A method for quantitatively predicting premature creep failure of high-temperature alloy based on error distribution

Abstract

The application discloses a method for quantitatively predicting premature creep failure of high-temperature alloy based on error distribution, comprising the following steps: predicting creep rupture performance to obtain predicted creep rupture time and stress; obtaining predicted time error and predicted stress error by using the predicted creep rupture performance and experimental data; fitting the predicted error by using logarithmic-logistic distribution and establishing improved logarithmic-logistic distribution; obtaining statistical data of the predicted error distribution, including the width of the distribution, the probability and position of outliers, etc., by using the improved logarithmic-logistic distribution to fit the predicted error; according to the effectiveness of the distribution fitting graph and the regression graph evaluation method, determining the best fitting method according to the width of the distribution, and predicting the probability of premature creep failure of the material according to the position of the outliers. The method is simple and efficient, can quantitatively predict premature creep failure of most austenitic steel, high-chromium steel and nickel-based alloy, and has the characteristics of stable and reliable results.

Description

Technical Field

[0001] This invention relates to the field of creep performance analysis technology for high-temperature alloys, and in particular to a method for quantitatively predicting premature creep failure of high-temperature alloys based on error distribution. Background Technology

[0002] Creep performance is a crucial design parameter for metallic structural materials used in thermal power generation, nuclear power, and rocket launches. To ensure the safe and reliable operation of fossil fuel power plants, it is necessary to periodically assess the degree of creep damage in critical components. This assessment helps prevent premature and unexpected failures and is known as residual creep life assessment. In the field of high-temperature engineering materials, premature creep failure refers to the possibility of a component or material failing due to creep at a rate far below its expected lifespan. Therefore, identifying outliers in creep damage is essential for assessing the risk of premature failure.

[0003] With the rise of machine learning algorithms, they have been widely applied in the field of materials structure and property research. Previous research has used machine learning algorithms to fit, analyze, and extrapolate creep fracture data of materials. Simultaneously, soft-constraint machine learning algorithms and constrained time-temperature parameter methods have been proposed. By adding constraints to machine learning algorithms and traditional time-temperature parameter methods, the long-term service life of materials can be more reasonably fitted and predicted.

[0004] Premature creep failure in materials is often overlooked; however, it is actually of great importance in many engineering fields. Premature creep failure can severely impact the safety and reliability of components or materials in energy sectors such as power plants. Most current research methods focus on predicting and designing for the average lifespan of materials. However, if the probability of premature creep failure could be obtained, the service life safety factor of materials could be designed more effectively, leading to more rational utilization of material properties. Summary of the Invention

[0005] To overcome the limitations of the prior art, this invention provides a method for quantitatively predicting premature creep failure of high-temperature alloys based on error distribution. This method can quantitatively predict premature creep failure of most austenitic steels, high-chromium steels, and nickel-based alloys. Furthermore, this method can be combined with the material's service conditions to assess the probability of premature creep failure and features stable and reliable results.

[0006] To achieve the above objectives, the present invention adopts the following technical solution.

[0007] A method for quantitatively predicting premature creep failure of high-temperature alloys based on error distribution includes the following steps.

[0008] S1 uses a soft-constraint machine learning algorithm and a constraint time-temperature parameter method to predict creep fracture performance, and obtains the predicted creep fracture time and creep fracture stress.

[0009] S2 uses the predicted creep fracture performance and experimental data to obtain the prediction time error and prediction stress error, etc.

[0010] S3 uses a log-log distribution to fit the prediction error and establishes an improved log-log distribution.

[0011] S4. The prediction error is fitted using an improved log-log distribution to obtain statistical data on the prediction error distribution, including the width of the distribution, the probability and location of outliers, etc.

[0012] S5. Evaluate the effectiveness of this method based on the distribution fitting plot and regression plot, determine the best fitting method based on the width of the distribution, and predict the probability of premature creep failure of the material based on the location of outliers.

[0013] In step S1, the soft-constrained machine learning algorithm is based on the patent "A method for predicting the creep properties of high-temperature alloys based on a soft-constrained neural network model" (application publication number: CN114563268A) and the paper "Application of softconstrained machine learning algorithms for creep rupture prediction of anaustenitic heat resistant steel Sanicro 25", Journal of Materials Research and Technology 22 (2023) 923-937.

[0014] In step S1, the constrained time-temperature parameter method is based on the paper “Error estimates in extrapolation of creep rupture data and its application to an austenitic stainless steel”, Materials at High Temperatures 39(2) (2022) 181-191.

[0015] In step S1, five soft-constraint machine learning algorithms are employed, including Soft-Constraint Bayesian Regularized Neural Network (SCBRNN), Soft-Constraint Levenberg-Marquardt Neural Network (SCLMNN), Soft-Constraint BFGNN (SCBFGNN), Soft-Constraint Finite-Memory BFGNN Regression Machine (SCLBFGSNNMR), and Soft-Constraint Support Vector Regression Machine (SCSVMR). Five time-temperature parameter methods (TTPs) are also used: Larson-Miller (LM) TTP, Orr-Sherby-Dorn (OSD) TTP, Manson-Succop (MS) TTP, Sud Aviation (SA) TTP, and Goldhoff-Sherby (GS) TTP. The parameter settings for these 10 models are detailed in published papers and will not be elaborated upon here.

[0016] In step S1, the present invention mainly uses the above 10 methods to predict the creep fracture properties of materials, and obtains the predicted creep fracture time and creep fracture stress, as well as the m value related to the first derivative of the predicted creep curve.

[0017] ... Formula 1

[0018] Where m is the negative of the reciprocal of the first derivative of the creep curve, and t R σ represents the creep rupture time, and σ represents the creep rupture stress. The creep curve refers to a curve where the horizontal axis represents the creep rupture time and the vertical axis represents the creep rupture stress.

[0019] On the other hand, the soft-constraint machine learning algorithm used in step S1 of the present invention is not limited to the fitting method in the aforementioned papers and patents. Other fitting methods in the prior art can also be used to obtain predicted creep fracture performance data, including predicted creep fracture time and creep fracture stress.

[0020] In step S2, the prediction time error is expressed as:

[0021] ... Formula 2

[0022] Where r p To predict the time error, t rpred and t rexp These are the predicted creep rupture time and the experimental creep rupture time, respectively.

[0023] The predicted stress error is expressed as:

[0024] ... Formula 3

[0025] or ... Formula 4

[0026] Where u s To predict stress error, σ rpred and σ rexp These are the predicted creep fracture stress and the experimental creep fracture stress, respectively; r p The prediction time error is represented by m, which is the negative of the reciprocal of the first derivative of the creep curve. The value of m is taken as the average value under different temperature conditions at 10,000 hours. The prediction stress error is expressed as the ratio of the predicted creep fracture stress to the experimental creep fracture stress (Formula 3), which can also be obtained by converting the prediction time error, i.e., Formula 4.

[0027] In step S3, the improved log-logarithmic distribution is obtained as follows:

[0028] S31, firstly, the log-logarithmic distribution is used to fit the prediction time error to obtain the width of the distribution (w). d0 ).

[0029] S32, for prediction time errors falling within 10 -1.5wd0 - 10 1.5wd0 Data outside the interval is assigned a weight of 3.

[0030] S33, using the log-log distribution to refit all prediction time error data, an improved log-log distribution is obtained.

[0031] In step S4, the improved log-logistic distribution obtained in step S3 is used to fit the prediction time error, and the corresponding statistical data of the error distribution are obtained, including the distribution width (w). d The probability and location of outliers, etc. Specific parameters are obtained as follows:

[0032] S41, the width (w) of the distribution can be directly obtained from the fitting results of the improved log-logarithmic distribution. d ).

[0033] S42, Prediction time error (r) p The probabilities of outliers were set to 0.5%, 2%, and 5%, and their corresponding locations could be directly obtained from the distribution fitting curve. The corresponding predicted stress error (u) s The outlier value can be calculated using formula 4 in step S2.

[0034] Step S5 mainly includes the following steps:

[0035] S51, using the statistical distribution data of the prediction time error obtained in step S4, plot a distribution fitting graph, plot the probability density distribution of the prediction time error, and plot the distribution graph of the prediction error fitted using the improved log-logarithmic distribution. Observe whether the fitted curve can characterize the distribution of the prediction error.

[0036] S52, using the predicted creep rupture time and experimental creep rupture time obtained in step S1, a regression plot is drawn, with the experimental creep rupture time on the horizontal axis and the predicted creep rupture time on the vertical axis. A line with a prediction error of 1 is also plotted within the plot as a reference. In the regression plot, the locations of the 5% outliers obtained in step S42 are used as half of the scatter plot.

[0037] S53. The effectiveness of the fitting method is qualitatively evaluated by observing whether the fitted curve in step S51 can characterize the predicted error distribution. The effectiveness of the fitting method is quantitatively evaluated by comparing whether the 5% outlier values ​​in step S52 are consistent with the experimental values. To effectively characterize the predicted error distribution, the error of the outlier values ​​should be controlled within ±1%.

[0038] S54, based on the width (w) of the improved log-logarithmic distribution obtained in step S41 d To determine the best fitting method for 10 models, where w d The smaller the value, the better the fit.

[0039] S55, based on the location of outliers in the prediction time error or prediction stress error obtained in step S42, predict the probability of premature creep failure of the material. The probability of premature creep failure should conform to the probability and location of the outliers. Its accuracy can be verified through step S53.

[0040] The advantages of this invention are:

[0041] A novel soft-constraint machine learning algorithm yields reliable and reasonable fitting results, which are then used for prediction error analysis. By proposing an improved log-logistic distribution and fitting the prediction error, the prediction error distribution can be reasonably characterized. This allows for the quantitative determination of the probability and location of outliers. Analysis of the regression plot quantitatively characterizes the accuracy of the proposed method. This provides a reliable method for assessing premature creep failure in materials.

[0042] This invention will greatly advance the assessment and prediction of residual creep life of high-temperature metallic structural materials used in thermal power, nuclear power, and other fields, and is of great significance to commercially available austenitic stainless steel, high-chromium steel, and nickel-based alloys currently under construction and in use. By combining this with the service conditions of materials in service, the residual life of materials can be quantitatively assessed, providing technical guidance for component maintenance.

[0043] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit the invention. Attached Figure Description

[0044] The accompanying drawings, which are incorporated in and form a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention. In the drawings:

[0045] Figure 1 This is a flowchart illustrating a method for quantitatively predicting premature creep failure of high-temperature alloys based on error distribution, as described in an embodiment of the present invention.

[0046] Figure 2 The figure shows the results of fitting the creep fracture performance of Super304H austenitic stainless steel using 10 models (five soft-constraint machine learning algorithms and five constraint time-temperature parameter methods) in an embodiment of the present invention.

[0047] Figure 3 This is a graph showing the prediction time error of Super304H fitted using an improved log-logarithmic distribution in an embodiment of the present invention. The vertical axis represents the time error probability density, and the horizontal axis represents the prediction time error; the predicted creep fracture time is obtained using the SCBRNN method.

[0048] Figure 4 This is a regression plot of the predicted creep rupture time and the experimental creep rupture time in an embodiment of the present invention. The vertical axis represents the predicted creep rupture time, the horizontal axis represents the experimental creep rupture time, the dashed line is the reference line with a prediction error of 1, and the solid line represents the position of the 5% outlier as half of the scatter plot (labeled "Position of 5% Outliners" in the figure). The predicted creep rupture time is obtained using the SCBRNN method. Detailed Implementation

[0049] The technical solutions of the embodiments of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. The embodiments can be implemented in various ways and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided to make the present invention more comprehensive and complete, and to fully convey exemplary embodiments to those skilled in the art.

[0050] Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. Numerous specific details are provided in the following description to give a full understanding of embodiments of the invention. However, those skilled in the art will recognize that the technical solutions of the invention can be practiced without one or more of the specific details, or other methods, components, steps, etc., can be employed. In other instances, well-known methods, implementations, or operations are not shown or described in detail to avoid obscuring various aspects of the invention.

[0051] Example

[0052] In this embodiment of the invention, the risk of premature creep failure was assessed using Super304H austenitic stainless steel as an example.

[0053] It should be noted that the creep test data for Super304H austenitic stainless steel comes from the Landolt-Börnstein literature database. Experimental data can also originate from open-source databases, published literature, actual production data from enterprises, or experimental data from laboratories in universities or research institutes. The source of the experimental data is not protected by this patent.

[0054] Embodiments of the present invention provide a method for quantitatively predicting premature creep failure of high-temperature alloys based on error distribution, such as... Figure 1 The diagram shown is a schematic flowchart of the method. As can be seen from the flowchart, it mainly includes steps S1-S5. The invention will be further described below with reference to the accompanying drawings and the specific execution of each step in the embodiments.

[0055] S1 uses a soft-constraint machine learning algorithm and a constraint time-temperature parameter method to predict creep fracture performance, and obtains the predicted creep fracture time and creep fracture stress.

[0056] Specifically, ten models were used to analyze and fit data on the creep fracture performance of Super304H, including creep fracture stress and creep fracture time at different temperatures. Five soft-constraint machine learning algorithms were employed: Soft-Constraint Bayesian Regularized Neural Network (SCBRNN), Soft-Constraint Levenberg-Marquardt Neural Network (SCLMNN), Soft-Constraint BFGNN (SCBFGNN), Soft-Constraint Finite-Memory BFGNN Regression Machine (SCLBFGSNNMR), and Soft-Constraint Support Vector Regression Machine (SCSVMR). Five time-temperature parameter (TTP) methods were used: Larson-Miller (LM) TTP, Orr-Sherby-Dorn (OSD) TTP, Manson-Succop (MS) TTP, Sud Aviation (SA) TTP, and Goldhoff-Sherby (GS) TTP.

[0057] In step S1, the soft-constrained machine learning algorithm is based on the patent "A Method for Predicting Creep Properties of High-Temperature Alloys Based on Soft-Constrained Neural Network Models" (Application Publication No.: CN114563268A) and the paper "Application of softconstrained machine learning algorithms for creep rupture prediction of anaustenitic heat resistant steel Sanicro 25", Journal of Materials Research and Technology 22 (2023) 923-937. The constraint time-temperature parameter method is based on the paper "Error estimates in extrapolation of creep rupture data and its application to an austenitic stainless steel", Materials at High Temperatures 39(2) (2022) 181-191. The parameter settings for these 10 models are described in detail in the published papers and will not be repeated here.

[0058] Specifically, in step S1, the above 10 methods are mainly used to predict the creep rupture properties of the material and obtain the predicted creep rupture time t. rpred and creep fracture stress σ pred And the value of m related to the first derivative of the predicted creep curve.

[0059]

[0060] Where m is the negative of the reciprocal of the first derivative of the creep curve, and t R σ represents the creep rupture time, and σ represents the creep rupture stress. The creep curve refers to a curve where the horizontal axis represents the creep rupture time and the vertical axis represents the creep rupture stress.

[0061] In this embodiment, the input parameters for the test temperatures were 600 ºC, 650 ºC, 700 ºC, and 750 ºC. The results are as follows: Figure 2 As shown, the horizontal axis represents creep fracture time, and the vertical axis represents creep fracture stress. "Exp" represents experimental data, SCMLAs represents soft-constraint machine learning algorithms, and TTP represents the constraint time-temperature parameter method. The legend lists five soft-constraint machine learning algorithms and five constraint time-temperature parameter methods.

[0062] S2 uses the predicted creep fracture performance and experimental data to obtain the prediction time error and prediction stress error, etc.

[0063] Specifically, the prediction time error is expressed as:

[0064]

[0065] Where r p To predict the time error, t rpred and t rexp These are the predicted creep rupture time and the experimental creep rupture time, respectively.

[0066] The predicted stress error is expressed as:

[0067]

[0068] or

[0069] Where u s To predict stress error, σ rpred and σ rexp These are the predicted creep fracture stress and the experimental creep fracture stress, respectively; r p The prediction time error is represented by m, which is the negative of the reciprocal of the first derivative of the creep curve. The value of m is taken as the average value under different temperature conditions at 10,000 hours. The prediction stress error is expressed as the ratio of the predicted creep fracture stress to the experimental creep fracture stress, which can also be obtained by converting the prediction time error, as shown in the formula above.

[0070] In this embodiment, t rpred This refers to the predicted creep rupture time obtained in step S1, i.e., the lines in different formats shown in Figure 1, t. rexp The creep fracture time is the experimental data for Super304H. m is a coefficient related to the first derivative of the creep curve obtained in step S1. Specifically, in this embodiment, for Super304H, it is taken as the average value at 600 ºC, 650 ºC, 700 ºC, and 750 ºC over 10,000 hours.

[0071] S3, fit the prediction error using a log-log distribution and establish an improved log-log distribution. This includes the following steps:

[0072] S31, firstly, the log-logarithmic distribution is used to fit the prediction time error to obtain the width of the distribution (w). d0 ).

[0073] Specifically, the width of the distribution (w) is obtained by fitting the prediction time error data obtained in step S2 using a log-logistic distribution. d0 ).

[0074] S32, for prediction time errors falling within 10 -1.5wd0 - 10 1.5wd0 Data outside the interval is assigned a weight of 3.

[0075] Specifically, based on the width (w) of the distribution in step S31 d0 Find the interval 10 -1.5wd0 - 10 1.5wd0 The range. Then for the prediction time error data falling within 10... -1.5wd0 - 10 1.5wd0 Data outside the specified interval is weighted by 3. This yields new prediction error data.

[0076] S33, using the log-log distribution to refit all prediction time error data, an improved log-log distribution is obtained.

[0077] Specifically, the new prediction time error data obtained in step S32 is fitted using a log-logarithmic distribution to obtain an improved log-logarithmic distribution.

[0078] It should be noted that steps S31-S33 need to be performed for all 10 different models.

[0079] S4 uses an improved log-log distribution to fit the prediction error, obtaining statistical data on the prediction error distribution, including the width of the distribution, the probability and location of outliers, etc. The specific parameters are obtained as follows:

[0080] S41, the width (w) of the distribution can be directly obtained from the fitting results of the improved log-logarithmic distribution. d ).

[0081] Specifically, the 10 models were analyzed, and the width of the distribution for each model was obtained. This result can be obtained directly after fitting the prediction error data.

[0082] S42, Prediction time error (r) p The probabilities of outliers were set to 0.5%, 2%, and 5%, and their corresponding locations could be directly obtained from the distribution fitting curve. The corresponding predicted stress error (u) s The outlier value can be calculated using the formula in step S2.

[0083] Specifically, the outlier probabilities of the prediction error of Super304H are set to 0.5%, 2%, and 5%, and their corresponding locations can be directly obtained through curve fitting. The corresponding predicted stress error (u) s The outliers can be calculated using the formula in step S2. Specifically, the location of the outlier with a predicted stress error of 0.5% for Super304H is calculated using the formula in step S2.

[0084] It is important to note that steps S41-S42 are required for the analysis of all 10 different models. Specifically, Table 1 below shows the statistical data obtained after fitting the data using an improved log-logistic distribution for the 10 different methods: five soft-constraint machine learning algorithms and five constrained time-temperature parameter methods. Where w... d r is the width of the distribution. p 5% outliers, r p 2% outliers, r p 0.5% outliers represent the probabilities and locations of outliers with prediction time errors of 5%, 2%, and 0.5%, respectively; s 0.5% outliers represent the probability and location of outliers in the predicted stress error of 0.5%.

[0085] Table 1. Statistical data obtained after analyzing the prediction time error distribution of Super304H austenitic stainless steel in this embodiment.

[0086]

[0087] S5. Evaluate the effectiveness of the method based on the distribution fitting plot and regression plot, determine the best fitting method based on the width of the distribution, and predict the probability of premature creep failure of the material based on the location of outliers. This mainly includes the following steps:

[0088] S51, using the statistical distribution data of the prediction time error obtained in step S4, plot a distribution fitting graph, plot the probability density distribution of the prediction time error, and plot the distribution of the prediction error fitted using the improved log-logarithmic distribution. Observe whether the fitted curve can characterize the distribution of the prediction error.

[0089] Specifically, such as Figure 3 The figure shows the prediction time error of Super304H fitted using an improved log-logarithmic distribution. The vertical axis represents the probability density of the time error, and the horizontal axis represents the prediction time error. The predicted creep fracture time was obtained using the SCBRNN method. The solid line curve is the fitted curve (labeled "log-logarithmic distribution fitting" in the figure), which can effectively characterize the probability density distribution of the prediction error.

[0090] S52, using the predicted creep rupture time and experimental creep rupture time obtained in step S1, a regression plot is drawn, with the experimental creep rupture time on the horizontal axis and the predicted creep rupture time on the vertical axis. A line with a prediction error of 1 is also plotted within the plot as a reference. In the regression plot, the locations of the 5% outliers obtained in step S42 are used as half of the scatter plot.

[0091] Specific examples Figure 4The figure shows a regression plot of predicted creep rupture time and experimental creep rupture time for Super304H austenitic stainless steel. The vertical axis represents the predicted creep rupture time, and the horizontal axis represents the experimental creep rupture time. The dashed line is the reference line of 1, and the solid line represents the position of the 5% outlier as half of the scatter plot (labeled "Position of 5% Outliners" in the figure). The predicted creep rupture time was obtained using the SCBRNN method.

[0092] S53. The effectiveness of the fitting method is qualitatively evaluated by observing whether the fitted curve in step S51 can characterize the predicted error distribution. The effectiveness of the fitting method is quantitatively evaluated by comparing whether the 5% outlier values ​​in step S52 are consistent with the experimental values. To effectively characterize the predicted error distribution, the error of the outlier values ​​should be controlled within ±1%.

[0093] Specifically, through comparison Figure 3 The effectiveness of a fitting method for predicting the distribution of time error can be qualitatively evaluated. Figure 4 The consistency between the fitting method and the experimental data can be quantitatively evaluated. In this embodiment, there are 234 data points for Super304H austenitic steel, with 12 data points in the 5% outlier category. The figure shows 12 data points falling outside the 5% outlier category, therefore, the error is 0, indicating that the method can effectively characterize the probability and location of outliers.

[0094] S54, based on the width (w) of the improved log-logarithmic distribution obtained in step S41 d To determine the best fitting method for 10 models, where w d The smaller the value, the better the fit.

[0095] Specifically, as shown in Table 1 above, there are 10 methods for analyzing the creep performance data of Super304H, categorized by their distribution width (w). d The data is sorted from smallest to largest. Therefore, SCBRNN is the best fitting method.

[0096] S55, based on the location of outliers in the prediction time error or prediction stress error obtained in step S42, predict the probability of premature creep failure of the material. The probability of premature creep failure should conform to the probability and location of the outliers. Its accuracy can be verified through step S53.

[0097] As shown in the data in Table 1 above, r p 5% outliers, r p 2% outliers, r p The 0.5% outliers represent the probabilities and locations of outliers with prediction time errors of 5%, 2%, and 0.5%, respectively. s0.5% outliers represent the probability and location of outliers in the predicted stress error of 0.5%.

[0098] Specifically, in step S54, SCBRNN has been determined as the best-fit method. For the SCBRNN method, by predicting the location of outliers with a stress error of 0.5% (2.7), it can be determined that there is a 0.5% probability that the material's creep fracture stress will be approximately 37% of the average predicted stress value, that is, a creep fracture stress lower than 63% of the average predicted stress value. On the other hand, by predicting the location of outliers with a time error of 5% (2.0), it can be determined that there is a 5% probability that the material's creep fracture time will be 50% of the predicted creep fracture time. Through the above examples and analysis, the risk of premature creep failure of materials can be quantitatively predicted.

[0099] Based on the above analysis and conclusions, it can be seen that this method can be combined with the service conditions of materials currently in service, thereby quantitatively determining the probability of premature creep failure.

[0100] It should be noted that the example image Figure 3 and Figure 4 The table shows the best fitting method for SCBRNN with Super304H. Other methods will also produce similar fitting distribution and regression plots. Table 1 shows that the error in the distribution width for different methods ranges from 0.24 to 0.3.

[0101] It is important to note that while SCBRNN is the best-fit method for Super304H, it may not be effective for other materials. The inventors have successfully applied this method to various materials, including austenitic stainless steels such as Super304H, Sanicro 25, T321H, TP316H, and T347H; high-chromium steels such as TP91, TP92, and SUH616B; and nickel-based alloys such as 13Cr4.5MoTi6Al. Furthermore, it has been found that the best-fit method is not fixed for different materials and data, nor is there a fixed ranking. Therefore, comprehensive comparison and analysis are needed to determine the best-fit method.

[0102] This invention demonstrates significant effectiveness in assessing and predicting the residual life of existing commercially available metallic structural materials. It can quantitatively predict the risk of premature creep failure under long-term high-temperature service conditions and has been successfully applied to various materials. It provides important guidance for assessing the residual life of currently in-service materials and has significant application value for the construction and maintenance of power plants.

[0103] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the following claims.

[0104] It should be understood that this disclosure is not limited to the precise structures and results described above and shown in the accompanying drawings, and various modifications and alterations may be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A method for quantitatively predicting premature creep failure of a high temperature alloy based on error distribution, characterized in that, The method comprises the following steps: S1, obtaining predicted creep rupture performance data including at least predicted creep rupture time and predicted creep rupture stress by using a soft constraint machine learning algorithm and a fitting method of constraint time-temperature parameter method to obtain predicted creep rupture performance; wherein, according to the predicted creep rupture performance data, an m value related to a first derivative of a predicted creep curve is obtained, wherein m is a negative value of the inverse of the first derivative of the creep curve, t R is the creep rupture time, σ is the creep rupture stress; the creep curve is a curve with the creep rupture time as the horizontal coordinate and the creep rupture stress as the vertical coordinate; S2, obtaining a prediction time error and a prediction stress error using the predicted creep rupture property and the experimental data; the prediction time error is expressed as: where r p is the prediction time error, t rpred and t rexp are the predicted creep rupture time and the experimental creep rupture time, respectively; The predicted stress error is expressed as: , where u s To predict stress error, σ rpred and σ rexp These are the predicted creep fracture stress and the experimental creep fracture stress, respectively; m is the value of m obtained in step S1, and is taken as the average value under different temperature conditions at 10,000 hours. S3, fitting the prediction error by using a log-logistic distribution to establish an improved log-logistic distribution; S31, first use log-logistic distribution fitting prediction time error, get the width of the distribution (w d0 ); S32, if the prediction time error falls within 10 -1.5wd0 - 10 1.5wd0 Data outside the interval is assigned a weight of 3; S33, re-fitting all the prediction time error data by using the log-logistic distribution to obtain the improved log-logistic distribution; S4, fitting the prediction error by using the improved log-logistic distribution to obtain statistical data of the prediction error distribution, at least including a width of the distribution, a probability and a position of an outlier; and specific parameter obtaining steps are as follows: S41, the width (w) of the distribution can be directly obtained from the fitting results of the improved log-logistic distribution d ); S42, the probability of outliers of the prediction time error (r p ) is set to 0.5%, 2%, 5%, and the corresponding position can be directly obtained from the distribution fitting curve; the outliers of the corresponding prediction stress error (u s ) are calculated by the formula in step S2; S5, evaluating effectiveness of the distribution fitting graph and the regression graph, determining the best fitting method according to the width of the distribution, and predicting a probability of premature creep failure of the material according to the position of the outlier; and the method mainly comprises the following steps: S51, drawing a distribution fitting graph by using the statistical distribution data of the prediction time error obtained in step S4, drawing a probability density distribution of the prediction time error, and drawing a distribution graph of the prediction error fitted by using the improved log-logistic distribution; and observing whether the fitting curve can represent the distribution of the prediction error; S52, drawing a regression graph by using the predicted creep rupture time and the experimental creep rupture time obtained in step S1, wherein an abscissa is the experimental creep rupture time, an ordinate is the predicted creep rupture time, and a line with a prediction error of 1 is drawn as a reference in the regression graph; and in the regression graph, the position of the 5% outlier obtained in step S42 is used as a half of a scatter band; S53, qualitatively evaluating effectiveness of the fitting method by observing whether the fitting curve in the fitting graph in step S51 can represent the distribution of the prediction error; and quantitatively evaluating effectiveness of the fitting method by comparing whether the 5% outlier in step S52 is consistent with the experimental value; and in order to effectively represent the distribution of the prediction error, an error of the outlier is controlled within ±1%; S54, determine the best fitting method in step 1 according to the width (w d ) of the improved log-logistic distribution obtained in step S41, wherein w d is smaller represents the better fitting effect; S55, predicting a probability of premature creep failure of the material according to the position of the outlier of the prediction time error or the prediction stress error obtained in step S42; the probability of premature creep failure of the material should be consistent with the probability and the position of the outlier; and accuracy is verified by step S53.

2. The method of claim 1, wherein the method is characterized by, In step S1, the soft constraint machine learning algorithm includes at least one of a soft constraint Bayesian regularization neural network, a soft constraint Levenberg-Marquardt neural network, a soft constraint BFGNN, a soft constraint limited memory BFGNN regression machine, and a soft constraint support vector regression machine; and the constraint time-temperature parameter method includes at least one of a Larson-Miller parameter method, an Orr-Sherby-Dorn parameter method, a Manson-Succop parameter method, a Sud Aviation parameter method, and a Goldhoff-Sherby parameter method.

3. The method of claim 1, wherein the method is characterized by: The high-temperature alloy at least includes austenitic steel, high-chromium steel, and nickel-based alloy.

Images