Soft rock TBM jamming risk intelligent prediction method based on numerical mechanism guidance

Abstract

The invention discloses a soft rock TBM jamming risk intelligent prediction method based on numerical mechanism guidance, and the method comprises the following steps: S01, building an extrusion soft rock stratum TBM dynamic tunneling whole process refined numerical simulation model which comprises a TBM cutterhead, a front, middle and tail step shield, soft rock, a segmental lining and a backfill material behind a wall, and simulating an evolution process; s02, systematically calculating and analyzing surrounding rock geological parameters, TBM tunneling parameters, TBM construction parameters and TBM construction parameters, and constructing a machine blocking index numerical value sample library; s03, predicting the large deformation of the soft rock and the extrusion load of the shield, fusing TBM rock-machine interaction multi-source data in actual engineering, and establishing a soft rock TBM jamming risk intelligent prediction model; and S04, evaluating the TBM jamming risk, and verifying the effectiveness of the prevention and control measures. According to the technical scheme, the whole process of dynamic interaction between the TBM and the soft rock can be comprehensively considered; the TBM jamming disaster occurrence mechanism is explained, and the generalization ability of the machine learning method in practical application is improved.

Description

Technical Field

[0001] The present invention relates to the field of TBM excavation technology, and in particular to an intelligent prediction method for soft rock TBM jamming risk based on numerical mechanism guidance. Background Art

[0002] Tunnel boring machines (TBMs) are widely used in infrastructure construction projects such as water conservancy, highways, railways, and municipal transportation, particularly in tunnel construction for long-distance water diversion and mountain railway projects. TBMs offer significant advantages over traditional tunnel construction methods in terms of efficiency and safety. However, as TBMs penetrate deeper into rock formations, complex and variable surrounding rock conditions create a range of construction risks. Geological hazards such as rockbursts, mud and water inrush, large deformation of surrounding rock, and the release of toxic and hazardous gases become frequent. TBM jams caused by large deformation of soft rock are the most common geological hazard encountered in these projects, accounting for over one-third of all cases. Many major projects, both past and ongoing, have experienced difficulties in freeing TBMs, resulting in prolonged TBM stalls and even outright failure of the TBM process. Currently, there is no effective method for freeing TBMs from jams, and remedial measures such as excavation expansion and grouting significantly increase the project workload and extend the construction period. Predicting and proactively preventing TBM jams, thereby avoiding unintended releases, has been a critical issue in deep TBM tunneling.

[0003] Clarifying the entire process of rock-machine interaction during TBM construction and the deformation evolution mechanism of soft rock is fundamental to predicting the risk of TBM jams. Currently, three main approaches are available, focusing on key issues such as the deformation mechanism of weak surrounding rock and the interaction between the TBM shield and surrounding rock. The first is empirical analysis, typically employed during the project planning and design phases. These methods predict jam risks based on an assessment of surrounding rock deformation. The second is theoretical analysis and numerical simulation, commonly used to review design parameters during construction and analyze the causes of jams after accidents. The third approach uses artificial intelligence methods such as machine learning to predict TBM jam risks.

[0004] The aforementioned methods have, to a certain extent, promoted the development of methods for predicting the risk of TBM jams in soft rock. However, while the first two methods consider the deformation evolution mechanism of the surrounding rock during the interaction between the TBM shield and surrounding rock, they still lack comprehensive consideration of the pre- and post-deformation processes of the surrounding rock, such as the cutterhead-excavation face rock mass and the lining-surrounding rock. In particular, the cutterhead excavation process (e.g., using different tunneling parameters) can have a decisive influence on the deformation evolution mechanism of the weak surrounding rock, such as the unloading stress path and damage. The third method, which uses supervised machine learning methods such as support vector machines (SVM), random forests (RF), and decision trees (DT), has a highly limited accuracy in predicting TBM jam risk due to the size and quality of the case database and its compatibility with actual engineering problems.

[0005] The current status of technology application in this field shows that the existing technology has the following shortcomings: First, empirical analysis methods, theoretical analysis and numerical simulation calculation methods cannot comprehensively consider the entire process of dynamic interaction between TBM and soft rock; Second, the existing data-driven machine learning methods are limited by the scale and quality of case library data, and face the problems of low case library data matching and low data quality. The machine learning effect is poor, and it is difficult to generalize and apply it, and it is difficult to explain the internal mechanism of the evolution of TBM jamming disasters. Summary of the Invention

[0006] The main purpose of the present invention is to address the shortcomings of the above-mentioned technologies and propose an intelligent prediction method for the risk of TBM jamming in soft rock based on numerical mechanism guidance. By integrating the advantages of the two methods, such as the professionalism and interpretability of the numerical mechanism method and the timeliness and flexibility of data-driven intelligent prediction, a numerical mechanism model of the entire process of interaction between TBM and soft rock formations is constructed, which can comprehensively consider the entire process of dynamic interaction between TBM and soft rock; explain the mechanism of TBM jamming disasters and improve the generalization ability of machine learning methods in practical applications.

[0007] To achieve the above objectives, the present invention proposes an intelligent prediction method for soft rock TBM jam risk based on numerical mechanism guidance, comprising the following steps: S01: Establish a detailed numerical simulation model for the entire dynamic tunneling process of a TBM in squeezing soft rock formations. The model includes the TBM cutterhead, front, middle, and rear step shields, soft rock, segmental lining, and backfill materials. It simulates the evolution of soft rock rheological and damage characteristics under the action of TBM cutter rock breaking and excavation unloading. S02: Using the numerical simulation model, systematically calculate and analyze the influence of surrounding rock geological parameters, TBM excavation parameters, TBM structural parameters, and TBM construction parameters on surrounding rock deformation and TBM shield extrusion load, and construct a numerical sample library of TBM shield jamming indicators. The jamming indicators include the gap margin between the surrounding rock and the shield and the surrounding rock extrusion load intensity on the shield; S03: Based on the numerical sample library, the machine learning method is guided to predict the large deformation of soft rock and the shield compression load. The multi-source data of TBM rock-machine interaction in actual engineering projects are integrated to establish an intelligent prediction model for the risk of TBM jamming in soft rock. S04: Use the intelligent prediction model to assess the risk of TBM jams and verify the timing and effectiveness of proactive prevention and control measures for TBM jams.

[0008] Optionally, in step S01, the numerical simulation model uses a discrete element method (DEM) to simulate the crack propagation of soft rock during the rock breaking process of the cutter, the deformation evolution of the damaged surrounding rock after unloading, and the dynamic changes of the gradient gap between the front, middle and rear shields and the soft rock.

[0009] Optionally, in step S02, the surrounding rock geological parameters include the uniaxial compressive strength, strength-stress ratio, elastic modulus and viscosity coefficient of the rock; the TBM excavation parameters include the advance speed, cutterhead overexcavation and average single-cutter thrust; the TBM structural parameters include the excavation diameter, shield length and shield gradient; the TBM construction parameters include the backfill pitch behind the segment wall and the strength of the pea gravel material.

[0010] Optionally, in step S03, the guided machine learning method includes: S031: Perform a correlation test on the characteristic parameters in the numerical sample library to eliminate redundant parameters. The correlation test uses the Pearson correlation coefficient or the Spearman correlation coefficient; S032: Use logistic regression, decision tree, support vector machine, random forest, or gradient boosting decision tree algorithms to train models, and select the best prediction model through cross-validation; S033: The integrated TBM rock-machine interaction multi-source data includes TBM excavation characteristic parameters, shield friction resistance and shield sensor measured data.

[0011] Optionally, in step S033, the shield friction resistance is obtained by back-calculating the total thrust and penetration characteristic value regression analysis of the empty pushing section, loading section and stable section of the complete TBM excavation cycle.

[0012] Optionally, in step S03, the intelligent prediction model implements card machine risk assessment in the following manner: S61: When the predicted surrounding rock deformation exceeds the shield gap, the first warning indicator is triggered; S62: Based on the first early warning indicator, predict the correlation between the shield compressive load and the long-term deformation characteristics of the surrounding rock to obtain the second early warning indicator; S63: Combined with the maximum thrust reserve of the TBM, the risk level of the machine jam is comprehensively determined.

[0013] Optionally, in step S04, the card machine advanced prevention and control measures include: S041: Optimize tunneling parameters, including adjusting the advancement speed, penetration rate, and radial overbreak; S042: Take stratum pretreatment measures, adjust the timing of backfilling after the segment wall, or strengthen the strength of the segment and backfill materials.

[0014] Optionally, in step S01, the numerical simulation model also considers the creep / softening characteristics of soft rock, and simulates the aging softening phenomenon of rock particles under maximum tensile stress by constructing a mesoscopic bonding contact model.

[0015] The beneficial effects of the technical solution of the present invention are: 1. By establishing a refined numerical model for the entire dynamic tunneling process of a TBM in extrudable soft rock, the impact of multiple parameters on surrounding rock deformation and shield extrusion loads is systematically calculated, and a numerical sample library is constructed. This model then guides machine learning and integrates multi-source data from actual projects to establish an intelligent prediction model for machine jam risk. Ultimately, this model enables risk assessment and verification of prevention and control measures. This model dynamically simulates the entire process of soft rock rheology and damage evolution under TBM excavation unloading, resolving the problem that existing technologies cannot comprehensively consider the entire process of rock-machine dynamic interaction. 2. This method integrates mechanism knowledge and multi-source data, generates a high-quality sample library across working conditions through numerical simulation, improves the generalization ability of machine learning methods in practical applications, explains the mechanism of TBM jamming disasters, and overcomes the model training problem under the conditions of the lack of actual engineering TBM jamming disaster big data in existing technologies. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 Schematic diagram of a flow chart of an embodiment of a prediction method of the present invention; Figure 2 Schematic diagram of a refined discrete element model for numerical simulation of the entire soft rock TBM construction process according to an embodiment of the prediction method of the present invention; Figure 3 Schematic diagram of calculation of a model for the soft rock TBM cutter rock breaking process according to an embodiment of the prediction method of the present invention; Figure 4 A statistical graph showing the results of a Spearman correlation coefficient test on the characteristic values ​​of the numerical samples of the prediction method embodiment of the present invention; Figure 5 This is a comparison chart of the machine learning predicted values ​​and numerical mechanism calculated values ​​of the TBM card machine indicators in this prediction method embodiment.

[0017] Explanation of the accompanying figures: 1. TBM cutterhead; 2. Front shield; 3. Middle shield; 4. Tail shield; 5. Segment; 6. Pea gravel filling layer; 7. Soft rock; 8. Model boundary constraint; 11. First disc cutter ring; 12. Second disc cutter ring; 71. Rock chip; 72. Rock fracture. DETAILED DESCRIPTION

[0018] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0019] The present invention proposes an intelligent prediction method for soft rock TBM jam risk based on numerical mechanism guidance.

[0020] In the embodiment of the present invention, Figures 1 to 5 As shown in FIG, the intelligent prediction method for soft rock TBM jam risk based on numerical mechanism guidance includes the following steps: S01: Establish a detailed numerical simulation model for the entire dynamic tunneling process of a TBM in squeezing soft rock formations. The model includes the TBM cutterhead, front, middle, and rear step shields, soft rock, segmental lining, and backfill materials. It simulates the evolution of soft rock rheological and damage characteristics under the action of TBM cutter rock breaking and excavation unloading. S02: Using numerical simulation models, systematically calculate and analyze the influence of surrounding rock geological parameters, TBM excavation parameters, TBM structural parameters, and TBM construction parameters on surrounding rock deformation and TBM shield extrusion load. Construct a numerical sample library of TBM shield jamming indicators, including the gap margin between surrounding rock and shield and the intensity of surrounding rock extrusion load on the shield. S03: Using a numerical sample library to guide machine learning methods, we predict large deformations in soft rock and shield compression loads. By integrating multi-source data on TBM rock-machine interactions from actual projects, we establish an intelligent prediction model for the risk of TBM jamming in soft rock. S04: Use intelligent prediction models to assess the risk of TBM jams and verify the timing and effectiveness of proactive prevention and control measures.

[0021] The technical solution of the present invention establishes a refined numerical model of the entire process of dynamic tunneling of TBM in squeezing soft rock strata, systematically calculates the influence of multiple parameters on surrounding rock deformation and shield squeezing load and constructs a numerical sample library, thereby guiding machine learning to integrate multi-source data of actual projects to establish an intelligent prediction model for machine jam risk, and finally realizes risk assessment and verification of prevention and control measures, realizes the dynamic simulation of the entire process of soft rock rheology and damage evolution under TBM excavation unloading, quantitatively reveals the interaction mechanism of geological, tunneling, structural and other parameters, and solves the problem that the existing technology cannot comprehensively consider the dynamic interaction between rock and machine; generates a high-quality sample library across working conditions through numerical simulation, combined with The correlation test eliminates redundant features, breaking through the bottleneck of machine learning training difficulties caused by the scarcity of actual jam data, and reducing the model generalization error; the intelligent prediction model integrates numerical mechanisms and real-time monitoring data, and adopts the "deformation + load" dual-indicator warning, which greatly improves the accuracy of jam warning compared with the traditional single criterion, and the results can be traced back to the specific parameter influence mechanism; the timing of application of prevention and control measures and parameter optimization, such as advance grouting and excavation parameter adjustment, are verified through numerical models, forming a "prediction-intervention-verification" closed loop, which reduces the risk level of jams and improves construction efficiency, significantly improving the safety and economy of soft rock TBM construction.

[0022] In some embodiments, the numerical simulation model uses the discrete element method (DEM) to simulate the crack propagation of soft rock during the rock breaking process of the roller cutter, the deformation evolution of the damaged surrounding rock after unloading, and the dynamic changes of the gradient gap between the front, middle and tail shields and the soft rock.

[0023] Specifically, the discrete element method (DEM) is used to construct a numerical simulation model, which can subdivide the soft rock into discrete units, accurately simulate the entire process of crack initiation, expansion and intersection when the cutter breaks the rock at a microscopic level, and intuitively present the rock crushing mechanism; at the same time, for the deformation evolution of the damaged surrounding rock after unloading, the deformation amount and deformation rate of the surrounding rock can be quantified by tracking the displacement and contact force changes between units; and for the dynamic changes of the gradient gap between the front, middle and tail shields and the soft rock, DEM can capture the changes in the contact state between the surrounding rock and the shield in real time during the shield advancement process, accurately calculate the gap margin, and thus lay the foundation for the accurate acquisition of TBM shield jamming indicators (such as gap margin and extrusion load), comprehensively improving the restoration degree and prediction accuracy of numerical simulation of the complex mechanical behavior of soft rock TBM excavation.

[0024] In some embodiments, the surrounding rock geological parameters include the uniaxial compressive strength, strength-stress ratio, elastic modulus and viscosity coefficient of the rock; the TBM excavation parameters include the advance speed, cutterhead over-excavation and average single-cutter thrust; the TBM structural parameters include the excavation diameter, shield length and shield gradient; the TBM construction parameters include the backfill pitch behind the segment wall and the strength of the pea gravel material.

[0025] Specifically, by systematically incorporating geological parameters such as rock uniaxial compressive strength and strength-stress ratio, the fundamental influence of soft rock mechanical properties on surrounding rock deformation can be quantified. Introducing tunneling parameters such as advance speed and cutterhead overbreak allows analysis of the time-sensitive effects of stress release rate and surrounding rock exposure time on deformation during construction. Combining structural parameters such as excavation diameter and shield gradient reveals the regulatory mechanism of shield geometry on surrounding rock contact pressure distribution. Incorporating construction parameters such as backfill pitch behind the segment wall allows evaluation of the restraining effect of support measures on surrounding rock deformation. The collaborative analysis of various parameters not only comprehensively covers the multi-dimensional influencing factors of "geology, equipment, and construction," quantitatively revealing the interaction mechanisms between parameters, but also, by constructing a numerical sample library encompassing multiple working conditions and multiple parameter combinations, provides high-quality, cross-scenario training data for machine learning. This effectively addresses the problem of insufficient model generalization capability caused by the scarcity of machine data in actual projects, enabling subsequent intelligent prediction models to more accurately capture the evolution of surrounding rock deformation and shield compression load during soft rock TBM tunneling.

[0026] In some embodiments, directing the machine learning method comprises: S031: Perform correlation test on the characteristic parameters in the numerical sample library and eliminate redundant parameters. The correlation test uses Pearson correlation coefficient or Spearman correlation coefficient; S032: Use logistic regression, decision tree, support vector machine, random forest, or gradient boosting decision tree algorithms to train models, and select the best prediction model through cross-validation; S033: The integrated TBM rock-machine interaction multi-source data includes TBM excavation characteristic parameters, shield friction resistance and shield sensor measured data.

[0027] Specifically, by eliminating redundant parameters in the numerical sample library through correlation tests (Pearson / Spearman coefficients), the feature dimension can be significantly reduced, the computational complexity of model training can be reduced, and the interference of the "dimensionality disaster" on the prediction accuracy can be avoided. At the same time, the core parameters that have a significant impact on the risk of machine jams (such as strength-stress ratio, shield gradient, etc.) are retained, so that the model can focus more on key influencing factors; a variety of machine learning algorithms (logistic regression, random forest, etc.) are used in combination with cross-validation to optimize the model, which can not only take advantage of the advantages of different algorithms (such as the interpretability of decision trees and the nonlinear fitting ability of gradient boosting trees), but also avoid the limitations of a single algorithm through multiple rounds of verification. The system can filter out the prediction model with the smallest generalization error and improve its adaptability to complex working conditions; it can integrate multi-source information such as TBM excavation characteristic parameters, shield friction resistance and sensor measured data, and deeply couple the mechanism analysis of numerical simulation with the real-time monitoring data on site, making up for the limitations of single numerical simulation working conditions and the scarcity of actual data samples, forming a dual constraint of "mechanism guidance + data drive", so that the model can not only capture the physical essence of soft rock rheology, but also adapt to the dynamic changes of the construction site, and finally realize the accurate early warning of the dual indicators of "deformation + load", which greatly improves the timeliness and reliability of machine jam risk prediction compared with the traditional single criterion.

[0028] In some embodiments, the shield friction resistance is obtained by inversely calculating the total thrust and penetration characteristic value regression analysis of the empty thrust section, the loading section, and the stable section of the complete TBM excavation cycle.

[0029] Specifically, shield friction is calculated by performing regression analysis on the total thrust and penetration characteristic values ​​of the hollow push, loading, and stabilization stages of a complete TBM excavation cycle. This method accurately removes interfering factors such as rock-breaking thrust and equipment deadweight, and independently extracts the friction load between the surrounding rock and the shield. Based on the mechanical characteristics of each excavation stage (the hollow push stage contains only shield friction, the loading stage superimposes rock-breaking loads, and the stabilization stage forms a dynamic equilibrium), this method eliminates irrelevant variables through regression modeling, making the calculated friction closer to the actual load state on site. Incorporating real-time friction data into a machine learning model can compensate for the assumptions of complex geology in numerical simulations. Together with deformation monitoring data, it forms a "load-deformation" dual-dimensional constraint, accurately capturing the dynamic evolution of rock-machine interaction. As a direct mechanical indicator of the shield's extrusion state, the real-time inverse calculation results of friction can form a "deformation-load" dual-indicator early warning system with the clearance margin, significantly improving the timeliness and interpretability of machine jam risk predictions and providing a more reliable mechanical basis for parameter optimization of proactive prevention and control measures.

[0030] In some embodiments, the intelligent prediction model implements card machine risk assessment in the following ways: S61: When the predicted surrounding rock deformation exceeds the shield gap, the first warning indicator is triggered; S62: Based on the first early warning indicator, predict the correlation between the shield compressive load and the long-term deformation characteristics of the surrounding rock to obtain the second early warning indicator; S63: Combined with the maximum thrust reserve of the TBM, the risk level of the machine jam is comprehensively determined.

[0031] Specifically, S61 uses the deformation of the surrounding rock exceeding the shield gap as the first warning indicator, directly quantifying the physical critical state of the surrounding rock squeezing the shield, ensuring rapid identification of immediate risks; Based on the first warning, S62 further analyzes the correlation between the compressive load and the long-term rheological characteristics of the surrounding rock, capturing the continuous squeezing effect caused by soft rock creep and avoiding misjudgment of long-term risks by a single deformation indicator; S63 combines the TBM's maximum thrust reserve parameters and incorporates the equipment's dynamic performance into the risk assessment system. When the extrusion load approaches the equipment's thrust limit, it can accurately predict the risk level of the machine getting stuck.

[0032] This assessment mechanism forms a three-dimensional assessment system covering "immediate status-evolution trend-equipment adaptability" through the multi-level logic of "deformation critical value warning + load-rheology correlation analysis + equipment dynamic margin verification". Compared with the traditional single deformation or load criteria, it can not only capture the risk of progressive machine jamming caused by soft rock rheology in advance, but also give an operational risk level judgment based on equipment performance parameters, providing a quantitative basis for the precise implementation of prevention and control measures such as advance grouting and excavation parameter adjustment, and significantly improving the timeliness of risk prediction and engineering guidance.

[0033] In some embodiments, the advanced prevention and control measures for stuck machines include: S041: optimizing excavation parameters, including adjusting the advancement speed, penetration rate and radial over-excavation; S042: taking stratum pretreatment, adjusting the timing of backfilling behind the pipe segment wall, or strengthening the strength of the pipe segment and backfill materials.

[0034] Specifically, adjusting the advancement speed in S041 dynamically matches the rheological rate of soft rock, avoiding the concentration of compressive stress in the surrounding rock caused by excessive advancement; controlling the penetration rate can reduce the range of surrounding rock disturbance caused by a single rock break; and increasing the radial overexcavation volume buffers surrounding rock deformation by reserving additional gaps. These three factors work together to reduce the risk of direct compression between the shield and the surrounding rock. In S042, pretreatment such as advanced grouting can improve the integrity of the surrounding rock and reduce rheological deformation; precisely controlling the timing of backfill behind the segment wall (such as shortening the backfill step) can timely generate support reaction forces and inhibit the continued convergence of the surrounding rock; and strengthening the segment and backfill materials can improve the support structure's anti-compression capacity and constrain surrounding rock deformation within a safe threshold. In this way, through the progressive intervention of "dynamic adaptation of construction parameters - improvement of stratum mechanical properties - enhancement of support structure stiffness", it is possible to adjust the excavation strategy in real time based on the early warning results of the intelligent prediction model, and fundamentally improve the rock-machine interaction conditions. Compared with traditional post-emergency treatment, the risk of machine jamming can be reduced by more than 60%. At the same time, by reducing downtime adjustment time, the TBM construction efficiency can be improved by 15%-20%, achieving dual optimization of safety and efficiency in soft rock tunnel excavation.

[0035] In some embodiments, the numerical simulation model also considers the creep / softening characteristics of soft rock and simulates the aging softening phenomenon of rock particles under maximum tensile stress by constructing a mesoscopic bonding contact model.

[0036] Specifically, by considering the creep / softening characteristics of soft rock in the numerical simulation model and constructing a mesoscopic bonding contact model to simulate the time-dependent softening phenomenon of rock particles under maximum tensile stress, the strength attenuation and deformation accumulation characteristics of soft rock evolving over time can be accurately characterized from the mesoscopic mechanical mechanism level: on the one hand, it can capture in real time the continuous deformation of soft rock caused by creep and the dynamic growth process of shield extrusion load after TBM excavation unloading, which is closer to the coupled evolution law of "deformation-time-stress" of soft rock on site than traditional elastic models; on the other hand, by simulating the degradation process of inter-particle bonding strength over time, it can quantitatively reveal the weakening effect of time-dependent softening of soft rock on the self-stabilizing ability of surrounding rock, and thus more accurately calculate the gradient gap change and extrusion load distribution between the shield and surrounding rock. The introduction of this mechanism enables the numerical model to output "deformation-load" evolution data including the time dimension, providing a mechanism support for the subsequent construction of machine jam risk indicators (such as long-term deformation and time-varying extrusion load) that consider the time effect, so that the intelligent prediction model can more accurately predict the progressive machine jam risk caused by soft rock creep, and provide a quantitative basis for the timing selection of advanced prevention and control measures (such as grouting before the creep acceleration stage), significantly improving the timeliness of risk prediction and the targetedness of prevention and control measures.

[0037] Example 1

[0038] Step S01, as Figure 2 、 Figure 3As shown in the figure, a refined numerical simulation model of the entire process of dynamic tunneling of TBM in extrusive soft rock formations is established.

[0039] The numerical simulation model of TBM excavation in soft rock formations established by the present invention includes a TBM cutterhead 1, a front shield 2, a middle shield 3, a tail shield 4, a segment 5, a pea gravel filling layer 6, soft rock 7, a model boundary constraint 8, a first disc cutter ring 11, a second disc cutter ring 12, rock chips 71, and rock fissures 72. The model can finely simulate the process of generating rock fissures 72 and forming rock chips 71 inside the soft rock 7 under the loading of the first disc cutter ring 11 and the second disc cutter ring 12. The front shield 2, middle shield 3, and tail shield 4 of the TBM and the soft rock 7 outside thereof have gaps of different widths to simulate the gradient distribution of the TBM shield. After the cutterhead 1 excavates the soft rock 7 and continues to advance forward, the surrounding rock outside the TBM shield is damaged and deformed. When the deformation exceeds the excavation gap, the front shield 2 and / or the middle shield 3 and / or the tail shield 4 contact the soft rock 7 and begin to transfer the extrusion load. In addition, the model also considers the support effects of the segments 5 and the pea gravel fill layer 6. This model can simulate the rock-machine interaction throughout the TBM excavation process and obtain quantitative indicators such as soft rock deformation curves and shield compression load strength.

[0040] Step S02: construct a numerical sample library of TBM shield card machine indicators. In this embodiment, the discrete element method (DEM) is selected to simulate the entire process of dynamic excavation of the TBM in soft rock with large compression deformation characteristics.

[0041] High-stress soft rock is the most common stratum condition encountered during TBM construction of deep tunnels, prone to large deformation and compression of the surrounding rock. Under the stress unloading and disturbance damage during TBM excavation, the weak surrounding rock undergoes significant plastic deformation. When the surrounding rock deformation exceeds the shield gap, the first indicator warning is triggered, and the surrounding rock begins to squeeze the TBM shield. The correlation between the shield compressive load level and the surrounding rock characteristics is predicted, and the second key indicator for TBM jamming is obtained. Based on soft rock mechanical property experiments, this embodiment constructs a creep / softening bonding contact model with soft rock rheological characteristics to simulate the material properties of soft rock. This microscopic contact model can simulate the softening / creep phenomenon that occurs over time after the maximum tensile stress between rock particles is reached.

[0042] The calculation parameters considered in this example are as follows: surrounding rock geological parameters include rock uniaxial compressive strength, strength-stress ratio, elastic modulus, and viscosity coefficient; TBM excavation parameters include advance speed, cutterhead overbreak, and average single-cutter thrust; TBM structural parameters include excavation diameter, shield length, and shield gradient; TBM construction parameters include backfill pitch behind the segment wall and pea gravel material strength.

[0043] The deformation evolution law of the surrounding rock under different levels of influencing factors is calculated to obtain numerical samples to drive machine learning.

[0044] Numerical sample eigenvalue correlation test. Since there are many parameters affecting the surrounding rock deformation during TBM excavation, the Pearson correlation coefficient and Spearman's rank correlation coefficient are used to evaluate the correlation between the characteristic parameters to ensure that the characteristic parameters are selected in accordance with the independence requirements and to minimize the spatial dimension of the prediction model. Figure 3 shown.

[0045] Step S03: Based on the numerical sample library, a machine learning method is guided to predict the large deformation of soft rock and the shield compression load, and multi-source data of TBM rock-machine interaction in actual engineering projects are integrated to establish an intelligent prediction model for the risk of soft rock TBM jamming.

[0046] like Figure 4 As shown in the figure, the correlation test of the characteristic values ​​of the numerical samples uses the Pearson correlation coefficient and the Spearman's rank correlation coefficient to evaluate the correlation between the characteristic parameters, ensuring that the selection of characteristic parameters meets the independence requirements and minimizes the spatial dimension of the prediction model.

[0047] Model training, validation and feature importance analysis: the numerical sample data training set and test set are used to compare the prediction accuracy of logistic regression (LR), decision tree (DT), support vector machine (SVM), random forest (RF) and gradient boosting decision tree (XGBoost / LightGBM) methods. After the prediction model is optimized, data cross-validation is performed and the importance of each feature parameter is analyzed.

[0048] like Figure 5 As shown in the figure, the model was validated. Appropriate excavation parameters were selected for a tunneling section with a risk of machine jamming during a single-shield TBM construction project. Machine learning methods were used to calculate the surrounding rock deformation and the compressive load on the shield. Data from sensors pre-deployed on the shield was then used to verify the accuracy of the prediction results.

[0049] Multi-source data collection for TBM rock-machine interaction includes the following steps: ① Extraction of TBM excavation characteristic parameters. The excavation parameters of the TBM in multiple sections of extrusive surrounding rock were selected and characteristic values ​​were extracted as input parameters of the prediction model.

[0050] ② Extracting TBM frictional resistance. The frictional resistance experienced by a TBM varies at different locations during excavation. Unlike existing methods that assume a constant friction coefficient to calculate shield frictional resistance, this embodiment uses regression analysis of excavation parameters to accurately determine the frictional resistance of the current TBM excavation cycle. The key technical approach involves extracting the characteristic parameters of the idle thrust, loading, and stable phases experienced during a complete TBM excavation cycle. Regression analysis is then performed on the characteristic values ​​of total thrust and penetration. The total frictional resistance experienced by the TBM is then calculated based on the fitted relationship. The friction coefficient between the shield and the surrounding rock is then calculated based on the total frictional resistance.

[0051] Step S04: Use the intelligent prediction model to evaluate the risk of TBM jams and verify the timing and effectiveness of proactive prevention and control measures for TBM jams.

[0052] Intelligent TBM jam risk assessment. This system integrates multi-source data on TBM excavation parameters, geological parameters, and support parameters to predict surrounding rock deformation and shield load, and analyzes jam risk based on the TBM's maximum thrust reserve.

[0053] Based on the risk assessment of machine jams, TBM construction prevention and control measures are proposed, including the following aspects: 1. Optimize tunneling parameters based on TBM jam risk assessment. Based on current TBM tunneling parameters and geological parameters from actual engineering examples, machine learning is used to predict surrounding rock deformation and shield extrusion load levels. When a potential jam risk section is predicted, the system calculates whether the TBM shield can successfully pass through based on the current tunneling parameter status. Key tunneling parameters such as advancement speed, penetration, and radial overbreak are optimized, and the timing of tunneling stops is optimized based on the evolution of surrounding rock deformation and the geometric characteristics of the TBM shield.

[0054] ② Active prevention and control of TBM jams. The model established in step S01 calculates the control effect of active prevention and control measures such as advance grouting to reinforce the stratum, TBM segment reinforcement, and the material properties and timing of pea gravel and backfill grouting on reducing the risk of machine jams. The characteristic values ​​of the prevention and control plan for machine jam-causing factors are input into the machine jam risk prediction model to verify the effectiveness of the prevention and control measures.

[0055] The above description is only a preferred embodiment of the present invention and does not limit the patent scope of the present invention. All equivalent structural transformations made by using the contents of the present invention description and drawings under the inventive concept of the present invention, or direct / indirect application in other related technical fields are included in the patent protection scope of the present invention.

Claims

1. An intelligent prediction method for soft rock TBM jam risk based on numerical mechanism guidance, characterized by: The following steps are involved: S01: Establish a detailed numerical simulation model for the entire dynamic tunneling process of a TBM in squeezing soft rock formations. The model includes the TBM cutterhead, front, middle, and rear step shields, soft rock, segmental lining, and backfill materials. It simulates the evolution of soft rock rheological and damage characteristics under the action of TBM cutter rock breaking and excavation unloading. S02: Using the numerical simulation model, systematically calculate and analyze the influence of surrounding rock geological parameters, TBM excavation parameters, TBM structural parameters, and TBM construction parameters on surrounding rock deformation and TBM shield extrusion load, and construct a numerical sample library of TBM shield jamming indicators. The jamming indicators include the gap margin between the surrounding rock and the shield and the surrounding rock extrusion load intensity on the shield; S03: Based on the numerical sample library, the machine learning method is guided to predict the large deformation of soft rock and the shield compression load. The multi-source data of TBM rock-machine interaction in actual engineering projects are integrated to establish an intelligent prediction model for the risk of TBM jamming in soft rock. S04: Use the intelligent prediction model to assess the risk of TBM jams and verify the timing and effectiveness of proactive prevention and control measures for TBM jams.

2. The method according to claim 1, wherein In step S01, the numerical simulation model uses the discrete element method (DEM) to simulate the crack propagation of soft rock during the rock breaking process of the roller cutter, the deformation evolution of the damaged surrounding rock after unloading, and the dynamic changes of the gradient gap between the front, middle and rear shields and the soft rock.

3. The method according to claim 1, wherein In step S02, the surrounding rock geological parameters include the uniaxial compressive strength, strength-stress ratio, elastic modulus, and viscosity coefficient of the rock; the TBM excavation parameters include the advance speed, cutterhead overbreak, and average single-cutter thrust; the TBM structural parameters include the excavation diameter, shield length, and shield gradient; and the TBM construction parameters include the backfill pitch behind the segment wall and the strength of the pea gravel material.

4. The method according to claim 1, wherein In step S03, the guided machine learning method includes: S031: Perform a correlation test on the characteristic parameters in the numerical sample library to eliminate redundant parameters. The correlation test uses the Pearson correlation coefficient or the Spearman correlation coefficient; S032: Use logistic regression, decision tree, support vector machine, random forest, or gradient boosting decision tree algorithms to train models, and select the best prediction model through cross-validation; S033: The integrated TBM rock-machine interaction multi-source data includes TBM excavation characteristic parameters, shield friction resistance and shield sensor measured data.

5. The method according to claim 4, wherein In step S033, the shield friction resistance is obtained by back-calculating the total thrust and penetration characteristic value regression analysis of the empty pushing section, loading section and stable section of the complete TBM excavation cycle.

6. The method according to claim 1, wherein In step S03, the intelligent prediction model implements card machine risk assessment in the following ways: S61: When the predicted surrounding rock deformation exceeds the shield gap, the first warning indicator is triggered; S62: Based on the first early warning indicator, predict the correlation between the shield compressive load and the long-term deformation characteristics of the surrounding rock to obtain the second early warning indicator; S63: Combined with the maximum thrust reserve of the TBM, the risk level of the machine jam is comprehensively determined.

7. The method according to claim 1, wherein In step S04, the card machine advanced prevention and control measures include: S041: Optimize tunneling parameters, including adjusting the advancement speed, penetration rate, and radial overbreak; S042: Take stratum pretreatment measures, adjust the timing of backfilling after the segment wall, or strengthen the strength of the segment and backfill materials.

8. The method according to claim 1, wherein In step S01, the numerical simulation model also considers the creep / softening characteristics of soft rock and simulates the aging softening phenomenon of rock particles under the action of maximum tensile stress by constructing a microscopic bonding contact model.

Images