Structural fracture damage simulation method and system based on deep learning

Abstract

The invention belongs to the field of structural fracture damage simulation, and provides a structural fracture damage simulation method and system based on deep learning in order to solve the problem that the whole-process prediction precision of complex fracture behaviors depends on network architecture design and an optimization algorithm, and the universality is limited. The structure fracture damage simulation method based on deep learning comprises the following steps: constructing a neural network model and setting initial parameters; optimizing initial parameters of the neural network model based on a sample set and a loss function to obtain a trained neural network model; and on the basis of current structure geometric parameters, material parameters, boundary conditions and initial crack information, crack initiation, expansion and bifurcation combination are predicted in a structural domain by using the trained neural network model, and a displacement field, a stress field and a damage evolution result are obtained. The adaptability of the model to different working conditions can be enhanced by introducing multi-scale and multi-physical field information.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of structural fracture damage simulation, and particularly relates to a structural fracture damage simulation method and system based on deep learning. BACKGROUND

[0002] The statements in this section merely provide background information related to the present application and do not necessarily constitute prior art.

[0003] In the field of computer-aided engineering (CAE), structural analysis and fracture damage prediction have long relied on traditional finite element method (FEM). The FEM method realizes the prediction of structural stress, strain and crack development by weakly equivalent partial differential equations to integral equations and numerical integration on discrete grids. However, in the processing of FEM method, high-precision simulation requires fine grid division in key areas such as crack tip and high stress concentration area, which requires large manual input and complex operation. Due to the approximation of the weak equivalence method, the FEM simulation results have deviations in the prediction of complex fracture behaviors such as crack initiation, propagation, branching and merging. When the model, material or load conditions change, the grid needs to be re-divided and the complete numerical solution needs to be executed, which lacks computational reusability. Although the integration of deep learning and computer-aided engineering can improve the calculation accuracy and efficiency, there are still problems such as limited universality of the full-process prediction accuracy of complex fracture behaviors depending on network architecture design and optimization algorithm. SUMMARY

[0004] In order to solve the above technical problems, the present application provides a structural fracture damage simulation method and system based on deep learning, which can enhance the adaptability of the model to different working conditions.

[0005] In order to achieve the above purpose, the present application adopts the following technical solutions: The first aspect of the present application provides a structural fracture damage simulation method based on deep learning.

[0006] In one or more embodiments, a structural fracture damage simulation method based on deep learning is provided, comprising: constructing a neural network model and setting initial parameters; optimizing the initial parameters of the neural network model based on a sample set and a loss function to obtain a trained neural network model; wherein the samples in the sample set are composed of structural geometric parameters, material parameters, boundary conditions, initial crack information, displacement field, stress field and damage evolution results; the loss function is a loss function considering the partial differential equation residual of stress tensor, and the unified target of physical consistency, distribution consistency and prior constraint; Based on the current structure geometric parameters, material parameters, boundary conditions and initial crack information, the trained neural network model is used to predict the initiation, propagation and bifurcation and merging of cracks in the structure domain to obtain the displacement field, stress field and damage evolution results.

[0007] As an implementation form, the construction process of the loss function is: An energy functional of the structure is constructed based on the principle of minimizing energy, wherein the energy functional of the structure is obtained by multiplying a tensile energy part and a learnable degradation function representing the automatic adjustment of the degradation rate of fracture, and then adding a compression energy part; According to the crack characteristic length of the structure and the effective fracture toughness of the structure, a discretized energy functional of the structure is obtained. The energy density in the energy functional of the structure is rasterized into a probability distribution, the second-order Wassertein distance is calculated, and then combined with the partial differential equation residual of the stress tensor, the irreversible damage constraint, and the square of the difference between the derivative of the discretized energy functional of the damaged structure with respect to time and the external force input power, to construct the loss function.

[0008] As an implementation form, the crack characteristic length of the structure is determined by the size of the strain gradient and the base length parameter.

[0009] As an implementation form, the effective fracture toughness of the structure is represented by the cumulative sum of the weights learned by the neural network models of the I-type crack mode fracture toughness and the II-type crack mode fracture toughness.

[0010] As an implementation form, the expression of the loss function is: ; Wherein, is a partial differential equation residual operator, is a stress tensor; , , is a weight coefficient; is an external force input power; is an irreversible damage constraint; is a probability distribution; is a derivative with respect to time; is a discretized energy functional; are displacement field and damage variable, respectively.

[0011] As an implementation form, the expression of the discretized energy functional is: ; In the formula, is the strain energy per unit after numerical discretization of the structure. is a normalization constant; represents a structure in which damage occurs; is an effective fracture toughness; is a crack characteristic length; is a damage variable gradient; is a proportion of a tensile crack in a crack mode.

[0012] As an implementation, the effective fracture toughness is expressed as: ; ; wherein, is a mode I crack mode fracture toughness; is a mode II crack mode fracture toughness; is a strain tensor; is a tensile energy part; is a compression energy part; is a weight learned by a neural network model.

[0013] As an implementation, an expression of an irreversible damage constraint is: ; wherein, is an irreversible damage constraint; is a damage variable; is a historical maximum damage value; is a smoothing parameter; represents a structure in which damage occurs.

[0014] A second aspect of the present application provides a deep learning-based structure fracture damage simulation system.

[0015] In one or more embodiments, a deep learning-based structure fracture damage simulation system comprises: a neural network model construction module configured to construct a neural network model and set initial parameters; a neural network model training module configured to optimize the initial parameters of the neural network model based on a sample set and a loss function to obtain a trained neural network model; wherein a sample in the sample set is composed of structure geometric parameters, material parameters, boundary conditions, initial crack information, displacement fields, stress fields, and damage evolution results; the loss function is a loss function considering the partial differential equation residual of the stress tensor, and the unified target of physical consistency, distribution consistency, and prior constraint; A structure fracture damage simulation prediction module is configured to predict the initiation, propagation and bifurcation and coalescence of cracks in the structure domain based on the current structure geometric parameters, material parameters, boundary conditions and initial crack information, and by using the trained neural network model, to obtain the displacement field, stress field and damage evolution results.

[0016] A third aspect of the present application provides an electronic device.

[0017] An electronic device includes a memory, a processor, and a computer program stored on the memory and executable on the processor, and the processor implements the steps of the deep learning-based structure fracture damage simulation method as described above when executing the program.

[0018] Compared with the prior art, the present application has the following advantages: The present application trains the neural network model by using the partial differential equation residual considering the stress tensor, and the loss function of the unified target of physical consistency, distribution consistency and prior constraint, realizes the introduction of multi-scale and multi-physical field information, enhances the adaptability of the neural network model to different working conditions of structure damage, and can ensure that the prediction result is physically reasonable and high-precision; the trained neural network model is used to process the current structure geometric parameters, material parameters, boundary conditions and initial crack information, to simulate and predict the corresponding displacement field, stress field and damage evolution results, to realize the purpose of accurately capturing complex crack behavior (initiation, propagation, bifurcation and coalescence), and to improve the reliability of structure engineering fracture damage simulation prediction. BRIEF DESCRIPTION OF DRAWINGS

[0019] The drawings accompanying the specification of the present application form a part thereof and serve to provide further understanding of the present application, together with the description of the exemplary embodiments of the present application and their description, and do not constitute improper limitations on the present application.

[0020] Figure 1 is a flowchart of the deep learning-based structure fracture damage simulation method of the embodiment of the present application; Figure 2 is a structure schematic diagram of the deep learning-based structure fracture damage simulation system of the embodiment of the present application; Figure 3 is a schematic diagram of an electronic device of the embodiment of the present application. DETAILED DESCRIPTION

[0021] The present application will be further described below in conjunction with the drawings and embodiments.

[0022] It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as generally understood by those skilled in the art to which the present application belongs.

[0023] It is to be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments consistent with the present application. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and / or "comprising," when used in this specification, specify the presence of stated features, steps, operations, elements, components, and / or groups thereof, but do not preclude the presence or addition of one or more other features, steps, operations, elements, components, and / or groups thereof.

[0024] Terminology: The structural member of the present application can be a metal structural member, such as a bolt / screw, a shaft, a special-shaped member, and a welded structural member, etc. The structural member of the present application can also be a composite material structural member, such as a carbon fiber machine cover, a space carbon fiber tube, etc. After the selected structural member, its structural geometric parameters, material parameters, and boundary conditions can be obtained according to the properties of the structural member, and the initial crack information can be obtained by existing detection technologies such as optical microscope observation method and ultrasonic detection method.

[0025] Embodiment one Figure 1 is a flowchart of a structural fracture damage simulation method based on deep learning of the present embodiment, as Figure 1 The structural fracture damage simulation method based on deep learning of the present embodiment can include the following steps S101-S103.

[0026] Among them, the specific implementation process of steps S101-S103 is as follows: Step S101: Construct a neural network model and set initial parameters.

[0027] The input of the neural network model is the structural geometric parameters, material parameters, boundary conditions, and initial crack information; the output is the displacement field, stress field, and damage evolution result. The damage evolution result is the damage variable.

[0028] It should be noted that the specific structure of the neural network model can be selected according to the actual situation.

[0029] Step S102: Based on the sample set and the loss function, the initial parameters of the neural network model are optimized to obtain a trained neural network model; wherein the samples in the sample set are composed of structural geometric parameters, material parameters, boundary conditions, initial crack information, displacement field, stress field and damage evolution result; the loss function is a loss function considering the partial differential equation residual of stress tensor, and the unified target of physical consistency, distribution consistency and prior constraint.

[0030] In the embodiments of the present application, the sample set can be constructed according to experimental data and historical data.

[0031] In the present embodiment, the construction process of the loss function is as follows: Step a: construct the energy functional of the structure based on the criterion of minimizing energy; wherein the energy functional of the structure is obtained by multiplying the tensile energy part and the learnable degradation function representing the automatic adjustment of the fracture degradation rate, and then adding the compressive energy part; The energy functional of the structure is: Wherein, is the structure with damage; is the damage variable, 0 represents perfect, and 1 represents complete fracture failure; is the strain tensor; is the tensile energy part; is the compressive energy part; is the learnable degradation function, which is determined by a neural network model, and can automatically adjust the fracture degradation rate.

[0032] Step b: obtain the discretized energy functional of the structure according to the crack characteristic length of the structure and the effective fracture toughness of the structure.

[0033] Wherein, the crack characteristic length of the structure is determined by the size of the strain gradient and the basic length parameter.

[0034] According to the energy functional of the structure , the crack characteristic length of the structure is determined by the crack characteristic length of the structure and the effective fracture toughness of the structure. ; Wherein, l 0 is the basic length parameter, which is the characteristic length parameter used in classical fracture mechanics, and its value is usually 3-5 times the average size of the mesh near the crack propagation; is a learnable coefficient; is the size of the strain gradient.

[0035] The characteristic length controls the width of the crack zone. When the strain is severe, the crack is more concentrated (i.e. becomes smaller); in the area where the strain is relatively gentle, the crack is more dispersed (i.e. becomes larger).

[0036] Through the neural network model, the spatial gradient (i.e. the spatial variation of the load form) can be learned and relied on, so that the toughness continuously, derivable and self-adaptive transitions between I-type crack (opening-type crack) and II-type crack (shear-type crack) modes, and adapts to shear-tension coupled fracture. The effective fracture toughness of the structure is represented as: ; ; ​​wherein, is the mode I crack fracture toughness; is the mode II crack fracture toughness; is the strain tensor; is the tensile energy fraction; is the compressive energy fraction; is the weight learned by the neural network model.

[0037] It should be noted here that both the mode I crack fracture toughness (opening mode crack) and the mode II crack fracture toughness (slip mode crack) can be characterized by the strain energy release rate. The effective fracture toughness of the structure can also be characterized by the strain energy release rate.

[0038] The mode I crack fracture toughness (opening mode crack) and the mode II crack fracture toughness (slip mode crack) are obtained by standard testing on the corresponding structure. For example, by the end notched flexure method (ENF method), by preloading a layered specimen in three-point bending, measuring the load-displacement curve during crack propagation, and calculating the strain energy release rate.

[0039] Step c: lattice the energy density in the energy functional of the structure into a probability distribution, calculate the second order Wasserstein distance, and then combine the partial differential equation residual of the stress tensor, the irreversible damage constraint, and the square of the difference between the derivative of the discretized energy functional of the damaged structure with respect to time and the external force input power to construct a loss function.

[0040] The energy functional of the structure is discretized, and the expression of the discretized energy functional is as follows: In the formula, is the strain energy per unit after numerical discretization of the structure; is a normalization constant; represents the structure that has been damaged; is the effective fracture toughness; is the crack characteristic length; is the damage variable gradient; is the proportion of tensile cracks in the crack mode. In the engineering application of the phase field fracture model, the value of 8 / 3 is often taken.

[0041] In engineering structures, cracks usually cannot heal themselves after they are generated. To ensure the irreversibility of the cracks, the irreversible damage constraint is as follows: In the formula, is the irreversible damage constraint; ​​is the damage variable; is the historical maximum damage value; is the smoothing parameter; denotes the structure that has occurred damage.

[0042] The energy density of the energy functional of the structure is gridded into a probability distribution, and the differentiable Sinkhorn approximation is used to calculate W2: ; wherein, is the model predicted energy distribution; is the reference energy distribution obtained by experiment or high-precision simulation; is the second-order Wasserstein distance. This method makes the predicted crack dissipation zone distribution consistent with the reference result.

[0043] Specifically, the expression of the loss function is: ; wherein, is the partial differential equation residual operator, is the stress tensor; , , is the weight coefficient; is the external force input power; is the irreversible damage constraint; is the probability distribution; is the derivative with respect to time; is the discretized energy functional; are the displacement field and the damage variable, respectively.

[0044] The first term in the loss function ensures that the mechanical equilibrium is satisfied; the second term ensures that the crack is irreversible; the third term ensures that the energy distribution is aligned; and the fourth term ensures energy conservation. The embodiment introduces distribution level alignment to improve the accuracy of crack patterns.

[0045] It should be noted that the trained model is migrated to similar structures or new load conditions to achieve rapid prediction without complete retraining.

[0046] Step S103: Based on the current structure geometric parameters, material parameters, boundary conditions and initial crack information, the trained neural network model is used to predict the initiation, propagation and bifurcation and merging of cracks in the structure domain to obtain the displacement field, stress field and damage evolution results.

[0047] This embodiment uses the residual of the partial differential equation that considers the stress tensor and a loss function with the unified goal of physical consistency, distribution consistency and prior constraints to train the neural network model, thereby introducing multi-scale and multi-physical field information, enhancing the adaptability of the neural network model to different working conditions of structural damage, and ensuring that the prediction results are physically reasonable and highly accurate; the trained neural network model is used to process the current structural geometric parameters, material parameters, boundary conditions and initial crack information, and simulate and predict the corresponding displacement field, stress field and damage evolution results, thereby achieving the purpose of accurately capturing complex crack behaviors (initiation, extension, bifurcation and merging), and improving the reliability of structural engineering fracture damage simulation prediction.

[0048] Example 2 like Figure 2 As shown, the structural fracture damage simulation system based on deep learning provided in this embodiment can be implemented in a software manner. The structural fracture damage simulation system based on deep learning includes the following software modules: The following is an introduction to the functions of each software module in the deep learning-based structural fracture damage simulation system: A neural network model building module 201 is used to build a neural network model and set initial parameters; A neural network model training module 202 is configured to optimize the initial parameters of the neural network model based on a sample set and a loss function to obtain a trained neural network model; wherein the samples in the sample set are composed of structural geometric parameters, material parameters, boundary conditions, initial crack information, displacement field, stress field, and damage evolution results; and the loss function is a loss function that considers the residual of the partial differential equation of the stress tensor and unifies the objectives of physical consistency, distribution consistency, and prior constraints; The structural fracture damage simulation prediction module 203 is used to predict the initiation, expansion, bifurcation and merging of cracks in the structural domain based on the current structural geometric parameters, material parameters, boundary conditions and initial crack information using the trained neural network model to obtain the displacement field, stress field and damage evolution results.

[0049] Specifically, in the neural network model training module 202, in an embodiment of the present invention, the sample set can be constructed based on experimental data and historical data.

[0050] The structure here can be a metal structural part or other structural parts. Those skilled in the art can make specific settings according to actual conditions, which will not be described in detail here.

[0051] In this embodiment, the process of constructing the loss function is: Step a: Constructing the energy functional of the structure based on the criterion of minimizing energy; wherein the energy functional of the structure is obtained by multiplying the tensile energy part and the learnable degradation function representing the automatically adjusted fracture degradation rate, and then adding the compressive energy part; The energy functional of the structure is: ; wherein, is the structure with damage; is the damage variable, 0 represents perfect, and 1 represents complete fracture failure; is the strain tensor; is the tensile energy part; is the compressive energy part; is the learnable degradation function, which is determined by a neural network model, and can automatically adjust the fracture degradation rate.

[0052] Step b: Obtaining the discretized energy functional of the structure according to the crack characteristic length of the structure and the effective fracture toughness of the structure.

[0053] wherein the crack characteristic length of the structure is determined by the size of the strain gradient and the basic length parameter.

[0054] According to the energy functional of the structure , the crack characteristic length of the structure is determined by performing structure fracture mechanics analysis: ; l 0 is the basic length parameter, which is the characteristic length parameter used in classical fracture mechanics, and its value is usually 3-5 times the average size of the mesh near the crack propagation; is a learnable coefficient; is the size of the strain gradient.

[0055] The characteristic length controls the width of the crack band. When the strain is severe, the crack is more concentrated (i.e. becomes smaller); in the area where the strain is relatively gentle, the crack is more dispersed (i.e. becomes larger).

[0056] Through the neural network model, the spatial gradient (i.e. the spatial variation of the load form) can be learned and relied on, so that the toughness continuously, derivable and self-adaptive transitions between I-type crack (opening-type crack) and II-type crack (shear-type crack) modes, and adapts to shear-tension coupled fracture. The effective fracture toughness of the structure is represented as: ; ; is the I-type crack mode fracture toughness;​​​ is the mode II crack fracture toughness; is the strain tensor; is the tensile energy fraction; is the compressive energy fraction; is the weight learned by the neural network model.

[0057] Step c: discretize the energy density in the energy functional of the structure into a probability distribution, calculate the second-order Wasserstein distance, and combine the partial differential equation residual of the stress tensor, the irreversible damage constraint, and the square of the difference between the derivative of the discretized energy functional of the damaged structure with respect to time and the external force input power to construct the loss function.

[0058] Discretize the energy functional of the structure Discretize the energy functional of the structure The expression of the discretized energy functional is: ; In the formula, is the strain energy per unit after numerical discretization of the structure; is a normalization constant; represents the damaged structure; is the effective fracture toughness; is the crack characteristic length; is the damage variable gradient; is the proportion of tensile cracks in the crack mode. In the engineering application of the phase field fracture model, the value of 8 / 3 is often taken.

[0059] In engineering structures, cracks usually cannot heal themselves after they are generated. To ensure the irreversibility of the cracks, the irreversible damage constraint is as follows: ; In the formula, is the irreversible damage constraint; is the damage variable; is the historical maximum damage value; is the smoothing parameter; represents the damaged structure.

[0060] Discretize the energy density of the energy functional of the structure into a probability distribution, and calculate W2 using the differentiable Sinkhorn approximation: ; In the formula, is the energy distribution predicted by the model; is the reference energy distribution obtained by experiment or high-precision simulation; is the second-order Wasserstein distance. This method keeps the predicted crack dissipation zone distribution relatively consistent with the reference result.

[0061] This method keeps the predicted crack dissipation zone distribution relatively consistent with the reference result.

[0062] Specifically, the expression of the loss function is: ; wherein, is a partial differential equation residual operator, is a stress tensor; 、 、 is a weight coefficient; is an external force input power; is an irreversible damage constraint; is a probability distribution; is a derivative with respect to time; is a discretized energy functional; are displacement field and damage variable respectively.

[0063] The first term in the loss function guarantees to meet the mechanical equilibrium; the second term guarantees the crack irreversibility; the third term guarantees the energy distribution alignment; and the fourth term guarantees the energy conservation. The embodiment introduces the distribution level alignment, thereby improving the accuracy of the crack pattern.

[0064] It should be noted that the trained model is migrated to similar structures or new load conditions to realize rapid prediction without complete retraining.

[0065] The deep learning-based structural fracture damage simulation system trains the neural network model by using the partial differential equation residual considering the stress tensor, and the loss function with unified objectives of physical consistency, distribution consistency and prior constraint, realizes the introduction of multi-scale and multi-physical field information, enhances the adaptability of the neural network model to different working conditions of structural damage, and can guarantee that the prediction result is physically reasonable and high-precision; the trained neural network model is used to process the current structure geometric parameters, material parameters, boundary conditions and initial crack information, simulate and predict the corresponding displacement field, stress field and damage evolution result, realize the purpose of accurately capturing complex crack behavior, and improve the reliability of structural engineering fracture damage simulation and prediction.

[0066] It should be noted that each module in the deep learning-based structural fracture damage simulation system of the embodiment corresponds to each step in the deep learning-based structural fracture damage simulation method in the above embodiment, and the specific implementation process is the same, which will not be repeated here.

[0067] Embodiment three The structure of the electronic device of the embodiment of the present application is described in detail as follows, Figure 3 The schematic diagram of the component structure of the electronic device provided by the embodiment of the present application can be understood, Figure 3 Only the exemplary structure of the electronic device is shown, not all the structures, and the shown part or all of the structures can be implemented as needed.

[0068] The electronic device provided by the embodiment of the present application comprises at least one processor 301, a memory 302, a user interface 303 and at least one network interface 304. The various components in the deep learning-based structure fracture damage simulation system are coupled together through a bus system 305. It can be understood that the bus system 305 is used to realize the connection communication between the components. In addition to including a data bus, the bus system 305 also includes a power bus, a control bus and a status signal bus. However, for the purpose of clear illustration, all the buses are marked as the bus system 305 in Figure 3 .

[0069] The user interface 303 can include a display, a keyboard, a mouse, a trackball, a click wheel, a key, a button, a touchpad or a touch screen, etc.

[0070] It can be understood that the memory 302 can be a volatile memory or a non-volatile memory, and can also include both volatile and non-volatile memories. The memory 302 in the embodiment of the present application can store data to support the operation of the terminal. Examples of these data include any computer programs for operating on the terminal, such as an operating system and an application program. The operating system contains various system programs, such as a framework layer, a core library layer, a driver layer, etc., for implementing various basic services and processing hardware-based tasks. The application program can include various application programs.

[0071] In some embodiments, the deep learning-based structural fracture damage simulation system provided by the embodiments of the present invention can be implemented using a combination of software and hardware. As an example, the deep learning-based structural fracture damage simulation system provided by the embodiments of the present invention can be a processor in the form of a hardware decoding processor, which is programmed to execute the deep learning-based structural fracture damage simulation method provided by the embodiments of the present invention. For example, the processor in the form of a hardware decoding processor can be one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), or other electronic components.

[0072] As an example, the processor 301 can be an integrated circuit chip with signal processing capabilities, such as a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., where the general-purpose processor can be a microprocessor or any conventional processor, etc.

[0073] As an example of hardware implementation of the deep learning-based structural fracture damage simulation system provided in an embodiment of the present invention, the device provided in an embodiment of the present invention can be directly executed by a processor 301 in the form of a hardware decoding processor. For example, the deep learning-based structural fracture damage simulation method provided in an embodiment of the present invention can be implemented by one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), or other electronic components.

[0074] The memory 302 in the embodiment of the present invention is used to store various types of data to support the operation of the structural fracture damage simulation system based on deep learning, or to store data for executing Figure 1program code of the method. Examples of these data include any executable instructions for operating on a deep learning-based structural fracture damage simulation system, such as an executable instruction, a program implementing the deep learning-based structural fracture damage simulation method of the embodiments of the present application can be included in the executable instruction.

[0075] In particular, according to the embodiments of the present application, the processes described above with reference to the flowcharts can be implemented as a computer software program. For example, the embodiments of the present application include a computer program product including a computer program carried on a computer-readable medium, the computer program including instructions for performing the processes described above with reference to the flowcharts. Figure 1 program code of the method. In such embodiments, the computer program can be downloaded and installed from a network by the communication section, and / or installed from a detachable medium. When the computer program is executed by the central processing unit, various functions defined in the apparatus of the present application are executed.

[0076] The present application is described with reference to the flowcharts and / or block diagrams of the methods, apparatus (system) and computer program products of the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of the flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing apparatus to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing apparatus produce a device that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 one flow or multiple flows and / or blocks Figure 1 Figure 1 an apparatus that implements the functions specified in one block or multiple blocks.

[0077] The above merely describes the preferred embodiments of the present application and is not intended to limit the present application. The present application can have various modifications and changes for those skilled in the art. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A deep learning-based structural fracture damage simulation method, characterized by, The method comprises the following steps: constructing a neural network model and setting initial parameters; optimizing the initial parameters of the neural network model based on a sample set and a loss function to obtain a trained neural network model; wherein the samples in the sample set are composed of structural geometric parameters, material parameters, boundary conditions, initial crack information, displacement fields, stress fields and damage evolution results; the loss function is a loss function considering the partial differential equation residual of the stress tensor, and the unified target of physical consistency, distribution consistency and prior constraint; based on the current structural geometric parameters, material parameters, boundary conditions and initial crack information, the trained neural network model is used to predict the initiation, propagation and bifurcation and merging of cracks in the structure domain to obtain the displacement field, stress field and damage evolution result. 2.The deep learning-based structural fracture damage simulation method of claim 1, wherein, The construction process of the loss function is as follows: An energy functional of a structure is constructed based on the criterion of minimizing energy; wherein the energy functional of the structure is obtained by multiplying the tensile energy part and the learnable degradation function representing the automatic adjustment of the fracture degradation speed, and then adding the compression energy part; According to the crack characteristic length of the structure and the effective fracture toughness of the structure, a discretized energy functional of the structure is obtained; The energy density in the energy functional of the structure is gridded into a probability distribution, the second order Wasserstein distance is calculated, and then combined with the partial differential equation residual of the stress tensor, the irreversible damage constraint, and the square of the difference between the derivative of the discretized energy functional of the damaged structure with respect to time and the external force input power, the loss function is constructed. 3.The deep learning-based structural fracture damage simulation method of claim 2, wherein, The crack characteristic length of the structure is determined by the size of the strain gradient and the basic length parameter. 4.The deep learning-based structure fracture damage simulation method of claim 2, wherein, The effective fracture toughness of the structure is represented by the cumulative sum of the weights learned by the neural network model of the mode I crack fracture toughness and the mode II crack fracture toughness. 5.The deep learning-based structural fracture damage simulation method of claim 1 or 2, wherein, The expression of the loss function is: ； wherein, is a partial differential equation residual operator, is a stress tensor; , , is a weight coefficient; is an external force input power; is an irreversible damage constraint; is a probability distribution; is a derivative with respect to time; is a discretized energy functional; are displacement field and damage variable, respectively. 6.The deep learning-based structural fracture damage simulation method of claim 5, wherein, Discretized energy functional The expression for the discretized energy functional is ； In the formula, is the strain energy per unit after numerical discretization of the structure; is a normalization constant; represents the structure that has occurred damage; is the effective fracture toughness; is the crack characteristic length; is the damage variable gradient; is the proportion of tensile cracks in the crack mode. 7.The deep learning-based structural fracture damage simulation method of claim 6, wherein, effective fracture toughness The expression for the effective fracture toughness is: ； ； wherein, is the mode I crack fracture toughness; is the mode II crack fracture toughness; is the strain tensor; is the tensile energy part; is the compressive energy part; is a weight learned by the neural network model. 8.The deep learning-based structural fracture damage simulation method of claim 1, wherein, The expression of the irreversible damage constraint is: ； wherein, is an irreversible damage constraint; is a damage variable; is a historical maximum damage value; is a smoothing parameter; denotes a structure in which damage occurs. 9.A deep learning-based structural fracture damage simulation system, characterized in that, The method comprises the following steps: a neural network model construction module for constructing a neural network model and setting initial parameters; a neural network model training module for optimizing the initial parameters of the neural network model based on a sample set and a loss function to obtain a trained neural network model; wherein the samples in the sample set are composed of structural geometric parameters, material parameters, boundary conditions, initial crack information, displacement fields, stress fields and damage evolution results; the loss function is a loss function considering the partial differential equation residual of the stress tensor, and the unified target of physical consistency, distribution consistency and prior constraint; a structure fracture damage simulation and prediction module for predicting the initiation, propagation and bifurcation and merging of cracks in the structure domain based on the current structural geometric parameters, material parameters, boundary conditions and initial crack information, using the trained neural network model to obtain the displacement field, stress field and damage evolution result.

10. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the program to realize the steps in the deep learning-based structure fracture damage simulation method according to any one of claims 1-8.

Images