Deformation risk discrimination method for soft surrounding rock tunnel based on shear dilation-shear constitutive

Abstract

The present application relates to the field of tunnel engineering geology and support design, and discloses a soft surrounding rock tunnel deformation risk discrimination method based on shear dilation-shear constitutive, comprising the following steps: constructing a nonlinear coupling model between the shear dilation angle and the shear stress, normal stress and joint parameters, and obtaining the shear dilation angle; establishing a volume strain rate discrimination formula; inverting the initial stress tensor field; reconstructing the irregular tunnel boundary; constructing a risk grade discrimination model and outputting the risk grade; matching the support scheme such as support stiffness, anchor rod parameters and spray layer thickness according to the risk grade; establishing a model to predict the risk trend; generating support adjustment suggestions; collecting monitoring data to dynamically correct the model parameters; and integrating all modules in the deployment system. The present application introduces the coupling relationship between the shear dilation angle and the shear strength, establishes a coupled constitutive discrimination model, and constructs a risk grading system and support correction strategy associated with support response, thereby realizing active early warning and dynamic adjustment of support stiffness in high-risk sections.

Description

Technical Field

[0001] This invention relates to the field of tunnel engineering geology and support design technology, specifically to a risk assessment method for soft surrounding rock tunnel engineering, which introduces shear dilatation-shear coupling relationship to realize the classification of stratum deformation risk and support correction decision. Background Technology

[0002] Currently, tunnels in weak surrounding rock generally face problems such as large deformation, crown settlement, and arch bulging. Traditional deformation judgment methods mostly rely on empirical formulas, initial surrounding rock classification, or linear extrapolation of monitoring values. These methods fail to accurately reflect the deformation evolution mechanism of the surrounding rock under complex initial geostress, joint structures, and construction disturbances. In particular, they neglect the dominant role of shear dilatation in the plastic evolution stage, leading to delayed support design and measures, which in turn cause arch failure or support failure.

[0003] Tunnels in weak surrounding rock are widely distributed in engineering practice under geological conditions such as mudstone, shale, carbonaceous slate, and fracture zones. Due to their poor structural integrity, well-developed joints and fissures, low strength, and significant influence from water softening, they are highly susceptible to irreversible and significant deformation problems under excavation disturbance or stress redistribution. Specifically, these problems manifest as follows:

[0004] Crown settlement: The settlement is caused by the imbalance between the gravity of the overlying surrounding rock and the support reaction force.

[0005] Arching and bulging: Due to the weakening of lateral constraints, the rock mass expands towards the free surface;

[0006] Face convergence: Under the release of confining pressure, the formation ahead experiences shear-dilatation failure;

[0007] Delayed support deformation: The initial support parameters are based on prior grading experience and fail to fully respond to the actual deformation trend, often leading to accidents such as anchor pull-out, spraying layer cracking, or even lining perforation.

[0008] Although geological classification methods (such as the Q system and RMR system) or linear discrimination methods based on monitoring data are widely used in current engineering, the following technical bottlenecks exist:

[0009] 1. Initial geological classification ignores dynamic evolution process: The state of the surrounding rock is not static and unchanging. Under construction disturbance and confining pressure release conditions, it will undergo strain softening and strength degradation. Existing classification methods cannot reflect this process.

[0010] 2. Neglect of dilatation: During shearing, weak rock masses generally exhibit volume expansion (dilatation), which is an important precursor to the expansion of the plastic zone. However, most constitutive models do not explicitly consider the evolution of the dilatation angle.

[0011] 3. Rough boundary and geostress modeling: Traditional two-dimensional simplified or assumed homogeneous models ignore the influence of irregular cross sections, joint control zones and multi-source initial stresses, resulting in a large gap between the discrimination model and reality.

[0012] 4. Disconnect between risk and support: Current support measures are often "corrected" after the fact rather than "feedforward design" based on risk, and lack a systematic risk level and support parameter response mapping mechanism. Summary of the Invention

[0013] Based on the above description, this invention provides a method for judging the deformation risk of tunnels in weak surrounding rock based on dilatation-shear constitutive model. It proposes to introduce the coupling relationship between dilatation angle and shear strength, establish a coupled constitutive discrimination model, and construct a risk classification system and support correction strategy associated with support response to achieve proactive early warning and dynamic adjustment of support stiffness in high-risk sections.

[0014] The technical solution of this invention to solve the above-mentioned technical problems is as follows: A method for determining the deformation risk of tunnels in weak surrounding rock based on dilatation-shear constitutive model, comprising the following steps:

[0015] A nonlinear coupling model between the dilatation angle and shear stress, normal stress, and joint parameters is constructed to obtain the dilatation angle. A volumetric strain rate discrimination formula is established based on the dilatation angle. The initial geostress tensor field is inverted using multi-source monitoring data. The irregular tunnel boundary is reconstructed using the Bezier or non-uniform rational B-spline method. A risk level discrimination model integrating multiple parameters is constructed to output risk levels I to IV. Support schemes such as support stiffness, anchor bolt parameters, and shotcrete thickness are matched according to the risk level.

[0016] Based on time-series monitoring data, establish long short-term memory or backpropagation neural network models to predict risk trends;

[0017] Based on the prediction results, support adjustment suggestions are generated;

[0018] Collect monitoring data to dynamically adjust model parameters;

[0019] Integrate each module into a deployment system with monitoring, calculation, and output functions.

[0020] This invention constructs a method for identifying deformation risks in tunnels in weak surrounding rock, with dilatation-shear coupling mechanism as the core and integrating multi-source data inversion, intelligent discrimination algorithm, support linkage strategy and model self-correction capability. This method enables proactive identification, prediction and early warning and support response for complex surrounding rock conditions, significantly improving the level of intelligent construction and engineering safety.

[0021] Based on the above technical solution, the present invention can be further improved as follows.

[0022] Furthermore, the nonlinear coupling model between the dilatation angle and shear stress, normal stress, and joint parameters is an empirical formula, which characterizes the relationship that the dilatation angle ψ increases with increasing shear stress and decreases with increasing normal stress.

[0023] Furthermore, the joint parameters include the joint roughness coefficient and the compressive strength of the joint wall surface, to reflect the influence of the joint surface morphology on the shear dilatation angle ψ;

[0024] The nonlinear coupling model uses an exponential or hyperbolic function to fit the relationship between the dilatation angle ψ and the shear stress, normal stress, and joint parameters, and the model coefficients are determined by regression from experimental data.

[0025] Furthermore, the volumetric strain rate discrimination formula based on the dilatation angle is: Volumetric strain rate = Shear strain rate × tan(ψ), where ψ is the dilatation angle;

[0026] Using the volumetric strain rate discrimination formula, when the calculated volumetric strain rate is positive, the risk of dilatational deformation of the surrounding rock is determined, and when it is negative, the trend of shear contraction deformation of the surrounding rock is determined.

[0027] Furthermore, the multi-source monitoring data includes surrounding rock displacement monitoring data, rock mass stress monitoring data, and geological survey data, and the initial geostress tensor field is obtained by inverting the above multi-source data.

[0028] The inversion of the initial geostress tensor field is achieved by establishing a numerical model that includes the effects of tunnel excavation and using an iterative algorithm to adjust the initial stress value until the deformation results calculated by the model match the multi-source monitoring data.

[0029] Furthermore, the reconstruction of the irregular tunnel boundary adopts the Bezier curve fitting method, which performs smooth fitting on the collected discrete points of the tunnel contour to obtain a continuous and smooth tunnel boundary curve.

[0030] The reconstruction of the irregular tunnel boundary adopts a non-uniform rational B-spline curve model. By introducing control vertex and node vectors, the tunnel boundary is accurately described to adapt to the complex curve tunnel boundary shape.

[0031] Furthermore, the input parameters of the multi-parameter risk level discrimination model include shear dilatation angle, volumetric strain rate, tunnel surrounding rock displacement, support internal force, and groundwater seepage pressure, wherein the influence weight of each parameter on the risk level is determined through model training;

[0032] Establish a linkage between the risk level and tunnel support parameters, and pre-store the corresponding support schemes for risk levels I to IV.

[0033] Furthermore, the risk trend prediction model based on time-series monitoring data is a long short-term memory neural network model, which receives monitoring parameters input in time series and outputs predicted values ​​of risk level or risk index for future time periods.

[0034] Furthermore, the risk trend prediction model is a model based on an error backpropagation neural network. It uses historical monitoring data to train and establish a nonlinear mapping relationship between input monitoring parameters and subsequent risk levels in order to predict future changes in risk levels.

[0035] The backpropagation neural network model includes an input layer, at least one hidden layer, and an output layer. The input layer obtains the values ​​of multiple current monitoring parameters, and the output layer generates the corresponding predicted risk level or risk indicator.

[0036] Furthermore, the support adjustment suggestions generated based on the prediction results include specific measures to adjust the support parameters for different predicted risk levels: when the predicted risk level increases, suggestions are output to increase the stiffness of the support structure, add or lengthen anchor bolts, and thicken the shotcrete layer; when the predicted risk level decreases, suggestions are output to delay or reduce support reinforcement measures.

[0037] Compared with the prior art, the technical solution of this application has the following beneficial technical effects:

[0038] 1. By introducing the shear dilatation angle into the main control variable for risk discrimination, the limitations of existing models that do not consider the plastic expansion mechanism are overcome, and constitutive-driven discrimination of deformation trends is achieved.

[0039] 2. Achieve integrated modeling of multi-source stress and geometric conditions, significantly improving the geometric accuracy and stress fitting accuracy of risk identification and numerical analysis.

[0040] 3. Construct a linkage mechanism between risk level and support parameters to achieve a logical closed loop from "judgment" to "adjustment" and improve the initiative and matching of support response.

[0041] 4. Introducing deep learning methods for risk trend prediction enables feedforward early warning capabilities, effectively extending the response window and enhancing prevention capabilities.

[0042] 5. It possesses the ability to self-correct the model and deploy the system, enabling it to adapt to the long-term evolution of the construction environment and form a sustainable intelligent support system. Attached Figure Description

[0043] Figure 1 A flowchart illustrating the overall process of the method for determining the deformation risk of tunnels in weak surrounding rock based on dilatation-shear constitutive model provided in this embodiment of the invention.

[0044] Figure 2The initial geostress tensor inversion modeling logic diagram for the deformation risk discrimination method of weak surrounding rock tunnel based on dilatation-shear constitutive model provided in the embodiments of the present invention;

[0045] Figure 3 This is a structural diagram of a multi-parameter risk level discrimination model for a method for judging the deformation risk of tunnels in weak surrounding rock based on dilatation-shear constitutive model provided in an embodiment of the present invention.

[0046] Figure 4 A diagram showing the coupling relationship between dilatation angle and shear stress in a method for determining the deformation risk of tunnels in weak surrounding rock based on dilatation-shear constitutive model provided in this embodiment of the invention.

[0047] Figure 5 The BP neural network (static type) model structure diagram of the deformation risk discrimination method for weak surrounding rock tunnels based on dilatation-shear constitutive model provided in the embodiments of the present invention;

[0048] Figure 6 This invention provides a modular deployment and integration system for a method to determine the deformation risk of tunnels in weak surrounding rock based on dilatation-shear constitutive model. Detailed Implementation

[0049] To facilitate understanding of this application, a more complete description will be provided below with reference to the accompanying drawings, which illustrate embodiments of the present application. However, the present application can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that the disclosure of this application will be thorough and complete.

[0050] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application.

[0051] Example:

[0052] A method for determining the deformation risk of tunnels in weak surrounding rock based on dilatation-shear constitutive model includes the following steps:

[0053] A nonlinear coupling model between the dilatation angle and shear stress, normal stress, and joint parameters is constructed to obtain the dilatation angle.

[0054] A volumetric strain rate discrimination formula was established based on the dilatation angle; the initial geostress tensor field was inverted using multi-source monitoring data.

[0055] Irregular tunnel boundaries are reconstructed using Bezier or NURBS (Non-Uniform Rational B-Splines) methods;

[0056] Construct a risk level discrimination model that integrates multiple parameters and outputs risk levels from I to IV;

[0057] Support solutions, such as support stiffness, anchor bolt parameters, and spray layer thickness, are matched according to the risk level.

[0058] LSTM (Long Short-Term Memory) or BP (Backpropagation Neural Network) models are built based on time-series monitoring data to predict risk trends;

[0059] Based on the prediction results, support adjustment suggestions are generated;

[0060] Collect monitoring data to dynamically adjust model parameters;

[0061] Integrate each module into a deployment system with monitoring, calculation, and output functions.

[0062] More specifically:

[0063] 1) Nonlinear constitutive relationship between dilatation angle and shear stress:

[0064] In this invention, the shear failure behavior of weak surrounding rock is considered to be solely determined by shear strength parameters (such as cohesion). internal friction angle The shear stress is determined by the shear angle and is also significantly influenced by the volumetric change behavior during the evolution of the rock mass's internal structure—the "dilatation effect." Traditional Mohr-Coulomb and Hoek-Brown constitutive models are mostly based on the "failure criterion," neglecting the volumetric expansion characteristics during the rock mass's shear deformation process and making it difficult to identify the critical evolution signals in the plastic expansion stage. Therefore, this invention first starts with the constitutive modeling of the dilatation angle and proposes a nonlinear coupling relationship model between the dilatation angle and shear stress, normal stress, and joint parameters.

[0065] 1.1) Among them, the shear dilatation angle ( The shear stress (or normal stress) is used to measure the tendency of rock masses or joint surfaces to expand in volume during shearing. Its physical meaning is the angle of normal displacement caused by a unit shear displacement. Its value is affected not only by shear stress and normal stress, but also by the joint surface roughness (…). ) and shear strength ( Closely related to ).

[0066] Based on the results of in-situ direct shear tests and triaxial shear tests, it can be observed that the shear dilatation angle usually fluctuates within the range of 5° to 30°, and as the shear stress increases, the shear dilatation angle shows a certain upward trend, while the increase of normal stress plays a role in inhibiting the expansion of the shear dilatation angle.

[0067] 1.2) Nonlinear coupling modeling of shear dilatation angle

[0068] This invention proposes the following empirical formula for predicting the shear dilatation angle:

[0069]

[0070] in:

[0071] Shear expansion angle, in degrees;

[0072] Shear stress (MPa);

[0073] Normal stress (MPa);

[0074] : Joint surface roughness coefficient;

[0075] Shear strength of joint surface (MPa);

[0076] The fitting coefficients are determined by nonlinear regression using multiple sets of experimental data.

[0077] In regression data analysis of multiple engineering projects, typical fitting results are as follows:

[0078]

[0079] This model demonstrates the following patterns:

[0080] As shear stress increases, the angle of shear dilatation rises. ;

[0081] As normal stress increases, the shear dilatation angle decreases. ;

[0082] and The larger the value, the rougher the joint surface, the higher the strength, and the larger the shear dilatation angle.

[0083] 1.3) Applicable conditions and engineering adaptability of the shear dilatation angle model

[0084] This model is applicable under the following conditions:

[0085] Weak surrounding rocks: shale, carbonaceous slate, mudstone, etc., with obvious joint structures;

[0086] Shear-controlled deformation segment: such as arch failure, waist shear sliding area;

[0087] Preferential introduction should be given to water-bearing conditions. Correction factor as moisture content decreases.

[0088] In practical applications, the dilatation angle model can be embedded in the finite element / difference method simulation platform to dynamically update the constitutive response, thereby capturing real processes such as strain localization, volume expansion, and shear slip.

[0089] 1.4) Advantages compared to traditional models

[0090]

[0091] In summary, the nonlinear coupling modeling method for dilatation angle proposed in this invention establishes a parameter link between joint structure, shear stress, and volume expansion, and constructs a constitutive response basis that truly reflects the shear evolution path of weak surrounding rock. This is a core prerequisite for subsequent volumetric strain rate discrimination and risk level determination.

[0092] 2) Volumetric strain rate criterion based on dilatation angle

[0093] In tunnel engineering with weak surrounding rock, traditional monitoring systems often use quantities such as "convergence displacement," "crown settlement," and "support strain" as the basis for risk warning. However, these indicators are mostly lagging responses and cannot reflect the critical instability trend of the surrounding rock in a timely manner. Based on shear dilatation angle modeling, this invention proposes a risk precursor identification method with volumetric strain rate as the main controlling indicator. This method can be used to identify potential fracture and expansion phenomena during shear deformation in advance, and is the basis for dynamic risk level identification.

[0094] 2.1) Volumetric Strain Rate (denoted as...) The ratio (1 / day) describes the volume change per unit volume of a rock mass per unit time, commonly expressed in units of 1 / day or 1 / h. During tunnel excavation, the surrounding rock may undergo three stages: elastic compression, shear sliding, and dilatation.

[0095] Elastic compression stage The volume decreases;

[0096] Critical state stage: Volume stable;

[0097] Dilatation and expansion phase: Volume expansion indicates the expansion of the plastic zone.

[0098] therefore, The positive and negative values ​​and their growth trends can serve as direct indicators for judging whether a rock mass has entered an unstable shear state.

[0099] 2.2) Establish the volumetric strain rate criterion formula

[0100] Based on the definition of dilatation angle and the theory of shear strain, this invention proposes the following formula for expressing volumetric strain rate:

[0101]

[0102] in:

[0103] Volumetric strain rate;

[0104] Shear expansion angle (unit: degrees or radians);

[0105] : Shear strain rate (which can be obtained by inversion of the shear deformation rate field);

[0106] : Indicates the proportion of volume expansion caused by a unit shear displacement.

[0107] This formula demonstrates that the larger the shear dilatation angle or the higher the shear strain rate, the greater the volumetric strain rate, and the higher the risk.

[0108] 2.3) Engineering Calculation Methods for Shear Strain Rate

[0109] Shear strain rate It can be obtained in the following two ways:

[0110] (1) Field monitoring and derivation: based on the shear displacement difference between adjacent monitoring points , and the distance between measuring points and time interval calculate:

[0111]

[0112] (2) Numerical simulation inversion: Using platforms such as FLAC3D or ABAQUS, extract the historical shear strain data of the monitoring unit to obtain the discrete shear strain sequence, and then derive the rate.

[0113] 2.4) Determination of volumetric strain rate threshold

[0114] Based on experience from comparing on-site tests and historical data, when:

[0115] The surrounding rock is basically stable (risk level I).

[0116] There is a trend of shear expansion (risk level II);

[0117] Significant shear dilatation necessitates enhanced support measures (Risk Level III).

[0118] Critical shear dilatation state, warning of instability (risk level IV).

[0119] 2.5) Examples of Engineering Applications

[0120] In Case Study 1 of Section 11, the shear strain rate of the arch waist in a carbonaceous slate tunnel Shear expansion angle ,but:

[0121]

[0122] The system determined the risk level to be Level II, outputting the suggestion of "controllable deformation but enhanced monitoring is required", and prepared to initiate anchor bolt lengthening measures.

[0123]

[0124] 2.6) Advantages compared with traditional discrimination methods

[0125] In summary, by introducing a volumetric strain rate discriminant index, this invention provides a deformation risk assessment method that is both based on physical mechanisms and can be used for real-time driving of field monitoring data, thereby improving the foresight and responsiveness of risk prediction for tunnels in weak surrounding rock. This formula is not only applicable to traditional monitoring data interpretation but also suitable for integration with LSTM / BP prediction models as a key risk factor in the training set.

[0126] 3) Specific details of inverting the initial geostress tensor field from multi-source data

[0127] 3.1) In tunnel construction, the stability of the surrounding rock largely depends on its initial geostress environment. The magnitude and direction of the geostress tensor not only affect the generation of shear stress but also determine the induction conditions for fracture propagation, shear dilatation, and structural instability. Traditional methods often simplify geostress to a hydrostatic pressure field or a homogeneous initial stress field, considering only the principal compressive stress and neglecting the differences in stress distribution in different directions and the guiding effect of geological discontinuities. This coarse modeling cannot truly reflect the risk evolution characteristics under special scenarios such as fault zones, joint sets, and shallow-buried bias pressure.

[0128] This invention introduces a multi-source monitoring data inversion method to establish an inversion mechanism that truly reflects the three-dimensional spatial geostress tensor field. It integrates monitoring, surveying, and construction feedback data to achieve adaptive identification of the initial stress tensor.

[0129] 3.2) Definition and composition of stress tensor field

[0130] The three-dimensional geostress tensor field is described by the following six components:

[0131]

[0132] in:

[0133] : These are the normal stresses in the X, Y, and Z directions, respectively;

[0134] : This represents the shear stress component, taking into account factors such as eccentric pressure and eccentric load;

[0135] Each force is a spatial distribution function, denoted as . .

[0136] 3.3) Data Sources and Integration Methods

[0137] The inversion algorithm described in this invention integrates the following four types of data sources to construct the initial boundary conditions and objective function of the tensor field:

[0138] (1) Drilling stress gauge test data: provides the direction and magnitude of local principal stresses at certain depths;

[0139] (2) Acoustic wave test / seismic wave velocity data: inverting the geostress gradient through wave velocity anomalies;

[0140] Construction deformation feedback data: The distribution of arch crown settlement and arch waist convergence velocity reflects uneven stress direction;

[0141] (3) Cross-section laser scanning and joint parameter library: obtain the direction of joint distribution, which is used to guide the weight distribution of non-uniform stress field in each direction.

[0142] After the above data is unified by coordinates, it enters the data fusion algorithm to construct an objective function for stress inversion.

[0143] 3.4) Inversion Method Steps

[0144] (1) Initial tensor field assumptions: set the direction of principal stress and the magnitude of boundary principal stress (e.g., based on regional geological data or experience from previous excavation sections).

[0145] (2) Constructing a stress-deformation mapping model: Input the assumed stress field into the finite difference / finite element model to simulate the deformation field caused by construction;

[0146] (3) Definition of objective function:

[0147]

[0148] in:

[0149] Monitoring deformation Model predicts deformation;

[0150] Measured stress value Model stress results;

[0151] Weighting factors are set based on data quality.

[0152] (4) Inversion Iteration: Genetic algorithm, particle swarm algorithm or gradient descent method are used to continuously adjust The free variables in the objective function Minimum.

[0153] (5) Convergence judgment: When the error threshold is less than 5% or the maximum number of iterations reaches the upper limit, the stress field inversion is considered successful.

[0154] 3.5) Local stress anomaly identification mechanism

[0155] To account for the influence of joints / fault zones, this invention also introduces an anisotropic correction function. Used to adjust the stress transmission capability in different directions:

[0156]

[0157] For example: the direction in which joints run parallel normal direction This reflects the weakening effect of weak joint surfaces.

[0158] 3.6) Engineering Verification

[0159] In Case Study 2 (Section 12), a shallow-buried fault tunnel experienced excessive settlement in the early stages of construction. Initial stress inversion revealed that the Z-axis stress in the assumed tensor was underestimated by 18%, and the crown compressive strength was insufficiently preset, leading to premature failure of the shotcrete layer. After inversion correction, the arch support parameters were optimized, ultimately reducing the settlement rate by more than 60%.

[0160] In summary, the multi-source data-driven initial geostress tensor inversion method presented in this section achieves a shift from "static assumptions" to "dynamic identification," thereby enhancing the model's predictive capabilities. Its greatest advantages lie in: integrating multiple data sources to adapt to diverse field environments; supporting modeling of complex joint-fault-controlled areas; and embedding subsequent dilatation angle models and risk level models to achieve full-chain data-driven operation.

[0161] 4) Boundary fitting and reconstruction method for irregular tunnel cross-sections

[0162] 4.1) Overview of Technical Issues

[0163] Traditional tunnel deformation analysis is mostly based on simplified assumptions of standard circular or horseshoe-shaped cross-sections. Its main purpose is to simplify structural stress and ensure symmetrical support arrangements. However, in actual engineering, tunnels in weak surrounding rock sections often experience significant deviations from the ideal model in cross-section due to the following factors:

[0164] Uneven distribution of geological joints leads to biased excavation.

[0165] Local construction such as small pipes and advanced support causes initial boundary disturbances;

[0166] Hydraulic disturbances or secondary effects of earthquakes cause asymmetric expansion of the lining structure;

[0167] Irregular tunnel face excavation or abnormal surface subsidence pulls the boundary.

[0168] If a symmetrical circular cross-section is still used for risk assessment and support design in the above situations, it will directly lead to prediction errors, support mismatches, and even induce surrounding rock instability in weak areas. Therefore, it is necessary to introduce a geometric reconstruction method that can accurately describe irregular cross-sections as the basis for subsequent analysis and modeling.

[0169] 4.2) Cross-sectional profile data acquisition method

[0170] This invention employs a high-precision tunnel cross-section laser scanning device (such as a lidar system or image measurement system) and combines the following steps to acquire the cross-section contour:

[0171] (1) Setting up monitoring stations: At the working face or in the section where the initial support is completed, each m sets up a scanning station;

[0172] (2) Point cloud collection: Collect no less than 2000 boundary scattered points (XYZ coordinates) for each cross section;

[0173] (3) Coordinate transformation: Project the local coordinate system onto the longitudinal centerline coordinate system of the tunnel and normalize it;

[0174] (4) Noise removal: A point cloud filtering algorithm based on Euclidean distance threshold is used to remove interference such as falling objects and equipment shadows.

[0175] 4.3) Bezier curve fitting algorithm

[0176] To achieve smooth reconstruction of the cross-sectional profile, this invention preferably uses the Bezier curve fitting method as a preliminary processing method, as follows:

[0177] (1) Control point setting: Sample the collected point cloud at equal intervals according to arc length, and initially set... Control points ;

[0178] (2) Curve construction: The cross-sectional boundary is constructed using the nth-order Bezier formula:

[0179]

[0180] (3) Error evaluation and optimization: Calculate the average fitting error (RMSE) between the curve boundary and the point cloud edge. If it exceeds 5cm, add control points or optimize the weights until the fitting error meets the engineering requirements.

[0181] (4) Closure treatment: When the difference between the start and end points of the cross section is greater than 5cm, a closed curve treatment is performed to ensure the structural integrity.

[0182] 4.4) NURBS 3D Surface Reconstruction Method

[0183] If a spatially continuous cross-sectional model is required to serve three-dimensional tensor inversion and support layout simulation, this invention further employs non-uniform rational B-spline (NURBS) surfaces for reconstruction. The core steps include:

[0184] (1) Section integration: Select continuous A fitted two-dimensional cross-sectional Bezier curve is used as the input spline;

[0185] (2) Establish node vectors and control mesh, and construct a 3D NURBS surface:

[0186]

[0187] in:

[0188] : Control vertex;

[0189] B-spline basis functions;

[0190] Weighting coefficient.

[0191] (3) Output format: The cross-section is finally generated as a geometric model in STEP or IGES format, which can be directly imported into BIM platform, FLAC3D, etc. for subsequent coupling analysis.

[0192] 4.5) Applicability and Practical Effects of the Project

[0193] In Case 2(12), the tunnel cross-section exhibits an asymmetrical state with a sudden change in the right arch waist and a concave arch crown. Using Bezier curves and NURBS fitting, the RMSE was controlled within 2.7 cm, achieving over 4 times the accuracy compared to the traditional circular arc approximation method (error 11.5 cm). Subsequent calculations of the shear dilatation angle and anchor bolt placement analysis were all based on the actual contour, effectively avoiding anchorage failure and voids in the shotcrete layer.

[0194]

[0195] 4.6) Method Comparison and Innovations

[0196] In summary, the irregular tunnel cross-section fitting and reconstruction method proposed in this invention constructs a fine geometric model based on point cloud data and mathematical curve algorithms, and has the following characteristics: it can adapt to complex surrounding rock morphologies such as joint development and bias deformation; it has strong error control capabilities and is suitable for refined support analysis; it is highly compatible with BIM, monitoring systems, and analysis platforms; and it can serve as a unified geometric basis for other modules (such as stress inversion and shear dilatation calculation).

[0197] 5) Multi-parameter fusion risk level discrimination model

[0198] 5.1) Overview of Technical Issues

[0199] In tunnels with weak surrounding rock, single indicators (such as crown settlement rate and arch waist convergence) often fail to effectively characterize complex deformation risks. Especially under the combined effects of initial support, construction disturbances, and changes in ground stress, the surrounding rock response exhibits high nonlinearity and coupling. Traditional risk level assessment methods (such as empirical grading methods) lack a unified mathematical model, failing to reflect the intrinsic mechanism between dilatation, shear, and volumetric deformation, and also failing to achieve quantitative alignment with support parameters.

[0200] This invention proposes a multi-parameter fusion risk level discrimination model, which constructs a risk function that can be used for level classification and dynamic response based on four types of indicators driven by dilatation angle: volumetric strain rate, principal stress change, shear concentration, and disturbance factor, in order to replace the original method based on empirical classification or linear threshold judgment.

[0201] 5.2) Overall Structure of the Discriminant Model

[0202] The risk level discrimination model constructed in this invention takes the following form:

[0203]

[0204] in:

[0205] Risk index;

[0206] Volumetric strain rate reflects shear dilatation strength;

[0207] Shear strain concentration (the ratio of maximum to minimum shear strain in the shear zone);

[0208] Maximum principal stress increment;

[0209] Disturbance factors (comprehensive factors such as construction step distance and face disturbance);

[0210] : Weighting coefficients for each parameter.

[0211] 5.3) Parameter Acquisition and Processing Methods

[0212] (1) Volumetric strain rate

[0213] The calculations were obtained from Section 2 above, using real-time shear dilatation angle and shear strain rate data.

[0214] (2) Shear strain concentration

[0215] Extract shear strain values ​​at different locations from simulation models or measured strain data, and calculate the concentration factor:

[0216]

[0217] in To prevent extremely small constants with a denominator of 0, a value of 10⁻ is recommended. 6 .

[0218] (3) Principal stress variation

[0219] Defined as the ratio of the maximum principal stress change value to the initial principal stress within the monitoring area, it reflects the degree of stress disturbance.

[0220] (4) Disturbance factor

[0221] Defined as a standardized score for factors such as construction disturbance level (high-frequency blasting, exposure time of the free face, and initial support closure time), with a value range of... It is automatically generated by the supervision record and construction information system.

[0222] 5.4) Risk Level Classification Rules

[0223]

[0224] This invention is based on a risk index The interval range is divided into levels as follows:

[0225] Note: Parameter weights Parameters can be adjusted according to the project type and geological conditions; the default recommendation is... .

[0226] 5.5) Case Validation

[0227] In Case 1, the measured parameters during the construction of a soft rock tunnel are as follows:

[0228]

[0229]

[0230]

[0231]

[0232] Substituting into the calculation, we get:

[0233]

[0234] The system determined that the risk level of this section was Level IV, triggering emergency support linkage and timely controlling the subsequent instability of the arch.

[0235] 5.6) Advantages of the multi-parameter fusion risk level discrimination model: It can integrate multiple monitoring and model data to avoid the limitations of single-variable judgment; it supports dynamic updates of risk levels and can be embedded in the support recommendation module; the parameters are adjustable and it is applicable to various rock types and construction methods; it forms a coherent mechanism chain with the shear dilatation model and LSTM trend prediction.

[0236] In summary, the multi-parameter fusion-based risk level discrimination model proposed in this invention is a multi-factor discrimination mechanism that uses the dilatation effect as the main line and integrates shear concentration, stress disturbance, and construction disturbance. It effectively solves the problems of "response lag, coarse judgment, and difficulty in linkage" in existing methods, and possesses high real-time performance, scalability, and engineering applicability.

[0237] 6) Risk level and support parameter linkage mechanism

[0238] 6.1) Technical Background and Design Concept

[0239] Traditional tunnel support design is mostly based on a static grading design approach of "surrounding rock level → reference support scheme," typically represented by the "initial support-pre-support-lining" structural configuration table in the New Austrian Tunneling Method (NATM). While simple and easy to use, it suffers from the following serious shortcomings:

[0240] Risk identification and support response "broken chain": risk assessment is lagging and design is not adjusted in a timely manner;

[0241] The dynamic evolution of the same surrounding rock grade is ignored: the surrounding rock condition often changes significantly at different construction stages;

[0242] Lack of quantitative linkage mechanism: No mathematical mapping has been formed between "risk indicators → support stiffness / anchor parameters".

[0243] To this end, the present invention constructs a support parameter linkage mechanism based on real-time risk level discrimination results, which realizes automatic matching and suggestion generation of key parameters such as support stiffness, anchor bolt arrangement and spray layer thickness.

[0244] 6.2) Linkage Mechanism Structure

[0245] This invention employs a one-to-one correspondence mechanism between "risk level" and "support response parameters," using the risk level output by the risk discrimination model. Adjust the following three core support parameters in a coordinated manner:

[0246] (1) Initial support stiffness (Spray coating thickness, concrete grade, and reinforcement configuration);

[0247] (2) Anchor system parameters (anchor length, anchor spacing, anchor type);

[0248] (3) Step distance control parameters (initial support closure step distance, grouting section length, etc.);

[0249] Define the support parameter response function:

[0250]

[0251] in:

[0252] Risk level;

[0253] Anchor bolt length;

[0254] Anchor bolt spacing;

[0255] Thickness of the shotcrete layer;

[0256] : Empirical regression function or tabular mapping function.

[0257]

[0258] 6.3) Typical Support Parameter Response Table

[0259] Note: This table shows the recommended configuration. Specific parameters may be adjusted locally based on lithology, burial depth, and design specifications.

[0260] 6.4) Example of support parameter calculation

[0261] Taking a risk index of 0.61 (classified as Level IV) in Section 5 as an example, the system automatically generates the following adjustment suggestions:

[0262] The thickness of the shotcrete was increased from the original design of 60mm to 120mm;

[0263] Original The anchor bolt L=2.0m was adjusted to L=4.0m, and a resin end anchor was installed;

[0264] The anchor bolt spacing was reduced from 1.5m×1.5m to 0.8m×0.8m;

[0265] A double-layer H-shaped steel arch frame was added, reducing the closed step distance from 1.2m to 0.6m.

[0266] It is recommended that the project be included in the construction organization plan after being reviewed by the technical team, and that its implementation and feedback effects be tracked by the support automatic recording system.

[0267] 6.5) Model Structure and Data Interface

[0268] In this invention, the linkage mechanism module operates through the following system structure:

[0269] (1) Input layer: Receives real-time risk assessment results from the risk assessment model;

[0270] (2) Decision engine: calls parameter mapping table or fitting function based on risk level;

[0271] (3) Output interface: Generate support suggestions in JSON or table structure for use by BIM platform or construction system;

[0272] (4) Feedback loop: Compare the support implementation status with the subsequent risk level to evaluate the support response effect.

[0273] 6.6) Engineering Application Effects

[0274] In Case 1, a section of high-dilatation surrounding rock triggered a Level III risk. The system recommended increasing the shotcrete thickness from 80mm to 100mm and lengthening the anchor bolts to 3.5m. After actual application, the deformation rate decreased by 45%; the anchor bolts were not pulled out; the integrity of the shotcrete layer improved, and no more cracks appeared in the arch.

[0275] This indicates that the support parameter response linkage mechanism has good practicality and timeliness.

[0276] In summary, the "risk level-support parameter linkage mechanism" of this invention realizes a dynamic linkage closed loop between the risk discrimination model and the support design system, overcoming the limitations of traditional design's "post-event response and unchanging presets," and has the following advantages: it can be embedded in an intelligent construction platform to support real-time support optimization; the parameter table structure is clear, facilitating rapid execution on-site; it supports gradual learning and feedback correction, improving adaptability to long-cycle construction; and it is deeply coupled with the dilatation-shear model and risk index discrimination module to form a complete technology chain.

[0277] 7) Risk trend prediction method based on LSTM / BP model

[0278] 7.1) Technical Background and Application Requirements

[0279] In tunnels with weak surrounding rock, the deformation behavior of the surrounding rock exhibits strong time-varying characteristics, nonlinear evolution, and coupled driving forces. The deformation rate and shear dilatation behavior often show abrupt changes in stages. Traditional methods often use "current displacement exceeding a threshold" to determine the risk level, but this is a result-based judgment with significant lag, making it difficult to achieve "preventive" support optimization.

[0280] Therefore, it is necessary to establish a trend prediction method based on time series monitoring data, to mine risk evolution patterns from known monitoring indicators, and to predict the evolution path of risk levels in advance, so as to pre-set a window period for support response.

[0281] This invention innovatively introduces two types of models: Long Short-Term Memory (LSTM) network and Backpropagation Neural Network (BP) to achieve nonlinear prediction of the risk trend of tunnels in weak surrounding rock based on features such as dilatation angle, volumetric strain rate, and stress tensor change.

[0282] 7.2) Input and output structure of the prediction model

[0283] (1) Input feature variable (X)

[0284]

[0285] Based on the core indicators of the discriminant model in Section 5), near-term... The following monitoring data sequence for the day:

[0286] (2) Output the predicted target (Y)

[0287] Predicting the future Risk level for each day within a day (7 days recommended) or risk index This is to assist in generating support adjustment recommendations.

[0288] 7.3) Model Selection and Structural Design

[0289] (1) LSTM model (temporal neural network)

[0290] Suitable for processing long-term, highly fluctuating nonlinear sequence data. The network structure is as follows:

[0291] Input layer: sequence length × feature dimension (e.g., 7 days × 5 indicators);

[0292] LSTM hidden layer: two stacked layers, including memory gate structure;

[0293] Fully connected output layer: Predicting daily risk index for the next 7 days .

[0294] Advantages: It has memory properties and can identify long-term lag effects.

[0295] (2) BP neural network (static type)

[0296] Suitable for scenarios with low data dimensionality and relatively flat trends. The structure is as follows:

[0297] Input layer: Extracts the feature vector of the current data (e.g., 5 items);

[0298] Hidden layers: 1-2 layers of neurons;

[0299] Output layer: Predicts a single target value, such as .

[0300] Advantages: Simple structure, fast convergence, suitable for rapid initial judgment.

[0301] 7.4) Model Training and Optimization

[0302] (1) Data preprocessing: Standardize all input metrics (z-score or min-max); imput missing data; remove outlier data;

[0303] (2) Training set construction: Historical monitoring data > 30 days, samples are constructed using a rolling window method;

[0304] (3) Model loss function:

[0305] For risk index prediction: MSE (mean squared error);

[0306] For risk level prediction: Cross Entropy or Focal Loss.

[0307] (4) Parameter optimization: Use Adam or RMSProp optimizers, with a typical learning rate of 0.001 and a training cycle of 50 to 100 rounds.

[0308] 7.5) Example of Risk Trend Prediction

[0309] In Case 2, the prediction for a certain tunnel section is as follows:

[0310] Input: Average shear dilatation angle over the past 7 days is 12.5°, and volumetric strain rate is approximately... The settlement of the vaulted arch is accelerated;

[0311] LSTM model output: Risk level changes from Level II to Level III to Level IV within the next 3 days;

[0312] The system issued an early warning signal two days in advance, and the construction unit immediately strengthened the arch waist reinforcement and grouting support, successfully suppressing the continued increase in deformation rate.

[0313] 7.6) Prediction Error Assessment and Correction Mechanism

[0314] This invention sets up a continuous prediction window and a sliding test method to evaluate model accuracy:

[0315] Accuracy indicators: MAPE (Mean Absolute Percentage Error) <10%, or risk level hit rate ≥85%;

[0316] If the deviation exceeds two levels consecutively, the model weight retraining mechanism will be automatically activated.

[0317] System feedback mechanism: If the prediction deviation exceeds the threshold, new data is automatically collected and the neural network weights are corrected.

[0318] In summary, the LSTM / BP prediction model constructed in this invention has the following advantages: it integrates the main control index of dilatation with actual monitoring values, achieving theoretical-practice fusion; it supports multi-step prediction and reserves decision response time; it can be embedded in the monitoring system to realize the feedforward linkage of "risk trend - support strategy"; the model can learn and correct itself, adapting to the changing evolution law of surrounding rock.

[0319] 8) Mechanism for Generating Support Adjustment Suggestions

[0320] 8.1) Technical Background and Objectives

[0321] In tunnel construction with weak surrounding rock, the timeliness and specificity of support measures directly affect deformation control and safety stability. Traditional support designs, once formulated, are often implemented according to static drawings, lacking a dynamic adjustment mechanism. Especially when the risk level changes rapidly (such as a sudden expansion of the shear dilatation zone), the original design cannot respond in time, which can easily lead to failures such as anchor bolt shearing and shotcrete cracking.

[0322] Therefore, after completing risk assessment and trend prediction, this invention further establishes a support suggestion generation mechanism. Based on the risk level or risk trend output, it generates a support parameter combination suggestion that matches the risk level or risk trend, which is used to guide front-line technicians to adjust support strategies in a timely manner and improve the initiative of on-site decision-making.

[0323] 8.2) Input basis and logical source

[0324] The logic for generating support recommendations is based on the following two input sources:

[0325] (1) Current risk level The results from the discriminant model in Section 5 can be used as follows: class;

[0326] (2) Future trend prediction : If the predicted value is higher than the current level, the "advance adjustment" is initiated from the prediction results of the LSTM or BP model in Section 7.

[0327] If it exists If necessary, switch to the corresponding level of support recommendation in advance and implement reinforcement measures that "pre-evolutionize".

[0328] 8.3) Structure of Support Recommendations

[0329] The support recommendations are output in the form of a structured parameter table, including the following key sub-items:

[0330]

[0331] All recommendations are based on the Risk Level Response Table (see Section 6) and the optimal combination is automatically selected by combining the actual measurement point response.

[0332] 8.4) Output Mechanism and Platform Integration

[0333] The output format of the support recommendations supports the following structured methods:

[0334] JSON structured data: Facilitates integration with BIM platforms and digital construction site systems;

[0335] PDF / Excel report format: for handover briefings, construction shift handovers, or archiving;

[0336] Automatic reminder interface: When the risk level reaches Level III or above, the system can send reminders to the project chief engineer, supervising engineer, and construction team leader, specifying the adjustment items and specific locations.

[0337] 8.5) Feedback and Correction Mechanism

[0338] To improve the effectiveness and adaptability of the recommendations, the system is designed with the following feedback loop:

[0339] (1) Implementation confirmation: Construction personnel mark whether the recommendations have been implemented on the BIM platform or mobile terminal;

[0340] (2) Effect evaluation: Compare the monitoring data before and after the support adjustment, such as the change in settlement rate and the change in anchor strain;

[0341] (3) Recommendation for correction: If the risk still increases after implementation, the system can adjust the recommended parameters, increase the support level or propose double reinforcement measures;

[0342] (4) Model optimization closed loop: The parameter weights are corrected by feedback results to provide learning samples for future suggestion generation.

[0343] 8.6) Engineering Application Effects

[0344] In Case 2, based on LSTM predictions, the system anticipates that the risk level in segment K3+560 will rise to Level IV within the next 3 days, and outputs the following recommendations 48 hours in advance:

[0345] The anchor bolt arrangement has been adjusted from 2.5m / 1.5m×1.5m to 3.5m / 1.0m×1.0m;

[0346] The thickness of the spray layer has been increased from 80mm to 120mm for double layers;

[0347] The step length was shortened from 1.2m to 0.6m.

[0348] After implementation, the settlement rate of the arch decreased from 11 mm / d to 3.2 mm / d, effectively preventing support instability.

[0349] In summary, the support adjustment suggestion generation mechanism constructed by this invention has the following advantages: it realizes a closed loop of "risk identification - suggestion output - execution confirmation"; it supports linkage with trend prediction and has feedforward and responsiveness; the suggestion parameter structure is clear, which is convenient for integration, communication and implementation; and it supports a feedback learning mechanism and has dynamic adaptation and self-optimization capabilities.

[0350] 9) The dynamic correction mechanism for model parameters is as follows:

[0351] The method for determining the deformation risk of tunnels in weak surrounding rock as described in this invention further includes a dynamic correction mechanism for model parameters to maintain the adaptability and timeliness of the risk determination and prediction model. By real-time acquisition and residual calculation of core feature variables in monitoring data, and based on an error threshold control strategy, the shear dilatation angle prediction function, the weight parameters of the risk determination model, and the structural weights of the deep learning model are automatically iteratively corrected, thereby achieving continuous optimization of model performance and improvement of environmental adaptability.

[0352] The dynamic adjustment mechanism for model parameters includes the following steps:

[0353] 9.1) Dynamic correction of the shear dilation angle function:

[0354] Based on the residuals between the newly acquired actual dilatation angle measurements and the model predictions, if the mean or standard deviation of the residual sequence exceeds a preset tolerance (e.g., ...), If the dilatation angle prediction model is not found, then the refitting process of the model will be initiated.

[0355] Preferably, methods such as least squares regression, polynomial regression, or hyperbolic fitting are used to target... The fitting coefficients in the function are updated and optimized, and the updated function is immediately used in subsequent predictions and risk analysis.

[0356] 9.2) Correction of parameter weights in the risk discrimination function:

[0357] For the risk discriminant function:

[0358]

[0359] The weighting coefficients in Conduct periodic assessments and redistributions.

[0360] If the predicted risk level differs from the actual risk level by more than one level in multiple consecutive monitoring periods, the system will automatically adjust the weights of each item based on the error backpropagation mechanism or the minimum error fitting method to ensure that high-impact factors (such as disturbance stress and shear dilatation rate) are reasonably represented in the risk function.

[0361] 9.3) Dynamic fine-tuning of LSTM or BP neural network models:

[0362] For deep learning models (such as LSTM networks or BP neural networks) used for risk trend prediction, when the mean squared error (MSE) of the prediction results or the accuracy of the grade prediction decreases beyond a set threshold (e.g., error > 10%), the system automatically triggers a fine-tuning process, including but not limited to:

[0363] Incremental learning: New monitoring data is used for incremental training of the network;

[0364] Local weight optimization: Freeze some network layers and only update the weights of the last fully connected layer;

[0365] Model structure correction: Adjust the number of hidden layer nodes or the learning rate based on the cross-validation results;

[0366] Transfer learning: When mismatch is severe, a pre-trained model can be loaded and the output layer can be retrained.

[0367] 9.4) Error Judgment and Correction Trigger Mechanism:

[0368] The dynamic correction mechanism relies on a set of continuous residual statistics and triggering criteria, including:

[0369] The mean residual of the shear dilatation angle exceeds the set angle;

[0370] The cumulative risk assessment error exceeds the set level;

[0371] The prediction model's accuracy declined after three consecutive rounds of misjudgments.

[0372] Or set the number of days (e.g., every...) (Daily) Regular mandatory updates.

[0373] Through the above mechanism, this invention enables the model parameters to be continuously updated with on-site feedback information during the construction period, ensuring that the deformation risk identification, trend judgment and support response capabilities always maintain engineering adaptability, time effectiveness and parameter accuracy, significantly improving the system's engineering reliability and long-term intelligent operation capability.

[0374] 10) Modular deployment and integration system

[0375] The method for determining the deformation risk of tunnels in weak surrounding rock as described in this invention further includes a set of system integration modules and a deployment scheme adapted to the site conditions of weak surrounding rock. The core functions such as shear dilatation angle modeling, in-situ stress inversion, risk determination, trend prediction, support suggestions, and model correction are encapsulated into standardized system modules, forming an intelligent deployment system that can operate for a long time at the tunnel construction site. It has data acquisition, analysis and calculation, parameter output and feedback closed-loop functions, and supports engineering adaptation to multiple geological conditions, multiple construction stages and multiple monitoring systems.

[0376] The system deployment scheme includes the following structural modules and functional components:

[0377] 10.1) Overall System Structure

[0378] (1) Data acquisition module

[0379] Connecting field sensors and monitoring terminals enables real-time acquisition of multi-dimensional data such as shear stress, normal stress, crown settlement, anchor strain, and joint parameters. Supported data sources include, but are not limited to:

[0380] IoT terminals such as fiber optic inclinometers, strain gauges, and pressure cells;

[0381] Geological exploration data (JRC, JCS, etc.);

[0382] Digital camera / point cloud scanning device (for acquiring cross-sectional morphology);

[0383] Numerical simulation data (such as tensor fields output by FLAC3D / ANSYS / ABAQUS).

[0384] BIM system call interface.

[0385] (2) The model operation module integrates the following sub-models: shear dilatation angle prediction sub-module; volumetric strain rate and plastic zone discrimination sub-module; multi-source initial geostress tensor inversion sub-module; irregular cross-section reconstruction sub-module; risk index and level discrimination sub-module; trend prediction sub-module (LSTM / BP network); support suggestion output sub-module; model dynamic correction sub-module.

[0386] (3) The decision output module is used to transform the analysis results into construction suggestions and early warning information, including: risk level prompts; automatic recommendation of support parameters; trend development charts and retrospective analysis; alarm logic (SMS / APP / visual panel prompts).

[0387] (4) The feedback correction module receives the actual construction execution data and subsequent monitoring data in real time, compares them with the original judgment results, forms the basis for model correction, and constructs a closed-loop working mechanism of "judgment-execution-feedback-correction".

[0388] 10.2) Module Packaging and Deployment Methods

[0389] (1) Software structure

[0390] The system software implements the core computing logic in Python / Java, adopts a modular architecture (such as Flask + TensorFlow + Numpy), and has good scalability and portability. It supports Windows / Linux operating systems.

[0391] (2) Hardware adaptation

[0392] It supports deployment on edge computing terminals (such as NVIDIA Jetson, industrial-grade control computers) at the construction site, or on centralized cloud servers; it can be flexibly configured according to the conditions of the construction site.

[0393] (3) Data interface specifications

[0394] The system adopts a standard data interface in JSON / XML format and supports interconnection with mainstream digital construction platforms (such as BIM5D, smart tunnel platforms) and geological cloud platforms.

[0395] (4) Operating mode and life cycle:

[0396] Supports continuous execution and segmented task execution;

[0397] The data acquisition cycle is configurable (typically 15 minutes to 6 hours).

[0398] The model's parameters can be updated as construction progresses;

[0399] The risk response mechanism can be embedded in the scheduling system to automatically trigger on-site responses.

[0400] 10.3) Deployment Instance

[0401] In Case Two, the system was deployed in a railway tunnel in a mountainous area. The section was equipped with one industrial control computer, 12 sets of data acquisition terminals, and 5 sets of communication relays, covering all high-risk areas and achieving the following objectives:

[0402] The overall data collection rate reached 98.3%;

[0403] The risk trend prediction rate reached 87.5% with an advance warning rate of 72 hours.

[0404] The adoption rate of support recommendations reached 92.1%, and the deformation rate of abnormal sections was controlled within the safe threshold.

[0405] The system has been running continuously and stably for more than 120 days without interruption or false alarms.

[0406] In summary, this invention, by constructing a modular deployment structure and integrating deformation mechanism modeling, risk identification, predictive analysis, and support response functions, forms an integrated hardware and software engineering-grade intelligent system with the following advantages: High integration: clear module structure and unified functional interface; Engineering deployability: adaptable to various construction site monitoring methods and geological conditions; Strong scalability: easy to upgrade algorithm models or connect to new equipment; High real-time performance: supports automatic linkage between risk identification and support response; Strong sustainable engineering operation capability: supports synchronous deployment at the edge and remote locations with high fault tolerance.

[0407] 11) Case 1: A tunnel in a high-stress, weak surrounding rock tunnel - shear dilatation judgment-guided support reinforcement

[0408] 11.1) Project Background

[0409] The tunnel is located in a mountainous area in southwestern China, with a total length of 3.6 km. It traverses carbonaceous slate interbedded with silty shale, characterized by a loose structure, well-developed joints, and a surrounding rock integrity index RMi<20, classifying the rock mass as Class V weak surrounding rock. The excavation depth is approximately 380 m, with initial ground stress concentration at the site, resulting in a settlement rate of 10 mm / d at the arch crown and significant bulging at the arch waist.

[0410] 11.2) Implementation steps and application modules

[0411] (1) Modeling of shear dilatation angle and early warning of volumetric strain

[0412] The joint JRC=812 and JCS=46MPa were obtained based on in-situ shear tests.

[0413] Establish a dilatation angle coupling model: The range of shear expansion angle is obtained. ;

[0414] After real-time monitoring of shear stress and normal stress, calculations are performed. ;

[0415] when Reaching 2.3×10⁻ 4 When the risk level is / day, the system determines it to be Level II and initiates support parameter optimization.

[0416] (2) Stress tensor inversion and boundary fitting

[0417] A three-dimensional initial stress tensor field was constructed using a borehole stress gauge and acoustic pressure measurement.

[0418] The tunnel cross-section was modeled using point cloud scanning, and the real boundary was reconstructed using NURBS, with a fitting error of <2cm.

[0419] (3) Risk level assessment and support linkage adjustment

[0420] The risk level was determined to be Level II. The system output the following support plan: increase the thickness of the sprayed layer from 60mm to 100mm, increase the density of the anchor bolts from φ22 L=2.5m to L=3.5m, and reduce the spacing from 1.5m×1.5m to 1.0m×1.0m.

[0421] 11.3) Actual effect

[0422] The deformation rate was controlled within 3 mm / d, and the bulging of the arched waist was significantly alleviated;

[0423] The support structure is free of cracks and anchor bolt pull-out, and the lining thickness meets the convergence limit requirements.

[0424] The system's accuracy in predicting risk level trends exceeds 90%, providing timely guidance for adjustments.

[0425] 12) Case 2: A shallow fault zone tunnel - Boundary reconstruction + LSTM prediction combined with support optimization

[0426] 12.1) Project Background

[0427] The tunnel passes through an active fault zone, with a burial depth of 20-35m. The strata are mainly composed of silty clay interbedded with gravel and broken carbonaceous shale interbedded with mud. The groundwater is abundant and easily softened. The tunnel has an irregular horseshoe-shaped cross-section. After excavation, the early convergence was severe, and the daily variation of the arch waist displacement reached 14mm.

[0428] 12.2) Implementation steps and application modules

[0429] (1) Irregular boundary modeling

[0430] Laser point cloud cross-section scanning is performed every 2m to obtain scattered points on the cross-section;

[0431] Bezier curves and NURBS fitting are used to fit the section boundary, which is then embedded into the BIM platform to form a recognizable geometric model.

[0432] The model is used to simulate the support layout and invert the maximum free deformation zone at the boundary.

[0433] (2) Dynamic identification of shear dilatation angle

[0434] Based on the on-site parameters: MPa MPa, JRC=10, JCS=5MPa;

[0435] The predicted dilatation angle ψ≈18°, resulting in a volumetric strain rate of 3.1×10⁻⁻⁻⁶. 4 / day, triggering Level III risk;

[0436] (3) Construction of LSTM prediction model

[0437] Input the displacement, stress, and volumetric deformation monitoring data for the past 30 days;

[0438] After model training, the risk level is predicted to rise to Level IV (instability warning) within 7 days.

[0439] Output support optimization suggestions: Start steel arch frame + double-layer spraying + advanced small pipe grouting.

[0440] 12.3) Actual effect

[0441] Successful early warning and advance deployment of Level IV support system to prevent deformation from exceeding limits;

[0442] The settlement rate of the arch was reduced from 12 mm / d to less than 3 mm / d.

[0443] The anchor bolts did not shear off, the sprayed layer had no perforations, and the support system operated stably until the initial support was completed.

[0444] The system is used for three adjacent bidding sections, and the average prediction error does not exceed one risk level.

[0445] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for determining the deformation risk of tunnels in weak surrounding rock based on dilatation-shear constitutive model, characterized in that, Includes the following steps: A nonlinear coupling model between the dilatation angle and shear stress, normal stress, and joint parameters is constructed to obtain the dilatation angle; a volumetric strain rate discrimination formula is established based on the dilatation angle; the initial geostress tensor field is inverted using multi-source monitoring data; the irregular tunnel boundary is reconstructed using the Bezier or non-uniform rational B-spline method; a risk level discrimination model integrating multiple parameters is constructed to output risk levels I to IV. Support solutions, such as support stiffness, anchor bolt parameters, and spray layer thickness, are matched according to the risk level. Based on time-series monitoring data, establish long short-term memory or backpropagation neural network models to predict risk trends; Based on the prediction results, support adjustment suggestions are generated; Collect monitoring data to dynamically adjust model parameters; Integrate each module into a deployment system with monitoring, calculation, and output functions; The nonlinear coupling model between the dilatation angle and shear stress, normal stress, and joint parameters is an empirical formula. ;in: Shear expansion angle, in degrees; Shear stress (MPa); Normal stress (MPa); : Joint surface roughness coefficient; Shear strength of joint surface (MPa); The fitting coefficients were determined by nonlinear regression using multiple sets of experimental data. This formula characterizes the relationship that the shear dilatation angle ψ increases with increasing shear stress and decreases with increasing normal stress; The joint parameters include the joint roughness coefficient and the compressive strength of the joint wall surface, to reflect the influence of the joint surface morphology on the shear dilatation angle ψ; The nonlinear coupling model uses an exponential or hyperbolic function to fit the relationship between the dilatation angle ψ and the shear stress, normal stress, and joint parameters, and the model coefficients are determined by regression from experimental data.

2. The method according to claim 1, characterized in that, The volumetric strain rate discrimination formula based on the dilatation angle is: Volumetric strain rate = Shear strain rate × tan(ψ), where ψ is the dilatation angle; Using the volumetric strain rate discrimination formula, when the calculated volumetric strain rate is positive, the risk of dilatational deformation of the surrounding rock is determined, and when it is negative, the trend of shear contraction deformation of the surrounding rock is determined.

3. The method according to claim 1, characterized in that, The multi-source monitoring data includes surrounding rock displacement monitoring data, rock mass stress monitoring data, and geological survey data. The initial geostress tensor field is obtained by inverting the above multi-source monitoring data. The inversion of the initial geostress tensor field is achieved by establishing a numerical model that includes the effects of tunnel excavation and using an iterative algorithm to adjust the initial stress value until the deformation results calculated by the model match the multi-source monitoring data.

4. The method according to claim 1, characterized in that, The reconstruction of the irregular tunnel boundary adopts the Bezier curve fitting method, which performs smooth fitting on the collected discrete points of the tunnel contour to obtain a continuous and smooth tunnel boundary curve. The reconstruction of the irregular tunnel boundary adopts a non-uniform rational B-spline curve model. By introducing control vertex and node vectors, the tunnel boundary is accurately described to adapt to the complex curve tunnel boundary shape.

5. The method according to claim 1, characterized in that, The risk trend prediction model based on time-series monitoring data is a long short-term memory neural network model. This model receives monitoring parameters input in time series and outputs predicted values ​​of risk level or risk index for future time periods.

6. The method according to claim 1, characterized in that, The risk trend prediction model is a model based on an error backpropagation neural network. It establishes a nonlinear mapping relationship between input monitoring parameters and subsequent risk levels by using historical monitoring data for training, so as to predict future changes in risk levels. The backpropagation neural network model includes an input layer, at least one hidden layer, and an output layer. The input layer obtains the values ​​of multiple current monitoring parameters, and the output layer generates the corresponding predicted risk level or risk indicator.

7. The method according to claim 1, characterized in that, The support adjustment suggestions generated based on the prediction results include specific measures to adjust support parameters for different predicted risk levels: when the predicted risk level increases, suggestions are output to increase the stiffness of the support structure, add or lengthen anchor bolts, and thicken the shotcrete layer; when the predicted risk level decreases, suggestions are output to delay or reduce support reinforcement measures.

Images