Multi-element alloy corrosion performance extrapolation prediction method based on GA-QRF

Abstract

The invention provides a multi-element alloy corrosion performance extrapolation prediction method based on GA-QRF, and relates to the field of alloy corrosion performance prediction, and the method comprises the following steps: obtaining corrosion data of similar multi-element alloys, and carrying out structured arrangement; establishing a QRF model based on the sorted similar multi-element alloy corrosion data and a quantile regression forest; optimizing the QRF model based on a genetic algorithm to obtain a GA-QRF model; carrying out structural arrangement on to-be-evaluated alloy components and service environment data; and based on the GA-QRF model and the sorted to-be-evaluated alloy components and service environment data, carrying out corrosion performance extrapolation prediction, and predicting the interval range of the corrosion characterization quantity of the to-be-evaluated alloy. According to the multi-element alloy corrosion performance extrapolation prediction method based on GA-QRF, rapid corrosion performance evaluation of newly researched and developed metal or metal not subjected to tests is achieved, and information can be provided for alloy material service life analysis and maintenance decision.

Description

Technical Field

[0001] The invention relates to the field of alloy corrosion performance prediction, and in particular to a multi-element alloy corrosion performance extrapolation prediction method based on GA-QRF. Background Art

[0002] "Multi-element alloy" refers to a metal material composed of two or more different elements; compared with a single metal, a multi-element alloy has a more complex chemical composition and crystal structure; these different metal elements dissolve in each other in the alloy and form new crystal structures and properties.

[0003] "Similar alloys" refer to a class of alloys with similar composition. The main difference between similar alloys lies in the difference in the content of various components. By adjusting the content of different components, the performance and characteristics of the alloy can be changed. This adjustment can affect the hardness, strength, corrosion resistance, conductivity and other aspects of the alloy.

[0004] Multi-element alloy materials have broad application prospects due to their excellent performance characteristics. With the continuous expansion and refinement of engineering needs, the research and development of new metal alloys has also shown a booming trend; under this situation, evaluating the corrosion performance of metal materials has become a very challenging research difficulty, and a faster and more efficient evaluation method is needed to meet the rapidly developing engineering field's requirements for multi-element alloy material performance analysis, and provide support for its service life assessment and maintenance decision-making.

[0005] Traditional methods for evaluating and predicting metal corrosion performance are mainly carried out through two different approaches: one is field testing, and the other is laboratory accelerated testing; field testing collects information on the service environment and service time, which is used to model the corrosion degradation of the test metal materials, and then evaluate and predict their corrosion performance; while accelerated testing is based on controllable test conditions, by accelerating the metal corrosion process, quickly obtaining metal corrosion data, and performing corrosion modeling; the accuracy of accelerated testing needs to be combined with field data for verification; the above methods have been successfully applied to developed metals, and a wealth of multi-element alloy corrosion performance data has been accumulated; however, with the rapid emergence of newly developed multi-element alloy materials, these methods require a large investment in both time and test costs, making it difficult to achieve rapid evaluation and prediction of multi-element alloy corrosion performance. Summary of the invention

[0006] The purpose of the present invention is to provide a GA-QRF-based multi-element alloy corrosion performance extrapolation prediction method, which is suitable for rapid evaluation, analysis and prediction of the corrosion performance of newly developed / untested similar alloys.

[0007] To achieve the above object, the present invention provides a multi-element alloy corrosion performance extrapolation prediction method based on GA-QRF, comprising the following steps:

[0008] Obtain corrosion data of similar multi-element alloys and organize them in a structured manner;

[0009] The QRF model was established based on the collated corrosion data of similar multi-element alloys and quantile regression forest;

[0010] The GA-QRF model is obtained by optimizing the QRF model based on genetic algorithm, and the GA-QRF model is trained and predicted using the corrosion data of similar multi-element alloys.

[0011] Structural organization of data on alloy composition and service environment to be evaluated;

[0012] Based on the GA-QRF model and the collated data on the composition and service environment of the alloy to be evaluated, the corrosion performance is extrapolated and predicted, and the corrosion characterization value prediction range of the alloy to be evaluated is determined.

[0013] Preferably, the corrosion data of the same type of multi-element alloys includes composition information, service environment information, service time information and corresponding corrosion parameter information of the same type of alloys, and the corrosion data of the same type of multi-element alloys is structured and stored:

[0014]

[0015] Among them, i is the structured data index; is the composition data of the multi-element alloy material under the corresponding index i, which is a 1×m1-dimensional vector, where m1 is the number of elements involved in the same type of alloy; is the service environment data corresponding to index i, which is a 1×m2-dimensional vector, where m2 is the number of environmental factors sensitive to alloy corrosion; t i is the corrosion time corresponding to index i; C i is the corrosion performance characterization quantity corresponding to index i.

[0016] Preferably, a QRF model is established based on quantile regression forest. With t as input and C as the optimization target, the model expression is as follows:

[0017]

[0018] in, is the composition data of multi-element alloy materials, is the service environment data, t is the corrosion time, f represents the quantile regression forest model, is the output of the quantile regression forest and has the following form:

[0019]

[0020] Among them, X represents the The combination of t, x is the specific input value, is the cumulative probability density function of output y when the input is x, w i (x) means the predicted result y is greater than the true value Y of the i-th sample in the training set i probability;

[0021] The output of the quantile regression forest is the cumulative probability density function of the prediction results, and the corresponding prediction means and prediction intervals are expressed as follows:

[0022]

[0023] I(x)=[Q α / 2 (x), Q 1-α / 2 (x)]

[0024]

[0025] Among them, μ(x) and I(x) correspond to the mean and interval of the predicted output when the input is x, 1-α represents the confidence level, n represents the number of samples in the training set, and Q α (x) is the one-sided α lower quantile of the predicted output, Q α / 2 (x) represents and Q 1-α / 2 (x) represents the bilateral α upper and lower quantiles of the predicted output, and inf{} represents the lower bound of the value.

[0026] Preferably, the QRF model includes the number of regression decision trees n tree 、Number of random sampling features of nodes m try and the maximum tree depth d tree Three hyperparameters.

[0027] Preferably, the GA-QRF model is obtained by optimizing the QRF model based on a genetic algorithm, and the GA-QRF model is trained and predicted using the corrosion data of the same multi-element alloy, including

[0028] Determine the range of the three hyperparameters as n tree ∈[n min ,n max ]、m try ∈[m min ,m max ] and d tree ∈[d min ,d max ], where n min 、n max 、m min 、m max ,d min ,dmax They correspond to the maximum and minimum values ​​of the three hyperparameters respectively, and then randomly select values ​​in each hyperparameter range to obtain the initial hyperparameter population P0, which is recorded as:

[0029]

[0030] in, They represent the jth individual in the initial population, and s is the number of individuals in each generation;

[0031] After the initial hyperparameter population generation is completed, the individuals in the population are encoded in the form of chromosomes and the binary representation is used to represent the Encode and string them together;

[0032] Afterwards, for each hyperparameter combination represented by an individual in the population, a QRF model is constructed, and K-fold cross-validation training is performed on the training set data to predict and evaluate the performance of the hyperparameter model. The performance evaluation index is the goodness of fit R 2 , which is defined as follows:

[0033]

[0034] Where K is the number of cross-validation folds, r k represents the number of samples contained in the k-th fold, y ki represents the true value of the i-th sample in the k-th fold, represents the predicted value of the i-th sample of the k-th fold, is the average value of the true values ​​of all samples in the kth fold;

[0035] According to the performance evaluation results of all individuals in the initial population, the individuals in the initial population P0 are subjected to selection, crossover and mutation operations. After completing the selection, crossover and mutation operations, the new generation population P is obtained. l , l represents the lth generation population;

[0036] For the generated new population P l , repeat the cross-validation evaluation and selection, crossover, and mutation work until the final population P l The performance evaluation results of the best samples in R 2 Reach the specified target value or P l With P l+k The best sample performance evaluation result R 2 When the difference is less than the specified value ε, the genetic algorithm optimization ends and the optimal hyperparameter combination (n best ,m best ,d best ), and the corresponding QRF model optimized by genetic algorithm is recorded as GA-QRF model.

[0037] Preferably, the alloy composition and service environment data to be evaluated are structured and collated, and the structuring specification is the same as the collation specification of the corrosion data of the same multi-element alloy, and the alloy composition and service environment data to be evaluated are structured and collated.

[0038]

[0039] in, The composition data of the alloy to be evaluated is consistent with the recording format of the composition of similar multi-element alloys. For alloy elements not included, the data is filled with 0; is the actual service environment of the alloy to be evaluated; t e The service life of the metal to be evaluated.

[0040] Therefore, the present invention adopts the above-mentioned method for predicting the corrosion performance of multi-element alloys based on GA-QRF extrapolation, and the technical effects are as follows:

[0041] (1) The present invention builds a model based on historical test data of similar alloys, and fully explores the relationship between alloy composition information, environmental service information, corrosion time information and alloy corrosion performance.

[0042] (2) The performance extrapolation prediction method proposed in the present invention can realize rapid corrosion performance analysis, evaluation and prediction of newly developed / untested alloys.

[0043] (3) Compared with the method of modeling alloy corrosion performance based on experiments, the correction method of the present invention shortens the performance evaluation modeling time, solves the hysteresis problem of the experimental method, and has broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 is a flow chart of the method of the present invention;

[0045] Figure 2 This is a distribution diagram of corrosion performance data of the same alloy in Example 1; Figure 2 (a) is the distribution diagram of corrosion performance data of similar alloys in Beijing; Figure 2 (b) is the distribution diagram of corrosion performance data of similar alloys in Qingdao; Figure 2 (c) is the distribution diagram of corrosion performance data of similar alloys in Wuhan; Figure 2 (d) is the distribution diagram of corrosion performance data of similar alloys in Jiangjin;

[0046] Figure 2 (e) is the distribution diagram of corrosion performance data of similar alloys in Guangzhou; Figure 2 (f) is the distribution diagram of corrosion performance data of similar alloys in Qionghai;

[0047] Figure 3The prediction results of the quantile regression forest model optimized by genetic algorithm; Figure 3 (a) is the effect diagram of the GA-QRF model training set of the same alloy in Example 1; Figure 3 (b) is the effect diagram of the GA-QRF model test set of the same alloy in Example 1.

[0048] Figure 4 The corrosion performance prediction diagram of the two alloys to be evaluated; Figure 4 (a) is the corrosion performance prediction diagram of 945 steel; Figure 4 (b) is the corrosion performance prediction diagram of 30CrMnSiA steel. DETAILED DESCRIPTION

[0049] The technical solution of the present invention is further described below through the accompanying drawings and embodiments.

[0050] Unless otherwise defined, technical or scientific terms used in the present invention shall have the common meanings understood by one having ordinary skills in the field to which the present invention belongs.

[0051] Embodiment 1

[0052] like Figure 1 As shown, the present invention provides a multi-element alloy corrosion performance extrapolation prediction method based on GA-QRF, comprising the following steps:

[0053] Step 1: Obtain corrosion data of similar multi-element alloys and organize them in a structured manner;

[0054] The data of similar alloy metals used in the cases of this method were obtained from the National Materials Corrosion and Protection Science Data Center; the alloy category selected in the case was low-carbon steel, and a total of 25 low-carbon steel alloys were collected under 6 service environments. The corrosion performance was characterized by the average annual corrosion rate.

[0055] The main elements of low carbon steel alloys are Fe, C, Mn, Si, P, S, Cr, Ni, Cu and Mo. The composition information of the 25 alloys selected in the case is shown in Table 1:

[0056] Table 1 Composition information of similar alloys

[0057]

[0058]

[0059] The corrosion of low carbon steel is mainly affected by factors such as temperature, humidity, chloride ion deposition rate and sulfur dioxide deposition rate. In this case, the environmental information statistics of the six historical field test sites are shown in Table 2:

[0060] Table 2 Test site environmental data

[0061]

[0062] The corrosion performance of the above materials is characterized by the annual average corrosion rate, which is expressed in μm / a. The corrosion rates of similar metals collected are as follows: Figure 2 As shown; the above data are combined to complete the structural arrangement of the corrosion data of the same type of multi-element alloy in step 1;

[0063] Step 2: Establish a QRF model based on the collated corrosion data of similar multi-element alloys and quantile regression forest;

[0064] After completing the data structured arrangement in step 1, the quantile regression forest is used to establish the basic model (QRF) of multi-element alloy composition information, service environment information, corrosion time information and corrosion performance characterization quantity, and the quantile regression forest is used to complete the construction of this nonlinear model; in the quantile regression forest, the number of regression decision trees n of the basic model hyperparameters is set tree =100, number of random sampling features of nodes m try = 10 and the maximum tree depth d tree =10; Use the data set compiled in step 1 and divide it into a training set and a test set in a ratio of 80% and 20% respectively, and use the training set to train the basic model.

[0065] Step 3, optimizing the QRF model based on genetic algorithm to obtain the GA-QRF model;

[0066] After completing step 2 to establish the basic QRF model, the genetic algorithm is used to tune the above three hyperparameters. The range of the three hyperparameters is limited to n tree ∈[100,500], m try ∈[1,12], d tree ∈[3,15], the optimization stopping criterion of the genetic algorithm is set as That is, the performance evaluation difference of the 10 generations of population is less than 10 -3 The final hyperparameter tuning result of the genetic algorithm is the number of regression decision trees n tree =386, number of random sampling features of nodes m try =8 and the maximum depth of the tree is d tree =9; Use the tuned GA-QRF model to predict and verify the training set and test set, and get the prediction results of the quantile regression forest model optimized by the genetic algorithm as follows Figure 3 (a) and Figure 3 (b) shows the goodness of fit of the training set R 2 =1.000, the confidence interval coverage rate obtained with 95% confidence level reaches 100%; the goodness of fit of the test set R 2=0.903, and the coverage rate of 95% confidence interval reaches 89.3%; the proposed method shows good prediction performance on the historical data of multi-element alloys.

[0067] Step 4: Structural arrangement of the alloy composition and service environment data to be evaluated;

[0068] In this embodiment, two alloys that have not been tested in the field are selected as materials to be evaluated, and their corrosion properties are extrapolated and predicted; the selected alloys are 945 steel and 30CrMnSiA steel, and their alloy element composition contents are shown in Table 3; the service environment of the metal to be evaluated is set to be located in the Yellow Sea, and its temperature is set to 12°C, the relative humidity is set to 72%, and the chloride ion deposition rate is set to 26 mg / m 2 / d, sulfur dioxide deposition rate is 68mg / m 2 / d;

[0069] Table 3 Composition information of metal alloys to be evaluated

[0070] Alloy Name C Mn Si P S Cr Ni Cu Mo Fe 945 0.13 1.20 0.44 0.02 0.01 0.60 0.35 1.15 0.20 Balance 30CrMnSiA 0.31 0.95 1.05 0.025 0.025 0.95 0.25 0.3 0 Balance

[0071] Step 5: Based on the GA-QRF model and the sorted data on the composition and service environment of the alloy to be evaluated, the corrosion performance is extrapolated and predicted, and the corrosion characterization value of the alloy to be evaluated is predicted.

[0072] For the two alloys selected in step 4, the GA-QRF model trained in step 3 is used to perform corrosion performance extrapolation prediction; the evaluation time is set to 1 year, 2 years, 4 years, 8 years and 16 years, and the composition information, service environment information and evaluation time information in step 4 are integrated and input into the GA-QRF model determined in step 3 to achieve the extrapolation prediction of its corrosion performance. The results are shown in the figure. Figure 4 (a) and Figure 4 (b)

[0073] Therefore, the present invention adopts the above-mentioned GA-QRF-based multi-element alloy corrosion performance extrapolation prediction method, and solves the lag of alloy corrosion performance evaluation based on experimental methods through in-depth mining of historical similar alloy data, and realizes rapid corrosion performance evaluation of newly developed / untested metals, which helps to provide information for alloy material service life analysis and maintenance decision-making, and can be used to support the research and development of similar alloy materials.

[0074] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A multi-element alloy corrosion performance extrapolation prediction method based on GA-QRF, characterized in that: The following steps are involved: Obtain corrosion data of similar multi-element alloys and organize them in a structured manner; The QRF model was established based on the collated corrosion data of similar multi-element alloys and quantile regression forest; The GA-QRF model is obtained by optimizing the QRF model based on genetic algorithm, and the GA-QRF model is trained and predicted using the corrosion data of similar multi-element alloys. Structural organization of data on alloy composition and service environment to be evaluated; Based on the GA-QRF model and the collated data on the composition and service environment of the alloy to be evaluated, the corrosion performance is extrapolated and predicted, and the corrosion characterization value prediction range of the alloy to be evaluated is determined.

2. The method for predicting corrosion performance of multi-element alloys based on GA-QRF according to claim 1 is characterized in that: The corrosion data of similar multi-element alloys includes the composition information, service environment information, service time information and corresponding corrosion parameter information of similar alloys, and the corrosion data of similar multi-element alloys are structured and saved: Among them, i is the structured data index; is the composition data of the multi-element alloy material under the corresponding index i, which is a 1×m1-dimensional vector, where m1 is the number of elements involved in the same type of alloy; is the service environment data corresponding to index i, which is a 1×m2-dimensional vector, where m2 is the number of environmental factors sensitive to alloy corrosion; t i is the corrosion time corresponding to index i; C i is the corrosion performance characterization quantity corresponding to index i.

3. The method for extrapolating and predicting corrosion performance of multi-element alloys based on GA-QRF according to claim 1 is characterized in that: The QRF model is established based on the quantile regression forest. With t as input and C as the optimization target, the model expression is as follows: in, is the composition data of multi-element alloy materials, is the service environment data, t is the corrosion time, f represents the quantile regression forest model, is the output of the quantile regression forest and has the following form: Among them, X represents the The combination of t, x is the specific input value, is the cumulative probability density function of output y when the input is x, w i (x) means the predicted result y is greater than the true value Y of the i-th sample in the training set i probability; The output of the quantile regression forest is the cumulative probability density function of the prediction results, and the corresponding prediction means and prediction intervals are expressed as follows: I(x)=[Q α / 2 (x),Q 1-α / 2 (x)] Among them, μ(x) and I(x) correspond to the mean and interval of the predicted output when the input is x, 1-α represents the confidence level, n represents the number of samples in the training set, and Q α (x) is the one-sided α lower quantile of the predicted output, Q α / 2 (x) represents and Q 1-α / 2 (x) represents the bilateral α upper and lower quantiles of the predicted output, and inf{} represents the lower bound of the value.

4. The method for extrapolating and predicting corrosion performance of multi-element alloys based on GA-QRF according to claim 1 is characterized in that: The QRF model includes the number of regression decision trees n tree 、Number of random sampling features of nodes m try and the maximum tree depth d tree Three hyperparameters.

5. The method for extrapolating and predicting corrosion performance of multi-element alloys based on GA-QRF according to claim 4 is characterized in that: The GA-QRF model is obtained by optimizing the QRF model based on genetic algorithm, and the GA-QRF model is trained and predicted using the corrosion data of similar multi-element alloys, including Determine the range of the three hyperparameters as n tree ∈[n min ,n max ]、m try ∈[m min ,m max ] and d tree ∈[d min ,d max ], where n min 、n max 、m min 、m max ,d min ,d max They correspond to the maximum and minimum values ​​of the three hyperparameters respectively, and then randomly select values ​​in each hyperparameter range to obtain the initial hyperparameter population P0, which is recorded as: in, They represent the jth individual in the initial population, and s is the number of individuals in each generation; After the initial hyperparameter population generation is completed, the individuals in the population are encoded in the form of chromosomes and the binary representation is used to represent the Encode and string them together; Afterwards, for each hyperparameter combination represented by an individual in the population, a QRF model is constructed, and K-fold cross-validation training is performed on the training set data to predict and evaluate the performance of the hyperparameter model. The performance evaluation index is the goodness of fit R 2 , which is defined as follows: Where K is the number of cross-validation folds, r k represents the number of samples contained in the k-th fold, y ki represents the true value of the i-th sample in the k-th fold, represents the predicted value of the i-th sample of the k-th fold, is the average value of the true values ​​of all samples in the kth fold; According to the performance evaluation results of all individuals in the initial population, the individuals in the initial population P0 are subjected to selection, crossover and mutation operations. After completing the selection, crossover and mutation operations, the new generation population P is obtained. l , l represents the lth generation population; For the generated new population P l , repeat the cross-validation evaluation and selection, crossover, and mutation work until the final population P l The performance evaluation results of the best samples in R 2 Reach the specified target value or P l With P l+k The best sample performance evaluation result R 2 When the difference is less than the specified value ε, the genetic algorithm optimization ends and the optimal hyperparameter combination (n best ,m best ,d best ), and the corresponding QRF model optimized by genetic algorithm is recorded as GA-QRF model.

6. The method for predicting corrosion performance of multi-element alloys based on GA-QRF according to claim 1, characterized in that: The data of alloy composition and service environment to be evaluated are structured and collated. The structured specifications are the same as those for the corrosion data of similar multi-element alloys. in, The composition data of the alloy to be evaluated is consistent with the recording format of the composition of similar multi-element alloys. For alloy elements not included, the data is filled with 0; is the actual service environment of the alloy to be evaluated; t e The service life of the metal to be evaluated.

Images