High-temperature component reliability interval evaluation method, device, equipment, medium and product

Abstract

The application discloses a high-temperature component reliability interval evaluation method and device, equipment, medium and product, relates to the technical field of reliability evaluation, and the method comprises the following steps: performing probability damage evaluation on a high-temperature component to obtain stress and strain probability distribution of a dangerous point of the high-temperature component; a plurality of physical information neural network models are constructed and trained with the creep fatigue test conditions of the high-temperature component as input and test life as output; according to the life prediction results of each physical information neural network model, the life prediction interval and basic trust distribution of each model are constructed; the fusion life prediction interval is obtained by fusing through the Dempster method; and according to the fusion life prediction interval and the stress and strain probability distribution, the reliability interval evaluation result of the life of the high-temperature component is obtained. The application considers the model uncertainty caused by the selected model, has the advantages of high prediction accuracy, strong engineering applicability and good physical explanation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of reliability evaluation, in particular to a high-temperature component reliability interval evaluation method, device, equipment, medium and product. BACKGROUND

[0002] With the complication and extremization of long-term service conditions of high-temperature rotating mechanical components such as steam turbines and gas turbines, accurate evaluation of the reliability of high-temperature rotating mechanical components has become a key problem to ensure the safe and reliable operation of equipment. In the reliability evaluation process, the traditional method often only relies on a single model to give point estimates, ignoring the uncertainty differences between different models, which can easily lead to distorted results. How to effectively characterize and fuse model uncertainty and construct a more robust reliability evaluation method to provide reliable basis for safety analysis and maintenance strategy optimization of high-temperature rotating mechanical components is an important problem to be solved at present.

[0003] In creep-fatigue life prediction, the traditional physical method is simple and easy to use, and has strong generalization ability, but the fitting and prediction accuracy is not high; the data model has high prediction accuracy, but is highly dependent on the quality of the data set, and may appear in the prediction process. The prediction of the anti-physical law. Physical-data model has high accuracy and good generalization ability, and also has physical interpretability. However, no matter which kind of model can only give point estimates of life prediction, and cannot reflect the uncertainty of the prediction value, and is easily affected by extreme data, which further limits the accuracy of reliability evaluation. SUMMARY

[0004] The purpose of the present application is to provide a high-temperature component reliability interval evaluation method, device, equipment, medium and product, which can better realize the reliability evaluation of high-temperature rotating mechanical components, and has the advantages of high prediction accuracy, strong engineering applicability and good physical interpretability.

[0005] To achieve the above purpose, the present application provides the following solutions:

[0006] In a first aspect, the present application provides a high-temperature component reliability interval evaluation method, comprising:

[0007] Performing probability damage assessment on the high-temperature component to obtain the stress and strain probability distribution of the dangerous point of the high-temperature component.

[0008] Constructing and training a plurality of physical information neural network models with the creep-fatigue test conditions of the high-temperature component as input and the test life as output; the physical information neural network model includes a fatigue life prediction model and a creep life prediction model.

[0009] According to the life prediction results of each physical information neural network model, the life prediction interval and basic trust distribution of each model are constructed.

[0010] The Dempster method is used for fusion based on the life prediction interval and basic trust distribution of each model to obtain a fused life prediction interval.

[0011] According to the fused life prediction interval and the stress and strain probability distribution, a reliability interval evaluation result of the high-temperature component life is obtained.

[0012] In a second aspect, the application provides a high-temperature component reliability interval evaluation device, which comprises a probability damage evaluation module, a model training module, an interval construction and trust distribution module, a fusion module, and a reliability interval evaluation module.

[0013] The probability damage evaluation module is used for probability damage evaluation of the high-temperature component to obtain stress and strain probability distribution of the dangerous point of the high-temperature component.

[0014] The model training module is used for constructing and training a plurality of physical information neural network models with the creep fatigue test conditions of the high-temperature component as input and the test life as output; the physical information neural network models comprise fatigue life prediction models and creep life prediction models.

[0015] The interval construction and trust distribution module is used for constructing life prediction intervals and basic trust distributions of each model according to life prediction results of each physical information neural network model.

[0016] The fusion module is used for fusion based on the life prediction interval and basic trust distribution of each model by the Dempster method to obtain a fused life prediction interval.

[0017] The reliability interval evaluation module is used for obtaining a reliability interval evaluation result of the high-temperature component life according to the fused life prediction interval and the stress and strain probability distribution.

[0018] In a third aspect, the application provides a computer device, which comprises a memory and a processor to store a computer program on the memory and run the computer program on the processor, and the processor executes the computer program to realize steps of the high-temperature component reliability interval evaluation method in any one of the above.

[0019] In a fourth aspect, the application provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to realize steps of the high-temperature component reliability interval evaluation method in any one of the above.

[0020] In a fifth aspect, the application provides a computer program product, which comprises a computer program, and the computer program is executed by a processor to realize steps of the high-temperature component reliability interval evaluation method in any one of the above.

[0021] According to the specific embodiments provided in the application, the application discloses the following technical effects:

[0022] The application provides a high-temperature component reliability interval evaluation method, device, equipment, medium and product. By obtaining the stress and strain probability distribution of the dangerous point of the high-temperature component instead of a single determined value, the engineering practicability and decision robustness of the evaluation method are significantly improved. A plurality of physical information neural network models are constructed and trained, the cognitive uncertainty caused by the selected models is considered, and comprehensive reliability analysis is performed. Based on the life prediction interval and basic trust allocation of each model, the Dempster method is used for fusion to obtain a fused life prediction interval. The Dempster evidence theory is used to quantify the cognitive uncertainty. According to the fused life prediction interval and the stress and strain probability distribution, the reliability interval evaluation result of the high-temperature component life is obtained. The obtained reliability interval evaluation result can cover the real situation better than the traditional point estimation, reduces the misjudgment risk caused by ignoring the cognitive uncertainty, and makes the evaluation result more comprehensive and real. BRIEF DESCRIPTION OF DRAWINGS

[0023] In order to more clearly illustrate the technical solutions in the embodiments of the application or the prior art, the drawings needed in the embodiments will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor.

[0024] Figure 1 A flowchart of a high-temperature component reliability interval evaluation method provided by an embodiment of the application is shown in the figure.

[0025] Figure 2 A detailed flowchart of a high-temperature component reliability interval evaluation method provided by an embodiment of the application is shown in the figure.

[0026] Figure 3 A schematic diagram of finite element model meshing of a steam turbine rotor provided by an embodiment of the application is shown in the figure.

[0027] Figure 4 A finite element displacement cloud map of a steam turbine rotor provided by an embodiment of the application is shown in the figure.

[0028] Figure 5 A material creep fatigue test data set scatter plot provided by an embodiment of the application is shown in the figure. (a) is a scatter distribution of strain range, (b) is a scatter distribution of plastic strain range, (c) is a scatter distribution of stress range, (d) is a scatter distribution of average stress, (e) is a scatter distribution of strain rate, (f) is a scatter distribution of tensile load holding time, and (g) is a scatter distribution of compressive load holding time.

[0029] Figure 6 A schematic diagram of interval reliability evaluation results provided by an embodiment of the present application is shown in the following table.

[0030] Figure 7 A structural schematic diagram of a computer device provided by an embodiment of the present application is shown in the following table. DETAILED DESCRIPTION

[0031] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0032] The above purposes, features and advantages of the present application can be more obvious and easy to understand. The present application will be described in further detail below with reference to the drawings and specific embodiments.

[0033] First, the technical terms related to the embodiments of the present application are introduced.

[0034] The input of the life prediction model of the related art is generally the test condition or working condition, and the output is generally the predicted life. However, the creep-fatigue life has a very large dispersion. The same test condition or working condition can also result in a very large actual life. Therefore, the single-value life predicted cannot be completely accurate, and there is a certain uncertainty between the predicted life and the real life. In the present application, the physical information neural network and the evidence theory are combined to reflect this uncertainty. The physical information neural network is a data-physical driven life prediction method, and the evidence theory is a quantitative means of cognitive uncertainty.

[0035] In an exemplary embodiment, as shown in Figure 1 A high-temperature component reliability interval evaluation method is provided, including the following steps 201-205. Wherein:

[0036] Step 201, performing a probabilistic damage assessment on the high-temperature component to obtain the stress and strain probability distribution of the dangerous point of the high-temperature component.

[0037] Step 202, constructing and training a plurality of physical information neural network models with the creep-fatigue test condition of the high-temperature component as input and the test life as output; the physical information neural network model includes a fatigue life prediction model and a creep life prediction model.

[0038] In order to avoid high experimental costs, the creep fatigue test conditions of the application are based on the material grade high temperature component, the stress and strain of the dangerous point calculated according to the finite element model, the dangerous point around is regarded as a continuous medium, and the life prediction model of the material grade is applied for prediction.

[0039] Step 203, according to the life prediction results of each physical information neural network model, the life prediction interval and the basic trust distribution of each model are constructed.

[0040] Step 204, based on the life prediction interval and the basic trust distribution of each model, the fusion life prediction interval is obtained by the Dempster method.

[0041] Step 205, according to the fusion life prediction interval and the stress and strain probability distribution, the reliability interval evaluation result of the high temperature component life is obtained.

[0042] The implementation of the above steps 201 to 205, by obtaining the stress and strain probability distribution of the dangerous point of the high temperature component instead of a single determined value, significantly improves the engineering practicability and decision robustness of the evaluation method; construct and train multiple physical information neural network models, consider the cognitive uncertainty caused by the selection of different models, and conduct comprehensive reliability analysis; based on the life prediction interval and the basic trust distribution of each model, the fusion life prediction interval is obtained by the Dempster method, and the cognitive uncertainty is quantified by the Dempster evidence theory; according to the fusion life prediction interval and the stress and strain probability distribution, the reliability interval evaluation result of the high temperature component life is obtained, the obtained reliability interval evaluation result can cover the real situation better than the traditional point estimation, reduces the risk of misjudgment caused by ignoring cognitive uncertainty, and makes the evaluation result more comprehensive and real.

[0043] In another exemplary embodiment of the application, the above step 201 is replaced by the following steps 301~305:

[0044] Step 301, a three-dimensional finite element model is established based on the structural characteristics of the high temperature component.

[0045] Step 302, Latin hypercube sampling is used to extract multiple parameter combinations from the model parameters of the three-dimensional finite element model, and finite element simulation is carried out respectively to obtain the stress and strain response of the dangerous point of the high temperature component under different parameter combinations.

[0046] Step 303, the trained agent model is obtained by taking the parameter combination of the finite element simulation as input and the stress and strain response of the dangerous point as output.

[0047] Step 304, multiple random parameter combinations are obtained by Latin hypercube sampling.

[0048] Step 305, based on the trained surrogate model, calculate the stress and strain responses of the dangerous point under a plurality of random parameter combinations obtained by Latin hypercube sampling through Monte Carlo simulation, and obtain the stress and strain probability distribution of the dangerous point.

[0049] The probabilistic damage assessment of the high-temperature component (specifically, a high-temperature rotating machinery component, such as a steam turbine component) of the present application comprises:

[0050] S11: Establish a high-fidelity finite element model, assume that the model parameters follow a probability distribution, obtain different model parameter combinations through Latin hypercube sampling, and perform finite element simulation.

[0051] Optionally, the finite element model of the present application is generally a component thermal-structural coupling model established using ABAQUS software, and the model parameters include but are not limited to material performance parameters, temperature, rotational speed, geometric size, and external load.

[0052] S12: Based on the finite element simulation results, train a surrogate model that can quickly calculate the stress / strain response of the dangerous point through the model parameters.

[0053] Optionally, the surrogate model refers to a general interpolation regression method, including but not limited to neural network model, support vector machine regression, and Gaussian process regression.

[0054] S13: Through Monte Carlo simulation, calculate the stress / strain of the system under a large number of model parameter random combinations, and obtain the stress / strain probability distribution of the dangerous point under this working condition.

[0055] The training of the physical information neural network model of the present application comprises:

[0056] S21: Data preprocessing, physical equation parameter pre-fitting, construction of physical loss function, and initialization of model hyperparameters.

[0057] Among them, data preprocessing is mainly to divide the data into training set and test set for training and verification of the physical information neural network; the processed data are stress and strain responses (such as stress, strain, plastic strain, temperature, etc.) output in the finite element simulation.

[0058] The model hyperparameters include but are not limited to learning rate, Dropout probability, regularization weight, physical loss weight, and Adamw optimizer weight.

[0059] The physical equation refers to the fatigue life equation and the creep life equation, and the fatigue life equation includes but is not limited to the Manson-Coffin equation, the Morrow equation, and the Ostergren equation, and the creep life equation includes but is not limited to the time fraction method and the ductility depletion method.

[0060] S22: forward propagation of the neural network, calculation of data loss and physical loss, backward propagation of the neural network, until convergence.

[0061] The expression of the physical loss function is:

[0062] .

[0063] .

[0064] wherein, is the physical loss, is the number of samples, is the creep-fatigue life predicted by the neural network, is the creep-fatigue life predicted by the physical equation, and respectively represent the creep damage and fatigue damage calculated by the physical equation.

[0065] The total loss of the physical information neural network model is:

[0066] .

[0067] wherein, is the total loss of the physical information neural network model, is the data loss function, is the physical loss function, is the weight of the physical loss function.

[0068] S23: Bayesian optimization until the maximum number of iterations is reached, and the best model parameters (i.e. the best model hyperparameters obtained according to the optimization objective function) are output.

[0069] The Bayesian optimization objective function is:

[0070] .

[0071] wherein, is the Bayesian optimization objective function, is the determination coefficient of the test set, is the determination coefficient of the training set, and the objective function takes values in [-1, 1], and the larger the value, the better the model generalization ability.

[0072] The life interval evaluation based on evidence theory of the present application includes:

[0073] S31: constructing a life prediction interval and a basic belief assignment according to the results of cross-validation of the physical information neural network model.

[0074] The cross-validation described above is the model evaluation process, and the results of cross-validation are used here. For each physical information neural network with different physical constraints trained, a five-fold cross-validation result can be obtained. Based on this result, a model can be constructed to predict the lifetime prediction interval and basic confidence assignment for each input condition. The model used for cross-validation here is the model with the best model hyperparameters, that is, the best model obtained through training.

[0075] S32: Fuse the prediction results of two models using Dempster's rule, and then continue to fuse new model prediction results using the fused model prediction results until all model prediction results are fused.

[0076] S33: Based on the stress and strain at the critical point obtained by finite element method, the life range of the component under this working condition is evaluated.

[0077] In another exemplary embodiment of this application, the calculation formula for the lifetime prediction interval is:

[0078]

[0079]

[0080]

[0081] in, For the first The lifetime prediction range of each model. For the first The lower bound of the lifetime prediction range for each model. For the first The upper limit of the lifetime prediction range for each model. for Quantiles For 1- Quantiles For confidence quantile parameters, These are the input parameters for the physical information neural network model. Number the model. For cross-validation folds, The total number of folds for cross-validation. For the first The model has input parameters On The set of prediction results from cross-validation.

[0082] This application uses a given confidence quantile parameter (or confidence level). Obtain quantiles; calculate the upper limit of the lifespan prediction interval. and the lower limit of the life prediction range The lifespan prediction range is obtained. (or model prediction interval).

[0083] In another exemplary embodiment of the present application, the basic trust allocation is calculated according to the following formula:

[0084] .

[0085]

[0086]

[0087]

[0088] wherein, is the basic trust allocation function of the i-th model, i is the life prediction interval of the i-th model, is the credibility weight of the i-th model, is the model score of the i-th model, is the model number, is the dimensionless error indicator of the i-th model, is a parameter for controlling the discrimination degree of the model score, is the model error indicator of the i-th model, and are the minimum error indicator and the maximum error indicator in the n models, respectively. According to the dimensionless model error indicator , the model score is converted into the model score , the model score is normalized into the model credibility weight j , and the basic trust allocation of each model is constructed according to the model credibility weight .

[0089] .

[0090] In another exemplary embodiment of the present application, the above step 204 is replaced by the following steps 401-404:

[0091] Step 401: Calculate the conflict factor according to the basic trust allocation, and the calculation formula of the conflict factor is as follows:

[0092] .

[0093] wherein, is the conflict factor, ​​​​​​To identify a subset of the frame, i.e., the model a Lifespan prediction range , For another subset of the identification framework, namely the model b Lifetime prediction range , For the model a Basic trust allocation, For the model b The basic trust allocation.

[0094] Step 402: According to Dempster's rule, the basic trust assignments of each model are merged one by one until all models are merged, resulting in the merged basic trust assignments. The calculation formula for the merged basic trust assignments is as follows:

[0095] .

[0096]

[0097] in, For the allocation of basic trust after integration, For the first The basic trust allocation of each model =1, 2, ..., , To identify any subset within the frame, This is the Dempster fusion operator.

[0098] Step 403: Perform a Pignistic transformation on the fused basic trust allocation to obtain the Pignistic probability distribution of the predicted lifetime. The formula for calculating the Pignistic probability distribution of the predicted lifetime is as follows:

[0099] .

[0100] in, For the Pignistic probability, For the interval [ c 1 ,c 2], The lower bound of the interval, The upper bound of the interval is... For set M The basic trust allocation.

[0101] Step 404: Based on the preset confidence level, determine the fusion lifetime prediction interval from the Pignistic probability distribution.

[0102] This application allocates basic trust based on each model. Basic trust allocation for the two models and Obtain conflict factor According to Dempster's rule, the basic trust distribution of the two models is merged sequentially. The basic trust assignments of each model are obtained, resulting in a fused basic trust assignment. A pignistic transformation is then performed on the fused basic trust assignment to obtain the probability distribution for predicting lifetimes under a set of experimental conditions. Quantiles are then selected. (i.e., the preset confidence level), calculate the minimum width that satisfies this interval: .in, This represents the confidence probability based on the Pignistic probability. This represents the confidence interval for predicting lifespan.

[0103] In another exemplary embodiment of this application, step 205 is replaced by steps 501 to 505:

[0104] Step 501: Based on the fused lifetime prediction interval, obtain the conditional probability density function of lifetime.

[0105] Step 502: The stress and strain probability distributions are used as prior probability density functions.

[0106] Step 503: Calculate the marginal probability density function of the lifetime based on the conditional probability density function and the prior probability density function. The calculation formula is as follows:

[0107] .

[0108] in, For predicted equipment lifespan T Exceeding time t The marginal probability density function, Given input parameters under Piginistic probability x Predicted equipment lifespan T Exceeding time t The conditional probability density function, For input parameters The prior probability density function.

[0109] Step 504: Calculate the conditional reliability function based on the conditional probability density function. The calculation formula is as follows:

[0110] .

[0111] .

[0112] in, Given input parameters x Predicted devices int The conditional reliability function at time t. For probability operators, For time, Let lifespan be a random variable. Given output parameters x The lifespan random variable at that time.

[0113] Step 505: Based on the conditional reliability function and the prior probability density function, obtain the overall predicted reliability function for lifetime and the reliability interval evaluation curve. The calculation formula is:

[0114] .

[0115] in, Let be the overall predicted reliability function for lifetime at time t.

[0116] .

[0117] .

[0118] in, for t The lower bound reliability function of the lifetime at time t. Given input parameters x exist t The lower bound conditional reliability function at time t. for t The upper limit of the lifetime at any given time is determined by the reliability function. Given input parameters x exist t The upper limit conditional reliability function at time t.

[0119] This application refers to models that have already been fused. Given input parameters The predicted lifespan interval is a random variable and follows a distribution: (in, Given output parameters x The lifespan random variable at that time, Given input parameters under Piginistic probability x Predicted equipment lifespan T The conditional probability density function at that time); if the input parameters Then the marginal distribution of lifetimes is obtained. From this, the conditional survival function (or conditional reliability function) can be obtained. The predicted reliability of the lifetime is obtained after synthesis. Thus, the interval reliability is obtained. , Thus, the final interval reliability curve (or reliability interval evaluation curve) is obtained .

[0120] The application provides a high-temperature rotating machinery component reliability interval evaluation method considering model uncertainty. On the basis of a constructed physical information neural network, point estimation for life is expanded into interval estimation by using evidence theory, and further interval estimation of reliability is realized. The upper limit of the reliability interval reflects the most economical reliability evaluation, but there is potential risk. The lower limit of the reliability interval reflects the most conservative reliability evaluation, but there is excessive maintenance. The interval reliability evaluation provides more decision basis for operation and maintenance management of the high-temperature rotating machinery component, helps the factory to find the optimal balance between economy and safety according to the actual operation situation, and better arranges the equipment maintenance shutdown plan. The reliability interval evaluation method has the advantages of high calculation efficiency, accurate prediction, good engineering applicability and the like.

[0121] The high-temperature component reliability interval evaluation method of the application is described below by taking a steam turbine rotor as an example.

[0122] In this embodiment, 12Cr1MoV alloy is taken as the material of the rotor, and the working temperature is 600 DEG C. As shown in FIG. 1, the rotor is composed of a shaft, a disk and a connecting piece. Figure 2As shown, this invention provides a reliability interval assessment method for high-temperature rotating mechanical components considering model uncertainties, comprising the following steps: Probabilistic damage assessment of high-temperature rotating mechanical components: establishing a high-fidelity finite element model, identifying critical points of the high-temperature rotating mechanical components, accumulating multiple sets of finite element simulation results, training a surrogate model, and obtaining the stress-strain probability distribution of critical points of the component through Monte Carlo simulation; Establishment of a material-level physical information neural network model: training the neural network with experimental conditions as input variables and lifetime as the output variable, constructing multiple physical loss functions based on the linear damage accumulation rule, training multiple physical information neural network models, and constraining the training gradient with dynamic physical loss weights to ensure a stable training process; Reliability interval estimation based on evidence theory: calculating the error index of cross-validation of each physical information neural network, constructing a lifetime prediction interval and basic trust allocation based on this, fusing the prediction results of each model one by one using Dempster's rule to obtain a lifetime interval estimate, further incorporating the influence of random uncertainty, and completing the reliability interval assessment. Traditional reliability assessment methods for high-temperature rotating mechanical components often focus on the random uncertainties brought about by model parameters, while ignoring the cognitive uncertainty caused by different model selections, resulting in an incomplete reliability analysis of high-temperature rotating mechanical components. The reliability interval assessment method for high-temperature rotating mechanical components considering model uncertainties of this invention, in addition to considering the random uncertainties caused by loads and material parameters, also considers the model uncertainties caused by the selected physical model. This method can reflect the uncertainty and confidence interval of the predicted values, providing a more comprehensive basis and information for reliability assessment required in engineering. This reliability assessment method has fast calculation speed, good robustness, and good engineering applicability.

[0123] like Figure 3 As shown, a three-dimensional finite element model was established based on the structural characteristics of the turbine rotor, and the physical properties, elasticity, and plasticity of the 12Cr1MoV material as a function of temperature were set. Temperature field and centrifugal force field were applied to the turbine rotor to deal with the creep fatigue interaction during actual service.

[0124] like Figure 4 As shown, the most dangerous position of the turbine rotor and the stress and strain values ​​at that position are determined through finite element simulation.

[0125] Assume that the parameters in the finite element simulation follow a normal distribution with a coefficient of variation of 0.05. These parameters include: Young's modulus. Fatigue strength factor Fatigue ductility factor Ostergren model parameters Ostergren model parameters Stress relaxation equation parameters Stress relaxation equation parameters , maximum stress , temperature , 50 groups of parameter combinations were obtained by Latin hypercube sampling, and finite element simulation was carried out in turn.

[0126] 50 groups of parameter combinations of finite element simulation were taken as input variables, and stress and strain values of the dangerous point were taken as output variables to train the surrogate model. The model selected neural network, support vector machine regression, Gaussian process regression, response surface method and random forest regression. After comparison, the neural network model had the smallest error and the strongest generalization ability, and the neural network model was selected as the surrogate model.

[0127] The neural network structure was set to 32-16-8, and the input variables were Young's modulus , fatigue strength factor , fatigue ductility factor , Ostergren model parameters , Ostergren model parameters , stress relaxation equation parameters , stress relaxation equation parameters , maximum stress , temperature , and the output variables were the stress , strain , displacement ; Because the orders of magnitude of the input variables are too different, the data needs to be normalized before training; The data division ratio of the training set and the test set is 4:1; The activation function selects ReLu, the optimizer selects AdamW, the L2 regularization method is adopted, the training round is set to 500 rounds, and the model loss function selects MSE.

[0128] After the training of the surrogate model is completed, 10,000 groups of parameter combinations are obtained by Latin hypercube sampling, and the stress and strain probability distribution of the dangerous point is obtained by Monte Carlo simulation through the surrogate model. The stress obeys , unit: MPa, and the strain obeys , unit: %.

[0129] As Figure 5 shown, 127 groups of creep-fatigue test data of 12Cr1MoV material were collected in the embodiment, and the test temperature was 600℃. The data includes strain range, plastic strain range, stress range, average stress, strain rate, tensile holding time, compression holding time and life.

[0130] The physical information neural network is trained with the data set, the network structure is set to 16-8-4, the input variables are strain range, plastic strain range, stress range, strain rate, tensile holding time and compression holding time, and the output variable is service life; the data division ratio of the training set and the test set is 4:1; the activation function is ReLu, the optimizer is AdamW, the L2 regularization method is adopted, and the training round is set to 500 rounds.

[0131] The total loss function of the model is set as:

[0132] .

[0133] The total loss of the physical information neural network model is The data loss function is The physical loss function is The weight of the physical loss function is. The total loss function participates in the forward propagation and back propagation of the model.

[0134] The expression of the physical loss function is:

[0135] .

[0136] .

[0137] The physical loss is The number of samples is The creep fatigue life predicted by the neural network is The creep fatigue life predicted by the physical equation is

[0138] The optional fatigue life prediction model of the model is the Manson-Coffin model (MC, the same below), the Morrow model (MO, the same below) and the Ostergren model (OS, the same below); the optional creep life prediction model is the Time-Fractional model (TF, the same below) and the Ductile Exhaustion model (DE, the same below). The damage expressions of the above five damage models are as follows:

[0139] .

[0140] .

[0141] .

[0142] .

[0143] ​​ .

[0144] where, is the total strain amplitude, is the fatigue strength exponent, is the fatigue ductility exponent, is the fatigue strength coefficient (or fatigue strength factor), is the material elastic modulus (or Young’s modulus), is the fatigue ductility coefficient (or fatigue ductility factor), is the fatigue life or number of cycles to failure, is the single-cycle fatigue damage, is the total strain range, is the mean stress, is a material constant (or Ostergren model parameter), is the plastic strain energy density per cycle, is a material constant (or Ostergren model parameter), is the creep damage calculated by the time fraction method, is the relaxation rate parameter, is the hold time, is a material constant, is the peak stress, is the stress relaxation equation parameter, is the time, is a material constant, is the creep damage calculated by the ductility exhaustion method, is a material constant, is a material constant.

[0145] Any one of the above fatigue damage models (or fatigue life prediction models) and creep damage models (or creep life prediction models) are combined (there are 6 combinations in total), and the model parameters need to be fitted in advance before the training starts. The specific values are shown in Table 1.

[0146] Table 1: Physical loss equation parameters

[0147]

[0148] The creep-fatigue life can be calculated using the linear damage accumulation rule:

[0149] .

[0150] where, and represent the creep damage and fatigue damage calculated by the physical equation, respectively.

[0151] Model selection learning rate, dropout probability, regularization weight, physical loss weight, optimizer weight are hyperparameters, Bayesian optimization method is used for 200 rounds of hyperparameter search, and the optimization target is:

[0152] .

[0153] wherein, is the test set determination coefficient, is the training set determination coefficient, the target function takes value [-1, 1], the larger the value, the better the model generalization ability. In order to ensure the robustness of the model, the average value of the model score obtained by ten-fold cross-validation in the training process is taken as the model score.

[0154] The search range of hyperparameters is shown in Table 2:

[0155] Table 2: Hyperparameter search range

[0156]

[0157] During training, the dynamic physical loss weight method is used to control the speed of gradient descent. When the norm of the gradient exceeds 1, the physical loss weight is multiplied by a coefficient less than 1 to avoid training failure caused by too large gradient.

[0158] After each model training is completed, ten-fold cross-validation is performed with the optimal hyperparameters, and the model error indicators are shown in Table 3. Error indicators include: mean square error, root mean square error, mean absolute error, mean absolute percentage error, determination coefficient and two times error band accuracy.

[0159] Table 3: Error indicators of physical information neural network model

[0160]

[0161] Compared with ordinary neural network models that do not contain physical information, physical information neural networks that integrate physical equations perform better in terms of accuracy and overall fitting degree in creep-fatigue life prediction.

[0162] After ten-fold cross-validation, each test condition obtains 6 models of 10 predicted lives, a total of 60 data, based on which life prediction intervals and basic trust distribution are constructed.

[0163] For each model's 10 predicted life data, a confidence quantile parameter is given, and the quantile number is obtained:

[0164]

[0165]

[0166] where, is the lower bound of the life prediction interval of the th model, is the upper bound of the life prediction interval of the th model, is the th quantile, is the 1- th quantile, is the confidence quantile parameter, is the input parameter of the physical information neural network model, is the model number, is the cross-validation fold number, is the total cross-validation fold number, is the prediction result set of the th model on the input parameter in the th cross-validation fold.

[0167] The model prediction interval can be obtained as: .

[0168] The basic trust allocation method is as follows. First, the dimensionless model error indicator is:

[0169]

[0170] where, is the dimensionless error indicator of the th model, is the model error indicator of the th model, and the mean absolute error (MAE) is selected here, and are the minimum error indicator and the maximum error indicator in the j th model, respectively. It should be noted that the error selected here is the error of the model for a set of test conditions and life, rather than the overall error indicator of the model. Then, the model score is converted as:

[0171] .

[0172] where, is the parameter for controlling the discrimination degree of the model score, which is taken as 0.3 here. The normalization is the model credibility weight:

[0173]

[0174] where, is the credibility weight of the th model, is the credibility weight of the model score of the i-th model.

[0175] Basic belief assignment of the i-th model:

[0176] .

[0177] where, is the basic belief assignment function of the i-th model, i is the life prediction interval of the i-th model. Basic belief assignments of the two models and

[0178] have a conflict factor:

[0179] .

[0180] where, is the conflict factor, is a subset of the recognition framework, i.e., the life prediction interval of the model a , is another subset of the recognition framework, i.e., the life prediction interval of the model . b According to the Dempster rule, the basic belief assignments of the two models are fused:

[0181]

[0182]

[0183] where, is an arbitrary subset in the recognition framework, is the Dempster fusion operator.

[0184] The basic belief assignments of the six models are sequentially fused to obtain:

[0185] .

[0186] where, is the fused basic belief assignment.

[0187] The Pignistic transformation is performed on the basic belief assignment to obtain the probability distribution of the predicted life for a set of test conditions:

[0188] .

[0189] where, is the Pignistic probability, is the interval , ​​​​ is the lower bound of the interval, is the upper bound of the interval, is the set of basic beliefs assigned to the set

[0190] Select quantile , calculate the minimum width of the interval that satisfies the interval:

[0191] .

[0192] where, is the belief probability based on Pignistic probability, is the confidence interval of the predicted life.

[0193] It is calculated that the interval width that satisfies the condition of 95% interval coverage is 238.

[0194] For the model that has been fused , given the input stress and strain , the interval of the predicted life obeys the distribution:

[0195] .

[0196] where, is the life random variable given the input stress and strain , and is the conditional probability density function of life under Piginistic probability.

[0197] From the aforementioned Monte Carlo simulation, the stress obeys , unit: MPa, and the strain obeys , unit: %, then the marginal distribution of life is:

[0198] .

[0199] where, is the predicted device life T over time t , the marginal probability density function of life, is the conditional probability density function of the predicted device life x given the input parameters T under Piginistic probability, is the prior probability density function of the input parameters .

[0200] Thus, the conditional survival function (or conditional reliability function) is obtained: ​

[0201] .

[0202] wherein, is the predicted conditional reliability function of the device at time x is the probability operator, t is the time, is the life random variable. The integrated prediction reliability of life is:

[0203]

[0204] .

[0205] wherein, is the population predicted reliability function of life at time t

[0206] The interval reliability is:

[0207] .

[0208] .

[0209] wherein, is the lower bound reliability function of life at time t is the probability operator, x is the lower bound conditional reliability function of the device at time t is the upper bound reliability function of life at time t is the probability operator, x is the upper bound conditional reliability function of the device at time t The final interval reliability curve (or reliability interval assessment curve) is obtained as shown in

[0210] . Figure 6 Based on the same inventive concept, the embodiments of the present application further provide a high-temperature component reliability interval assessment device for implementing the high-temperature component reliability interval assessment method described above. The implementation scheme of the device for solving the problem is similar to the implementation scheme described in the above method, and therefore the specific limitations in one or more high-temperature component reliability interval assessment device embodiments provided below can refer to the limitations of the high-temperature component reliability interval assessment method described above, which will not be repeated here.

[0211] Based on the same inventive concept, the embodiments of the present application further provide a high-temperature component reliability interval assessment device for implementing the high-temperature component reliability interval assessment method described above. The implementation scheme of the device for solving the problem is similar to the implementation scheme described in the above method, and therefore the specific limitations in one or more high-temperature component reliability interval assessment device embodiments provided below can refer to the limitations of the high-temperature component reliability interval assessment method described above, which will not be repeated here.

[0212] ​​​​​​​In an exemplary embodiment, a high-temperature component reliability interval evaluation device is provided, comprising: a probabilistic damage evaluation module, a model training module, an interval construction and trust allocation module, a fusion module, and a reliability interval evaluation module.

[0213] The probabilistic damage evaluation module is configured to perform probabilistic damage evaluation on the high-temperature component to obtain stress and strain probability distributions of the high-temperature component at dangerous points.

[0214] The model training module is configured to construct and train a plurality of physical information neural network models with creep-fatigue test conditions of the high-temperature component as input and test life as output; the physical information neural network models include a fatigue life prediction model and a creep life prediction model.

[0215] The interval construction and trust allocation module is configured to construct life prediction intervals and basic trust allocations of each model according to life prediction results of each physical information neural network model.

[0216] The fusion module is configured to fuse the life prediction intervals and basic trust allocations of each model through the Dempster method to obtain a fused life prediction interval.

[0217] The reliability interval evaluation module is configured to obtain a reliability interval evaluation result of the high-temperature component life according to the fused life prediction interval and the stress and strain probability distributions.

[0218] In an exemplary embodiment, a computer device can be a server or a terminal, and its internal structure diagram can be as shown in Figure 7 The computer device includes a processor, a memory, an input / output interface (I / O), and a communication interface. The processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The database of the computer device is configured to store high-temperature component reliability interval evaluation data. The input / output interface of the computer device is configured to exchange information between the processor and external devices. The communication interface of the computer device is configured to communicate with external terminals through a network connection. The computer program is executed by the processor to implement a high-temperature component reliability interval evaluation method.

[0219] Those skilled in the art can understand that, Figure 7The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0220] In an exemplary embodiment, a computer device is also provided, including a memory and a processor, the memory storing a computer program, and the processor implementing the steps in the above method embodiments when executing the computer program.

[0221] In an exemplary embodiment, a computer readable storage medium is provided, storing a computer program, and the computer program implements the steps in the above method embodiments when executed by a processor.

[0222] In an exemplary embodiment, a computer program product is provided, including a computer program, and the computer program implements the steps in the above method embodiments when executed by a processor.

[0223] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties, and the collection, use and processing of related data need to comply with relevant regulations.

[0224] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when the computer program is executed, the processes of the above-mentioned embodiments of the methods can be included. Any reference to a memory, a database or other medium used in the embodiments provided in the present application can include at least one of a non-volatile and a volatile memory. The non-volatile memory can include a read-only memory (ROM), a magnetic tape, a floppy disk, a flash memory, an optical storage, a high-density embedded non-volatile memory, a resistive random access memory (ReRAM), a magnetoresistive random access memory (MRAM), a ferroelectric random access memory (FRAM), a phase change memory (PCM), a graphene memory, etc. The volatile memory can include a random access memory (RAM) or an external cache memory, etc. As an illustration but not limitation, the RAM can be in various forms, such as a static random access memory (SRAM) or a dynamic random access memory (DRAM), etc.

[0225] The database involved in the embodiments provided in the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a blockchain, etc., without being limited thereto. The processor involved in the embodiments provided in the present application can be a general processor, a central processor, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., without being limited thereto.

[0226] The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, all possible combinations of the technical features in the above embodiments are not described, however, as long as the combinations of the technical features do not exist contradictory, they should be considered as the scope of the present application.

[0227] The principles and implementation modes of the present application are described by applying specific examples in the present application. The above-mentioned embodiments are only used to help understand the method and its core idea of the present application; meanwhile, for those skilled in the art, according to the idea of the present application, the specific implementation mode and application range can be changed. In conclusion, the content of the present application should not be understood as a limitation.

Claims

1. A method for evaluating the reliability range of high-temperature components, characterized in that, include: Probabilistic damage assessment is performed on high-temperature components to obtain the stress and strain probability distribution at critical points of the high-temperature components; Using the creep fatigue test conditions of high-temperature components as input and the test life as output, multiple physical information neural network models are constructed and trained; the physical information neural network models include fatigue life prediction models and creep life prediction models. Based on the lifetime prediction results of each physical information neural network model, the lifetime prediction range and basic trust allocation of each model are constructed. Based on the lifetime prediction intervals and basic confidence assignments of each model, the Dempster method is used to fuse them to obtain the fused lifetime prediction interval. Based on the fusion life prediction range and the stress and strain probability distribution, the reliability range assessment results of the high-temperature component life are obtained.

2. The method for evaluating the reliability range of high-temperature components according to claim 1, characterized in that, The formula for calculating the lifetime prediction interval is: in, For the first The lifetime prediction range of each model. For the first The lower bound of the lifetime prediction range for each model. For the first The upper limit of the lifetime prediction range for each model. for Quantiles For 1- Quantiles For confidence quantile parameters, These are the input parameters for the physical information neural network model. Number the model. For cross-validation folds, The total number of folds for cross-validation. For the first The model has input parameters On The set of prediction results from cross-validation.

3. The method for evaluating the reliability range of high-temperature components according to claim 1, characterized in that, The formula for calculating the basic trust allocation is as follows: ； in, For the first i The basic trust assignment function of the model, For the first The lifetime prediction range of each model. For the first The credibility weights of each model For the first The model score of each model. Number the model. For the first The dimensionless error index of the model To control the parameters of the model score discrimination, For the first The model error index of each model. and They are respectively j The minimum and maximum error indices in each model.

4. The method for evaluating the reliability range of high-temperature components according to claim 1, characterized in that, Based on the lifetime prediction intervals and basic confidence assignments of each model, the Dempster method is used to fuse them to obtain the fused lifetime prediction interval, which specifically includes: The conflict factor is calculated based on the basic trust allocation, and the formula for calculating the conflict factor is as follows: ； in, As a conflict factor, To identify a subset of the frame, i.e., the model a Lifespan prediction range , For another subset of the identification framework, namely the model b Lifetime prediction range , For the model a Basic trust allocation, For the model b Basic trust allocation; According to Dempster's rule, the basic trust assignments of each model are merged one by one until all models are merged, resulting in the merged basic trust assignment. The formula for calculating the merged basic trust assignment is as follows: ； in, For the allocation of basic trust after integration, For the first The basic trust allocation of each model =1, 2, ..., , To identify any subset within the frame, For Dempster fusion operator; The fused basic trust allocation is subjected to a Pignistic transformation to obtain the Pignistic probability distribution of predicted lifetime. The formula for calculating the Pignistic probability distribution of predicted lifetime is as follows: ； in, For the Pignistic probability, For the interval [ , ], The lower bound of the interval, The upper bound of the interval is... For set M Basic trust allocation; Based on a preset confidence level, the fusion lifetime prediction interval is determined from the Pignistic probability distribution.

5. The method for evaluating the reliability range of high-temperature components according to claim 1, characterized in that, Based on the fusion life prediction range and the stress and strain probability distribution, the reliability range assessment results of the high-temperature component life are obtained, specifically including: Based on the fusion lifetime prediction interval, the conditional probability density function of lifetime is obtained; The probability distributions of stress and strain are used as prior probability density functions; Based on the conditional probability density function and the prior probability density function, the marginal probability density function of the lifetime is calculated using the following formula: ； in, For predicted equipment lifespan T Exceeding time t The marginal probability density function, Given input parameters under Piginistic probability x Predicted equipment lifespan T Exceeding time t The conditional probability density function, For input parameters The prior probability density function; The conditional reliability function is calculated based on the conditional probability density function, and the formula is as follows: ； ； in, Given input parameters x Predicted devices in t The conditional reliability function at time t. For probability operators, For time, Let lifespan be a random variable. For given output parameters x Lifetime random variable; Based on the conditional reliability function and the prior probability density function, the overall predicted reliability function for lifetime and the reliability interval assessment curve are obtained. The calculation formula is: ； in, for t The overall predictive reliability function for lifetime at time point; ； ； in, for t The lower bound reliability function of the lifetime at time t. Given input parameters x exist t The lower bound conditional reliability function at time t. for t The upper limit of the lifetime at any given time is determined by the reliability function. Given input parameters x exist t The upper limit conditional reliability function at time t.

6. The method for evaluating the reliability range of high-temperature components according to claim 1, characterized in that, Probabilistic damage assessment is performed on high-temperature components to obtain the stress and strain probability distribution at critical points of the components, specifically including: A three-dimensional finite element model was established based on the structural characteristics of the high-temperature component. Latin hypercube sampling was used to extract multiple sets of parameter combinations from the model parameters of the three-dimensional finite element model, and finite element simulations were performed to obtain the stress and strain response of the high-temperature component at the critical point under different parameter combinations. Using the parameter combination of finite element simulation as input and the stress and strain response at the critical point as output, a surrogate model is trained to obtain a well-trained surrogate model. Multiple combinations of random parameters were obtained through Latin hypercube sampling; Monte Carlo simulation, based on a trained surrogate model, is used to calculate the stress and strain responses of critical points under multiple random parameter combinations obtained by Latin hypercube sampling, and the stress and strain probability distributions of critical points are obtained.

7. A high-temperature component reliability range assessment device, characterized in that, include: The probabilistic damage assessment module is used to perform probabilistic damage assessment on high-temperature components and obtain the stress and strain probability distribution at the critical points of the high-temperature components. The model training module is used to construct and train multiple physical information neural network models with the creep fatigue test conditions of high-temperature components as input and the test life as output; the physical information neural network models include fatigue life prediction models and creep life prediction models. The interval construction and trust allocation module is used to construct the lifetime prediction interval and basic trust allocation for each model based on the lifetime prediction results of each physical information neural network model. The fusion module is used to fuse the lifetime prediction intervals and basic trust assignments of each model using the Dempster method to obtain the fused lifetime prediction interval. The reliability range assessment module is used to obtain the reliability range assessment results of the high-temperature component life based on the fusion life prediction range and the stress and strain probability distribution.

8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the high-temperature component reliability range assessment method according to any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the high-temperature component reliability range assessment method as described in any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the high-temperature component reliability range assessment method as described in any one of claims 1-6.

Images