A Special-shaped servo structural agent and undercutting construction method for shield tunnel ramp junction

Abstract

In the construction and renovation processes of underground spaces in dense urban environments, there is an increasing demand for space-intensive layouts and special-shaped cross-sections. To address the requirements for special-shaped cross-section design and lightweight structures, this study, against the backdrop of construction of underground road ramp junctions, proposes a servo structural agent for special-shaped cross-sections and its underground excavation method with no foundation pit construction by integrating traditional underground structures with intelligent perception and servo strut control. This agent embeds servo components into the structural system and performs active force control through sensing of stress and deformation. Compared with traditional structures, servo structural agent can adaptively balance the load changes induced by excavation, thereby reducing component size and deformation. On this basis, an undercutting construction method and process are proposed, utilizing dual shield tunneling parallel construction and underground excavation to replace the traditional large working shaft schemes. Through numerical simulations of typical conditions, the servo structural agent is shown to reduce maximum deformation by 30% and peak internal forces by 10%. The related work provides novel approaches for underground excavation construction and safety control in urban areas with dense underground road networks.

Graphical abstract

Keywords

servo structural agent / special-shaped cross-section / shield tunnel / undercutting construction / servo strut

1 Introduction

The rapid urbanization process has increased demands for space-intensive layouts and special-shaped underground infrastructure. Urban underground road, as a critical transportation infrastructure, play a critical role in alleviating surface traffic congestion and improving urban mobility efficiency, with their construction scale and quantity expanding as a key trend in urban renewal [1]. To enhance regional connectivity, ramp tunnels intersecting with surface roads are often incorporated into main tunnels. However, abrupt changes in traffic direction at these junctions necessitate special-shaped cross-sections, resulting in complex tunnel geometries that have become a critical limiting factor in underground road projects [2]. Structures with special-shaped cross-sections exhibit geometric asymmetry, structural asymmetry, and asymmetric loading, rendering them more susceptible to damage than standard single-bore tunnels [3]. Additionally, shield tunnel ramp junctions feature localized stress concentration during initial construction [4], significant deformation at junctions and arch crowns [5,6], and requirements for temporary supports and longitudinal connections [7]. Therefore, structural safety control during construction of special-shaped cross-section structures at ramp junctions warrants significant attention.

Conventional construction methods for ramp junctions in underground roads often rely on open-cut excavation and caisson sinking, which suffer from prolonged timelines, large footprints, and environmental impacts, making them unsuitable for dense urban cores [8,9]. Consequently, underground excavation techniques are increasingly adopted, with Japan pioneering methods like shield cutting, large-diameter curved pipe roofing, and arched pipe umbrella techniques for Yamate Tunnel ramps [10,11], and China developing innovations like the pipe-roof pre-construction [12,13], freeze-sealing pipe roof [14], pile-beam-arch [9,15,16], cavern-pile [17], open-cut combined with underground excavation [18], and mechanized lateral cutting methods [19]. These have been applied in metro station expansions and cross-passage constructions, with emerging projects adopting special-shaped and varying cross-sectional geometries.

Current underground excavation methods rely on passive reinforcement technologies and large-scale rigid supporting structures, such as steel tube slabs [20], steel ring girders [19], internal support [21], hydraulic support trolley [22], which lack proactive adjustment capabilities and are insufficient for extreme loading scenarios. Recent servo active control advancements, applied in deep excavation engineering [23–27] and connecting passage expansions [28], enable dual control of deformation and load. Field studies show hydraulic servo-enhanced steel supports actively adjust axial loads, significantly reducing lateral wall displacements, bending moments, and ground settlements [25,29,30], with innovations like capsule expansion [31], elastic support points method [32], and corner and arching effects [33] to optimize support efficiency. However, such active control has not been applied to urban underground road shield tunnels, despite its potential for managing complex mechanical interactions and stringent deformation control requirements of special-shaped cross-sections.

Concurrently, the integration of sensing and control technologies in underground engineering enables the acquisition of feedback information. Such as intelligent fiber optic sensors for real-time structural monitoring and adaptive load regulation [34–36]; synergistic applications of smart materials [37] and deep learning models [38,39] to enhance reinforcement reliability. The rapid development of artificial intelligence in recent years has led to the emergence of the concept of agent in structural sensing and control methods. An agent refers to any entity that perceives its environment through sensors and acts upon that environment through actuators. It has evolved from direct responses to immediate perception into a dynamic adaptive process with autonomous learning that optimizes behavior through experience [40,41].

Furthermore, Intelligent Structural Systems (ISS) with similar concepts are transitioning from theory to practice, initially focusing on embedding smart materials or sensors for structural control and health monitoring [42,43]. Recent advancements include modular vision-based and large language model-driven agents for enhanced environmental perception [44], with studies improving self-sensing, self-adaptive, and self-healing capabilities [45], early warning and damage localization via deep learning [46], and hybrid control strategies integrating deep reinforcement learning and gradient-based optimization [47]. Nonetheless, existing ISS and agents lack strategies for perceiving structural stress states and implementing active control in complex underground road projects involving shield tunnel ramp junctions.

In summary, current underground excavation methods, projects, and structural reinforcement technologies in dense urban underground road cores still face the following challenges. 1) Passive rigid supporting structures exhibit limited adaptability for effective deformation control in complex special-shaped tunnel cross-sections. 2) ISS lack strategies and mechanisms integrating stress state perception with active control.

To address these challenges, this study focuses on the complex mechanical characteristics and deformation control requirements of special-shaped cross-sections in shield tunnel ramp junctions. We propose a Servo Structural Agent (SSA) integrated with sensing-control units and a combined permanent-temporary structural design.

The main contribution of this study can be summarized as follows.

1) A self-adaptive control strategy for shield tunnel ramp junction

The proposed control strategy for special-shaped cross-section structures adheres to a framework encompassing environmental parameter sensing, intelligent reinforcement learning, and action policy generation, offering a novel adaptive approach for underground space structures.

2) An intelligent undercutting excavation construction methodology

A novel full undercutting construction method proposed, integrating dual shield tunneling parallel construction and underground excavation to replace the traditional large working shaft schemes, offers renovation and expansion solutions for dense urban cores.

3) Special-shaped SSA mechanical and deformation behavior

Through numerical simulations, the mechanical behavior and deformation patterns of the special-shaped SSA are analyzed, validating the advantages of the proposed method and structure over conventional supports in deformation and stress control.

2 Special shaped cross-section servo structural agent

To address safety control challenges in future scenarios with complex stress conditions and stringent deformation requirements, this study proposes a SSA for shield tunnels with special-shaped cross-sections. Focusing on the construction of junction zone in urban underground roads, the SSA integrates three core components: shield tunnel segments, intelligent sensing units, and self-adaptive servo control mechanisms.

2.1 Structural agent principles and components

The structural agent refers to an integrated structural system comprising sensing, control, and conventional structural units, designed to achieve self-adaptive optimization of stress and deformation. It dynamically senses stress states and generates control strategies in response to environmental changes, as illustrated in Fig. 1.

The core innovation lies in the synergistic integration of sensing, control, and structural systems. By employing reinforcement learning algorithms, the system dynamically adjusts control unit actions based on real-time structural states, enabling adaptive optimization of mechanical performance. For special-shaped cross-sections in dense urban underground junction zones, this system mitigates complex stress variations during excavation (e.g., soil pressure fluctuations), minimizing structural deformation and stress concentrations, thereby enhancing construction safety and efficiency. The system integrates deformation and axial-force sensing units and servo control units to form a structural agent for shield tunnels with special-shaped cross-sections. Key components are as follows.

1) Integrated structural system

The design integrates servo struts with special-shaped segment configurations (Fig. 2). For special-shaped cross-sections expanded using dual shields (detailed in Section 3), the system combines segment lining, servo struts, and support trusses in a way that works for both permanent and temporary applications. In special-shaped segment structures, servo strut members are embedded to facilitate mechanical load transfer and are designed as permanent components retained during construction and operation. These components remain locked under normal conditions but can be activated for stress regulation when required. Additionally, temporary servo strut carriages can be implemented during construction to provide comprehensive structural bracing, ensuring stability.

2) Sensing units

Sensing units, serve as the “neural terminals” of the structural agent, detecting dynamic changes in structural stress states caused induced by variations in the loading environment and providing data support for subsequent active control strategies. As shown in Fig. 3, key structural members are equipped with laser sensors, tilt sensors, and mechanical sensors to monitor deformation of support members and segment convergence.

The placement of sensing units requires optimized design based on structural mechanics and construction requirements. Typical locations include segment joints, critical nodes of servo struts, and high-stress zones (areas most sensitive to mechanical changes). Permanent servo components incorporate embedded sensors integrated with structural members to ensure stability throughout construction and operation phases.

Real-time data transmission is achieved through dedicated signal lines connecting sensing units to servo control units, enabling analysis of displacement and axial force measurements.

3) Control units and systems

Servo components function as control units of structural agent, comprising support heads, hydraulic cylinders, pressure sensors, displacement sensors, and base plates (Fig. 4). High-strength bolts connect support heads to tunnel segments, with additional steel bolt reinforcement at segment joints. For non-circular special-shaped segments, support heads are strategically positioned at geometric transition zones and critical joints. Each head incorporates pressure and displacement sensors, field-tested for real-time servo parameter adjustments. Precise hydraulic pressure regulation allows the support heads to provide accurate support forces, achieving simultaneous optimization of structural stress and deformation.

The control system integrates servo struts with programmable logic controllers (PLC). Servo struts offer adjustable axial load capacity [24–29], while PLCs [48,49] employ reinforcement learning algorithms (detailed in Subsection 2.2) to dynamically modify support forces and truss configurations based on real-time monitoring data, enhancing structural stability and safety.

2.2 Self-adaptive control methodology

Following the operational framework of environmental parameter perception, intelligent reinforcement learning, and action strategy generation, the structural agent implements structural perception and active control through servo struts and PLCs, establishing a coordinated structure-perception-control system (Fig. 5). The control mechanism operates through the following process.

1) Environment. Structural state perception. Displacement and pressure sensors at support head locations continuously monitor strain, displacement, and axial forces in servo struts and tunnel segments, capturing real-time structural responses to enable immediate perception of segment behavior.

2) Learning. Control strategy learning. PLCs compare monitoring real-time data against preset thresholds (e.g., maximum allowable displacement and axial force safety limits) while analyzing correlations between axial force and displacement. Reinforcement learning algorithms determine required adjustments to jacking displacement and thrust force, generating optimized servo control parameters.

3) Action. Active structural control. Hydraulic pressure regulation in cylinder precisely controls support head displacement and thrust magnitude. Dynamic adjustments at segment joints and support nodes actively compensate for additional stresses induced by ground deformation, maintaining force equilibrium while mitigating issues such as segment misalignment and joint separation caused by localized stress concentrations.

The servo control strategy learning algorithm is realized through a reinforcement learning network model, as shown in Fig. 6. Data on servo load application are collected from servo struts. Numerical simulations or physical experiments are employed to model tunnel environmental states. The tunnel environment state includes the structure’s own deformation and stress, as well as the surrounding water and soil loads and applied active loads (as shown in Fig. 1). The environmental observation state provides input observations for the reinforcement learning value network and policy network. Using these inputs and predefined policy gradients, iterative training updates the value and policy networks until predefined training criteria, including structural displacement and constraint conditions, are satisfied. Continuous updates based on environmental reward-penalty feedback optimize policy network parameters until the reward values stabilize and converge, ultimately outputting servo parameters generated by the policy network.

In practical applications, as tunnel excavation progresses, continuous monitoring of strain, displacement, and axial force parameters. Dynamic adjustments to servo control parameters based on stress variations during each construction step achieve uniform stress distribution and minimal deformation throughout the construction process.

3 Junction zone intelligent undercutting excavation construction method

3.1 Undercutting excavation method for junction zone

To address the complex connection configurations of ramp junctions in dense urban underground roads, a novel undercutting excavation method is proposed through integrating special-shaped cross-section SSA (see Fig. 7 for schematic diagram of underground road junction zone).

The construction stages are as follows (Fig. 8).

1) Construct main and ramp tunnels using shield tunneling (Fig. 8(a)), with smooth transitions at junction sections. Thicken junction interfaces and embed pipe jacking guide frames to ensure precise positioning of jacking pipes.

2) Pipe curtain structures are constructed at transition interface cross-sections as shown in Fig. 8(b), incorporating frozen pipes for ground consolidation through soil freezing reinforcement.

3) Deploy special-shaped cross-section servo struts or support trolleys within tunnels in junction zones as shown in Fig. 8(c), dynamically adjusting support forces based on real-time segment stress and convergence variations.

4) Curve pipe jacking method is employed to thrust arched beam structures with box sections from main tunnels into ramp tunnels, as shown in Fig. 8(d). Outer jacking pipes serve as temporary supports, while embedded frozen pipes freeze surrounding soil to form frozen soil curtains.

5) Upon completion of jacking, inner frozen pipes are removed, while outer pipes are remained as permanent lining skeletons. The combined action of the frozen soil curtain and segments ensures tunnel stability and load transfer, as shown in Fig. 8(e).

6) Construction progresses incrementally inward, excavating soil within the tunnel while assembling steel-reinforced concrete segments. The width and curvature of these segments are gradually adjusted during advancement to form a smooth transition in junction zones.

7) End walls are constructed at tunnel transition interface section junctions to maintain overall tunnel stability through complete enclosure.

8) Following removal of residual soils in junction zones, temporary servo struts are dismantled while permanent struts are retained as part of operation phase (Figs. 8(f) and 8(h)), achieving full undercutting excavation of special-shaped cross-sections in junction zones (Fig. 8(g)).

3.2 Shield-curve pipe jacking-freezing-structure agent

To enhance construction safety and efficiency, this section proposes a coordinated construction technology integrating shield tunneling, curve pipe jacking, ground freezing, and the SSA, as illustrated in Fig. 9.

Shield tunneling serves as the primary excavation method for main and ramp tunnels, with excavation conducted sequentially through junction zones. Curve pipe jacking method is the main technique for connecting the main and ramp tunnels. Ground freezing provides post-jacking reinforcement, while SSA act as the principal load-bearing component during construction, ensuring precise load distribution.

The coordinated construction of these four technologies adheres to the principles of spatiotemporal coupling and dynamic regulation. Shield tunneling establishes the primary tunnel alignment, while during the pipe jacking process, the SSA and freezing curtains jointly maintain dynamic mechanical equilibrium. The SSA actively compensates for construction-induced stress variations (e.g., jacking friction and surrounding rock stress fluctuations) and coupled stress fields, converting frost heave effects and jacking stresses into controlled structural prestress. Upon achieving mechanical stability in special-shaped cross-sections, temporary servo supports are removed, with permanent servo struts remained at tunnel junctions. Supplemental structural members are installed as needed to form a new system, ensuring long-term operational stability without compromising structural integrity.

4 Simulation of special-shaped servo structural agent

This section details the numerical modeling framework and implementation procedures of the SSA with special-shaped cross-sections. Finite element analysis using Abaqus software simulates the mechanical behavior of servo struts in a critical construction stage of the undercutting excavation of shield tunnel ramp junctions.

4.1 Simulation modeling and parameters value

The numerical model focuses on the most critical construction stage following curve pipe jacking in the novel undercutting excavation method. The tunnel cross-section employs special-shaped segments with localized thickening, referencing the staggered segment assembly pattern from Japanese urban underground road practices (Fig. 10).

Key modeling introduction are as follows.

1) Main tunnel (12 m external diameter) and ramp tunnel (9 m external diameter) segments are connected via 500 mm thick rectangular arched beams (curve pipe-jacking structures), with 1 m clearance between tunnels.

2) Segment bolts (30 mm diameter, 660 m length, 50 mm embedment, 60° installation angle) modeled using 2D truss elements.

3) The soil domain (60 m × 85 m, 22.5 m burial depth) is modeled using 2D shell elements.

4) The reinforcement structure at the joint, used for pipe jacking guidance and load transfer, adopts the same concrete material as the jacking pipes and segments.

5) The SSA comprises servo struts and support trusses. Servo struts are modeled by a “pre-applied axial force + truss element” approach (see Subsection 4.2 for details), while the support trusses are modeled with 2D truss elements featuring a cross-sectional area of 0.05 m².

Material parameters for the numerical simulation are shown in Table 1. Soil layer parameters are based on soft clay data from Shanghai, modeled as a homogeneous soil mass using the Mohr-Coulomb constitutive model with CPE4 2D shell elements (Tables 1 and 2). To reflect the stress and deformation patterns and trends of the structural system, the segment and pipe jacking structures utilize C50 concrete, following the relevant code [50], and are simulated using the concrete linear elastic constitutive model with CPE4I 2D shell elements (Table 1). The servo struts and bolts utilize Q235 steel and Grade 8.8 steel, respectively, and both materials employ a double-fold linear elastic constitutive model, with yield strengths of 235 and 640 MPa, and ultimate strengths of 441 and 800 MPa.

4.2 Servo struts modeling

The influencing factors of servo technology on deformation control include the geometric parameters of servo struts, construction processes, and the servo axial load application process [51]. In foundation pit engineering, servo combined load supports can suppress excavation-induced deformations at each respective construction stage [29]. As excavation depth increases, soil pressure gradually intensifies during foundation pit construction stages. In contrast, for shield tunnel construction stages, changes in cross-sectional shape represent the primary factor influencing structural deformation and loading conditions, as different cross-sectional forms necessitate distinct structural configurations. The special-shaped cross-section ramp junction project in this study primarily involves three main cross-sectional forms, as illustrated in Fig. 8; therefore, the post-curve pipe jacking construction stage and the most adverse scenario are selected as the primary objects of numerical simulation.

Three numerical approaches for simulating servo struts are evaluated.

1) Point load element method. Design axial forces are simulated using concentrated point load elements [24,25]. While suitable for foundation pits where strut self-weight is negligible, it fails to address axial force losses caused in tunnel engineering applications.

2) Thermal expansion analogy method. Axial force adjustments are simulated by altering strut lengths through heating and cooling based on the coefficient of thermal expansion [26]. Although this approach effectively replicates dynamic length variations during servo adjustments, the induced temperature-stress coupling conflicts with this study’s focus on post-adjustment mechanical effects.

3) Pre-stress and truss element method. Combining pre-applied axial forces with truss elements [52,53]. This method compensates for axial force losses through real-time monitoring and force replenishment. Initial strut forces are calculated using standard truss elements, compared with design values, and discrepancies are corrected via pre-stress adjustments.

After comparative analysis, the hybrid “pre-stress + truss element” method is adopted to simulate special-shaped SSA behaviors in this study.

4.3 Boundary conditions and constraints

The finite element model’s boundary conditions are implemented through displacement constraints on the soil: the soil base is restrained in both x and y directions, while the soil sides are restrained in the x-direction. Contact and constraint types between special-shaped cross-sectional structures include surface-to-surface contact between special-shaped segments and surrounding soil, surface-to-surface contact between adjacent segments, tie constraints between internal servo struts and segments, and embedded constraints (50mm embedded length) between circumferential bolts and segments. Detailed descriptions of structural constraints and contact types are listed in Table 3.

4.4 Simulation conditions

Given the complex asymmetric mechanical behavior of special-shaped cross-sections in shield tunnel ramp junctions, this study employs the structural-stratum method for numerical modeling, with an overburden depth of 20 m. To evaluate the performance of the SSA, the most critical construction stage post-pipe jacking is analyzed under three working conditions (Fig. 11).

1) No support condition. After pipe-jacking completion, segments directly bear the full earth pressure without any support.

2) Internal support condition. A rigid framework comprising diagonal and cross braces is installed in special-shaped cross-sections.

3) Servo support condition. Self-adaptive servo struts with support trusses dynamically adjust support forces (ranging from 50 to 200 kN) in special-shaped cross-sections.

The two-dimensional computational model for the construction stage after the advancement of the curved pipe jacking is shown in Fig. 12. It mainly includes soil, segment structure, pipe jacking structure, support truss, and bolts, adopting three reinforcement conditions: no support, traditional internal support, and servo support. Displacement constraints in the x and y directions are applied to the bottom of the soil, and displacement constraints in the x direction are applied to the sides of the soil. The model includes 66338 mesh elements and 66889 nodes, with the soil near the tunnel structure being refined.

4.5 Numerical simulation results

Stress and displacement contours for the three working conditions are shown in Fig. 13. Following curve pipe jacking construction and geostress equilibrium, unsupported tunnel segments exhibited maximum compressive stress (3.82 MPa) and tensile stress (15.26 MPa) at the main tunnel segment-pipe joint and waist region, with peak displacement (25.11 mm) occurring at the tunnel invert. Critical control zones were identified at segment-pipe joints, waist, crown and invert. Under conventional internal support, compressive stress increased slightly to 4.29 MPa and tensile stress decreased to 7.33 MPa, with structural displacement declining to 11.33 mm. The servo support condition further optimized performance, achieving a compressive stress of 2.26 MPa, a tensile stress of 9.03 MPa, and a displacement of 7.67 mm, demonstrating superior deformation control compared to conventional methods. Both supports significantly mitigated stresses, with servo systems exhibiting enhanced geometric stability.

5 Comparison and discussions

5.1 Segment structural mechanical behavior

5.1.1 Segment deformation

The deformation modes of special-shaped cross-section segments in junctions are shown in Fig. 14, where positive values indicate outward deformation, whereas the negative. Under unsupported condition, segments exhibited asymmetric elliptical deformation with significant inhomogeneity and peak displacements due to local instability. Maximum deformations occurred at the waist regions (17.02 mm for main tunnel, 12.38 mm for ramp tunnel) and bottom zones (24.93 mm for main tunnel, 15.96 mm for ramp tunnel), characterized by outward deformation at the waist and inward deformation at the crown and invert.

The implementation of supports significantly mitigated deformations. For the main tunnel, which exhibited greater radial deformation, the peak radial deformation under servo supports (7.65 mm) demonstrated a 30.9% reduction compared to conventional internal supports (11.07 mm). While conventional supports failed to address asymmetric deformation, particularly at the mechanically vulnerable 45°waist zones, servo support demonstrates 1.8 times greater effectiveness in radial deformation control, achieving more uniform stress redistribution.

The displacement comparison for main tunnel segments in special-shaped cross-sections is shown in Table 4. Under the unsupported condition, peak lateral and vertical displacements at the segment waist and bottom reached 17.14 and 25.07 mm, respectively. Both conventional internal supports and servo supports effectively controlled convergence deformations: conventional supports reduced lateral and vertical displacements to 6.1 and 11.32 mm (approximately 50% reduction), while servo supports further reduced displacements to 3.98 and 7.66 mm (approximately 70% reduction). Servo support demonstrated significantly superior deformation control compared to conventional methods, particularly in mitigating asymmetric deformation at mechanically vulnerable zones.

5.1.2 Segment stress

The bending moments distributions of special-shaped cross-section segments in junctions are shown in Fig. 15, where positive values indicate the outer in compression and the inner in tension, whereas negative. Under the unsupported condition, segments exhibited significant asymmetric bending moment distributions. Positive moments were concentrated at the waist regions and crown joints, peaking at 746.4 (main tunnel) and 666.7 kN·m (ramp tunnel) at the outer waist. Negative moments localized near the crown and invert joints, reaching −647.2 (main tunnel) and −442.5 kN·m (ramp tunnel). Critical mechanical vulnerabilities were observed at the 45°outer waist zones due to high moment concentrations.

The implementation of supports altered global bending moment patterns and shifting inflection points. Conventional supports reduced negative moments at joints but retained substantial positive moments at the waist. In contrast, servo supports achieved more uniform stress distributions, reducing bending moments at the 45°outer waist zones and mitigating stress concentrations in mechanically vulnerable regions.

The axial force distributions of special-shaped cross-section segments in junctions are shown in Fig. 16. Under the unsupported condition, abrupt axial force variations were observed at segment joints, exhibiting a pronounced asymmetric pear-shaped distribution. The implementation of supports improved force uniformity and reduced peak axial forces at the outer waist regions, with servo support demonstrating superior performance in mitigating force concentrations compared to conventional support.

The internal force comparison for main tunnel segments in special-shaped cross-sections is shown in Table 5. Under the unsupported condition, the maximum positive bending moment at the waist and minimum negative bending moment at joints reached 746.4 and −647.2 kN·m, respectively, with a peak axial force of 3431.2 kN. With conventional internal supports, these values reduced to 405.3 kN·m (45.7% reduction), −333.8 kN·m, and 2944.4kN (14.2% reduction). Servo support further decreased the bending moments to 204.8 (55.7% reduction) and −330.7 kN·m, and axial force to 2733.3 kN (20.3% reduction), demonstrating superior reinforcement efficacy. The reduced circumferential unevenness of axial forces confirmed that servo supports achieved more uniform stress redistribution through active force compensation, outperforming conventional methods in structural homogenization.

5.2 Pipe jacking structural mechanical behavior

5.2.1 Pipe jacking deformation

The deformation comparison of pipe-jacking structures in junction zone are shown in Table 6. Under the unsupported condition, maximum lateral and vertical deformations reached 2.86 and 7.67 mm, respectively. Conventional internal supports reduced these values to 1.31 (lateral) and 2.09 mm (vertical), while servo supports further decreased deformations to 0.26 (lateral) and 2.62 mm (vertical). Servo supports achieved deformation control efficacy exceeding 70%, significantly outperforming conventional methods and demonstrating its superior capability in mitigating structural deformations of pipe-jacking structure.

5.2.2 Pipe jacking stress

The internal force comparison of pipe-jacking structures in junction zone are shown in Table 7. Under the no-support condition, the maximum bending moment and axial force reached 82.94 kN·m and 691.7 kN, respectively. Conventional internal supports increased the bending moment slightly to 92.44 kN·m while reducing axial force to 245.7 kN (64.5% reduction), whereas servo support decreased these values to 74.58 kN·m (bending moment) and 201.9 kN (axial force, 70.8% reduction). The results indicate that supports have limited influence on bending moments but significantly mitigate axial forces. Servo supports demonstrate superior control over internal forces by actively compensating axial load transfer, reducing stress concentrations at joints and minimizing force transmission to supporting structures.

5.3 Support trusses mechanical behavior

5.3.1 Support nodes

A comparison of lateral and vertical displacements at support nodes in the special-shaped cross-sections of main tunnel junction zones (Table 8) reveals that conventional internal supports, relying primarily on structural rigidity, exhibit displacement magnitudes ranging from 1.51 to 10.63 mm. In contrast, servo supports achieve refined displacement control through active force regulation, limiting truss node displacements to a narrower range of 1.4 to 9.33 mm, achieving an approximate 26.8% reduction in peak displacements. This demonstrates the superior efficacy of servo supports in mitigating deformation at critical structural nodes compared to conventional methods.

5.3.2 Support members

The axial forces in the support members within the main tunnel are significantly greater than those in the ramp tunnel. Therefore, a comparison is made of the axial forces in the diagonal and vertical rigid supports under traditional support methods and the servo support truss members. In conventional internal supports, the main tunnel’s diagonal, vertical, and lateral supports experience axial forces of 65.13, 1014.06, and 457.36 kN, respectively. The servo supports reduce the axial force range in the main tunnel support trusses to 58.03 to 712.76 kN, achieving approximate an 29.7% reduction in peak axial forces compared to conventional supports. This reduction lowers the steel material requirements and structural rigidity demands, facilitating a more lightweight design while maintaining load-bearing capacity.

5.4 Engineering implications and challenges

The construction constraints of ramp junctions in dense urban underground road systems pose significant challenges for urban renewal [2]. The SSA proposed in this study provides a novel strategy for ensuring structural deformation control in full undercutting excavation of shield tunnels with special-shaped cross-sections. This control approach, incorporating adaptive mechanisms aligned with emerging reinforcement learning concepts, demonstrates through numerical simulations that, compared to conventional support structures, SSA effectively reduces radial deformation peaks in segment (by approximately 30%) and internal forces (by approximately 10%) through multi-point active loading. By integrating permanent-temporary servo struts and support trusses, it optimizes force distribution in mechanically vulnerable zones (e.g., 45° waist regions), reducing node displacements and axial forces in support trusses (by approximately 30%). This comprehensive analysis of SSA with special-shaped cross-sections facilitates lightweight structural designs for urban underground road reconstruction, enhancing the value of active load equipment, such as servo technology, in future shield tunnel projects.

Despite the demonstrated mechanical performance in numerical simulations, the proposed method simplifies the interaction and properties of the structure and soil environment. Moreover, the application of servo structural systems in shield tunnel projects remains limited, lacking field data validation to assess SSA performance in practical engineering. Therefore, future research will involve field tests in the ongoing Shanghai North Cross Passage Project in China, combined with model tests, numerical simulations, and reinforcement learning algorithms for comprehensive validation.

6 Conclusions

This study focuses addresses challenges of ensuring structural safety in special-shaped cross-section shield tunnel ramp junctions under full undercutting excavation, proposing a SSA and a full undercutting construction method. The deformation and stress behavior of this structural system were conducted using numerical simulations. Main conclusions are as follows.

1) SSA for special-shaped cross-sections achieves adaptive control strategies through the integration of sensing, control, and structural systems. Combined with the full underground excavation method incorporating servo supports and a permanent-temporary integrated structural system, it provides a novel paradigm for the design and construction of complex cross-section structures in urban underground spaces.

2) Servo struts, compared to conventional internal supports, reduce radial deformation peaks in tunnel segments by 30% through multi-point active force application; they decrease deformation peaks in pipe jacking structures by 20%; suppress deformations in mechanically vulnerable zones and support nodes; and enhance the overall deformation control capability of the structure.

3) Servo struts, compared to conventional internal supports, actively compensate for asymmetric loads, reducing maximum bending moments in tunnel segments by 10% and maximum axial forces by 6%; decreasing maximum axial forces in pipe jacking structures by 10%; reducing axial forces in support members by 30%, thereby decreasing steel consumption; and improving internal force transmission at junctions.

4) While numerical simulations demonstrate SSA’s efficacy, the lack of field validation limits its practical assessment. Future research should focus on integrating RL algorithms to optimize servo parameters dynamically, adapting to varying cross-sectional geometries and geotechnical conditions, and combining model tests with analytical solutions for comprehensive validation.

References

[1] Broere W . Urban underground space: Solving the problems of today’s cities. Tunnelling and Underground Space Technology, 2016, 55: 245–248

[2] Fang TGuan LFeng L. Research on undercutting method of shield-driven underground expressways in urban core zones with soft soil layers. Chinese Journal of Underground Space and Engineering, 2022, 18(S2): 792–799, 812 (in Chinese)

[3] Yang H , Zhou W , Liu C , Jiang X , Yu L . Large-scale shaking table test of seismic response laws for a shallow double-arch tunnel under unsymmetrical pressure with a damping layer. Advances in Civil Engineering, 2020, 2020(1): 8899276

[4] Hsiao F Y , Wang C L , Chern J C . Numerical simulation of rock deformation for support design in tunnel intersection area. Tunnelling and Underground Space Technology, 2009, 24(1): 14–21

[5] Li Y , Jin X , Lv Z , Dong J , Guo J . Deformation and mechanical characteristics of tunnel lining in tunnel intersection between subway station tunnel and construction tunnel. Tunnelling and Underground Space Technology, 2016, 56: 22–33

[6] Tan Z , Li Z , Tang W , Chen X , Duan J . Research on stress characteristics of segment structure during the construction of the large-diameter shield tunnel and cross-passage. Symmetry, 2020, 12(8): 1246

[7] Lu P , Yuan D , Luo W , Jin D , Liu M . Centrifugal model test study on the mechanical characteristics of shield tunnels influenced by different types of openings for cross passages. Applied Sciences, 2022, 12(13): 6421

[8] Ma C , Peng F . Some aspects on the planning of complex underground roads for motor vehicles in Chinese cities. Tunnelling and Underground Space Technology, 2018, 82: 592–612

[9] Cao L , Fang Q , Zhang D , Chen T . Subway station construction using combined shield and shallow tunnelling method: Case study of Gaojiayuan station in Beijing. Tunnelling and Underground Space Technology, 2018, 82: 627–635

[10] Hiroshi D , Akira S , Miyama D . Design and construction of enlarging shield tunnel sections of large dimensional shield tunnels for the on-open-cut method. Tunnelling and Underground Space Technology, 2006, 21(3): 249

[11] Namikawa YK TTerashima JInoue YKoseki MMiyashita Matsumoto . Construction of expressway branch junction structure using non-cut-and- cover enlargement method to combine two shield tunnels in sedimentary soft rock. In: ITA-AITES Wond Tunnel Congress. London: Taylor & Francis Group, 2016

[12] LI YZhang KHuang CLi ZDeng M. Analysis of surface subsidence of tunnel built by pipe-roof pre-construction method. Rock and Soil Mechanics, 2011, 32(12): 3701–3707 (in Chinese)

[13] LI YZhang KHuang C. Monitoring of lining structure of tunnels built by using pipe-roof pre-construction method. Chinese Journal of Geotechnical Engineering, 2012, 34(8): 1541–1547 (in Chinese)

[14] Zhang P , Ma B , Zeng C , Xie H , Li X , Wang D . Key techniques for the largest curved pipe jacking roof to date: A case study of Gongbei tunnel. Tunnelling and Underground Space Technology, 2016, 59: 134–145

[15] Liu X , Liu Y , Qu W , Tu Y . Internal force calculation and supporting parameters sensitivity analysis of side piles in the subway station excavated by Pile-Beam-Arch method. Tunnelling and Underground Space Technology, 2016, 56: 186–201

[16] Xu X , Li Z , Fang Q , Zheng H . Challenges and countermeasures for using pile-beam-arch approach to enlarge large-diameter shield tunnel to subway station. Tunnelling and Underground Space Technology, 2020, 98: 103326

[17] Zhang M , Liu Y , Fan L , Li P . Performance of constructing a double-deck subway station by combining the shield method and cavern–pile method. Tunnelling and Underground Space Technology, 2017, 67: 120–131

[18] Yang Y , Liao S , Liu M , Wu D , Pan W , Li H . A new construction method for metro stations in dense urban areas in Shanghai soft ground: Open-cut shafts combined with quasi-rectangular jacking boxes. Tunnelling and Underground Space Technology, 2022, 125: 104530

[19] Zheng Z , Xue X , Su D , Chen J , Qiu T , Chen P , Xie D , Su L , Chen W , Huang S . et al. Instability mechanism and reinforcement measures for segments of Ultra-Large diameter shield tunnels when constructing cross passages by mechanical methods. Tunnelling and Underground Space Technology, 2024, 154: 106125

[20] Jia P , Nie Y , Shi P , Jiang X , Lu B , Zhao W . Flexural performance of a novel pipe-roof structure and optimization of key parameters. Journal of Constructional Steel Research, 2022, 199: 107594

[21] Liu X , Gao Y , Ramos G , Turmo J , Zhu Y . Full-scale tests on segmental lining with mechanically cut cross-passage opening. Tunnelling and Underground Space Technology, 2024, 151: 105874

[22] Wang SZhu YYu Z. Design of automatic hydraulic support trolley for overlapped shield tunnel based on SolidWorks Simulation. Tunnel Construction, 2019, 39(2): 294–301 (in Chinese)

[23] Di H , Guo H , Zhou S , Chen J , Wen L . Investigation of the axial force compensation and deformation control effect of servo steel struts in a deep foundation pit excavation in soft clay. Advances in Civil Engineering, 2019, 2019(1): 5476354

[24] Li M , Demeijer O , Chen J . Effectiveness of servo struts in controlling excavation-induced wall deflection and ground settlement. Acta Geotechnica, 2020, 15(9): 2575–2590

[25] Nangulama H K , Jian Z . Deformation control monitoring of basement excavation at field construction site: A case of hydraulic servo steel enhancement geotechnology. Advances in Civil Engineering, 2022, 2022(1): 6234581

[26] Xu C , Hou J , Liu B , Lei F , Song L . Research on coordinated relationship between deformation and force in shaft foundation pit support structures. Buildings, 2024, 14(11): 3438

[27] Zhao Y , Chen X , Hu B , Huang L , Lu G , Yao H . Automatic monitoring and control of excavation disturbance of an ultra-deep foundation pit extremely adjacent to metro tunnels. Tunnelling and Underground Space Technology, 2023, 142: 105445

[28] Sun YLiu MLu PChen RPang K. Review of key technical issues in construction of cross passage in shield tunnel. Modern Tunnelling Technology, 2021, 58(S1): 293–302 (in Chinese)

[29] Nangulama H K , Jian Z , Xiao Z , Zhu J , Yuan F . Stage-by-stage control effect field analysis of steel material servo enhanced support system on lateral displacement and bending moment during deep basement excavation. Case Studies in Construction Materials, 2022, 16: e01068

[30] Haundi T , Nangulama H K , Mbewe V R . Site characterisation, deep basement support, construction, and deformation control. Geotechnical and Geological Engineering, 2024, 42(3): 1611–1622

[31] Zheng G , Huang J , Diao Y , Yan Y , Jiao C , Zhang L , Jia J , Peng J . Field investigation of deflection characteristics and control strategies of a short floating pile adjacent to deep excavation. Journal of Geotechnical and Geoenvironmental Engineering, 2025, 151(4): 05025001

[32] Li S , Wang F , Le Y , Que Q , Su Y , Lin H . Design and deformation pattern simulation of deep excavation support structures. Frontiers in Earth Science, 2024, 12: 1416957

[33] Dong X , Zhou F , Yoo C , Wang X , Zhu R . Suggested deformation control approach for braced excavations incorporating corner and soil arching effects. Tunnelling and Underground Space Technology, 2026, 167: 107059

[34] Guo X , Wang B , Wang Z , Yu W , Ma Z , Yang T . Application of the microclamped Fiber Bragg Grating (FBG) sensor in rock bolt support quality monitoring. Advances in Civil Engineering, 2020, 2020(1): 3419835

[35] Liu Q , Guo J , Liu L , Huang K , Tian W , Li X . Optimization analysis of smart steel-plastic geogrid support for tunnel. Advances in Civil Engineering, 2020, 2020(1): 6661807

[36] Zhang X , Zhu H , Jiang X , Broere W . Distributed fiber optic sensors for tunnel monitoring: A state-of-the-art review. Journal of Rock Mechanics and Geotechnical Engineering, 2024, 16(9): 3841–3863

[37] Li X , Xue W , Fu C , Yao Z , Liu X . Mechanical properties of high-performance steel-fibre-reinforced concrete and its application in underground mine engineering. Materials, 2019, 12(15): 2470

[38] You X , Yan G , Alnowibet K A , El-Meligy M . Artificial neural network-enhanced mathematical simulation based on Carrera unified formulation for dynamic analysis and structural integrity assessment of advanced nanocomposites reinforced tunnel structures. Mechanics of Advanced Materials and Structures, 2025, 32: 4882–4892

[39] Li T , Shu J , Wang Y . Deformation prediction of underground engineering support structures via the ST-CNN-LSTM model. Journal of Civil Structural Health Monitoring, 2025, 15(6): 1901–1919

[40] Barbati M , Bruno G , Genovese A . Applications of agent-based models for optimization problems: A literature review. Expert Systems with Applications, 2012, 39(5): 6020–6028

[41] Mazzetto S . Interdisciplinary Perspectives on Agent-Based Modeling in the Architecture, Engineering, and Construction Industry: A Comprehensive Review. Buildings, 2024, 14(11): 3480

[42] Sun M , Staszewski W J , Swamy R N . Smart Sensing Technologies for Structural Health Monitoring of Civil Engineering Structures. Advances in Civil Engineering, 2010, 2010(1): 724962

[43] Korkmaz S . A review of active structural control: challenges for engineering informatics. Computers and Structures, 2011, 89(23–24): 2113–2132

[44] Xi Z , Chen W , Guo X , He W , Ding Y , Hong B , Zhang M , Wang J , Jin S , Zhou E . et al. The rise and potential of large language model based agents: A survey. Science China. Information Sciences, 2025, 68(2): 121101

[45] Azad M M , Kim S , Cheon Y B , Kim H S . Intelligent structural health monitoring of composite structures using machine learning, deep learning, and transfer learning: A review. Advanced Composite Materials, 2024, 33(2): 162–188

[46] Seventekidis P , Giagopoulos D , Arailopoulos A , Markogiannaki O . Structural Health Monitoring using deep learning with optimal finite element model generated data. Mechanical Systems and Signal Processing, 2020, 145: 106972

[47] Panda J , Chopra M , Matsagar V , Chakraborty S . Continuous control of structural vibrations using hybrid deep reinforcement learning policy. Expert Systems with Applications, 2024, 252: 124075

[48] Wang S , Xu L , Zhang X , Long L , Zhuang X . Development and field analysis of a novel servo concrete bracing system for deep foundation pit excavation. Buildings, 2024, 14(6): 1674

[49] Wang C , Lin G , Yao H , Ding H , Tang X , Lin Z , Wang X . Observation and analysis of deformation control performance of hydraulic servo steel support for a soft soil pit—A case study of a pit project in Hangzhou. Buildings, 2025, 15(7): 1002

[50] GB/T 50010-2010. Code for Design of Concrete Structures. Beijing: China Architecture and Building Press, 2010

[51] Sun JBai T. Control methods for lateral deformation of retaining structure of deep foundation pit in soft soil. Tunnel Construction, 2019, 39(S2): 308–317 (in Chinese)

[52] Wang ZLi ZCai LLi HXu H. Research on application technology of steel support servo system for foundation pit. Tunnel Construction, 2020, 40(S2): 10–22 (in Chinese)

[53] Li HLiao STang YShen M. Deformation characteristics and stress path analysis of divided foundation pit in soft soils. Journal of Tongji University (Natural Science), 2021, 49(8): 1116–1127 (in Chinese)

RIGHTS & PERMISSIONS

Higher Education Press