CN112214933A - Fatigue performance prediction method based on machine learning - Google Patents
Fatigue performance prediction method based on machine learning Download PDFInfo
- Publication number
- CN112214933A CN112214933A CN202011123975.5A CN202011123975A CN112214933A CN 112214933 A CN112214933 A CN 112214933A CN 202011123975 A CN202011123975 A CN 202011123975A CN 112214933 A CN112214933 A CN 112214933A
- Authority
- CN
- China
- Prior art keywords
- model
- training
- sample data
- data
- fatigue performance
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 51
- 238000010801 machine learning Methods 0.000 title claims abstract description 27
- 238000012549 training Methods 0.000 claims abstract description 45
- 239000000463 material Substances 0.000 claims abstract description 44
- 238000012360 testing method Methods 0.000 claims abstract description 17
- 238000012795 verification Methods 0.000 claims abstract description 10
- 238000010200 validation analysis Methods 0.000 claims abstract description 6
- 238000010206 sensitivity analysis Methods 0.000 claims abstract description 4
- 230000008569 process Effects 0.000 claims description 12
- 238000012545 processing Methods 0.000 claims description 9
- 238000004458 analytical method Methods 0.000 claims description 8
- 238000000513 principal component analysis Methods 0.000 claims description 8
- 230000006870 function Effects 0.000 claims description 7
- 210000002569 neuron Anatomy 0.000 claims description 6
- 238000013507 mapping Methods 0.000 claims description 5
- 239000011159 matrix material Substances 0.000 claims description 5
- 239000000203 mixture Substances 0.000 claims description 5
- 230000009467 reduction Effects 0.000 claims description 5
- 238000010438 heat treatment Methods 0.000 claims description 4
- 238000010606 normalization Methods 0.000 claims description 4
- 238000005070 sampling Methods 0.000 claims description 4
- 238000012706 support-vector machine Methods 0.000 claims description 4
- 229910052799 carbon Inorganic materials 0.000 claims description 3
- 230000008859 change Effects 0.000 claims description 3
- 229910052804 chromium Inorganic materials 0.000 claims description 3
- 229910052802 copper Inorganic materials 0.000 claims description 3
- 238000012417 linear regression Methods 0.000 claims description 3
- 229910052748 manganese Inorganic materials 0.000 claims description 3
- 229910052759 nickel Inorganic materials 0.000 claims description 3
- 210000004205 output neuron Anatomy 0.000 claims description 3
- 229910052698 phosphorus Inorganic materials 0.000 claims description 3
- 230000000704 physical effect Effects 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 3
- 229910052710 silicon Inorganic materials 0.000 claims description 3
- 229910052717 sulfur Inorganic materials 0.000 claims description 3
- 238000003062 neural network model Methods 0.000 claims description 2
- 238000012847 principal component analysis method Methods 0.000 claims description 2
- 238000009826 distribution Methods 0.000 description 6
- 238000002360 preparation method Methods 0.000 description 5
- 238000013461 design Methods 0.000 description 4
- 229910000831 Steel Inorganic materials 0.000 description 3
- 230000007547 defect Effects 0.000 description 3
- 238000011161 development Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000009661 fatigue test Methods 0.000 description 3
- 239000007769 metal material Substances 0.000 description 3
- 238000005457 optimization Methods 0.000 description 3
- 238000010791 quenching Methods 0.000 description 3
- 230000000171 quenching effect Effects 0.000 description 3
- 239000010959 steel Substances 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 238000007405 data analysis Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000008520 organization Effects 0.000 description 2
- 238000012827 research and development Methods 0.000 description 2
- 238000009864 tensile test Methods 0.000 description 2
- 238000012897 Levenberg–Marquardt algorithm Methods 0.000 description 1
- 238000012356 Product development Methods 0.000 description 1
- 230000001133 acceleration Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000005255 carburizing Methods 0.000 description 1
- 238000004140 cleaning Methods 0.000 description 1
- 230000001186 cumulative effect Effects 0.000 description 1
- 238000013144 data compression Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000001125 extrusion Methods 0.000 description 1
- 238000005242 forging Methods 0.000 description 1
- 230000001939 inductive effect Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000011089 mechanical engineering Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000005096 rolling process Methods 0.000 description 1
- 238000005204 segregation Methods 0.000 description 1
- 238000012163 sequencing technique Methods 0.000 description 1
- 238000004904 shortening Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 238000000638 solvent extraction Methods 0.000 description 1
- 238000005496 tempering Methods 0.000 description 1
- 238000004154 testing of material Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/10—Machine learning using kernel methods, e.g. support vector machines [SVM]
-
- 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
- G06N3/045—Combinations of networks
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16C—COMPUTATIONAL CHEMISTRY; CHEMOINFORMATICS; COMPUTATIONAL MATERIALS SCIENCE
- G16C60/00—Computational materials science, i.e. ICT specially adapted for investigating the physical or chemical properties of materials or phenomena associated with their design, synthesis, processing, characterisation or utilisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2111/00—Details relating to CAD techniques
- G06F2111/06—Multi-objective optimisation, e.g. Pareto optimisation using simulated annealing [SA], ant colony algorithms or genetic algorithms [GA]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/04—Ageing analysis or optimisation against ageing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/18—Manufacturability analysis or optimisation for manufacturability
Abstract
The invention discloses a fatigue performance prediction method based on machine learning, which comprises the steps of firstly collecting material data to form initial sample data, adopting a Sobol global sensitivity analysis method to evaluate the contribution degree of each characteristic variable to the variance of an output variable, and evaluating the contribution degree of each characteristic variable to the variance of the output variableThe importance of the training sample data is ranked, and then key characteristic variables are screened out to form training sample data; dividing training sample data into a training set, a verification set and a test set for training a model; performing model training learning based on training set and validation set data, obtaining corresponding model parameters, outputting model based on existing training set/validation set learning training, and adopting determination coefficient R2And quantitatively evaluating the prediction performance of the model, completing the establishment of the model, and predicting the fatigue performance by using the trained model. The invention utilizes more comprehensive material information to establish the nonlinear corresponding relation between the fatigue performance and the material information and is applied to fatigue performance prediction in higher precision and generalized environment.
Description
Technical Field
The invention belongs to the technical field of fatigue performance evaluation, and particularly relates to a fatigue performance prediction method based on machine learning.
Background
In the industrial fields of automobiles, aerospace and the like, fatigue failure is a main mode of failure of key structural parts. The section of the mechanical engineering handbook of China "structural fatigue strength design" indicates that: more than 80% of mechanical components are in fatigue failure. The fatigue performance data of the material is important design information for product development, but the cost of the material durability test is high and the period is long (the test period of a single fatigue curve is more than 1 month, and the test cost is more than 10 ten thousand RMB (without material cost and processing cost)). Since 1870, the link between fatigue strength and tensile properties of metallic materials has been widely explored, and the main objective is to predict the fatigue strength of materials quickly by using tensile tests with lower test cost, but this technical route has obvious defects. Firstly, the deformation uniformity of the tensile deformation is obviously higher than that of the fatigue deformation, in fact, most of the fatigue deformation is caused by local defects or stress concentration, and the fatigue deformation with smaller stress amplitude is particularly obvious; secondly, the deformation mechanisms of fatigue deformation and tensile deformation are in complex association, and the fatigue deformation and the tensile deformation are only associated under specific conditions, so that more parameters except the tensile property are generally required to predict the fatigue performance; third, the effect and degree of effect of changes in composition and process parameters on tensile and fatigue properties are not necessarily the same or similar.
With the rapid development of computer technology and data analysis technology and the rapid growth of the scale of material testing and simulation data, the concept of "Materials information" has been attracting attention. The advanced modern data analysis statistics and modeling method gradually enters the traditional material research and development field, and the method has a primary value in the aspects of reducing development cost, shortening development period and the like. The method is based on basic information (components, process, hardness, organization and other information) of the material and completely depends on data driving, accurately and quickly predicts the fatigue life of the material, and has great significance for research and development of new materials, acceleration of material tests and forward optimization design of industrial products.
Patent publication No. CN109855959A discloses a method for predicting fatigue strength of a metal material, which comprises the following steps: 1. selecting materials of the same series for tensile property test; 2. testing fatigue performance; 3. parameter fitting: determining the sigma of the material by using the measured tensile and fatigue datay/σbAnd σw/σyValue (note: sigma)yIs the yield strength, σbIs the tensile strength, σwIs fatigue strength) and then by σy/σbThe value being abscissa, in σw/σyValue plotting sigma for ordinatew/σy—σy/σbObtaining parameters omega and C (slope and intercept respectively) through linear fitting; 4. determining the value of sigma y/sigma b of the material through the tensile property of the material to be predicted, and further determining the sigma of the material through the fitted straight linew/σyValue, thereby obtaining fatigue strength σ of the corresponding materialwAnd (5) predicting the value. The method realizes the purpose of predicting the fatigue strength of the same series of metal materials by utilizing a tensile test and a small amount of fatigue tests by establishing a corresponding relation model among the fatigue strength, the yield strength and the tensile strength. However, the corresponding relation between the fatigue strength and the tensile property is only established by the technology, and the relation model adopts the simplest linear fitting and is influenced by factors such as manufacturing process, macro segregation, component adjustment and the like, so that the method cannot obtain a satisfactory prediction result, and the generalization capability of the model is poor.
Disclosure of Invention
In order to overcome the defects in the prior art, the invention provides a fatigue performance prediction method based on machine learning, which utilizes more comprehensive material information to establish the nonlinear corresponding relation between the fatigue performance and the material information and is applied to fatigue performance prediction in higher precision and generalized environment.
The technical scheme adopted by the invention is as follows:
a fatigue performance prediction method based on machine learning comprises the following steps,
collecting material data, wherein the material data comprises related data of material composition, microstructure parameters, heat treatment process parameters, processing process parameters, material mechanical properties and material physical properties to form initial sample data, and is expressed as S { (x { (X)1，y1)；(x2，y2)；…(xn，yn)}，xiIs a characteristic variable of the ith group of samples, i is 1, 2. y isiFor group i samples at 106Fatigue strength at cycle life, and xiAs input vector for machine learning, yiAs an output vector for machine learning; and carrying out normalization processing on the sample data;
evaluating the contribution degree of each characteristic variable to the variance of the output variable by adopting a Sobol global sensitivity analysis method, and obtaining the contribution amount of the change of each characteristic variable to the variance of the label variable through variance analysis; sorting the importance of the characteristic variables, further screening out key characteristic variables and forming training sample data;
dividing training sample data into a training set, a verification set and a test set for training a model; performing model training learning based on training set and validation set data, obtaining corresponding model parameters, outputting model based on existing training set/validation set learning training, and adopting determination coefficient R2And quantitatively evaluating the prediction performance of the model, completing the establishment of the model, and predicting the fatigue performance by using the trained model.
Further, a characteristic variable x in the initial sample dataiThe dimensions v of (a) are 13, and the weight percentages of the 8 elements C, Si, Mn, P, S, Cr, Ni and Cu, the yield strength, the tensile strength, the elongation, the reduction of area and the Brinell hardness are respectively.
Further, normalization processing is performed on the characteristic values in the sample data by adopting a range transform method, and the normalization processing is expressed as:
wherein x isi，jNormalized for the jth feature in the ith set of samples,
and
respectively the minimum value and the maximum value of the jth characteristic in all sample data.
And further, performing kernel principal component analysis on the normalized sample data by adopting a principal component analysis method, mapping a characteristic variable matrix in the initial sample data from a low-dimensional space to a high-dimensional linear space to realize linearization of nonlinear numbers, and compressing and removing part of redundant noise information in the data.
Further, the existing characteristic variable matrix is mapped to a high-dimensional linear space from a low-dimensional space through an RBF kernel function to realize the linearization of nonlinear numbers:
RBF kernel function:
where the kernel parameter σ is a real number, x, predefined in advanceiAnd xjThe normalized sample data of the ith group and the jth group.
Further, a large number of samples are generated in a given characteristic variable analysis space through Sobol sampling, and the analysis result is 106Determining sample size according to computing power of computer, with recommended value of 10vAnd v is the total number of characteristic variables.
Furthermore, a BP neural network model is selected as the model, and a two-layer feedforward network comprises an S-shaped hidden neuron and a linear output neuron and is suitable for solving the multidimensional mapping problem; and setting the number of hidden neurons as 10, training and learning the data of the training set and the verification set, acquiring corresponding model parameters, and outputting the NNF model based on the existing training set/verification set learning and training.
Further, the model is selected from linear regression, fine tree, quadratic support vector machine regression or rational quadratic gaussian process regression.
The invention has the beneficial effects that:
1. compared with the traditional fatigue performance prediction method, the method adopted by the invention mainly depends on the existing data in the material database, and can obtain a fatigue performance prediction result with higher accuracy by combining a small amount of fatigue tests. The characteristic of the invention not only saves a large amount of testing cost in engineering application, but also can give full play to the function of the existing database and help engineering technicians to understand and master the relation and rule among different material data.
2. Compared with a fatigue performance prediction method which solely depends on machine learning, the multi-fidelity optimization method adopted by the invention has higher reliability and cost performance. Although machine learning can establish a nonlinear correspondence relationship between different pieces of information of a material, when the relevance between the pieces of information is low, problems such as overfitting and the like are often caused, so that the generalization capability of a model is poor, and the actual prediction effect is poor. The multi-fidelity optimization method disclosed by the invention compensates and optimizes the problem, a machine learning model obtained by database data training is limited to be a low-fidelity model, and fatigue performance of a small amount of real tests is used as high-fidelity data and used for calibrating the low-fidelity model, so that the contradiction of machine learning overfitting and high fatigue test cost is well balanced.
3. The invention can be used for optimizing the material preparation process. The present invention is currently labeled as fatigue performance, and may be labeled with virtually any performance or organization parameter. The preparation process parameters of the material can be input as a model, and conversely, the preparation process parameters can be optimized by optimizing the label result, so that the industrial application target of improving the material preparation level is realized.
Drawings
FIG. 1 is a flow diagram of fatigue performance prediction based on machine learning.
FIG. 2 is a graph showing the result of principal component analysis by KPCA.
FIG. 3 is a feature variable importance ranking diagram.
FIG. 4 is a sample distribution plot for a training set and a test set.
FIG. 5 is a graph comparing predicted values with experimental values.
Fig. 6 error distribution histogram.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
A fatigue performance prediction method based on machine learning comprises the following steps:
1. data preparation and preprocessing. First, material data is collected, the material data including material composition, microstructural parametersData such as heat treatment process parameters, processing process parameters, mechanical properties of materials, physical properties of materials and the like; for example, a certain steel material needs to be collected, and the composition, tensile and fatigue data of the steel material need to be collected, so that the generalization prediction capability of the model is improved, and the collected data series cover different technological processes such as heat treatment (such as quenching, tempering, normalizing, inductive quenching, isothermal quenching, carburizing and the like) and plastic working (such as rolling, forging, extrusion and the like). Obtaining initial sample data with perfect information through cleaning and screening, namely S { (x)1，y1)；(x2，y2)；…(xn，yn)}. Wherein the total number n of samples is 105; x is the number ofi∈Rv( i 1, 2.. times.n) is a characteristic variable of the ith group of samples, and the characteristic variable xiContains 8 main elements (i.e. C, Si, Mn, P, S, Cr, Ni and Cu) in weight percentage, yield strength, tensile strength, elongation, reduction of area and Brinell hardness, so that the dimension v of the characteristic variable is 13; a characteristic variable xiAs an input vector for machine learning; y isiIs the i-th sample at 106Fatigue strength at cycle life and as an output vector for machine learning (also known as a "label").
Then, normalizing all sample characteristic values by adopting a range transform method, wherein a calculation formula of the range transform method is as follows:
wherein x isi,jNormalized for the jth feature in the ith set of samples,
and
respectively the minimum value and the maximum value of the jth characteristic in all sample data.
2. And (4) analyzing the main components. Kernel Principal Component Analysis (KPCA) was performed on the normalized sample data using Kernel Principal Component Analysis. And mapping the existing characteristic variable matrix from a low-dimensional space to a high-dimensional linear space through an RBF kernel function to realize linearization of nonlinear numbers.
RBF kernel function:
where the kernel parameter σ is a real number, x, predefined in advanceiAnd xjThe normalized sample data of the ith group and the jth group. And performing dimensionality reduction on the data mapped to the high-dimensional linear space by adopting principal component analysis, realizing data compression and removing partial redundant noise information in the data. The principal component contribution of the high-dimensional linear spatial data after principal component analysis is shown in fig. 2. It can be seen that the cumulative interpretation ability of the first, second, and third pivot elements for 13 feature variables in the original data set has reached 91.7%, which means that 13-dimensional feature variables in the original data set can be effectively reduced to 3-dimensional, and since the feature variables in this case are of few types and all have important physical meanings, the original feature variables are not subjected to dimension reduction processing during model training.
3. Feature importance ranking. And evaluating the contribution degree of each characteristic variable to the variance of the output variable by adopting a Sobol global sensitivity analysis method, and sequencing the importance of the characteristic variables. Generating a large number of samples in a given characteristic variable analysis space through Sobol sampling, wherein the sample size is determined according to the computing power of a computer, and the recommended value is 10vAnd v is the total number of characteristic variables. And predicting the label of each sample by adopting a machine learning model, and obtaining the contribution of the change of each characteristic variable to the variance of the label variable through variance analysis. As shown in FIG. 3, the Yield strength (Yield strength), Hardness (Hardness) and tensile strength (Ultimate tensile strength) were 10 times higher than those of the conventional steel sheet6Fatigue strength at cycle life is quite closely related in nature.
4. Training set/test set partitioning. Whether the training set is reasonable or not has an important influence on the generalization capability of the model. Principal component analysis results show that the first principal component and the second principal component have 87% of interpretation capability on the characteristics of the original data set, and the first principal component plan and the second principal component plan shown in fig. 4 can visually reflect the basic characteristics of a characteristic space. In the present embodiment, the samples have uniform feature distribution, so that a random sampling method is adopted, and about 70% of the samples are selected as a training set (i.e., 74 samples), about 10% of the samples are selected as a verification set (i.e., 10 samples), and about 20% of the samples are selected as a test set (i.e., 21 samples), and the test sets are uniformly distributed in the coverage of the training set.
5. And selecting and training a model. In the scheme, a BP Neural network Fitting method (Neural network Fitting) based on a Levenberg-marquardt algorithm is selected, and the two-layer feedforward network comprises an S-type hidden neuron and a linear output neuron, so that the method is suitable for solving the problem of multi-dimensional mapping. And (3) using an NNF method, setting the number of hidden neurons to be 10, training and learning the data of the training set and the verification set, acquiring corresponding model parameters, and outputting an NNF model trained based on the existing training set/verification set learning. Machine learning methods also include linear regression, fine trees, quadratic Support Vector Machine (SVM) regression, rational quadratic gaussian process regression, and the like.
6. And (6) evaluating the model. In this case, the coefficient R is determined2To quantitatively evaluate the model predicted performance. The calculation formula is as follows:
as shown in FIG. 5, the test set prediction result R in this case20.91, overall data set prediction result R20.90, the model prediction performance is good. In order to further grasp the distribution of the deviation of the model with respect to the predicted results of all samples, an error distribution histogram of the model predicted results is plotted with the number of samples in a certain error interval as ordinate and the corresponding error interval as abscissa, as shown in fig. 6. Generally, the error histogram should be in unimodal form, distributed axisymmetrically with X ═ 0, and the more accurate the prediction, the higher the peak value. By determining the coefficient R2And the values are combined with the error distribution histogram and the prediction/experimental value comparison graph, so that the prediction performance of the training model can be basically judged.
The above embodiments are only used for illustrating the design idea and features of the present invention, and the purpose of the present invention is to enable those skilled in the art to understand the content of the present invention and implement the present invention accordingly, and the protection scope of the present invention is not limited to the above embodiments. Therefore, all equivalent changes and modifications made in accordance with the principles and concepts disclosed herein are intended to be included within the scope of the present invention.
Claims (8)
1. A fatigue performance prediction method based on machine learning is characterized in that material data are collected, wherein the material data comprise related data of material composition, microstructure parameters, heat treatment process parameters, processing process parameters, material mechanical properties and material physical properties to form initial sample data, and the initial sample data are expressed as S { (x {)1，y1)；(x2，y2)；…(xn，yn)}，xiIs a characteristic variable of the ith group of samples, i is 1, 2. y isiFor group i samples at 106Fatigue strength at cycle life, and xiAs input vector for machine learning, yiAs an output vector for machine learning; and carrying out normalization processing on the sample data;
evaluating the contribution degree of each characteristic variable to the variance of the output variable by adopting a Sobol global sensitivity analysis method, and obtaining the contribution amount of the change of each characteristic variable to the variance of the label variable through variance analysis; sorting the importance of the characteristic variables, and further screening out key characteristic variables to form training sample data;
dividing training sample data into a training set, a verification set and a test set for training a model; performing model training learning based on training set and validation set data, obtaining corresponding model parameters, outputting model based on existing training set/validation set learning training, and adopting determination coefficient R2And quantitatively evaluating the prediction performance of the model, completing the establishment of the model, and predicting the fatigue performance by using the trained model.
2. The method according to claim 1, wherein the feature variable x in the initial sample data is a feature variable x in the fatigue performance prediction method based on machine learningiHas a dimension v of 13, respectivelyThe weight percentage of 8 elements of C, Si, Mn, P, S, Cr, Ni and Cu, yield strength, tensile strength, elongation, reduction of area and Brinell hardness.
3. The method for predicting fatigue performance based on machine learning according to claim 1, wherein the eigenvalue in the sample data is normalized by a range transform method, which is expressed as:
wherein x isi,jNormalized for the jth feature in the ith set of samples,
and
respectively the minimum value and the maximum value of the jth characteristic in all sample data.
4. The fatigue performance prediction method based on machine learning according to claim 1, characterized in that a principal component analysis method is adopted to perform kernel principal component analysis on the normalized sample data, a characteristic variable matrix in the initial sample data is mapped from a low-dimensional space to a high-dimensional linear space to realize linearization of nonlinear numbers, and partial redundant noise information in the data is compressed and removed.
5. The machine learning-based fatigue performance prediction method of claim 4, wherein the existing characteristic variable matrix is mapped from a low-dimensional space to a high-dimensional linear space through an RBF kernel function to realize linearization of nonlinear numbers:
RBF kernel function:
where the kernel parameter σ is a real number, x, predefined in advanceiAnd xjThe normalized sample data of the ith group and the jth group.
6. The method for predicting fatigue performance based on machine learning as claimed in any one of claims 1-5, wherein a large number of samples are generated in a given characteristic variable analysis space by Sobol sampling, and the analysis result is 106Determining sample size according to computing power of computer, with recommended value of 10vAnd v is the total number of characteristic variables.
7. The machine learning-based fatigue performance prediction method of claim 6, wherein the model is a BP neural network model, and the two-layer feedforward network comprises S-type hidden neurons and linear output neurons, and is suitable for solving the multidimensional mapping problem; and setting the number of hidden neurons as 10, training and learning the data of the training set and the verification set, acquiring corresponding model parameters, and outputting the NNF model based on the existing training set/verification set learning and training.
8. The method of claim 7, wherein the model is selected from linear regression, fine tree, quadratic support vector machine regression, and rational quadratic Gaussian process regression.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011123975.5A CN112214933B (en) | 2020-10-20 | 2020-10-20 | Fatigue performance prediction method based on machine learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011123975.5A CN112214933B (en) | 2020-10-20 | 2020-10-20 | Fatigue performance prediction method based on machine learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112214933A true CN112214933A (en) | 2021-01-12 |
| CN112214933B CN112214933B (en) | 2024-04-19 |
Family
ID=74055991
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011123975.5A Active CN112214933B (en) | 2020-10-20 | 2020-10-20 | Fatigue performance prediction method based on machine learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112214933B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113345538A (en) * | 2021-06-23 | 2021-09-03 | 北京理工大学重庆创新中心 | Material performance prediction method based on dynamic selection training set |
| CN113807016A (en) * | 2021-09-22 | 2021-12-17 | 华东理工大学 | Data-driven engineering material ultra-high cycle fatigue life prediction method |
| CN114486515A (en) * | 2021-12-15 | 2022-05-13 | 中国科学院金属研究所 | Creep graphite cast iron fatigue strength prediction method based on microstructure and tensile property |
| CN114781250A (en) * | 2022-04-11 | 2022-07-22 | 西南交通大学 | Multi-factor influence environment fatigue life prediction method based on machine learning |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107704920A (en) * | 2017-10-16 | 2018-02-16 | 安徽工业大学 | One kind is based on BP neural network roll alloy contact prediction of fatigue behaviour method |
| CN108153999A (en) * | 2018-01-26 | 2018-06-12 | 大连交通大学 | A kind of welding point fatigue life intelligent Forecasting based on fatigue characteristic domain |
| CN110910972A (en) * | 2019-11-20 | 2020-03-24 | 长沙理工大学 | Fatigue stress concentration coefficient prediction method based on Gaussian process |
| CN111460731A (en) * | 2020-03-30 | 2020-07-28 | 苏州热工研究院有限公司 | Development method of RPV material irradiation embrittlement prediction model based on machine learning algorithm |
| CN111695260A (en) * | 2020-06-12 | 2020-09-22 | 上海大学 | Material performance prediction method and system |
| AU2020101854A4 (en) * | 2020-08-17 | 2020-09-24 | China Communications Construction Co., Ltd. | A method for predicting concrete durability based on data mining and artificial intelligence algorithm |
-
2020
- 2020-10-20 CN CN202011123975.5A patent/CN112214933B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107704920A (en) * | 2017-10-16 | 2018-02-16 | 安徽工业大学 | One kind is based on BP neural network roll alloy contact prediction of fatigue behaviour method |
| CN108153999A (en) * | 2018-01-26 | 2018-06-12 | 大连交通大学 | A kind of welding point fatigue life intelligent Forecasting based on fatigue characteristic domain |
| CN110910972A (en) * | 2019-11-20 | 2020-03-24 | 长沙理工大学 | Fatigue stress concentration coefficient prediction method based on Gaussian process |
| CN111460731A (en) * | 2020-03-30 | 2020-07-28 | 苏州热工研究院有限公司 | Development method of RPV material irradiation embrittlement prediction model based on machine learning algorithm |
| CN111695260A (en) * | 2020-06-12 | 2020-09-22 | 上海大学 | Material performance prediction method and system |
| AU2020101854A4 (en) * | 2020-08-17 | 2020-09-24 | China Communications Construction Co., Ltd. | A method for predicting concrete durability based on data mining and artificial intelligence algorithm |
Non-Patent Citations (4)
| Title |
|---|
| K GENEL等: "Application of artificial neural network for predicting strain-life fatigue properties of steels on the basis of tensile tests", INTERNATIONAL JOURNAL OF FATIGUE, vol. 26, no. 10, 31 October 2004 (2004-10-31), pages 1027 - 1035 * |
| MYLLYKOSKI P等: "Development of Prediction Model for Mechanical Properties of Batch Annealed Thin Steel Strip by using Artificial Neural Network Modelling", JOURNAL MATERIAL PROCESSING TECHNOLOGY, vol. 60, no. 4, 31 December 1996 (1996-12-31), pages 399 - 404 * |
| 左秀荣, 姜茂发, 薛向欣: "人工神经网络在钢铁材料力学性能预测方面的应用", 特殊钢, no. 05, 21 October 2004 (2004-10-21), pages 26 - 29 * |
| 齐秀美;: "基于BP神经网络的热轧低碳钢力学性能预测", 河北冶金, no. 09, 28 September 2020 (2020-09-28), pages 55 - 57 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113345538A (en) * | 2021-06-23 | 2021-09-03 | 北京理工大学重庆创新中心 | Material performance prediction method based on dynamic selection training set |
| CN113345538B (en) * | 2021-06-23 | 2022-09-30 | 北京理工大学重庆创新中心 | Material performance prediction method based on dynamic selection training set |
| CN113807016A (en) * | 2021-09-22 | 2021-12-17 | 华东理工大学 | Data-driven engineering material ultra-high cycle fatigue life prediction method |
| CN114486515A (en) * | 2021-12-15 | 2022-05-13 | 中国科学院金属研究所 | Creep graphite cast iron fatigue strength prediction method based on microstructure and tensile property |
| CN114486515B (en) * | 2021-12-15 | 2023-05-30 | 中国科学院金属研究所 | Vermicular cast iron fatigue strength prediction method based on microstructure and tensile property |
| CN114781250A (en) * | 2022-04-11 | 2022-07-22 | 西南交通大学 | Multi-factor influence environment fatigue life prediction method based on machine learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112214933B (en) | 2024-04-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112214933B (en) | Fatigue performance prediction method based on machine learning | |
| Wang et al. | Fatigue-life prediction of additively manufactured metals by continuous damage mechanics (CDM)-informed machine learning with sensitive features | |
| Quan et al. | Artificial neural network modeling to evaluate the dynamic flow stress of 7050 aluminum alloy | |
| Yun | Prediction model of algal blooms using logistic regression and confusion matrix | |
| Derazkola et al. | Review on modeling and simulation of dynamic recrystallization of martensitic stainless steels during bulk hot deformation | |
| CN112949005B (en) | High-strength and high-toughness steel design method based on machine learning guidance | |
| CN111310348A (en) | Material constitutive model prediction method based on PSO-LSSVM | |
| McBride et al. | Modelling tensile properties of gamma-based titanium aluminides using artificial neural network | |
| Hong et al. | Exploration of machine learning to predict hot ductility of cast steel from chemical composition and thermal conditions | |
| Liu et al. | Machine learning for extending capability of mechanical characterization to improve springback prediction of a quenching and partitioning steel | |
| CN116502455A (en) | Process parameter determination method and system for laser selective melting technology | |
| Puchi-Cabrera et al. | Machine learning aided nanoindentation: A review of the current state and future perspectives | |
| Jimenez-Martinez et al. | Fatigue Life Prediction of Aluminum Using Artificial Neural Network. | |
| CN114580286A (en) | Neural network model optimization method based on stainless steel ultra-thin strip annealing process | |
| Zhong et al. | A process-data-driven BP neural network model for predicting interval-valued fatigue life of metals | |
| Wu et al. | Deformation resistance prediction of tandem cold rolling based on grey wolf optimization and support vector regression | |
| CN116484745A (en) | Design method of hot-rolled low-alloy steel for physical metallurgy guidance industry big data mining | |
| Wei et al. | High-cycle fatigue SN curve prediction of steels based on a transfer learning-guided convolutional neural network | |
| CN114692507A (en) | Counting data soft measurement modeling method based on stacking Poisson self-encoder network | |
| CN112241832A (en) | Product quality grading evaluation standard design method and system | |
| Yan et al. | AP-GAN-DNN based creep fracture life prediction for 7050 aluminum | |
| CN111063401A (en) | BP neural network-based thermal treatment state Mg-Zn-Zr alloy structure and mechanical property prediction method | |
| Gocheva-Ilieva | Using MARS for modeling and predicting tensile strength of low alloy steels | |
| Duan et al. | Analysis of High-Cycle Fatigue Life Prediction of 304 Stainless Steel Based on Deep Learning | |
| Kulina et al. | Ensemble Learning for Predicting the Tensile Strength of Alloy Steels from Chemical Composition and Processing Parameters |
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 | ||
| GR01 | Patent grant |