1. Introduction
Large-diameter shield tunnels are extensively employed in the construction of urban transportation infrastructure due to the rapid development of modern metropolitan areas. Currently, the commonly used design methods for shield tunnel segments, such as the modified idiom and beam spring models, primarily focus on stress analysis during regular operation and do not account for the impact of various factors during the construction phase. Empirical evidence from numerous shield engineering projects indicates that there is a transitional period from segment assembly into a ring to the final consolidation of both the grout and the segment, enabling them to withstand the earth’s pressure collectively. During this stage, the mechanical behavior of the segment becomes intricate, encompassing factors such as Jack thrust, grouting pressure, slurry buoyancy, formation stress, and adverse internal forces resulting from segment assembly. Hence, there is a need to conduct a systematic investigation into the mechanical behavior of segments and their vital influencing factors, considering the interactions among the shield machine, segment, and surrounding soil. Additionally, it is crucial to accurately assess the initial stress state of the tunnel structure by comprehensively understanding the laws governing tunnel structure deformation, internal forces, and external loads.
Due to the construction sequence (tunneling posture, assembly process, synchronous grouting), the conventional theoretical analysis and numerical simulation struggle to accurately depict the stress state of the shield segment during this stage. In the construction and operation of large-diameter shields, manual measurements are frequently employed to monitor tunnel deformation and changes in internal forces. Nevertheless, manual measurements are conducted using a handheld reader once or twice a day, resulting in low monitoring accuracy and susceptible monitoring results. Numerous scholars have extensively researched tunnel engineering monitoring systems. Jiang et al. [
1] proposed a real-time monitoring system for surface deformation induced by shield construction in homogeneous strata, combining static-level measurements and single-point displacement meters. This system has been successfully implemented in Beijing Metro Line 14 and the new airport line. Cui and Tan [
2] conducted long-term settlement monitoring for Shanghai Tunnel Line 1 since 1995, revealing that long-term service can lead to differential settlement in the tunnel’s longitudinal direction, thus causing segment disease. Huang et al. [
3] developed a real-time monitoring system to analyze the interaction between cutter head excavation and surrounding rock during shield tunneling, which was applied to a tunnel project in Lanzhou, and they also proposed a back-calculation method to estimate the surrounding rock pressure. Zheng et al. [
4] used optical fiber sensors to monitor the longitudinal strain, curvature, and joint deformation of segments during the tunneling of Foshan Metro Line 2. They established a segment deformation prediction model based on the beam theory, considering the variation in formation parameters. Hu et al. [
5] conducted field monitoring and finite element analysis to study ground displacement, as well as longitudinal and transverse deformation of an existing tunnel when a new shield tunnel in Tianjin Metro intersected with it. The calculation results indicate minimal deformation in the lower structure of the existing tunnel. Gue et al. [
6] monitored and analyzed the mechanical response of the existing tunnel structure during the process of building a new tunnel through an existing tunnel. Clayton et al. [
7] used displacement sensors to monitor the settlement caused by the tunnel crossing the airport process Wang et al. [
8] monitored the hoisting process of the underwater inlet and outlet of a hydraulic tunnel in a power plant and simulated the deformation of the segment joint during the lifting process. Guo et al. [
9] utilized GPS and Beidou systems to monitor riverbed deformation caused by underwater shield tunneling, achieving millimeter-level monitoring accuracy. Wang et al. [
10] proposed a stress analysis method for three-dimensional multilayer elastic materials, which combines domain decomposition technology and employs the generalized finite difference method to calculate the displacement and stress of subregions. The method is then compared with the finite element method and the boundary element method. The results indicate that the finite difference method exhibits high accuracy and efficiency in addressing nonlinear problems. Kabir and Aghdam [
11] proposed a solution method based on the Bezier MultiStep Method to solve the nonlinear vibration and postbuckling configuration control equations of composite Euler–Bernoulli reinforced beams with graphene nanoplatelets. This method demonstrates high accuracy and robustness. Bert and Malik [
12] proposed a numerical solution method, the differential quadrature method, for analyzing composite laminated plates. They utilized the first-order shear deformation plate theory to solve and verify the accuracy of the proposed method for the vibration of symmetrically laminated cross-ply plates.
In recent years, there has been a growing interest among researchers in establishing inversion analysis models based on field-monitored physical quantity data. These models aim to derive the initial parameters of the project site and propose theoretical prediction models that closely approximate the actual service state of the tunnel. Callisto and Ricci [
13] utilized measured damage data from an earthquake-damaged tunnel structure in Italy. They employed the equivalent static method to perform an inversion analysis of the earthquake-induced load and validated the proposed method using a coupled dynamic finite element method that considered the interaction between the soil and structure. Fakhimi et al. [
14] introduced a numerical parameter inversion analysis method. They performed a matching inversion analysis of numerical simulation results and the measured convergence value of a tunnel in Iran to determine the in situ stress and soil cohesion of the tunnel’s location. These values were then compared with the results of direct shear and plate load tests. Kim et al. [
15] developed an inversion analysis method for the concrete lining ground load, considering the construction process, using convergence data from the concrete segment. Sharifzadeh et al. [
16] employed a single-variable optimization algorithm to directly analyze displacement and simulate the creep characteristics of the surrounding rock in the tunnel. Lee et al. [
17,
18] utilized the numerical inversion analysis method to investigate the load distribution of the surrounding rock during the excavation of a multiarch tunnel. They established a correlation between the height of the load loose circle and the deformation modulus of the rock mass and derived a calculation method for the surrounding rock load in a multiarch tunnel. Yan et al. [
19] presented an inversion analysis method for determining the water and soil load acting on the exterior of the tunnel. They based their method on strain data from monitored segment sections on-site and assumed a linear change in the water and soil load. The least square method was employed for inversion analysis, allowing for the calculation of the stress in the tunnel section.
The rapid development of artificial intelligence technology has led to the gradual integration of machine learning algorithms into the prediction and parameter inversion analysis processes of tunnel engineering. Ye et al. [
20] proposed static leveling to monitor the real-time floating amount of segments throughout the entire process of shield tunneling. They introduced a machine learning algorithm that combines particle swarm optimization and the random forest method to predict the floating process of the segment, achieving an R2 value of 0.915. Ye et al. [
21] monitored the surface settlement resulting from shield tunneling using Fiber Bragg grating sensors. They utilized the TS-BPNN algorithm to predict the subsequent settlement. Ye et al. [
22] compiled a segment floating process database that incorporates geological conditions, tunnel geometry, and shield tunneling parameters. They employed a variety of machine learning algorithms to predict the floating process. Niu et al. [
23] proposed an inversion analysis method that combines model testing and numerical simulation, based on the measured water and soil load distribution in a railway tunnel located in Foshan. They presented the load distribution characteristics of the tunnel segment in the soft soil layer. Vardakos et al. [
24] utilized a simulated annealing algorithm and numerical simulation software to conduct an inversion analysis of the tunnel’s deformation and plastic zone. This approach has been employed in specific engineering calculations. Yu et al. [
25] used a BP neural network to perform an inversion analysis of the displacement of a pile foundation. They obtained the elastic modulus, Poisson’s ratio, cohesion, and internal friction angle of a landslide. Zhao and Feng [
26] employed particle swarm optimization and multilinear regression to conduct an inversion analysis of the direction and magnitude of in situ stress along the tunnel. This analysis was based on geological conditions and measured stress.
Currently, the monitoring and mechanical analysis of tunnel segment structures primarily concentrate on the operational stage after tunnel construction is completed, with less consideration given to the mechanical behavior of shield tunnel segments during construction processes. Additionally, the soil pressure acting on the segment significantly impacts segment design, particularly for large-diameter underwater shield tunnels. A notable disparity exists between the current segment design theory and the actual stress state, making it challenging to ascertain whether the existing segment design load method can ensure the safety and reliability of the segment during both construction and operation. The development of deep learning technology and the proposal of inversion analysis theory offer potential solutions to this issue.
This paper establishes an on-site monitoring system for the Internet of Things using embedded optical fiber sensors. The variation in water and earth pressure and mechanical response law of the segment ring in the stage of shield assembly and removal from the shield tail are analyzed. A Bayesian-genetic algorithm inversion analysis model of water and earth pressure was proposed. Specifically, the optimization objective function for inversion analysis is developed by modifying the idiom, which establishes a relationship between the segment’s internal force and external load. The measured internal force data are then analyzed using a genetic algorithm. Additionally, a Bayesian algorithm is employed to optimize various parameters, such as population number, gene length, crossover probability, and mutation probability, within the genetic algorithm. This optimization aims to enhance computational efficiency and accuracy. Finally, the inversion analysis load is compared with the field-measured value, and recommendations regarding the load values are provided.
2. Project Overview
2.1. Project Background
The Qinwang tunnel is situated 3 km downstream from the Fuyang bridge. This tunnel is the first road–rail tunnel in Zhejiang Province of China. It spans a total length of 3066 m, with 2868 m comprising the main tunnel and 1254 m forming the shield length in the river crossing section, as shown in
Figure 1. In the river crossing section, a double-tube circular shield is employed. The shield cutter head has an outer diameter of 15.8 m, while the shield tunnel has an outer diameter of 15.2 m and an inner diameter of 13.9 m. The segment is constructed of C60 reinforced concrete and takes the form of a flat plate structure. It incorporates a universal wedge ring with a wedge amount of 52 mm, has a ring width of 2 m, a segment thickness of 0.65 m, and employs a main reinforcement HRB400e with a diameter of 28 mm. The outer main reinforcement of the segment has a protective layer thickness of 0.05 m, while the inner main reinforcement has a protective layer thickness of 0.04 m. Staggered assembly is achieved using inclined bolt connections, and the tunnel is constructed using the 7 (standard block) + 2 (connecting block) + 1 (key block) block mode.
The Qinwang tunnel predominantly traverses a stratum of pebble soil, characterized by uneven particle distribution, and challenges in transmitting force between particles. This soil layer is a classic example of mechanical instability. The overlying soil layer predominantly consists of silty clay, silt, and fine sand, exhibiting strong permeability. The riverbed elevation above the Qinwang tunnel typically ranges from −5.43 m to 3.64 m. In these geological conditions, the operation of the Qinwang tunnel is primarily susceptible to risks including uneven settlement deformation, segment cracking, and water seepage. Consequently, the design of the Qinwang tunnel employed the beam–spring model for calculations, incorporating earth and water separation in load calculation. Simultaneously, a safety factor of 1.25 was considered for structural anti-floating verification. The specific geological parameters are presented in
Figure 2.
2.2. Design and Implementation of Monitoring System
To investigate the mechanical behavior and key influencing factors of the segment under the interaction between the shield machine and the surrounding soil, it is necessary to analyze the stress state of the shield machine during tunneling. The primary sources of mechanical behavior in the shield machine segment can be attributed to four factors, as shown in
Figure 3:
Jacking force L1: During the construction of a shield tunnel, each jack of the shield machine exerts force on the segment through a pallet, which divides the cross-section into different areas. However, the thrust in each area varies during excavation and assembly. Measured data indicates that the thrust at the bottom of the tunnel arch can be twice as high as the thrust at the top of the tunnel arch [
27].
Shield tail brush force L2: The shield tail brush, typically made of steel wire, is installed at the rear of the shield shell of the shield machine. During synchronous grouting of the shield machine, the shield tail seal is employed to prevent slurry from flowing back into the shield machine. As the tunneling distance increases, the shield tail brush becomes enveloped by hardened slurry, resulting in increased stiffness and generating force reactions on the segment.
Shield machine shell force L3: Due to the influence of the tunnel design axis during the tunneling process, the shield machine may not always align perfectly parallel to the segment. This misalignment is particularly notable on uphill and downhill slopes, where the segment may meet the shield shell of the shield machine, resulting in a force exerted on the segment.
Friction force L4 and extrusion force L5 of adjacent segment: When the segment is assembled, they are subject to boundary conditions imposed by the adjacent segment. Longitudinally, the adjacent segment exerts an extrusion force on the assembled segment, while laterally, a frictional force is generated.
Throughout the shield tunnel construction, the evolving construction process leads to gradual changes in the acting forces L1~L5, delineated into three phases: Phase 1 involves the segment being situated within the shield shell, where the jack thrust L1 directly influences the segment, inducing longitudinal stress fluctuations. Phase 2: The segment is gradually removed from the shield tail. The brush force L2 and shield shell force L3 act on the outer side of the segment, causing internal stress fluctuations in the segment along with the jack thrust L1. Phase 3: The segment dislodges from the shield tail. The L1~L3 forces gradually dissipate, and the stress on the segment gradually balances under the action of earth and water pressure. In the three phases, both the frictional force L4 and the extrusion force L5 act on the segment, affecting the stress fluctuations. Consequently, the monitoring system design involves placing an earth pressure gauge between the segment to monitor L1 and L5. Additionally, an earth pressure gauge and seepage pressure gauge are positioned outside the segment to monitor L2 and L3. An earth pressure gauge is positioned between blocks to monitor L4.
The experimental ring is buried at the left and right tunnel rings, R41, R264, R600, and R609. To achieve monitoring objectives, earth pressure gauges and seepage pressure gauges are positioned on the outer side of each segment to track changes in the earth and water pressures. Concrete strain gauges and steel stress gauges are also installed on each segment to monitor changes in internal forces. Additionally, upon completion of segment assembly, a bolt stress gauge is installed at a designated location to monitor changes in bolt stress during shield tunneling construction. This study focuses on the R600, providing a detailed introduction to the design, installation, and data acquisition process of the on-site Internet of Things monitoring system. Furthermore, it analyzes the mechanical behavior of an ultra-large-diameter shield tunnel segment during the construction stage. The division of blocks and the distribution of embedded sensors in the test ring segment are illustrated in
Figure 4.
The resolution and accuracy of sensors are shown in
Table 1.
All sensors are embedded in the segment factory according to the design specifications, as shown in
Figure 5. Fiber Bragg grating sensors are interconnected using optical cables, which converge to the concealed box in a series configuration. To ensure data quality, both ends of the optical cable are connected to the concealed box, with 16 m of excess length reserved at one end and 5 m reserved at the other end for backup purposes. The osmometer cable is also connected to the concealed box. Once the segment casting and curing are completed, the concrete covering the earth pressure gauge and osmometer is carefully chiseled off. A handheld demodulator is then used to collect the wavelength and frequency readings from the sensor, which are compared with the preset values to verify the reliability of the sensor during the monitoring process.
The segment design wedge reaches its maximum at block B4 and its minimum at block F. Therefore, during the segment assembly process, B4 blocks are first assembled, followed by the left and right blocks, and finally, the F blocks are assembled. To ensure the collection of necessary data during the shield construction process, the demodulator and distribution box are installed at the designated position of the tunnel arch waist (R556 R557) one day in advance.
For the installation of the left half-ring instrument and demodulator, testers 1 and 2 connect the embedded sensors in the left half-ring to the demodulator using collection cables in blocks. In this process, priority is given to connecting the earth pressure gauges between the rings to monitor the influence of Jack thrust on the stress of the segment ring. After the installation of the embedded sensors, testers 1 and 2 install the bolt stress meter and connect its cables to the demodulator in blocks. Meanwhile, testers 3 and 4 use a handheld reader to collect initial values from the sensors at the left waist of the tunnel, which are then connected to the flange plate. The flange plate is further connected to the demodulator to enable automatic data collection, with an acquisition frequency of 1 s per time. Additionally, tester 6 is assigned to check the demodulator data in real time. Testers 4 and 5 follow a similar procedure to assemble the right half-ring instrument and demodulator to the left flange, as shown in
Figure 6.
5. Conclusions
This paper presents a study conducted on a super-large-diameter highway–rail combined shield tunnel project. An on-site monitoring system based on the Internet of Things (IoT) is established to continuously monitor the water and soil loads, as well as internal forces during the process of segment assembly and removal from the shield tail. The measured results are then utilized to invert the water and soil loads, leading to the following main conclusions:
(1) The earth and water pressure on the outer side of the segment undergo two stages of fluctuation and stabilization. Once the segment is removed from the shield tail, it undergoes dynamic adjustments due to the combined effects of the longitudinal component of the jack thrust, brush force on the shield tail, shield shell force, formation pressure, grouting pressure, and segment friction generated by adjacent assembled segments. Eventually, the loads gradually stabilize. The peak load during the fluctuation stage is twice the load observed during the stable stage. Greater initial internal forces will significantly influence forces during tunnel operation. Excessive earth and water pressure in the fluctuating section can lead to segment damage and seepage, ultimately impacting the tunnel’s long-term functionality and robustness.
(2) After the segment is removed from the shield tail, the internal forces experience violent fluctuations before reaching a state of stability. The peak value of the axial force during the fluctuation stage is eight times higher than the stable value. This fluctuation is primarily attributed to changes in boundary conditions (such as jack thrust and extrusion pressure on the front and rear segment) during the removal process. Similarly, the peak value of the bending moment during the fluctuation stage is five times higher than the stable value, mainly due to variations in load conditions (shield machine force, grouting pressure, and water and soil pressure) during the segment ring extraction from the shield tail.
(3) To invert the water and soil loads on the outer side of the segment, a Bayesian genetic algorithm is proposed. The obtained earth pressure and water pressure values exhibit errors of 76.21% and 17.86%, respectively, when compared with the measured values. The inversion algorithm exhibits a 51.16% reduction in calculation error compared with the full coverage pressure theory, indicating that the proposed algorithm provides results closer to the actual measurements.