Rock mass chemical corrosion simulation method and system based on neural network near field dynamics

Abstract

The invention provides a rock mass chemical corrosion simulation method and system based on neural network near-field dynamics, and relates to the technical field of fractured rock mass chemical corrosion simulation, and the method comprises the steps: building a solid mechanical field model based on a solid field near-field dynamics theory of a Hoek-Brown failure criterion, and building a chemical field model based on a neural network near-field dynamics theory; in a solid field, average principal stress among material points is calculated in real time based on a Hoek-Brown failure criterion, a nonlinear shear tensile failure relation in a stress state is obtained, in a chemical field, a concentration distribution prediction model is incorporated into the chemical field, and in an iteration time step of each solid mechanical field, a nonlinear shear tensile failure relation is obtained. And predicting the concentration distribution of the substance points in real time by adopting a concentration distribution prediction model, calculating a corresponding stress state, updating damage parameters, feeding back the updated damage parameters to the concentration distribution prediction model, and iteratively predicting the concentration distribution of the substance points in the next time step until an iteration termination condition is reached, thereby realizing a stress-corrosion coupling simulation calculation process.

Description

Technical Field

[0001] This disclosure relates to the field of chemical corrosion simulation technology for fractured rock masses, specifically to a method and system for simulating chemical corrosion of rock masses based on neural network near-field dynamics. Background Technology

[0002] The statements in this section are merely background information relating to this disclosure and do not necessarily constitute prior art.

[0003] In geotechnical engineering, underground energy storage, nuclear waste geological disposal, and slope engineering, fractured rock masses are exposed to chemically corrosive environments such as groundwater, acidic solutions, or salts for extended periods. These chemical media undergo complex chemical reactions with the rock mass's mineral components (such as dissolution, precipitation, and ion exchange), leading to significant deterioration of the rock's mechanical properties (such as strength and stiffness) and potentially inducing new microcracks or accelerating the propagation of existing cracks. This strong coupling between chemical corrosion damage and stress-deformation (chemical-mechanical coupling) is a key factor affecting the long-term stability and safety of engineering structures. Therefore, accurately and efficiently simulating the mechanical response and damage evolution of fractured rock masses under chemically corrosive environments is crucial for assessing the long-term service performance of engineering structures, predicting disaster risks, and optimizing protective designs.

[0004] However, existing methods for simulating chemical corrosion processes still have the following problems: (1) High nonlinearity and strong coupling: The chemical corrosion rate is usually strongly affected by multiple factors such as solution concentration, stress state and crack aperture, and these factors feed back to each other, forming a highly nonlinear and strongly coupled system. Traditional step-by-step iterative or weakly coupled simulation methods are difficult to accurately capture this complex interaction mechanism.

[0005] (2) Bottleneck of long time duration and computational efficiency: Actual rock chemical corrosion is a slow, long-term process. Even if advanced computational methods can effectively handle fracture problems, in order to achieve the necessary accuracy, it is usually necessary to perform massive time-step iterative calculations when simulating this slow-evolving, strongly nonlinearly coupled chemical-mechanical process. This results in extremely high computational costs and unacceptable simulation time, which seriously restricts the application of this method in the analysis of large-scale practical engineering problems. Summary of the Invention

[0006] To address the aforementioned issues, this disclosure proposes a method and system for simulating chemical corrosion of rock masses based on neural network peri-field dynamics. The Hoek-Brown failure criterion is introduced into peri-field dynamics theory, and a deep learning peri-field dynamics calculation framework for force-chemical coupling of fractured rock masses is constructed. The Hoek-Brown criterion and tensile strength criterion accurately describe the nonlinear shear-tensile failure of rock masses under complex stress states, thereby realizing stress-corrosion coupled simulation.

[0007] According to some embodiments, the present disclosure adopts the following technical solutions: A method for simulating chemical corrosion of rock masses based on neural network near-field dynamics includes: A solid mechanical field model is constructed based on the Hoek-Brown failure criterion and a chemical field model is constructed based on the neural network peri-field dynamics theory. In the solid field, the average principal stress between material points is calculated in real time based on the Hoek-Brown failure criterion to obtain the nonlinear shear-tensile failure relationship under stress. In the chemical field, the concentration distribution prediction model is incorporated into the chemical field. In each iteration time step of the solid mechanical field, the concentration distribution prediction model is used to predict the concentration distribution of material points in real time and calculate the corresponding stress state, update the damage parameters, and use the updated damage parameters to feed back to the concentration distribution prediction model to iteratively predict the concentration distribution of material points in the next time step until the iteration termination condition is reached, thus realizing the stress-corrosion coupled simulation calculation process.

[0008] According to some embodiments, the present disclosure adopts the following technical solutions: A rock mass chemical corrosion simulation system based on neural network near-field dynamics includes: The model building module is used to construct solid mechanical field models based on the Hoek-Brown failure criterion and solid field peri-field dynamics theory, and to construct chemical field models based on neural network peri-field dynamics theory. The coupled simulation module is used to calculate the average principal stress between material points in real time in a solid field based on the Hoek-Brown failure criterion, and obtain the nonlinear shear-tensile failure relationship under stress state. In a chemical field, the concentration distribution prediction model is incorporated into the chemical field. In each iteration time step of the solid mechanical field, the concentration distribution prediction model is used to predict the concentration distribution of material points in real time, calculate the corresponding stress state, update the damage parameters, and use the updated damage parameters to feed back to the concentration distribution prediction model to iteratively predict the concentration distribution of material points in the next time step until the iteration termination condition is reached, thus realizing the stress-corrosion coupled simulation calculation process.

[0009] According to some embodiments, the present disclosure adopts the following technical solutions: A computer program product includes a computer program that, when executed by a processor, implements the aforementioned method for simulating rock chemical corrosion based on neural network near-field dynamics.

[0010] According to some embodiments, the present disclosure adopts the following technical solutions: A non-transitory computer-readable storage medium is provided for storing computer instructions, which, when executed by a processor, implement the aforementioned rock mass chemical corrosion simulation method based on neural network near-field dynamics.

[0011] According to some embodiments, the present disclosure adopts the following technical solutions: 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 chemical corrosion simulation method based on neural network near-field dynamics.

[0012] Compared with the prior art, the beneficial effects of this disclosure are as follows: This invention discloses a method for simulating chemical corrosion of rock mass based on neural network peridynamics. It introduces peridynamic theory based on the Hoek-Brown failure criterion and establishes a stress analysis solution model by introducing a nonlocal differential operator. Compared with the classical energy release rate criterion of peridynamics, the Hoek-Brown criterion and tensile strength criterion can more accurately describe the nonlinear shear-tensile failure of rock mass under complex stress state, and overcome the defect that a single criterion cannot distinguish the failure mode.

[0013] This disclosed method for simulating chemical corrosion of rock masses based on neural network near-field dynamics employs a concentration distribution prediction model to predict the concentration distribution of material points in real time within a chemical field, calculates the corresponding stress state, updates damage parameters, and feeds the updated damage parameters back to the concentration distribution prediction model. This effectively avoids the redundancy of traditional iterative calculations in near-field dynamic chemical corrosion simulations, achieving rapid concentration prediction in stress-corrosion coupled simulations. The neural network can efficiently obtain the concentration distribution at each solid time step, significantly reducing computation time and resource consumption, and improving simulation efficiency.

[0014] This disclosed method for simulating chemical corrosion of rock masses based on neural network near-field dynamics captures the complex nonlinear relationship between stress and corrosion by training near-field dynamic seepage results through neural networks. Compared with traditional methods, the neural network model has stronger adaptability and can handle complex scenarios with different rock mass types, fracture distributions, and solid phase concentration characteristics, thus improving the accuracy of simulation results. Attached Figure Description

[0015] The accompanying drawings, which form part of this disclosure, are used to provide a further understanding of this disclosure. The illustrative embodiments of this disclosure and their descriptions are used to explain this disclosure and do not constitute an undue limitation of this disclosure.

[0016] Figure 1This is a schematic diagram of the near-field dynamics of material points according to an embodiment of this disclosure; Figure 2 This is a flowchart illustrating the training process of a fully connected neural network for rock chemical corrosion according to an embodiment of this disclosure. Figure 3 This is a stress-corrosion coupling calculation framework for near-field dynamics in an embodiment of this disclosure. Detailed Implementation

[0017] The present disclosure will be further described below with reference to the accompanying drawings and embodiments.

[0018] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of this disclosure. 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 disclosure pertains.

[0019] 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 exemplary embodiments according to this disclosure. 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.

[0020] Example 1 One embodiment of this disclosure provides a method for simulating chemical corrosion of rock masses based on neural network near-field dynamics, the method steps of which include: Step 1: Construct a solid mechanical field model based on the peri-field dynamics theory of solid fields according to the Hoek-Brown failure criterion, and construct a chemical field model based on the peri-field dynamics theory of neural networks; Step 2: In the solid field, the average principal stress between material points is calculated in real time based on the Hoek-Brown failure criterion to obtain the nonlinear shear-tensile failure relationship under stress. In the chemical field, the concentration distribution prediction model is incorporated into the chemical field. In each iteration time step of the solid mechanical field, the concentration distribution prediction model is used to predict the concentration distribution of material points in real time and calculate the corresponding stress state, update the damage parameters, and use the updated damage parameters to feed back to the concentration distribution prediction model to iteratively predict the concentration distribution of material points in the next time step until the iteration termination condition is reached, thus realizing the stress-corrosion coupled simulation calculation process.

[0021] As one embodiment, the rock mass chemical corrosion simulation method based on neural network peridynamics disclosed herein introduces peridynamic theory based on the Hoek-Brown failure criterion, combines it with chemical field peridynamic equations, and predicts the corrosion process of fractured rock masses through a fully connected neural network, forming a deep learning peridynamic framework for chemical corrosion damage. The specific implementation process is as follows: Step 1: Introduce the Hoek-Brown failure criterion into the solid field peri-field dynamics theory. First, use the solid field peri-field dynamics theory based on the Hoek-Brown failure criterion to perform rock mass stress analysis and calculation. Specifically, the stress calculation process in the peri-field dynamics theory based on the Hoek-Brown failure criterion is as follows: (1) Where, coordinates , and Relative position vector The amount, For infinitesimal residuals, the function Given quantities - For the change, For partial differentials, the derivative can be calculated using formula (1) for two-dimensional analysis: (2) in, and functions respectively for and The derivative of for Other interacting matter points in the neighborhood The set, For two-dimensional spatial measurement, They are nonlocal orthogonal functions. and This corresponds to the order. Nonlocal function. Satisfies the following orthogonality: (3) Furthermore, the near-field dynamic stress calculation utilizes differential operators to represent the Navier equilibrium equations. In the two-dimensional case, the stress components can be constructed as follows: (4) (5) (6) in, and Divided into the elastic modulus and Poisson's ratio of the material, and These represent displacements in two coordinate directions, respectively.

[0022] Furthermore, the near-field dynamic governing equations and nonlocal force densities of the solid field are expressed as follows: (7) (8) in, for exist t acceleration at any moment For solid density, for Physical strength at the point, For the force density of classical peridynamics, , They are respectively , The position after deformation This represents the relative displacement of the material points after the shape change. Let the potential energy at a point be... The micromodulus of the model, Let be the elongation of the bond. Where, elongation... and strain energy density at They are represented as follows: (9) (10) in, Interaction points volume, W Let be the strain energy density at a point.

[0023] Furthermore, matter points and The characterization function of the bond breakage state is: When the function This indicates that the bond is not broken and there is a nonlocal force, when the function The time indicates that the bond breaks and the nonlocal force disappears. The degree of damage at point x is defined as: (11) when When x is broken, it means that all bonds at x are broken and the point is completely destroyed.

[0024] As one example, the Hoek-Brown failure criterion is constructed. For two-dimensional analysis, in and Calculate the maximum and minimum principal stresses at each location: (12) (13) Furthermore, when and When the stress conditions between them satisfy the following conditions, the determination is made. and The bonds between them are sheared and broken: (14) in, and They are respectively and The average value of the maximum and minimum principal stresses. Represents the uniaxial compressive strength of intact rock. , , a The parameters of the Hoek-Brown rock mass are derived from the geological strength index. GSI Disturbance factor D Rock brittleness constant Sure: (15) when and When the stress conditions between them satisfy the following conditions, the determination is made. and The bonds between them are subjected to tensile failure: (16) in, The uniaxial tensile strength of the rock is represented by the Brazilian splitting strength or the result of a direct tensile test. If no test is available, it can be estimated using the Hoek-Brown parameters. (17) During the iteration of the near-field dynamic equations in the solid field, the average principal stress between material points is calculated in real time, and the connecting bonds gradually undergo shear or tensile failure. The Hoek-Brown criterion and tensile strength criterion can more accurately describe the nonlinear shear and tensile failure of rock masses under complex stress states.

[0025] Step 2: In the chemical field, the concentration distribution prediction model is incorporated into the chemical field, and then coupled simulation calculations with the solid field are performed.

[0026] Specifically, in the traditional near-field dynamics theory of chemical field corrosion of rock masses, corrosion is regarded as a dissolution-diffusion process, with material points... exist t Solid concentration at time In the neighborhood, with other matter points Concentration information is transmitted through diffusion bonds, and its evolution equation is as follows: (18) in, For material points and The micro-diffusion coefficient between the two bonds depends on the properties of the material points at both ends of the bond (solid or liquid phase). The values ​​differ. The solid-liquid phase properties of a substance point can be determined by the corrosion damage at that point. d c Factor representation, d c A value of 1 indicates liquid phase properties and no mechanical interaction with surrounding material points. d c A value less than 1 indicates that the point is a solid phase property point. and d c The relationship between them is: (19) in, The coefficient of solubility is micro. The diffusion coefficient is a property of liquid phases and can be determined based on ion diffusivity. D calculate: (20) in, δ The neighborhood radius is used to describe the effect of the solid field stress state on the corrosion micro-solution coefficient. To mitigate the impact, a linear stress enhancement model is introduced: (twenty one) The dissolution rate constant under no stress. The hydrostatic stress at the material point is positive. β The stress sensitivity coefficient is determined experimentally. The reference stress is used. In two-dimensional analysis, the hydrostatic stress can be expressed as: When under tensile stress, the corrosion rate is amplified, and conversely, the corrosion rate is weakened. To represent the degree of structural corrosion damage, [the following is used:] d c Defined as concentration-dependent damage coefficient: (twenty two) in, This represents the solid concentration (moles per unit volume) when the solid phase is not corroded. This represents the saturation concentration of ions in the liquid phase. When the concentration of ions in the solid phase... After corrosion and dissolution, the concentration decreased to liquid phase saturation. When the material completely dissolves, the solid phase point transforms into the liquid phase point, and all mechanical interaction bonds connecting this material point break.

[0027] Furthermore, non-liquid phase matter points Before chemical corrosion occurs, the number of unbroken bonds in the vicinity is: N c,x Construct the initial concentration with the first term (the first term of the arithmetic sequence) as the initial concentration, and the last term as... ,common N c,x The concentrations of the items are randomly distributed using methods such as Fisher-Yates shuffling. N c,x Within each connecting bond. With matter points... Chemical corrosion occurs, the concentration continuously decreases, and when the substance becomes... When the concentration value is lower than the allocation array value, the corresponding interaction bond undergoes irreversible breakage; when it decreases to... At that point, the substance completely dissolves and transforms into a liquid phase.

[0028] Furthermore, this application incorporates a concentration distribution prediction model into a chemical field, establishes a fully connected neural network as a concentration distribution prediction model to predict the concentration of chemical corrosion process in fractured rock masses, and performs coupled simulation calculations with the solid field. Its architecture includes an input layer, hidden layers, and an output layer. The features in the input layer need to fully describe the information of the material points and ensure the input... With output Matching features, selecting spatial coordinates x, and considering the degree of corrosion damage. d c and boundary condition eigenvalues b bnd : (twenty three) in, sta () represents the standardization of input features. Furthermore, the number of neurons in the input layer equals the number of features. Further, to capture the non-linear relationship between input and output, multiple hidden layers are set, and appropriate methods are selected for hyperparameter tuning to determine the suitable network depth, including but not limited to grid search, random search, and Bayesian optimization.

[0029] Furthermore, all hidden layers use non-linear activation functions, including but not limited to ReLU, SELU, Sigmoid, and Tanh, whose activation functions are all suitable for model training. Among them, the ReLU activation function is relatively simple to compute and can accelerate training. (twenty four) When input dWhen 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. Furthermore, ReLU does not suffer from the vanishing gradient problem compared to Sigmoid and Tanh, making it suitable for deep networks. The output layer does not require a non-linear activation function. (25) Furthermore, to ensure the neural network can be trained effectively on the input data... b bnd Perform appropriate preprocessing and choose a normalization (Z-Score Normalization) method that can effectively accelerate training. This method can ensure that all input data have the same scale, thereby avoiding the instability of training caused by the excessive magnitude of certain features.

[0030] To expand the training set, the input data coordinates are randomly translated to increase data diversity, thereby improving the model's generalization ability. For two-dimensional analysis, in the coordinate... x 1 and x Add d to each of the two locations. x 1 and d x 2. The original coordinates are transformed into ( x 1+d x 1, x 2+ d x 2).

[0031] Furthermore, during the forward propagation process of model training, the input data... The predicted value is obtained through calculation by the network layer. The neural network has the following number of layers: L The calculation for each layer is then expressed as: (26) 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 Let be the weight matrix and bias vector in the l-th layer. 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 The calculation formula is: (27) in and The first iThe actual concentration value and the predicted concentration value of each substance point N This represents the total number of material points. Backpropagation updates the parameters by calculating the gradient of the loss function with respect to each parameter and then using gradient descent. The gradients of the output layer and the hidden layer are as follows: (28) (29) in, For the first l Linear combination of layers, And represents the gradient term, Let be the derivative of the activation function, and ⊙ denote element-wise multiplication. The weight matrix and bias vector are as follows: (30) To avoid constructing a complete derivative expression, the gradient is solved using automatic differentiation by AD.

[0032] Step 3: Perform coupled simulation calculations of chemical field and solid field. Use concentration distribution prediction model to predict the concentration distribution of material points in real time, calculate the corresponding stress state, update the damage parameters, and use the updated damage parameters to feed back to the concentration distribution prediction model to iteratively predict the concentration distribution of material points in the next time step until the iteration termination condition is reached, thus realizing the coupled simulation calculation process of mechanical-chemical (stress-corrosion) of fractured rock mass.

[0033] Specifically, in the deep learning near-field dynamics algorithm for coupled mechanics and chemical processes in fractured rock masses, the trained solid concentration distribution prediction model of the fractured rock mass is incorporated into the near-field dynamics stress-corrosion coupled calculation. During each time step of the solid mechanics field solution, the concentration distribution prediction model is used to quickly obtain the concentration distribution of each material point, eliminating the need for numerous fluid iteration steps in the chemical field corrosion near-field dynamics method, thus greatly improving computational efficiency. Under corrosion, the solid phase of the rock mass gradually dissolves into a liquid phase, leading to the movement of the solid-liquid phase boundary, causing changes in the displacement and stress fields, resulting in stress concentration. Under stress, the fractured rock mass fractures according to the Hoek-Brown and tensile criteria. As damage accumulates, corrosion damage is more likely to penetrate inwards. Simultaneously, based on the different solid-liquid phases and stress states of the material points, the corrosion diffusion coefficient and micro-dissolution coefficient are recalculated and distributed, thereby altering the material's corrosion process.

[0034] As one embodiment, the simulation method of the rock mass chemical corrosion simulation method based on neural network near-field dynamics disclosed herein includes: Step (1): Construct the solution domain, including: initializing the solution domain, defining the geometry of the fractured rock mass, including parameters such as the thickness, length, distribution and aperture of the rock mass; setting the physical properties of the rock mass and fluids, such as the rock mass's elastic modulus, Poisson's ratio, solid concentration, compressive strength, tensile strength, geological strength index, disturbance factor, rock brittleness constant, as well as the saturation concentration and ion diffusion rate of non-rock masses, to ensure that the solution model can truly reflect the chemical-stress coupling behavior of the fractures.

[0035] Step (2): Initialize the calculation parameters, including: set the calculation parameters according to the structural characteristics of the fractured rock mass, including the number of material points, neighborhood radius, boundary conditions, etc.; configure the required nonlocal differential operators, bond micromodulus, strain energy density, corrosion microdissolution coefficient, etc.; discretize the solution domain according to these calculation parameters and generate the coordinates of the material points, and determine the adjacent material points in the neighborhood of each material point; initialize the bonds and nonlocal forces of the material points in their near-field neighborhood, initialize the material point displacement, initial corrosion damage coefficient, etc.

[0036] Step (3): Training the concentration distribution prediction model, including: using the numerical results of near-field kinetic chemical corrosion simulation as the training set, changing various boundary conditions to increase the number of training sets, and dividing the test set; normalizing the boundary feature values ​​and other data; determining the network structure and initializing the weights and biases; inputting the training set data into the neural network, calculating the predicted value and loss function value through forward propagation; using automatic differentiation to update the weights and biases of the neural network based on the loss function value, and performing backpropagation; repeating the above process until the loss function value converges or reaches the preset number of training rounds; using the test set data to evaluate the final performance of the model, and adjusting the network structure if the model performance is poor.

[0037] Step (4): Perform the chemical-corrosion coupling simulation process, including: deploying the trained prediction model and carrying out the force-chemical coupling simulation of the fractured rock mass; in each iteration time step of the solid mechanical field, directly use the neural network model to quickly obtain the concentration of material points; calculate the nonlocal force, acceleration, displacement, stress and other information of each material point; if the stress state of the material point satisfies the Hoek-Brown criterion or the tensile strength criterion, the bond nonlocal force between material points disappears and the damage parameters are updated; after updating the solid mechanical field, feed the information back to the prediction model and obtain the material point concentration distribution of the next iteration time step, realize the stress-corrosion coupling calculation process, until the calculation termination condition is triggered.

[0038] Step (5): Coupled result analysis, output the rock mass-chemical corrosion coupled simulation results, visualize and analyze the crack propagation process and concentration distribution.

[0039] Example 2 One embodiment of this disclosure provides a method for simulating chemical corrosion of rock masses based on neural network near-field dynamics. This embodiment uses a two-dimensional plane strain limestone fracture specimen, and the specific simulation calculation process is as follows: Step 1: Constructing the computational model. A two-dimensional plane strain limestone fracture specimen was selected, with uniform geometric dimensions: length L = 0.20 m, height h = 0.05 m. A crack with a 60-degree angle to the horizontal was pre-fabricated in the center of the specimen, with an initial crack length of 0.01 m. The crack surface was treated with "empty bonds," meaning that all bonds between material points on both sides of the crack were pre-broken, and their nonlocal interaction forces disappeared. The entire computational domain was discretized using a uniform square grid with a material point spacing of 1 mm, totaling 10,000 material points. The neighborhood radius δ of each point was set to 3.015 mm to meet the convergence requirements of near-field dynamics while also considering computational efficiency.

[0040] Step 2: Set material parameters and boundary conditions. Rock elastic modulus 14 GPa, Poisson's ratio ν = 0.3, density... Geological strength index GSI =65, perturbation factor D=0.7, rock brittleness constant is taken as 12, rock compressive strength is 60MPa, tensile strength is 0.60MPa. An increasing downward vertical displacement is applied to the upper boundary, with a final value of 0.30mm over time. The lower boundary has a fixed vertical displacement, and the left and right boundaries are free boundaries. Displacement is output in real time during the calculation, and stress at each material point is calculated using a nonlocal differential operator.

[0041] Step 3: Destruction Assessment. Calculate the material points. and The average maximum and minimum principal stresses are determined when the stress condition satisfies the following condition: and The bonds between them are sheared and broken: (31) when and When the stress conditions between them satisfy the following conditions, the determination is made. and The bonds between them are subjected to tensile failure: (32) Step 4: Result Analysis. The specimen initially exhibited a linear elastic response under horizontal displacement loading. When the displacement reached 0.05 mm, tensile stress concentration first appeared at the fracture tip, and the first tensile bond fractured instantaneously. The crack extended upwards for 3 mm perpendicular to the direction of the maximum principal tensile stress, forming a narrow tensile fracture band. With continued loading to 0.12 mm, the shear slip on both sides of the fracture surface, combined with osmotic pressure, satisfied the shear failure condition; shear bonds failed in batches, and the crack extended obliquely inwards for 9 mm at a 45° angle. When the displacement increased to 0.20 mm, a stepped crack path of tension-shear-retension was formed; a continuous shear band eventually formed in the middle of the specimen, connected at both ends by tensile cracks, resulting in an overall V-shaped failure. This embodiment demonstrates the ability of the Hoek-Brown criterion near-field dynamics framework to distinguish and evolve shear and tensile failure modes in fractured rock masses under boundary loads.

[0042] Example 3 One embodiment of this disclosure provides a method for simulating chemical corrosion of rock masses based on neural network near-field dynamics. This embodiment uses an underground acidic reservoir as a prototype, and the specific implementation process is as follows: Step 1: Scenario and Objective. This embodiment uses an underground acidic reservoir as a prototype. The reservoir measures 10m × 10m, and the lithology is fine-grained quartz sandstone with a uniaxial compressive strength of 60MPa, a tensile strength of 4.5MPa, and an elastic modulus of 18GPa. The reservoir is filled with a sulfuric acid solution with a pH of 3.5 and an ion diffusion coefficient of... The fracture network is statistically generated from measured joint orientations, with a maximum extension length of 6m. The model aims to achieve a high-precision mapping from "spatial coordinates + corrosion damage degree + boundary condition indicators" to "solid molar concentration" by training a multi-layer fully connected neural network. This will replace the traditional corrosion iterative solution in subsequent stress-corrosion coupled calculations, significantly improving computational efficiency while ensuring engineering accuracy, with the relative error controlled within 5%.

[0043] Step 2: Training Data Generation. To obtain sufficiently rich and physically reliable training samples, a database was first generated using traditional chemical field near-field dynamics. The model was discretized using a Cartesian grid with a spatial spacing of 0.05 m, generating 40,000 material points. The neighborhood radius was set to three times the spatial spacing to ensure complete coverage of nonlocal interactions. Boundary conditions were set as follows: a fixed concentration of 0.6 mol / L at the left boundary, zero flux at the right boundary, and symmetry between the upper and lower boundaries; different combinations of horizontal and vertical pressures were applied to the lateral stresses; initial fractures could be set at the left boundary. Further, Latin hypercube sampling was used to obtain 10 levels of concentration at the left boundary and 10 levels of initial corrosion damage in the rock mass; 4 levels of different pressure combinations were applied in the horizontal and vertical directions, resulting in 400 combinations. After calculation, coordinates, current corrosion damage level, boundary markers, and solid molar concentration were extracted at each node. The simulation results were divided into training, validation and test sets in an 8:1:1 ratio. Some data were normalized, the coordinates were randomly shifted, and uniform noise was applied to the penetration rate. An additional 20% of samples were generated to improve the robustness of the model.

[0044] Step 3: Configure the network structure. A fully connected feedforward architecture is adopted. Input layer: 5 neurons, corresponding to normalized spatial coordinates, erosion damage, and boundary features. Hidden layer 1: 64 neurons, ReLU activation function, weights initialized with a uniform Xavier distribution, bias initialized to 0, followed by BatchNorm1d for stable distribution. Hidden layer 2: 128 neurons, ReLU activation function, Dropout probability 0.1 to suppress overfitting. Hidden layer 3: 64 neurons, ReLU activation function, followed by BatchNorm1d again. Output layer: 1 neuron, linear activation, directly outputting the solid molar concentration. C Complete training was performed on a single RTX-3060 graphics card.

[0045] Step 4: Training parameter settings. Loss function: Mean Squared Error (MSE). Optimizer: Adam, initial learning rate. The performance decays by 0.5 every 50 epochs. The batch size is 2048, balancing GPU memory usage and convergence speed. The maximum training epochs are 300 epochs. The validation set MSE is monitored; if the MSE decreases within 15 consecutive epochs... Then terminate early. Gradient clipping: L2 norm upper limit of 5 to prevent gradient explosion.

[0046] Step 5: Model Validation and Error Analysis. The average relative error on the test set was 4.6%, and the maximum error was 7.9%, meeting engineering requirements. Further physical consistency verification was conducted: the predicted concentration was substituted into the model to calculate the diffusion flux at the next time step, and compared with the true value; the average error was 6.8%. Sensitivity analysis was performed: a 10% change in inlet concentration resulted in a prediction error change of less than 1%; a 20% change in fracture aperture resulted in an error change of less than 2%, indicating that the model has good robustness.

[0047] Example 4 One embodiment of this disclosure provides a specific application process of a rock mass chemical corrosion simulation method based on neural network near-field dynamics, including the following implementation details: Step 1: Model Embedding. The trained solid concentration prediction network is embedded as an independent module into the near-field dynamics main program; the network weights are loaded once when the program starts, and the concentration field is obtained by calling the network throughout the program without the need for re-initialization.

[0048] Step 2: Time Step Loop. At the beginning of each solid mechanics field time step, first read the spatial coordinates, corrosion damage values, and boundary features of all material points; input these features in batches into the loaded prediction network to obtain the full-field solid concentration distribution in milliseconds.

[0049] Step 3: Phase Update. Based on the predicted concentration and preset threshold, determine the solid-liquid phase of each material point: those with concentrations above the threshold remain in the solid phase, while those below the threshold become liquid phases; simultaneously update the corrosion damage coefficient and record the new position of the solid-liquid interface.

[0050] Step 4: Solving the mechanical field. Based on the updated phase distribution, recalculate the nonlocal forces, displacements, and stresses between material points; if the stress at a certain point satisfies the Hoek-Brown shear failure or tensile failure criterion, then break the corresponding bond and accumulate damage to achieve crack propagation.

[0051] Step 5: Parameter reallocation. After crack propagation and phase change are completed, the corrosion diffusion coefficient and micro-solidification coefficient are reallocated to each material point based on the new solid-liquid ratio and local stress level, so that the subsequent concentration prediction is consistent with the actual corrosion rate.

[0052] Step 6: Information Feedback. The displacement, stress, and new damage values ​​obtained in Step 4 are used as one of the input features for the next step and fed back to the prediction network; the network continues to output a new concentration field in the next time step, realizing the continuous advancement of stress-corrosion bidirectional coupling.

[0053] Step 7: Termination Judgment. Repeat steps 2 to 6 until the total simulation time reaches the preset value or the overall damage index meets the termination condition; output the final crack morphology, concentration cloud map and mechanical response to provide a basis for engineering safety evaluation.

[0054] Example 5 One embodiment of this disclosure provides a rock mass chemical corrosion simulation system based on neural network near-field dynamics, comprising: The model building module is used to introduce the near-field dynamics theory of solid field based on the Hoek-Brown failure criterion and the near-field dynamics theory of chemical field corrosion of rock mass to construct a mechanical-chemical coupling simulation model of fractured rock mass; The coupled simulation module is used to calculate the average principal stress between material points in real time in a solid field based on the Hoek-Brown failure criterion, and obtain the nonlinear shear-tensile failure relationship under stress state. In a chemical field, the concentration distribution prediction model is incorporated into the chemical field. In each iteration time step of the solid mechanical field, the concentration distribution prediction model is used to predict the concentration distribution of material points in real time, calculate the corresponding stress state, update the damage parameters, and use the updated damage parameters to feed back to the concentration distribution prediction model to iteratively predict the concentration distribution of material points in the next time step until the iteration termination condition is reached, thus realizing the stress-corrosion coupled simulation calculation process.

[0055] Example 6 One embodiment of this disclosure provides a computer program product, including a computer program that, when executed by a processor, implements the aforementioned rock mass chemical corrosion simulation method based on neural network near-field dynamics.

[0056] Example 7 One embodiment of this disclosure provides a non-transitory computer-readable storage medium for storing computer instructions, which, when executed by a processor, implement the rock mass chemical corrosion simulation method based on neural network near-field dynamics.

[0057] Example 8 One embodiment of this disclosure 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 mass chemical corrosion simulation method based on neural network near-field dynamics.

[0058] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. 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, create a machine 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.

[0059] 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.

[0060] While the specific embodiments of this disclosure have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of this disclosure. 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 this disclosure are still within the scope of protection of this disclosure.

Claims

1. A method for simulating chemical corrosion of rock mass based on neural network near-field dynamics, characterized in that, include: We introduce the peri-field dynamics theory of solid field based on the Hoek-Brown failure criterion and the peri-field dynamics theory of chemical field corrosion of rock mass to construct a mechanical-chemical coupling simulation model of fractured rock mass; In the solid field, the average principal stress between material points is calculated in real time based on the Hoek-Brown failure criterion to obtain the nonlinear shear-tensile failure relationship under stress. In the chemical field, the concentration distribution prediction model is incorporated into the chemical field. In each iteration time step of the solid mechanical field, the concentration distribution prediction model is used to predict the concentration distribution of material points in real time and calculate the corresponding stress state, update the damage parameters, and use the updated damage parameters to feed back to the concentration distribution prediction model to iteratively predict the concentration distribution of material points in the next time step until the iteration termination condition is reached, thus realizing the stress-corrosion coupled simulation calculation process.

2. The rock mass chemical corrosion simulation method based on neural network near-field dynamics as described in claim 1, characterized in that, In the theory of near-field dynamics of solid fields, the Hoek-Brown failure criterion is introduced. First, the stress analysis of rock mass is calculated using the theory of near-field dynamics of solid fields based on the Hoek-Brown failure criterion. The degree of damage to material points is defined. Based on the Hoek-Brown failure criterion, for two-dimensional analysis, the maximum and minimum principal stresses of material points are calculated. During the iteration process of the near-field dynamics equations of solid fields, the average principal stress between material points is calculated in real time, and the nonlinear relationship of the gradual shear or tensile failure of the connecting bonds is obtained.

3. The rock mass chemical corrosion simulation method based on neural network near-field dynamics as described in claim 1, characterized in that, The concentration distribution prediction model consists of an input layer, a hidden layer, and an output layer. The features of the input layer need to fully describe the information of the material points and ensure that the input and output features match. Spatial coordinates, corrosion damage degree, and boundary condition feature values ​​are selected. To capture the nonlinear relationship between the input and output, multiple hidden layers are set, and hyperparameters are adjusted to determine the appropriate network depth. All hidden layers use nonlinear activation functions.

4. The rock mass chemical corrosion simulation method based on neural network near-field dynamics as described in claim 3, characterized in that, The numerical results of near-field kinetic chemical corrosion simulations are used as the training set. Various boundary conditions are changed to increase the size of the training set, while a test set is also created. The boundary feature values ​​are normalized to determine the network structure and initialize the weights and biases. The training set data is then 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, and backpropagation is performed. The above process is repeated until the loss function values ​​converge or the preset number of training rounds is reached, thus completing the training of the concentration distribution prediction model.

5. The rock mass chemical corrosion simulation method based on neural network near-field dynamics as described in claim 1, characterized in that, The input to the concentration distribution prediction model is spatial coordinates, the degree of corrosion damage, and boundary condition characteristic values. During each time step of the solid mechanics field solution, the concentration distribution prediction model is used to obtain the concentration distribution of each material point. Under corrosion, the solid phase medium of the rock mass gradually dissolves into a liquid phase, causing the solid-liquid phase boundary to move, resulting in changes in the displacement field and stress field, thus generating stress concentration. Under stress, the fractured rock mass satisfies the Hoek-Brown and tensile criteria and fractures. As damage accumulates, corrosion damage is more likely to occur internally. At the same time, based on the different solid-liquid phases and stress states of the material points, the corrosion diffusion coefficient and micro-dissolution coefficient are recalculated and distributed, thereby changing the corrosion process of the material.

6. The rock mass chemical corrosion simulation method based on neural network near-field dynamics as described in claim 5, characterized in that, In each iteration time step of the solid mechanical field, the concentration of material points is quickly obtained directly using the concentration distribution prediction model; The nonlocal forces, accelerations, displacements, and stresses of each material point are calculated. If the stress state of a material point satisfies the Hoek-Brown criterion or the tensile strength criterion, the nonlocal forces between material points disappear, and the damage parameters are updated. After updating the solid mechanical field, the information is fed back to the concentration distribution prediction model, and the concentration distribution of material points in the next iteration time step is obtained, realizing the stress-corrosion coupled calculation process until the calculation termination condition is triggered.

7. A rock mass chemical corrosion simulation system based on neural network near-field dynamics, characterized in that, include: The model building module is used to introduce the near-field dynamics theory of solid field based on the Hoek-Brown failure criterion and the near-field dynamics theory of chemical field corrosion of rock mass to construct a mechanical-chemical coupling simulation model of fractured rock mass; The coupled simulation module is used to calculate the average principal stress between material points in real time in a solid field based on the Hoek-Brown failure criterion, and obtain the nonlinear shear-tensile failure relationship under stress state. In a chemical field, the concentration distribution prediction model is incorporated into the chemical field. In each iteration time step of the solid mechanical field, the concentration distribution prediction model is used to predict the concentration distribution of material points in real time, calculate the corresponding stress state, update the damage parameters, and use the updated damage parameters to feed back to the concentration distribution prediction model to iteratively predict the concentration distribution of material points in the next time step until the iteration termination condition is reached, thus realizing the stress-corrosion coupled simulation calculation process.

8. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the rock mass chemical corrosion simulation method based on neural network near-field dynamics as described in any one of claims 1-6.

9. 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 mass chemical corrosion simulation method based on neural network near-field dynamics as described in any one of claims 1-6.

10. 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 chemical corrosion based on neural network near-field dynamics as described in any one of claims 1-6.

Images