42CrMo fatigue crack growth rate prediction method based on machine learning

Abstract

The invention discloses a 42CrMo fatigue crack propagation rate prediction method based on machine learning, and the method comprises the steps: obtaining a corresponding relation among a stress intensity factor, a stress ratio and a fatigue crack propagation rate of a 42CrMo material through a fatigue crack propagation test, and taking the corresponding relation as an original data set; dividing the preprocessed original data set into a training set and a test set, constructing a fatigue crack growth rate prediction model, training the model by adopting the training set, and evaluating the trained model by adopting the test set; and inputting the actually measured stress intensity factor and stress ratio of the 42CrMo material to be predicted into the prediction model obtained in the step 3, and outputting through the prediction model to obtain a predicted value of the fatigue crack growth rate of the 42CrMo material to be predicted. According to the method, the fatigue crack growth rate of the 42CrMo material can be efficiently and accurately predicted at low cost, so that the service life of the wind power bearing is prolonged, and potential safety hazards are reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of 42CrMo alloy life prediction, and in particular to a 42CrMo fatigue crack growth rate prediction method based on machine learning. Background Art

[0002] 42CrMo alloy has excellent comprehensive mechanical properties, high strength and toughness, and has price advantages compared with other medium and high alloy steels. It is widely used in wind power, hydropower and other fields to manufacture key basic components such as bearings and gears. Due to the harsh working conditions of bearings, gears and other parts, they are prone to fatigue damage due to alternating loads and impact loads. After three stages of fatigue crack initiation, steady-state crack expansion and rapid instability expansion, fatigue fracture accidents will eventually occur. Therefore, a crack growth rate model is established to predict the fatigue crack growth rate of 42CrMo to provide a basis for structural safety evaluation.

[0003] In recent years, machine learning methods have become cutting-edge directions and hot areas such as material life, performance prediction, and new material design, and have significant advantages in prediction efficiency and cost control. However, the existing technology lacks a machine learning solution for 42CrMo alloy life prediction, so it is urgent to develop a 42CrMo fatigue crack growth rate prediction solution based on machine learning. Summary of the invention

[0004] The purpose of the present invention is to provide a 42CrMo fatigue crack growth rate prediction method based on machine learning, which can predict the fatigue crack growth rate of 42CrMo material at low cost, high efficiency and accurately, thereby increasing the service life of wind turbine bearings and reducing safety hazards.

[0005] The objective of the present invention is achieved through the following technical solutions:

[0006] A 42CrMo fatigue crack growth rate prediction method based on machine learning, the method comprising:

[0007] Step 1: Using fatigue crack growth test, obtain the corresponding relationship between stress intensity factor, stress ratio and fatigue crack growth rate of 42CrMo material, and use it as the original data set;

[0008] Step 2: preprocess the original data set obtained in step 1;

[0009] Step 3, dividing the preprocessed original data set into a training set and a test set, and constructing a fatigue crack growth rate prediction model, using the training set to train the model, using the test set to evaluate the trained model, and optimizing the parameter set of the prediction model through a hyperparameter optimization algorithm until the optimal prediction model is obtained;

[0010] Step 4: Input the actually measured stress intensity factor and stress ratio of the 42CrMo material to be predicted into the prediction model obtained in step 3, and obtain the predicted value of the fatigue crack growth rate of the 42CrMo material to be predicted through the output of the prediction model.

[0011] It can be seen from the technical solution provided by the present invention that the above method can predict the fatigue crack growth rate of 42CrMo material at low cost, high efficiency and accurately, thereby increasing the service life of wind power bearings and reducing safety hazards. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0013] Figure 1 A schematic flow chart of a method for predicting fatigue crack growth rate of 42CrMo based on machine learning provided in an embodiment of the present invention;

[0014] Figure 2 A relationship diagram between the actual value and the predicted value of the fatigue crack growth rate of the training set according to the embodiment of the present invention;

[0015] Figure 3 A graph showing the relationship between the actual value and the predicted value of the fatigue crack growth rate of the test set described in the embodiment of the present invention;

[0016] Figure 4 This is a numerical schematic diagram of using a prediction model to predict fatigue crack growth rate according to an embodiment of the present invention. DETAILED DESCRIPTION

[0017] The following is a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments, which does not constitute a limitation of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0018] like Figure 1 The figure shows a flow chart of a method for predicting fatigue crack growth rate of 42CrMo based on machine learning provided by an embodiment of the present invention, wherein the method comprises:

[0019] Step 1: Using fatigue crack growth test, obtain the corresponding relationship between stress intensity factor, stress ratio and fatigue crack growth rate of 42CrMo material, and use it as the original data set;

[0020] In this step, based on the fatigue test of the CT specimen by the fatigue testing machine, the stress intensity factor ΔK and the corresponding fatigue crack growth rate da / dN data of the 42CrMo material are obtained; the stress intensity factor ΔK, the stress ratio R and the corresponding crack growth rate da / dN are taken as a set of data, and the original data set is formed through multiple sets of data;

[0021] Among them, the fatigue testing machine used is a PLD-100kN microcomputer-controlled electro-hydraulic servo fatigue testing machine;

[0022] The specimen used in the fatigue test of the 42CrMo material to be predicted is the standard compact tensile specimen in GB / T6398-2017 Fatigue Crack Growth Method for Fatigue Test of Metallic Materials;

[0023] In a set of data, the stress intensity factor ΔK and the stress ratio R are input quantities, and the crack growth rate da / dN is the corresponding output quantity. For example, based on the stress ratio of 0.1, a total of 720 sets of data can constitute the original data set.

[0024] Step 2: preprocess the original data set obtained in step 1;

[0025] In this step, feature engineering and standardization are performed on the original data set. The stress ratio R is added to 1 and then squared. The stress intensity factor ΔK obtained through the test is logarithmically transformed. The specific data preprocessing process is as follows:

[0026] df[ΔK]=log(df[ΔK])

[0027] df[R]=sqrt(dfR+1).

[0028] Step 3, dividing the preprocessed original data set into a training set and a test set, and constructing a fatigue crack growth rate prediction model, using the training set to train the model, using the test set to evaluate the trained model, and optimizing the parameter set of the prediction model through a hyperparameter optimization algorithm until the optimal prediction model is obtained;

[0029] In this step, 80% of the original data set after preprocessing is used as the training set and 20% as the test set;

[0030] The fatigue crack growth rate prediction model constructed is an extreme gradient boosting method XGBoost regression prediction model, and the parameter set of the prediction model includes: the maximum depth of the tree (max-depth), the learning rate (learning-rate), the random feature sampling ratio of each tree (colsample-bytree), the sample ratio of each tree (subsample), the minimum loss reduction of node splitting (gamma), the minimum weight required in the child node (min-child-weight), the L1 regularization term weight (reg-alpha), the L2 regularization term weight (reg-lambda) and the number of trees (n-estimators);

[0031] The process of training the model using the training set is as follows: the preprocessed training set data is imported into the constructed fatigue crack growth rate prediction model through the data reading command, and the performance of the model under each parameter combination is compared by traversing all specified parameter combinations to find the optimal parameters, and the constructed prediction model is retrained with the optimal parameters.

[0032] After the prediction model is trained, the model determination coefficient R is used. 2 , mean square error MSE, mean absolute error MAE are used as evaluation indicators of model prediction accuracy to determine whether the prediction model meets the requirements;

[0033] Among them, the model determination coefficient R 2 for:

[0034]

[0035] In the above formula, n is the number of data groups in the test set; y j is the crack growth rate of the jth group of data in the test set; The crack growth rate predicted by the model for the jth group of data in the test set; is the average value of all crack growth rates in the test set;

[0036] The mean absolute error MAE is:

[0037]

[0038] In the above formula, n is the number of data groups in the test set; y i is the actual value of fatigue crack growth rate of the i-th sample; is the predicted value of fatigue crack growth rate of the i-th sample; Represents the absolute error between the predicted value and the actual value of the i-th sample;

[0039] The mean square error MSE is:

[0040]

[0041] In the above formula, n is the number of data groups in the test set; y i is the actual value of fatigue crack growth rate of the i-th sample; is the predicted value of fatigue crack growth rate of the i-th sample; It represents the square of the error between the predicted value and the actual value of the i-th sample.

[0042] In the specific implementation, the hyperparameter optimization algorithm includes defining the search space, evaluating the objective function, and using the Bayesian optimization algorithm to iteratively search for the best hyperparameters to improve the prediction accuracy and generalization ability of the model. The core of hyperparameter optimization is to find a set of parameter values ​​that minimize the objective function (RMSE). In the model of the present invention, the hyperparameter space is first defined, and the possible value ranges of the model's hyperparameters are defined, max-depth: [3, 20]; learning-rate: [0.001, 0.005]; colsample-bytree: [0.75, 1.0]; subsample: [0.75, 1.0];

[0043] gamma:[0,2]; min-child-weight:[1,200]; reg-alpha:[0.1,20]; reg-lambda:[0.1,20], n-estimators:[500,5000]; then use cross validation (K-fold cross validation), set K=10, and set the training set (X train , Y train ) is divided into K parts, and K-1 parts are used for training each time, and 1 part is used for verification:

[0044] (X train , Y train ), (X val , Y val ) = train_val_split(X train , Y train，i )

[0045] When training the model, for each copy, train the model and make predictions:

[0046]

[0047] represents the predicted value of the test set; f represents the model function; X val represents the input data of the test set; θ represents the hyperparameter of the model;

[0048]

[0049] The meaning of this formula is: find the optimal value of the independent variable x by the ratio of the minimum loss function to the regularization term constraint function; l(x) represents the loss function related to the model prediction performance, namely the mean square error (MSE); g(x) represents the constraint of the regularization term; where l(x) represents the nature of the defined loss function.

[0050] Calculate the mean square error (MSE) of each validation, and finally return the average MSE of all folds as the performance indicator of this set of hyperparameters; based on the evaluation results of each time, update the prior distribution of the model to make it more inclined to search for hyperparameters that may perform better, and use the updated distribution to sample new hyperparameters to evaluate the model performance again; repeat the above process until the preset number of iterations is reached or the stopping condition is met, so as to obtain the optimal parameters (optimal solution) of the model.

[0051] For example, the maximum depth of the adjusted tree (max-depth) = 17, the learning rate (learning-rate) = 0.0013126, the random feature sampling ratio of each tree (colsample-bytree) = 0.8335, the sample ratio of each tree (subsample) = 0.8391, the minimum loss reduction of node splitting (gamm,a) = 0.727325, the minimum weight required in the child node (min-child-weight) = 92, the L1 regularization term weight (reg-alpha) = 8.557021, the L2 regularization term weight (reg-lambda) = 9.106427, the number of trees (n-estimators) = 3627 (these parameters are based on the defined hyperparameter space, and the corresponding values ​​that meet the model of the present invention are directly output according to the search);

[0052] The coefficient of determination R of the optimal model 2 =0.9101, mean square error MSE = 2.4098e-08, mean absolute error MAE = 0.0001189.

[0053] Step 4: Input the actually measured stress intensity factor and stress ratio of the 42CrMo material to be predicted into the prediction model obtained in step 3, and obtain the predicted value of the fatigue crack growth rate of the 42CrMo material to be predicted through the output of the prediction model.

[0054] In the specific implementation, as shown in Table 1, based on the trained prediction model, the stress intensity factor ΔK and stress ratio R to be predicted are input for prediction, and the corresponding fatigue crack growth rate da / dN is obtained. The curve is as follows: Figure 4 shown.

[0055] Table 1 Prediction model prediction input variables and results

[0056]

[0057] like Figure 2 FIG. 4 is a graph showing the relationship between the actual value and the predicted value of the fatigue crack growth rate of the training set according to the embodiment of the present invention. Figure 3 FIG. 4 is a graph showing the relationship between the actual value and the predicted value of the fatigue crack growth rate of the test set according to the embodiment of the present invention. Figure 4 The figure shows a numerical schematic diagram of the fatigue crack growth rate predicted by the prediction model according to an embodiment of the present invention. The dotted line indicates that the predicted value is consistent with the actual value. The closer the data point is to the dotted line, the higher the prediction accuracy of the model. According to the evaluation index of the model, it is shown that the model basically meets expectations and the prediction results are good.

[0058] Depend on Figure 2 , Figure 3 as well as Figure 4 It can be seen that most of the data points fall near the dotted line, but some data points are far away from the dotted line, showing a large dispersion. The prediction accuracy can be improved by increasing the sample size of the data set or optimizing the model using a more complex optimization algorithm.

[0059] It is worth noting that the contents not described in detail in the embodiments of the present invention belong to the prior art known to professional and technical personnel in the field.

[0060] In summary, the method described in the embodiment of the present invention can fully consider the nonlinear problem in the process of predicting the fatigue crack growth rate of 42CrMo alloy. Under the same stress ratio conditions, the crack growth rate of the sample can be predicted in a portable, efficient and accurate manner, avoiding the calculation error of the fatigue crack growth rate caused by traditional linear fitting, reducing the limitations of the application scope of the existing fatigue crack growth rate theoretical equation, and reducing the calculation cost and test cost.

[0061] In addition, a person skilled in the art can understand that all or part of the steps in the above-mentioned embodiment method can be implemented by instructing related hardware through a program, and the corresponding program can be stored in a computer-readable storage medium. The above-mentioned storage medium can be a read-only memory, a disk or an optical disk, etc.

[0062] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by any technician familiar with the technical field within the technical scope disclosed in the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims. The information disclosed in the background technology section of this article is only intended to deepen the understanding of the overall background technology of the present invention, and should not be regarded as an admission or in any form that the information constitutes prior art known to those skilled in the art.

Claims

1. A 42CrMo fatigue crack growth rate prediction method based on machine learning, characterized in that: The method comprises: Step 1: Using fatigue crack growth test, obtain the corresponding relationship between stress intensity factor, stress ratio and fatigue crack growth rate of 42CrMo material, and use it as the original data set; Step 2: preprocess the original data set obtained in step 1; Step 3, dividing the preprocessed original data set into a training set and a test set, and constructing a fatigue crack growth rate prediction model, using the training set to train the model, using the test set to evaluate the trained model, and optimizing the parameter set of the prediction model through a hyperparameter optimization algorithm until the optimal prediction model is obtained; Step 4: Input the actually measured stress intensity factor and stress ratio of the 42CrMo material to be predicted into the prediction model obtained in step 3, and obtain the predicted value of the fatigue crack growth rate of the 42CrMo material to be predicted through the output of the prediction model.

2. The 42CrMo fatigue crack growth rate prediction method based on machine learning according to claim 1 is characterized in that: In step 1, based on the fatigue test of the CT specimen by the fatigue testing machine, the stress intensity factor ΔK and the corresponding fatigue crack growth rate da / dN data of the 42CrMo material are obtained; the stress intensity factor ΔK, the stress ratio R and the corresponding crack growth rate da / dN are taken as a set of data, and the original data set is formed through multiple sets of data; Among them, the fatigue testing machine used is a PLD-100kN microcomputer-controlled electro-hydraulic servo fatigue testing machine; The specimen used in the fatigue test of the 42CrMo material to be predicted is the standard compact tensile specimen in GB / T6398-2017 Fatigue Crack Growth Method for Fatigue Test of Metallic Materials; In a set of data, the stress intensity factor ΔK and the stress ratio R are input quantities, and the crack growth rate da / dN is the corresponding output quantity.

3. The 42CrMo fatigue crack growth rate prediction method based on machine learning according to claim 1 is characterized in that: In step 2, feature engineering and standardization are performed on the original data set. The stress ratio R is added to 1 and then squared. The stress intensity factor ΔK obtained through the test is logarithmically transformed. The specific data preprocessing process is as follows: df[ΔK]=log(df[ΔK]) df[R]=sqrt(dfRR+1).

4. The 42CrMo fatigue crack growth rate prediction method based on machine learning according to claim 1 is characterized in that: In step 3, 80% of the original data set after preprocessing is used as the training set and 20% as the test set; The fatigue crack growth rate prediction model constructed is an extreme gradient boosting method XGBoost regression prediction model, and the parameter set of the prediction model includes: the maximum depth of the tree, the learning rate, the random feature sampling ratio of each tree, the sample ratio of each tree, the minimum loss reduction of node splitting, the minimum weight sum required in the child node, the L1 regularization term weight, the L2 regularization term weight and the number of trees; The process of training the model using the training set is: The preprocessed training set data is imported into the constructed fatigue crack growth rate prediction model through the data reading command. The performance of the model under each parameter combination is compared by traversing all specified parameter combinations to find the optimal parameters, and the constructed prediction model is retrained with the optimal parameters.

5. The 42CrMo fatigue crack growth rate prediction method based on machine learning according to claim 1 is characterized in that: In step 3, after the prediction model is trained, the model determination coefficient R 2 , mean square error MSE, mean absolute error MAE are used as evaluation indicators of model prediction accuracy to determine whether the prediction model meets the requirements; Among them, the model determination coefficient R 2 for: In the above formula, n is the number of data groups in the test set; y j is the crack growth rate of the jth group of data in the test set; The crack growth rate predicted by the model for the jth group of data in the test set; is the average value of all crack growth rates in the test set; The mean absolute error MAE is: In the above formula, n is the number of data groups in the test set; y i is the actual value of fatigue crack growth rate of the i-th sample; is the predicted value of fatigue crack growth rate of the i-th sample; Represents the absolute error between the predicted value and the actual value of the i-th sample; The mean square error MSE is: In the above formula, n is the number of data groups in the test set; y i is the actual value of fatigue crack growth rate of the i-th sample; is the predicted value of fatigue crack growth rate of the i-th sample; It represents the square of the error between the predicted value and the actual value of the i-th sample.

Images