] G. Girmscheid, C. Schexnayder, "Drill and Blast Tunneling Practices," Practice Periodical on Structural Design and Construction, vol. 7 no. 3, pp. 125-133, 2002.

[3] Z. Sun, D. Zhang, Q. Fang, "Technologies for Large Cross-Section Subsea Tunnel Construction Using Drilling and Blasting Method," Tunnelling and Underground Space Technology, vol. 141,DOI: 10.1016/j.tust.2023.105161, 2023.

[4] H.-B. Wang, R.-J. Zhang, F.-J. Tao, J.-J. Zheng, "Study on Long-Term Surface Settlement Induced by Shield Tunneling in under-Consolidated Soft Ground," Tunnelling and Underground Space Technology, vol. 148,DOI: 10.1016/j.tust.2024.105772, 2024.

[5] N. Jiao, X. Wan, J. Ding, X. Zhang, C. Xue, "Mechanical Properties and Microstructure of Lime-Treated Shield Tunnel Muck Improved With Carbide Slag and Soda Residue," Construction and Building Materials, vol. 428,DOI: 10.1016/j.conbuildmat.2024.136419, 2024.

[6] K. Ying, F. Ye, Y. Li, X. Liang, E. Su, X. Han, "Backfill Grouting Diffusion Law of Shield Tunnel considering Porous Media With Nonuniform Porosity," Tunnelling and Underground Space Technology, vol. 127,DOI: 10.1016/j.tust.2022.104607, 2022.

[7] P. Li, C. Q. Wang, G. X. Fan, Y. Liu, J. C. Xiong, "Study on Time-Varying Viscosity of Cement and Sodium Silicate Grout under Seawater Condition," Construction and Building Materials, vol. 269,DOI: 10.1016/j.conbuildmat.2020.121223, 2021.

[8] H. Qin, Y. Tang, Z. Wang, X. Xie, D. Zhang, "Shield Tunnel Grouting Layer Estimation Using Sliding Window Probabilistic Inversion of GPR Data," Tunnelling and Underground Space Technology, vol. 112,DOI: 10.1016/j.tust.2021.103913, 2021.

[9] Z. Shi, X. Bu, K. Zeng, X. Xie, H. Zeng, "Optimization Study for Cement-Based Grout Mixture Ratio of Shield Tunnel considering Spatiotemporal Evolution of Grout Buoyancy," Construction and Building Materials, vol. 442,DOI: 10.1016/j.conbuildmat.2024.137600, 2024.

[10] S. K. Dwivedi, M. Vishwakarma, A. Soni, "Advances and Researches on Non Destructive Testing: A Review," Materials Today: Proceedings, vol. 5 no. 2, pp. 3690-3698, DOI: 10.1016/j.matpr.2017.11.620, 2018.

[11] A. M. Alani, L. Lantini, "Recent Advances in Tree Root Mapping and Assessment Using Non-Destructive Testing Methods: A Focus on Ground Penetrating Radar," Surveys in Geophysics, vol. 41 no. 3, pp. 605-646, DOI: 10.1007/s10712-019-09548-6, 2020.

[12] X. Xu, Y. Lei, F. Yang, "Railway Subgrade Defect Automatic Recognition Method Based on Improved Faster R-CNN," Scientific Programming, vol. 2018, 2018.

[13] B. Kravitz, M. Mooney, J. Karlovsek, I. Danielson, A. Hedayat, "Void Detection in Two-Component Annulus Grout behind a Pre-Cast Segmental Tunnel Liner Using Ground Penetrating Radar," Tunnelling and Underground Space Technology, vol. 83, pp. 381-392, DOI: 10.1016/j.tust.2018.09.032, 2019.

[14] X. Jiang, H. Zhu, Z. Yan, "A State-of-Art Review on Development and Progress of Backfill Grouting Materials for Shield Tunneling," Developments in the Built Environment, vol. 16,DOI: 10.1016/J.DIBE.2023.100250, 2023.

[15] D. Rui, "Research on Fiber Sensing of Health Monitoring for Steel Tube-Confined Concrete Arch Bridge," China Civil Engineering Journal, vol. 38, pp. 69-74, 2005.

[16] L. Xiang, H. L. Zhou, Z. Shu, S. H. Tan, G. Q. Liang, J. Zhu, "GPR Evaluation of the Damaoshan Highway Tunnel: A Case Study," NDT & E International, vol. 59, pp. 68-76, DOI: 10.1016/j.ndteint.2013.05.004, 2013.

[17] H. Qin, X. Xie, Y. Tang, Z. Wang, "Experimental Study on GPR Detection of Voids inside and behind Tunnel Linings," Journal of Environmental and Engineering Geophysics, vol. 25 no. 1, pp. 65-74, DOI: 10.2113/JEEG18-085, 2020.

[18] M. Fu’an, T. Fu, H. Shangjun, "GPR Test and Application Based on Shield Tunnel Segment Groutingsimulation Platform," Mineral Resources Geology, vol. 36, pp. 854-864, 2022.

[19] L. Jianying, "Research on Radar Detection Technology and Migration of Hidden Defects behind the Shield Tunnel," 2021.

[20] K. Li, X. Y. Xie, B. Zhou, C. F. Huang, Y. H. Zhou, M. Li, "Model Test Study on GPR Detection for Cement-Sodium Silicate Synchronous Grouting in Shield Tunnel," Zhongguo Gonglu Xuebao/China Journal of Highway and Transport, vol. 37, pp. 330-341, DOI: 10.19721/J.CNKI.1001-7372.2024.03.021, 2024.

[21] A. Benedetto, F. Tosti, L. Bianchini Ciampoli, F. D’Amico, "An Overview of Ground-Penetrating Radar Signal Processing Techniques for Road Inspections," Signal Processing, vol. 132, pp. 201-209, DOI: 10.1016/j.sigpro.2016.05.016, 2017.

[22] Z. Tong, J. Gao, D. Yuan, "Advances of Deep Learning Applications in Ground-Penetrating Radar: A Survey," Construction and Building Materials, vol. 258,DOI: 10.1016/j.conbuildmat.2020.120371, 2020.

[23] J. Zou, X. Xie, B. Zhou, "Large Rectangular Cross-Section Tunnel Undercrossing Urban Road by Micro Pipe Jacking and Joint Assembly Structure (MPJ & JAS) Method in Soft Soils," Scientific Reports, vol. 14 no. 1,DOI: 10.1038/s41598-024-55754-7, 2024.

[24] R. H. Cole, "Time Domain Reflectometry," Annual Review of Physical Chemistry, vol. 28 no. 1, pp. 283-300, DOI: 10.1146/annurev.pc.28.100177.001435, 1977.

[25] K. Noborio, "Measurement of Soil Water Content and Electrical Conductivity by Time Domain Reflectometry: A Review," Computers and Electronics in Agriculture, vol. 31 no. 3, pp. 213-237, DOI: 10.1016/S0168-1699(00)00184-8, 2001.

[26] S. B. Jones, J. M. Wraith, D. Or, "Time Domain Reflectometry Measurement Principles and Applications," Hydrological Processes, vol. 16, pp. 141-153, 2002.

[27] A. Neal, "Ground-Penetrating Radar and Its use in Sedimentology: Principles, Problems and Progress," Earth-Science Reviews, vol. 66 no. 3-4, pp. 261-330, DOI: 10.1016/j.earscirev.2004.01.004, 2004.

[28] D. Daniels, Ground Penetrating Radar, 2004.

[29] S. Hong, D. Chen, B. Dong, "Numerical Simulation and Mechanism Analysis of GPR-Based Reinforcement Corrosion Detection," Construction and Building Materials, vol. 317,DOI: 10.1016/j.conbuildmat.2021.125913, 2022.

[30] gprMax: Electromagnetic simulation software, (n.d.)., . (accessed October 4, 2023) http://www.gprmax.com/

[31] K. Li, X. Xie, B. Zhou, "Thickness Regression for Backfill Grouting of Shield Tunnels Based on GPR Data and CatBoost & BO-TPE: A Full-Scale Model Test Study," Underground Space, vol. 17, pp. 100-119, DOI: 10.1016/J.UNDSP.2023.10.003, 2024.

[32] H. M. Jol, Ground Penetrating Radar Theory and Applications, 2008.

[33] US Patent for Loaded-to-frame detection equipment and method for backfill grouting of shield tunnel Patent (Patent # 11,181,633 issued November 23, 2021) - Justia Patents Search, (n.d.). (accessed February 28, 2023). https://patents.justia.com/patent/11181633

[34] K. Li, X. Xie, X. Zhang, B. Zhou, T. Qu, L. Zeng, "Application of intelligent real-time GPR monitoring system in the quality detection of backfill grouting of shield tunnel. In19th International Conference on Ground Penetrating Radar," pp. 83-86, DOI: 10.1190/gpr2022, 2022.

[35] R. W. Harseno, S.-J. Lee, S.-H. Kee, S. Kim, "Evaluation of Air-Cavities Behind Concrete Tunnel Linings Using GPR Measurements," Remote Sensing, vol. 14 no. 21,DOI: 10.3390/rs14215348, 2022.

[36] L. Hou, Q. Zhang, Y. Du, "Width Estimation of Hidden Cracks in Tunnel Lining Based on Time-Frequency Analysis of GPR Data and Back Propagation Neural Network Optimized by Genetic Algorithm," Automation in Construction, vol. 162,DOI: 10.1016/j.autcon.2024.105394, 2024.

[37] C. Li, M. Li, X. Yang, "Boundary Recognition of Tunnel Lining Void From Ground-Penetrating Radar Data," Journal of Geophysics and Engineering, vol. 20 no. 4, pp. 723-734, DOI: 10.1093/jge/gxad041, 2023.

[38] X. Xie, J. Zhai, B. Zhou, "Back-Fill Grouting Quality Evaluation of the Shield Tunnel Using Ground Penetrating Radar With Bi-Frequency Back Projection Method," Automation in Construction, vol. 121,DOI: 10.1016/j.autcon.2020.103435, 2021.

[39] L. Zeng, X. Zhang, X. Xie, B. Zhou, C. Xu, S. Lambot, "Measuring Annular Thickness of Backfill Grouting Behind Shield Tunnel Lining Based on GPR Monitoring and Data Mining," Automation in Construction, vol. 150,DOI: 10.1016/j.autcon.2023.104811, 2023.

[40] K. Li, X. Xie, Y. Zhou, B. Zhou, "GPR-AI Master: Cloud-Based Intelligent Analysis System for Ground Penetrating Radar Data of Backfill Grouting in Shield Tunnel (Chinese Software)," 2024.

[41] K. Li, X. Xie, B. Zhou, "GPR-AI Master: GPR Data Analysis Utilizing Deep Learning and ECS for Backfill Grouting Detection in Shield Tunnels," ,DOI: 10.1109/PIERS62282.2024.10617955, 2024.