A fatigue damage prediction method and system enhanced by a cross-scale model

Abstract

The application provides a fatigue damage prediction method and system enhanced by a cross-scale model, and belongs to the field of material fatigue performance prediction. The method comprises the following steps: constructing a discrete dislocation dynamics model and a cyclic crystal plasticity model, calibrating parameters of the discrete dislocation dynamics model; determining to-be-optimized parameters of the cyclic crystal plasticity model and inversely calculating optimal values of the to-be-optimized parameters to obtain a calibrated cross-scale mechanical model; introducing initial microstructure characteristics of a material into the cross-scale mechanical model to obtain an extended data set containing multi-scale characteristics and fatigue damage results; and constructing a full connection neural network prediction model, training the model based on the extended data set, calculating Shapley values of input characteristics by using a model interpretability tool, and outputting contribution degrees of the multi-scale microstructure characteristics to fatigue damage. The application realizes cross-scale interpretable prediction from a micro mechanism to macro performance, and provides a new direction for precise evaluation of fatigue life and microstructure optimization design of heterogeneous materials.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of material fatigue performance prediction, and particularly relates to a fatigue damage prediction method and system enhanced by a cross-scale model. 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 fields of aerospace, high-end equipment manufacturing, etc., long-life key components under extreme working conditions have put forward strict requirements on the comprehensive performance of materials. Such components are subjected to cyclic loading for a long time, and fatigue damage accumulation is the core bottleneck restricting their repeated use and shortening the service life. For key structural materials with high strength and good toughness, the fatigue performance of the materials under the use environment directly determines the reliability and safety of the equipment.

[0004] To solve the problem of material fatigue performance prediction, various prediction models have been developed in the industry: for rapid engineering evaluation scenarios, stress / strain-life models have been formed, which can quickly give engineering estimation results of fatigue life; for fatigue analysis of components with defects, Paris models and incremental fatigue damage models have been proposed, which can focus on the defect expansion process to carry out damage assessment; in the aspect of micro-mechanism research, the phase field model can realize micro-defect sensitivity analysis, and the crystal plasticity model and multi-scale model provide support for revealing the deformation mechanism at the micro level; with the development of artificial intelligence technology, data-driven or mechanism-driven machine learning models have also been applied to process rapid optimization and fatigue performance prediction, significantly improving the prediction efficiency.

[0005] However, the existing various fatigue performance prediction models still have obvious limitations, which are difficult to meet the precise prediction demand of fatigue performance under extreme working conditions: on the one hand, traditional models generally do not consider the cross-scale internal correlation of microstructure in the process of thermal-mechanical cyclic plastic deformation and damage, and cannot establish a direct link between micro dislocation behavior and macro fatigue performance, making it difficult to reveal the core correlation mechanism of dislocation short-range reaction and low-cycle fatigue performance; on the other hand, data-driven machine learning models rely on a large number of sample data, which are difficult to clarify the complex action relationship of cross-scale microstructure under limited sample conditions, while mechanism-driven machine learning models only model the physical mechanism at a single scale, lacking the ability to fully describe the cross-scale deformation mechanism, and the model's explainability and prediction accuracy are limited. SUMMARY

[0006] To overcome the shortcomings of the prior art, the present application provides a fatigue damage prediction method and system enhanced by a cross-scale model, which can realize high-fidelity prediction of low-cycle fatigue damage, quantify the contribution of multi-scale microstructure to low-cycle fatigue damage, and reveal the correlation mechanism of dislocation short-range reaction and low-cycle fatigue damage.

[0007] To achieve the above object, one or more embodiments of the present application provide the following technical solutions:

[0008] The first aspect of the present application provides a fatigue damage prediction method enhanced by a cross-scale model.

[0009] A fatigue damage prediction method enhanced by a cross-scale model, comprising:

[0010] A discrete dislocation dynamics model and a cyclic crystal plasticity model are constructed, and the parameters of the discrete dislocation dynamics model are calibrated; the parameters to be optimized of the cyclic crystal plasticity model are determined, and the optimal values of the parameters to be optimized are inverted to obtain a calibrated cross-scale mechanical model.

[0011] The initial microstructure characteristics of the material are introduced into the calibrated cross-scale mechanical model to obtain an extended data set containing multi-scale characteristics and fatigue damage results.

[0012] A fully connected neural network prediction model is constructed, taking multi-scale microstructure characteristics, process parameters and external working conditions as inputs and taking low-cycle fatigue damage stress or life as output; the prediction model is trained based on the extended data set, and the Shapley value of the input characteristics is calculated using a model interpretability tool to output the quantitative contribution of the multi-scale microstructure characteristics to fatigue damage.

[0013] As a further technical solution, the discrete dislocation dynamics model is constructed, and the parameters of the discrete dislocation dynamics model are calibrated, comprising:

[0014] The parameters of the discrete dislocation dynamics model are obtained, and the parameter types are divided into dislocation slip parameters, interaction parameters, diffusion parameters and lattice parameters.

[0015] The dislocation density evolution law is extracted through the nanoindentation load-displacement curve, and the critical stress of dislocation slip, dislocation spacing and interaction parameters are back calculated using the stress-Burgers vector relationship formula.

[0016] The dislocation interaction, dislocation density and yield strength predicted by the discrete dislocation dynamics simulation are verified in combination with the dislocation interaction motion represented by TEM and the stress-strain relationship of the indentation experiment.

[0017] As a further technical solution, the parameters to be optimized of the cyclic crystal plasticity model are determined, and the optimal values of the parameters to be optimized are inverted to obtain a calibrated cross-scale mechanical model, comprising:

[0018] The parameters to be optimized of the cyclic crystal plasticity model and their value ranges are determined, and the parameters to be optimized include a dislocation multiplication coefficient and a damage evolution parameter.

[0019] The construction volume represents a unit model, and a cyclic load is applied for simulation to extract macro mechanical response data; the macro mechanical response data includes yield stress and final stress;

[0020] A machine learning model is trained based on the to-be-optimized parameters and the macro mechanical response data, a macro stress-strain curve is taken as an optimization target, an optimization algorithm is used to iteratively optimize the trained machine model, and an optimal value combination of the to-be-optimized parameters that minimizes the error between the simulation response and the experimental target is inversely calculated;

[0021] The optimal value combination is substituted into the cyclic crystal plasticity model to obtain a calibrated cross-scale mechanical model.

[0022] As a further technical solution, the initial microstructure features of the material are introduced into the calibrated cross-scale mechanical model to obtain an extended data set containing multi-scale features and fatigue damage results, including:

[0023] In combination with macro and micro experimental characterization, the initial microstructure features are introduced into the RVE model to obtain a wide-area RVE model;

[0024] According to the calibrated cross-scale mechanical model parameters, the evolution processes of dislocation short-range reactions, slip system activation and evolution and damage factor evolution of the heterogeneous material under the action of cyclic loads under different stress states and temperatures are simulated, the correlation mechanism between micro dislocation behavior and macro low-cycle fatigue damage is revealed, and an extended data set containing multi-scale features and fatigue damage results is obtained.

[0025] As a further technical solution, the initial microstructure features include at least one of an initial dislocation configuration, a grain size, a micro-orientation and a heterogeneous phase distribution.

[0026] As a further technical solution, the Shapley value of the input features is calculated by using a model interpretability tool, specifically: the SHAP tool is used to calculate the Shapley value of the calibrated parameters on the macro yield stress and the final stress, the marginal contribution of each parameter is analyzed, and the parameters are adjusted according to the contribution analysis result and then substituted into the model for verification.

[0027] The second aspect of the present application provides a fatigue damage prediction system enhanced by a cross-scale model.

[0028] A fatigue damage prediction system enhanced by a cross-scale model, comprising:

[0029] The parameter calibration module is configured to: construct a discrete dislocation dynamics model and a cyclic crystal plasticity model, calibrate the parameters of the discrete dislocation dynamics model; determine the to-be-optimized parameters of the cyclic crystal plasticity model and inversely calculate the optimal values of the to-be-optimized parameters to obtain a calibrated cross-scale mechanical model;

[0030] The mechanism analysis module is configured to: import initial microstructure characteristics of the material into the calibrated cross-scale mechanical model to obtain an extended data set containing multi-scale characteristics and fatigue damage results;

[0031] The contribution evaluation module is configured to: construct a fully connected neural network prediction model with multi-scale microstructure characteristics, process parameters and external working conditions as inputs and low-cycle fatigue damage stress or life as outputs; train the prediction model based on the extended data set, calculate the Shapley value of the input characteristics using a model interpretability tool, and output the quantitative contribution of the multi-scale microstructure characteristics to the fatigue damage.

[0032] The third aspect of the present application provides a computer readable storage medium having a program stored thereon, the program being executed by a processor to implement the steps of the cross-scale model enhanced fatigue damage prediction method according to the first aspect of the present application.

[0033] The fourth aspect of the present application provides an electronic device comprising a memory, a processor and a program stored on the memory and executable on the processor, wherein the processor executes the program to implement the steps of the cross-scale model enhanced fatigue damage prediction method according to the first aspect of the present application.

[0034] The fifth aspect of the present application provides a computer program product comprising computer programs / instructions, which are executed by a processor to implement the steps of the cross-scale model enhanced fatigue damage prediction method according to the first aspect of the present application.

[0035] The above one or more technical solutions have the following beneficial effects:

[0036] (1) The present application divides the multi-type key parameters of the discrete dislocation dynamics model, combines the nanoindentation experiment, TEM characterization and stress-Burgers vector relationship formula to complete the basic parameter backstepping and verification, and then generates sufficient sample data by using the RVE model, trains by using the SVR and FFNN double machine learning models, and couples the genetic algorithm to realize efficient solution of the optimized parameters of the cyclic crystal plasticity model. At the same time, the SHAP tool is used to quantify the parameter marginal effect, and the macro stress-strain curve is used for verification, which significantly improves the efficiency and accuracy of the cross-scale model parameter calibration, solves the problems of complex and low precision in the traditional parameter calibration process, and provides a high credibility model basis for low-cycle fatigue damage prediction.

[0037] (2) Relying on macro-micro experimental characterization to construct a wide-area RVE model containing macro-micro crystallographic information of the heterogeneous zone, based on the calibrated cross-scale mechanical model parameters, the dislocation behavior law, slip system evolution and damage factor change in the cyclic plastic deformation process of the material under different working conditions and multi-scale microstructure parameters are systematically analyzed. The internal correlation between micro-zone dislocation short-range reaction and low-cycle fatigue damage under variable temperature and complex stress state is clarified; a cross-scale mechanism-driven fully connected neural network model is constructed, which comprehensively incorporates working condition information, complex microstructure information and LP-DED process parameters, and the Shapley value of each parameter is accurately calculated by SHAP method, the marginal effect and contribution ratio of micro dislocation configuration, meso-grain size / orientation and process parameters on low-cycle fatigue damage are clearly quantified, the "black box" problem of traditional machine learning model is solved, and the model is enhanced. The explainability provides a clear quantitative basis for the microstructure design, process optimization and fatigue life improvement of heterogeneous Cu / Ni alloy.

[0038] Advantages of the additional aspects of the application will be in part apparent from the following description, will in part be apparent from the following description, or will be learned by practice of the application. BRIEF DESCRIPTION OF DRAWINGS

[0039] The accompanying drawings, which form a part of this specification, are included to provide a further understanding of the application, and are incorporated by reference herein. The embodiments illustrated in the drawings are presented by way of example in

[0040] Figure 1 A method flowchart for a first embodiment.

[0041] Figure 2 A system structure diagram for a second embodiment. DETAILED DESCRIPTION

[0042] It should be noted that the following detailed description is merely exemplary and is intended to provide further description of the application. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the application belongs.

[0043] It should be noted that the terms used herein are merely intended to describe specific embodiments and are not intended to limit the exemplary embodiments according to the present application.

[0044] The embodiments in the present application and the features in the embodiments can be combined with each other without conflict.

[0045] Embodiment one

[0046] The embodiment discloses a fatigue damage prediction method enhanced by a cross-scale model, realizes high-fidelity prediction of low-cycle fatigue damage and quantitative characterization of multi-scale microstructure contribution by constructing a cross-scale mechanical model and combining a machine learning technology. The method comprises the following steps: efficient parameter calibration and verification of the cross-scale model, analysis of the correlation mechanism of short-range dislocation reaction and low-cycle fatigue damage, and evaluation of the multi-scale microstructure contribution based on the cross-scale mechanism driven machine learning model. Through the discrete dislocation-crystal plasticity and machine learning dual-drive model, the internal correlation of the cross-scale microstructure is clarified, the model interpretability is enhanced, and a new effective way is provided for low-cycle fatigue life prediction of heterogeneous alloys.

[0047] Specifically, as shown in the figure, a fatigue damage prediction method enhanced by a cross-scale model comprises the following steps: Figure 1

[0048] In step S1, a discrete dislocation dynamics model and a cyclic crystal plasticity model are constructed, and the parameters of the discrete dislocation dynamics model are calibrated; the to-be-optimized parameters of the cyclic crystal plasticity model are determined, and the optimal value of the to-be-optimized parameters is inversely calculated to obtain a calibrated cross-scale mechanical model.

[0049] The discrete dislocation dynamics model and the cyclic crystal plasticity model are respectively built, and the discrete dislocation dynamics model is used to describe the sliding, interaction and other behaviors of micro dislocations. The construction process comprises the following steps: at the micro scale, the dislocation is represented as a discrete line segment unit, and the sliding, interaction and evolution behaviors of the dislocation under an applied stress field and a dislocation interaction stress field are explicitly considered; the dislocation movement speed is determined by the difference between the resolved shear stress and the critical sliding stress, and the model comprehensively considers mechanisms such as dislocation pile-up, cross-sliding, annihilation and Frank-Read source activation. The total length of the dislocation line in the calculation domain is calculated by statistical calculation to obtain the dislocation density as a micro evolution parameter.

[0050] The construction of the cyclic crystal plasticity model comprises the following steps: at the mesoscale, a cyclic crystal plasticity constitutive model is introduced, the total strain is decomposed into an elastic strain and a plastic strain, the plastic strain is obtained by superimposing the shear strain of each slip system, and the slip rate is controlled by the resolved shear stress and the critical shear stress. The critical shear stress is associated with the dislocation density through the Taylor relationship, so that the micro dislocation evolution information is introduced into the crystal plasticity model. Meanwhile, a continuous damage variable is introduced into the cyclic crystal plasticity model, the damage evolution equation is driven by the cyclic plastic strain amplitude, and the dislocation density is used as a damage acceleration factor.

[0051] In the above manner, the dislocation density output by the discrete dislocation dynamics model is coupled into the cyclic crystal plasticity model as an input parameter, the two models together form a cross-scale mechanical model, and the micro dislocation behavior and the macro cyclic plasticity and damage accumulation are cooperatively described.

[0052] ​Subsequently, the discrete dislocation dynamics model parameters are calibrated, and the discrete dislocation dynamics model parameters are obtained, and the parameter types are divided into dislocation slip parameters, interaction parameters, diffusion parameters and lattice parameters;

[0053] Through the nanoindentation load-displacement curve, the evolution law of dislocation density is extracted, and the critical stress of dislocation slip, dislocation spacing and interaction parameters are inversely deduced by using the stress-Burgers vector relationship formula.

[0054] Firstly, the hardness H of the material is calculated according to the indentation load P and the contact area A, and the hardness is converted into the equivalent flow stress by using the Tabor relationship. Subsequently, the quantitative relationship between the flow stress and the dislocation density is established based on the Taylor relationship, and the evolution law of the dislocation density is inversely deduced by the stress level corresponding to different indentation depths. Further, the average dislocation spacing can be represented by the reciprocal square root of the dislocation density, and the critical stress of dislocation slip is calculated accordingly. At the same time, by analyzing the change of the hardening slope in the indentation curve, the dislocation interaction coefficient and the dislocation proliferation related parameters can be inversely deduced.

[0055] The above process realizes the quantitative inversion from the nanoindentation experimental results to the dislocation slip, dislocation spacing and interaction parameters.

[0056] Subsequently, combined with the dislocation interaction motion represented by TEM and the stress-strain relationship of the indentation experiment, the dislocation interaction, dislocation density and yield strength predicted by the discrete dislocation dynamics simulation are verified, which is used to calibrate the dislocation dynamics parameters and ensure the reliability and robustness of the discrete dislocation dynamics model, specifically:

[0057] Firstly, the initial dislocation density, dislocation configuration and main dislocation type of the material are determined according to the TEM observation results, and the initial conditions of the model are given by combining the lattice constant and Burgers vector and other material intrinsic parameters. Subsequently, the nanoindentation loading process is simulated in the discrete dislocation dynamics model, and the dislocation density evolution, dislocation interaction morphology and corresponding simulated load-displacement curve are obtained. The simulation results are compared with the experimental results: on the one hand, the dislocation pile-up, interaction and annihilation behavior observed in TEM are compared to verify the prediction ability of the model for the dislocation evolution mechanism; on the other hand, the simulated load-displacement curve is fitted with the experimental curve to verify the prediction accuracy of the model for the yield behavior and hardening characteristics. By minimizing the error between the simulation results and the experimental results, the dislocation mobility, interaction coefficient and annihilation parameters are iteratively corrected. When the model can stably reproduce the experimental trend under different loading conditions, the calibration of the discrete dislocation dynamics model parameters is completed, thereby ensuring the reliability and robustness of the model. For example, the force-displacement curve of the indentation experiment is used to verify whether the parameters of the discrete dislocation dynamics simulation are appropriate.

[0058] For the cyclic crystal plasticity model, the elastic modulus, Poisson's ratio and the size of the Burgers vector are determined by the experimental curve or the intrinsic properties of the material. And the model parameters related to the evolution of dislocation density, such as dislocation multiplication coefficient, activation energy coefficient, dislocation crossing obstacle frequency, viscous coefficient, etc., and the damage evolution related parameters damage factor are obtained. The two types of parameters are taken as the optimization parameters and the value variation range is determined.

[0059] A volume representative element (RVE) model (size 1mm 3mm 3mm) containing a geometric shape of tensile-shear stress state is established, the grain orientation

[001] is taken as the cyclic tensile loading direction, the number of elements and the type of elements are determined, the cyclic load of strain control is applied, and the strain rate is consistent with the in-situ cyclic experiment. The yield stress and the final stress are extracted as the output quantities by using the established cross-scale mechanical model. The simulation process is repeated multiple times to generate 1000 groups of data for subsequent training and verification of the machine learning model.

[0060] Based on the rational analysis of the data set, machine learning models such as support vector regression (SVR) model and feedforward neural network (FFNN) are used as the machine learning model to be trained, the optimization parameters of the cyclic crystal plasticity model are selected as the input, and the yield stress and the final stress are selected as the output. Among them, the SVR adopts a polynomial kernel function, the FFNN adopts a mean square error as a loss function, the coefficient of determination R2 is used to describe the regression effect, the learning rate is 0.001, and the training period is 100. All data are 80% as a training set and 20% as a test set. The SVR and FFNN models are trained, and the model performance is evaluated by mean square error and R2;

[0061] Based on the trained machine learning model, genetic algorithm is used to determine the parameters of the cross-scale mechanical model. Specifically, the support vector regression model or the feedforward neural network model is used to establish the mapping relationship between the parameters of the cyclic crystal plasticity model and the macroscopic mechanical response (yield stress, final stress), so as to construct a proxy model to replace the high computational cost of traditional finite element or crystal plasticity model in the optimization process.

[0062] On this basis, the genetic algorithm is introduced to perform global search on the optimization parameters, wherein the individual code of the genetic algorithm corresponds to the dislocation multiplication coefficient and the damage evolution parameters in the cyclic crystal plasticity model, and the fitness function is composed of the error between the prediction results of the proxy model and the experimental target curve. The iteration number of the genetic algorithm is set to 500, the population size is 1000, the crossover rate and the mutation rate are 60% and 10% respectively, the result of the evaluation formula is taken as the error, and the optimization parameters of the model are solved, and the convergence speed and the final accuracy of the genetic algorithm enhanced by the machine learning model are analyzed.

[0063] With the help of SHAP tool, the Shapley values of the model parameters under the yield stress and the final stress are calculated by the machine learning model, and the marginal effect of the optimized model parameters on the yield stress and the final stress is analyzed. According to the feature analysis result, the corresponding parameters are changed, and the yield stress and the final stress are simulated by bringing the changed parameters into the cyclic crystal plasticity model. The macro stress-strain curve is compared and verified, and the calibrated parameters are further corrected.

[0064] Step S2, the initial microstructure characteristics of the material are introduced into the calibrated cross-scale mechanical model to obtain an extended data set containing multi-scale characteristics and fatigue damage results;

[0065] Combined with macro and micro experimental characterization, the initial dislocation configuration, microstructure morphology, micro-orientation and other information are introduced into the RVE model to ensure that the experimental initial state and the model initial value are consistent, and a wide-area RVE model considering heterogeneous zone macro and micro crystallographic information is established. Specifically: first, according to the macro and micro experimental characterization results, the initial dislocation density, grain size distribution, grain orientation and heterogeneous phase space distribution are parameterized described, and they are mapped as initial state variables in the volume representative element (RVE) model, wherein the grain orientation is assigned by orientation matrix or Euler angle form, and the dislocation density is introduced as a slip system related internal state variable.

[0066] Subsequently, the boundary conditions and loading paths consistent with the experiment are established in the RVE model, and the above initial microstructure parameters are input into the cross-scale mechanical model as the model initial conditions, so that the discrete dislocation dynamics model and the cyclic crystal plasticity model are consistent at the initial time.

[0067] Under the action of cyclic load, the model is calculated by time step or cycle step, the discrete dislocation dynamics model is used to update the dislocation density and dislocation interaction behavior, and the cyclic crystal plasticity model is used to calculate the slip system activation, plastic strain accumulation and damage variable evolution based on the updated dislocation density. Finally, the macro stress response and fatigue damage results under the corresponding working condition are output.

[0068] According to the above calibrated cross-scale mechanical model parameters, under different strain amplitudes, stress states, normal / high temperatures, the cross-scale mechanical model is used to analyze the absorption, penetration, jamming law of the dislocation and grain boundary phase interface, the evolution of slip system activation, the distribution of strain shear band and the evolution law of low cycle fatigue damage factor in the cyclic plastic deformation process of Cu / Ni heterogeneous zone under different grain sizes, phase sizes and initial micro-orientation. Finally, an extended data set containing multi-scale characteristics and fatigue damage results is obtained, specifically:

[0069] After the initial microstructure characteristics and external working conditions are given, the cross-scale mechanical model first calls the discrete dislocation dynamics model to calculate the slip, pile-up, interaction and evolution process of dislocations under cyclic loading at the microscale, and obtains the dislocation density and dislocation configuration information with the evolution of the cycle.

[0070] Then, the dislocation density is passed as an internal state variable to the cyclic crystal plasticity model to update the critical shear stress and plastic slip rate of each slip system, so as to calculate the plastic strain response and cyclic hardening or softening behavior at the mesoscale. At the same time, based on the cyclic plastic strain amplitude and dislocation density evolution, a damage evolution equation is introduced to update the fatigue damage variable.

[0071] By repeating the above calculation process under different strain amplitude, stress state and temperature conditions, the cross-scale mechanical model can output the macroscopic stress-strain response and low-cycle fatigue damage results under the corresponding working conditions, and further form an extended data set containing multi-scale characteristics and fatigue damage results.

[0072] Step S3, a full connection neural network prediction model is constructed with multi-scale microstructure characteristics, process parameters and external working conditions as input and low-cycle fatigue damage stress or life as output; the prediction model is trained based on the extended data set, the Shapley value of the input features is calculated using the model interpretability tool, and the quantitative contribution of the multi-scale microstructure characteristics to the fatigue damage is output.

[0073] Based on the full connection neural network model, the low-cycle fatigue damage stress is taken as the output, and the different working conditions (strain amplitude, stress state), complex microstructure information (grain size / orientation, dislocation configuration) and LP-DED process (laser power, scanning speed) are considered in the input. Among them, the feature data set of grain size / orientation and dislocation configuration is derived from the simulation of the cross-scale mechanical model. The hyperparameters and training settings (network structure, optimizer, regularization method, loss function and learning rate) of the full connection neural network are set, and then the model performance is evaluated by mean square error and R2;

[0074] The full connection neural network model is explained based on the calculated Shapley value, the Shapley value of the dislocation configuration, grain size / orientation and LP-DED process parameters under the low-cycle fatigue damage stress is calculated by the full connection neural network model, the marginal effect of the dislocation configuration, grain size / orientation and LP-DED process parameters on the low-cycle fatigue damage is analyzed, and then the contribution of the microstructure to the low-cycle fatigue damage is quantified.

[0075] Embodiment two

[0076] The embodiment discloses a fatigue damage prediction system enhanced by a cross-scale model;

[0077] As shown in Figure 2 A cross-scale model enhanced fatigue damage prediction system comprises:

[0078] A parameter calibration module is configured to construct a discrete dislocation dynamics model and a cyclic crystal plasticity model, calibrate parameters of the discrete dislocation dynamics model, determine to-be-optimized parameters of the cyclic crystal plasticity model, and inverse the optimal values of the to-be-optimized parameters to obtain a calibrated cross-scale mechanical model.

[0079] A mechanism analysis module is configured to import initial microstructure features of a material into the calibrated cross-scale mechanical model to obtain an extended data set containing multi-scale features and fatigue damage results.

[0080] A contribution evaluation module is configured to construct a fully connected neural network prediction model taking multi-scale microstructure features, process parameters and external working conditions as inputs and taking low-cycle fatigue damage stress or life as output, train the prediction model based on the extended data set, calculate Shapley values of input features by using a model explainability tool, and output quantitative contributions of multi-scale microstructure features to fatigue damage.

[0081] Embodiment three

[0082] The purpose of the embodiment is to provide a computer readable storage medium.

[0083] A computer readable storage medium has a computer program stored thereon, the program being executed by a processor to implement the steps in the cross-scale model enhanced fatigue damage prediction method of embodiment one.

[0084] Embodiment four

[0085] The purpose of the embodiment is to provide an electronic device.

[0086] An electronic device comprises a memory, a processor, and a program stored on the memory and executable on the processor, the processor executing the program to implement the steps in the cross-scale model enhanced fatigue damage prediction method of embodiment one.

[0087] Embodiment five

[0088] Embodiment five of the present application provides a computer program product comprising computer programs / instructions, which are executed by a processor to implement the steps in the cross-scale model enhanced fatigue damage prediction method of embodiment one.

[0089] The steps involved in the apparatuses of embodiments two, three, four and five above correspond to the method of embodiment one, and the specific implementation can be seen from the relevant description of embodiment one. The term "computer readable storage medium" should be understood as including a single medium or multiple media of one or more instruction sets; it should also be understood as including any medium capable of storing, encoding or carrying a set of instructions for execution by a processor and causing the processor to perform any of the methods in the present application.

[0090] Those skilled in the art should understand that each module or step of the present application described above can be realized by a general computer device, or alternatively, they can be realized by program codes executable by a computing device, so that they can be stored in a storage device for execution by a computing device, or they can be respectively manufactured into individual integrated circuit modules, or a plurality of modules or steps among them can be manufactured into a single integrated circuit module to realize. The present application is not limited to any specific combination of hardware and software.

[0091] Although the specific embodiments of the present application are described above in combination with the drawings, the description is not a limitation on the scope of protection of the present application, and those skilled in the art should understand that various modifications or variations made by those skilled in the art on the basis of the technical solutions of the present application without creative labor are still within the scope of protection of the present application.

Claims

1. A method for fatigue damage prediction enhanced by a cross-scale model, characterized in that, The method comprises the following steps: A discrete dislocation dynamics model and a cyclic crystal plasticity model are constructed, and parameters of the discrete dislocation dynamics model are calibrated, including: obtaining the parameters of the discrete dislocation dynamics model, and dividing the parameter types into dislocation slip parameters, interaction parameters, diffusion parameters and lattice parameters; the dislocation density evolution law is extracted from the nanoindentation load-displacement curve, and the critical stress of dislocation slip, the dislocation spacing and the interaction parameters are inversely deduced by using the stress-Burgers vector relationship formula; the dislocation interaction, dislocation density and yield strength predicted by the discrete dislocation dynamics simulation are verified in combination with the dislocation interaction motion represented by TEM and the stress-strain relationship of the indentation experiment; The parameters to be optimized of the cyclic crystal plasticity model are determined, and the optimal values of the parameters to be optimized are inversely calculated to obtain a calibrated cross-scale mechanical model, specifically: the parameters to be optimized of the cyclic crystal plasticity model and the value ranges of the parameters are determined, the parameters to be optimized include a dislocation multiplication coefficient and a damage evolution parameter; a volume representative element model is constructed, and a cyclic load is applied for simulation, and macro mechanical response data are extracted; the macro mechanical response data include a yield stress and a final stress; a machine learning model is trained based on the parameters to be optimized and the macro mechanical response data, the macro stress-strain curve is taken as an optimization target, and an optimization algorithm is used to iteratively optimize the trained machine model, and the optimal value combination of the parameters to be optimized that makes the simulation response and the experimental target error minimum is inversely calculated; the optimal value combination is substituted into the cyclic crystal plasticity model to obtain the calibrated cross-scale mechanical model; Initial microstructure characteristics of the material are introduced into the calibrated cross-scale mechanical model to obtain an extended data set containing multi-scale characteristics and fatigue damage results. A full connection neural network prediction model is constructed, taking multi-scale microstructure characteristics, process parameters and external working conditions as inputs and taking low cycle fatigue damage stress or life as output; the prediction model is trained based on the extended data set, and the Shapley value of the input characteristics is calculated by using a model interpretability tool, and the quantitative contribution of the multi-scale microstructure characteristics to the fatigue damage is output.

2. The cross-scale model enhanced fatigue damage prediction method of claim 1, wherein, Initial microstructure characteristics of the material are introduced into the calibrated cross-scale mechanical model to obtain an extended data set containing multi-scale characteristics and fatigue damage results, including: In combination with macro and micro experimental characterization, initial microstructure characteristics are introduced into an RVE model to obtain a wide-area RVE model; According to the calibrated cross-scale mechanical model parameters, the short-range reaction of dislocations, the activation and evolution of slip systems and the evolution process of damage factors of the heterogeneous material under the action of cyclic load under different stress states and temperatures are simulated, the correlation mechanism between micro dislocation behavior and macro low cycle fatigue damage is revealed, and an extended data set containing multi-scale characteristics and fatigue damage results is obtained.

3. The cross-scale model enhanced fatigue damage prediction method of claim 1, wherein, The initial microstructure characteristics include at least one of an initial dislocation configuration, a grain size, a micro-orientation and a heterogeneous phase distribution.

4. The cross-scale model enhanced fatigue damage prediction method of claim 1, wherein, The Shapley values of the input features are calculated by using a model explainability tool, specifically, the SHAP tool is used to calculate the Shapley values of the calibrated parameters for the macro yield stress and the final stress, the marginal contribution of each parameter is analyzed, and the parameters are adjusted according to the contribution analysis result and then are input into the model for verification.

5. A fatigue damage prediction system enhanced by a cross-scale model, characterized by, The method comprises the following steps: The parameter calibration module is configured to: construct a discrete dislocation dynamics model and a cyclic crystal plasticity model, calibrate the parameters of the discrete dislocation dynamics model, including: obtaining the parameters of the discrete dislocation dynamics model, dividing the parameter types into dislocation slip parameters, interaction parameters, diffusion parameters and lattice parameters; extracting the dislocation density evolution law through the nanoindentation load-displacement curve, and inversely calculating the critical stress of dislocation slip, dislocation spacing and interaction parameters by using a stress-Burgers vector relationship formula; combining the dislocation interaction movement represented by TEM and the stress-strain relationship represented by the indentation experiment to verify the dislocation interaction, dislocation density and yield strength predicted by the discrete dislocation dynamics simulation; The method further comprises the following steps: determining the to-be-optimized parameters of the cyclic crystal plasticity model and inversely calculating the optimal values of the to-be-optimized parameters to obtain a calibrated cross-scale mechanical model, specifically: determining the to-be-optimized parameters of the cyclic crystal plasticity model and the value ranges of the to-be-optimized parameters, the to-be-optimized parameters including a dislocation proliferation coefficient and a damage evolution parameter; constructing a volume representative unit model and applying a cyclic load for simulation to extract macro mechanical response data; the macro mechanical response data including a yield stress and a final stress; training a machine learning model based on the to-be-optimized parameters and the macro mechanical response data, taking the macro stress-strain curve as an optimization target, and iteratively optimizing the trained machine model by using an optimization algorithm to inversely calculate an optimal value combination of the to-be-optimized parameters that makes the simulation response and the experimental target have the minimum error; and substituting the optimal value combination into the cyclic crystal plasticity model to obtain the calibrated cross-scale mechanical model; The mechanism analysis module is configured to: input the initial microstructure features of the material into the calibrated cross-scale mechanical model to obtain an extended data set containing multi-scale features and fatigue damage results; The contribution evaluation module is configured to: construct a full connection neural network prediction model taking multi-scale microstructure features, process parameters and external working conditions as inputs and taking low cycle fatigue damage stress or life as an output; train the prediction model based on the extended data set, calculate the Shapley values of the input features by using a model explainability tool, and output the quantitative contribution of the multi-scale microstructure features to the fatigue damage.

6. A computer-readable storage medium having stored thereon a program, characterized in that, The program is executed by the processor to implement the steps in the fatigue damage prediction method of the cross-scale model enhancement according to any one of claims 1-4.

7. An electronic device comprising a memory, a processor, and a program stored in the memory and executable on the processor, characterized by The processor executes the program to implement the steps in the fatigue damage prediction method of the cross-scale model enhancement according to any one of claims 1-4.

8. A computer program product comprising computer programs / instructions, characterized in that, The computer program / instruction is executed by the processor to implement the steps in the fatigue damage prediction method of the cross-scale model enhancement according to any one of claims 1-4.

Images