CN114563268B - Method for predicting creep performance of high-temperature alloy based on soft constraint neural network model - Google Patents
Method for predicting creep performance of high-temperature alloy based on soft constraint neural network model Download PDFInfo
- Publication number
- CN114563268B CN114563268B CN202210161213.7A CN202210161213A CN114563268B CN 114563268 B CN114563268 B CN 114563268B CN 202210161213 A CN202210161213 A CN 202210161213A CN 114563268 B CN114563268 B CN 114563268B
- Authority
- CN
- China
- Prior art keywords
- creep
- neural network
- data
- term
- constraint
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000003062 neural network model Methods 0.000 title claims abstract description 41
- 238000000034 method Methods 0.000 title claims abstract description 32
- 229910045601 alloy Inorganic materials 0.000 title claims abstract description 15
- 239000000956 alloy Substances 0.000 title claims abstract description 15
- 230000007774 longterm Effects 0.000 claims abstract description 32
- 238000013528 artificial neural network Methods 0.000 claims abstract description 31
- 239000000463 material Substances 0.000 claims abstract description 27
- 238000013213 extrapolation Methods 0.000 claims abstract description 19
- 238000002474 experimental method Methods 0.000 claims abstract description 16
- 238000012360 testing method Methods 0.000 claims description 37
- 238000012549 training Methods 0.000 claims description 16
- 210000002569 neuron Anatomy 0.000 claims description 12
- 230000008014 freezing Effects 0.000 claims description 6
- 238000007710 freezing Methods 0.000 claims description 6
- 238000010586 diagram Methods 0.000 claims description 4
- 238000010200 validation analysis Methods 0.000 claims description 4
- 238000012795 verification Methods 0.000 claims description 2
- 238000012827 research and development Methods 0.000 abstract description 3
- 239000010935 stainless steel Substances 0.000 abstract 1
- 229910001220 stainless steel Inorganic materials 0.000 abstract 1
- 230000006870 function Effects 0.000 description 6
- 238000011156 evaluation Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 238000004519 manufacturing process Methods 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 229910000897 Babbitt (metal) Inorganic materials 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 2
- 229910000963 austenitic stainless steel Inorganic materials 0.000 description 2
- 238000010248 power generation Methods 0.000 description 2
- 229910000601 superalloy Inorganic materials 0.000 description 2
- 230000006978 adaptation Effects 0.000 description 1
- 239000012612 commercial material Substances 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000001627 detrimental effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000004069 differentiation Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 230000008018 melting Effects 0.000 description 1
- 238000002844 melting Methods 0.000 description 1
- 239000007769 metal material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000012031 short term test Methods 0.000 description 1
- 238000010998 test method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N3/00—Investigating strength properties of solid materials by application of mechanical stress
- G01N3/08—Investigating strength properties of solid materials by application of mechanical stress by applying steady tensile or compressive forces
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N3/00—Investigating strength properties of solid materials by application of mechanical stress
- G01N3/08—Investigating strength properties of solid materials by application of mechanical stress by applying steady tensile or compressive forces
- G01N3/18—Performing tests at high or low temperatures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N2203/00—Investigating strength properties of solid materials by application of mechanical stress
- G01N2203/0058—Kind of property studied
- G01N2203/0069—Fatigue, creep, strain-stress relations or elastic constants
- G01N2203/0071—Creep
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N2203/00—Investigating strength properties of solid materials by application of mechanical stress
- G01N2203/0058—Kind of property studied
- G01N2203/0069—Fatigue, creep, strain-stress relations or elastic constants
- G01N2203/0073—Fatigue
Abstract
The invention discloses a method for predicting creep performance of a high-temperature alloy based on a soft constraint neural network model, which comprises the following steps: constructing a Bayesian regularized neural network model by setting a neural network structure, input parameters and output parameters; adding constraint conditions required for the first derivative and the second derivative of the creep-strength creep life curve, and establishing a soft constraint Bayes regularized neural network model; fitting short-term creep experiment data by using a soft constraint Bayes regularized neural network model, and searching and obtaining a scheme meeting constraint condition requirements; extrapolating a model result by using the obtained scheme, and predicting the long-term creep property of the material; and comparing the model prediction result with experimental data, and verifying the accuracy of the model. The method can realize simple and efficient fitting and extrapolation, can be used for predicting the long-term creep properties of most commercial stainless steel, new materials in the research and development stage at present and other high-temperature alloys, and has stable and reliable results.
Description
Technical Field
The invention relates to the technical field of creep performance analysis of high-temperature alloys, in particular to a method for predicting creep performance of a high-temperature alloy based on a soft constraint Bayes regularized neural network model.
Background
Creep property is an important performance index of high-temperature pressure-bearing structural materials used in the fields of thermal power generation, nuclear power generation, petrochemical industry, aerospace and the like. The high-temperature metal material serving as an important component of thermal power and nuclear power generally needs to be in service for 30 years or more under high temperature and high pressure for a long time, and the creep performance is not only related to national economy, but also related to national safe production and efficient production. Unlike conventional tensile mechanical properties, creep generally refers to a mechanical behavior in which plastic deformation slowly occurs under high pressure conditions at high temperature (typically greater than 0.5T m,Tm is the melting point of the material). Creep failure may typically be greater than 10 years or longer at lower stresses. While creep failure also typically occurs at stresses below yield strength. In order to increase the efficiency of power plants, new materials are being developed worldwide to accommodate the higher operating efficiency of power plants, i.e. higher temperatures and stresses. For new materials, it is important to predict the creep life because experiments for up to 10 years are performed, which are not only costly, but also very detrimental to material development.
The prior applied models are Norton's formula, master formula of European creep Cooperation committee of time temperature parameter method (TTP), and the recently proposed Wilshire formula. These are all empirical models fitted by parameters. The fitting effect is good, however, the extrapolation result is limited. At present, theoretical models based on a microscopic mechanism of a material exist, but more physical parameters are needed to be relied on, so that the application of the material is affected to a certain extent.
Neural networks are currently being used widely in the materials field as an algorithm for machine learning. Neural networks can often ignore the complex intrinsic nature, build a model without parameters, and build better fit parameters between input and output. Many researchers have tried to simply apply neural networks to fit experimental data and achieve better local fitting effects, as well as many resulting in overfitting. And neural networks are mostly used for interpolation but not for extrapolation. Because extrapolation often produces unexpected results that violate physical reality.
Disclosure of Invention
In order to overcome the authority in the prior art, the invention provides a method for predicting the creep performance of the high-temperature alloy based on a soft constraint Bayesian regularized neural network model, which can realize simple and efficient fitting and extrapolation and predict the long-term creep performance of a material.
In order to achieve the above purpose, the present invention adopts the following technical scheme.
A method for predicting creep performance of a high-temperature alloy based on a soft constraint Bayesian regularized neural network model comprises the following steps.
S1, constructing a Bayesian regularized neural network model by setting parameters such as a neural network structure, input, output and the like;
S2, adding constraint conditions required for the first derivative and the second derivative of the creep-strength creep life curve, and establishing a soft constraint Bayes regularized neural network model;
s3, fitting short-term creep experiment data by using a soft constraint Bayes regularized neural network model, and searching and obtaining a scheme meeting constraint condition requirements;
s4, extrapolating a model prediction result by using the obtained scheme, and predicting the long-term creep property of the material;
s5, comparing the model prediction result with experimental data, and verifying the model accuracy.
In step S1, parameters such as the neural network structure, input and output are set, wherein the input parameters are test stress and test temperature, and the output parameters are creep rupture time. Wherein creep rupture time refers to the time that a material breaks under given test temperature and test stress conditions.
In step S1, the neural network structure is set to 2-N-1, where N is the number of hidden layer neurons, and its setting range is typically between 3-20.
In step S1, classification of data is performed using 70:15:15, dividing the data into training data, validation data and test data.
In step S1, the training method adopted is a bayesian regularization method.
In step S2, the abscissa of the creep-strength creep life curve is the creep rupture life or creep rupture time, and the ordinate is the creep rupture strength or creep rupture stress.
In step S2, the constraint conditions of the first derivative and the second derivative are as follows:
/>
Where m is the negative of the inverse of the first derivative of the creep-life curve of creep strength, t R is the creep-rupture time, σ is the creep-rupture strength.
In step S2, constraint conditions of a first derivative and a second derivative are added to the creep-strength creep life curve, wherein the obtaining of the first derivative and the second derivative of the creep-strength creep life curve includes the following steps:
S21, fitting short-term creep experiment data by using a Bayesian regularized neural network according to the description in the step S1, obtaining a network structure and freezing the network structure;
s22, under the condition of corresponding test temperature, establishing a group of continuous data of simulated stress, and obtaining a simulated creep stress creep life curve serving as a prediction curve by utilizing the frozen network structure in the step S21;
s23, fitting the prediction curve obtained in the step S22 by using a polynomial, and obtaining a fitting function;
s24, deriving the fitting function obtained in the step S23, and respectively obtaining a first derivative and a second derivative.
The constraint conditions are obtained by the above steps S21, S22, S23, S24.
In step S2, constraint conditions of the first derivative and the second derivative are added to the creep-strength creep life curve. The constraint condition is added to the training of the neural network by a method for constraining the training result of the Bayesian regularized neural network, namely, soft constraint. Soft constraints are here typically implemented as a program that alters the training process of the neural network, as opposed to hard constraints.
In step S3, the short-term creep experiment data is fitted by using the soft constraint bayesian regularized neural network model established in step S1 and step S2, including the following settings:
Setting the number of neurons of a hidden layer of the neural network within a certain interval range, and generally between 3 and 20;
setting random data flow in a certain range; random data stream here refers to a set of random data used to generate initial fitting coefficients;
Data were divided into three groups at 70:15:15 for training, validation and testing.
A solution is then sought within this range that can meet the constraints in step S2.
In the step S3, the soft constraint Bayesian regularized neural network model established in the step S1 and the step S2 is utilized to fit short-term creep experiment data, and a scheme meeting the constraint condition requirement in the step S2 is found and obtained. Mainly comprises the following steps: the constraint condition is obtained through steps S21, S22, S23, S24, and compared with the constraint condition requirement. The meeting of the requirements is a scheme of the model.
In the step S3, the soft constraint Bayesian regularized neural network model established in the step S1 and the step S2 is utilized to fit short-term creep experiment data, and a scheme meeting the constraint condition requirement in the step S2 is found and obtained. Mainly comprises the following steps: the constraint condition is obtained through steps S21, S22, S23, S24, and compared with the constraint condition requirement. If the requirements are not satisfied, searching for the next set of tests is automatically performed within the number of hidden layer neurons and the random data stream range set in step S3 until a scheme is found.
In step S4, the model structure is extrapolated by using the obtained scheme, and the long-term creep performance is predicted. Mainly comprises the following steps:
S41, fitting short-term creep experiment data by utilizing the scheme searched in the step S3 to obtain a network structure, and freezing the obtained network structure.
S42, under the condition of corresponding test temperature, establishing a group of continuous data of simulated stress, and obtaining a simulated creep stress creep life curve which is a prediction curve by utilizing the frozen network structure in the step S41;
s43, extrapolating the result of the prediction curve to a long-term creep life; long-term creep performance of the material is predicted.
In step S5, the model prediction result is compared with the experimental data, and the accuracy of the model is verified. The method comprises the following steps: according to the creep life and creep strength under long-term conditions obtained in step S4, comparison is made with experimental data under usable long-term conditions. And checking the accuracy of the model by using the correlation coefficient or the error distribution diagram. The correlation coefficient under extrapolation conditions should be greater than 75%. And when extrapolated to long-term conditions, the result still needs to meet the constraints.
The "short term" of the short term experimental data may be any experimental data of a short time, and 10000 hours or less is adopted in the embodiment of the present invention, but the scheme of the present invention is not limited to 10000 hours, but may be 1000 hours or more. Also, "long term" in the above long term conditions, long term creep life and the like is not limited to 10000 to 100000 hours used in the examples.
The invention has the advantages that:
by introducing constraint conditions of the first derivative and the second derivative of the creep strength creep life in a simple Bayesian regularized neural network, the problems of local overfitting caused by the neural network and extrapolation results which do not accord with physical reality significance are avoided. And compared with the traditional post-evaluation test method, the method greatly simplifies the evaluation method of creep life extrapolation.
Unlike hard-constrained neural networks, soft-constrained neural networks have lower demands on material domain researchers and material domain engineering specialists. The purpose can be achieved only by adding the prediction result of the soft constraint limiting model.
Therefore, the invention is greatly advanced to be used for the evaluation and prediction of the long-term creep property of the high-temperature pressure-bearing metal structural material in the fields of thermal power, nuclear power and the like, and has great significance for the current commercial materials and new materials in the laboratory research and development stage. And the invention can provide guiding reference and help for similar complex engineering problems in other fields.
It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention as claimed.
Drawings
The accompanying drawings, which are incorporated in and form a part of the specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention. In the drawings:
FIG. 1 is a flow chart of a method for predicting high temperature alloy creep performance based on a soft constraint Bayesian regularized neural network model in an embodiment of the present invention.
FIG. 2 is a graph of results of a Bayesian regularized neural network model fitting short term (10000 hours or less) creep test data in an embodiment of the present invention.
FIG. 3 is a graph of a prediction curve (less than or equal to 10000 hours) and an extrapolation result (10000-100000 hours) obtained under the set test temperature and test stress conditions by using a soft constraint Bayesian regularized neural network model in the embodiment of the invention.
FIG. 4 is a graph showing the relationship between m values, i.e., negative values of the inverse of the first derivative of the creep life curve, and creep life in the constraint condition of soft constraint of the Bayesian regularized neural network model in accordance with an embodiment of the present invention.
FIG. 5 is a graph showing the relationship between the second derivative of the creep life curve and the creep life in the embodiment of the present invention.
FIG. 6 is a graph comparing creep strength predicted and extrapolated to 100000 hours using a soft-constraint Bayesian regularized neural network model with experimental data in an embodiment of the present invention.
FIG. 7 is a graph of correlation coefficients between extrapolation results and experimental values of a soft constraint Bayesian regularized neural network model in an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will now be clearly and completely described with reference to the accompanying drawings in the embodiments of the present invention. It will be apparent that the described embodiments are only some, but not all, of the embodiments of the present invention. Other figures may be derived from these figures without inventive effort for a person of ordinary skill in the art. The embodiments may be implemented in many ways and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the example embodiments to those skilled in the art.
Furthermore, the described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to give a thorough understanding of embodiments of the invention. One skilled in the relevant art will recognize, however, that the inventive aspects may be practiced without one or more of the specific details, or with other methods, components, steps, etc. In other instances, well-known methods, implementations, or operations are not shown or described in detail to avoid obscuring aspects of the invention.
Examples
In the embodiment of the invention, super304H austenitic stainless steel is taken as an example to predict the long-term creep life. It should be stated that the creep test data of Super304H austenitic stainless steel is derived from Landolt-B and rnstein literature database. The experimental data can be derived from open source databases, published documents, actual production data of enterprises, experimental data of laboratories of universities, research institutions and the like. The source of experimental data is not protected by this patent.
The embodiment of the invention provides a method for predicting the creep performance of a high-temperature alloy based on a soft constraint Bayesian regularized neural network model, which is shown in a schematic flow chart of FIG. 1. From this flow chart, it is seen that the present invention is further described below with reference to the drawings and the specific execution of the steps of the embodiments, mainly comprising the steps S1-S5.
S1, constructing a Bayesian regularized neural network model by setting a neural network structure, input parameters and output parameters.
Specifically, the input parameters are test stress and test temperature, the creep rupture time is an output parameter, a Bayesian regularization method is adopted as a training method, and 70 is adopted for classifying data: 15:15, dividing the data into training data, verification data and test data, and setting the neural network structure to be 2-N-1, wherein N is the number of hidden layer neurons, and the setting range is 3-20. When the method is used for predicting the creep performance of the superalloy, the input parameters and the output parameters can be obtained through short-term tests or literature investigation.
In this embodiment, the neural network structure is set to 2-N-1, where the number N of hidden layer neurons is set to 6-16. The random data stream is set in the range 1015-1114. The input parameter test temperature is 600 ℃,650 ℃,700 ℃ and 750 ℃. The set of data for the established simulated stress is 10000 consecutive stress data from the minimum to the maximum of the stress test values, specifically the minimum stress value 39 MPa, the maximum stress value 357 MPa. In the step S1 of this embodiment, a bayesian regularized neural network model is constructed under the above conditions.
S2, constraint conditions required for the first derivative and the second derivative of the creep-strength creep life curve are added, and a soft constraint Bayes regularized neural network model is established.
First, constraint conditions are obtained by the following steps.
And S21, fitting short-term creep experiment data by using the Bayesian regularized neural network model constructed in the S1 to obtain a network structure and freezing the network structure. In this embodiment, the short-term creep test data refers to test data with a creep life of 10000 hours or less. The result of fitting the short-term 10000-hour creep experimental data is shown in fig. 2 by using the Bayesian regularized neural network and the data setting obtained in the step S1, wherein 'Exp' in the figure represents experimental data and 'NN' in the figure represents a Bayesian regularized neural network model prediction result.
S22, under the condition of corresponding test temperatures (600 ℃,650 ℃,700 ℃ and 750 ℃) a set of continuous simulated stress data is established, 10000 continuous stress data are established from a minimum stress value 39 MPa to a maximum stress value 357 MPa of a stress experiment value, the frozen network structure in the step S21 is utilized to obtain a simulated creep stress creep life curve, and a creep strength and creep life prediction curve, such as a part of the curve with an abscissa of less than or equal to 10 4 hours in FIG. 3, is drawn.
S23, fitting the prediction curve obtained in the step S22 by using a polynomial, and obtaining a fitting function; specifically, in this embodiment, the predictive curve obtained in step S22, that is, the portion of the curve shown in fig. 3 having the abscissa of 10 4 hours or less is fitted with the 8 th order polynomial, and the fitting function is obtained.
S24, deriving the fitting function obtained in the step S23, and respectively obtaining a first derivative and a second derivative. Specifically, in the present embodiment, the first derivative and the second derivative of the fitting function obtained in S23 are directly obtained by differentiation. And the portion of the curve whose derivative is 10 4 hours or less on the abscissa in fig. 4 and 5 is plotted. Wherein fig. 4 shows the value of m in the constraint, i.e. the negative value of the inverse of the first derivative, as shown in the following equation. FIG. 5 shows the second derivative of the creep-strength creep life curve.
Thus, constraints of the first derivative and the second derivative are obtained, and the following is required:
/>
Where m is the negative of the inverse of the first derivative of the creep-life curve of creep strength, t R is the creep-rupture time, σ is the creep-rupture strength.
These results are satisfactory as can be seen from the portion of the plot of fig. 4,5 having an abscissa of 10 4 hours or less.
And adding the constraint conditions into the training of the Bayesian regularization neural network through a method formula of constraining the training result of the Bayesian regularization neural network, namely, soft constraint, so as to obtain a soft constraint Bayesian regularization neural network model.
And S3, fitting short-term creep experiment data by using the soft constraint Bayes regularized neural network model finally established in the step S2, and searching and obtaining a scheme meeting constraint condition requirements.
Specifically, first, the short-term creep experimental data is fitted by using the soft constraint bayesian regularized neural network model established in S2, wherein the following settings are made: the number of neurons of a hidden layer of the neural network is set within a certain interval range, generally between 3 and 20, and the number of neurons of the hidden layer of the neural network is set between 6 and 16 in the embodiment; setting a random data stream within a certain range, wherein the range of the random data stream is 1015-1114 in the embodiment, the random data stream refers to a random data set for generating an initial fitting coefficient, input parameters are test stress and test temperature, and output parameters are creep rupture time; for all experimental data, the training data, validation data and test data were divided into three groups at a ratio of 70:15:15.
Then, a scheme that can satisfy the constraint condition in step S2 is sought within the above range. The method comprises the following steps: comparing with constraint condition requirements, and if the requirements are met, obtaining a model scheme; if the requirements are not met, searching for the next set of tests is automatically performed within the range of the number of hidden layer neurons and the random data stream set in the step S3 until a scheme is found. In this embodiment, a set of schemes is found, wherein the random data stream is 1020, the number of neurons in the hidden layer is 15, the polynomial of the fitted prediction curve is 8 th order polynomial, and the results are shown in fig. 2-5, wherein fig. 3-5 are the part of the curve with the abscissa of 10 4 hours or less.
S4, extrapolating a model result by using the obtained scheme, and predicting the long-term creep property of the material. Mainly comprises the following steps:
S41, fitting short-term creep experiment data by utilizing the scheme searched in the step S3 to obtain a network structure, and freezing the obtained network structure.
S42, under the condition of corresponding test temperature, establishing a set of continuous data of simulated stress, and obtaining a simulated creep stress creep life curve serving as a prediction curve by utilizing the frozen network structure in the step S41. In this embodiment, the prediction curve obtained by fitting the short-term creep test data of 10000 hours or less, that is, the portion of the curve having the abscissa of 10 4 hours or less in fig. 3, is obtained based on the result of fitting the short-term creep test data of 10000 hours or less.
S43, extrapolating the result of the prediction curve to a long-term creep life; long-term creep performance of the material is predicted. In this embodiment, the predicted curve obtained in S42 is extrapolated to 100000 hours, and the result of the extrapolated predicted curve is shown as the part of the curve having the abscissa of 10 4~105 hours in fig. 3. At the same time, the result of the extrapolation up to 100000 hours was examined, and the constraint of the extrapolated result was still met, as in the part of the curve with the abscissa of 10 4~105 hours in fig. 4 and 5.
S5, comparing the model prediction result with experimental data, and verifying the accuracy of the model.
The method comprises the following steps: according to the creep life and creep strength under long-term conditions obtained in step S4, comparison is made with experimental data under usable long-term conditions. And checking the accuracy of the model by using the correlation coefficient or the error distribution diagram. The correlation coefficient under extrapolation conditions should be greater than 75%. And when extrapolated to long-term conditions, the result still needs to meet the constraints.
Specifically, in this embodiment, the result predicted by using the soft constraint bayesian regularized neural network model is compared with experimental data, and the comparison result is shown in fig. 6: includes a comparison of 10000 hours or less predictions (the "NN Fit" curve at different temperatures in the figure) and creep strength extrapolated from 10000 hours to 100000 hours (the "NN Extrap" curve at different temperatures in the figure) with experimental data (the "Exp" scatter plot at different temperatures in the figure). . Fig. 7 shows correlation coefficients between extrapolation results and experimental values of the soft constraint bayesian regularized neural network model of the present embodiment, in which "Data" is a Data scatter distribution diagram with a target value on the abscissa and a predicted value on the crowded coordinate, a straight line marked with "Fit" is a fitting straight line between the target value and the predicted value, and a straight line marked with "y=t" is a straight line when the predicted value is equal to the target value, so that the correlation coefficients between the extrapolation results and the experimental values of the present embodiment are 87% and the satisfied correlation coefficients are greater than 75%. It can also be seen from the figure that the extrapolation results are stable and reliable.
The invention has obvious effect on the evaluation, analysis and prediction of the long-term service performance of the existing commercial high-temperature pressure-bearing metal structural material. Has important guiding significance for the research of new materials in the research and development stage at present. The long-term creep property of the material can be predicted by combining short-term experimental data, and a great amount of time and economic cost are saved. And greatly promotes the research of extrapolation technology of the superalloy material under long-term service conditions.
Other embodiments of the disclosure will be apparent to those skilled in the art from consideration of the specification and practice of the disclosure herein. This application is intended to cover any variations, uses, or adaptations of the disclosure following, in general, the principles of the disclosure and including such departures from the present disclosure as come within known or customary practice within the art to which the disclosure pertains. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the disclosure being indicated by the following claims.
It is to be understood that the disclosure is not limited to the precise construction and results set forth above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.
Claims (5)
1. A method for predicting creep performance of a high-temperature alloy based on a soft constraint Bayesian regularized neural network model is characterized by comprising the following steps:
S1, a Bayesian regularization neural network model is constructed by setting a neural network structure, input parameters and output parameters and adopting a Bayesian regularization training method;
S2, adding constraint conditions required for the first derivative and the second derivative of the creep-strength creep life curve, and establishing a soft constraint Bayes regularized neural network model;
Wherein, the creep strength creep life curve refers to the abscissa being the creep rupture life or creep rupture time, and the ordinate being the creep rupture strength or creep rupture stress;
The soft constraint refers to that the constraint condition is added into the training of the neural network through a method for constraining the training result of the Bayesian regularized neural network, namely, the soft constraint;
s3, fitting short-term creep experiment data by using the soft constraint Bayes regularized neural network model established in the S2, and searching and obtaining a scheme meeting constraint condition requirements;
s4, extrapolating a model prediction result by using the obtained scheme, and predicting the long-term creep property of the material;
s5, comparing the model prediction result with experimental data, and verifying the accuracy of the model;
In the step S1, input parameters are test stress and test temperature, and output parameters are creep rupture time; wherein creep rupture time refers to the time that a material breaks under given test temperature and test stress conditions;
In the step S1, the neural network structure is set to be 2-N-1, wherein N is the number of hidden layer neurons, and N is between 3 and 20; classification of data, using 70:15:15, dividing the data into training data, verification data and test data;
in step S2, the constraint conditions of the first derivative and the second derivative are as follows:
，/>，
where m is the negative of the inverse of the first derivative of the creep-life curve of creep strength, t R is the creep-rupture time, σ is the creep-rupture strength;
in step S2, the steps of obtaining the first derivative and the second derivative are as follows:
S21, fitting short-term creep experiment data by using the Bayesian regularized neural network constructed in the S1 to obtain a network structure and freezing the network structure;
s22, under the condition of corresponding test temperature, establishing a group of continuous data of simulated stress, and obtaining a simulated creep stress creep life curve serving as a prediction curve by utilizing the frozen network structure in the step S21;
s23, fitting the prediction curve obtained in the step S22 by using a polynomial, and obtaining a fitting function;
s24, deriving a fitting function, and respectively obtaining a first derivative and a second derivative.
2. The method for predicting the creep performance of a high-temperature alloy based on a soft constraint bayesian regularized neural network model according to claim 1, wherein in the step S3, when the soft constraint bayesian regularized neural network model established by using the step S2 is used for fitting short-term creep experimental data:
Setting the number of neurons of a hidden layer of the neural network within an interval range of 3-20;
setting random data flow in a certain range; random data stream here refers to a set of random data used to generate initial fitting coefficients;
The data were divided into three groups of training data, validation data and test data at a ratio of 70:15:15.
3. The method for predicting the creep performance of the high-temperature alloy based on the soft constraint Bayesian regularized neural network model as claimed in claim 1, wherein the scheme for searching and obtaining the solution meeting the constraint condition requirement in S3 refers to:
comparing the short-term creep experiment data fitted in the step S3 with the constraint condition requirements in the step S2, and if the requirements are met, obtaining a scheme of the model; if the requirements are not met, searching for the next group of tests in the searching range automatically until a scheme is found;
wherein the search range refers to the number of hidden layer neurons and the random data stream range set in step S3.
4. The method for predicting creep performance of a high-temperature alloy based on a soft constraint bayesian regularized neural network model according to claim 1, wherein in step S4, the model structure is extrapolated and the long-term creep performance is predicted by using the obtained scheme, and the method comprises the following steps:
S41, fitting short-term creep experiment data by utilizing the scheme searched in the step S3 to obtain a network structure, and freezing the obtained network structure;
s42, under the condition of corresponding test temperature, establishing a group of continuous data of simulated stress, and obtaining a simulated creep stress creep life curve, namely a prediction curve, by utilizing the frozen network structure in the step S41;
s43, extrapolating the result of the prediction curve to a long-term creep life; long-term creep performance of the material is predicted.
5. The method for predicting creep performance of a high-temperature alloy based on a soft constraint bayesian regularized neural network model according to claim 1, wherein in the step S5, the model prediction result is compared with experimental data, and the accuracy of the model is verified, which means that:
Comparing the long-term creep life and the long-term creep strength obtained in the step S4 with experimental data under long-term conditions, and checking the accuracy of the model by using a correlation coefficient or an error distribution diagram; wherein the correlation coefficient under the extrapolation condition is more than 75%, and the extrapolation result still meets the requirement of the constraint condition in the step S2 when the extrapolation is performed under the long-term condition.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210161213.7A CN114563268B (en) | 2022-02-22 | 2022-02-22 | Method for predicting creep performance of high-temperature alloy based on soft constraint neural network model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210161213.7A CN114563268B (en) | 2022-02-22 | 2022-02-22 | Method for predicting creep performance of high-temperature alloy based on soft constraint neural network model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114563268A CN114563268A (en) | 2022-05-31 |
| CN114563268B true CN114563268B (en) | 2024-04-26 |
Family
ID=81713615
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210161213.7A Active CN114563268B (en) | 2022-02-22 | 2022-02-22 | Method for predicting creep performance of high-temperature alloy based on soft constraint neural network model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114563268B (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106529017A (en) * | 2016-11-08 | 2017-03-22 | 华南理工大学 | High-temperature creep deformation prediction and creep damage analysis method for high-chrome steel component |
| CN111415010A (en) * | 2020-03-20 | 2020-07-14 | 广东电网有限责任公司阳江供电局 | Bayesian neural network-based wind turbine generator parameter identification method |
| CN113252465A (en) * | 2021-05-20 | 2021-08-13 | 天津理工大学 | M-H method-based heat-resistant steel creep life prediction method |
| CN114021481A (en) * | 2021-11-19 | 2022-02-08 | 华东理工大学 | Creep fatigue life prediction method based on converged physical neural network |
-
2022
- 2022-02-22 CN CN202210161213.7A patent/CN114563268B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106529017A (en) * | 2016-11-08 | 2017-03-22 | 华南理工大学 | High-temperature creep deformation prediction and creep damage analysis method for high-chrome steel component |
| CN111415010A (en) * | 2020-03-20 | 2020-07-14 | 广东电网有限责任公司阳江供电局 | Bayesian neural network-based wind turbine generator parameter identification method |
| CN113252465A (en) * | 2021-05-20 | 2021-08-13 | 天津理工大学 | M-H method-based heat-resistant steel creep life prediction method |
| CN114021481A (en) * | 2021-11-19 | 2022-02-08 | 华东理工大学 | Creep fatigue life prediction method based on converged physical neural network |
Non-Patent Citations (2)
| Title |
|---|
| 基于神经网络的镍基高温合金蠕变断裂寿命研究;于洋洋 等;计算机应用技术;20051231;第25卷(第2期);96-98 * |
| 用人工神经网络法预测镍基单晶高温合金的蠕变断裂寿命;李军伟 等;金属学报;20040331;第40卷(第3期);258-262 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114563268A (en) | 2022-05-31 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zhang et al. | A physics-informed neural network for creep-fatigue life prediction of components at elevated temperatures | |
| Duan et al. | A multivariate grey prediction model based on energy logistic equation and its application in energy prediction in China | |
| Tong et al. | Coal consumption forecasting using an optimized grey model: The case of the world's top three coal consumers | |
| CN114021481B (en) | Creep fatigue life prediction method based on fusion physical neural network | |
| CN112147432A (en) | BiLSTM module based on attention mechanism, transformer state diagnosis method and system | |
| Wang et al. | Reliability assessment for system-level turbine disc structure using LRPIM-based surrogate model considering multi-failure modes correlation | |
| Zhang et al. | Active extremum Kriging-based multi-level linkage reliability analysis and its application in aeroengine mechanism systems | |
| CN112069685A (en) | Interpretability-considered complex electromechanical system health assessment method and system | |
| CN111695288A (en) | Transformer fault diagnosis method based on Apriori-BP algorithm | |
| Wang et al. | Reinforced quantum-behaved particle swarm-optimized neural network for cross-sectional distortion prediction of novel variable-diameter-die-formed metal bent tubes | |
| CN114563268B (en) | Method for predicting creep performance of high-temperature alloy based on soft constraint neural network model | |
| Gu et al. | Creep-fatigue reliability assessment for high-temperature components fusing on-line monitoring data and physics-of-failure by engineering damage mechanics approach | |
| CN105808910A (en) | Irradiation deformation calculation method of zirconium-niobium pressure pipe in CANDU reactor | |
| Li et al. | Fatigue reliability analysis of aeroengine blade-disc systems using physics-informed ensemble learning | |
| Jia et al. | Microstructural evolution and constitutive models of 9CrMoCoB heat-resistant steel during high-temperature deformation | |
| Chu et al. | A minimum-cost modeling method for nonlinear industrial process based on multimodel migration and bayesian model averaging method | |
| Peng et al. | Dynamic recrystallization behavior under inhomogeneous thermomechanical deformation state | |
| Chen et al. | A novel fractional Hausdorff discrete Grey model for forecasting the renewable energy consumption | |
| CN116050663A (en) | Rock burst intensity level prediction method based on GD-DNN model | |
| CN115630582A (en) | Multi-sliding-window model fused soft rock tunnel surrounding rock deformation prediction method and equipment | |
| Zhou et al. | A framework for fatigue reliability analysis of high-pressure turbine blades | |
| Liu et al. | Transformer fault diagnosis model based on iterative nearest neighbor interpolation and ensemble learning | |
| Liu et al. | Very high‐cycle fatigue life prediction of high‐strength steel based on machine learning | |
| CN114707398A (en) | Method for predicting creep performance of metal structure material based on hard constraint neural network model | |
| CN114662370B (en) | Time-varying structure reliability analysis self-adaptive PCK method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| CB03 | Change of inventor or designer information | ||
| CB03 | Change of inventor or designer information |
Inventor after: He Junjing Inventor after: Zhou Yanjun Inventor after: Qin Haiying Inventor before: He Junjing Inventor before: Zhou Yanjun Inventor before: Long Qidong Inventor before: Qin Haiying |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |