A surrounding rock design method and system for a tunnel impact test

Abstract

The application provides a surrounding rock design method and system for a tunnel impact test, and relates to engineering structure test technology, and comprises the following steps: constructing an impact tunnel numerical model based on pre-test data; simulating based on the impact tunnel numerical model to obtain the influence on the surrounding rock under different impact conditions; performing deep learning based on the impact conditions and the geometric information of the area affected by the impact to construct a prediction model; inputting a to-be-tested condition into the prediction model to obtain predicted influence geometric information; and obtaining surrounding rock design parameters through a pre-constructed first model according to the predicted influence geometric information. Through quantitative analysis of simulation and deep learning, the method avoids the subjectivity of traditional experience design, makes the surrounding rock design more scientific and targeted, can effectively simulate the response of the surrounding rock under the impact of the tunnel, and guarantees the accuracy of the test results.

Description

Technical Field

[0001] This invention relates to the field of engineering structure testing technology, specifically a method and system for designing surrounding rock for tunnel impact testing. Background Technology

[0002] With the advancement of transportation infrastructure construction, the number of completed tunnels is increasing, and train speeds are rising, significantly increasing the probability of derailment and collisions. The resulting losses are also more severe, necessitating research on tunnel impact response. Previous research on tunnel impact testing primarily employed numerical simulations. However, numerical simulations involve high-speed contact problems, complex numerical modeling, and a large computational workload; furthermore, the influencing factors of train impacts with shield tunnels are extremely complex. Therefore, tunnel impact experiments are needed to provide data support for design and research.

[0003] When a tunnel is impacted, the interaction between the surrounding rock and the tunnel affects the tunnel's load-bearing capacity. Furthermore, the properties and boundary parameters of the surrounding rock significantly influence the tunnel's response. However, existing impact testing devices only consider the impact when a train contacts the tunnel plane, neglecting the interaction between the tunnel and the surrounding rock during the impact process. Because the surrounding rock model in the testing device has boundaries, wave reflection and superposition effects exist. Setting different surrounding rock parameters during the test leads to different obtained response data, meaning the data deviates from the ideal real-world situation. Therefore, in experiments studying the interaction between surrounding rock and tunnel, accurately designing the surrounding rock to obtain reliable response data has become an urgent problem to be solved. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for designing the surrounding rock in tunnel impact tests, thereby improving the aforementioned problems. To achieve this purpose, the technical solution adopted by this invention is as follows:

[0005] In a first aspect, this application provides a method for designing the surrounding rock for tunnel impact tests, including:

[0006] A numerical model of the impact tunnel was constructed based on preliminary test data;

[0007] Simulations were performed based on a numerical model of an impact tunnel to obtain the impact results on the surrounding rock under different impact conditions; the impact results include the geometric information of the impact-affected area.

[0008] A prediction model is constructed based on deep learning of the impact conditions and the geometric information of the corresponding impact-affected areas.

[0009] Input the working condition to be tested into the prediction model to obtain the predicted influence geometric information;

[0010] According to the predicted influence geometric information, the surrounding rock design parameters are obtained through the pre-constructed first model, and the surrounding rock design parameters include back plate distance, bedrock height, soil body burial depth and plate shape arc.

[0011] In a second aspect, the application also provides a surrounding rock design system for a tunnel impact test, and the system comprises:

[0012] A first module is configured to construct an impact tunnel numerical model based on pre-test data;

[0013] A second module is configured to simulate based on the impact tunnel numerical model to obtain influence results on surrounding rock under different impact conditions; the influence results include regional geometric information affected by the impact;

[0014] A third module is configured to perform deep learning based on the impact conditions and the corresponding regional geometric information affected by the impact to construct a prediction model;

[0015] A fourth module is configured to input a to-be-tested condition into the prediction model to obtain predicted influence geometric information;

[0016] A fifth module is configured to obtain surrounding rock design parameters, including back plate distance, bedrock height, soil body burial depth and plate shape arc, according to the predicted influence geometric information and through the pre-constructed first model.

[0017] Beneficial effects:

[0018] The method can quickly and accurately output surrounding rock influence geometric information according to a to-be-tested condition by collecting data based on simulation and establishing a prediction model by deep learning; the prediction geometric information is converted into specific design parameters such as back plate distance and bedrock height through the pre-constructed first model, and the parameters cover the dimensions of soil body and plate shape arc; the quantification analysis of simulation and deep learning avoids the subjectivity of traditional experience design, makes the surrounding rock design more scientific and targeted, effectively simulates the surrounding rock response under the impact of a tunnel, and guarantees the accuracy of test results.

[0019] Other features and advantages of the present application will be illustrated in the following description, and some will become apparent from the description, or will be understood by those skilled in the art through implementation of the embodiments of the present application. BRIEF DESCRIPTION OF DRAWINGS

[0020] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments, and it should be understood that the following drawings only show some embodiments of the present application, and therefore should not be regarded as a limitation to the scope, and for those skilled in the art, other related drawings can also be obtained without creative labor on the premise of the drawings.

[0021] Figure 1A flow chart of the surrounding rock design method for the tunnel impact test in the embodiment;

[0022] Figure 2 A structural schematic diagram of the impact test device in the embodiment;

[0023] Figure 3 A structural diagram of the surrounding rock design equipment for the tunnel impact test in the embodiment.

[0024] Symbol explanation: 1-air cannon, 2-acceleration pipeline, 3-rail, 4-base, 5-infrared speedometer, 6-semi-tunnel model, 7-soil filling box, 8-top cover, 9-soil filling box base, 800-surrounding rock design equipment for the tunnel impact test, 801-processor, 802-memory, 803-multimedia assembly, 804-I / O interface, 805-communication assembly. DETAILED DESCRIPTION

[0025] To make the objects, technical solutions and advantages of the embodiments of the present application clearer, the following will be combined with the drawings in the embodiments of the present application to make a clear and complete description of the technical solutions in the embodiments of the present application. Obviously, the described embodiments are some of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application described and shown in the drawings can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present application.

[0026] It should be noted that: similar labels and letters represent similar items in the following drawings, therefore, once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings.

[0027] Embodiment 1:

[0028] Referring to Figure 1 , the embodiment provides a surrounding rock design method for a tunnel impact test, comprising steps S100, S200, S300, S400 and S500.

[0029] S100, constructing an impact tunnel numerical model based on pre-test data;

[0030] When the experiment starts, the position and angle of the impact tunnel need to be set, a pre-loading test is performed, a low-strength test block with the same mass is used to impact a force response plate at the position of the impact tunnel at a target speed, the response plate is equipped with an acceleration , stress , pressure Pressure range The relevant sensors are then used to collect data via a high-frequency acquisition system. The collected data is processed according to the following analysis workflow:

[0031] First, calculate the internal forces of the structure based on stress. :

[0032] ;

[0033] By measuring acceleration The inertial force of the structure itself during the impact was calculated. :

[0034] ;

[0035] Then based on pressure With pressure range Receive the impact force of the train ;

[0036] Based on the dynamics formula, the external forces on the structure are obtained. :

[0037] ;

[0038] Then, based on the formula for the resultant force of external forces, the impact force is corrected to obtain the impact force acting on the soil:

[0039] ;

[0040] Based on Newton's second law, a numerical analysis model is established to solve the equations of motion of the continuum and to establish a database of surrounding rock design parameters.

[0041] ;

[0042] : Quality matrix; : Acceleration vector of surrounding rock nodes; : Surrounding rock damping matrix; : Velocity vector of surrounding rock nodes; Rock stiffness matrix; : Displacement vector of surrounding rock nodes; : External force vector at the surrounding rock node.

[0043] Based on the above equations, a finite element numerical model of a semi-infinite soil is established.

[0044] S200. Simulations are performed based on a numerical model of an impact tunnel to obtain the impact results on the surrounding rock under different impact conditions; the impact results include the geometric information of the impact-affected area.

[0045] Different impact conditions are set for simulation to obtain the damage data of the tunnel and the influence results of the surrounding rock, the influence results including geometric information of an area affected by the impact, for example, a diffusion range of a stress wave generated by the impact in the soil; details are as follows:

[0046] According to the peak stress of each grid in the simulation result, an affected area is identified, and surface triangular grids of the affected area are extracted; here, in order to accurately identify the affected area, in addition to the peak stress, data such as plastic strain, peak particle velocity, and energy density cloud map can also be considered; the entire grid space is binarized according to a preset threshold to obtain the affected area, and then surface triangular grids are extracted;

[0047] The surface triangular grids are uniformly resampled to generate a preset number of three-dimensional points to obtain a point cloud;

[0048] Each point cloud is rotated and aligned according to the tunnel axis, the point cloud is translated so that its centroid is at a preset origin, and the point cloud is scaled to fall within a preset unit sphere.

[0049] S300, based on the impact condition and the corresponding area geometric information affected by the impact, deep learning is performed to construct a prediction model;

[0050] The area geometric information includes three-dimensional point cloud information; this step specifically includes:

[0051] The impact condition is input into a preset encoder to output a target latent vector;

[0052] The impact condition includes impact force, area, speed, impact angle, impact position, rock type, tunnel thickness, and longitudinal curvature; the condition parameters are encoded and spliced to obtain a condition parameter vector; a fully connected layer is used as an encoder, the condition parameter vector is input, and through a series of fully connected layers, its dimension is gradually increased, and nonlinearity is introduced. The last fully connected layer maps the dimension to a preset latent space size, and the output is the target latent vector;

[0053] The decoder works as follows:

[0054] A uniform two-dimensional grid is created; for example, a (H, 2) matrix is generated for a point cloud with a point number H=2048;

[0055] The target latent vector is copied multiple times to construct a latent tensor, which is spliced with the two-dimensional grid to obtain a first tensor, wherein the number of copies is determined according to the size of the two-dimensional grid; if the size of the two-dimensional grid is (H, 2), the target latent vector is copied H times; the first tensor indicates that each two-dimensional grid point is assigned global condition information;

[0056] The first tensor is processed by the first multi-layer perception to obtain an estimated three-dimensional point cloud; the first obtained three-dimensional point cloud is relatively rough and only has a rough shape estimation;

[0057] The target latent vector is spliced with the estimated three-dimensional point cloud to obtain a second tensor;

[0058] The second tensor is processed by the second multi-layer perception to obtain a predicted three-dimensional point cloud.

[0059] After the above encoder-decoder architecture is built, the encoder and the decoder need to be trained using known sample data; on the one hand, the Chamfer Distance is calculated as a loss term, which is used to measure the similarity between the actual point cloud and the predicted point cloud (that is, the order of points does not affect); on the other hand, there may be a case where the shape and position are similar but the distribution direction is different, so a loss term for measuring the consistency of the direction needs to be introduced, which is constructed as follows:

[0060] The covariance matrix of the real sample point cloud is calculated based on the three-dimensional point cloud information, and the covariance matrix is decomposed to obtain real principal component vectors;

[0061] First, the point cloud needs to be translated to the center of mass, and the variance matrix is calculated, that is, the three coordinate axes of the point cloud are regarded as three random variables, the linear correlation between them is measured, and the principal direction (maximum variance direction) and the anisotropy degree are given.

[0062] The centered point cloud is represented as ;

[0063] ;

[0064] is the transpose of ;

[0065] ;

[0066] The diagonal elements are the variances of the point cloud in the x, y, and z directions, and the non-diagonal elements are the covariances, which represent the cooperative changes in different directions

[0067] The covariance matrix is decomposed:

[0068] ;

[0069] Λ is a 3x3 diagonal matrix, and the three values on the diagonal are the eigenvalues (λ1, λ2, λ3), and it is usually assumed that λ1 ≥ λ2 ≥ λ3. V Each column of V is the corresponding eigenvector (v1, v2, v3), that is, the real principal component vectors of the three-axis direction.

[0070] The predicted 3D point cloud generated by the second multilayer perceptron is obtained, the covariance matrix of the predicted 3D point cloud is calculated, and feature decomposition is performed to obtain the predicted principal component vector.

[0071] The predicted principal component vectors are calculated using the same method described above.

[0072] The principal direction loss term is constructed by calculating the angle between the true principal component vector and the predicted principal component vector. ;

[0073] The position loss term is constructed by calculating the chamfer distance between the real sample point cloud and the predicted 3D point cloud;

[0074] ;

[0075] in, This is the location loss value. The smaller the value, the more similar the two point clouds are in shape and spatial location; P For real sample point clouds, p It is a point in a real sample point cloud; Q To predict 3D point clouds, q To predict a point in a 3D point cloud; The square of the Euclidean distance; The total number of points in the real sample point cloud; To predict the total number of points in a 3D point cloud;

[0076] The loss function of the prediction model is constructed based on the main direction loss term and the position loss term. According to the loss function, the encoder is updated through backpropagation, and the parameters of the decoder are updated simultaneously. ;

[0077] S400. Input the working condition to be measured into the prediction model to obtain the predicted influence geometric information;

[0078] Input the working condition parameters to be tested into the model to obtain the predicted three-dimensional point cloud, which represents the spatial range of the surrounding rock that may be affected by the impact under the working condition. Based on this spatial range, select the surrounding rock design parameters to minimize the impact of unreasonable surrounding rock structure design on the test results.

[0079] S500. Based on the predicted influence geometry information, the surrounding rock design parameters are obtained through a pre-constructed first model. These parameters include backplate distance, bedrock height, soil depth, and plate shape curvature. The process includes the following steps:

[0080] First, features are extracted from the 3D point cloud:

[0081] The total volume and surface area of the predicted three-dimensional point cloud are extracted by convex hull estimation, and the ratio of the volume and surface area is calculated to obtain basic geometric features;

[0082] The maximum and minimum values of the point cloud in the three-axis direction of the tunnel are extracted respectively, and the span of the point cloud in the three-axis direction of the tunnel is calculated to obtain the spatial distribution feature; this feature is used to determine the absolute position and range of the influence area in space;

[0083] The minimum distance and direction of the point cloud centroid to the tunnel inner contour surface are calculated, and the depth of the point cloud in the direction of the vault, side wall and inverted arch is calculated respectively to obtain the position feature;

[0084] The shape feature is extracted by principal component analysis:

[0085] The principal component analysis first calculates the eigenvalues λ1, λ2, λ3 of each principal component in the same way as step S300;

[0086] Then, based on the eigenvalues, the linearity, planarity, sphericity, isotropy and other features are calculated;

[0087] The extracted features are spliced to obtain a point cloud feature vector, and the point cloud feature vector is input into the first model to obtain the surrounding rock design parameter.

[0088] The tunnel surrounding rock model of the present application adopts a half-tunnel model 6, which is installed on one side of a filling box 7, and artificial soil is filled in the filling box 7; the back plate is a plate member opposite to the half-tunnel model 6 of the filling box 7, and the back plate distance refers to the distance between the back plate and the half-tunnel model 6; in addition, the distance between the two side plates also needs to be designed, the bedrock height refers to the height of the half-tunnel model 6 from the bottom of the filling box 7; the soil body burial depth refers to the depth of the filled soil; the plate member shape curvature refers to the three plate members of the filling box 7, which can be designed as a curved surface, and the change of the curved surface curvature will also cause differences in the obtained response data, which is a parameter that needs to be accurately designed.

[0089] The construction process of the first model is as follows:

[0090] Obtain test samples, the test samples include historical surrounding rock design parameters and corresponding historical test data, the historical test data includes impact working condition data and real response data;

[0091] According to the impact working condition data and the historical surrounding rock design parameters, simulation is carried out to obtain simulation response data;

[0092] The real response data and the simulation response data are compared, for example, the cosine similarity is calculated, and the reliability of the historical surrounding rock design parameters is determined according to the comparison result; when the real response data is similar to the simulation response data, it means that the adopted surrounding rock design parameters are reasonable under this working condition, and this test can be used as a training sample for parameter learning.

[0093] According to the reliability of the historical surrounding rock design parameters, the test samples are screened to obtain training samples;

[0094] The training samples are used to train the gradient boosting tree to obtain a first model, and LightGBM or XGBoost model can be used. The relationship between the design parameters and the geometric features is not a simple linear relationship. Gradient boosting tree can automatically learn these complex and nonlinear mapping relationships by combining multiple weak learners, and is usually better than other models.

[0095] The embodiment also provides a collision test device for implementing the above method, referring to Figure 2 , which comprises an air cannon 1, an acceleration pipeline 2, a rail 3, a base 4, an infrared speedometer 5, a half-tunnel model 6, a filling box 7, a top cover 8 and a filling box base 9.

[0096] The air cannon 1 and the acceleration pipeline 2 form a launching device. When the control valve is opened, the air cannon 1 pushes the train model on the rail 3 out from the starting position of the acceleration pipeline, completes the acceleration process in the entire acceleration pipeline, and then the train model continues to run by inertia. The rail 3 is a steel member, and the train is clamped on the track to run forward, ensuring the running track of the train.

[0097] The base 4 has a lifting function and a rotating function. The lifting function is used to ensure the normal running of the train. Through the rotation of the base 4, the train can hit the tunnel at an angle of the side overturning posture.

[0098] The infrared speedometer 5 is installed at the end of the launching device and is used to monitor the impact speed of the train. The tunnel model is made of tile concrete. According to the requirements of the tunnel test, the diameter, thickness and longitudinal curvature of the tunnel are determined to make the half-tunnel model 6.

[0099] Embodiment 2

[0100] The embodiment provides a surrounding rock design system for a tunnel collision test, which comprises:

[0101] A first module is used to construct a collision tunnel numerical model based on pre-test data;

[0102] A second module is used to simulate based on the collision tunnel numerical model to obtain an influence result of surrounding rock under different collision working conditions; the influence result comprises regional geometric information affected by the collision;

[0103] A third module is used to perform deep learning based on the collision working condition and the corresponding regional geometric information affected by the collision to construct a prediction model;

[0104] A fourth module is used to input a to-be-tested working condition into the prediction model to obtain predicted influence geometric information;

[0105] a fifth module configured to obtain a surrounding rock design parameter by using the first pre-constructed model according to the predicted influence geometry information, the surrounding rock design parameter including a back plate distance, a bedrock height, a soil depth, and a plate shape curvature.

[0106] As an optional implementation, the region geometry information includes three-dimensional point cloud information; and the third module includes:

[0107] a first unit configured to input the impact working condition into a preset encoder to output a target latent vector;

[0108] a second unit configured to create a uniform two-dimensional grid;

[0109] a third unit configured to copy the target latent vector multiple times to construct a latent tensor, and splice the latent tensor with the two-dimensional grid to obtain a first tensor, wherein the number of times of copying is determined according to a size of the two-dimensional grid;

[0110] a fourth unit configured to process the first tensor by using a first multi-layer perception to obtain an estimated three-dimensional point cloud;

[0111] a fifth unit configured to splice the target latent vector with the estimated three-dimensional point cloud to obtain a second tensor;

[0112] a sixth unit configured to process the second tensor by using a second multi-layer perception to obtain a predicted three-dimensional point cloud.

[0113] As an optional implementation, the third module further includes:

[0114] a seventh unit configured to calculate a covariance matrix of a real sample point cloud based on the three-dimensional point cloud information, and perform eigenvalue decomposition on the covariance matrix to obtain a real principal component vector;

[0115] an eighth unit configured to obtain a predicted three-dimensional point cloud generated by the second multi-layer perception, calculate a covariance matrix of the predicted three-dimensional point cloud, and perform eigenvalue decomposition to obtain a predicted principal component vector;

[0116] a ninth unit configured to construct a principal direction loss term by calculating an included angle between the real principal component vector and the predicted principal component vector;

[0117] a tenth unit configured to construct a position loss term by calculating a chamfer distance between the real sample point cloud and the predicted three-dimensional point cloud;

[0118] an eleventh unit configured to construct a loss function of the prediction model based on the principal direction loss term and the position loss term, and update the encoder by back propagation according to the loss function.

[0119] As an optional implementation, the fifth module comprises:

[0120] The twelfth unit is configured to extract the total volume and surface area of the predicted three-dimensional point cloud by convex hull estimation, and calculate the ratio of the volume and the surface area to obtain basic geometric features;

[0121] The thirteenth unit is configured to extract the maximum and minimum values of the point cloud in the three-axis directions of the tunnel respectively, and calculate the span of the point cloud in the three-axis directions of the tunnel to obtain spatial distribution features;

[0122] The fourteenth unit is configured to calculate the minimum distance and direction of the point cloud centroid to the inner contour surface of the tunnel, and calculate the depth of the point cloud in the direction of the vault, the side wall and the inverted arch respectively to obtain position features;

[0123] The fifteenth unit is configured to extract shape features by principal component analysis;

[0124] The sixteenth unit is configured to splice the extracted features to obtain a point cloud feature vector, and input the point cloud feature vector into the first model to obtain the surrounding rock design parameters.

[0125] Embodiment 3:

[0126] Corresponding to the above method embodiment, the present embodiment also provides a surrounding rock design device for tunnel impact test. The surrounding rock design device for tunnel impact test described below can be mutually corresponding and referred to with the surrounding rock design method for tunnel impact test described above.

[0127] Figure 3 is a block diagram of a surrounding rock design device 800 for tunnel impact test according to an exemplary embodiment. As shown in Figure 3As shown, the tunnel impact test surrounding rock design device 800 includes a processor 801 and a memory 802. The tunnel impact test surrounding rock design device 800 can also include one or more of a multimedia component 803, an input / output (I / O) interface 804, and a communication component 805. Among them, the processor 801 is used to control the overall operation of the tunnel impact test surrounding rock design device 800 to complete all or part of the steps in the tunnel impact test surrounding rock design method described above. The memory 802 is used to store various types of data to support the operation of the tunnel impact test surrounding rock design device 800, which can include, for example, commands for operating any application or method on the tunnel impact test surrounding rock design device 800, and application-related data, such as contact data, sent and received messages, pictures, audio, video, and the like. The memory 802 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0128] The multimedia component 803 can include a screen and an audio component. Among them, the screen can be a touch screen, for example, and the audio component is used to output and / or input audio signals. For example, the audio component can include a microphone for receiving external audio signals.

[0129] The received audio signals can be further stored in the memory 802 or sent out through the communication component 805. The audio component also includes at least one speaker for outputting audio signals. The I / O interface 804 provides an interface between the processor 801 and other interface modules, which can be a keyboard, a mouse, a button, etc. These buttons can be virtual buttons or physical buttons. The communication component 805 is used for the tunnel impact test surrounding rock design device 800 to communicate with other devices through wired or wireless communication. The wireless communication, for example, Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G or 4G, or a combination of one or more of them, so the corresponding communication component 805 can include a Wi-Fi module, a Bluetooth module, an NFC module.

[0130] Embodiment 4:

[0131] Corresponding to the above method embodiment of the tunnel impact test surrounding rock design method, the embodiment also provides a readable storage medium. The readable storage medium described below can be correspondingly referred to the tunnel impact test surrounding rock design method described above.

[0132] A readable storage medium, the readable storage medium stores a computer program, and the computer program is executed by a processor to implement the steps of the above tunnel impact test surrounding rock design method embodiment.

[0133] The readable storage medium can be a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various readable storage media that can store program codes.

[0134] It should be noted that, in this paper, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between the entities or operations. Moreover, the term "include", "contain" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or device. Without more limitations, the element defined by the statement "including a" does not exclude the presence of other identical elements in the process, method, article or device including the element.

[0135] The above merely illustrates the specific embodiments of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of the changes or replacements within the technical range disclosed by the present application, which should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method of designing a surrounding rock for a tunnel impact test, characterized in that, The method comprises the following steps: constructing a numerical model of a tunnel impacted by a projectile based on pre-experiment data; conducting simulation based on the numerical model of the tunnel impacted by the projectile to obtain an impact result of surrounding rock under different impact conditions; the impact result comprises geometric information of an area affected by the impact; conducting deep learning based on the impact condition and the corresponding geometric information of the area affected by the impact, the geometric information comprising three-dimensional point cloud information, and constructing a prediction model, which comprises the following steps: inputting the impact condition into a preset encoder to output a target latent vector; creating a uniform two-dimensional grid; copying the target latent vector multiple times to construct a latent tensor, and splicing the latent tensor with the two-dimensional grid to obtain a first tensor, wherein the number of times of copying is determined according to the size of the two-dimensional grid; processing the first tensor by using a first multi-layer perceptron to obtain an estimated three-dimensional point cloud; splicing the target latent vector with the estimated three-dimensional point cloud to obtain a second tensor; processing the second tensor by using a second multi-layer perceptron to obtain a predicted three-dimensional point cloud; calculating a covariance matrix of a real sample point cloud based on the three-dimensional point cloud information, and performing eigenvalue decomposition on the covariance matrix to obtain a real principal component vector; obtaining the predicted three-dimensional point cloud generated by the second multi-layer perceptron, calculating a covariance matrix of the predicted three-dimensional point cloud, and performing eigenvalue decomposition to obtain a predicted principal component vector; constructing a principal direction loss term by calculating the included angle between the real principal component vector and the predicted principal component vector; constructing a position loss term by calculating the chamfer distance between the real sample point cloud and the predicted three-dimensional point cloud; constructing a loss function of the prediction model based on the principal direction loss term and the position loss term, and updating the encoder through back propagation according to the loss function; inputting a to-be-tested condition into the prediction model to obtain predicted impact geometric information; obtaining surrounding rock design parameters by using a pre-constructed first model according to the predicted impact geometric information, the surrounding rock design parameters comprising a back plate distance, a bedrock height, a soil body burial depth, and a plate shape curvature.

2. The method of designing a tunnel impact test of surrounding rock according to claim 1, characterized in that, The method comprises the following steps: extracting the total volume and surface area of the predicted three-dimensional point cloud by convex hull estimation, and calculating the ratio of the volume to the surface area to obtain basic geometric features; extracting the maximum and minimum values of the point cloud in the three-axis directions of the tunnel respectively, and calculating the span of the point cloud in the three-axis directions of the tunnel to obtain spatial distribution features; calculating the minimum distance and direction of the centroid of the point cloud to the inner contour surface of the tunnel, and calculating the depth of the point cloud in the direction of the vault, the side wall and the inverted arch respectively to obtain position features; extracting shape features by principal component analysis; splicing the extracted features to obtain a point cloud feature vector, and inputting the point cloud feature vector into the first model to obtain the surrounding rock design parameters.

3. The tunnel impact test of surrounding rock design method according to claim 1, characterized in that, The method comprises the following steps: obtaining a test sample, the test sample comprising historical surrounding rock design parameters and corresponding historical test data, the historical test data comprising impact condition data and real response data; conducting simulation according to the impact condition data and the historical surrounding rock design parameters to obtain simulation response data; comparing the real response data with the simulation response data, and determining the reliability of the historical surrounding rock design parameters according to the comparison result; The training sample is obtained by screening the test sample according to the reliability of historical surrounding rock design parameters; The first model is obtained by training the gradient boosting tree using the training sample.

4. The method of designing a tunnel impact test of surrounding rock according to claim 1, characterized in that, The influence results of the surrounding rock under different impact conditions are obtained, including: According to the peak stress of each grid in the simulation result, the affected area is identified, and the surface triangular grid of the affected area is extracted; The surface triangular grid is uniformly resampled to generate a preset number of three-dimensional points to obtain the point cloud; According to the tunnel axis, each point cloud is rotated and aligned, the centroid of the point cloud is translated to a preset origin, and the point cloud is scaled to fall within a preset unit sphere.

5. A system for designing a surrounding rock of a tunnel impact test, characterized by It comprises: The first module is configured to construct an impact tunnel numerical model based on pre-test data; The second module is configured to simulate based on the impact tunnel numerical model to obtain influence results of the surrounding rock under different impact conditions; the influence results include the geometric information of the area affected by the impact; The third module is configured to perform deep learning based on the impact condition and the corresponding geometric information of the area affected by the impact, and to construct a prediction model, wherein the geometric information includes three-dimensional point cloud information. The third module comprises: The first unit is configured to input the impact condition into a preset encoder to output a target latent vector; The second unit is configured to create a uniform two-dimensional grid; The third unit is configured to copy the target latent vector multiple times to construct a latent tensor, and splice the latent tensor with the two-dimensional grid to obtain a first tensor, wherein the number of times of copying is determined according to the size of the two-dimensional grid; The fourth unit is configured to process the first tensor using a first multi-layer perceptron to obtain an estimated three-dimensional point cloud; The fifth unit is configured to splice the target latent vector with the estimated three-dimensional point cloud to obtain a second tensor; The sixth unit is configured to process the second tensor using a second multi-layer perceptron to obtain a predicted three-dimensional point cloud; The seventh unit is configured to calculate the covariance matrix of the real sample point cloud based on the three-dimensional point cloud information, and perform eigenvalue decomposition on the covariance matrix to obtain a real principal component vector; The eighth unit is configured to obtain the predicted three-dimensional point cloud generated by the second multi-layer perceptron, calculate the covariance matrix of the predicted three-dimensional point cloud, and perform eigenvalue decomposition to obtain a predicted principal component vector; The ninth unit is configured to construct a principal direction loss term by calculating the included angle between the real principal component vector and the predicted principal component vector; The tenth unit is configured to construct a position loss term by calculating the chamfer distance between the real sample point cloud and the predicted three-dimensional point cloud; The eleventh unit is configured to construct a loss function of the prediction model based on the principal direction loss term and the position loss term, and update the encoder according to the loss function through back propagation; The fourth module is configured to input a to-be-tested condition into the prediction model to obtain predicted influence geometric information; The fifth module is configured to obtain the surrounding rock design parameters including the back plate distance, the bedrock height, the soil burial depth and the plate shape curvature based on the predicted influence geometric information and the pre-constructed first model.

6. The tunnel impact test of surrounding rock design system according to claim 5, characterized in that, The fifth module comprises: The twelfth unit is configured to extract the total volume and surface area of the predicted three-dimensional point cloud by convex hull estimation, calculate the ratio of the volume and the surface area, and obtain the basic geometric features. The thirteenth unit is configured to extract maximum and minimum values of the point cloud in three axial directions of the tunnel respectively, and calculate span of the point cloud in the three axial directions of the tunnel to obtain a spatial distribution feature; The fourteenth unit is configured to calculate minimum distance and direction from a point cloud centroid to a tunnel inner contour surface, and calculate depths of the point cloud in a vault, a sidewall and an invert direction respectively to obtain a position feature; The fifteenth unit is configured to extract a shape feature by principal component analysis; The sixteenth unit is configured to splice the extracted features to obtain a point cloud feature vector, and input the point cloud feature vector into a first model to obtain a surrounding rock design parameter.

Images