A method, device and medium for establishing a crack growth model based on data drive

Abstract

The present application discloses a method, device and medium for establishing a crack propagation model based on data-driven, which relates to the field of fatigue safety design of aviation structures. The method includes: obtaining multi-source crack propagation experimental data and classifying the data; obtaining an initial crack propagation model containing unknown correction terms; randomly constructing multiple functional expressions of basis functions in the correction terms according to preset operators and terminators; substituting the constructed basis functions into the initial crack propagation model to obtain alternative crack propagation models; constructing the initial crack propagation model population with the alternative crack propagation models; evolving the individuals in the initial crack propagation model population using gene crossover and mutation behavior, fitting the model parameters of each individual according to the crack propagation experimental data of each classification, and calculating the evaluation index of each individual to screen out the optimal crack propagation model. The present invention can reliably learn a unified crack propagation model that meets the requirements based on multi-source crack propagation experimental data.

Description

Technical Field

[0001] The present application relates to the field of fatigue safety design of aviation structures, and in particular to a method, device and medium for establishing a crack propagation model based on data-driven. Background Art

[0002] Crack propagation analysis plays an important role in the fatigue safety design of aviation structures. It helps determine the remaining life and safety inspection cycle of structures by predicting how cracks in structures gradually propagate under repeated loads during long-term use. Among them, a reliable crack propagation model is an important basis for crack propagation analysis. It is used to reflect the crack propagation rate of cracked structures under cyclic loads. One of the milestone examples is the introduction of the Paris model: ( is the stress intensity factor range, represents the extension of the crack front after a single load cycle, C and m are material correlation coefficients), which can describe the change in crack growth rate in different experiments from the initial size to the critical size through a simple single model, and to a considerable extent reveals the inherent similarity of complex crack growth behavior in the structure. Based on this unified model, engineers can more reliably analyze and predict the crack growth behavior in the structure, which greatly promotes the application of the structural damage tolerance design concept and improves the fatigue safety of metal structures in service.

[0003] However, the crack growth behavior in real structures is complex, not only showing significant differences in different stages, but also related to many factors such as crack type, loading method, material state and environmental conditions. In view of the limitations of the Paris law, scholars have proposed various improved models. For example, by introducing asymptotic terms for the slow crack growth stage and the fast crack growth stage to more comprehensively reflect the crack growth rate characteristics at different stages, in order to reflect the crack closure effect caused by the plastic deformation of the material, Establish corrections, and propose a variety of improved models to unify the small crack extension and long crack extension stages. In particular, there are also some improved models specifically established for factors that affect crack extension. For example, in order to consider the influence of crack type, a correction model for corner cracks / face cracks is established in the relevant technology; in order to consider the influence of load, the walker model introduces the influence of stress ratio, and the Willenborg model considers the influence of load interaction; in order to consider the difference in crack closure effect caused by material differences, a series of correction models are proposed in the relevant technology; in order to consider environmental factors, the relevant technology establishes corresponding correction models based on the Forman model, etc.

[0004] Although researchers at home and abroad have carried out extensive work on the construction of crack propagation models, there are still challenges when these models are used for crack propagation behavior analysis: on the one hand, most of the improved models based on the Paris model usually only improve specific features and cannot fully reflect various factors such as crack type, loading method, material state and environmental conditions, so the generalization of the model is challenging. On the other hand, although combined models such as NASGRO have better versatility by comprehensively considering various factors affecting crack propagation, they contain more than ten parameters in order to cover a variety of crack propagation characteristics. Most of these parameters lack clear physical meanings and require rich engineering experience to determine, which has considerable application difficulties. Therefore, to date, many scholars are still exploring the establishment of a more concise but universal unified crack propagation model to provide support for efficient and reliable crack propagation analysis. Summary of the invention

[0005] The purpose of this application is to provide a data-driven method, device and medium for establishing a crack propagation model, which can construct a concise and universal unified crack propagation model.

[0006] To achieve the above objectives, this application provides the following solutions.

[0007] In a first aspect, the present application provides a method for establishing a crack propagation model based on data-driven, comprising the following steps.

[0008] Acquire multi-source crack extension experimental data and perform data classification.

[0009] An initial crack growth model is obtained; the initial crack growth model includes a preset classical crack growth model and preset correction items; the preset correction items include at least one unknown basis function and a variable parameter configured for each basis function.

[0010] Functional expressions of multiple basis functions are randomly constructed according to a preset operator set and a preset terminator set; the terminators in the preset terminator set include input variables of the initial crack extension model and constant terms in the basis functions; when the basis functions are constructed, the value of the constant term is unknown.

[0011] The unknown basis functions in the initial crack propagation model are replaced by the constructed basis functions to obtain several alternative crack propagation models; each of the alternative crack propagation models includes a plurality of unknown variable parameters; the variable parameters of the alternative crack propagation models include at least the variable parameters configured by the basis functions; the variable parameters refer to the parameters in the alternative crack propagation models that will change with the experimental data.

[0012] A number of candidate crack extension models constitute an initial crack extension model population; each candidate crack extension model is regarded as an individual; each element of the basis function in the candidate crack extension model is regarded as a gene; each element includes an operator and a terminator in the constructed basis function.

[0013] Gene crossover and mutation behavior are used to evolve the individuals in the initial crack propagation model population, and the variable parameters of the model corresponding to each individual are fitted according to the crack propagation experimental data of each classification, and the evaluation index of each individual is calculated to screen out the optimal crack propagation model; the evaluation index includes model complexity, the number of model variable parameters, model fitting goodness and the degree of dispersion of changes in model variable parameter values.

[0014] In a second aspect, the present application provides a data-driven crack propagation model building device, comprising the following modules.

[0015] The multi-source data classification module is used to obtain multi-source crack extension experimental data and perform data classification.

[0016] The initial crack extension model acquisition module is used to obtain the initial crack extension model; the initial crack extension model includes a preset classical crack extension model and preset correction items; the preset correction items include at least one unknown basis function and variable parameters configured for each basis function.

[0017] A basis function construction module is used to randomly construct function expressions of multiple basis functions according to a preset operator set and a preset terminator set; the terminators in the preset terminator set include input variables of the initial crack extension model and constant terms in the basis function; when the basis function is constructed, the value of the constant term is unknown.

[0018] The alternative crack extension module construction module is used to replace the unknown basis functions in the initial crack extension model with the constructed basis functions to obtain several alternative crack extension models; each of the alternative crack extension models includes multiple unknown variable parameters; the variable parameters of the alternative crack extension model at least include the variable parameters of the basis function configuration; the variable parameters refer to the parameters in the alternative crack extension model that will change with the experimental data.

[0019] The population construction module is used to form an initial crack extension model population from a number of alternative crack extension models; each alternative crack extension model is regarded as an individual; each element of the basis function in the alternative crack extension model is regarded as a gene; each element includes an operator and a terminator in the constructed basis function.

[0020] The crack extension model optimization module is used to evolve the individuals in the initial crack extension model population by adopting gene crossover and mutation behavior, fit the variable parameters of the model corresponding to each individual according to the crack extension experimental data of each classification, and calculate the evaluation index of each individual to screen out the optimal crack extension model; the evaluation index includes model complexity, the number of model variable parameters, model fitting goodness and the degree of dispersion of changes in model variable parameter values.

[0021] In a third aspect, the present application provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-mentioned data-driven crack propagation model establishment method.

[0022] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned data-driven crack propagation model establishment method.

[0023] According to the specific embodiments provided in this application, this application discloses the following technical effects.

[0024] The present application provides a method, device and medium for establishing a crack extension model based on data-driven, the method comprising: obtaining multi-source crack extension experimental data and performing data classification; obtaining an initial crack extension model containing unknown correction terms; randomly constructing multiple function expressions of basis functions in the correction terms according to preset operators and preset terminators; substituting the constructed basis functions into the initial crack extension model to obtain alternative crack extension models; forming an initial crack extension model population with several alternative crack extension models; evolving individuals in the initial crack extension model population using gene crossover and mutation behavior, fitting the model parameters of each individual according to the crack extension experimental data of each classification, and calculating the evaluation index of each individual to screen out the optimal crack extension model. The present invention constructs multiple alternative crack extension models, fits model parameters (including variable parameters and constant parameters) using experimental data of different classifications, and calculates the evaluation index of each model, thereby screening individuals with excellent performance during the evolution process, and finally screening out the optimal crack extension model. The present invention can reliably learn a unified crack extension model that meets the requirements based on multi-source crack extension experimental data. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0026] Figure 1 This is an application environment diagram of a method for establishing a crack propagation model based on data-driven in one embodiment of the present application.

[0027] Figure 2 A schematic flow chart of a method for establishing a crack propagation model based on data drive provided in one embodiment of the present application.

[0028] Figure 3 A schematic diagram of the technical concept of a data-driven crack propagation model establishment method provided in one embodiment of the present application.

[0029] Figure 4 An example schematic diagram of basis functions in a loading alternative crack growth model provided in an embodiment of the present application.

[0030] Figure 5 An exemplary schematic diagram of generating a new model by gene crossover provided in one embodiment of the present application.

[0031] Figure 6 Two exemplary schematic diagrams of generating new models by gene mutation are provided in one embodiment of the present application.

[0032] Figure 7 A schematic diagram showing the characterization capability of the model obtained through data-driven learning provided in one embodiment of the present application for label 1 data.

[0033] Figure 8 A schematic diagram showing the representation capability of the model obtained through data-driven learning provided in one embodiment of the present application for label 2 data.

[0034] Fig. 9 A schematic diagram of functional modules of a data-driven crack propagation model building device provided in another embodiment of the present application.

[0035] Fig.10 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0036] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0037] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0038] The data-driven crack growth modeling method provided in the embodiments of the present application can be applied to Figure 1 In the application environment shown. Among them, the terminal communicates with the server through the network. The data storage system can store the data that the server needs to process. The data storage system can be set up separately, integrated on the server, or placed on the cloud or other servers. The terminal can send multi-source crack extension experimental data, the initial crack extension model, the preset operator set and the preset terminator set to the server. After the server receives the multi-source crack extension experimental data, the initial crack extension model, the preset operator set and the preset terminator set, the server classifies the multi-source crack extension experimental data, and randomly constructs a function expression form of multiple basis functions according to the preset operator set and the preset terminator set; replaces the unknown basis function in the initial crack extension model with the constructed basis function to obtain several alternative crack extension models; constitutes the initial crack extension model population with several alternative crack extension models; uses gene crossover and mutation behavior to evolve the individuals in the initial crack extension model population, and fits the variable parameters of the model corresponding to each individual according to the crack extension experimental data of each classification, and calculates the evaluation index of each individual to screen out the optimal crack extension model. The server can feed back the obtained optimal crack extension model to the terminal. In addition, in some embodiments, the data-driven crack extension model establishment method can also be implemented by the server or the terminal alone, such as the terminal can directly construct the optimal crack extension model for the multi-source crack extension experimental data, the initial crack extension model, the preset operator set and the preset terminator set, or the server can obtain the multi-source crack extension experimental data, the initial crack extension model, the preset operator set and the preset terminator set from the data storage system, and construct the optimal crack extension model.

[0039] The terminals may be, but are not limited to, various desktop computers, laptops, smart phones, tablet computers, IoT devices and portable wearable devices. IoT devices may be smart speakers, smart TVs, smart air conditioners, smart car-mounted devices, etc. Portable wearable devices may be smart watches, smart bracelets, head-mounted devices, etc. The server may be implemented as an independent server or a server cluster consisting of multiple servers, or a cloud server.

[0040] In an exemplary embodiment, Figure 2 and Figure 3 As shown, a method for establishing a crack growth model based on data drive is provided. The method is executed by a computer device, and can be executed by a computer device such as a terminal or a server alone, or by a terminal and a server together. In the embodiment of the present application, the method is applied to Figure 1The server in is taken as an example to illustrate, including the following steps 101 to 106.

[0041] Step 101, obtaining multi-source crack extension experimental data and performing data classification.

[0042] In this embodiment, multi-source crack growth experimental data are introduced to learn a concise unified crack growth model.

[0043] The established method is explained by generating data of existing classical models. Taking the influence of load as an example, experimental data often only contain limited stress ratios, while the crack growth model needs to be available under a wide range of stress ratios. When fitting parameters, the present invention constrains all stress ratio data to use the same fitting parameters, so that this set of parameters is used under any stress ratio. Specifically, the present invention classifies the input multi-source crack growth experimental data and establishes labels, requiring the model to comprehensively determine parameters for data with different labels instead of directly fitting parameters for each data, so that the determined model parameters can meet the needs of engineering applications.

[0044] The crack growth experimental data generated based on the NASGRO (HS) model are used to illustrate.

[0045] (1).

[0046] Among them, all the parameters and stress ratio settings of the generated data are shown in Table 1.

[0047] Table 1 Parameter settings of NASGRO (HS) model used to generate experimental data

[0048]

[0049] When generating experimental data, the model parameters are known. , , and In the case of a certain stress ratio (such as -0.2, 0.1 or 0.4, etc.), it can be concluded that and Different numerical combinations of the same set of model parameters , , and Next, the obtained and Different numerical combinations of are considered the same type of data and are given the same label.

[0050] In Table 1, it can be seen that the data of label 1 and label 2 are the data after merging multiple stress ratios, and there will be two sets of fitting parameters with different values ​​for the data of label 1 or the data of label 2.

[0051] Furthermore, additional noise is added to the data according to formula (2) to simulate the noise that may exist in the experimental data:

[0052] (2).

[0053] in, , represents the randomly applied noise, Used as a control for the variance of the applied noise, as an example, .

[0054] As an example, the present invention can classify data according to different experimental factors such as material grade, heat treatment form, sampling direction, experimental environment, loading conditions, etc.

[0055] Step 102, obtaining an initial crack growth model; the initial crack growth model includes a preset classical crack growth model and preset correction items; the preset correction items include at least one unknown basis function and a variable parameter configured for each basis function.

[0056] For step 102, a crack extension model search space is constructed based on domain knowledge, wherein the search space includes the form adopted by the model, the input variables of the model, the parameter range of the model, and the operation function that the model needs to adopt.

[0057] In order to ensure that the crack propagation model finally established can cover the crack propagation laws of various types of experimental data in step 101, engineers first determine the form of the model, the input variables of the model, the parameter range of the model, and the operation functions that the model needs to adopt based on the knowledge of the crack propagation field.

[0058] As for the form adopted by the model, as an example, since many existing crack growth models can be summarized as an improved form of the Paris law, the present invention first defines the initial crack growth model to be learned as an improved form of the Paris model:

[0059] (3).

[0060] in, is the variable parameter in the Paris model part, is the correction function for the Paris model, which is unknown and needs to be obtained through symbolic regression training. It contains constant terms, variable parameters, and the driving force for crack growth selected in the above example. and .

[0061] Furthermore, since most of the experimental data are expressed in a double logarithmic coordinate system, formula (3) is transformed into a double logarithmic coordinate system as the structural form of the crack growth model. At the same time, in order to more uniformly describe the variable parameters in the model in the following text, Corresponding replacement , and we get formula (4).

[0062] (4).

[0063] Furthermore, a linear combination of basis functions is used to represent the unknown correction term .

[0064] (5).

[0065] in, is the variable parameter to be determined that is newly introduced corresponding to these basis functions. The expression form of is unknown and needs to be randomly constructed according to the preset operator and the preset terminator in step 103. In addition, the expression form of the correction term shown in formula (5) is predetermined by engineers based on the knowledge of crack propagation. According to actual needs, the expression form of the correction term can be other forms, and the form shown in formula (5) does not have a limiting effect.

[0066] Furthermore, due to the direct and As input, after complex calculations, there may be confusing dimensions in formula (5), which is not conducive to the interpretability of the model. Therefore, in this embodiment, and The input variables of the model are defined as follows.

[0067] (6).

[0068] in, are two newly introduced parameters to be determined. In this embodiment, has the same dimension as the stress intensity factor, so after being processed by formula (6), formula (5) becomes a dimensionless correction function. In particular, due to It has the same dimension as the stress intensity factor, requiring the model to be constructed after Certain physical constraints should be met, so The following constraints were made.

[0069] (7).

[0070] Therefore, the parameters currently included in formula (5) are and , they are all parameters that can change with the experimental data to reflect the dispersion of multi-source experimental data due to differences in experimental factors.

[0071] Furthermore, in order to reflect the constants that may appear in the crack growth model, the present invention uses the randomly generated basis function Introducing symbols As representatives of constants in the model, their values, once determined, will not change with the data.

[0072] In the above example, in the search space of the initial crack extension model constructed based on the domain knowledge of the present invention, the operation functions may be addition, subtraction, multiplication, logarithm and reciprocal functions.

[0073] Step 103, randomly constructing functional expressions of multiple basis functions according to a preset operator set and a preset terminator set; the terminators in the preset terminator set include input variables of the initial crack extension model and constant terms in the basis functions; when the basis functions are constructed, the value of the constant term is unknown.

[0074] Step 104, using the constructed basis function to replace the unknown basis function in the initial crack propagation model to obtain several alternative crack propagation models; each of the alternative crack propagation models includes a plurality of unknown variable parameters; the variable parameters of the alternative crack propagation model at least include the variable parameters of the basis function configuration; the variable parameters refer to the parameters in the alternative crack propagation model that will change with the experimental data.

[0075] Step 105, several candidate crack extension models are used to form an initial crack extension model population; each candidate crack extension model is regarded as an individual; each element of the basis function in the candidate crack extension model is regarded as a gene; each element includes an operator and a terminator in the constructed basis function.

[0076] Step 106, using gene crossover and mutation behavior to evolve individuals in the initial crack extension model population, fitting the variable parameters of the model corresponding to each individual according to the crack extension experimental data of each classification, and calculating the evaluation index of each individual to screen out the optimal crack extension model; the evaluation index includes model complexity, the number of model variable parameters, model fitting goodness and the degree of dispersion of changes in model variable parameter values.

[0077] By implementing the above steps 101 to 106, the present invention can directly learn a concise unified crack propagation model from multi-source experimental data reflecting multiple experimental factors. First, the parameter-containing model search space is determined in combination with domain knowledge, and a reasonable model search space is constructed by drawing on existing typical work, thereby avoiding a large amount of time spent on examining those randomly generated models of little value, so as to quickly converge to the model actually needed. Secondly, a parameter-containing model is used for symbolic regression, and a unified form of crack propagation model is mined with multi-source experimental data of multiple experimental factors as input. Since the parameter-containing model can comprehensively utilize the changes in model variables and parameters to fully characterize the dispersion in the data, the crack propagation model learned can reflect the inherent similarity of multi-source experimental data. Finally, an evaluation and update process of parameter-containing alternative crack propagation models is established. By independently evaluating the parameter and structural form evolution of the alternative crack propagation model, each alternative crack propagation model is tested on different data respectively, and the structural generalization ability and parameter stability of the alternative model are comprehensively evaluated during the training process, and finally a unified crack propagation model that meets the requirements is learned.

[0078] In another exemplary embodiment of the present application, in step 103, taking the basis function shown in the above formula (5) as an example, when constructing the expression form of the basis function, the operator set includes “ , , , ",in is a custom reciprocal function. The terminator set includes " , , According to the form of formula (5), a large number of different parameter-containing models, namely, candidate crack growth models, are generated by randomly extracting operators and terminators. Several candidate crack growth models constitute a population.

[0079] Specifically, Figure 4 As shown, a process of randomly generating a basis function is shown. First, the operator " ”, which contains two inputs, and thus further loads a constant With another operator " ". Operator" "Two inputs are still required, so a constant is further loaded With another operator " ". Operator" "Only one input is needed, so a variable is loaded further At this point, there are no more symbols that need to be loaded, sampling is complete, and a basis function is generated Therefore, it is concluded that step 103, randomly constructing a plurality of function expressions of basis functions according to a preset operator set and a preset terminator set, specifically includes the following steps.

[0080] (a1) Randomly selecting an operator from the preset operator set to obtain a first operator.

[0081] (a2) extracting a terminator and a second operator from the preset terminator set and the preset operator set respectively according to the required input number of the currently extracted first operator.

[0082] (a3) If the currently extracted second operator and terminator cannot form a complete expression, the currently extracted second operator replaces the currently extracted first operator, and the process returns to step (a2) "according to the required number of inputs of the currently extracted first operator, extract the terminator and the second operator from the preset terminator set and the preset operator set respectively", until all the extracted operators and all the terminators form a complete expression, thereby obtaining a functional expression form of a basis function.

[0083] (a4) Return to the step of “randomly extracting an operator from the preset operator set to obtain a first operator” until a preset number of basis function expressions are extracted.

[0084] In another exemplary embodiment of the present application, the model population is evolved by combining genetic programming technology with a phased model screening strategy. By simulating the "survival of the fittest" strategy of biological population evolution, the model individuals with poor performance are gradually eliminated, and finally the individuals with excellent performance are screened out. The basis functions in formula (5) are regarded as multiple genes in the genetic programming-based method, and a complete candidate crack extension model is regarded as an individual. By simulating the gene crossover and mutation behavior existing in biological evolution, it is possible to generate new individuals based on existing individuals and update the population. Among them, Figure 5 It shows how to generate new model individuals from the previous generation population through genetic crossover. Figure 6 It shows how to generate new model individuals from the previous generation of population through gene mutation, where the mutation forms are more diverse, which may be a mutation of the local model tree or a mutation of a single algorithm / terminator. With the evolution of each generation of model population, the present invention establishes a phased model screening strategy. For the example of formula (4), as an example, set the first stage evolution to 25 generations, using the complexity and goodness of fit As the evaluation index of each candidate crack growth model, at the end of the first stage, there are as many models in the population as possible that can fit the different label data well but are not too complex. As an example, the second stage evolution is set to 25 generations, and these models with strong fitting ability are further screened in the second stage, using complexity The dispersion of the model's variable parameter values As the evaluation index of each candidate crack growth model. In addition, it is also required that all model individuals in the population in the second stage It should not be higher than 0.05 to further select those alternative crack growth models that fit different label data well and have small fluctuations in fitting parameters. As an example, set the third stage evolution to 10 generations and use the complexity Variable number of model parameters As the evaluation index of each candidate crack growth model. In addition, in the third stage, all model individuals in the population are required to shall not be higher than 0.05, and all model individuals It should not be higher than 1.0. Finally, a population of 3000 model individuals was used for symbolic regression, and the model evolved for a total of 60 generations to obtain the optimal model individual, as shown in the following formula.

[0085] (8).

[0086] According to the above description, in step 106, individuals in the initial crack propagation model population are evolved by using gene crossover and mutation behavior, variable parameters of the model corresponding to each individual are fitted according to the crack propagation experimental data of each classification, and evaluation indicators of each individual are calculated to screen out the optimal crack propagation model, which specifically includes the following steps.

[0087] (c1) For each individual in the current generation population, the variable parameters of the model corresponding to each individual are fitted according to the crack extension experimental data of each classification, and the model complexity and the model fitting goodness of each individual are calculated; initially, the current generation population is the initial crack extension model population.

[0088] (c2) According to the model complexity and the model goodness of fit, multiple individuals with excellent performance are selected from the current generation population and recorded as first excellent individuals.

[0089] As for the model goodness of fit, for an alternative crack growth model, the crack growth experimental data of each category are used for data fitting, and a goodness of fit is calculated for each category. The final model goodness of fit is determined based on each goodness of fit.

[0090] (c3) Gene crossover or gene mutation is performed according to the basis functions of the individuals in the current generation population to obtain the first new individual; the first new individual of the current generation and the first excellent individual of the current generation constitute the next generation crack propagation model population.

[0091] (c4) Let the next generation crack extension model population be the current generation population, and return to the step "for each individual in the current generation population, fit the variable parameters of the model corresponding to each individual according to the crack extension experimental data of each classification, and calculate the model complexity and model fitting goodness of each individual", until the current iteration number reaches the first preset iteration number. Steps (c1) to (c4) are the evolution process of the first stage.

[0092] (c5) Let the next generation crack propagation model population corresponding to the first preset number of iterations be the current generation population, fit the variable parameters of the model corresponding to each individual according to the crack propagation experimental data of each classification, and calculate the model fitting goodness of each individual in the current generation population, the model complexity and the dispersion of the change of the model variable parameter value.

[0093] (c6) Selecting a plurality of individuals with excellent performance from the current generation population according to the complexity of the model and the dispersion of the variation of the variable parameter values ​​of the model and recording them as second excellent individuals; the model fitting goodness of the second excellent individuals is less than or equal to the preset fitting goodness.

[0094] (c7) Gene crossover or gene mutation is performed according to the basis functions of the individuals in the current generation population to obtain the second new individual; the second new individual of the current generation and the second excellent individual of the current generation constitute the next generation crack propagation model population.

[0095] (c8) Let the next generation crack extension model population be the current population, and return to the step of "fitting the variable parameters of the model corresponding to each individual according to the crack extension experimental data of each classification, and calculating the model fitting goodness, model complexity and the dispersion degree of the change of the model variable parameter value of each individual in the current generation population", until the current iteration number reaches the second preset iteration number. Steps (c5) to (c8) are the evolution process of the second stage.

[0096] In order to determine the degree of dispersion of the model's variable parameter values, an alternative crack propagation model is fitted using the crack propagation experimental data of each category, and then the value of each variable parameter is calculated for each category of experimental data. Then, the degree of dispersion of the final model's variable parameter values ​​is determined based on the dispersion of the changes in the values ​​of each variable parameter.

[0097] (c9) The next generation crack extension model population corresponding to the second preset number of iterations is set as the current generation population, and the evaluation index of each individual in the current generation population is calculated.

[0098] (c10) According to the complexity of the model and the number of variable parameters of the model, multiple individuals with excellent performance are selected from the current generation population and recorded as third excellent individuals; the model fitting goodness of the third excellent individuals is less than or equal to the preset fitting goodness of fit, and the variation dispersion of the model variable parameter values ​​is less than or equal to the preset variation dispersion.

[0099] (c11) Gene crossover or gene mutation is performed according to the basis functions of the individuals in the current generation population to obtain the third new individual; the third new individual of the current generation and the third excellent individual of the current generation constitute the next generation crack extension model population.

[0100] (c12) Let the next generation crack extension model population be the current generation population, and return to the step of "fitting the variable parameters of the model corresponding to each individual according to the crack extension experimental data of each classification, and calculating each of the evaluation indicators of each individual in the current generation population", until the current iteration number reaches the third preset iteration number. Steps (c9) to (c12) are the evolution process of the third stage.

[0101] (c13) Selecting the best individual in the crack growth model population corresponding to the third preset number of iterations as the best crack growth model.

[0102] In another exemplary embodiment of the present application, two methods are used to screen individuals with excellent performance during the population evolution process, and the individuals with excellent performance in each stage include the excellent individuals screened by the two methods.

[0103] The first way is that for individuals with excellent performance in the first stage, individuals at the Pareto front are selected from individuals in the population according to the corresponding evaluation indicators and identified as "excellent performance". For individuals with excellent performance in the second stage, the constraints of model goodness of fit must be met, that is, from individuals whose model goodness of fit is less than or equal to the preset goodness of fit, individuals at the Pareto front are selected according to the corresponding evaluation indicators and identified as "excellent performance". For individuals with excellent performance in the third stage, the constraints of model goodness of fit and the degree of dispersion of the variation of the model variable parameter values ​​must be met, that is, from individuals whose model goodness of fit is less than or equal to the preset goodness of fit and whose degree of dispersion of the variation of the model variable parameter values ​​is less than or equal to the preset degree of dispersion of variation are selected according to the corresponding evaluation indicators and identified as "excellent performance".

[0104] The second method is that for individuals with excellent performance in the first stage, they are ranked from good to bad according to the model goodness of fit, and a certain proportion of individuals with a higher ranking are identified as "excellent performance". For individuals with excellent performance in the second stage, individuals with a model goodness of fit less than or equal to the preset goodness of fit are ranked from good to bad according to the dispersion of the model variable parameter values, and a certain proportion of individuals with a higher ranking are identified as "excellent performance". For individuals with excellent performance in the third stage, individuals with a model goodness of fit less than or equal to the preset goodness of fit and a dispersion of the model variable parameter values ​​less than or equal to the preset dispersion of the changes are ranked from good to bad according to the number of model variable parameters, and a certain proportion of individuals with a higher ranking are identified as "excellent performance".

[0105] In another exemplary embodiment of the present application, the number of individuals in each generation of population is the same, and excellent individuals are selected from the current generation population and directly used as individuals in the next generation population, while other individuals in the next generation population are generated by gene crossover and mutation of individuals in the current generation population, and a bidding competition method can be used specifically. Set the scale of the bidding competition (for example, set the scale of the bidding competition to 100), then each time 100 individuals are selected from the current generation population to compete in the bidding competition, and the individual with the best performance wins and is selected (in the bidding competition, the individual with the best goodness of fit in the first stage performs best, the individual with the smallest parameter variation range in the second stage performs best, and the individual with the least number of parameters in the third stage performs best), and the selected individual with the best performance in the bidding competition is subjected to gene mutation, gene crossover, or direct inheritance to generate new individuals of the current generation. If gene crossover is performed, two individuals with excellent performance need to be selected.

[0106] In another exemplary embodiment of the present application, parameter fitting is performed on the parameter-containing model to be assessed to obtain evaluation indicators. First, the parameters that can be changed for data with different labels in the model, as well as the constants that need to be kept consistent for all data, are clarified. Then, the variable parameters in the model are determined in the face of specific data. The constants in the model are first determined in the form of traversal, and on this basis, the variable parameters in the model are determined using an alternating optimization algorithm. Finally, all models are sorted based on the evaluation indicators of each parameter-containing model in the population.

[0107] Since the constants in the model are consistent for all cases, the constants in each candidate crack growth model are first assigned. The symbolic regression method constructed by the present invention will use a traversal method to assign constants, that is, consider all possible values ​​of all constants in the model to evaluate the model, and finally select the optimal constant value as the fixed constant of the model. For example, the generated model contains two constants , and the sampling space of the constant is [-1, 0, 1], then There will be 9 possible values ​​for , that is, the alternative crack growth model constructed by the invention will be specifically divided into 9 models to be evaluated after being assigned constant values, and then the model with the best performance will be selected based on the subsequent evaluation of these 9 models with different constant values, and the constant value of the best performing model will be selected as the assigned value of the model constant. For the purpose of improving interpretability, no arbitrarily given constants are given. The sampling space is not set to [-1, 0, 1] but initially set to [-1, 0, 1].

[0108] After the constants in the alternative crack growth model are assigned, in order to establish the evaluation of the model, there are also parameters used for dimensionless , and the parameters used as linear combinations of basis functions Need to be determined. Obviously due to The existence of , cannot ensure that all parameters of any alternative crack extension model in the population can be obtained based on linear regression, so the present invention specifically establishes a fitting process for these parameters. With variable parameters The parameter fitting is performed in the way of The quasi-Newton method is used to solve the problem. The linear regression method is used to solve the problem. Specifically, at the beginning, the parameters are given With parameters Initial value, use the experimental data of each classification to fit the parameters, and use the quasi-Newton method to solve the parameters To update the previous parameters , the currently updated parameters Assign to the model, then use the experimental data of each classification to fit the parameters, and use the linear regression method to solve the parameters To update the previous parameters , the currently updated parameters Assign the model, then use the experimental data of each classification to fit the parameters, and continue to use the quasi-Newton method to solve the parameters To update the previous parameters , constantly optimizing parameters alternately With variable parameters , until the error between the model's predicted data and the corresponding actual data converges, the parameter iteration optimization process ends, and the variable parameters are fitted. With variable parameters In particular, for the parameter As for the stress intensity factor, it may have a specific physical meaning because of its dimension. Therefore, except for formula (7), the present invention does not The parameters of are intended to cooperate with other constraints. For the above variable parameter fitting, the model fitting goodness can be further determined according to the experimental data of each classification, and the model with the optimal constant value in each model to be evaluated after the constant value is assigned can be screened out by using the model fitting goodness. After the optimal constant value in the model is determined, in the process of population evolution, further evaluation indicators can be obtained, including model complexity , model fit goodness , the dispersion of model parameter value changes And the number of model parameters , and the best individuals in the population are selected using the evaluation index. Therefore, before the first execution of step (c1) "fitting the variable parameters of the model corresponding to each individual according to the crack extension experimental data of each classification, and calculating the model complexity and model fitting goodness of each individual", the data-driven crack extension model establishment method further includes the following steps.

[0109] For each candidate crack growth model, the optimal constant term value combination among multiple constant term value combinations is selected according to the model fitting goodness; the constant term value combination refers to the value combination of each constant term in the basis function randomly determined according to the preset constant term value range; the fitting goodness is determined according to crack growth experimental data of different classifications. The crack growth model with the optimal constant term value combination is used to fit the model variable parameters.

[0110] In another exemplary embodiment of the present application, the step of "for each candidate crack extension model, screening the optimal constant term numerical combination among multiple constant term numerical combinations according to the model fit goodness" specifically includes the following steps.

[0111] (b1) For each candidate crack growth model, a numerical combination of the constant terms in the basis function is randomly determined according to a preset constant term value range.

[0112] (b2) For the alternative crack growth model under each numerical combination, the crack growth experimental data of each classification are used to fit the alternative crack growth model and the model fitting goodness is calculated.

[0113] (b3) Determining the optimal constant term numerical combination of each alternative crack growth model according to the model fit goodness of the alternative crack growth model under each numerical combination.

[0114] In another exemplary embodiment of the present application, for the parameter-containing model to be evaluated, whether there is correlation between the parameters is checked, and redundant parameters are removed based on the test results to improve the robustness and reliability of parameter fitting. Since the basis functions in formula (5) are randomly generated, they may be linearly correlated with each other and affect the fitting of the parameters. Therefore, in this embodiment, any two of the formula (5) are first checked. and Can it be simplified to the same expression, thereby removing the same basis functions and regaining independent variable parameters? Therefore, before executing the step of "selecting the optimal constant term numerical combination among multiple constant term numerical combinations according to the model fit goodness for each candidate crack growth model", the data-driven crack growth model establishment method further includes the following steps.

[0115] For each alternative crack propagation model, check whether any two basis functions can be simplified to the same expression; if so, remove any basis function of the two basis functions in the alternative crack propagation model that can be simplified to the same expression to obtain a simplified alternative crack propagation model, and use the simplified alternative crack propagation model to screen the optimal constant term numerical combination.

[0116] The following is a validation of the effectiveness of the unified crack growth model established by the method of the present invention. The performance of the formula (8) obtained by learning the technical method of the present invention on different types of crack growth experimental data is as follows: Figure 7 and Figure 8 The two sets of parameters fitted by formula (8) for the data of the two labels in Table 1 are shown in Table 2.

[0117] Table 2 The variable parameter fitting results of the model obtained by data-driven learning for the data in Table 1

[0118]

[0119] Figure 7 and Figure 8 The black dots in the figure represent the experimental data under different labels, and the surface area represents the crack propagation model after the specific parameters are determined according to the corresponding label data. Figure 7 and Figure 8 It reflects the characterization ability of the learned crack growth model on the respective label data.

[0120] Combination Figure 7 and Figure 8It can be seen that the crack growth model obtained based on data-driven learning performs well for both label data 1 and label data 2 in Table 1. In addition, the model can fit all stress ratio data within a certain label well using the same set of parameters, and can be used for stress ratios outside the training data, indicating that the crack growth model trained based on the data-driven method can meet the situation of stress ratio changes during variable amplitude load analysis in engineering applications.

[0121] Combining Table 1 and Table 2, it can be seen that the actual fitting parameters of the crack propagation model obtained based on data-driven learning are basically consistent with the parameters used to generate the data, indicating that the data-driven crack propagation model establishment method established in the present invention can not only find a model that meets the requirements from the noise data, but also stably and reliably obtain the fitting parameters for different data.

[0122] In this embodiment, a data-driven crack propagation model method is established, including a parameterized symbolic regression method, which can directly learn a concise unified crack propagation model from multi-source experimental data reflecting multiple experimental factors. First, a parameterized model is used instead of a specific model for symbolic regression, and the dispersion in the data is fully characterized by comprehensively utilizing the changes in model variables and variable parameters. The parameterized symbolic regression method established by the present invention can make full use of multi-source experimental data, so that the learned model can reflect its inherent similarity. Secondly, a parameterized model evaluation process is established specifically, which can independently carry out the evolution of the parameterized model and the parameter evaluation, so that each parameterized model can test different data, and comprehensively evaluate the structural generalization ability and parameter stability of the alternative model during the training process. Finally, combined with the model evolution based on genetic programming and the staged model screening, a controllable and intuitive model population evolution can be achieved, so that a unified crack propagation model that meets the requirements can be reliably learned.

[0123] The present application also provides an application scenario, which applies the above-mentioned data-driven crack propagation model establishment method. Specifically: The data-driven crack propagation model establishment method provided in this embodiment can be applied in the fatigue safety design scenario of aviation structures. In the fatigue safety analysis process of aircraft structures, different parts may use materials in different states, and the service environment may also be different. In order to ensure reliable fatigue analysis of aircraft structures, it is necessary to carry out crack propagation tests under factors such as multiple material states and multiple experimental environments. Therefore, based on the present invention, the crack propagation model that is most suitable for these data can be directly obtained from a large amount of experimental data, providing a reliable basis for the fatigue safety design of aircraft structures.

[0124] Based on the same inventive concept, the embodiment of the present application also provides a device for implementing the above-mentioned data-driven crack propagation model establishment method. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above-mentioned method, so the specific limitations in one or more data-driven crack propagation model establishment device embodiments provided below can refer to the limitations of the data-driven crack propagation model establishment method above, and will not be repeated here.

[0125] In an exemplary embodiment, Fig. 9 As shown, a data-driven crack propagation model building device is provided, including the following modules.

[0126] The multi-source data classification module M1 is used to obtain multi-source crack extension experimental data and perform data classification.

[0127] The initial crack extension model acquisition module M2 is used to obtain the initial crack extension model; the initial crack extension model includes a preset classical crack extension model and preset correction items; the preset correction items include at least one unknown basis function and variable parameters configured for each basis function.

[0128] The basis function construction module M3 is used to randomly construct function expressions of multiple basis functions according to a preset operator set and a preset terminator set; the terminators in the preset terminator set include input variables of the initial crack extension model and constant terms in the basis function; when the basis function is constructed, the value of the constant term is unknown.

[0129] The alternative crack extension module construction module M4 is used to replace the unknown basis functions in the initial crack extension model with the constructed basis functions to obtain several alternative crack extension models; each of the alternative crack extension models includes multiple unknown variable parameters; the variable parameters of the alternative crack extension models at least include the variable parameters of the basis function configuration.

[0130] The population construction module M5 is used to form an initial crack extension model population from a number of candidate crack extension models; each candidate crack extension model is regarded as an individual; each element of the basis function in the candidate crack extension model is regarded as a gene; each element includes an operator and a terminator in the constructed basis function.

[0131] The crack extension model optimization module M6 is used to evolve the individuals in the initial crack extension model population by adopting gene crossover and mutation behavior, fit the variable parameters of the model corresponding to each individual according to the crack extension experimental data of each classification, and calculate the evaluation index of each individual to screen out the optimal crack extension model; the evaluation index includes model complexity, the number of model variable parameters, model fitting goodness and the degree of dispersion of changes in model variable parameter values.

[0132] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Fig.10 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the constructed data-driven crack propagation model. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for establishing a crack propagation model based on data-driven is implemented.

[0133] Those skilled in the art will understand that Fig.10 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components. In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0134] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0135] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0136] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0137] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0138] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0139] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A method for establishing a crack growth model based on data-driven, characterized in that: The data-driven crack propagation model establishment method includes: Acquire multi-source crack extension experimental data and perform data classification; Acquire an initial crack growth model; the initial crack growth model includes a preset classical crack growth model and a preset correction term; the preset correction term includes at least one unknown basis function and a variable parameter configured for each basis function; Randomly constructing a plurality of function expressions of basis functions according to a preset operator set and a preset terminator set; the terminators in the preset terminator set include input variables of the initial crack extension model and constant terms in the basis functions; when the basis functions are constructed, the value of the constant term is unknown; Using the constructed basis function to replace the unknown basis function in the initial crack growth model, a number of alternative crack growth models are obtained; each of the alternative crack growth models includes a plurality of unknown variable parameters; the variable parameters of the alternative crack growth model at least include the variable parameters configured by the basis function; the variable parameters refer to the parameters in the alternative crack growth model that will change with the experimental data; A plurality of candidate crack extension models constitute an initial crack extension model population; each candidate crack extension model is regarded as an individual; each element of the basis function in the candidate crack extension model is regarded as a gene; each element includes an operator and a terminator in the constructed basis function; Gene crossover and mutation behavior are used to evolve the individuals in the initial crack propagation model population, and the variable parameters of the model corresponding to each individual are fitted according to the crack propagation experimental data of each classification, and the evaluation index of each individual is calculated to screen out the optimal crack propagation model; the evaluation index includes model complexity, the number of model variable parameters, model fitting goodness and the degree of dispersion of changes in model variable parameter values.

2. The method for establishing a crack growth model based on data drive according to claim 1, characterized in that: The function expressions of various basis functions are randomly constructed according to the preset operator set and the preset terminator set, including: Randomly select an operator from the preset operator set to obtain a first operator; Extracting a terminator and a second operator from the preset terminator set and the preset operator set respectively according to the required input number of the currently extracted first operator; If the currently extracted second operator and the terminator cannot form a complete expression, the currently extracted first operator is replaced by the currently extracted second operator, and the process returns to the step of "extracting the terminator and the second operator from the preset terminator set and the preset operator set respectively according to the required number of inputs of the currently extracted first operator", until all the extracted operators and all the terminators form a complete expression, and a function expression form of a base function is obtained; Return to step "randomly extract an operator from the preset operator set to obtain a first operator" until a function expression of a preset number of basis functions is extracted.

3. The method for establishing a crack growth model based on data drive according to claim 1, characterized in that: The individuals in the initial crack growth model population are evolved by using gene crossover and mutation behaviors, and the variable parameters of the model corresponding to each individual are fitted according to the crack growth experimental data of each classification, and the evaluation index of each individual is calculated to screen out the optimal crack growth model, which specifically includes: For each individual in the current generation population, the variable parameters of the model corresponding to each individual are fitted according to the crack extension experimental data of each classification, and the model complexity and the model fitting goodness of each individual are calculated; initially, the current generation population is the initial crack extension model population; According to the model complexity and the model goodness of fit, multiple individuals with excellent performance are selected from the current generation population and recorded as first excellent individuals; Gene crossover or gene mutation is performed according to the basis functions of individuals in the current generation population to obtain the first new individual; the first new individual of the current generation and the first excellent individual of the current generation constitute the next generation crack extension model population; The next generation crack extension model population is set as the current generation population, and the process returns to step "for each individual in the current generation population, the variable parameters of the model corresponding to each individual are fitted according to the crack extension experimental data of each classification, and the model complexity and the model fitting goodness of each individual are calculated", until the current number of iterations reaches the first preset number of iterations; The next generation crack extension model population corresponding to the first preset number of iterations is set as the current generation population, and the variable parameters of the model corresponding to each individual are fitted according to the crack extension experimental data of each classification, and the model fitting goodness, the model complexity and the dispersion degree of the change of the model variable parameter value of each individual in the current generation population are calculated; According to the complexity of the model and the dispersion of the change of the variable parameter value of the model, a plurality of individuals with excellent performance are selected from the current generation population and recorded as second excellent individuals; the model fitting goodness of the second excellent individuals is less than or equal to the preset fitting goodness; Gene crossover or gene mutation is performed according to the basis functions of individuals in the current generation population to obtain the second new individual; the second new individual of the current generation and the second excellent individual of the current generation constitute the next generation crack extension model population; The next generation crack extension model population is set as the current generation population, and the process returns to the step of "fitting the variable parameters of the model corresponding to each individual according to the crack extension experimental data of each classification, and calculating the model fitting goodness of fit, the model complexity and the dispersion degree of the change of the model variable parameter value of each individual in the current generation population", until the current number of iterations reaches the second preset number of iterations; The next generation crack extension model population corresponding to the second preset number of iterations is set as the current generation population, and the variable parameters of the model corresponding to each individual are fitted according to the crack extension experimental data of each classification, and the evaluation index of each individual in the current generation population is calculated; According to the complexity of the model and the number of variable parameters of the model, a plurality of individuals with excellent performance are selected from the current generation population and recorded as third excellent individuals; the model fitting goodness of the third excellent individuals is less than or equal to the preset fitting goodness, and the variation dispersion of the model variable parameter values ​​is less than or equal to the preset variation dispersion; Gene crossover or gene mutation is performed according to the basis functions of individuals in the current generation population to obtain the third new individual; the third new individual of the current generation and the third excellent individual of the current generation constitute the next generation crack extension model population; The next generation crack extension model population is set as the current generation population, and the process returns to the step of "fitting the variable parameters of the model corresponding to each individual according to the crack extension experimental data of each classification, and calculating the evaluation index of each individual in the current generation population", until the current number of iterations reaches the third preset number of iterations; The best individual in the crack growth model population corresponding to the third preset number of iterations is selected as the optimal crack growth model.

4. The method for establishing a crack growth model based on data drive according to claim 3, characterized in that: Before executing the step of "fitting the variable parameters of the model corresponding to each individual according to the crack extension experimental data of each classification, and calculating the model complexity and the model fitting goodness of each individual", the data-driven crack extension model establishment method further includes: For each candidate crack extension model, the optimal constant term numerical combination among multiple constant term numerical combinations is screened according to the model fitting goodness; the constant term numerical combination refers to the numerical combination of each constant term in the basis function randomly determined according to a preset constant term value range; the crack extension model with the optimal constant term numerical combination is used to fit the model variable parameters.

5. The method for establishing a crack growth model based on data drive according to claim 4, characterized in that: For each candidate crack growth model, the optimal constant term value combination among multiple constant term value combinations is selected according to the model fitting goodness, including: For each candidate crack growth model, randomly determine the numerical combination of the constant terms in the basis function according to the preset constant term value range, and obtain a number of known constant term crack growth models; Using the crack growth experimental data of each classification, model variable parameter fitting is performed on each known constant term crack growth model, and the model fitting goodness of each known constant term crack growth model is calculated; According to the model fitting goodness of each known constant term crack growth model, the optimal constant term numerical combination of the corresponding alternative crack growth model is determined.

6. The method for establishing a crack growth model based on data drive according to claim 4, characterized in that: Before executing the step of "selecting the optimal constant term numerical combination among multiple constant term numerical combinations according to the model fit goodness for each candidate crack growth model", the data-driven crack growth model establishment method further includes: For each alternative crack growth model, check whether any two basis functions can be simplified to the same expression; If so, any one of the two basis functions in the alternative crack growth model that can be simplified to the same expression is removed to obtain a simplified alternative crack growth model, and the simplified alternative crack growth model is used to perform optimal constant term numerical combination screening.

7. A data-driven crack growth modeling device, characterized in that: The data-driven crack propagation model building device includes: Multi-source data classification module, used to obtain multi-source crack extension experimental data and perform data classification; An initial crack extension model acquisition module is used to acquire an initial crack extension model; the initial crack extension model includes a preset classical crack extension model and a preset correction term; the preset correction term includes at least one unknown basis function and a variable parameter configured for each basis function; A basis function construction module, used for randomly constructing a plurality of basis function expression forms according to a preset operator set and a preset terminator set; the terminator in the preset terminator set includes an input variable of an initial crack extension model and a constant term in the basis function; when the basis function is constructed, the value of the constant term is unknown; An alternative crack extension module construction module is used to replace the unknown basis functions in the initial crack extension model with the constructed basis functions to obtain a number of alternative crack extension models; each of the alternative crack extension models includes a plurality of unknown variable parameters; the variable parameters of the alternative crack extension model at least include the variable parameters configured by the basis functions; the variable parameters refer to the parameters in the alternative crack extension model that will change with the experimental data; A population construction module is used to form an initial crack extension model population from a number of candidate crack extension models; each candidate crack extension model is regarded as an individual; each element of the basis function in the candidate crack extension model is regarded as a gene; each element includes an operator and a terminator in the constructed basis function; The crack extension model optimization module is used to evolve the individuals in the initial crack extension model population by adopting gene crossover and mutation behavior, fit the variable parameters of the model corresponding to each individual according to the crack extension experimental data of each classification, and calculate the evaluation index of each individual to screen out the optimal crack extension model; the evaluation index includes model complexity, the number of model variable parameters, model fitting goodness and the degree of dispersion of changes in model variable parameter values.

8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the data-driven crack propagation modeling method described in any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for establishing a crack growth model based on data drive described in any one of claims 1 to 6 is implemented.

Images