Layered rock deformation analysis method and device based on data driving

Abstract

The invention discloses a stratified rock deformation analysis method and device based on data driving. The stratified rock deformation analysis method comprises the steps that a stress-strain database of stratified rocks is established based on a test and data recognition algorithm; converting the data into a database of the stratified rock at the target angle by using a coordinate conversion formula of the database; constructing a statically indeterminate equation for the stress-strain data of each data point of the database, and taking a least square method solution as a balance factor matrix; and finally, calculating a stress field, a strain field and a displacement field of the stratified rock at the target angle under the loading and unloading effects by adopting a local convex set data driving calculation method considering the noise influence of the test data. The method is suitable for calculating deformation and stress distribution of the stratified rock at different angles under the action of external loading and unloading, deformation analysis of the stratified rock under the action of loading and unloading can be carried out without defining the constitutive relation of the stratified rock, and the method has important significance on research of rock mass which is a complex heterogeneous and anisotropic medium.

Description

Technical Field

[0001] The invention belongs to the technical field of layered rock deformation, and in particular relates to a layered rock deformation analysis method and device based on data drive. Background Art

[0002] In nature, rocks with layered structures account for two-thirds of the land area. Layered rock has obvious anisotropy characteristics, which are mainly affected by the direction of the bedding plane and the mechanical properties of the rock material. Correctly understanding the basic mechanical properties of layered rock is of great significance to the stability analysis and safety evaluation of rock engineering.

[0003] Before solving various numerical analysis methods for rock engineering, the constitutive model and mechanical parameters of geotechnical materials must be defined in advance, which directly affects the results of engineering stability evaluation. Due to the heterogeneity and anisotropy of rock bodies, there is no unified and mature constitutive model that can fully and correctly represent the constitutive characteristics of various types of rock bodies under any loading conditions. In addition, the proposal and verification of new constitutive models require high intellectual and time costs. Various empirical assumptions may also lead to calculation errors and a decrease in the scope of model applicability, while destroying the originality of the test data.

[0004] In view of the above problems, the present invention aims to provide a numerical simulation technology that does not require the definition of a rock mass constitutive model. This technology can be used for deformation analysis of layered rock masses under loading and unloading, as well as for excavation and reinforcement design of layered rock mass slopes, foundations, and underground caverns, and has important application value. Summary of the invention

[0005] The present invention is made to solve the above-mentioned problems, and its purpose is to provide a data-driven layered rock deformation analysis method, which aims to bypass the construction process of the traditional constitutive model and directly perform calculations based on large-scale experimental stress-strain data points, conservation laws and boundary conditions. It can greatly improve the ability of rock simulation prediction, and the calculation results are more consistent with the actual situation.

[0006] In order to achieve the above purpose, the present invention adopts the following scheme:

[0007] A data-driven layered rock deformation analysis method comprises the following steps:

[0008] Step 1: Design a uniaxial compression test of a grooved rock sample, collect the strain field and loading force information during the test, and establish an initial stress-strain database of layered rock based on a data recognition algorithm;

[0009] Step 2, converting the initial stress-strain database in step 1 to the target angle to obtain a layered rock conversion stress-strain database;

[0010] Step 3, constructing an indeterminate equation for the stress-strain data of each data point in the converted stress-strain database, and using the least squares solution as the equilibrium factor matrix;

[0011] Step 4: Use the local convex set data-driven calculation method that takes into account the influence of test data noise to calculate the stress field, strain field, and displacement field of layered rock at the target angle under loading and unloading.

[0012] Furthermore, the step of establishing an initial stress-strain database of layered rocks based on a data recognition algorithm in step 1 includes:

[0013] Step 1: Initialize mapping based on k-means algorithm;

[0014] Step 2: Obtain the strain field of each unit through DIC, and calculate the unit material strain based on the initial mapping result of Step 1;

[0015] Step 3: construct a large sparse matrix, and solve the large sparse matrix based on the preconditioned biconjugate gradient method to calculate the unit material stress and Lagrange multiplier;

[0016] Step 4: Based on the unit material stress and Lagrange multiplier, the stress field of each unit under X loading conditions is calculated;

[0017] Step 5: Arrange the disordered point cloud in order by constructing a kd tree to find the best mapping pointer for each unit under each loading condition and update it, and construct a global objective penalty function to keep the global objective penalty function as minimum;

[0018] Step 6: Repeat Steps 2-5 until the initialization mapping converges and remains unchanged.

[0019] Furthermore, the formula for calculating the unit material strain in Step 2 is as follows:

[0020]

[0021] in, is the unit material strain; ie X To provide the pointer of the initial cluster, e = 1, 2, 3...n is the number of the finite element numerical integration point; is the integral weight of the integral point e; is the strain field of the Xth loading condition at integration point e.

[0022] Furthermore, the large sparse matrix in Step 3 is:

[0023]

[0024] Where A is the diagonal block constant stiffness matrix, n is the total number of finite element numerical integration points, B e is the strain transformation matrix of the integration point e; C is the equilibrium factor matrix; For any Xth loading condition, S 1 , S 2 S X are the non-diagonal block matrices under the 1st, 2nd, and Xth loading conditions, respectively. is the integral weight of the integral point e; δ e is the signal function, N mat The number of material databases generated for the plan, ie X To provide pointers for the initial clustering, M e is the mapping state of the integration point e, λ 1 , 2 , X are the Lagrange multipliers under the 1st, 2nd and Xth loading conditions respectively; is the unit material stress; f 1 、f 2 f X They are the node external force load vector arrays under the 1st, 2nd and Xth loading conditions respectively.

[0025] The solution process using the biconjugate gradient method is: left-multiply and right-multiply the sparse matrix by a preprocessing matrix P to balance the numerical magnitude of each row and column.

[0026]

[0027] Where totalDof represents the total number of nodal degrees of freedom of the sampling model under various loading conditions, and S represents the independent stress component contained in each data point.

[0028] Furthermore, the stress field formula of the Xth loading condition at the integration point e calculated in Step 4 is as follows:

[0029]

[0030] in, is the stress field of the Xth loading condition at integration point e.

[0031] Furthermore, in Step 5, a kd tree is constructed to arrange the disordered point cloud in an orderly manner. The specific operations are as follows: first, the data of the mechanical state and the material state are homogenized; then, the material state is divided according to the dimensions in turn and the nodes are selected according to the median to establish a kd tree; finally, for each mechanical state, the material data point closest to it is queried in the kd tree and the mapping pointer is recorded;

[0032] The global objective penalty function is:

[0033]

[0034] Among them, G is the global objective penalty function; G X is the target penalty function under the Xth loading condition, is the strain field of the Xth loading condition at integration point e, is the unit material strain;

[0035] Furthermore, in step 2, the initial stress-strain database in step 1 is converted to the target angle based on the coordinate conversion formula to obtain the layered rock conversion stress-strain database. The coordinate conversion formula of the database is:

[0036]

[0037] in, represents the transformed rock strain; represents the transformed rock stress; θ is the angle between the global coordinate system and the local coordinate system; ε x is the strain in the x direction; ε y is the strain in the y direction; xy is the shear strain; x is the stress in the x direction; σ y is the stress in the y direction; τ xy is the shear stress.

[0038] Furthermore, in step 3, a hyperstatic equation is constructed by converting the stress-strain data of each data point in the stress-strain database, and the least squares solution is used as the equilibrium factor matrix:

[0039]

[0040] In the formula, idata is the number of data points, and represents the rock stress-strain data pair obtained from the experiment, C ijkl is the equilibrium constant matrix of the least squares solution, N is 1, 2, 3…idata is the data point number.

[0041] Further, step 4 includes: performing linear interpolation on several stress-strain material data points closest to the mechanical state point to construct a local convex set, and then projecting the mechanical state point onto the reconstructed related local convex manifold, and seeking the optimal material state search solution by non-negative least squares method;

[0042] For a given mechanical state point s z , first find the k material data points D(k,s z ), and use the following formula to calculate the weighting coefficient w i :

[0043]

[0044] use Represents the given mechanical state s z The corresponding optimal material data point is:

[0045]

[0046] in, Represents the distance from the mechanical state point s z The most recent i-th material data point, w i is the weighting coefficient of the i-th material data point.

[0047] In another aspect, the present invention provides a data-driven layered rock deformation analysis device, comprising:

[0048] A database establishment unit, which is used to collect the strain field and loading force information during the test and establish an initial stress-strain database of layered rocks based on a data recognition algorithm;

[0049] A coordinate conversion unit, which is used to convert the initial stress-strain database to a target angle to obtain a layered rock conversion stress-strain database;

[0050] An equation construction part, which is used to construct an indeterminate equation for the stress-strain data of each data point in the converted stress-strain database, and uses the least squares solution as a balance factor matrix;

[0051] The calculation part is used to calculate the stress field, strain field and displacement field of layered rock at the target angle under loading and unloading by adopting a local convex set data-driven calculation method that takes into account the influence of test data noise.

[0052] Compared with the prior art, the present invention has the following beneficial effects:

[0053] 1) The present invention can realize stress-strain analysis under external loading and unloading, as well as excavation and reinforcement design of layered rock slopes, foundations, and underground caverns, without defining the constitutive relationship of layered rock mass, which is a complex non-homogeneous, non-continuous, nonlinear, and anisotropic medium, and has important application value.

[0054] 2) The present invention proposes a method for generating a layered rock stress-strain database. Through full-field digital image measurement and external force load boundary conditions, the stress components of the mechanical state under each loading condition and the stress-strain database of layered rocks that are common under all loading conditions can be identified without using constitutive equations.

[0055] 3) The present invention uses the coordinate conversion formula of the database to broaden the application scope of the database. For the deformation problem of layered rocks at different angles, there is no need to conduct multiple tests. It is only necessary to convert the layered rock database at a certain angle to a suitable angle through the coordinate conversion formula.

[0056] 4) The present invention directly uses the stress-strain database to drive the simulation calculation, and uses the local convex set data-driven algorithm to reduce the impact of test noise. Compared with the traditional rock mass numerical simulation analysis, it overcomes the errors caused by empirical modeling, while retaining the originality of the test data, so it can better simulate the actual situation. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 A flow chart of a data-driven layered rock deformation analysis method according to the present invention;

[0058] Figure 2 This is a 0-degree layered rock size diagram involved in an embodiment of the present invention;

[0059] Figure 3 The stress scatter points and their projection diagrams generated for the training involved in the embodiment of the present invention;

[0060] Figure 4 A schematic diagram of the database coordinate conversion principle involved in an embodiment of the present invention;

[0061] Figure 5 This is a 30-degree layered rock size diagram involved in an embodiment of the present invention;

[0062] Figure 6 A simplified mechanics and finite element mesh schematic diagram of 30-degree layered rock involved in an embodiment of the present invention;

[0063] Figure 7 Comparison cloud diagrams and relative error diagrams of x-direction displacements of representative points involved in embodiments of the present invention: (a) is the x-direction displacement measured by DIC; (b) is the x-direction displacement calculated by LCDDCM; (c) is the relative error diagram in the x-direction;

[0064] Figure 8 Comparison cloud diagrams and relative error diagrams of the y-direction displacement of representative points involved in the embodiments of the present invention: (a) is the y-direction displacement measured by DIC; (b) is the y-direction displacement calculated by LCDDCM; (c) is the relative error diagram in the y-direction. DETAILED DESCRIPTION

[0065] In order to facilitate those skilled in the art to understand and implement the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and the present embodiment. It should be understood that the implementation examples described herein are only used to illustrate and explain the present invention and are not used to limit the present invention.

[0066] like Figure 1 As shown, the present invention provides a data-driven layered rock deformation analysis method, comprising the following steps:

[0067] Step 1, collect material stress-strain data under some loading conditions during the test, and establish an initial stress-strain database of layered rocks under all loading conditions based on the data recognition algorithm;

[0068] Among them, in step 1, by designing the experiment and collecting the loading force information and corresponding strain field information during the experiment, combined with the data identification algorithm (DDI) training, the stress field information under various loading conditions and the complete initial material stress-strain database common to all loading conditions are obtained. In the past, when obtaining the stress field, most of them relied on the assumption of uniform stress field, or in the case of non-uniform stress field, they relied on the inverse parameter identification method of the constitutive model. This method can directly obtain the stress-strain database without relying on the constitutive equation.

[0069] In step 1, the initial stress-strain database of layered rocks under all loading conditions is established based on the data recognition algorithm, including:

[0070] Step 1: Initialize mapping based on k-means algorithm;

[0071] Step 2: Calculate the unit material strain based on the initial mapping results of Step 1;

[0072]

[0073] in, is the unit material strain; ie X To provide the pointer of the initial cluster, e = 1, 2, 3...n is the number of the finite element numerical integration point; is the integral weight of the integral point e; is the strain field of the Xth loading condition at integration point e.

[0074] Step 3: construct a large sparse matrix, and solve the large sparse matrix based on the preconditioned biconjugate gradient method to calculate the unit material stress and Lagrange multiplier;

[0075] The large sparse matrix is:

[0076]

[0077] A is the diagonal block constant stiffness matrix, n is the total number of finite element numerical integration points, B e is the strain transformation matrix of the integration point e; C is the equilibrium factor matrix; For any Xth loading condition, S1 , S 2 S X are the non-diagonal block matrices under the 1st, 2nd, and Xth loading conditions, respectively. is the integral weight of the integral point e; δ e is the signal function, N mat The number of material databases generated for the plan, ie X To provide pointers for the initial clustering, M e is the mapping state of the integration point e, λ 1 , 2 , X are the Lagrange multipliers under the 1st, 2nd and Xth loading conditions respectively; is the unit material stress; f 1 、f 2 f X They are the node external force load vector arrays under the 1st, 2nd and Xth loading conditions respectively.

[0078] Step 4: Calculate the stress field of X loading conditions based on the unit material stress and Lagrange multiplier;

[0079]

[0080] in, is the stress field of the Xth loading condition at integration point e.

[0081] Step 5: Arrange the disordered point cloud in order by constructing a kd tree to find the best mapping pointer for each unit under each loading condition and update it, and construct a global objective penalty function to keep the global objective penalty function as minimum;

[0082]

[0083] Among them, G is the global objective penalty function; G X is the target penalty function under the Xth loading condition, is the strain field of the Xth loading condition at integration point e, is the unit material strain;

[0084] Step 6: Repeat Steps 2-5 until the initialization mapping converges and remains unchanged.

[0085] In Step 1, initialize the mapping ie X When the k-means algorithm is used to replace the original random initial mapping, the convergence efficiency of the data recognition algorithm is significantly improved. Specifically, the k-means algorithm is applied to the strain field In the example, according to the Euclidean distance squared metric, The strains are divided into several clusters with the smallest mutual difference. In order to improve the distribution effect, secondary distribution is used and the amount of data in each cluster is ensured to be roughly the same.

[0086] In Step 3, when solving large sparse matrices, the preconditioned biconjugate gradient method is used instead of the original left division method to achieve better convergence results. The magnitude differences between the elements of a large sparse matrix are large, which will cause the matrix to be severely ill-conditioned and close to singular. An unreasonable solution will lead to large calculation errors and be time-consuming. The specific operation is: multiply the sparse matrix by a preconditioned matrix P on the left and right to balance the numerical magnitude of each row and column; then use the biconjugate gradient method to solve.

[0087]

[0088] Where totalDof represents the total number of nodal degrees of freedom of the sampling model under various loading conditions, S represents the independent stress component contained in each data point (1 in one-dimensional problems, 3 in two-dimensional problems, and 6 in three-dimensional problems); N mat The number of material databases generated for the plan.

[0089] In Step 5, it is necessary to find an optimal mapping pointer for each unit in each loading situation. X This step is very expensive. By constructing a kd tree to arrange the disordered point cloud in an orderly manner, a faster and more efficient retrieval operation can be achieved compared to the conventional traversal method. The specific operation is: first, the data of the mechanical state and material state are homogenized, specifically Then the material state is divided according to the dimension in turn (the axis is selected as the dividing surface in turn with the depth of the tree) and the kd tree is established by selecting the node according to the median; finally, for each mechanical state, the material data point closest to it is queried in the kd tree and the ie X .

[0090] Step 2, converting the initial stress-strain database in step 1 to the target angle to obtain a layered rock conversion stress-strain database;

[0091] In step 2, for the layered rock deformation problem at various angles, there is no need to conduct separate experiments to obtain the stress-strain database at the corresponding angles. It only needs to be converted through the coordinate conversion formula of the database. The coordinate conversion formula of the database is:

[0092]

[0093] in, represents the transformed rock strain; represents the transformed rock stress; θ is the angle between the global coordinate system and the local coordinate system (clockwise is positive); ε x is the strain in the x direction; ε y is the strain in the y direction; xy is the shear strain; x is the stress in the x direction; σ y is the stress in the y direction; τ xy is the shear stress.

[0094] Step 3, constructing an indeterminate equation for the stress-strain data of each data point in the converted stress-strain database, and using the least squares solution as the equilibrium factor matrix;

[0095] In step 3, the hyperstatic equation is constructed by the stress-strain data of each data point in the database, and the least squares solution is used as the equilibrium factor matrix:

[0096]

[0097] In the formula, idata is the number of data points, and represents the rock stress-strain data pair obtained from the experiment, C ijkl is the equilibrium constant matrix of the least squares solution, N is 1, 2, 3…idata is the data point number.

[0098] Step 4: Use the local convex set data-driven calculation method that takes into account the influence of test data noise to calculate the stress field, strain field, and displacement field of layered rock at the target angle under loading and unloading.

[0099] In step 4, in order to reduce the impact of data noise caused by material samples, measurement methods, test environment or human factors during the test, a local convex set data-driven calculation method with k=5 is used to replace the traditional distance minimization data-driven calculation method, which can significantly reduce the impact of data noise and enhance the robustness and convergence stability of data-driven calculation.

[0100] The distance minimization data-driven calculation method uses spatial distance as the unit of measurement, and uses the information of a single data point for each unit to extract material properties. When there are data noise points in the material database, it is very easy to amplify the data noise caused by a single data point, thereby affecting the calculation accuracy. To alleviate this effect, a local convex set data-driven calculation method with k=5 is used. It first performs linear interpolation on several stress-strain material data points closest to the mechanical state point to construct a local convex set, and then projects the mechanical state point onto the reconstructed related local convex manifold, and uses the non-negative least squares method to find the optimal material state search solution.

[0101] For a given mechanical state point s z, first find the k material data points D(k,s z ), and use the following formula to calculate the weighting coefficient w i :

[0102]

[0103] use Represents the given mechanical state s z The corresponding optimal material data point is:

[0104]

[0105] in, Represents the distance from the mechanical state point s z The most recent i-th material data point, w i is the weighting coefficient of the i-th material data point.

[0106] LCDDCM can significantly reduce the impact of data noise by using the idea of ​​local convex construction, and enhance the robustness and convergence stability in data-driven calculations. In order to ensure the effectiveness of linear approximation, in principle, the number of closest points k selected should be greater than the expected dimension d of the underlying manifold (k = 5). It is worth noting that the values ​​of stress and strain in the database vary greatly. In order to ensure the correctness of the solution, it is necessary to homogenize the stress and strain before solving. Specifically,

[0107] Example 1

[0108] In this embodiment, the data-driven layered rock deformation analysis method provided by the present invention is used for calculation, and the specific steps are as follows:

[0109] Step 1: Cut a 0-degree layered rock sample with a length, width and thickness of 100×100×10 mm from the parent rock sample, and cut a groove in the rock sample to increase the inhomogeneity of the displacement field, such as Figure 2 As shown. The speckle field was prepared on the surface of the rock sample, and the rock sample was subjected to uniaxial compression using the rock mass test (RMT) system. During the test, the pressure sensor provided by the RMT was used to collect the loading force information, and the commercial digital image correlation (DIC) system was used to measure the strain field on the surface of the rock sample. The collected loading force information and the corresponding strain field information can be trained into the stress-strain database of the rock sample material in combination with the data identification algorithm (DDI). A total of N X (N X = 100) times of loading. The number of triangular meshes N into which the DIC surface analysis area is divided elem is 19408, and the total number of mechanical states calculated is N mecha=N elem ×N X = 1940800. To ensure the good performance of the DDI algorithm, the target generation rate selected for this training is 1%, and the number of material stress-strain databases obtained by the final training is N. * =N mecha / 100=19408. Figure 3 The trained stress scatter plot and its projections in various directions are shown.

[0110] Step 2: Use the database coordinate conversion formula to convert it to the database of layered rocks at the target angle (30 degrees). The principle of database coordinate conversion is as follows: Figure 4 As shown (taking stress component as an example). The coordinate transformation formula of the database is:

[0111]

[0112] in, represents the transformed rock strain; represents the transformed rock stress; θ is the angle between the global coordinate system and the local coordinate system (clockwise is positive); ε x is the strain in the x direction; ε y is the strain in the y direction; xy is the shear strain; x is the stress in the x direction; σ y is the stress in the y direction; τ xy is the shear stress.

[0113] Step 3, construct a hyperstatic equation by using the stress-strain data of each data point in the 30-degree database, and use the least squares solution as the equilibrium factor matrix:

[0114]

[0115] In the formula, idata is the number of data points, and represents the rock stress-strain data pair obtained from the experiment, C ijkl is the equilibrium constant matrix for the least squares solution.

[0116] Step 4: Cut and process a 30-degree layered rock sample with a length, width and thickness of 100×100×10 mm from the same parent rock sample, such as Figure 5 As shown in Figure 1. The speckle field is prepared on the surface of the rock sample, and the rock sample is subjected to uniaxial compression using a rock mass test (RMT) system. During the test, the displacement field on the surface of the rock sample is measured using a commercial digital image correlation (DIC) system. Figure 6The numerical calculation model of layered rock shown in the figure contains 4400 triangular units. The rock sample loading problem is solved using the 30-degree layered rock database generated above and the local convex set data-driven calculation method (LCDDCM) k=5 considering the influence of test data noise, and the data-driven results are compared with the test results obtained by DIC to verify the accuracy and reliability of the method.

[0117] Figure 7 The x-direction displacement comparison cloud map and relative error map of representative points are shown. Figure 8 The y-direction displacement comparison cloud map and relative error map of representative points are shown. By comparison, the displacement fields measured by LCDDCM and DIC are smooth and change similarly. At the same time, the displacement results obtained using the LCDDCM method are more obvious and symmetrical than the DIC results, because the data-driven results also meet the equilibrium and compatibility constraints, which can better represent the stress-strain relationship of layered rocks under loading and unloading.

[0118] Example 2

[0119] This embodiment provides a data-driven layered rock deformation analysis device, comprising:

[0120] A database establishment unit, which is used to collect material stress-strain data under some loading conditions during the test process, and establish an initial stress-strain database of layered rocks under all loading conditions based on a data recognition algorithm;

[0121] A coordinate conversion unit, which is used to convert the initial stress-strain database to a target angle to obtain a layered rock conversion stress-strain database;

[0122] An equation construction part, which is used to construct an indeterminate equation for the stress-strain data of each data point in the converted stress-strain database, and uses the least squares solution as a balance factor matrix;

[0123] The calculation part is used to calculate the stress field, strain field and displacement field of layered rock at the target angle under loading and unloading by adopting a local convex set data-driven calculation method that takes into account the influence of test data noise.

[0124] The above is only a preferred specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by any technician familiar with the technical field within the technical scope disclosed in the present application should be covered within the protection scope of the present application.

[0125] It should be understood that parts not elaborated in detail in this specification belong to the prior art.

[0126] It should be understood that the above description of the preferred embodiment is relatively detailed and cannot be regarded as limiting the scope of patent protection of the present invention. Under the enlightenment of the present invention, ordinary technicians in this field can also make substitutions or modifications without departing from the scope of protection of the claims of the present invention, which all fall within the scope of protection of the present invention. The scope of protection requested for the present invention shall be based on the attached claims.

Claims

1. A data-driven layered rock deformation analysis method, characterized in that: The steps include: Step 1: Design a uniaxial compression test of a grooved rock sample, collect the strain field and loading force information during the test, and establish an initial stress-strain database of layered rock based on a data recognition algorithm; Step 2, converting the initial stress-strain database in step 1 to the target angle to obtain a layered rock conversion stress-strain database; Step 3, constructing an indeterminate equation for the stress-strain data of each data point in the converted stress-strain database, and using the least squares solution as the equilibrium factor matrix; Step 4: Use the local convex set data-driven calculation method that takes into account the influence of test data noise to calculate the stress field, strain field, and displacement field of layered rock at the target angle under loading and unloading.

2. The data-driven layered rock deformation analysis method according to claim 1, characterized in that: The step of establishing an initial stress-strain database of layered rocks based on a data recognition algorithm in step 1 includes: Step 1: Initialize mapping based on k-means algorithm; Step 2: Obtain the strain field of each unit through DIC, and calculate the unit material strain based on the initial mapping result of Step 1; Step 3: construct a large sparse matrix, and solve the large sparse matrix based on the preconditioned biconjugate gradient method to calculate the unit material stress and Lagrange multiplier; Step 4: Based on the unit material stress and Lagrange multiplier, the stress field of each unit under X loading conditions is calculated; Step 5: Arrange the disordered point cloud in order by constructing a kd tree to find the best mapping pointer for each unit under each loading condition and update it, and construct a global objective penalty function to keep the global objective penalty function as minimum; Step 6: Repeat Steps 2-5 until the initialization mapping converges and remains unchanged.

3. A data-driven layered rock deformation analysis method according to claim 2, characterized in that: The formula for calculating the unit material strain in Step 2 is as follows: in, is the unit material strain; ie X To provide the pointer of the initial cluster, e = 1, 2, 3...n is the number of the finite element numerical integration point; is the integral weight of the integral point e; is the strain field of the Xth loading condition at integration point e.

4. The data-driven layered rock deformation analysis method according to claim 3 is characterized by: The large sparse matrix in Step 3 is: ie X Where A is the diagonal block constant stiffness matrix, n is the total number of finite element numerical integration points, B e is the strain transformation matrix of the integration point e; C is the equilibrium factor matrix; For any Xth loading condition, S 1 , S 2 S X are the non-diagonal block matrices under the 1st, 2nd, and Xth loading conditions, respectively. is the integral weight of the integral point e; δ e is the signal function, N mat The number of material databases generated for the plan, ie X To provide pointers for the initial clustering, M e is the mapping state of the integration point e, λ 1 , 2 , X are the Lagrange multipliers under the 1st, 2nd and Xth loading conditions respectively; is the unit material stress; f 1 、f 2 f X are the node external force load vector arrays under the 1st, 2nd and Xth loading conditions respectively; The solution process using the biconjugate gradient method is: left-multiply and right-multiply the sparse matrix by a preprocessing matrix P to balance the numerical magnitude of each row and column. Where totalDof represents the total number of nodal degrees of freedom of the sampling model under various loading conditions, and S represents the independent stress component contained in each data point.

5. The data-driven layered rock deformation analysis method according to claim 4 is characterized by: The stress field formula of the Xth loading condition of the integration point e calculated in Step 4 is as follows: in, is the stress field of the Xth loading condition at integration point e.

6. The data-driven layered rock deformation analysis method according to claim 5 is characterized by: In Step 5, a kd tree is constructed to arrange the disordered point cloud in an orderly manner. The specific operations are as follows: first, the data of the mechanical state and the material state are homogenized; then the material state is divided according to the dimensions in turn and the nodes are selected according to the median to establish a kd tree; finally, for each mechanical state, the material data point closest to it is queried in the kd tree and the mapping pointer is recorded; The global objective penalty function is: Among them, G is the global objective penalty function; G X is the target penalty function under the Xth loading condition, is the strain field of the Xth loading condition at integration point e, is the element material strain.

7. The data-driven layered rock deformation analysis method according to claim 1, characterized in that: In step 2, the initial stress-strain database in step 1 is converted to the target angle based on the coordinate conversion formula to obtain the layered rock conversion stress-strain database. The coordinate conversion formula of the database is: in, represents the transformed rock strain; represents the transformed rock stress; θ is the angle between the global coordinate system and the local coordinate system; ε x is the strain in the x direction; ε y is the strain in the y direction; xy is the shear strain; x is the stress in the x direction; σ y is the stress in the y direction; τ xy is the shear stress.

8. The data-driven layered rock deformation analysis method according to claim 1, characterized in that: In step 3, the hyperstatic equation is constructed by converting the stress-strain data of each data point in the stress-strain database, and the least squares solution is used as the equilibrium factor matrix: In the formula, idata is the number of data points, and represents the rock stress-strain data pair obtained from the experiment, C ijkl is the equilibrium constant matrix of the least squares solution, N is 1, 2, 3…idata is the data point number.

9. The data-driven layered rock deformation analysis method according to claim 1, characterized in that: Step 4 includes: Linear interpolation is performed on several stress-strain material data points closest to the mechanical state point to construct a local convex set, and then the mechanical state point is projected onto the reconstructed related local convex manifold, and the optimal material state search solution is sought by the non-negative least squares method; For a given mechanical state point s z , first find the k material data points D(k,s z ), and use the following formula to calculate the weighting coefficient w i : use Represents the given mechanical state s z The corresponding optimal material data point is: in, Represents the distance from the mechanical state point s z The most recent i-th material data point, w i is the weighting coefficient of the i-th material data point.

10. A data-driven layered rock deformation analysis device, characterized in that: include: A database establishment unit, which is used to collect the strain field and loading force information during the test and establish an initial stress-strain database of layered rocks based on a data recognition algorithm; A coordinate conversion unit, which is used to convert the initial stress-strain database to a target angle to obtain a layered rock conversion stress-strain database; An equation construction part, which is used to construct an indeterminate equation for the stress-strain data of each data point in the converted stress-strain database, and uses the least squares solution as a balance factor matrix; A calculation part, which is used to calculate the stress field, strain field and displacement field of layered rock at a target angle under loading and unloading by adopting a local convex set data-driven calculation method that takes into account the influence of test data noise; The data-driven layered rock deformation analysis device is used to execute the steps in the method described in any one of claims 1-9.

Images