CN117332676A - Fatigue performance prediction method based on self-adaptive feature selection - Google Patents

Abstract

The invention relates to a fatigue performance prediction method based on self-adaptive feature selection, which comprises the following steps: s1, collecting material related data, and dividing a data set into a training set and a testing set; s2, taking a correlation combination of the training set and the testing set as a characteristic weight, and taking sensitive characteristics of the characteristic weight as sample data; s3, inputting sample data into a support vector machine regression model for training, and inputting a sample test set into the trained support vector machine regression model for evaluating model performance; s4, inputting the combined test set into the integrated regression model, inputting the combined test set into the trained integrated regression model, and using the estimated model performance. The method can help identify the most relevant features of fatigue performance prediction, thereby potentially improving the accuracy of the prediction model, and the model can intensively consider key factors influencing fatigue behavior through selecting information features and eliminating irrelevant features.

Description

A fatigue performance prediction method based on adaptive feature selection

Technical field

The invention belongs to the technical field of material fatigue performance evaluation, and specifically relates to a fatigue performance prediction method based on adaptive feature selection.

Background technique

In modern industrial production, the fatigue performance of materials or parts is a very important physical quantity, which plays a vital role in ensuring the quality, life and safety of products. In the past, fatigue performance prediction was usually based on engineering experience and test data. This method has great dependence, low reliability and high cost. In recent years, the widespread application of machine learning methods has brought new opportunities for fatigue performance prediction.

However, there are still some technical problems that need to be solved in the process of applying machine learning to fatigue performance prediction. For example, how to preprocess input data and reduce data noise and errors; how to design and select appropriate feature extraction and processing techniques to capture important features of materials or parts; how to select appropriate machine learning algorithms and conduct Optimization to improve the accuracy, precision and robustness of fatigue performance prediction; and how to evaluate and verify prediction models to determine whether their quality and performance meet requirements, etc.

Contents of the invention

The purpose of the present invention is to overcome the shortcomings of the existing technology and provide a fatigue performance prediction method based on adaptive feature selection.

The technical problems solved by the present invention are achieved through the following technical solutions:

A fatigue performance prediction method based on adaptive feature selection, characterized in that: the steps of the prediction method are:

S1. Collect material-related data, normalize the feature data to form a data set, and divide the data set into a training set and a test set;

S2. Use mutual information to extract the correlation between features and target variables. Use the adaptive weighting method to combine the correlation between the training set and the test set as feature weights. Use the feature weights as the input of adaptive feature selection to screen sensitive features. , using sensitive features as sample data;

S3. Divide the sample data into a sample training set and a sample test set, input the sample training set into the support vector machine regression model for training, input the sample test set into the trained support vector machine regression model, and use the determination coefficient R ² to evaluate model performance;

S4. Use the enumeration method to combine non-sensitive features and sensitive features, divide the combined data into a combined training set and a combined test set, use BaggingRegressor to build an integrated regression model based on SVR, and input the combined test set into the integrated regression model. The combined test set was input into the trained ensemble regression model and the coefficient of determination ^R2 was used to evaluate model performance.

Moreover, the specific steps of S1 are:

The collected material-related data includes chemical composition, process parameters, upstream processing characteristics and corresponding fatigue strength. The formed data set is expressed as F={f _i,j ,y _i ,i=1,2,...,n;j =1,2,…,k},

Among them: n is the number of samples;

k is the number of features;

f _i,j is the j-th feature of the i-th sample;

y _i is the fatigue strength of the i-th sample;

Normalize the characteristics of the data and divide the normalized data set into a training set and a test set;

The normalization process is expressed as:

Among them: f' _i,j is the normalized value of the j-th feature in the i-th sample;

x _j is all the eigenvalues of the jth one.

Moreover, the specific steps of S2 are:

Mutual information is used to extract the correlation between the characteristic variables and fatigue intensity of the training set and test set. Mutual information is a concept that measures the degree of correlation between two random variables and is not limited to linear relationships. It is based on the concept of information theory and is used to describe the statistical dependence between two variables, that is, the degree of information sharing between them. The calculation of mutual information involves the joint probability distribution of two random variables and their respective marginal probability distributions. In continuous random variables, the calculation formula is as follows:

Among them: p(f _j ,Y) is the joint probability density function of f _j and Y;

p(f _j ) and p(Y) are the marginal probability density functions of f _j and Y respectively;

I(f _j ; Y) is the mutual information of f _j and Y, which is used to measure the correlation between two random variables f _j and Y;

The correlation between the training set and the test set is combined using adaptive weighting to form feature weights. The calculation formula is as follows:

W _j = αI ₁ (f _j ; Y) + βI ₂ (f _j ; Y)

Among them: W _j is the feature weight;

I ₁ (f _j ; Y) is the correlation of the training set;

I ₂ (f _j ; Y) is the correlation of the test set;

α, β are weight coefficients, α+β=1;

The threshold l _α is calculated based on the feature weight and significance level. Features whose feature weight W _j exceeds the threshold l _α are selected as sensitive features as sample data. The calculation formula is as follows:

Among them: μ is the mean value of the combination weight;

σ is the variance of the combination weight;

α is the chi-square distribution The significance level, 2μ ² /σ is the degree of freedom.

The advantages and beneficial effects of the present invention are:

1. The fatigue performance prediction method of the present invention based on adaptive feature selection can help identify the features most relevant to fatigue performance prediction, thereby potentially improving the accuracy of the prediction model. Through the selection of information features and the elimination of irrelevant features, The model is able to focus on the key factors that influence fatigue behavior.

2. The present invention's fatigue performance prediction method based on adaptive feature selection uses data to identify important features, allowing the model to adapt and learn from available information. This data-driven approach is useful when dealing with complex fatigue mechanisms that traditional models cannot fully capture. The method has higher accuracy and adaptability, and enumerating non-sensitive features can take into account the interaction and mutual influence between features, thereby gaining a more comprehensive understanding of the contribution of features to the target variable.

Description of drawings

Figure 1 is a flow chart of the present invention;

Figure 2 is a schematic diagram of feature screening in the present invention;

Figure 3 is a comparison chart of the prediction results of the original data and sensitive features of the present invention;

Figure 4 is a schematic diagram of the prediction results of the combined features of the present invention.

Detailed ways

The present invention will be further described in detail below through specific examples. The following examples are only descriptive, not restrictive, and cannot be used to limit the scope of the present invention.

As shown in Figure 1, a fatigue performance prediction method based on adaptive feature selection is characterized by: including the following steps:

Step S1: The collected material data includes chemical composition, process parameters, upstream processing characteristics and corresponding fatigue strength. The data set is expressed as F={fi _,j ,y _i ,i=1,2,...,n;j =1,2,…,k}, where: n is the number of samples, here is 437, k is the number of features, here is 25, including chemical composition variables (C, Si, Mn, P, S, Ni , Cr, Cu, Mo), process parameter variables (normalizing temperature, overhardening temperature, overhardening time, overhardening cooling rate, carburizing temperature, carburizing time, diffusion temperature, diffusion time, quenching medium temperature, tempering temperature , tempering time, tempering cooling rate), upstream processing characteristic variables (ratio of ingot to rod, area ratio of plastic deformation inclusions, area ratio of inclusions in discontinuous arrangements, area ratio of isolated inclusions), f _i,j is The j-th characteristic of the i-th sample, _yi is the rotational bending fatigue strength of the i-th sample.

The characteristics of the data are normalized, and the normalized data set is randomly divided into 80% training set and 20% test set. The normalization processing is expressed as:

Among them: f′ _i,j is the normalized value of the j-th feature in the i-th sample;

f _j is all the eigenvalues of column j.

Step S2: Use mutual information to evaluate the characteristic weight between characteristic variables and fatigue strength. In continuous random variables, the calculation formula is as follows:

Among them: p(f _j ,Y) is the joint probability density function of f _j and Y;

p(f _j ) and p(Y) are the marginal probability density functions of f _j and Y respectively;

The correlation between the training set and the test set is adaptively assigned a weight coefficient to calculate the feature weight W _j . The calculation formula is as follows:

W _j = αI ₁ (f _j ; Y) + βI ₂ (f _j ; Y)

Among them: W _j is the feature weight;

I ₁ (f _j ; Y) is the correlation of the training set;

I ₂ (f _j ; Y) is the correlation of the test set;

α, β are weight coefficients, α+β=1;

The threshold l _α is calculated based on the feature weight and significance level. The significance level is 0.01. Features whose feature weight W _j exceeds the threshold l _α are selected as sensitive features. The sample data is shown in Figure 2. The calculation formula is as follows:

Among them, μ is the mean value of the combination weight, σ is the variance of the combination weight, and α is the chi-square distribution. The significance level, 2μ ² /σ is the degree of freedom.

Step S3: Divide the sample data into 80% sample training set and 20% sample test set. The sample training set is input into the support vector regression model for training, the parameters of the model are obtained, and the sample test set is input into the trained model to use the coefficient of determination. R ² is used to evaluate the performance of the model, and the optimal combination is output through continuous iteration of the model.

The formula for calculating ^R2 is as follows:

Among them, _yi is the real fatigue intensity, To predict fatigue strength,/> is the average value of true fatigue strength.

The optimal weight coefficient combination is α = 0.65, β = 0.35, the test set prediction result of the original data R ² = 0.974, and the test set prediction result of the sensitive feature R ² = 0.961, as shown in Figure 3.

Although a large number of features are removed, the filtered sensitive features can still maintain good predictive ability, which shows that these features have high relevance and importance for the prediction of target variables, which further verifies the effectiveness of adaptive feature selection.

Step S4: Use the enumeration method to combine non-sensitive features and sensitive features. BaggingRegressor constructs 10 SVR base models by randomly sampling the combined training set with replacement, and averages the prediction results of each base model. To obtain the final ensemble prediction results, input the combined test set into the trained ensemble model and use the coefficient of determination R ² to evaluate the model performance.

The optimal combination of features is shown in Figure 2. By reasonably combining sensitive features and non-sensitive features to capture higher-level feature interactions and patterns, its determination coefficient R ² =0.983, as shown in Figure 4. There is a certain improvement compared to R ² for original data and sensitive data. These methods can help models better explain and predict data, and improve model performance in real-world applications.

Although the embodiments and drawings of the present invention have been disclosed for illustrative purposes, those skilled in the art will understand that various substitutions, changes and modifications are possible without departing from the spirit and scope of the present invention and the appended claims. , therefore, the scope of the present invention is not limited to the contents disclosed in the embodiments and drawings.

Claims (3)

1. A fatigue performance prediction method based on self-adaptive feature selection is characterized by comprising the following steps: the prediction method comprises the following steps:

s1, collecting material related data, performing normalization processing on characteristic data to form a data set, and dividing the data set into a training set and a testing set;

s2, extracting correlation between the features and the target variable by using mutual information, combining the correlation of the training set and the testing set by using a self-adaptive weighting method to serve as feature weights, taking the feature weights as input of self-adaptive feature selection, screening sensitive features, and taking the sensitive features as sample data;

s3, dividing sample data into a sample training set and a sample testing set, inputting the sample training set into a support vector machine regression model for training, and inputting the sample testing set into the trained support vector machine regression model to enableBy determining coefficient R ² To evaluate model performance;

s4, combining the non-sensitive features and the sensitive features by using an enumeration method, dividing the combined data into a combined training set and a combined testing set, constructing an integrated regression model based on SVR by using a BaggingReggresor, inputting the combined testing set into the integrated regression model, inputting the combined testing set into the trained integrated regression model, and using a decision coefficient R ² To evaluate model performance.

2. The fatigue performance prediction method based on adaptive feature selection according to claim 1, wherein: the specific steps of the S1 are as follows:

the collected material-related data contains chemical composition, process parameters, upstream process characteristics, and corresponding fatigue strength, and the resulting data set is denoted as f= { F _i,j ,y _i ,i＝1,2,…,n；j＝1,2,…,k}，

Wherein: n is the number of samples;

k is the number of features;

f _i,j a j-th feature that is the i-th sample;

y _i fatigue strength for the ith sample;

normalizing the characteristics of the data, and dividing the normalized data set into a training set and a testing set;

the normalization process is expressed as:

wherein: f's' _i,j Normalized values for the jth feature in the ith sample;

x _j all eigenvalues for the j-th.

3. The fatigue performance prediction method based on adaptive feature selection according to claim 1, wherein: the specific steps of the S2 are as follows:

mutual information is used to extract correlations between the characteristic variables of the training and test sets and the fatigue strength. Mutual information is a concept that measures the degree of association between two random variables and is not limited to linear relationships. It is a concept based on information theory, which is used to describe the statistical dependency between two variables, i.e. the degree of sharing of information between them. The computation of mutual information involves a joint probability distribution of two random variables and a respective edge probability distribution. In the continuous random variable, the calculation formula is as follows:

wherein: p (f) _j Y) is f _j And a joint probability density function of Y;

p(f _j ) And p (Y) is f _j And an edge probability density function of Y;

I(f _j the method comprises the steps of carrying out a first treatment on the surface of the Y) is f _j And Y, which is used to measure two random variables f _j And Y;

the correlation of the training set and the test set is combined by adopting a self-adaptive weighting method to form characteristic weights, and the calculation formula is as follows:

W _j ＝αI ₁ (f _j ；Y)+βI ₂ (f _j ；Y)

wherein: w (W) _j Is a characteristic weight;

I ₁ (f _j the method comprises the steps of carrying out a first treatment on the surface of the Y) is the correlation of the training set;

I ₂ (f _j the method comprises the steps of carrying out a first treatment on the surface of the Y) is the correlation of the test set;

α, β are weight coefficients, α+β=1;

calculating threshold value l according to feature weight and significance level _α Selecting a feature weight W _j Exceeding threshold l _α Is characterized by sensitive characteristics, and is used as sample data, and the calculation formula is as follows:

wherein: mu is the average value of the combination weights;

sigma is the variance of the combining weights;

alpha is chi-square distributionSignificance level of 2 mu ² And/sigma is the degree of freedom.