A method for quickly determining rock mass quality of TBM tunneling

Abstract

The application provides a kind of TBM excavation rock mass quality rapid determination method, it is related to surrounding rock quality measurement technical field.The application is based on the real-time working parameter of TBM, utilizes artificial neural network model, realizes the method for the quick determination of surrounding rock mass quality in front of working face.In the TBM tunnel that has been excavated, the excavation parameters of TBM are collected, including cutterhead rotating speed, cutterhead torque, total thrust, thrust cylinder pressure, excavation speed, cutterhead penetration, etc.Parameter such as roughness of filler in rock slag is collected, ground stress information is collected for ground stress inversion, the exposed rock mass between TBM cutterhead and segment erector is photographed by using annular camera to collect surrounding rock mass parameter information, and the above various data are artificially classified and labeled as training samples, based on training samples and Barton Q system classification, RBF (radial basis) neural network is used for deep learning training, and finally a kind of surrounding rock mass quality rapid determination method based on TBM excavation parameter can be obtained.

Description

Technical Field

[0001] This invention relates to the field of surrounding rock quality measurement technology, and in particular to a method for rapidly determining the quality of rock mass during TBM tunneling. Background Technology

[0002] "Exploring the depths of the Earth is a strategic scientific and technological problem that we must solve." To this end, my country is continuously developing deeper into the world in areas such as transportation, water conservancy and hydropower, and mining. Currently, my country has become the world's largest tunnel-producing country, accumulating rich engineering experience in tunnel design. Many of these deep and long tunnels are excavated using TBMs (Tunnel Boring Machines). The TBM method has become the mainstream method in deep and long tunnel construction, offering advantages such as fast excavation speed, low construction cost, minimal construction disturbance, and high safety. However, TBM construction is extremely sensitive to rock conditions. Inappropriate selection of excavation parameters during the excavation process can significantly affect the TBM's efficiency. When the TBM traverses sections with high-strength surrounding rock, it is necessary to promptly adjust excavation control parameters, such as total thrust, cutterhead torque, and cutterhead penetration. Otherwise, it will accelerate cutterhead wear, leading to increased construction costs, increased rock-breaking difficulty, and decreased excavation efficiency. Conversely, when traversing sections with low-strength and unstable surrounding rock, failure to adjust excavation parameters in a timely manner can easily result in TBM shield or cutterhead jamming. The essence of the above phenomenon is that the cutterhead rotates and cuts the rock mass, making it impossible to directly measure the quality of the rock mass in front of the tunnel face. The quality of the surrounding rock mass in front of the tunnel face that the TBM is about to excavate is unknown, which makes the TBM tunneling under the condition of unknown surrounding rock mass quality, and has a certain degree of blindness and subjectivity.

[0003] Therefore, timely acquisition of accurate indicators such as the quality of the surrounding rock mass for engineering rock mass classification is directly related to adjusting tunneling control parameters and construction plans. It is also crucial for preventing TBM jamming and excessive cutterhead wear, and for improving tunneling efficiency. Currently, there are various engineering rock mass classification systems worldwide, such as classification by uniaxial compressive strength, tunnel rock stability, and rock mass integrity. However, these classifications are all based on direct on-site measurements and are not comprehensive enough. Furthermore, the internationally widely used Paton-Q system classification still has the following problems:

[0004] 1. The different attitudes of the rock strata in the excavated cavern were not taken into account, namely the dip angle and strike of the rock strata.

[0005] 2. The RQD value in the Q system does not take into account the relationship between the geological structure surface orientation and the tunnel axis direction. The Q system does not consider rock strength and unfavorable combinations of structural surfaces when considering factors affecting the stability of the surrounding rock. The RQD value is significantly affected by the direction of the structural surface. The relationship between the tunnel excavation direction and the direction of the dominant joint set should be considered, especially the unfavorable combination relationship.

[0006] 3. The Q system design method does not consider the principal direction of geostress. With different principal directions of geostress, the stress concentration and failure location of the tunnel surrounding rock will inevitably be different. Under normal circumstances, after the tunnel is excavated, the tangential stress on the cross section increases, the radial stress decreases, while the axial stress remains almost unchanged. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention provides a method for rapidly determining the quality of rock mass during TBM tunneling. It employs on-site collection of tunneling data, rock mass data, and in-situ stress inversion, combined with deep learning from artificial neural networks, to achieve rapid and accurate determination of the surrounding rock mass parameters for TBM tunneling.

[0008] A method for rapidly determining the quality of rock mass during TBM tunneling includes the following steps:

[0009] Step 1: Collect TBM tunneling parameters during TBM tunneling, specifically including cutterhead rotation speed, cutterhead torque, total propulsion force, propulsion cylinder pressure, tunneling speed, and cutterhead penetration.

[0010] Step 2: Collect the rock debris generated during the excavation to obtain information about the rock mass, including the rock mass filling material, roughness, and whether the rock mass contains clay.

[0011] Step 3: Determine the direction of the tectonic stress field in the engineering area based on the measured ground stress points, and perform a three-dimensional ground stress field inversion analysis on the engineering area based on the measured ground stress inversion data to determine the direction of the tectonic stress field in the engineering area.

[0012] Step 4: Install a circumferential automatic camera between the TBM cutterhead and the segment installer to photograph and record the rock strata, and store the images in computer software. Manually evaluate the rock mass quality index RQD and the joint group coefficient J of the rock strata. n Roughness coefficient J of joints r The alteration coefficient J of the joint a The influence coefficient of groundwater J ω The stress influence coefficient (SRF) was manually classified, and these data were calculated together with the TBM tunneling parameters, rock debris information and stress inversion data in steps 1, 2 and 3, and recorded as training samples.

[0013] The calculation formula is:

[0014] Where k1 is the correlation coefficient between the orientation of the structural surface and the angle between the tunnel axis and k2 is the correlation coefficient between the principal directions of the ground stress.

[0015] Step 5: Establish a deep learning artificial network identification model to correlate six parameters—cutterhead rotation speed, cutterhead torque, total propulsion force, propulsion cylinder pressure, tunneling speed, and cutterhead penetration—with the geomechanical properties and geostress of the rock mass.

[0016] An RBF neural network is used to establish a deep learning artificial network identification model. The RBF neural network is a three-layer neural network, which includes an input layer, a hidden layer, and an output layer. The transformation from the input layer to the hidden layer is non-linear, while the transformation from the hidden layer to the output layer is linear.

[0017] The activation function of the RBF neural network is expressed as follows:

[0018] in, This is the p-th input sample; Let x be the center vector of the radial basis function of the i-th hidden layer node; p = 1, 2, ..., P, where P is the total number of samples; ||x P -c i || represents the Euclidean norm; σ represents the variance of the Gaussian function.

[0019] The TBM tunneling parameters in step 1 are normalized by converting them to numbers between [0,1]. Using a self-organizing center selection learning method and a supervised learning process, the weights between the hidden and output layers are calculated. First, h centers are selected as k-means clustering centers. For the radial basis function of the Gaussian kernel, the variance is calculated using the formula: c max The maximum distance between the selected center points is used to calculate the connection weights of neurons from the hidden layer to the output layer directly using the least squares method. This involves solving for the partial derivative of the loss function with respect to ω and setting it to zero, resulting in the following formula:

[0020]

[0021] The RBF neural network selects the tunneling parameters of the TBM and the inverted in-situ stress data as input layer sample parameters, while the H value representing the rock mass quality is used as the output layer sample parameter.

[0022] Step 6: Use TBM tunneling parameters, rock debris information, in-situ stress inversion data, and training samples to perform deep learning training on the deep learning artificial network identification model from Step 5;

[0023] The deep learning artificial network identification model is trained until the grid converges and the set error standard is reached, thus obtaining the trained deep learning artificial network identification model.

[0024] Step 7: Test the trained deep learning artificial network recognition model;

[0025] Input the TBM excavation parameters and ground stress data of an excavated tunnel, and compare the difference between the output H value and the actual surrounding rock mass quality of the tunnel within the set error threshold range. If the recognition requirements are met, proceed to step 8; otherwise, increase the training sample data, adjust the RBF neural network training parameters, and repeat steps 5-6 until a deep learning artificial network recognition model that meets the actual engineering requirements is obtained.

[0026] Step 8: Upload the real-time TBM tunneling parameters and ground stress to the computer, analyze and process them in the deep learning artificial network identification model, and then calculate the H value according to the calculation formula to complete the rock mass quality prediction.

[0027] The beneficial effects of adopting the above technical solution are as follows:

[0028] This invention provides a method for rapidly determining the quality of rock mass during TBM tunneling. Compared to existing methods, this invention improves the Paton Q system classification, considering factors such as different rock strata attitudes, the relationship between the attitude of structural planes and the tunnel axis, and the principal direction of ground stress. This method uses RBF (Radial Basis Function) neural networks for deep learning, which is more efficient than other neural networks and better suited for rapid parameter confirmation. Furthermore, it only requires human assistance during the artificial neural network learning process and can achieve a high degree of artificial intelligence after deep learning. Since the quality parameters of the rock mass in front of the tunnel face cannot be directly measured during TBM tunneling, the rock mass between the TBM cutterhead and the segment assembler can be used as a training sample. Tunneling parameters at the tunnel face (cutterhead rotation speed, cutterhead torque, total propulsion force, propulsion cylinder pressure, tunneling speed, cutterhead penetration) can be used as input parameters to predict the quality of the rock mass in front of the tunnel face, providing a basis for rapid tunneling and avoiding difficulties such as excessive cutter wear or machine jamming. Attached Figure Description

[0029] Figure 1 This is a flowchart of the method for rapidly determining the quality of rock mass during TBM tunneling in an embodiment of the present invention;

[0030] Figure 2 This is a schematic diagram of ground stress and tunnel axis in an embodiment of the present invention;

[0031] Figure 3 This is a simplified flowchart of the artificial neural network in an embodiment of the present invention;

[0032] Figure 4 This is a conceptual diagram of TBM construction in an embodiment of the present invention. Detailed Implementation

[0033] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0034] A method for rapid determination of rock mass quality during TBM tunneling, such as Figure 1 As shown, the specific steps include:

[0035] Step 1: Collect TBM tunneling parameters during TBM tunneling, specifically including cutterhead rotation speed, cutterhead torque, total propulsion force, propulsion cylinder pressure, tunneling speed, and cutterhead penetration.

[0036] In this embodiment, the cutterhead rotation speed, propulsion cylinder pressure, tunneling speed, cutterhead penetration depth, cutterhead torque, and total propulsion force can be monitored and obtained through the TBM control console.

[0037] Step 2: Collect the rock debris generated during the excavation to obtain information about the rock mass, including the rock mass filling material, roughness, and whether the rock mass contains clay.

[0038] In this embodiment, a residue collection device is used to pick up the rock crushed by the cutter head, and the rock debris is subjected to three-dimensional imaging by infrared ranging to identify information such as the filling material of the rock mass, the roughness of the joints, and whether the rock mass contains clay.

[0039] Step 3: Determine the direction of the tectonic stress field in the engineering area based on the measured ground stress points, and perform a three-dimensional ground stress field inversion analysis on the engineering area based on the measured ground stress inversion data to determine the direction of the tectonic stress field in the engineering area.

[0040] In this embodiment, specifically, samples are taken from the tunnel rock mass every 30-50 meters to obtain the in-situ stress data for that section. This data is then used to perform in-situ stress inversion using the basic principles of Particle Swarm Optimization-Differential Decomposition (PSO-DE) and FLAC3D modeling. Figure 2 As shown;

[0041] Step 4: Install a circumferential automatic camera between the TBM cutterhead and the segment installer to photograph and record the rock strata, and store the images in computer software. Manually evaluate the rock mass quality index RQD and the joint group coefficient J of the rock strata. n Roughness coefficient J of joints r The alteration coefficient J of the joint a The influence coefficient of groundwater J ω The stress influence coefficient (SRF) was manually classified, and these data were calculated together with the TBM tunneling parameters, rock debris information and stress inversion data in steps 1, 2 and 3, and recorded as training samples.

[0042] In this embodiment, a circumferential automatic photographic camera is installed to photograph and record the exposed, unsupported rock strata between the cutterhead and the segment installer during tunneling. The data is stored in real time on a computer, and the data is manually analyzed based on rock mass quality indicators (RQD) and joint group coefficients (J). n Roughness coefficient J of joints rThe alteration coefficient J of the joint a The influence coefficient of groundwater J ω The six categories of in-situ stress influence coefficient (SRF) indicators are labeled and classified, and matched with the TBM tunneling parameters, rock debris information and in-situ stress inversion data at this time to form training samples for use in training artificial neural networks.

[0043] The calculation formula is:

[0044] Where k1 is the correlation coefficient between the orientation of the structural plane and the angle between the tunnel axis and k2 is the correlation coefficient between the principal directions of the ground stress. See Table 1 for details.

[0045] Table 1. Coefficients of k1 and k2

[0046]

[0047]

[0048] The standards for the values ​​of each coefficient are shown in Tables 2 to 7 below.

[0049] Table 2 Description and values ​​of rock quality index RQD

[0050]

[0051] Table 3. Number of Joint Groups (J) n Description and values

[0052]

[0053] Table 4 Joint roughness coefficient (J) r Description and values

[0054]

[0055] Table 5 Joint alteration coefficient (J) a Description and values

[0056]

[0057]

[0058] Table 6. Water Reduction Coefficient for Water Treatment (J) ω Description and values

[0059]

[0060] Table 7. Description and Values ​​of Stress Reduction Factor (SRF)

[0061]

[0062]

[0063] Step 5: Establish a deep learning artificial network recognition model, such as... Figure 3 As shown, six parameters—cutterhead rotation speed, cutterhead torque, total propulsion force, propulsion cylinder pressure, tunneling speed, and cutterhead penetration—are correlated with the geomechanical properties and geostress of the rock mass.

[0064] To ensure rapid learning and meet the real-time requirements of drilling as much as possible, an RBF neural network is used to establish a deep learning artificial network identification model. The RBF neural network is a three-layer neural network, which includes an input layer, a hidden layer, and an output layer. The transformation from the input layer to the hidden layer is non-linear, while the transformation from the hidden layer to the output layer is linear.

[0065] The activation function of the RBF neural network is expressed as follows:

[0066] in, This is the p-th input sample; Let x be the center vector of the radial basis function of the i-th hidden layer node; p = 1, 2, ..., P, where P is the total number of samples; ||x P -c i ‖ represents the Euclidean norm; σ represents the variance of the Gaussian function.

[0067] To effectively utilize the characteristics of the sigmoid function to ensure the nonlinear effect of network neurons and reduce neural network errors, the TBM tunneling parameters in step 1 are normalized, i.e., the parameters are uniformly converted into numbers between [0,1]. In the self-organizing center selection learning method, a supervised learning process is used to solve for the weights between the hidden layer and the output layer. First, h centers are selected as k-means clustering centers. For the radial basis function of the Gaussian kernel, the variance is solved by the formula: c max The maximum distance between the selected center points is used to calculate the connection weights of neurons from the hidden layer to the output layer directly using the least squares method. This involves solving for the partial derivative of the loss function with respect to ω and setting it to zero, resulting in the following formula:

[0068]

[0069] The RBF neural network selects the tunneling parameters of the TBM and the inverted geostress data as input layer sample parameters, while the H value, which characterizes the rock mass quality, is used as the output layer sample parameter.

[0070] Step 6: Use TBM tunneling parameters, rock debris information, geostress inversion data, and training samples to perform deep learning training on the deep learning artificial network identification model from Step 5.

[0071] The deep learning artificial network identification model is trained until the grid converges and the set error standard is reached, thus obtaining the trained deep learning artificial network identification model; the sample data used during training are all parameters collected in steps 1, 2, 3, and 4.

[0072] Step 7: Test the trained deep learning artificial network recognition model;

[0073] Input the TBM excavation parameters and ground stress data for an already excavated tunnel, such as... Figure 4 As shown, compare the difference between the output H value and the actual surrounding rock mass quality of the tunnel within the set error threshold range. If the recognition requirements are met, proceed to step 8; otherwise, increase the training sample data, adjust the RBF neural network training parameters, and repeat steps 5-6 until a deep learning artificial network recognition model that meets the actual engineering requirements is obtained.

[0074] Step 8: Upload the real-time TBM tunneling parameters and ground stress to the computer, analyze and process them in the deep learning artificial network identification model, and then calculate the H value according to the calculation formula to complete the rock mass quality prediction.

[0075] After the test is completed, simply input the real-time TBM tunneling parameters during tunnel excavation, namely cutterhead rotation speed, cutterhead torque, total propulsion force, propulsion cylinder pressure, tunneling speed, cutterhead penetration, and inverted ground stress data, into the computer. After calculation and analysis by the artificial neural network, the H-value of the surrounding rock quality in front of the tunnel face can be predicted.

[0076] The above description is merely a preferred embodiment of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in the embodiments of this disclosure.

Claims

1. A method for rapidly determining the quality of rock mass during TBM tunneling, characterized in that, Includes the following steps: Step 1: Collect TBM tunneling parameters during TBM tunneling, specifically including cutterhead rotation speed, cutterhead torque, total propulsion force, propulsion cylinder pressure, tunneling speed, and cutterhead penetration. Step 2: Collect the rock debris generated during the excavation to obtain information about the rock mass, including the rock mass filling material, roughness, and whether the rock mass contains clay. Step 3: Determine the direction of the tectonic stress field in the engineering area based on the measured ground stress points, and perform a three-dimensional ground stress field inversion analysis on the engineering area based on the measured ground stress inversion data to determine the direction of the tectonic stress field in the engineering area. Step 4: Install a circumferential automatic camera between the TBM cutterhead and the segment installer to photograph and record the rock strata, and store the images in computer software. Manually evaluate the rock mass quality index RQD and the joint group coefficient J of the rock strata. n Roughness coefficient J of joints r The alteration coefficient J of the joint a The influence coefficient of groundwater J ω The stress influence coefficient (SRF) was manually classified, and these data were calculated together with the TBM tunneling parameters, rock debris information and stress inversion data in steps 1, 2 and 3, and recorded as training samples. The calculation formula is: ×k1×k2; Where k1 is the correlation coefficient between the orientation of the structural surface and the angle between the tunnel axis and k2 is the correlation coefficient between the principal directions of the ground stress. Step 5: Establish a deep learning artificial network identification model to correlate six parameters—cutterhead rotation speed, cutterhead torque, total propulsion force, propulsion cylinder pressure, tunneling speed, and cutterhead penetration—with the geomechanical properties and geostress of the rock mass. Specifically, the deep learning artificial network identification model is established using an RBF neural network; the RBF neural network is a three-layer neural network, which includes an input layer, a hidden layer, and an output layer; the transformation from the input layer to the hidden layer is non-linear, while the transformation from the hidden layer to the output layer is linear. The activation function of the RBF neural network is expressed as follows: ; in, This is the p-th input sample; Let p be the center vector of the radial basis function of the i-th hidden layer node; p = 1, 2, ..., P, where P is the total number of samples. It is a Euclidean norm; The variance of the Gaussian function; The TBM tunneling parameters in step 1 are normalized by converting them to numbers between [0,1]. Using a self-organizing center selection learning method and a supervised learning process, the weights between the hidden and output layers are calculated. First, h centers are selected as k-means clustering centers. For the radial basis function of the Gaussian kernel, the variance is calculated using the formula: ; ,c max The maximum distance between the selected center points is used to calculate the connection weights of neurons from the hidden layer to the output layer directly using the least squares method. This involves solving for the partial derivative of the loss function with respect to ω and setting it to zero, resulting in the following formula: ; ； ； The RBF neural network selects the tunneling parameters of the TBM and the inverted in-situ stress data as input layer sample parameters, while the H value representing the rock mass quality is used as the output layer sample parameter. Step 6: Use TBM tunneling parameters, rock debris information, in-situ stress inversion data, and training samples to perform deep learning training on the deep learning artificial network identification model from Step 5; The deep learning artificial network identification model is trained until the grid converges and the set error standard is reached, thus obtaining the trained deep learning artificial network identification model. Step 7: Test the trained deep learning artificial network recognition model; Input the TBM excavation parameters and ground stress data of an excavated tunnel, and compare the output H value with the difference between the actual surrounding rock mass quality of the tunnel and the set error threshold. If the recognition requirements are met, proceed to step 8. If the requirements are not met, increase the training sample data, adjust the RBF neural network training parameters, and repeat steps 5-6 until a deep learning artificial network recognition model that meets the actual engineering requirements is obtained. Step 8: Upload the real-time TBM tunneling parameters and ground stress to the computer, analyze and process them in the deep learning artificial network identification model, and then calculate the H value according to the calculation formula to complete the rock mass quality prediction.

Images