CN116842834A - Interpretable creep rupture life prediction method based on machine learning and SHAP value - Google Patents
Interpretable creep rupture life prediction method based on machine learning and SHAP value Download PDFInfo
- Publication number
- CN116842834A CN116842834A CN202310767414.6A CN202310767414A CN116842834A CN 116842834 A CN116842834 A CN 116842834A CN 202310767414 A CN202310767414 A CN 202310767414A CN 116842834 A CN116842834 A CN 116842834A
- Authority
- CN
- China
- Prior art keywords
- creep rupture
- creep
- shap
- machine learning
- input
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 62
- 238000010801 machine learning Methods 0.000 title claims abstract description 37
- 239000000463 material Substances 0.000 claims abstract description 29
- 238000004364 calculation method Methods 0.000 claims abstract description 15
- 230000007613 environmental effect Effects 0.000 claims abstract description 14
- 238000012549 training Methods 0.000 claims description 49
- 238000012360 testing method Methods 0.000 claims description 47
- 238000004519 manufacturing process Methods 0.000 claims description 7
- 238000000137 annealing Methods 0.000 claims description 6
- 238000002790 cross-validation Methods 0.000 claims description 4
- 230000002787 reinforcement Effects 0.000 claims description 2
- 230000001502 supplementing effect Effects 0.000 claims description 2
- 239000000126 substance Substances 0.000 abstract description 11
- 238000012216 screening Methods 0.000 abstract description 6
- 238000002474 experimental method Methods 0.000 abstract description 4
- 238000012545 processing Methods 0.000 abstract description 4
- 238000002360 preparation method Methods 0.000 abstract description 2
- 239000006185 dispersion Substances 0.000 description 5
- 229910052757 nitrogen Inorganic materials 0.000 description 5
- 229910052717 sulfur Inorganic materials 0.000 description 5
- 239000000203 mixture Substances 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- 229910000831 Steel Inorganic materials 0.000 description 3
- 238000010276 construction Methods 0.000 description 3
- 230000007774 longterm Effects 0.000 description 3
- 238000005457 optimization Methods 0.000 description 3
- 239000010959 steel Substances 0.000 description 3
- 229910052799 carbon Inorganic materials 0.000 description 2
- 238000010438 heat treatment Methods 0.000 description 2
- 229910052698 phosphorus Inorganic materials 0.000 description 2
- 238000012163 sequencing technique Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 229910052729 chemical element Inorganic materials 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000013213 extrapolation Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000005272 metallurgy Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000006116 polymerization reaction Methods 0.000 description 1
- 235000013619 trace mineral Nutrition 0.000 description 1
- 239000011573 trace mineral Substances 0.000 description 1
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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2113/00—Details relating to the application field
- G06F2113/26—Composites
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- General Engineering & Computer Science (AREA)
- Geometry (AREA)
- Computer Hardware Design (AREA)
- Data Mining & Analysis (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Testing Resistance To Weather, Investigating Materials By Mechanical Methods (AREA)
Abstract
The invention discloses an interpretable creep rupture life prediction method based on machine learning and SHAP values, which is characterized in that a creep rupture life data set of a material is widely collected through experiments and literature data, the influence of different input characteristics such as chemical components, microstructure parameters, preparation processing parameters, environmental factors and the like of the material on a creep rupture life prediction model is considered, and the marginal contribution of each input characteristic to a model result is calculated through the SHAP method, so that the influence degree of different characteristics on the creep life can be interpreted. Marginal contributions of individual input features to the model result can be interpreted by the SHAP method. Meanwhile, the optimal input characteristic screening step based on the SHAP value and the forward search method can reduce the redundancy degree of the model, improve the calculation efficiency and improve the prediction accuracy of the model.
Description
Technical Field
The invention belongs to the field of material life prediction, and relates to an interpretable creep rupture life prediction method based on machine learning and SHAP values.
Background
Modern process equipment such as reactors, heat exchangers, pressure pipes and the like are widely applied to various fields of energy, chemical industry, metallurgy and the like, and the service environment faced by equipment materials becomes more and more complex. In addition to withstanding the loads of equipment during start-up and shut-down, steady-state operation, equipment critical components are also subjected to severe high temperature environments. Under the long-term action of load and high temperature, a large number of creep voids can be generated in the material, and the growth and polymerization of the creep voids can form creep cracks, so that creep rupture failure occurs. Therefore, a reliable creep rupture life prediction method is established for materials facing long-term high-temperature service environment, the creep rupture life of the materials is accurately and efficiently predicted, and the method has important significance for guaranteeing the safety and reliability of high-temperature process equipment.
Conventional creep rupture life prediction methods mainly include time-temperature-parameter methods such as Larson-Miller parameter methods, which extrapolate based on endurance test results, and life extrapolation methods such as θ projection methods, which extrapolate based on creep curves. However, these conventional methods all require creep experiments to be performed in a laboratory, and the long-term creep rupture life of the material can be estimated by obtaining accelerated creep test data, thus consuming a lot of time and economic costs. On the other hand, the traditional creep rupture life prediction method cannot fully consider the influence of factors such as material chemical components, heat treatment and the like on the creep life, so that the popularization and application of the method in engineering are limited.
Disclosure of Invention
The invention aims to overcome the defects of the prior art, and provides an interpretable creep rupture life prediction method based on machine learning and SHAP values, so as to solve the problem that the influence of various factors on the creep life is difficult to fully consider in the traditional creep life prediction experimental method, and the prediction rate is low.
In order to achieve the purpose, the invention is realized by adopting the following technical scheme:
an interpretable creep rupture life prediction method based on machine learning and SHAP values, comprising the steps of:
step 1, dividing a data set into a training set and a testing set, wherein each data set comprises a plurality of input features and an output feature;
step 2, training a creep rupture life prediction model by using a training set as input data through an extreme gradient enhancement algorithm, and calculating a SHAP value of each input feature according to a training result; ordering the input features in order from high to low with SHAP values;
step 3, comprising the following sub-steps:
step 3.1, selecting the input characteristic with the largest SHAP value to form an input characteristic set, training a creep rupture life prediction model, and obtaining a relative root mean square error;
step 3.2, selecting one input feature from the rest input features each time by adopting a forward search method, supplementing the selected input feature into the previous input feature set, training a creep rupture life prediction model by an extreme gradient enhancement algorithm, and calculating a relative root mean square error;
step 3.3, repeating the step 3.2 until all input features are selected and stopped after calculation;
step 3.4, selecting the input feature set corresponding to the minimum relative root mean square error as the optimal feature set;
and 4, constructing a machine learning model for predicting the predicted creep life by adopting training set data through an optimal feature set, and obtaining a final predicted creep life fracture model after optimizing parameters.
The invention further improves that:
preferably, in step 1, the input features include material chemistry, microstructure parameters, manufacturing process parameters, and environmental impact parameters; the output is characterized by logarithmic creep rupture life.
Preferably, the microstructure parameters include grain size, inclusion size, annealing time, and material form.
Preferably, the environmental impact parameters include experimental stress and experimental temperature.
Preferably, in step 2, the SHAP value φ of each input feature i The calculation formula of (2) is as follows:
wherein M represents the number of input features,a set representing all permutation components of M features, < ->Representing the set of all feature components preceding feature k in rank R, f x (S)＝E[f(X)|X s ＝x S ]Representing the conditional expectation in case of a known feature subset S.
Preferably, in step 3.1 and step 3.2, the relative root mean square error RRMSE Train The method comprises the following steps:
where n is the number of data samples in the training set, y i Is the true value of the logarithmic creep rupture life in the training set, f (x i ) Is a predicted value of creep rupture logarithmic life.
Preferably, in step 4, a machine learning model of predicted creep life is constructed by an extreme gradient reinforcement algorithm.
Preferably, in step 4, parameters are optimized by a ten-fold cross-validation and grid search method.
Preferably, in step 4, based on the test set, a predicted creep life is obtained from the final predicted creep life fracture model, and the relative root mean square error RRMSE of the test set is calculated from the creep life Test Relative root mean square error RRMSE Test Smaller means higher prediction accuracy.
Preferably, the relative root mean square errorDifference RRMSE Test The calculation formula of (2) is as follows:
where n' is the number of data samples in the test set, y j Is the true value of the logarithmic lifetime of creep rupture in the test set, f (x j ) Is a predicted value of creep rupture logarithmic life.
Compared with the prior art, the invention has the following beneficial effects:
the invention discloses an interpretable creep rupture life prediction method based on machine learning and SHAP values, which is characterized in that a creep rupture life data set of a material is widely collected through experiments and literature data, the influence of different input characteristics such as material chemical components, material microstructure, preparation processing parameters, environmental factors and the like on a creep rupture life prediction model is considered, and the marginal contribution of each input characteristic to a model result is calculated through the SHAP method, so that the influence degree of different characteristics on the creep life can be interpreted. Meanwhile, in the feature screening step based on the SHAP value and the forward search method, the screened optimal input feature set is input into the creep life prediction model based on machine learning, so that the redundancy degree of the model is reduced, the life prediction time and economic cost are reduced, the calculation efficiency of the model is improved, the prediction precision of the model is enhanced, and the contribution of each input variable to the creep rupture life can be explained.
Drawings
FIG. 1 is a flow chart of an interpretable creep rupture life prediction method based on machine learning and SHAP values in accordance with the present invention;
FIG. 2 is a ranking chart of SHAP values of all input feature variables of a training set sample in the present invention;
FIG. 3 is a graph of the best feature screening results according to SHAP values and forward search methods of the present invention;
FIG. 4 is a graph of predicted and actual values of creep log life obtained based on a machine-learned creep rupture life prediction model using test set data, and relative root mean square error calculation results in accordance with the present invention;
FIG. 5 is a graph comparing predicted and actual values of creep life obtained based on a machine-learned creep rupture life prediction model at a 2-fold error dispersion band using test set data in the present invention.
Detailed Description
The invention is described in further detail below with reference to the drawings and to specific embodiments.
Firstly, creep life data sets are obtained through experimental and literature data and are randomly divided into training sets and test sets, the input characteristics of each data set comprise material chemical components, microstructure parameters, processing parameters and environmental parameters, and the output variables are creep logarithmic life; training a creep rupture life prediction model by using a machine learning method by adopting training set data, calculating SHAP values of all sample input features, and analyzing the importance of each feature and the influence of each feature on a prediction result; then, according to SHAP value result, adopting forward search method to screen the optimal characteristic set from the input characteristic set; thirdly, constructing a machine learning model for predicting creep life by using the screened optimal characteristics, and optimizing model parameters and structures; and finally, predicting the creep rupture life by using the optimized model according to the data of the optimal input characteristics in the test set, and evaluating the model precision. An advantage of the present invention is that marginal contributions of individual input features to the model result can be interpreted by the SHAP method. Meanwhile, the optimal input characteristic screening step based on the SHAP value and the forward search method can reduce the redundancy degree of the model, improve the calculation efficiency and improve the prediction accuracy of the model.
Referring to FIG. 1, the invention provides an interpretable creep rupture life prediction method based on machine learning and SHAP values, comprising the steps of:
s1: data set construction: obtaining creep life data sets, and randomly dividing the creep life data sets into 80% training sets and 20% testing sets, wherein each data set input characteristic set comprises characteristics such as material chemical components, microstructure parameters, manufacturing process parameters, environmental influence parameters and the like, and an output variable is creep logarithmic life; normalizing the input data and the output variable;
the creep life data set constructed in the step S1 shares a number of input features and 1 output variable (logarithmic creep rupture life); input features include mass fractions of material chemistry, microstructure parameters, manufacturing process parameters, and environmental impact parameters; the chemical composition contains C, mn, P, S, ni, si, cr, mo, N, al, V, nb and other chemical elements, microstructure factors comprise grain size, inclusion size, lattice constant, stacking fault energy and the like, manufacturing process parameters comprise normalizing temperature, normalizing time, annealing temperature, annealing time, material form and the like, and environmental influence parameters comprise experimental stress and experimental temperature.
The material forms in the step S1 include Tube, plate, pipe and the like, and are replaced by numbers 1, 2, 3 and the like in the data set.
S2: SHAP value calculation: training a creep rupture life prediction model by using the normalized training set data and using a machine learning method, calculating SHAP values of all sample input features, and analyzing the importance of each feature and the influence of each feature on a prediction result; the method comprises the following specific steps of:
s21: and (3) taking all sample input characteristics of a training set as input data, training a creep rupture life prediction model by using an XGBoost algorithm (eXtreme Gradient Boosting, extreme gradient enhancement algorithm) in machine learning, and enabling the prediction model trained by the algorithm to have higher prediction precision and more accurate prediction result.
S22: and according to the model training result of the step S21, calculating SHAP values of all sample input features of the training set, sequencing the input features according to the sequence from high to low of the SHAP values, and analyzing the importance of each input feature and the influence of each input feature on the model result.
Wherein, for each sample, the SHAP value φ of each input feature k is calculated i ：
Wherein M represents the number of input features,a set representing all permutation components of M features, < ->Representing the set of all feature components preceding feature k in rank R, f x (S)＝E[f(X)|X s ＝x S ]Representing the conditional expectation in case of a known feature subset S.
According to the SHAP value of each input feature obtained through calculation, the contribution degree of each feature to the model prediction result can be analyzed, and whether the influence of each feature on the creep life accords with the physical rule is explained.
S3: feature screening: according to SHAP value result, adopting forward search method to screen the optimal characteristic set from the input characteristic set; the method specifically comprises the following steps:
s31: according to the SHAP value result in step S22, firstly, selecting the feature with the largest SHAP value as the input feature set, training the creep rupture life prediction model by using XGBoost algorithm, and calculating the relative root mean square error RRMSE of the model to evaluate the model performance.
S32: selecting one from the rest unselected input features to be added into the input feature set in each increment by adopting a forward search method according to the sequence of SHAP values from high to low in the step S22, training a model by adopting an XGBoost algorithm and the same model parameters, and calculating a relative root mean square error RRMSE for training Train The process is repeated until all features have been selected.
Relative root mean square error RRMSE for training Train The method comprises the following steps:
where n is the data samples in the training setQuantity, y i Is the true value of the logarithmic creep rupture life in the training set, f (x i ) Is a predicted value of creep rupture logarithmic life.
S33: according to the result of step S32, a relative root mean square error RRMSE is selected Train And the input feature set corresponding to the minimum time is the optimal feature set.
S4: model construction: constructing a machine learning model for predicting creep life by using the screened optimal characteristics, and optimizing model parameters and structures;
constructing a machine learning model for predicting creep life by using the screened optimal feature set and training set data through an XGBoost algorithm; the optimal super parameters of the prediction model are sought by adopting a ten-fold cross validation and grid search method so as to realize the optimization of the model structure.
S5: model evaluation: and evaluating the precision of the finally established creep rupture life prediction model according to the data of the optimal input characteristics in the test set.
And obtaining a predicted value of the creep rupture logarithmic life by using the optimized creep rupture life prediction model and adopting data of optimal input characteristics in a test set, and comparing the predicted value with a corresponding true value. Calculation of the relative root mean square error RRMSE of the creep rupture life prediction model for testing in a test set Test And simultaneously drawing a 2-time life dispersion band to test the accuracy of the model prediction result.
Relative root mean square error RRMSE for testing Test The method comprises the following steps:
where n' is the number of data samples in the test set, y j Is the true value of the logarithmic lifetime of creep rupture in the test set, f (x j ) Is a predicted value of creep rupture logarithmic life.
The following is a further explanation in connection with specific examples:
example 1
Referring to FIG. 1, the invention provides an interpretable creep rupture life prediction method based on machine learning and SHAP values, which specifically comprises the following steps:
step S1: obtaining creep life data sets through experimental and literature data, wherein each data set input characteristic comprises material chemical components, processing parameters and environmental parameters, and the output variable is creep logarithmic life; normalizing the input data and the output variable; randomly dividing the data set into an 80% training set and a 20% testing set;
the material selected in this embodiment is 9Cr1MoVNb steel, and the constructed data set includes 503 sets of training set sample data and 125 sets of test set data.
The creep life data set constructed in the step S1 has 18 input characteristics, including 11 material chemical composition mass fractions, 5 manufacturing process parameters and 2 environmental impact parameters; the chemical composition comprises 11 elements such as C, mn, P, S, ni, cr, mo, N, al, V, nb, the manufacturing process parameters comprise 5 parameters such as normalizing temperature, normalizing time, annealing temperature, annealing time and material form, and the environmental influence parameters comprise experimental stress and experimental temperature; the material forms include Tube, plate and Pipe 3 forms, and 0, 1 and 2 are replaced in the original data set respectively.
Step S2: training a creep rupture life prediction model by using the normalized training set data and using a machine learning method, calculating SHAP values of all sample input features, and analyzing the importance of each feature and the influence of each feature on a prediction result;
the step S2 specifically includes:
step S21: and (3) taking all sample input characteristics of the training set as input data, and training a creep rupture life prediction model by using an XGBoost algorithm in machine learning.
Step S22: and according to the model training result of the step S21, calculating SHAP values of all sample input features of the training set, sequencing the input features according to the sequence from high to low of the SHAP values, and analyzing the importance of each feature and the influence of each feature on the model result.
Wherein the method comprises the steps ofFor each sample, the SHAP value φ for each input feature k is calculated i ：
Wherein M represents the number of input features,a set representing all permutation components of M features, < ->Representing the set of all feature components preceding feature k in rank R, f x (S)＝E[f(X)|X s ＝x S ]Representing the conditional expectation in case of a known feature subset S. According to the SHAP value of each input feature obtained through calculation, the contribution degree of each feature to the model prediction result can be analyzed, and whether the influence of each feature on the creep life accords with the physical rule is explained.
The SHAP value results for all sample input features are given in fig. 2. The larger the SHAP value of the input feature, the greater the degree of contribution of the feature to the creep life prediction model, and the greater the impact on the accuracy of the life prediction. FIG. 2 shows that the influence of 2 environmental parameters (i.e., experimental stress and experimental temperature) on the prediction model is ranked in the first two places, which are far higher than the chemical composition and heat treatment parameters of the material, while the V and Cr element contents are ranked in the 3 rd and 4 th places, respectively. In general, the greater the experimental stress, the higher the service temperature, and the shorter the creep rupture life of the material; the addition of the trace elements of V and Cr has important influence on enhancing the high-temperature mechanical property of the 9Cr1Mo steel, and particularly, the V can effectively improve the high-temperature durable strength and creep resistance of the steel. Therefore, in this embodiment, the calculation result of the SHAP value matches the physical rule, and the contribution degree or importance of each feature to the creep life prediction model is displayed.
Step S3: and (3) according to the SHAP value result, adopting a forward search method to screen the optimal feature set from the input feature set.
The step S3 specifically includes:
step S31: according to the SHAP value result in step S22, firstly, selecting the feature with the largest SHAP value as the input feature set, training the creep rupture life prediction model by using XGBoost algorithm, and calculating the relative root mean square error RRMSE of the model to evaluate the model performance.
Step S32: selecting one input feature set from the rest unselected input features by forward search method according to the sequence of SHAP values from high to low in step S22, incrementally adding one input feature set each time, training the model by XGBoost algorithm and the same model parameters, and calculating relative root mean square error RRMSE Train The process is repeated until all features have been selected.
Step S33: according to the result of step S32, a relative root mean square error RRMSE is selected Train And the input feature set corresponding to the minimum time is the optimal feature set.
The RRMSE Train The method comprises the following steps:
where n is the number of data samples in the training set, y i Is the true value of the logarithmic creep rupture life in the training set, f (x i ) Is a predicted value of creep rupture logarithmic life.
Fig. 3 shows the feature screening result obtained by using XGBoost algorithm and forward search method according to the SHAP value result in this embodiment. When the number of input features is 9, the model relative root mean square error RRMSE Train Minimum, 0.0902%, indicates that the model accuracy is highest at this time. Therefore, 9 features such as experimental stress, experimental temperature, V, cr, normalizing temperature, N, mn, S, and Nb were selected as the optimal input feature set.
Step S4: model construction: constructing a machine learning model for predicting creep life by using the screened optimal characteristics, and optimizing model parameters and structures;
the step S4 specifically includes: utilizing the screened optimal characteristic set (namely experimental stress, experimental temperature, V, cr, normalizing temperature, N, mn, S and Nb), adopting training set data, and constructing a machine learning model for predicting creep life through an XGBoost algorithm; searching optimal super parameters of a prediction model by adopting a ten-fold cross validation and grid search method so as to realize optimization of a model structure; the results of the optimized model kernel function and the hyper-parameters are shown in table 1.
Table 1: hyper-parametric optimization result of XGBoost model
Step S5: and evaluating the precision of the finally established creep rupture life prediction model according to the data of the optimal input characteristics in the test set.
The step S5 specifically includes: obtaining a creep fracture logarithmic life prediction value by using an optimized creep fracture life prediction model and adopting data (namely experimental stress, experimental temperature, V, cr, normalizing temperature, N, mn, S and Nb) of optimal input characteristics in a test set, and comparing the creep fracture logarithmic life prediction value with a corresponding true value; calculating the relative root mean square error RRMSE of the creep rupture life prediction model in the test set Test And simultaneously calculating a 2-time life dispersion band to test the accuracy of the model prediction result.
Said relative root mean square error RRMSE Test The method comprises the following steps:
where n' is the number of data samples in the test set, y j Is the true value of the logarithmic lifetime of creep rupture in the test set, f (x j ) Is a predicted value of creep rupture logarithmic life. Relative root mean square error RRMSE Test Smaller means higher prediction accuracy.
With this embodiment, the following predicted results can be obtained:
FIG. 4 shows a creep rupture life prediction model based on machine learning using test set dataAnd comparing the obtained predicted value and the actual value of the creep logarithmic life with the calculated result of the relative root mean square error RRMSE. As seen in the figure, the data points are all substantially close to the 45 dashed line, indicating that the predicted and actual values of creep life are very close. Relative root mean square error RRMSE of prediction result Test Only 0.0457, which shows that the constructed creep life prediction model has higher prediction precision.
FIG. 5 is a graph showing a comparison of predicted and actual values of creep life at a 2-fold error dispersion band using test set data based on a machine-learned creep rupture life prediction model. As can be seen, the vast majority of data points lie within a 2-fold error dispersion band, which means that the constructed machine-learning-based creep-rupture life prediction model can more accurately predict and evaluate the creep life of the material.
The foregoing description of the preferred embodiments of the invention is not intended to be limiting, but rather is intended to cover all modifications, equivalents, alternatives, and improvements that fall within the spirit and scope of the invention.
Claims (10)
1. A machine learning and SHAP value based interpretable creep rupture life prediction method comprising the steps of:
step 1, dividing a data set into a training set and a testing set, wherein each data set comprises a plurality of input features and an output feature;
step 2, training a creep rupture life prediction model by using a training set as input data through an extreme gradient enhancement algorithm, and calculating a SHAP value of each input feature according to a training result; ordering the input features in order from high to low with SHAP values;
step 3, comprising the following sub-steps:
step 3.1, selecting the input characteristic with the largest SHAP value to form an input characteristic set, training a creep rupture life prediction model, and obtaining a relative root mean square error;
step 3.2, selecting one input feature from the rest input features each time by adopting a forward search method, supplementing the selected input feature into the previous input feature set, training a creep rupture life prediction model by an extreme gradient enhancement algorithm, and calculating a relative root mean square error;
step 3.3, repeating the step 3.2 until all input features are selected and stopped after calculation;
step 3.4, selecting the input feature set corresponding to the minimum relative root mean square error as the optimal feature set;
and 4, constructing a machine learning model for predicting the predicted creep life by adopting training set data through an optimal feature set, and obtaining a final predicted creep life fracture model after optimizing parameters.
2. The machine learning and SHAP value-based interpretable creep rupture life prediction method according to claim 1, wherein in step 1, the input features include material chemistry, microstructure parameters, manufacturing process parameters, and environmental impact parameters; the output is characterized by logarithmic creep rupture life.
3. The machine learning and SHAP value-based interpretable creep rupture life prediction method according to claim 2, wherein the microstructure parameters include grain size, inclusion size, annealing time and material form.
4. The machine learning and SHAP value-based interpretable creep rupture life prediction method according to claim 2, wherein the environmental impact parameters include experimental stress and experimental temperature.
5. The machine learning and SHAP value based interpretable creep rupture life prediction method according to claim 1, wherein in step 2, the SHAP value Φ of each input feature i The calculation formula of (2) is as follows:
wherein M represents the number of input features,representing a set of all permutations of M features, P k R Representing the set of all feature components preceding feature k in rank R, f x (S)＝E[f(X)|X s ＝x S ]Representing the conditional expectation in case of a known feature subset S.
6. The method for machine learning and SHAP value based interpreted creep rupture life prediction as claimed in claim 1, wherein in step 3.1 and step 3.2, the relative root mean square error RRMSE is calculated Train The method comprises the following steps:
where n is the number of data samples in the training set, y i Is the true value of the logarithmic creep rupture life in the training set, f (x i ) Is a predicted value of creep rupture logarithmic life.
7. The method for predicting the life of an interpretable creep rupture based on machine learning and SHAP according to claim 1, wherein in step 4, a machine learning model for predicting the life of the creep is constructed by an extreme gradient reinforcement algorithm.
8. The machine learning and SHAP value-based interpretable creep rupture life prediction method according to claim 1, wherein in step 4, parameters are optimized by a ten-fold cross-validation and grid search method.
9. The machine learning and SHAP value based interpretable creep rupture life prediction method according to any one of claims 1 to 8, wherein in step 4, based on the test set, the final prediction is performedThe creep life rupture model obtains a predicted creep life by calculating the relative root mean square error RRMSE of the test set Test Relative root mean square error RRMSE Test Smaller means higher prediction accuracy.
10. The method for machine learning and SHAP value based interpretable creep rupture life prediction as claimed in claim 9, wherein said relative root mean square error RRMSE Test The calculation formula of (2) is as follows:
where n' is the number of data samples in the test set, y j Is the true value of the logarithmic lifetime of creep rupture in the test set, f (x j ) Is a predicted value of creep rupture logarithmic life.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310767414.6A CN116842834A (en) | 2023-06-27 | 2023-06-27 | Interpretable creep rupture life prediction method based on machine learning and SHAP value |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310767414.6A CN116842834A (en) | 2023-06-27 | 2023-06-27 | Interpretable creep rupture life prediction method based on machine learning and SHAP value |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116842834A true CN116842834A (en) | 2023-10-03 |
Family
ID=88171924
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310767414.6A Pending CN116842834A (en) | 2023-06-27 | 2023-06-27 | Interpretable creep rupture life prediction method based on machine learning and SHAP value |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116842834A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117390390A (en) * | 2023-10-19 | 2024-01-12 | 上海交通大学 | Nuclear power plant state prediction method based on interpretable machine learning |
-
2023
- 2023-06-27 CN CN202310767414.6A patent/CN116842834A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117390390A (en) * | 2023-10-19 | 2024-01-12 | 上海交通大学 | Nuclear power plant state prediction method based on interpretable machine learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113033105B (en) | Method for predicting fatigue performance of steel material based on mechanical theory guiding transfer learning | |
| CN116842834A (en) | Interpretable creep rupture life prediction method based on machine learning and SHAP value | |
| CN114021481B (en) | Creep fatigue life prediction method based on fusion physical neural network | |
| CN111310348A (en) | Material constitutive model prediction method based on PSO-LSSVM | |
| CN112730061B (en) | Multi-stage variable temperature and variable load creep life evaluation method | |
| Batsoulas | Cumulative fatigue damage: CDM‐based engineering rule and life prediction aspect | |
| Suzuki et al. | Application of computational tools in alloy design | |
| CN107657121B (en) | Aircraft structure performance prediction processing method and system based on corrosion level evaluation | |
| CN113252465B (en) | M-H method-based creep life prediction method for heat-resistant steel | |
| Fan et al. | Refined model for the stress-strain curve of austenitic stainless-steel materials at elevated temperatures | |
| CN114781250A (en) | Multi-factor influence environment fatigue life prediction method based on machine learning | |
| CN117037966A (en) | Method and system for intelligently researching and developing special steel | |
| CN109580918B (en) | Method for predicting molecular composition of naphtha | |
| CN116539459A (en) | Fatigue crack growth rate prediction method and system based on acoustic emission monitoring and machine learning | |
| Chen et al. | A cooperative PSO-DP approach for the maintenance planning and RBDO of deteriorating structures | |
| Messner et al. | Development of the Technical Basis of a Unified Viscoplastic Model of 316H Stainless Steel for Incorporation into ASME Division 5 | |
| CN115169241A (en) | TiAl alloy fatigue life prediction method driven by data-model | |
| CN108520167B (en) | Method and system for rapidly evaluating high-temperature life of G102 steel heating surface | |
| Gocheva-Ilieva | Using MARS for modeling and predicting tensile strength of low alloy steels | |
| Singh et al. | Design of multiple deferred state repetitive group sampling plan for inverse Weibull distribution based on life test | |
| CN112326540A (en) | Nuclear zirconium-4 alloy corrosion resistance prediction method based on BP network model, electronic equipment and storage medium | |
| KR20230056351A (en) | System and Method for estimating physical properties of power generating equipment part materials based on deep learning | |
| CN116682505B (en) | HRB400E steel mechanical property prediction method based on quantile regression forest | |
| da Silva et al. | A reliability model for non-isothermal isotropic damages | |
| CN114088517B (en) | Evaluation method for acceleration condition of material creep life test |
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 |