Fusion model construction method and health assessment method

Abstract

The invention provides a fusion model construction method and a health assessment method, and relates to the technical field of metal material life prediction. The fusion model construction method comprises the steps of performing multiple independent random assignment on crack growth parameters to obtain a plurality of first expressions of each failure mechanism model; according to the first measured data, performing Bayesian updating on the prior probability of each first expression; according to the plurality of first expressions and the posterior probability of each first expression, calculating to obtain a second expression of each failure mechanism model; according to the second measured data, performing Bayesian updating on the prior probability of each second expression; and according to the plurality of second expressions and the posterior probability of each second expression, calculating to obtain a fusion model. According to the method provided by the invention, the problem that the accuracy of a fatigue life prediction method based on the failure mechanism model is difficult to guarantee due to the inherent uncertainty of different failure mechanism model superposition metal material parameters is solved.

Description

Technical Field

[0001] The present application relates to the field of metal material life prediction, and in particular to a fusion model construction method and a health assessment method. Background Art

[0002] During the service life of metal materials in mechanical structures, the dominant failure mechanism is often fatigue failure. This failure mode is hidden and dangerous because even if the cyclic load amplitude to which the metal material is subjected is generally lower than its yield strength threshold, fatigue damage can still accumulate inadvertently, eventually leading to sudden failure events. In view of this, it is particularly important to conduct accurate health assessments of metal materials to avoid fatigue failure of metal materials.

[0003] In the prior art, the health assessment of metal materials is mainly based on measured data. However, the previous measured data of metal materials are usually measured data when the metal materials have not failed, and there are fewer measured data when they are on the verge of fatigue failure or have already failed. Therefore, it is difficult to conduct health assessment of metal materials based on measured data, and it is impossible to achieve a high-accuracy health assessment. Therefore, it is necessary to combine measured data and failure mechanism models to conduct health assessment of metal materials.

[0004] However, the selection of failure mechanism models is highly dependent on the professional insights and experience accumulation of technicians, and is essentially a method system that is deeply dependent on domain knowledge. The difference in the depth of technicians' understanding of failure mechanisms and the difference in experience in the recognition of failure modes will directly affect the selection and application of failure mechanism models. Different failure mechanism models plus the inherent uncertainty of metal material parameters make it difficult to guarantee the accuracy of health assessment methods based on failure mechanism models. Summary of the invention

[0005] The present application provides a fusion model construction method and a health assessment method to solve the problem that the accuracy of the health assessment method based on the failure mechanism model is difficult to guarantee due to the inherent uncertainty of the metal material's own parameters caused by different failure mechanism models.

[0006] The first aspect of the present application provides a fusion model construction method, the method comprising:

[0007] Performing multiple independent random assignments on multiple crack growth parameters of each of the preset multiple failure mechanism models to obtain a first expression of each failure mechanism model after each random assignment;

[0008] Acquire first measured data of a first fatigue crack of a target metal material, and perform Bayesian update on a priori probability of each preset first expression according to the first measured data to obtain a posterior probability of each first expression; wherein the first measured data is used to indicate crack propagation of the first fatigue crack, and the priori probability and posterior probability of each first expression are used to indicate the ability of the corresponding first expression to describe crack propagation;

[0009] According to the multiple first expressions of each failure mechanism model and the posterior probability of the corresponding first expressions, a second expression of each failure mechanism model is calculated;

[0010] Acquire second measured data of a second fatigue crack of the target metal material, and perform Bayesian update on the prior probability of each preset second expression according to the second measured data to obtain the posterior probability of each second expression;

[0011] According to the second expression of each failure mechanism model and the posterior probability of the corresponding second expression, a fusion model is calculated to predict the fatigue life of the target metal material according to the fusion model.

[0012] In a possible design, the first measured data includes: measured crack sizes of the target metal material in each of a plurality of first preset cycles, the first target cycle being any one of the plurality of first preset cycles; the first target expression being any one of a plurality of first expressions of the target failure mechanism model, the target failure mechanism model being any one of a plurality of failure mechanism models;

[0013] For the first target cycle and the first target expression, according to the first measured data, a Bayesian update is performed on the preset prior probability of each first expression to obtain the posterior probability of each first expression, including:

[0014] Obtaining a first measured crack size of the target metal material at a first target cycle according to the first measured data, and obtaining a first predicted crack size of the target metal material at the first target cycle according to the first target expression;

[0015] Obtaining a first likelihood function of the first target expression in a first target cycle according to an error between the first predicted crack size and the first measured crack size;

[0016] According to the first likelihood function, a Bayesian update is performed on the prior probability of the first target expression in the first target cycle to obtain the posterior probability of the first target expression in the first target cycle.

[0017] In a possible design, the prior probabilities of the multiple first expressions of the target failure mechanism model in the first cycle each obey the mean distribution; the likelihood functions of the multiple first expressions of the target failure mechanism model in the first target cycle each obey the normal distribution;

[0018] For the first target cycle and the target failure mechanism model, after performing a Bayesian update on the prior probability of the first target expression in the first target cycle according to the first likelihood function to obtain the posterior probability of the first target expression in the first target cycle, the method further includes:

[0019] The probability normalization process is performed on the respective posterior probabilities of the multiple first expressions of the target failure mechanism model in the first target cycle.

[0020] In a possible design, for the target failure mechanism model, a second expression of each failure mechanism model is calculated based on multiple first expressions of each failure mechanism model and the posterior probability of the corresponding first expression, including:

[0021] According to the respective posterior probabilities of the multiple first expressions of the target failure mechanism model, a weighted calculation is performed on the multiple crack growth parameters of each first expression of the target failure mechanism model to obtain a calculated value of each crack growth parameter;

[0022] A second expression of the target failure mechanism model is obtained according to the calculated values ​​of each of the plurality of crack growth parameters.

[0023] In a possible design, after calculating the second expression of each failure mechanism model according to the multiple first expressions of each failure mechanism model and the posterior probability of the corresponding first expression, the method further includes:

[0024] Obtaining third measured data of each of a plurality of third fatigue cracks of the target metal material; wherein the sampling time and / or sampling position of each third fatigue crack is different from that of the first fatigue crack;

[0025] According to the plurality of third measured data, parameters of the second expression of each failure mechanism model are updated respectively.

[0026] In a possible design, the second measured data includes: measured crack sizes of the target metal material in each of a plurality of second preset cycles, the second target cycle being any one of the plurality of second preset cycles;

[0027] For the second target cycle, according to the second measured data, the prior probability of each preset second expression is updated by Bayesian method to obtain the posterior probability of each second expression, including:

[0028] Obtaining a second measured crack size of the target metal material at a second target cycle according to the second measured data, and obtaining a second predicted crack size of the target metal material at the second target cycle predicted by each second expression according to a plurality of second expressions;

[0029] Obtaining a second likelihood function of each second expression at a second target cycle according to the errors between each second predicted crack size and the second measured crack size;

[0030] According to the second likelihood function of each second expression in the second target cycle, a Bayesian update is performed on the prior probability of each second expression in the second target cycle to obtain the posterior probability of each second expression in the second target cycle.

[0031] In a possible design, the prior probabilities of the plurality of second expressions in the first cycle follow a mean distribution; the errors of the plurality of second expressions in the second target cycle follow a normal distribution;

[0032] After performing Bayesian update on the prior probability of each second expression in the second target cycle according to the second likelihood function of each second expression in the second target cycle, the method further comprises:

[0033] Probability normalization is performed on the posterior probabilities of the multiple second expressions in the second target cycle.

[0034] In a possible design, a fusion model is calculated based on the second expression of each failure mechanism model and the posterior probability of the corresponding second expression, including:

[0035] For the second target cycle, a segmented fusion model of the second target cycle is obtained by weighted calculation according to the plurality of second expressions and the posterior probabilities of the corresponding second expressions in the second target cycle;

[0036] According to the segmented fusion model of each second preset cycle, a fusion model is obtained by fitting.

[0037] A second aspect of the present application provides a health assessment method, the method comprising:

[0038] Acquire fourth measured data of a fourth fatigue crack of the target metal material; wherein the fourth measured data includes: an initial crack size of the fourth fatigue crack;

[0039] According to the fourth measured data and the preset fusion model, at least one of the multiple health assessment indicators of the target metal material is evaluated; wherein the fusion model is a model constructed by the fusion model construction method of any one of the first aspects; the multiple health assessment indicators include: fatigue life, reliability and crack size.

[0040] In one possible design, the evaluation results include: the predicted crack size of the target metal material at each cycle;

[0041] After evaluating at least one of the plurality of health assessment indicators of the target metal material according to the fourth measured data and the preset fusion model, the method further includes:

[0042] A confidence interval is obtained according to the predicted crack size of the target metal material in each cycle and a preset prediction confidence level; wherein the confidence interval includes: an upper confidence crack size and a lower confidence crack size of the target metal material in each cycle.

[0043] The third aspect of the present application provides a fusion model construction device, the device comprising:

[0044] A random assignment module, used for performing multiple independent random assignments on multiple crack growth parameters of each of the preset multiple failure mechanism models, and obtaining a first expression of each failure mechanism model after each random assignment;

[0045] A first updating module is used to obtain first measured data of a first fatigue crack of a target metal material, and to perform a Bayesian update on a prior probability of each preset first expression according to the first measured data, so as to obtain a posterior probability of each first expression; wherein the first measured data is used to indicate a crack propagation condition of the first fatigue crack, and the prior probability and the posterior probability of each first expression are used to indicate the ability of the corresponding first expression to describe the crack propagation condition;

[0046] A first calculation module, used for calculating a second expression of each failure mechanism model according to a plurality of first expressions of each failure mechanism model and a posterior probability of the corresponding first expression;

[0047] A second updating module is used to obtain second measured data of a second fatigue crack of the target metal material, and perform Bayesian update on the prior probability of each preset second expression according to the second measured data to obtain a posterior probability of each second expression;

[0048] The second calculation module is used to calculate the fusion model according to the second expression of each failure mechanism model and the posterior probability of the corresponding second expression, so as to predict the fatigue life of the target metal material according to the fusion model.

[0049] A fourth aspect of the present application provides a health assessment device, the device comprising:

[0050] A data acquisition module, used to acquire fourth measured data of a fourth fatigue crack of the target metal material; wherein the fourth measured data includes: an initial crack size of the fourth fatigue crack;

[0051] A life prediction module is used to evaluate at least one of multiple health assessment indicators of the target metal material based on the fourth measured data and a preset fusion model; wherein the fusion model is a model constructed by the health assessment device of the third aspect; the multiple health assessment indicators include: fatigue life, reliability and crack size.

[0052] A fifth aspect of the present application provides an electronic device, comprising: a memory, and a memory communicatively connected to a processor;

[0053] Memory stores computer-executable instructions;

[0054] When the processor executes the computer-executable instructions stored in the memory, it is used to implement any fusion model construction method of the first aspect, or any health assessment method of the second aspect.

[0055] The sixth aspect of the present application provides a computer-readable storage medium, which stores computer execution instructions. When the computer execution instructions are executed by a processor, they are used to implement the fusion model construction method of any one of the first aspects, or the health assessment method of any one of the second aspects.

[0056] The seventh aspect of the present application provides a computer program product, including a computer program, which, when executed by a processor, is used to implement the fusion model construction method of any one of the first aspects, or the health assessment method of any one of the second aspects.

[0057] The present application provides a fusion model construction method and health assessment method, the fusion model construction method includes: performing multiple independent random assignments on crack growth parameters to obtain multiple first expressions of each failure mechanism model; performing Bayesian update on the prior probability of each first expression according to the first measured data; calculating the second expression of each failure mechanism model according to the multiple first expressions and the posterior probability of each first expression; performing Bayesian update on the prior probability of each second expression according to the second measured data; calculating the fusion model according to the multiple second expressions and the posterior probability of each second expression. The following technical effects are achieved: performing Bayesian update on the prior probabilities of the multiple first expressions of each failure mechanism model solves the problem of uncertainty of multiple crack growth parameters of the failure mechanism model; performing Bayesian update on the prior probabilities of the second expressions of each of the multiple failure mechanism models solves the problem that different failure mechanism models cannot be accurately applied to various objects and scenarios; determining the expression form of the likelihood function through measured data, and implementing the parameter update of the failure mechanism model through Bayesian update, thereby improving the accuracy of fatigue life prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the technical solutions in the present application or the prior art, a brief introduction is given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0059] Figure 1 Schematic diagram of a through crack;

[0060] Figure 2 Schematic diagram of cracks with multiple side surface cracks;

[0061] Figure 3 Schematic diagram of a single-side surface crack;

[0062] Figure 4 Schematic diagram of the process of building a fusion model provided in the embodiment of the present application Figure 1 ;

[0063] Figure 5 Schematic diagram of the process of building a fusion model provided in the embodiment of the present application Figure 2 ;

[0064] Figure 6 A schematic diagram of a cluster of curves of the Paris-Erdogan model provided in an embodiment of the present application;

[0065] Figure 7 Prior cumulative probability distribution provided for the embodiment of the present application Figure 1 ;

[0066] Figure 8 A comparison diagram of the posterior probability distribution provided in the embodiment of the present application and before the update;

[0067] Fig. 9 Comparison of the a posteriori reliability provided in the embodiment of the present application with that before the update Figure 1 ;

[0068] Fig.10 Schematic diagram of the process of building a fusion model provided in the embodiment of the present application Figure 3 ;

[0069] Fig.11 A comparison chart of the fusion model, prediction model and measured data provided in the embodiments of the present application;

[0070] Fig.12 A flowchart of a health assessment method provided in an embodiment of the present application;

[0071] Fig.13 A schematic diagram of a confidence interval provided for an embodiment of the present application;

[0072] Fig.14 Prior cumulative probability distribution provided for the embodiment of the present application Figure 2 ;

[0073] Fig.15 A posterior probability distribution diagram provided for an embodiment of the present application;

[0074] Fig.16 Comparison of the a posteriori reliability provided in the embodiment of the present application with that before the update Figure 2 . DETAILED DESCRIPTION

[0075] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0076] In the present application, words such as "first" and "second" are used to distinguish between identical or similar items with substantially the same functions and effects. Those skilled in the art will understand that words such as "first" and "second" do not limit the quantity and execution order, and words such as "first" and "second" do not necessarily limit the difference. It should be noted that in the present application, words such as "exemplary" or "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described as "exemplary" or "for example" in the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a specific way. In the present application, "at least one" refers to one or more, and "more" refers to two or more.

[0077] It should be noted that the "at..." in this application can be the instant when a certain situation occurs, or it can be a period of time after a certain situation occurs, and this application does not make specific restrictions on this. In addition, the fusion model construction method and health assessment method provided in this application are only used as examples, and the fusion model construction method and health assessment method can also include more or less content.

[0078] In order to clearly describe the technical solution of this application, some terms and technologies involved in this application are briefly introduced below:

[0079] Fatigue failure: refers to the phenomenon that the metal material of a mechanical structure suddenly breaks or its performance significantly degrades after a long period of cumulative damage under cyclic loads below its yield strength. This failure mode is usually not easy to detect in time, so it is hidden and dangerous.

[0080] Failure mechanism model: refers to a mathematical model or theoretical framework used to describe and predict fatigue failure of metal materials under specific conditions. The failure mechanism model can simulate and analyze the development process of fatigue damage based on factors such as the material's microstructure, mechanical properties, and load conditions, thereby predicting the fatigue life of the material.

[0081] Domain knowledge: refers to the professional knowledge, theories, methods and experience accumulated in a specific discipline or professional field. Domain knowledge includes an in-depth understanding of metal material properties, fatigue damage mechanisms, failure analysis, and related experimental techniques and data processing methods.

[0082] Bayesian updating: is a method to update the probability of an event or hypothesis after new information or data is known. Bayesian updating is based on Bayesian inference and calculates the posterior probability by combining the prior probability and the likelihood function of the new information.

[0083] Likelihood function: A function that describes the likelihood of a model parameter taking a specific value given observed data. Likelihood functions are often used to estimate model parameters, and the parameter values ​​that best fit the observed data are found by maximizing the likelihood function.

[0084] The technical solution of the present application is described in detail with specific embodiments below. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The present application will be described below in conjunction with the accompanying drawings. In order to clearly understand the technical solution of the present application, the solution of the prior art is first described in detail.

[0085] In recent years, with the development of science and technology, the application of mechanical structures in various industries has been widely popularized. Many complex operations in real life require the cooperation of mechanical structures and technicians or the completion of mechanical structures alone. In this case, the requirements for the reliability and life of mechanical structures are getting higher and higher. Especially for mechanical structures operating in some extreme environments, their operating conditions are harsh, maintenance and guarantee are difficult, their functional structures are complex, and there are many types of working conditions. Mechanical structures are prone to functional or performance degradation, and even failures with serious consequences. Therefore, how to conduct health status assessment for mechanical structures is the current research focus.

[0086] Metal materials for mechanical structures are one of the research directions. During the service life of metal materials, the dominant failure mechanism is often fatigue failure. This failure mode is hidden and dangerous because even if the amplitude of the cyclic load to which the metal material is subjected is generally lower than its yield strength threshold, fatigue damage can still accumulate inadvertently, eventually leading to sudden failure events. It is particularly important to perform accurate health assessments of metal materials, that is, to predict the number of stress or strain cycles that the metal material experiences when fatigue failure occurs, or the crack size under the working load, as well as the reliability level of the current state.

[0087] The health assessment method based on the failure mechanism model, relying on the status monitoring, status evaluation and measured data of metal materials, can comprehensively judge the current health status of metal materials, and further estimate key indicators such as the evolution trend of their structural performance, the time of fault initiation, the remaining service life and the reliability level.

[0088] In the prior art, the health assessment of metal materials is mainly based on measured data. However, the previous measured data of metal materials are usually measured data when the metal materials have not failed, and there are fewer measured data when they are on the verge of fatigue failure or have already failed. Therefore, it is difficult to conduct health assessment of metal materials based on measured data, and it is impossible to achieve a high-accuracy health assessment. Therefore, it is necessary to combine measured data and failure mechanism models to conduct health assessment of metal materials.

[0089] A method for fault prediction and maintenance planning based on global Bayesian updating comprises the following steps:

[0090] S101. Establish a life prediction model based on nonlinear Wiener process;

[0091] S102, establishing a dynamic predictive maintenance strategy for equipment based on the parameters in the global Bayesian update life prediction model;

[0092] S103. Calculation of cost rate function based on renewal-reward theory;

[0093] S104, optimizing and solving the predictive maintenance model based on genetic algorithm, that is, using the genetic algorithm (GA) toolbox in MATLAB to solve the minimum expected cost rate and the optimal parameter combination of the model.

[0094] However, this method has the following problems: in the parameter update step, historical information and real-time monitoring data are needed to obtain online and offline parameters. This requires technicians to have certain practical experience, such as how to select sensor models, how to arrange sensors on the experimental device, and how to send and process sensor data.

[0095] Another method for evaluating the health of alloy materials based on crack growth rate includes the following steps:

[0096] S201. Using a continuous piecewise log-linear crack growth rate model to describe crack growth rate data of the alloy;

[0097] S202, introducing the Heaviside step function and the transition stress intensity factor amplitude into the classic Paris model to establish a continuous segmented model;

[0098] S203, performing a crack growth experiment on the arc additively manufactured titanium alloy material to obtain crack growth rate data of the alloy material specimen;

[0099] S204. Perform statistical analysis on the experimental results to verify the accuracy of the model prediction results.

[0100] However, this method has the following problems: this method is used to predict the fatigue life of titanium alloy materials, and the material manufacturing process requires technicians to have sufficient physical background knowledge of the analysis object; in addition, the damage tolerance method is used to predict fatigue life, and the single mechanism model is improved. Affected by manufacturing process parameters and manufacturing process parameters, the single mechanism model cannot fully describe the actual extension of the actual crack. It can be seen that the health assessment method based on the failure mechanism model relies on the professional insights and experience accumulation of technicians, and is essentially a method system that deeply relies on domain knowledge. The difference in the depth of technicians' understanding of failure mechanisms and the difference in experience in the recognition of failure modes will directly affect the selection and application of failure mechanism models. For example, when predicting the fatigue life of metal materials, when the initial crack is monitored, the geometric and stress characteristics of the cracked structural parts and the crack extension can be analyzed based on the damage location determined by the initial crack, and the appropriate structural damage mechanism model can be selected, such as the Paris-Erdogan model, the Trantina-Johnson model or the Forman model. Then, the parameters of the selected failure mechanism model are updated and corrected in real time using prior knowledge and measured data. However, actual engineering structures are usually very complex, involving multiple links such as raw material selection, processing and manufacturing, installation and commissioning, with many parameters, and the influence of the uncertainty of the metal material's own parameters needs to be considered; in addition, different metal materials have different fatigue performances and corresponding fatigue mechanisms. A single failure mechanism model cannot cover all crack propagation situations. Therefore, the existing technology has different failure mechanism models superimposed on the inherent uncertainty of the metal material's own parameters, resulting in the problem that the accuracy of the health assessment method based on the failure mechanism model is difficult to guarantee.

[0101] Therefore, in response to the above technical problems, it was found in the study that in order to solve this problem, the multi-model fusion method is applied to fatigue life prediction, and a health assessment method based on the fusion model is proposed. Specifically, based on the above multiple failure mechanism models and measured data, a fusion model close to or consistent with the measured data is constructed, and this is used to deal with various uncertainties in fatigue life prediction. The fusion model can output a prediction curve that conforms to the actual crack extension, and give a prediction of the crack size distribution under the corresponding cycle, as well as the life prediction results of the corresponding confidence limit.

[0102] Based on the above creative findings, the technical solution of the present application is proposed.

[0103] The following introduces the application scenarios of the fusion model construction method and health assessment method provided in this application.

[0104] The fusion model construction method and health assessment method are applied to the prediction of the central crack of a finite width plate. Based on Bayesian inference and failure mechanism model, various uncertainties in fatigue life prediction can be handled. Take the classic Paris-Erdogan model as an example. The crack growth rate of the Paris-Erdogan model is shown in the following formula:

[0105]

[0106] Among them, a is the crack size, N is the number of cycles, △k is the stress intensity factor, and c and m are crack growth parameters. The randomization of c and m can reflect the uncertainty of the metal material itself, the initial crack size a0 can reflect the uncertainty of the initial crack measurement, and the randomization of △k can reflect the uncertainty of the load. The calculation formula of △k is:

[0107]

[0108] Among them, β is the correction coefficient. For different crack positions and forces, the calculation formula of β is different; △σ k is the stress variation acting on the finite width plate.

[0109] Figure 1 is a schematic diagram of a through crack. Figure 1 As shown in Figure 1, a through crack is a crack that runs through the entire thickness of a plate with a limited width. This type of crack is usually considered a serious defect because it will seriously affect the strength and durability of the component. When the crack is a through crack, the calculation formula for β is:

[0110]

[0111] Among them, β=1.00 (a<<h), σ=6M / bh 2 , b is the plate thickness, and h is half of the plate width.

[0112] Figure 2 Figure 1 is a schematic diagram of a crack with multiple side surface cracks. Figure 2 As shown in the figure, multi-side surface cracks refer to cracks that appear on multiple surfaces of a finite width plate. These cracks do not completely penetrate the finite width plate in the thickness direction, but may be connected or close to each other on multiple sides. When the crack is a multi-side surface crack, the calculation formula of β is:

[0113]

[0114] Among them, β=1.12 (a<<h).

[0115] Figure 3 Figure 1 is a schematic diagram of a single-side surface crack. Figure 3 As shown in the figure, a single-sided surface crack refers to a crack that only appears on one surface of a finite width plate, and this crack does not completely penetrate the finite width plate in the thickness direction. When the crack is a single-sided surface crack, the calculation formula of β is:

[0116]

[0117] Among them, β=1.12 (a<<h).

[0118] When the initial crack size a0 and the measured crack size a N When , the corresponding calculation formula for the number of cycles N is:

[0119]

[0120] Crack size a corresponding to the number of cycles N N The calculation formula is:

[0121]

[0122] The embodiments of the present application are introduced below in conjunction with the drawings in the specification.

[0123] Figure 4 Schematic diagram of the process of building a fusion model provided in the embodiment of the present application Figure 1 .like Figure 4 As shown, in the embodiment of the present application, the execution subject may be a fusion model construction device, which may be located in an electronic device, which may be a data processing server. The fusion model construction method provided in the embodiment of the present application includes the following steps:

[0124] S301, performing multiple independent random assignments on multiple crack growth parameters of respective preset multiple failure mechanism models, and obtaining a first expression of each failure mechanism model after each random assignment.

[0125] Specifically, in an experimental environment, multiple crack growth parameters c and m of the failure mechanism model are usually measured in the experimental environment. Due to the differences between the actual use of the mechanical structure and the experimental environment, the c and m obtained in the experimental environment cannot fully reflect the actual situation and there is uncertainty. Therefore, multiple independent random assignments are made to c and m.

[0126] Taking a failure mechanism model as an example, for each random assignment of c and m, a first expression of the model can be obtained; for n independent random assignments, n first expressions of the model can be obtained. Correspondingly, for multiple failure mechanism models, for n independent random assignments, ∑n first expressions can be obtained.

[0127] S302. Obtain the first measured data of the first fatigue crack of the target metal material.

[0128] Specifically, the first measured data is used to indicate the crack propagation of the first fatigue crack. Then, according to the first measured data, the initial crack size of the first fatigue crack can be known. Based on the initial crack size of the first fatigue crack, the fatigue life prediction curves of each of the ∑n first expressions, that is, the a-N curves, can be obtained. Bayesian inference assumes that the uncertainties of c and m both follow a certain probability distribution, thus making the prediction more accurate. In the prior art, the determination of prior data usually relies on experimental tests, previous statistical data, and the professional insights and experience accumulation of technicians. When there is a lack of sufficient measured data, it can be assumed to be a certain uniform distribution. Then, for each failure mechanism model, the probability that each a-N curve is the actual crack growth curve is equal, all being 1 / n. These ∑n a-N curves constitute the prior data of Bayesian inference. Therefore, according to the first measured data, Bayesian update can be performed on these prior data.

[0129] S303. According to the first measured data, perform Bayesian update on the prior probability of each preset first expression to obtain the posterior probability of each first expression.

[0130] Specifically, for each a-N curve, the calculation formula of Bayesian inference is:

[0131]

[0132] Among them, P(a-N) is the prior probability of each a-N curve, and in the first cycle, its value is 1 / n; P(measured data 丨 a-N curve) is the likelihood function, and P(measured data) is a constant term closely related to the first measured data; P(a-N 丨 measured data) is the posterior probability of each a-N curve. The prior probability and posterior probability of each first expression are both used to indicate the corresponding first expression's ability to describe the crack propagation situation.

[0133] For each failure mechanism model, the probability that the multiple aN curves of the model are actual crack extension curves is equal to 1, so P (measured data) as a constant related to the first measured data does not need to be calculated. When the prior probability of the aN curve is known (i.e. 1 / n), the first measurement is performed to obtain the initial crack size, and the posterior probability of the aN curve can be obtained according to the Bayesian update.

[0134] When the actual measurement is performed again, the first posterior probability of the aN curve can be used as the prior probability of the second actual measurement, and the posterior probability of the second actual measurement can be obtained again according to the Bayesian update. Continuous actual measurements can be performed, and the probability of the aN curve being the actual crack extension can be continuously updated according to the Bayesian update to obtain a more reasonable fatigue prediction result.

[0135] S304 . Calculate a second expression for each failure mechanism model according to the multiple first expressions for each failure mechanism model and the posterior probability of the corresponding first expressions.

[0136] Specifically, after performing Bayesian updating and obtaining the posterior probability of each aN curve, the mean μ of c can be calculated. c , and the mean μ of m m , the calculation formula is:

[0137] μ c =∑cP(c)

[0138] μ m =∑mP(m)

[0139] Among them, P(c) and P(m) represent the posterior probabilities of c and m in each aN curve. c , as c of the second expression of the model, and μ of each failure mechanism model m , as m of the second expression of the model, the second expression of each failure mechanism model can be obtained.

[0140] S305. Obtain second measured data of a second fatigue crack of the target metal material.

[0141] Specifically, the second measured data is used to indicate the crack extension of the second fatigue crack. Based on the second measured data, the initial crack size of the second fatigue crack can be known; wherein the sampling time and / or sampling position of the second fatigue crack is different from that of the first fatigue crack.

[0142] Based on the initial crack size of the second measured data, the a-N curves of each of the multiple second expressions can be obtained. These a-N curves constitute the prior data for Bayesian inference. The greater the prior probability of the second expression, the more the corresponding failure mechanism model can describe the actual crack growth situation. The prior probability of the failure mechanism model is determined by the actual environment. Since the usage environment and applicable stage of each model are different, the prior probabilities are also different. In fatigue life prediction, due to the uncertainties such as the crack occurrence location, stress and deformation mode, and crack size, it is difficult to determine which model is more applicable, and it can only be assumed that all models have equal probabilities. Therefore, based on the second measured data, the prior data of each second expression can be updated by Bayesian method.

[0143] S306. According to the second measured data, perform Bayesian update on the prior probability of each preset second expression to obtain the posterior probability of each second expression.

[0144] Specifically, for each a-N curve, the calculation formula for Bayesian update is:

[0145]

[0146] where \(i\leq k\), \(Pr(M k )\) is the prior probability of the \(k\)th failure mechanism model \(M k \) among the multiple failure mechanism models \(M\), and \(k\) is a positive integer; \(L(D|M k )\) is the likelihood function of \(M k \) under the second measured data \(D\). The magnitude of the likelihood function value of \(M k \) reflects the ability of \(M k \) to describe the actual crack growth situation compared with other failure mechanism models. If the likelihood function value of \(M k \) is high, its ability to describe the actual crack growth situation is strong compared with other failure mechanism models; if the likelihood function value of \(M k \) is low, its ability to describe the actual crack growth situation is weak compared with other failure mechanism models. Since in fatigue life prediction, the prior probabilities of the models are all equal, the likelihood function then becomes the only metric that can distinguish the ability of each model to describe the actual crack growth.

[0147] S307. Calculate the fusion model based on the second expression of each failure mechanism model and the corresponding posterior probability of the second expression, so as to predict the fatigue life of the target metal material according to the fusion model.

[0148] Specifically, the uncertainty of failure mechanism model selection can be quantified by introducing the model probability method, and the impact of failure mechanism model selection uncertainty on fatigue life prediction can be handled by model fusion. This method probabilistically quantifies the ability of a set of deterministic failure mechanism models to describe fatigue life prediction, and then combines multiple models into a single fusion model by adjusting the posterior model probability of each failure mechanism model to predict fatigue life.

[0149] The method of obtaining the fusion model is essentially a probability and weighted method. First, it is assumed that each failure mechanism model has the same ability to describe the actual crack extension, that is, the prior probability is equal; then, the posterior probability of each failure mechanism model is updated through the second measured data, that is, the weight of each failure mechanism model is adjusted according to the second measured data, and the crack extension is described through the weighted fusion model.

[0150] The embodiment of the present application provides a method for constructing a fusion model, which includes: performing multiple independent random assignments on crack growth parameters to obtain multiple first expressions of each failure mechanism model; performing Bayesian updates on the prior probability of each first expression according to the first measured data; calculating the second expression of each failure mechanism model according to the multiple first expressions and the posterior probability of each first expression; performing Bayesian updates on the prior probability of each second expression according to the second measured data; calculating the fusion model according to the multiple second expressions and the posterior probability of each second expression. The following technical effects are achieved: performing Bayesian updates on the prior probabilities of the multiple first expressions of each failure mechanism model solves the problem of uncertainty of the multiple crack growth parameters of the failure mechanism model; performing Bayesian updates on the prior probabilities of the second expressions of each of the multiple failure mechanism models solves the problem that different failure mechanism models cannot be accurately applied to various objects and scenarios; determining the expression form of the likelihood function through measured data, and implementing the parameter update of the failure mechanism model through Bayesian updates, thereby improving the accuracy of fatigue life prediction.

[0151] Figure 5 Schematic diagram of the process of building a fusion model provided in the embodiment of the present application Figure 2 ,like Figure 5 As shown, the fusion model construction method provided in the embodiment of the present application is Figure 4On the basis of the fusion model construction method provided in the embodiment, the first measured data includes: the measured crack size of the target metal material in each of the multiple first preset cycles, the first target cycle is any one of the multiple first preset cycles; the first target expression is any one of the multiple first expressions of the target failure mechanism model, and the target failure mechanism model is any one of the multiple failure mechanism models; for the first target cycle and the first target expression, S303 includes:

[0152] S401. Obtain a first measured crack size of a target metal material in a first target cycle according to first measured data.

[0153] Specifically, due to the lack of sufficient test information and previous statistical data, the first measured crack size of the target metal material in the first target cycle can be obtained based on the first measured data.

[0154] S402. Obtain a first predicted crack size of a target metal material in a first target cycle according to a first target expression.

[0155] It should be noted that there is no execution order for S401 and S402. S401 may be executed first and then S402 as provided in the embodiment of the present application, or S402 may be executed first and then S401, or S401 and S402 may be executed simultaneously.

[0156] S403. Obtain a first likelihood function of the first target expression in the first target cycle according to an error between the first predicted crack size and the first measured crack size.

[0157] Specifically, according to Bayesian inference, the posterior probability of the first target expression in the first target cycle is proportional to the product of the first likelihood function and the prior probability of the first target cycle. The determination of the first likelihood function is related to the first measured data. Here, it is assumed that the first likelihood function obeys a normal distribution. The calculation formula of the first likelihood function is:

[0158]

[0159] Among them, a h is the actual crack size of the first target cycle, a is the first predicted crack size of the first target cycle, s is the variance, the value of s is related to the first measured data, and the value of s in the embodiment of the present application can be 1. According to the above formula, the value of the first likelihood function depends on a. h The error between a and a. When the error is small, the value of the first likelihood function is large, and the maximum value is 1, which means that a h is consistent with a; when the error is large, the value of the first likelihood function is small, and the maximum value is approximately 0, which means that a hThe difference from a is huge. Therefore, the first likelihood function can be regarded as the weight of each first expression being an actual expression. The smaller the error, the larger the value of the first likelihood function, and the greater the possibility that the first expression is an actual expression.

[0160] S404. Perform a Bayesian update on the prior probability of the first target expression in the first target cycle according to the first likelihood function to obtain the posterior probability of the first target expression in the first target cycle.

[0161] It should be noted that, for other first preset cycles outside the first target cycle, and other first expressions outside the first target expression, the calculation method of the posterior probability is similar, and will not be repeated in this embodiment.

[0162] In a possible design, the prior probabilities of the multiple first expressions of the target failure mechanism model in the first cycle obey the mean distribution; the likelihood functions of the multiple first expressions of the target failure mechanism model in the first target cycle obey the normal distribution; for the first target cycle and the target failure mechanism model, after S404, the fusion model construction method further includes:

[0163] S405 , performing probability normalization processing on the respective posterior probabilities of the multiple first expressions of the target failure mechanism model in the first target cycle.

[0164] Specifically, the probability sum of the aN curves of the multiple first expressions is respectively the actual crack extension curve is equal to 1, that is, the posterior probabilities of the multiple first expressions are required to add up to 1. Therefore, each time after the posterior probability of the first expression is obtained, probability normalization processing is performed.

[0165] In a possible design, for the target failure mechanism model, S304 includes:

[0166] According to the respective posterior probabilities of the multiple first expressions of the target failure mechanism model, a weighted calculation is performed on the multiple crack growth parameters of each first expression of the target failure mechanism model to obtain a calculated value of each crack growth parameter.

[0167] A second expression of the target failure mechanism model is obtained according to the calculated values ​​of each of the plurality of crack growth parameters.

[0168] Specifically, in the above embodiment, the mean value μ of c is calculated as c , and the mean μ of m m , which is equivalent to a weighted calculation. In addition, we can also use μ c , calculate the variance σ of c c , and according to μ m , calculate the variance σ of m m , the calculation formula is:

[0169] σ c =∑(c-μ c ) 2 P(c)

[0170] σ m =∑(m-μ m ) 2 P(m)

[0171] Among them, σ c and σ m The degree of discretization of the second expression used to represent the target failure mechanism model.

[0172] It should be noted that, for other failure mechanism models other than the target failure mechanism model, the calculation method of the respective second expressions is similar, and will not be described in detail in this embodiment.

[0173] In a possible design, after S304, the fusion model building method further includes:

[0174] Third measured data of each of a plurality of third fatigue cracks of the target metal material is obtained.

[0175] According to the plurality of third measured data, parameters of the second expression of each failure mechanism model are updated respectively.

[0176] Specifically, the sampling time and / or sampling position of each third fatigue crack are different from those of the first fatigue crack. Due to the measurement error of the detection equipment, the initial crack size detected is also uncertain. Therefore, for the information such as the crack initiation point and crack size obtained at different sampling times and / or sampling positions in the actual measurement, the above steps are repeated to obtain the real-time update of the target failure mechanism model of each key part of the target metal material, and the model is corrected accordingly. The technical effect is: more accurate and reliable crack size prediction and remaining life assessment can be made than before the update.

[0177] The following will provide a further detailed description of how to obtain the second expression of the Paris-Erdogan model based on multiple first expressions of the Paris-Erdogan model in conjunction with a specific embodiment.

[0178] Taking the through crack on the fuselage panel as an example, according to the first measured data, a0=10mm; in addition, according to the parameter table of the target metal material, the limit crack size a can be obtained. c =25.6mm. The parameter table of the target metal material is shown in Table 1.

[0179] Table 1:

[0180] parameter Physical meaning Value / Interval unit Distribution Type <![CDATA[a0]]> Initial crack size 10 mm Fixed value β Correction factor 1 Fixed value <![CDATA[△σ h ]]> stress 78.6 MPa Fixed value c Failure Parameters <![CDATA[U(2.1×10 -10 ,4.1×10 -10 )]]> random variable m Failure Parameters U(3.32，3.52) random variable <![CDATA[a c ]]> Limit crack size 25.6 mm Fixed value <![CDATA[△K th ]]> Threshold 1.39 <![CDATA[MPam 0.5 ]]> Fixed value

[0181] For the prior distribution, the first expression of the Paris-Erdogan model is obtained by sampling the probability distribution of c and m. If the sample size is large enough, the set of all first expressions can reflect the actual crack extension. In order to explain the method more concisely, 10 groups of samples are taken as an example to perform Bayesian update. The 10 groups of samples [m, c] are: [3.35, 2.1×10 -10 ], [3.35, 2.5×10 -10 ], [3.35, 3×10 -10 ], [3.35, 3.5×10 -10 ], [3.35, 4.1×10 -10 ], [3.45, 2.1×10 -10 ], [3.45, 2.5×10 -10 ], [3.45, 3×10 -10 ], [3.45, 3.5×10 -10 ] and [3.45, 4.1×10 -10 ]. Figure 6 Schematic diagram of aN curve cluster of the Paris-Erdogan model provided in the embodiment of the present application. The aN curves of the 10 groups of samples [m, c] are as follows Figure 6 shown.

[0182] The prior information of these 10 aN curves is shown in Table 2.

[0183] Table 2:

[0184]

[0185] From Table 2 and Figure 6 It can be seen that before 1879 cycles, no aN curve has a predicted crack size exceeding a c ; At 2000 cycles, only [3.45, 4.1×10 -10 ] the predicted crack size exceeds a c , the corresponding fatigue life is 1879 cycles, so the prior reliability of the target metal material at 2000 cycles is 0.9. At 3000 cycles, there are 5 groups [3.35, 3.5×10 -10 ], [3.35, 4.1×10 -10 ], [3.45, 3×10 -10 ], [3.45, 3.5×10 -10 ] and [3.45, 4.1×10 -10 ] the predicted crack size exceeds a c, so the a priori reliability of the target metal material at 3000 cycles is 0.5. The predicted crack size of all aN curves will exceed a after 4875 cycles. c . Figure 7 Prior cumulative probability distribution provided for the embodiment of the present application Figure 1 ,like Figure 7 As shown, the prior probability distribution of crack size at 250, 1000, 2000 and 3000 cycles can be obtained.

[0186] After that, the prior probability is updated by Bayesian to obtain the posterior probability. The calculation formula of the posterior probability is:

[0187]

[0188] Where p(c, m) is the prior probability, which takes a value of 0.1. When the actual crack size is detected to be 11.2 mm after 250 cycles, the posterior probability can be obtained according to the above formula. The posterior information of these 10 aN curves is shown in Table 3.

[0189] Table 3:

[0190]

[0191] From Tables 2 and 3, it can be seen that the posterior cumulative probability distribution of the predicted crack size and the posterior reliability of the fatigue life are compared with those before the update. Figure 8 A comparison diagram of the posterior probability distribution provided in the embodiment of the present application and before the update; Fig. 9 Comparison of the a posteriori reliability provided in the embodiment of the present application with that before the update Figure 1 .like Figure 8 As shown in Figure 2, the posterior cumulative probability distribution of the predicted crack size is compared with that before the update. Fig. 9 As shown, the relationship between the reliability R(t) and the number of cycles is shown, compared with that before the update, where the posterior reliability of the fatigue life is shown as the solid line in the figure, and the prior reliability is shown as the dotted line in the figure.

[0192] Finally, we calculate μ c =3.37×10 -10 , σ c =0.466×10 -10 , μ m =3.44,σ m =0.003359.

[0193] It should be noted that the above is just an example of 10 groups of samples to briefly introduce the process of obtaining the second expression. When c and m are randomly assigned multiple times independently, the aN curve under a given confidence level can be obtained.

[0194] Fig.10 Schematic diagram of the process of building a fusion model provided in the embodiment of the present application Figure 3 ,like Fig.10 As shown, the fusion model construction method provided in the embodiment of the present application is Figure 4 Based on the embodiment, the second measured data includes: the measured crack size of the target metal material in multiple second preset cycles, and the second target cycle is any one of the multiple second preset cycles; for the second target cycle, S306 includes:

[0195] S501. Obtain a second measured crack size of a target metal material in a second target cycle according to second measured data.

[0196] S502. Obtain, according to the plurality of second expressions, a second predicted crack size of the target metal material in a second target cycle predicted by each second expression.

[0197] S503. Obtain a second likelihood function of each second expression in a second target cycle according to the errors between each second predicted crack size and the second measured crack size.

[0198] Specifically, the prior probability of the failure mechanism model in the second target cycle is the ability of the model to describe the crack propagation phenomenon. If the model can fully describe the crack propagation phenomenon, the prior probability of the model is 1. Assuming that there are multiple failure mechanism models that can be used to describe the crack propagation phenomenon, and there is a second measured data D, then the prior probability of the model can be updated using Bayesian inference.

[0199] Since the failure mechanism model is only an approximate description of the actual crack extension, it is inevitable that there is an error between each second predicted crack size and the second measured crack size. Before the second measured data is obtained, the error is unknown. The unknown error is generally described by the uncertainty method. More specifically, the uncertainty of the error is usually described by a certain probability distribution. In statistics, the error is usually assumed to obey a normal distribution with a mean of 0 and a constant variance.

[0200] The calculation formula for the error between each second predicted crack size and the second measured crack size is:

[0201] y=f k +ε k

[0202] ε k ~N(0,σ k 2 )

[0203] Where y is the second measured crack size, f kis a failure mechanism model M that describes y k The second predicted crack size, ε k is the error between y and f k ε k follows a normal distribution with a mean of 0 and a variance of σ k 2 The purpose of introducing ε k is to describe the uncertainty difference between the second predicted crack size and the second measured crack size. When ε k takes the mean value of 0, it does not change the f k of M k , that is, M k is the actual model. This also reflects that when each failure mechanism model is established, it is considered that its second predicted crack size is most likely to reflect the second measured crack size. ε k reflects the mathematical relationship between M k and y. According to this formula, it can be seen that y also follows a normal distribution. Then, based on M k the prior distribution is:

[0204]

[0205] ε kn = d n - f kn

[0206] where d n is a measured crack size of the second measured data D = {d1,..., d N}. f kn is the predicted crack size of M k corresponding to d n . ε kn is a random variable that follows a normal distribution with a mean of 0 and a standard deviation of σ k . Assuming that the measured data are independent of each other, since the likelihood function L(D|σ k , M k ) is the product of the likelihood function L(d n |σ k , M k ), then L(D|σ k , M k ) can be expressed as:

[0207]

[0208] Then the second likelihood function L(D|M k ) can be obtained from the marginal distribution of the integral of L(D|σ k , M k ) with respect to the parameter. The calculation formula is:

[0209] L(D|M k ) = ∫L(D|σ k , M k )g(σ k |M k )dσ k

[0210] where g(σ k |M k ) is the conditional distribution of the unknown parameter σ k given M k . For given D and M k , σ k can be obtained by the maximum likelihood principle.

[0211] S504. Update the prior probability of each second expression in the second target loop according to the second likelihood function of each second expression in the second target loop to obtain the posterior probability of each second expression in the second target loop.

[0212] Specifically, the simplest way to estimate model parameters using the likelihood function is to find a set of parameters that maximize the value of the likelihood function. This estimation method is called Maximum Likelihood Estimation (MLE). MLE makes the occurrence probability of the observed data the highest. From the perspective of computational simplicity, we usually take the logarithm of the likelihood function first and then maximize it. The parameter estimates obtained by these two methods are the same. The logarithmic form of the likelihood function is called the log-likelihood function. The log-likelihood function is the sum of the logarithms of the probability densities of each observed value, while the likelihood function is the product of the sampling probability densities of each observed value.

[0213] σ k 's maximum likelihood estimate value is obviously the maximum possible value that makes D detected, and can be obtained by setting the logarithm of L(d n |σ k , M k ) equal to 0 with respect to σ k . The calculation formula is:

[0214]

[0215] Solving gives the expression for σ k 2 as:

[0216]

[0217] where σ k 2 is the parameter obtained by fitting the second measured data and can also be denoted as σ' k2 Substitute the maximum likelihood estimate value of σ k 2 into the expression of L(D|σ k , M k ), and the expression of L(D|M k ) can be obtained as follows:

[0218]

[0219] It should be noted that the condition for this expression to hold is that the errors are independent and follow a normal distribution.

[0220] Under the assumption that the prior probability follows a uniform distribution, that is, each failure mechanism model can completely describe the second measured data, the calculation formula for the posterior probability of the second expression can be obtained as follows:

[0221]

[0222] It should be noted that for other second preset cycles outside the second target cycle, the calculation method of the posterior probability is similar, and will not be elaborated in this embodiment.

[0223] In a possible design, the prior probabilities of multiple second expressions in the first cycle follow a mean distribution; the errors of multiple second expressions in the second target cycle follow a normal distribution; then after S504, the fusion model construction method further includes:

[0224] S505. Perform probability normalization on the posterior probabilities of multiple second expressions in the second target cycle respectively.

[0225] Specifically, the sum of the prior probabilities and posterior probabilities of all failure mechanism models is 1, so each time after obtaining the posterior probabilities of all failure mechanism models, probability normalization needs to be performed.

[0226] In a possible design, S307 includes:

[0227] For the second target cycle, based on multiple second expressions and the posterior probabilities of the corresponding second expressions in the second target cycle, calculate the segmented fusion model of the second target cycle by weighted calculation.

[0228] Fit the fusion model according to the segmented fusion models of each second preset cycle.

[0229] Specifically, during the construction of the fusion model, only the first predicted crack size distribution under the preset number of cycles of the failure mechanism model is fused, which is a probabilistic treatment and does not involve physical mechanisms, nor is an analytical model obtained. The posterior probability of the second expression essentially reflects the influence magnitude of each failure mechanism model in the fusion model, that is, the weight. The greater the posterior probability of the model, the higher the weight. Therefore, the analytical expression of the fusion model can be expressed as:

[0230] f(y丨D)=∑Pr(M k 丨D)f(y丨M k ,D)

[0231] where f(y丨M k ,D) is the prediction of the second measured data by M k on the premise of obtaining D, and Pr(M k 丨D) is the posterior probability of M k after obtaining D, which is the weight coefficient of M k relative to other models. According to this expression, the expression for the description of the crack growth situation by the fusion model can be obtained as:

[0232] g(y丨D)=∑Pr(M k 丨D)g(y丨M k ,D)

[0233] where g(y丨M k ,D) is the prediction of the second measured data by M k on the premise of obtaining D.

[0234] If g(y丨M k ,D) has a mean E(y丨D) and a variance Var(y丨D), then the formulas for the mean and variance of the fusion model g(y丨D) are:

[0235] E(y丨D)=∑Pr(M k 丨D)E(y丨M k ,D)

[0236] Var(y丨D)

[0237] =∑Pr(M k 丨D)Var(y丨M k ,D)

[0238] +∑Pr(M k 丨D)[E(M k 丨D)-E(M k 丨D)] 2

[0239] where ∑Pr(M k|D)Var(y|M k , D) is the within-model variance, which reflects the uncertainty level of each failure mechanism model. Pr(M k |D)[E(M k |D) - E(M k |D)] 2 is the between-model variance, which reflects the uncertainty of the model form, that is, the uncertainty level of the target failure mechanism model relative to the fusion model.

[0240] Since g(y|M k , D) follows a normal distribution, the linear combination g(y|D) of g(y|M k , D) also follows a normal distribution. The distribution expression of the fusion model g(y|D) is:

[0241] g(y|D) = N{∑Pr(M k |D)f k , ∑Pr(M k |D)σ′ k 2 , ∑Pr(M k |D)[f k - E(y|D)] 2}

[0242] The following will further elaborate on obtaining the fusion model according to multiple second expressions in combination with specific embodiments.

[0243] The above embodiments only use the Paris-Erdogan model to describe fatigue crack growth, but the Paris-Erdogan model does not consider the influence of the stress intensity factor threshold on crack growth. Here, two fatigue growth models, the Paris-Erdogan model and the Trantina-Johnson model, are comprehensively considered to obtain the fusion model. Among them, the expression of the Trantina-Johnson model is:

[0244]

[0245] Assume that c and m follow a uniform distribution and are independent of each other, and take c = 4.1×10 -10 , m = 3.49. Then the a-N curves of the Paris-Erdogan model and the Trantina-Johnson model and the measured crack curve are as Fig.11 shown, Fig.11 is the comparison diagram of the fusion model, prediction model and measured data provided by the embodiment of the present application. Among them, the a-N curve of the fusion model is shown as the dotted line in the figure, and the a-N curve of the measured data is shown as the dash-dotted line in the figure. From Fig.11It can be seen that when using a single failure mechanism model to describe the fatigue crack growth phenomenon, the prediction curve is not accurate enough. Each failure mechanism model can only partially describe the fatigue crack growth phenomenon. The Paris-Erdogan model can well describe the actual fatigue crack growth phenomenon when the number of cycles is low, below about 1000 cycles. However, at high cycle numbers, the crack grows exponentially and too fast, and the error from the curve of the second measured data also becomes larger and larger. The prediction curve of the Trantina-Johnson model is lower than the actual fatigue crack growth curve.

[0246] Using Bayesian inference and combining with the second measured data, a fusion model that combines the Paris-Erdogan model and the Trantina-Johnson model can be obtained. From the formula of the mean E(y丨D) of the fusion model, the mean curve of the fitted fusion model is obtained and compared with the Paris-Erdogan model and the Trantina-Johnson model. After combining the two using Bayesian inference with the second measured data, they complement each other, so the accuracy of the fusion model is improved.

[0247] After that, the posterior probability of the fusion model is calculated. According to the formula for calculating the posterior probability, the posterior model probability of the Paris-Erdogan model is 0.9029, and the posterior model probability of the Trantina-Johnson model is 0.0971. A comparison of the Paris-Erdogan model, the Trantina-Johnson model, and the fusion model is shown in Table 4.

[0248] Table 4

[0249]

[0250] According to the posterior probabilities of the Paris-Erdogan model and the Trantina-Johnson model, the expression of the fusion model can be obtained as:

[0251]

[0252] Fig.12 This is a schematic flowchart of the health assessment method provided by the embodiment of the present application. As Fig.12 shown, in the embodiment of the present application, the execution subject can be a health assessment device, which can be located in an electronic device, and this device can be a data processing server. Then, the health assessment method provided by the embodiment of the present application includes the following steps:

[0253] S601. Obtain the fourth measured data of the fourth fatigue crack of the target metal material.

[0254] The fourth measured data includes: an initial crack size of a fourth fatigue crack.

[0255] S602: Evaluate at least one of a plurality of health assessment indicators of the target metal material according to the fourth measured data and a preset fusion model.

[0256] The fusion model is a model constructed by the fusion model construction method of the above embodiment. Multiple health assessment indicators include: fatigue life, reliability and crack size, among which the fatigue life and crack size of the target metal material are predicted similarly to the above embodiment, and the embodiments of this application will not be repeated.

[0257] After predicting the fatigue life and crack size of the target metal material, the reliability of the prediction result can be verified based on the fourth measured data.

[0258] Specifically, the fusion model comprehensively considers multiple failure mechanism models, combines the advantages of each failure mechanism model, and more comprehensively describes the actual fatigue crack propagation phenomenon. However, model fusion only takes into account the differences between models, while the parameters within the model reflect the model's understanding of the fatigue crack propagation phenomenon, which is not reflected in the fusion model. The fusion model, as well as the confidence interval obtained based on the fusion model, are completely derived from mathematical derivations, divorced from the physical essence, and have excessive dispersion and negative values. This requires a Bayesian update of the prior probability of the fusion model based on the fusion model to obtain the posterior probability of the fusion model.

[0259] In one possible design, the evaluation results include: the predicted crack size of the target metal material at each cycle;

[0260] After S602, the health assessment method further includes:

[0261] S603: Obtain a confidence interval according to the predicted crack size of the target metal material in each cycle and a preset prediction confidence level.

[0262] Specifically, for the fatigue life prediction of the target metal material, in order to reflect its reliability, a confidence interval needs to be given. The confidence interval includes: the upper confidence crack size and the lower confidence crack size of the target metal material in each cycle. Assuming that the prediction confidence is 95%, the expression of the confidence interval is:

[0263]

[0264] Fig.13 A schematic diagram of the confidence interval provided in the embodiment of the present application. Fig.13It can be seen that the aN curve of the fusion model is shown as the solid line in the figure, the aN curve of the confidence interval is shown as the dotted line in the figure, and the aN curve of the measured data is shown as the dotted line in the figure. The confidence interval with a confidence level of 95% completely contains the actual fatigue crack growth curve, and the accuracy is improved compared with the single failure mechanism model.

[0265] Fig.14 Prior cumulative probability distribution provided for the embodiment of the present application Figure 2 When the number of [m, c] is much larger than 10, for example, 1000, the prior reliability of the fusion model in 250, 1000, 2000 and 3000 cycles is as follows: Fig.14 shown.

[0266] According to the fourth measured data, the measured crack size of each cycle of the fourth fatigue crack is shown in Table 5.

[0267] Table 5

[0268]

[0269] Fig.15 A posterior probability distribution diagram provided for an embodiment of the present application; Fig.16 Comparison of the a posteriori reliability provided in the embodiment of the present application with that before the update Figure 2 .like Fig.15 As shown, the posterior probability distribution of the predicted crack size is shown; Fig.16 As shown, the relationship between reliability and number of cycles is shown, compared with that before updating, where the posterior reliability of fatigue life is shown as the solid line in the figure, and the prior reliability is shown as the dotted line in the figure.

[0270] In the embodiment of the present application, the fusion model construction device may be located in an electronic device. The fusion model construction device includes:

[0271] A random assignment module, used for performing multiple independent random assignments on multiple crack growth parameters of each of the preset multiple failure mechanism models, and obtaining a first expression of each failure mechanism model after each random assignment;

[0272] A first updating module is used to obtain first measured data of a first fatigue crack of a target metal material, and to perform a Bayesian update on a prior probability of each preset first expression according to the first measured data, so as to obtain a posterior probability of each first expression; wherein the first measured data is used to indicate a crack propagation condition of the first fatigue crack, and the prior probability and the posterior probability of each first expression are used to indicate the ability of the corresponding first expression to describe the crack propagation condition;

[0273] A first calculation module, used for calculating a second expression of each failure mechanism model according to a plurality of first expressions of each failure mechanism model and a posterior probability of the corresponding first expression;

[0274] A second updating module is used to obtain second measured data of a second fatigue crack of the target metal material, and perform Bayesian update on the prior probability of each preset second expression according to the second measured data to obtain a posterior probability of each second expression;

[0275] The second calculation module is used to calculate the fusion model according to the second expression of each failure mechanism model and the posterior probability of the corresponding second expression, so as to predict the fatigue life of the target metal material according to the fusion model.

[0276] In a possible design, the first measured data includes: measured crack sizes of the target metal material in each of a plurality of first preset cycles, the first target cycle being any one of the plurality of first preset cycles; the first target expression being any one of a plurality of first expressions of the target failure mechanism model, the target failure mechanism model being any one of a plurality of failure mechanism models;

[0277] For the first target loop and the first target expression, the first update module includes:

[0278] A first prediction module, used to obtain a first measured crack size of the target metal material in a first target cycle according to the first measured data, and to obtain a first predicted crack size of the target metal material in the first target cycle according to the first target expression;

[0279] A first error module, used for obtaining a first likelihood function of a first target expression in a first target cycle according to an error between a first predicted crack size and a first measured crack size;

[0280] The third updating module is used to perform Bayesian update on the prior probability of the first target expression in the first target cycle according to the first likelihood function to obtain the posterior probability of the first target expression in the first target cycle.

[0281] In a possible design, the prior probabilities of the multiple first expressions of the target failure mechanism model in the first cycle each obey the mean distribution; the likelihood functions of the multiple first expressions of the target failure mechanism model in the first target cycle each obey the normal distribution;

[0282] For the first target cycle and the target failure mechanism model, the fusion model building device also includes:

[0283] The first normalization module is used to perform probability normalization processing on the respective posterior probabilities of the plurality of first expressions of the target failure mechanism model in the first target cycle.

[0284] In a possible design, for a target failure mechanism model, a first calculation module includes:

[0285] A first weighting module, used for performing weighted calculation on multiple crack growth parameters of each first expression of the target failure mechanism model according to respective posterior probabilities of the multiple first expressions of the target failure mechanism model, to obtain a calculated value of each crack growth parameter;

[0286] The third calculation module is used to obtain a second expression of the target failure mechanism model according to the calculated values ​​of each of the plurality of crack growth parameters.

[0287] In a possible design, the fusion model building device further includes:

[0288] A first acquisition module is used to acquire third measured data of each of a plurality of third fatigue cracks of the target metal material; wherein the sampling time and / or sampling position of each third fatigue crack is different from that of the first fatigue crack;

[0289] The fourth updating module is used to update the parameters of the second expression of each failure mechanism model according to the plurality of third measured data.

[0290] In a possible design, the second measured data includes: measured crack sizes of the target metal material in each of a plurality of second preset cycles, the second target cycle being any one of the plurality of second preset cycles;

[0291] For the second target cycle, the second update module includes:

[0292] A second prediction module is used to obtain a second measured crack size of the target metal material in a second target cycle according to the second measured data, and to obtain a second predicted crack size of the target metal material in the second target cycle predicted by each second expression according to a plurality of second expressions;

[0293] A second error module, used for obtaining a second likelihood function of each second expression in a second target cycle according to the error between each second predicted crack size and the second measured crack size;

[0294] The fifth updating module is used to perform Bayesian update on the prior probability of each second expression in the second target cycle according to the second likelihood function of each second expression in the second target cycle, so as to obtain the posterior probability of each second expression in the second target cycle.

[0295] In a possible design, the prior probabilities of the plurality of second expressions in the first cycle follow a mean distribution; the errors of the plurality of second expressions in the second target cycle follow a normal distribution;

[0296] The fusion model building device also includes:

[0297] The second normalization module is used to perform probability normalization processing on the respective posterior probabilities of the plurality of second expressions in the second target cycle.

[0298] In one possible design, the second computing module includes:

[0299] A second weighting module is used for obtaining a segmented fusion model of the second target cycle by weighted calculation according to a plurality of second expressions and a posterior probability of the corresponding second expressions in the second target cycle for the second target cycle;

[0300] The fourth calculation module is used to fit the fusion model according to the segmented fusion model of each second preset cycle.

[0301] The fusion model construction device provided in the embodiment of the present application can execute Figures 4 to 11 The technical solution of the method embodiment shown in the figure has the same implementation principle and technical effect as Figures 4 to 11 The method embodiments shown are similar and will not be described in detail in the embodiments of the present application.

[0302] In the embodiment of the present application, the health assessment device may be located in an electronic device. The health assessment device includes:

[0303] A data acquisition module, used to acquire fourth measured data of a fourth fatigue crack of the target metal material; wherein the fourth measured data includes: an initial crack size of the fourth fatigue crack;

[0304] A life prediction module is used to evaluate at least one of multiple health assessment indicators of the target metal material based on the fourth measured data and a preset fusion model; wherein the fusion model is a model constructed by the health assessment device of the above embodiment; the multiple health assessment indicators include: fatigue life, reliability and crack size.

[0305] In one possible design, the fatigue life prediction results include: the predicted crack size of the target metal material at each cycle;

[0306] The health assessment device further comprises:

[0307] The confidence interval module is used to obtain a confidence interval based on the predicted crack size of the target metal material in each cycle and a preset prediction confidence level; wherein the confidence interval includes: an upper confidence crack size and a lower confidence crack size of the target metal material in each cycle.

[0308] The health assessment device provided in the embodiment of the present application can perform Figures 12 to 16 The technical solution of the method embodiment shown in the figure has the same implementation principle and technical effect as Figures 12 to 16 The method embodiments shown are similar and will not be described in detail in the embodiments of the present application.

[0309] The embodiment of the present application also provides an electronic device, the electronic device comprising: the electronic device comprising: at least one processor and a memory. The electronic device also comprises a communication component. The processor, the memory and the communication component are connected via a bus.

[0310] In the specific implementation process, at least one processor executes the computer execution instructions stored in the memory, so that the at least one processor is used to implement the fusion model construction method and health assessment method of the above-mentioned embodiment.

[0311] The specific implementation process of the processor can be found in the above-mentioned method embodiment, and its implementation principle and technical effect are similar, so the embodiments of this application will not be repeated here.

[0312] In the above embodiments, it should be understood that the processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), etc. The general-purpose processor may be a microprocessor or the processor may be any conventional processor, etc. The steps of the method disclosed in the application may be directly embodied as being executed by a hardware processor, or may be executed by a combination of hardware and software modules in the processor.

[0313] The memory may include a high-speed RAM memory, and may also include a non-volatile storage NVM, such as at least one disk storage.

[0314] The bus may be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, the bus in the drawings of the present application is not limited to only one bus or one type of bus.

[0315] The above functions implemented by the electronic device and the main control device introduce the scheme provided by the embodiment of the present application. It is understandable that in order to implement the above functions, the electronic device or the main control device includes a hardware structure and / or software module corresponding to each function. In combination with the units and algorithm steps of each example described in the embodiment disclosed in the embodiment of the present application, the embodiment of the present application can be implemented in the form of hardware or a combination of hardware and computer software. Whether a function is executed in the form of hardware or computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the technical solution of the embodiment of the present application.

[0316] The embodiment of the present application also provides a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, they are used to implement the fusion model construction method and health assessment method of the above embodiment. In the specific implementation of the above fusion model construction method and health assessment method, each module can be implemented as a processor.

[0317] The above-mentioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk. The readable storage medium can be any available medium that can be accessed by a general or special-purpose computer.

[0318] An exemplary readable storage medium is coupled to a processor so that the processor can read information from the readable storage medium and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can be located in an application specific integrated circuit (ASIC). Of course, the processor and the readable storage medium can also exist as discrete components in an electronic device or a main control device.

[0319] The embodiment of the present application also provides a computer program product, including a computer program, which, when executed by a processor, is used to implement the fusion model construction method and health assessment method of the above-mentioned embodiment.

[0320] The computer program is stored in a readable storage medium. At least one processor can read the computer program from the readable storage medium. At least one processor executes the computer program to execute the solution provided in any of the above embodiments.

[0321] Those skilled in the art can understand that all or part of the steps of implementing the above-mentioned application embodiments can be completed by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, the steps of the above-mentioned method embodiment are executed; and the aforementioned storage medium includes: ROM, RAM, disk or optical disk and other media that can store program codes.

[0322] So far, the technical solution of the present application has been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present application is obviously not limited to these specific embodiments, and the above embodiments are only used to illustrate the technical solution of the present application rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for constructing a fusion model, characterized in that: The method comprises: Performing multiple independent random assignments on multiple crack growth parameters of each of the preset multiple failure mechanism models to obtain a first expression of each failure mechanism model after each random assignment; Acquire first measured data of a first fatigue crack of a target metal material, and perform Bayesian update on a preset prior probability of each first expression according to the first measured data to obtain a posterior probability of each first expression; wherein the first measured data is used to indicate a crack propagation condition of the first fatigue crack, and the prior probability and the posterior probability of each first expression are used to indicate the ability of the corresponding first expression to describe the crack propagation condition; Calculate the second expression of each failure mechanism model according to the multiple first expressions of each failure mechanism model and the posterior probability of the corresponding first expression; Acquire second measured data of a second fatigue crack of the target metal material, and perform Bayesian update on a preset prior probability of each of the second expressions according to the second measured data to obtain a posterior probability of each of the second expressions; A fusion model is calculated based on the second expression of each failure mechanism model and the posterior probability of the corresponding second expression, so as to predict the fatigue life of the target metal material based on the fusion model.

2. The method according to claim 1, characterized in that The first measured data includes: the measured crack size of the target metal material in each of a plurality of first preset cycles, the first target cycle being any one of the plurality of first preset cycles; the first target expression being any one of a plurality of first expressions of a target failure mechanism model, the target failure mechanism model being any one of the plurality of failure mechanism models; For the first target cycle and the first target expression, performing Bayesian updating on the preset prior probability of each of the first expressions according to the first measured data to obtain the posterior probability of each of the first expressions includes: Obtaining a first measured crack size of the target metal material in the first target cycle according to the first measured data, and obtaining a first predicted crack size of the target metal material in the first target cycle according to the first target expression; Obtaining a first likelihood function of the first target expression in the first target cycle according to an error between the first predicted crack size and the first measured crack size; According to the first likelihood function, a Bayesian update is performed on the prior probability of the first target expression in the first target cycle to obtain the posterior probability of the first target expression in the first target cycle.

3. The method according to claim 2, characterized in that The prior probabilities of the first expressions of the target failure mechanism model in the first cycle obey the mean distribution; The likelihood functions of the plurality of first expressions of the target failure mechanism model in the first target cycle respectively obey a normal distribution; For the first target cycle and the target failure mechanism model, after performing Bayesian updating on the prior probability of the first target expression in the first target cycle according to the first likelihood function to obtain the posterior probability of the first target expression in the first target cycle, the method further includes: Probability normalization processing is performed on the respective posterior probabilities of the plurality of first expressions of the target failure mechanism model in the first target cycle.

4. The method according to claim 3, characterized in that: For the target failure mechanism model, the second expression of each failure mechanism model is calculated based on multiple first expressions of each failure mechanism model and the posterior probability of the corresponding first expression, including: According to the respective posterior probabilities of the plurality of the first expressions of the target failure mechanism model, a weighted calculation is performed on the plurality of crack growth parameters of each of the first expressions of the target failure mechanism model to obtain a calculated value of each of the crack growth parameters; A second expression of the target failure mechanism model is obtained according to the calculated values ​​of each of the plurality of crack growth parameters.

5. The method according to any one of claims 1 to 4, characterized in that: After calculating the second expression of each failure mechanism model according to the plurality of first expressions of each failure mechanism model and the posterior probability of the corresponding first expression, the method further comprises: Obtaining third measured data of each of a plurality of third fatigue cracks of the target metal material; wherein the sampling time and / or sampling position of each of the third fatigue cracks is different from that of the first fatigue crack; According to the plurality of third measured data, parameters of the second expression of each failure mechanism model are updated respectively.

6. The method according to claim 1, characterized in that The second measured data includes: measured crack sizes of the target metal material in each of a plurality of second preset cycles, the second target cycle being any one of the plurality of second preset cycles; For the second target cycle, performing Bayesian updating on the preset prior probability of each second expression according to the second measured data to obtain the posterior probability of each second expression includes: Obtaining a second measured crack size of the target metal material in the second target cycle according to the second measured data, and obtaining a second predicted crack size of the target metal material in the second target cycle predicted by each of the second expressions according to a plurality of the second expressions; Obtaining a second likelihood function of each second expression in the second target cycle according to the errors between each second predicted crack size and the second measured crack size; According to the second likelihood function of each second expression in the second target cycle, a Bayesian update is performed on the prior probability of each second expression in the second target cycle to obtain the posterior probability of each second expression in the second target cycle.

7. The method according to claim 6, characterized in that The prior probabilities of the plurality of the second expressions in the first cycle obey the mean distribution; the errors of the plurality of the second expressions in the second target cycle obey the normal distribution; After performing Bayesian updating on the prior probability of each second expression in the second target cycle according to the second likelihood function of each second expression in the second target cycle, the method further comprises: Probability normalization is performed on the respective posterior probabilities of the plurality of the second expressions in the second target cycle.

8. The method according to claim 7, characterized in that The fusion model is calculated based on the second expression of each failure mechanism model and the posterior probability of the corresponding second expression, including: For the second target cycle, according to the plurality of second expressions and the posterior probabilities of the corresponding second expressions in the second target cycle, a segmented fusion model of the second target cycle is obtained by weighted calculation; The fusion model is obtained by fitting according to the segmented fusion model of each second preset cycle.

9. A health assessment method, characterized in that: The method comprises: Acquire fourth measured data of a fourth fatigue crack of the target metal material; wherein the fourth measured data includes: an initial crack size of the fourth fatigue crack; According to the fourth measured data and the preset fusion model, at least one of the multiple health assessment indicators of the target metal material is evaluated; wherein the fusion model is a model constructed by the method described in any one of claims 1-8; the multiple health assessment indicators include: fatigue life, reliability and crack size.

10. The method according to claim 9, characterized in that The evaluation results include: predicted crack size of the target metal material at each cycle; After evaluating at least one of the plurality of health assessment indicators of the target metal material according to the fourth measured data and the preset fusion model, the method further includes: A confidence interval is obtained according to the predicted crack size of the target metal material in each cycle and a preset prediction confidence level; wherein the confidence interval includes: an upper confidence crack size and a lower confidence crack size of the target metal material in each cycle.

Images