Deeply-buried water-rich layered surrounding rock tunnel TBM (Tunnel Boring Machine) tunneling jamming risk prediction, prevention and control method

Abstract

The invention provides a deeply-buried water-rich layered surrounding rock tunnel TBM tunneling jamming risk prediction, prevention and control method, and belongs to the technical field of tunnel engineering intelligent construction. The invention provides a multi-source perception-intelligent prediction-dynamic regulation and control three-in-one prevention and control system aiming at the jamming risk caused by dynamic coupling of surrounding rock deformation and cutterhead load in TBM tunneling under deep-buried, water-rich and layered surrounding rock geological conditions. The method comprises the following steps: constructing a layered surrounding rock three-dimensional geomechanical model based on micro-seismic monitoring and a geological radar; 12-dimensional parameters such as cutterhead torque, surrounding rock convergence deformation and water inflow are collected in real time; a LightGBM integrated learning algorithm is adopted to realize card machine risk grading prediction; and according to the risk level, dynamically triggering prevention and control measures such as grouting reinforcement, drainage pressure reduction and propulsion parameter optimization, and correcting the model through a closed-loop feedback mechanism. According to the method, the early warning accuracy rate is larger than or equal to 90%, the machine jamming accident rate can be reduced by 60% or above, and the TBM monthly footage is improved by 25%-40%.

Description

Technical Field

[0001] This invention relates to the field of tunnel engineering safety monitoring technology, specifically to a method for predicting and preventing risks of TBM (Tunnel Boring Machine) jamming in deep-buried, water-rich, layered surrounding rock tunnels. Background Technology

[0002] In the construction of deep-buried, water-rich, layered tunnel boring machines (TBMs) in the surrounding rock, the coupling effect of high ground stress (>20MPa) and water-rich environment (water inflow ≥30m³ / h) triggers large deformations of the surrounding rock (>50mm) and a surge in asymmetric loads on the cutterhead (peak value 28MPa), leading to frequent machine jamming accidents. Existing technologies have significant shortcomings in addressing these issues. Traditional monitoring methods fragment geological forecasts, machine parameters, and hydrological data, failing to quantify the dynamic coupling effect of seepage pressure, deformation, and load. Early warning relies on fixed thresholds or human experience, with an average response lag of >45 minutes. Remedial measures are passive and delayed, only resorting to grouting after a machine jam, resulting in losses exceeding 5 million yuan per accident. Existing research has not solved the problems of multi-source information fusion modeling and proactive prevention and control, necessitating the construction of an intelligent prediction and prevention system that integrates geology, machine technology, and hydrology. Summary of the Invention

[0003] In view of the above-mentioned defects in the existing technology, the purpose of this invention is to provide a method for predicting and preventing the risk of TBM jamming in deep-buried water-rich layered surrounding rock tunnels. This method can be used to predict the risk of TBM jamming, with an early warning accuracy of ≥90%, which can reduce the jamming accident rate by more than 60% and increase the monthly TBM advance by 25%~40%.

[0004] To achieve the above-mentioned technical features, the objective of this invention is as follows: A method for predicting and preventing risks of TBM (Tunnel Boring Machine) jamming in deeply buried, water-rich, layered surrounding rock tunnels, comprising the following steps: S1, Geological Information Fusion: The in-situ stress field is inverted through microseismic monitoring, and the surrounding rock stability index is constructed by combining ground-penetrating radar and borehole data. Model: ; in, To score the rock mass, This refers to rock mass quality indicators. Pore ​​water pressure, It represents the uniaxial compressive strength of the rock. S2, Real-time multi-parameter monitoring: Synchronously collects cutterhead torque, surrounding rock convergence deformation and water inflow; S3, Risk Prediction: Input multi-dimensional feature vectors into the LightGBM classification model and output the risk level. The model training data includes several sets of historical card machine case data. S4, Dynamic Prevention and Control: Implement tiered strategies based on risk levels; S5, Closed-loop feedback: Based on prevention and control effectiveness coefficient Dynamically update model parameters: ; in, K To control the energy dissipation coefficient, Δk For process deformation rate, k Initial deformation rate.

[0005] Preferably, the accuracy of the geostress field in S1 is ±5°.

[0006] Preferably, the inversion of the geostress field in S1 uses the microseismic event location method to establish the bedding dip angle. θ The mapping relationship with the direction of maximum principal stress is such that the measurement error of the bedding dip angle is ≤3°.

[0007] Preferably, the range of the cutter head torque collected in S2 is 0-35kN·m, and the sampling rate is 10Hz; The range of the collected surrounding rock convergence deformation is 0-60mm, and the sampling rate is 5Hz; The range of the collected water inflow is 0-5 MPa, and the sampling rate is 2 Hz.

[0008] Preferably, the multidimensional feature vector in S3 has 12 dimensions, including: torque fluctuation rate, RSI index, water inflow rate, propulsion gradient, deformation rate, bedding dip angle, pore water pressure rate, cutterhead rotation speed fluctuation rate, rock mass integrity coefficient, tunneling speed rate, seepage pressure gradient, and cutter wear.

[0009] Preferably, the LightGBM classification model structure in S3 is as follows: (1) Input layer: 12-dimensional feature vector; (2) Hidden layers: Tree 1: max_depth=7, Tree 2: num_leaves=64, Tree n: n_estimators=500; (3) Output layer: probability distribution of different risk levels.

[0010] Preferably, the risk levels are divided into: 0-safe / 1-low risk / 2-high risk.

[0011] Preferably, the implementation of the tiered strategy based on risk level in step S4 specifically involves: (1) Level 1 risk: Optimize propulsion speed and set propulsion force ≤ 12MPa; (2) Level 2 risk: Radial grouting reinforcement, with a water-cement ratio of 0.8~0.9:1; (3) Level 3 risk: shutdown + advanced pipe roof support, with a pipe roof diameter of at least 108mm and a ring spacing of less than 0.8m.

[0012] Preferably, the torque warning threshold in S4 is based on the stratification dip angle. θ Dynamic adjustment: ; in, T 0 is the baseline threshold, 22 kN·m is taken for sandstone strata, and 18 kN·m is taken for mudstone; This is the torque warning threshold.

[0013] Preferably, the closed-loop feedback in S5 includes a Case Reasoning Base (CBR), and the case similarity calculation formula is: ; in, For case similarity, For feature weights, For the current eigenvalue, These are the feature values ​​of historical cases. The largest eigenvalue, i This is the eigenvalue index.

[0014] The present invention has the following beneficial effects: 1. Firstly, this invention significantly improves the safety and efficiency of TBM construction by fusing geological, machine, and hydrological multi-source information and using the LightGBM intelligent prediction model: Firstly, it achieves early warning of machine jamming risk (≥30 minutes, accuracy ≥90%), reducing the accident rate from the traditional 1.2 times / km to 0.3 times / km, a reduction of over 60%.

[0015] 2. Secondly, this invention improves tunneling continuity through dynamic control strategies (such as tilt angle adaptive grouting), increasing monthly advance from 210 meters to 290 meters, and improving efficiency by 38%.

[0016] 3. Finally, the invention adopts a closed-loop feedback mechanism to reduce emergency response costs by approximately RMB 1.5 million per incident, and comprehensively saves engineering costs of RMB 3 million per kilometer.

[0017] 4. In particular, this invention controls the deformation of the surrounding rock within 30mm in deep-buried, water-rich layered rock (burial depth > 500m, water pressure > 2MPa), and reduces the asymmetric load on the cutterhead by 45%, fundamentally solving the problem of machine jamming caused by the coupling of high ground stress and seepage softening. Attached Figure Description

[0018] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0019] Figure 1 This is a system architecture diagram of the present invention; Figure 2 This is a flowchart of the LightGBM model of the present invention; Figure 3The bedding angle of this invention θ Curve showing the relationship between torque threshold and torque threshold.

[0020] Figure 1 In China: S1. Geological information fusion; S2. Real-time multi-parameter monitoring; S3. Risk prediction; S4. Dynamic prevention and control; S5. Closed-loop feedback; Figure 2 In the diagram: 1-Input layer; 2-Hidden layer; 3-Output layer. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0022] Example 1: Please see Figure 1-2 This embodiment provides a method for predicting and preventing the risk of TBM jamming in deeply buried, water-rich, layered rock tunnels. The method includes the following steps: S1, Geological Information Fusion: The in-situ stress field is inverted through microseismic monitoring (accuracy ±5°), and the surrounding rock stability index is constructed by combining ground-penetrating radar and borehole data. Model: ; in, To score the rock mass, This refers to rock mass quality indicators. Pore ​​water pressure, It represents the uniaxial compressive strength of the rock. S2, synchronously collects cutterhead torque (range 0-35kN.m, sampling rate 10Hz), surrounding rock convergence deformation (range 0-60mm, sampling rate 5Hz), and water inflow (range 0-5MPa, sampling rate 2Hz). S3, Risk Prediction: Input the 12-dimensional feature vector into the LightGBM classification model and output the risk level (0-safe / 1-low risk / 2-high risk). The model training data includes 217 historical card machine cases.

[0023] S4, Dynamic Prevention and Control: Implement tiered strategies based on risk levels; (1) Level 1 risk: Optimize propulsion speed (propulsion force ≤ 12MPa); (2) Level 2 risk: Radial grouting reinforcement (water-cement ratio 0.8:1); (3) Level 3 risk: shutdown + advanced pipe roof support (pipe roof diameter 108mm, ring spacing 0.8m).

[0024] S5, Closed-loop feedback: Based on prevention and control effectiveness coefficient Dynamically update model parameters: ; in, K To control the energy dissipation coefficient, Δk For process deformation rate, k Initial deformation rate.

[0025] Furthermore, the inversion of the geostress field in S1 employs the microseismic event location method to establish the bedding dip angle. θ The mapping relationship with the direction of maximum principal stress, and the measurement error of bedding dip angle ≤ 3°.

[0026] Furthermore, the multidimensional feature vector in S3 has 12 dimensions, including: torque fluctuation rate, RSI index, water inflow rate, propulsion gradient, deformation rate, bedding dip angle, pore water pressure rate, cutterhead rotation speed fluctuation rate, rock mass integrity coefficient, tunneling speed rate, seepage pressure gradient, and tool wear.

[0027] Furthermore, the LightGBM classification model structure in S3 is as follows: (1) Input layer: 12-dimensional feature vector; (2) Hidden layers: Tree 1: max_depth=7, Tree 2: num_leaves=64, Tree n: n_estimators=500; (3) Output layer: probability distribution of different risk levels.

[0028] Furthermore, the torque warning threshold in S4 is based on the stratification angle. θ Dynamic adjustment: ; in, T 0 is the baseline threshold, 22 kN·m is taken for sandstone strata, and 18 kN·m is taken for mudstone; This is the torque warning threshold.

[0029] Furthermore, the closed-loop feedback in S5 includes a Case Reasoning Base (CBR), and the case similarity calculation formula is as follows: ; in, For case similarity, For feature weights, For the current eigenvalue, These are the feature values ​​of historical cases. The largest eigenvalue, i This is the eigenvalue index.

[0030] Example 2: A water diversion tunnel project has the following geological conditions: burial depth of 620m, alternating layers of mudstone and sandstone (layer thickness of 0.3-1.2m), and water inflow of 80m³ / h.

[0031] A method for predicting and preventing TBM jamming risks in deep-buried, water-rich, layered rock tunnels is proposed. This method addresses the jamming risk caused by the dynamic coupling of surrounding rock deformation and cutterhead load during TBM tunneling under geological conditions of deep burial (depth ≥ 300m), water-rich (water inflow ≥ 30m³ / h), and layered surrounding rock (layer thickness 0.2-2m). A three-dimensional geomechanical model of the layered surrounding rock is proposed, incorporating multi-source sensing, intelligent prediction, and dynamic control. The method includes: constructing a three-dimensional geomechanical model of the layered surrounding rock based on microseismic monitoring and ground-penetrating radar; real-time acquisition of 12 parameters including torque fluctuation rate, RSI index, water inflow rate change, propulsion gradient, deformation rate, bedding dip angle, pore water pressure change rate, cutterhead rotation speed fluctuation rate, rock mass integrity coefficient, tunneling speed change rate, seepage pressure gradient, and cutter wear; using the LightGBM ensemble learning algorithm to predict jamming risk levels (safe / low risk / high risk); dynamically triggering control measures such as grouting reinforcement, drainage depressurization, and propulsion parameter optimization based on the risk level; and correcting the model through a closed-loop feedback mechanism. This method has an early warning accuracy of ≥90%, can reduce the machine jamming accident rate by more than 60%, and increase the monthly footage of TBM by 25%~40%.

[0032] like Figure 1 As shown, a method for predicting and preventing risks of TBM (Tunnel Boring Machine) jamming in deeply buried, water-rich, layered surrounding rock tunnels is characterized by the following steps: S1. Geological Information Fusion: The geostress field is inverted through microseismic monitoring (accuracy ±5°), and a surrounding rock stability index (RSI) model is constructed by combining ground-penetrating radar and borehole data. ; in RMR To score the rock mass, Q This refers to rock mass quality indicators. P w Pore ​​water pressure, σ c It represents the uniaxial compressive strength of the rock.

[0033] S2. Real-time multi-parameter monitoring: Synchronously collect cutterhead torque (range 0-35kN·m, sampling rate 10Hz), surrounding rock convergence deformation (range 0-60mm, sampling rate 5Hz), and water inflow (range 0-5MPa, sampling rate 2Hz).

[0034] S3. Risk Prediction: Input the 12-dimensional feature vector into the LightGBM classification model and output the risk level (0-safe / 1-low risk / 2-high risk). The model training data includes 217 historical card machine cases.

[0035] S4. Dynamic Prevention and Control: Implement tiered strategies based on risk levels. (1) Level 1 risk: Optimize propulsion speed (propulsion force ≤ 12MPa); (2) Level 2 risk: Radial grouting reinforcement (water-cement ratio 0.8:1); (3) Level 3 risk: shutdown + advanced pipe roof support (pipe roof diameter 108mm, ring spacing 0.8m).

[0036] S5. Closed-loop feedback: Based on prevention and control effectiveness coefficient K Dynamically update model parameters: ; in, K To control the energy dissipation coefficient, Δk For process deformation rate, k Initial deformation rate.

[0037] In step S1, the geostress field inversion adopts the microseismic event location method to establish the mapping relationship between the bedding dip angle θ and the direction of the maximum principal stress, with the bedding dip angle measurement error ≤3°.

[0038] The LightGBM model structure in step S3 is as follows: (1) 1 input layer: 12-dimensional features (including torque fluctuation rate, RSI index, water inflow rate, propulsion gradient, deformation rate, bedding dip angle, pore water pressure rate, cutterhead rotation speed fluctuation rate, rock mass integrity coefficient, tunneling speed rate, seepage pressure gradient, and tool wear). (2) Hidden layers: Tree 1: max_depth=7, Tree 2: num_leaves=64, Tree n: n_estimators=500; (3) 3 output layers: 3 types of risk probability distributions.

[0039] In step S4, the torque warning threshold is dynamically adjusted according to the stratification dip angle θ. ; in T 0 is the baseline threshold (22 kN·m for sandstone strata and 18 kN·m for mudstone). The closed-loop feedback in step S5 includes a Case Reasoning Base (CBR), and the case similarity calculation formula is as follows: ; in, For case similarity, For feature weights, For the current eigenvalue, These are the feature values ​​of historical cases. The largest eigenvalue, iThis is the eigenvalue index.

[0040] The system automatically executes the following: advance speed 80→45mm / min; radial grouting (diffusion radius 1.8m); drainage hole spacing increased to 1.5m; In summary, by adopting the above technical solution, the beneficial effects of this invention are as follows: This invention significantly improves the safety and efficiency of TBM construction through the fusion of geological, machine, and hydrological multi-source information and the LightGBM intelligent prediction model. Firstly, it achieves advanced early warning of machine jamming risks (≥30 minutes, accuracy ≥90%), reducing the accident rate from the traditional 1.2 times / km to 0.3 times / km, a reduction of over 60%. Secondly, dynamic prevention and control strategies (such as angle-adaptive grouting) improve tunneling continuity, increasing monthly advance from 210 meters to 290 meters, an efficiency improvement of 38%. Finally, the closed-loop feedback mechanism reduces emergency response costs by approximately 1.5 million yuan per incident, resulting in comprehensive savings of 3 million yuan per kilometer in project costs. Especially in deeply buried, water-rich, layered surrounding rock (depth > 500m, water pressure > 2MPa), the deformation of the surrounding rock is controlled within 30mm, and the asymmetric load on the cutterhead decreases by 45%, fundamentally solving the industry problem of machine jamming caused by the coupling of high ground stress and seepage softening.

[0041] The above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for predicting and preventing risks of TBM (Tunnel Boring Machine) jamming in deeply buried, water-rich, layered rock tunnels, characterized in that... Includes the following steps: S1, Geological Information Fusion: The in-situ stress field is inverted through microseismic monitoring, and the surrounding rock stability index is constructed by combining ground-penetrating radar and borehole data. Model: ； in, To score the rock mass, This refers to rock mass quality indicators. Pore ​​water pressure, It represents the uniaxial compressive strength of the rock. S2, Real-time multi-parameter monitoring: Synchronously collects cutterhead torque, surrounding rock convergence deformation and water inflow; S3, Risk Prediction: Input multi-dimensional feature vectors into the LightGBM classification model and output the risk level. The model training data includes several sets of historical card machine case data. S4, Dynamic Prevention and Control: Implement tiered strategies based on risk levels; S5, Closed-loop feedback: Based on prevention and control effectiveness coefficient Dynamically update model parameters: ； in, K To control the energy dissipation coefficient, Δk For process deformation rate, k Initial deformation rate.

2. The method for predicting and preventing risks of TBM tunneling jams in deep-buried, water-rich, layered rock tunnels according to claim 1, characterized in that, The accuracy of the geostress field in S1 is ±5°.

3. The method for predicting and preventing risks of TBM tunneling jams in deeply buried, water-rich, layered rock tunnels according to claim 1, characterized in that, The inversion of the geostress field in S1 uses the microseismic event location method to establish the bedding dip angle. θ The mapping relationship with the direction of maximum principal stress is such that the measurement error of the bedding dip angle is ≤3°.

4. The method for predicting and preventing risks of TBM tunneling jams in deeply buried, water-rich, layered rock tunnels according to claim 1, characterized in that, The range of the cutter head torque collected in S2 is 0-35kN·m, and the sampling rate is 10Hz; The range of the collected surrounding rock convergence deformation is 0-60mm, and the sampling rate is 5Hz; The range of the collected water inflow is 0-5 MPa, and the sampling rate is 2 Hz.

5. The method for predicting and preventing risks of TBM tunneling jams in deeply buried, water-rich, layered rock tunnels according to claim 1, characterized in that, The S3 multidimensional feature vector has 12 dimensions, including: torque fluctuation rate, RSI index, water inflow rate, propulsion gradient, deformation rate, bedding dip angle, pore water pressure rate, cutterhead rotation speed fluctuation rate, rock mass integrity coefficient, tunneling speed rate, seepage pressure gradient, and tool wear.

6. The method for predicting and preventing risks of TBM tunneling jams in deeply buried, water-rich, layered rock tunnels according to claim 5, is characterized in that... The structure of the LightGBM classification model in S3 is as follows: (1) Input layer: 12-dimensional feature vector; (2) Hidden layers: Tree 1: max_depth=7, Tree 2: num_leaves=64, Tree n: n_estimators=500; (3) Output layer: probability distribution of different risk levels.

7. The method for predicting and preventing risks of TBM tunneling jams in deep-buried, water-rich, layered rock tunnels according to claim 6, characterized in that, The risk levels are divided into: 0 - safe / 1 - low risk / 2 - high risk.

8. The method for predicting and preventing risks of TBM tunneling jams in deep-buried, water-rich, layered rock tunnels according to claim 7, characterized in that, The specific implementation of the risk-level tiered strategy in S4 is as follows: (1) Level 1 risk: Optimize propulsion speed and set propulsion force ≤ 12MPa; (2) Level 2 risk: Radial grouting reinforcement, with a water-cement ratio of 0.8~0.9:1; (3) Level 3 risk: shutdown + advanced pipe roof support, with a pipe roof diameter of at least 108mm and a ring spacing of less than 0.8m.

9. The method for predicting and preventing risks of TBM tunneling jams in deeply buried, water-rich, layered rock tunnels according to claim 1, characterized in that, The torque warning threshold in S4 is based on the stratification dip angle. θ Dynamic adjustment: ； in, T 0 is the baseline threshold, 22 kN·m is taken for sandstone strata, and 18 kN·m is taken for mudstone; This is the torque warning threshold.

10. The method for predicting and preventing risks of TBM tunneling jams in deeply buried, water-rich, layered rock tunnels according to claim 1, characterized in that, The closed-loop feedback in S5 includes a Case Reasoning Base (CBR), and the case similarity calculation formula is as follows: ； in, For case similarity, For feature weights, For the current eigenvalue, These are the feature values ​​of historical cases. The largest eigenvalue, i This is the eigenvalue index.

Images