Rock creep process simulation method and system based on deep learning near-field dynamics

Abstract

This invention provides a method and system for simulating rock creep processes based on deep learning-based near-field dynamics, involving the interdisciplinary fields of rock mechanics, computational mechanics, and artificial intelligence. The method includes: acquiring parameter information of the target rock to be simulated, including the rock's geometry and physical properties; discretizing the target rock into several material points based on the parameter information to construct a creep damage model; performing time-step iterative simulation of the rock creep process based on the initialized creep damage model, ultimately obtaining the displacement field, stress field, and creep strain distribution of the target rock over time. This invention significantly improves computational efficiency by establishing a neural network mapping from bond geometric variables to force density, replacing the complex iterative calculations in traditional near-field dynamics. Furthermore, by combining the creep damage model with damage evolution criteria, it achieves high-precision simulation of the entire creep process of fractured rock masses.

Description

Technical Field

[0001] This invention relates to the interdisciplinary fields of rock mechanics, computational mechanics, and artificial intelligence, specifically to a method and system for simulating rock creep processes based on deep learning-based near-field dynamics. Background Technology

[0002] In geotechnical engineering, hydraulic engineering, and underground resource extraction, the creep behavior of rock under long-term constant geostress is a key factor leading to slope instability, large tunnel deformation, and goaf collapse. Rock mass, as a typical heterogeneous and discontinuous medium, is rich in joints, fissures, and other defects that gradually accumulate damage under long-term loading, resulting in deterioration of its mechanical properties and exhibiting significant nonlinear creep characteristics. Therefore, accurately simulating the entire process of creep deformation and damage evolution of fractured rock mass under long-term loading has significant theoretical and engineering value for the long-term stability assessment of geotechnical engineering.

[0003] Currently, numerical simulations of rock mass creep behavior mainly employ traditional methods such as the finite element method or the discrete element method. However, traditional continuum mechanics methods struggle to naturally describe the initiation, propagation, and penetration of microcracks during creep when dealing with fractured rock masses. While the peridynamics (PD) method can effectively circumvent crack tip singularities through nonlocal integral equations and is naturally suitable for simulating the fracture and damage processes of rock masses, it often requires complex iterative calculations of the interaction forces between each pair of material points when solving the creep-damage coupled constitutive relationship. This results in extremely low computational efficiency and severely restricts its widespread application in long-term creep simulations of complex rock mass engineering.

[0004] Therefore, existing rock creep simulation methods are inadequate in terms of computational efficiency, damage evolution description, and handling of complex fractures, which limits their practical application in engineering. Summary of the Invention

[0005] To address the aforementioned problems, this invention proposes a method and system for simulating rock creep processes based on deep learning-based peri-field dynamics. By establishing a neural network mapping from bond geometric variables to force density, it replaces the complex iterative calculations in traditional peri-field dynamics, significantly improving computational efficiency. Furthermore, by combining the Lemaitre creep damage model and damage evolution criteria, it achieves high-precision simulation of the entire creep process of fractured rock masses.

[0006] According to some embodiments, the present invention adopts the following technical solution:

[0007] Methods for simulating rock creep processes based on deep learning-based near-field dynamics include:

[0008] Obtain parameter information of the target rock to be simulated, including the rock's geometry and the physical properties of the rock mass;

[0009] Based on the parameter information, the target rock is discretized into several material points to construct a creep damage model of the target rock;

[0010] Based on the initialized creep damage model, time-step iterative simulation of rock creep process is performed to obtain the displacement field, stress field and creep strain distribution of target rock as it evolves over time.

[0011] In each iteration time step, firstly, based on the physical field information of the current time step, the force density components between each material point pair are predicted using a neural network model to obtain the force density distribution of the current time step; then, based on the force density distribution, the physical field information is updated using unconventional near-field dynamics theory; finally, based on the updated physical field information, it is determined whether the connection bond between the material point pairs has broken, the damage parameters are updated based on the determination result, and the iteration proceeds to the next time step until the termination condition is triggered.

[0012] According to some embodiments, the present invention adopts the following technical solution:

[0013] A rock creep simulation system based on deep learning-based near-field dynamics includes:

[0014] The information acquisition module is configured to acquire parameter information of the target rock to be simulated, including the rock's geometry and the physical properties of the rock mass.

[0015] The model building module is configured to: discretize the target rock into several material points based on the parameter information, and build a creep damage model of the target rock;

[0016] The iterative simulation module is configured to perform time-step iterative simulation of the rock creep process based on the initialized creep damage model, and finally obtain the displacement field, stress field and creep strain distribution of the target rock over time.

[0017] In each iteration time step, firstly, based on the physical field information of the current time step, the force density components between each material point pair are predicted using a neural network model to obtain the force density distribution of the current time step; then, based on the force density distribution, the physical field information is updated using unconventional near-field dynamics theory; finally, based on the updated physical field information, it is determined whether the connection bond between the material point pairs has broken, the damage parameters are updated based on the determination result, and the iteration proceeds to the next time step until the termination condition is triggered.

[0018] According to some embodiments, the present invention adopts the following technical solution:

[0019] A computer program product includes a computer program that, when executed by a processor, implements the rock creep process simulation method based on deep learning peri-field dynamics.

[0020] According to some embodiments, the present invention adopts the following technical solution:

[0021] A non-transitory computer-readable storage medium is provided for storing computer instructions, which, when executed by a processor, implement the rock creep process simulation method based on deep learning peri-field dynamics.

[0022] According to some embodiments, the present invention adopts the following technical solution:

[0023] An electronic device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the rock creep process simulation method based on deep learning peri-field dynamics.

[0024] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0025] This invention significantly improves the computational efficiency of rock creep simulation and reduces the computational cost of long-scale simulations by directly predicting the force density components between material pairs using a neural network model at each time step, replacing the complex iterative calculation process in traditional peri-field dynamics. Simultaneously, based on the predicted force density distribution, unconventional peri-field dynamics theory is used to update the displacement field, stress field, and creep strain distribution. At each time step, the updated physical field information is used to determine whether the bonds between material pairs have broken, thereby dynamically updating the damage parameters. This allows for more precise tracking of damage accumulation and failure behavior during creep, improving the physical reliability of the simulation results. Furthermore, because this method uses discrete material point descriptions within a peri-field dynamics framework, it is naturally applicable to target rocks with complex geometries and non-uniform rock mass properties, avoiding the numerical difficulties of traditional continuous medium methods in handling fracture discontinuities. This achieves efficient and accurate simulation of the entire rock creep process. Attached Figure Description

[0026] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0027] Figure 1 This is a flowchart of the method in Example 1.

[0028] Figure 2This is a flowchart of the method in Example 3. Detailed Implementation

[0029] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0030] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0031] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0032] Example 1

[0033] One embodiment of the present invention provides a method for simulating rock creep processes based on deep learning-based near-field dynamics, such as... Figure 1 As shown, it includes:

[0034] Step 1: Obtain parameter information of the target rock to be simulated, including the rock's geometry and physical properties;

[0035] Step 2: Based on the parameter information, the target rock is discretized into several material points to construct a creep damage model of the target rock;

[0036] Step 3: Based on the initialized creep damage model, perform time-step iterative simulation of the rock creep process to finally obtain the displacement field, stress field and creep strain distribution of the target rock over time.

[0037] In each iteration time step, firstly, based on the physical field information of the current time step, the force density components between each material point pair are predicted using a neural network model to obtain the force density distribution of the current time step; then, based on the force density distribution, the physical field information is updated using unconventional near-field dynamics theory; finally, based on the updated physical field information, it is determined whether the connection bond between the material point pairs has broken, the damage parameters are updated based on the determination result, and the iteration proceeds to the next time step until the termination condition is triggered.

[0038] As one embodiment, the rock creep simulation method based on deep learning peri-field dynamics of the present invention significantly improves computational efficiency by establishing a neural network mapping from bond geometry variables to force density, replacing the complex iterative calculations in traditional peri-field dynamics. Furthermore, by combining the Lemaitre creep damage model and damage evolution criteria, it achieves high-precision simulation of the entire creep process of fractured rock masses. The specific implementation process is described below:

[0039] I. Unconventional Perifield Dynamics Theory

[0040] By introducing unconventional peridynamics theory and continuous medium theory, we construct unconventional peridynamic equations and force density expressions.

[0041] Specifically, in the theory of unconventional peri-field dynamics, any point of matter in space... In a radius of δ nonlocal scope Inside, the unconventional peri-field dynamics equations are:

[0042] (1)

[0043] In the formula, For rock mass density, For matter points exist t acceleration at any moment for Other interacting matter points in the neighborhood exist The force density applied on, Conversely. represent nonlocal scope, Let V be the volume of a point within the domain. for Physical density at the location.

[0044] Material point pairs in the initial coordinate system and relative position vector The relative displacement vector when deformation occurs According to the definition of near-field dynamic state, the reference position state Defined as:

[0045] (2)

[0046] Transformation Defined as:

[0047] (3)

[0048] Then matter point The nonlocal deformation gradient F at point is:

[0049] (4)

[0050] (5)

[0051] In the formula For vectors The determined scalar weight function, For tensor product, It is a positive definite symmetric shape matrix. Its inverse matrix. Equation (1) can be used to obtain the force density by being equivalent to the strain energy density in continuous medium theory. The expression:

[0052] (6)

[0053] In the formula, P is the first Piola–Kirchhoff stress tensor, which, under the assumption of small deformation and small rotation, is related to the Cauchy stress tensor. equal:

[0054] (7)

[0055] In the formula, C is the elastic tensor, and is the matrix double inner product. Therefore, equation (6) can also be expressed as:

[0056] (8)

[0057] II. Peri-field dynamics theory of creep in fractured rock masses

[0058] Based on equation (7), considering the Lemaitre rock creep damage model, the Cauchy strain tensor is... Decomposed into the sum of elastic strain and creep strain:

[0059] (9)

[0060] In the formula, For elastic strain tensor, For the creep strain tensor, the parameters are used. D This describes the degree of damage at a point in the fractured rock mass, with a value ranging from 0 to 1, corresponding to no damage and complete failure, respectively.

[0061] When a material point is damaged by the external environment, the Lemaitre equivalent stress can be further expressed as:

[0062] (10)

[0063] Furthermore, the Lemaitre equivalent Mises stress With deviatoric stress They are respectively:

[0064] (11)

[0065] (12)

[0066] Under multiple stress conditions in fractured rock masses, the creep rate is: for:

[0067] (13)

[0068] In the formula, t For time, d For differentiation operators, A The creep correlation coefficient is... n Stress index This is the reference stress.

[0069] The constitutive equation for the rock damage stage is:

[0070] (14)

[0071] Small deformation and strain under quasi-static conditions Increment It can be broken down into:

[0072] (15)

[0073] In the formula and The elastic and creep strain increments are respectively.

[0074] Time increment The creep strain increments before and after the change are:

[0075] (16)

[0076] The nonlocal deformation gradient before and after the time increment, and the increment of the strain tensor. The relationship between them is:

[0077] (17)

[0078] Then consider the stress increment during the damage parameter stage. for:

[0079] (18)

[0080] in The stress before and after the time increment is the time increment. , The relationship between them is:

[0081] (19)

[0082] Substituting into equation (8), we obtain the force density of the creep damage model of the fractured rock mass in the next time step:

[0083] (20)

[0084] Once the force density is determined, the acceleration and displacement can be calculated according to equation (1).

[0085] III. Perifield Dynamics Theory Based on Deep Learning

[0086] Based on deep learning-based perifield dynamics theory, this method improves upon the low efficiency of iterative calculations using rock creep damage model variables in unconventional perifield dynamics theories by establishing a neural network mapping from bond geometry variables to force density. Taking a two-dimensional problem as an example, material points... and The deformation gradient tensor F between the two objects under the action of external forces can be expressed as:

[0087] (twenty one)

[0088] In the formula, and It represents the principal deformation elongation in the plane, and its direction is consistent with the coordinate axis. There is no shear deformation, and there is only tension (greater than 1), compression (less than 1), or no deformation (equal to 1) along two orthogonal directions.

[0089] Furthermore, it is assumed that the deformation gradient tensor is determined solely by the material points. and The relative position vectors between The contribution, given the deformation gradient tensor, can be obtained through geometric relations. Elongation of the deformed key Elongation of bond and the corner of the key The elongation of the bond is defined as:

[0090] (twenty two)

[0091] The relationship with the deformation gradient tensor is as follows: Then the key's angle for:

[0092] (twenty three)

[0093] Furthermore, matter points and The key between Corresponding force density vector Equation (20) decomposes the force density vector into normal components. and tangential components Both are along the deformation after and The direction of the bond and its perpendicular tangential direction, The positive direction is defined as starting from the material point. to the point of matter The counterclockwise direction is defined as Positive direction.

[0094] Furthermore, a fully connected neural network is used to construct the mapping relationship. Its architecture includes an input layer, hidden layers, and an output layer. The input features of the input layer need to fully describe the deformation information of the rock material points and ensure a match between the input and output features. Therefore, the input features of the input layer include, but are not limited to, the bond length before deformation. Elongation of bond Key corner Creep correlation coefficient A, creep strain determinant Deformation gradient determinant The output of the output layer includes the normal component of the force density vector. With tangential component .

[0095] Furthermore, in order to capture the nonlinear relationship between input and output, multiple hidden layers are set up, and appropriate methods are selected for hyperparameter tuning to determine the appropriate network depth, including but not limited to grid search, random search, Bayesian optimization, etc.

[0096] Furthermore, all hidden layers use non-linear activation functions, including but not limited to ReLU, SELU, Sigmoid, and Tanh. These activation functions are all capable of handling model training tasks. Among them, the ReLU activation function is relatively simple to compute and can accelerate training.

[0097] (twenty four)

[0098] When input d When the value is greater than 0, the output is d When input d When the value is less than or equal to 0, the output is 0. In addition, ReLU does not have the gradient vanishing problem and can be applied to both shallow and deep networks.

[0099] Furthermore, during the forward propagation process of model training, the input data... The predicted value is obtained through calculation using the hidden layer. If the number of hidden layers is L, then the calculation for each layer is expressed as:

[0100] (25)

[0101] in l As a hidden layer, The output of each hidden layer is used as the input of the next layer. The activation function is ReLU for the hidden layer and linear activation for the output layer. and For the first l The weight matrix and bias vector in the layer.

[0102] Optimal parameters and The loss function is determined by minimizing the residual between the predicted and actual values ​​or by achieving the target accuracy, using mean squared error (MSE) as the loss function. J MSE :

[0103] (26)

[0104] in and The first i The true and predicted normal components of the dataset. and The first i The actual and predicted tangential components of the dataset. N The number of samples in the training set. Backpropagation calculates the gradient of the loss function with respect to each parameter and updates the parameters using gradient descent. The gradients of the output layer and the hidden layer are respectively:

[0105] (27)

[0106] (28)

[0107] in For the first l Linear combination of layers, And represents the gradient term, is the derivative of the activation function, and ⊙ denotes element-wise multiplication.

[0108] The weight matrix and bias vector are as follows:

[0109] (29)

[0110] To avoid constructing a complete derivative expression, Automatic Differentiation is used to solve the gradient calculation.

[0111] Furthermore, a dataset was generated based on the typical deformation states of rock creep under plane stress conditions. 70% of the dataset was used for training, 15% for validation, and 15% for prediction.

[0112] IV. Criteria for Determining Rock Mass Failure;

[0113] Furthermore, in the criteria for judging rock mass failure, material points... Damage degree parameters at the site D It can be defined as:

[0114] (30)

[0115] In the formula For matter points In time t Damage parameters at time, For matter points and Functions that define the broken bond states between connections. The bonding bonds remain intact, and interactions exist. When the bonding bonds break, the interaction disappears.

[0116] Furthermore, bond fracture is determined using a combination of the maximum tensile strength and the Mohr-Coulomb criterion.

[0117] (Tensile failure) (31)

[0118] (Shear failure) (32)

[0119] In the formula, and They are matter points and The average minimum principal stress and average maximum principal stress between these values ​​can be obtained from the equivalence relations in mechanics of materials. For the tensile strength of the rock, c For cohesion, The internal friction angle is denoted as , and fracture occurs when any two material points satisfy one of the terms in equations (31) and (32) during the calculation process.

[0120] The rock creep process simulation method proposed in this embodiment has the following advantages compared with the existing technology:

[0121] (1) Significantly improved computational efficiency: By constructing a direct mapping from deformation state to force density through deep learning, the complex iterative calculation process in traditional near-field dynamics is avoided, which greatly improves the computational speed of rock creep simulation.

[0122] (2) More accurate damage evolution simulation: Combining the Lemaitre creep damage model with the fracture determination method based on the maximum tensile strength and the Mohr-Coulomb criterion, it can more realistically reflect the damage accumulation and failure behavior of fractured rock mass during the creep process.

[0123] (3) Applicable to complex fracture networks: This method is based on the near-field dynamics theory of nonlocal interaction, and is naturally applicable to the creep simulation of multi-fracture and non-uniform rock masses, overcoming the limitations of traditional continuous medium methods in fracture treatment.

[0124] Example 2

[0125] One embodiment of the present invention provides a specific process for constructing and training a neural network model in a rock creep simulation method based on deep learning near-field dynamics, using a two-dimensional homogeneous granite slab as the object, including:

[0126] 1. Overview of the example.

[0127] This embodiment uses a two-dimensional homogeneous granite slab with geometric dimensions of 0.4 m × 0.2 m to simulate the long-term creep behavior of a complete rock mass under uniaxial compressive load. The model's bottom is completely constrained in the y-direction displacement, while a constant compressive stress is applied to the top. The model aims to achieve a high-precision mapping from bond geometric variables and creep state to force density components by training a deep neural network, thereby replacing traditional near-field dynamic iterative calculations in subsequent creep damage simulations, significantly improving computational efficiency while ensuring accuracy.

[0128] 2. Data preparation.

[0129] To construct a sufficient and physically reliable training dataset, samples are first generated using a traditional near-field dynamics method combined with the Lemaitre creep damage model. The specific steps are as follows:

[0130] (1) Set multiple loading conditions (compressive stress at 5 levels) and material creep parameters (creep coefficient A and stress index at 4 levels each).

[0131] (2) Transient creep calculations were performed for each parameter combination; the model was uniformly discretized with a particle spacing of 2 mm and a neighborhood radius of 6 mm. After the calculations were completed, the input features of each pair of interacting particles at different time steps (including initial bond length, bond elongation, bond rotation angle, creep correlation coefficient A, current creep strain determinant, and deformation gradient determinant) and the corresponding true force density normal and tangential components were extracted.

[0132] (3) Divide all simulation results into training set, validation set and test set in a ratio of 7:1.5:1.5 to ensure statistical consistency of data distribution.

[0133] 3. Neural network structure.

[0134] The network employs a fully connected feedforward architecture, with the following configuration: the input layer has 6 neurons, corresponding to 6 input features; hidden layer 1 has 128 neurons using ReLU activation; hidden layer 2 has 256 neurons using ReLU activation; hidden layer 3 has 128 neurons using ReLU activation; and the output layer has 2 neurons (one for the normal component and one for the tangential component of the stress density), using linear activation. All weights are initialized using a He normal distribution, and the bias is initialized to 0. To prevent overfitting, a Dropout layer is added after hidden layer 2, with a dropout probability set to 0.1.

[0135] 4. Training configuration.

[0136] The loss function used was mean squared error (MSE). The Adam optimizer was used during training, with an initial learning rate of 1e-3, and cosine annealing was employed to adjust the learning rate. The batch size was set to 512, and the total number of training epochs was 3000. An early stopping mechanism was implemented, terminating training when the validation set loss failed to decrease for zero consecutive epochs. All training was completed on a single NVIDIA RTX 5080 GPU.

[0137] 5. Post-training verification and simulation application.

[0138] Evaluation results on the independent test set show that the mean absolute error of the force density component is 0.08 N / m³, and the coefficient of determination R² reaches 0.998, indicating that the network has established a highly accurate mapping relationship. Deploying the trained model into a complete creep damage simulation pipeline successfully reproduces the entire process of creep strain accumulation in a rock slab under long-term loading, the evolution of damage from zero, and the eventual appearance of a localized failure zone in the middle of the slab. Compared with the benchmark solution using only traditional iterative methods, this method reduces the computation time by approximately 40 times while achieving the same accuracy (relative error of displacement field <2%).

[0139] Example 3

[0140] One embodiment of the present invention provides a specific example of a method for simulating rock creep processes based on deep learning-based near-field dynamics, such as... Figure 2 As shown, it includes:

[0141] A1 solution domain construction:

[0142] Furthermore, the main contents of constructing the A1 solution domain are: determining the geometric solution domain and boundary conditions, including the geometric shape information of the rock mass such as width, length, and fractures, and selecting the coordinate system and loading or constraint method. Physical properties of the rock mass, such as elastic modulus, Poisson's ratio, density, maximum tensile strength, cohesion, friction angle, and creep parameters, are set.

[0143] A2 parameter initialization;

[0144] Furthermore, the main contents of A2 parameter initialization are as follows: discretize the solution domain into a set of material points, determine the point spacing and neighborhood radius, and construct a neighborhood for each material point. Initialize the reference position state, deformation state, shape matrix, and deformation gradient of each point. Set the body force density and calculation time step, and initialize state variables such as displacement, velocity, elastic strain, creep strain, stress, and damage variables at each material point.

[0145] A3 model training:

[0146] Furthermore, the A3 model training utilizes the method described in Example 2 to train the model, which is then deployed to traditional iterative computation to improve computational efficiency. Dataset generation: Considering various loading and creep conditions, including stretching, shearing, and different creep correlation coefficients, the parameters of the input and output layers corresponding to each condition are calculated to form a dataset, which is then divided into a test set and a validation set. The network structure is determined, and weights and biases are initialized. Simultaneously, the training set data is input into the neural network, and the predicted values ​​and loss function values ​​are calculated through forward propagation. Automatic differentiation is used to update the weights and biases of the neural network based on the loss function values, followed by backpropagation. This process is repeated until the loss function values ​​converge or the preset number of training rounds is reached. The final performance of the model is evaluated using the test set data; if the model performance is unsatisfactory, the network structure is adjusted.

[0147] A4 Iteration and Update:

[0148] Furthermore, the A4 loop iterates and updates, traversing the interaction of each pair of material points during each time step iteration. It calls the already deployed model, quickly obtains the unconventional force density components through input parameters, and updates information such as material point acceleration, displacement, stress, and strain. Subsequently, bond fracture is determined. When the maximum tensile strength or the Mohr-Coulomb criterion is satisfied, the interaction disappears, the damage parameters are updated, and the process proceeds to the next time step iteration.

[0149] A5 Results Analysis:

[0150] Furthermore, the main contents of the A4 results analysis are: outputting and analyzing the stress field, displacement field, creep strain, and damage distribution that evolve over time.

[0151] Example 4

[0152] One embodiment of the present invention provides a rock mass experimental verification process for a rock creep process simulation method based on deep learning near-field dynamics. The method uses a homogeneous rock mass without initial macroscopic cracks as the research object to verify the effectiveness and computational efficiency of the method of the present invention in simulating the creep deformation and damage evolution process of rock mass under long-term load.

[0153] First, following step A1, a two-dimensional plane stress condition rock mass solution domain is constructed. The geometric model is a rectangular rock mass with a reasonable length and height ratio. The rock mass boundary conditions adopt a method of fixing the bottom and applying a constant normal stress at the top, with the left and right boundaries set as free boundaries. A coordinate system is selected and the loading direction is determined. Material parameters of the rock mass, such as elastic modulus, Poisson's ratio, density, maximum tensile strength, cohesion, internal friction angle, creep parameters, and reference stress, are initialized.

[0154] Secondly, the solution domain is discretized according to step A2, dividing the continuous rock mass into a uniformly distributed set of material points. The spacing between material points and the neighborhood radius δ are set to construct a nonlocal domain for each material point. The reference position state, deformation state, shape matrix and its inverse matrix of each material point are initialized, and the displacement, velocity, elastic strain, creep strain, stress and damage parameters of each material point are set to zero at the initial moment.

[0155] Subsequently, following step A3, the force density is rapidly predicted using the deep learning near-field dynamics method described in step S3. Prior to this, the rapid force density prediction model is established as follows: Based on typical creep loading states under plane stress conditions, a training dataset is generated, including input features such as initial bond length, bond elongation, bond rotation angle, creep parameters, and deformation gradient-related invariants, as well as the corresponding normal and tangential force density components as outputs. The dataset is divided into training, validation, and test sets to complete the training of the neural network model and deploy it into the near-field dynamics calculation process.

[0156] During the time step progression, all material point pairs are traversed according to step A4. The trained neural network model is invoked to quickly obtain the force density components of each bond, and the acceleration, velocity, and displacement information of the material points are updated accordingly. Subsequently, the elastic strain, creep strain, and stress state are updated based on the Lemaitre creep damage model, and the bond fracture is determined according to the maximum tensile strength and the Mohr-Coulomb criterion described in step S4, with damage parameters updated in real time.

[0157] Finally, following step A5, the displacement field, stress field, creep strain distribution, and damage evolution characteristics of the rock mass over time are output and analyzed. The results show that under constant load, the rock mass exhibits typical initial creep, stable creep, and accelerated creep stages. The method of this invention can stably simulate the creep deformation process of the rock mass and significantly improves computational efficiency compared to traditional near-field dynamic creep calculation methods.

[0158] Example 5

[0159] One embodiment of the present invention provides a rock creep process simulation system based on deep learning near-field dynamics, comprising:

[0160] The information acquisition module is configured to acquire parameter information of the target rock to be simulated, including the rock's geometry and the physical properties of the rock mass.

[0161] The model building module is configured to: discretize the target rock into several material points based on the parameter information, and build a creep damage model of the target rock;

[0162] The iterative simulation module is configured to perform time-step iterative simulation of the rock creep process based on the initialized creep damage model, and finally obtain the displacement field, stress field and creep strain distribution of the target rock over time.

[0163] In each iteration time step, firstly, based on the physical field information of the current time step, the force density components between each material point pair are predicted using a neural network model to obtain the force density distribution of the current time step; then, based on the force density distribution, the physical field information is updated using unconventional near-field dynamics theory; finally, based on the updated physical field information, it is determined whether the connection bond between the material point pairs has broken, the damage parameters are updated based on the determination result, and the iteration proceeds to the next time step until the termination condition is triggered.

[0164] Example 6

[0165] One embodiment of the present invention provides a computer program product, including a computer program that, when executed by a processor, implements the rock creep process simulation method based on deep learning peri-field dynamics.

[0166] Example 7

[0167] In one embodiment of the present invention, a non-transitory computer-readable storage medium is provided for storing computer instructions. When the computer instructions are executed by a processor, they implement the rock creep process simulation method based on deep learning peri-field dynamics.

[0168] Example 8

[0169] One embodiment of the present invention provides an electronic device, including: a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the rock creep process simulation method based on deep learning peri-field dynamics.

[0170] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0171] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0172] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A method for simulating a creep process of rock based on deep learning near-field dynamics, characterized in that, include: Obtain parameter information of the target rock to be simulated, including the rock's geometry and the physical properties of the rock mass; Based on the parameter information, the target rock is discretized into several material points to construct a creep damage model of the target rock; Based on the initialized creep damage model, time-step iterative simulation of rock creep process is performed to obtain the displacement field, stress field and creep strain distribution of target rock as it evolves over time. In each iteration time step, firstly, based on the physical field information of the current time step, the force density components between each material point pair are predicted using a neural network model to obtain the force density distribution of the current time step; then, based on the force density distribution, the physical field information is updated using unconventional near-field dynamics theory; finally, based on the updated physical field information, it is determined whether the connection bond between the material point pairs has broken, the damage parameters are updated based on the determination result, and the iteration proceeds to the next time step until the termination condition is triggered. The construction of the creep damage model for the target rock specifically includes: The Cauchy strain tensor is decomposed into the sum of the elastic strain tensor and the creep strain tensor. Damage parameters are introduced to describe the degree of damage at a point in the fractured rock mass, and the Lemaitre equivalent stress is used to characterize the material response under damage conditions. The rate of change of the creep strain tensor is expressed as: where t is time, d is differential operator, A is creep related coefficient, n is stress exponent, is the reference stress, , are the Lemaitre equivalent Mises stress and deviatoric stress, respectively.

2. The deep learning based near-field dynamics rock creep process simulation method of claim 1, wherein, The neural network model uses a fully connected neural network to construct the mapping relationship between bond geometry variables and force density, and consists of an input layer, a hidden layer, and an output layer; The input layer includes the bond length before deformation, bond elongation, bond rotation angle, creep correlation coefficient, creep strain determinant, and deformation gradient determinant. The output layer outputs the normal and tangential components of the force density vector.

3. The rock creep process simulation method based on deep learning peri-field dynamics as described in claim 1, characterized in that, The determination of whether the bonds between substance point pairs have broken specifically includes: Based on the maximum tensile strength criterion and the Mohr-Coulomb criterion, tensile failure and shear failure are determined. When any two material points satisfy either criterion, the bond is determined to be broken and the interaction disappears.

4. The rock creep simulation method based on deep learning peri-field dynamics as described in claim 1, characterized in that, The method of updating physical field information through unconventional near-field dynamics theory specifically includes: Based on the force density distribution, the acceleration, velocity and displacement information of the material point are updated through the unconventional near-field dynamic equation, and the elastic strain, creep strain and stress state are updated through the creep damage model.

5. A rock creep process simulation system based on deep learning peri-field dynamics, characterized in that, The method for simulating rock creep processes based on deep learning-based near-field dynamics as described in any one of claims 1-4 includes: The information acquisition module is configured to acquire parameter information of the target rock to be simulated, including the rock's geometry and the physical properties of the rock mass. The model building module is configured to: discretize the target rock into several material points based on the parameter information, and build a creep damage model of the target rock; The iterative simulation module is configured to perform time-step iterative simulation of the rock creep process based on the initialized creep damage model, and finally obtain the displacement field, stress field and creep strain distribution of the target rock over time. In each iteration time step, firstly, based on the physical field information of the current time step, the force density components between each material point pair are predicted using a neural network model to obtain the force density distribution of the current time step; then, based on the force density distribution, the physical field information is updated using unconventional near-field dynamics theory; finally, based on the updated physical field information, it is determined whether the connection bond between the material point pairs has broken, the damage parameters are updated based on the determination result, and the iteration proceeds to the next time step until the termination condition is triggered.

6. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the rock creep process simulation method based on deep learning near-field dynamics as described in any one of claims 1-4.

7. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium is used to store computer instructions, which, when executed by a processor, implement the rock creep process simulation method based on deep learning peri-field dynamics as described in any one of claims 1-4.

8. An electronic device, characterized in that, include: The device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to perform the simulation method for rock creep process based on deep learning near-field dynamics as described in any one of claims 1-4.

Images