Accurate control method for content of microelement Mg in high-temperature alloy based on artificial neural network

Abstract

The invention discloses a method for accurately controlling the content of microelement Mg in a high-temperature alloy based on an artificial neural network. The method comprises the following steps: firstly, determining the influence rule of the Mg element addition amount and smelting process parameters on the content of the microelement Mg in the high-temperature alloy through experiment or field production data, and obtaining a machine learning data set; then, statistical analysis is conducted on the multi-factor nonlinear coupling relation between input and output through an artificial neural network machine learning algorithm, an Mg content prediction model based on the artificial neural network is established, and the artificial neural network structure and related algorithms are further optimized; and finally, predicting the Mg content under a certain process condition by utilizing the optimized Mg content prediction model, further adjusting alloy smelting process parameters, adjusting the Mg element in the alloy to the target content, and realizing accurate control of the Mg content of the trace element. According to the method, the trace element Mg in the high-temperature alloy is accurately and stably controlled to reach the target content, so that casting production of the high-temperature alloy is guided.

Description

Technical Field

[0001] The invention belongs to the technical field of high-temperature alloy casting, and in particular relates to a method for accurately controlling the content of trace element Mg in a high-temperature alloy based on an artificial neural network. Background Art

[0002] Trace element Mg is generally considered to be a beneficial element in high-temperature alloys. It has an important influence on the solidification characteristic temperatures of high-temperature alloys, such as the solid-liquid phase temperature, solidification temperature range, and dendrite overlap point temperature, and can directly change the solidification process, solidification structure and mechanical properties of the alloy. Studies have shown that trace element Mg can purify high-temperature alloy melts and reduce the harmful effects of impurities such as O, S, and P in the alloy. At the same time, adding trace element Mg to high-temperature alloys can significantly change the size and distribution of the γˊ phase in the alloy, spheroidize and disperse primary carbides, reduce the grain boundary energy and the phase boundary energy of carbide-matrix, thereby improving the alloy interface bonding force and significantly improving the creep performance and endurance life of the alloy. In addition, the addition of Mg can also control the number, morphology and distribution characteristics of the δ phase in GH4169 alloy, thereby improving the mechanical properties of the alloy. The study also found that there is an optimal content range of trace element Mg in high-temperature alloys. If its content in the alloy is too low, it cannot play a positive role, while if its content is too high, it may have a negative impact on the alloy properties. It can be seen that it is necessary to accurately and stably control the trace element Mg in the high-temperature alloy to the target content.

[0003] However, due to the high saturated vapor pressure of Mg element and its great volatilization tendency, and the many factors that affect the volatilization rate of Mg during vacuum induction melting, it is difficult to achieve accurate control of Mg content. In the early stage, researchers mainly conducted regression analysis on the influence of smelting process parameters on Mg content based on the volatilization dynamics of Mg element to establish empirical formulas and mathematical models for content prediction, so as to guide the control of Mg content in actual production. However, simple regression fitting analysis is difficult to better describe the complex nonlinear coupling relationship between multivariate input and output, which makes the traditional empirical formula and mathematical model for Mg content prediction still insufficient in terms of prediction accuracy. Therefore, it is urgent to establish a precise control method for the trace element Mg content in high-temperature alloys.

[0004] As a deep learning algorithm model, artificial neural network can realize modeling and prediction of complex nonlinear systems through data training. It has good technical advantages in describing the nonlinear coupling relationship of multiple factors and has been successfully applied to the accurate prediction of material structure and properties. Therefore, by using artificial neural network training to fit the complex nonlinear relationship between smelting process parameters and Mg element recovery rate, and establishing a Mg content prediction model, it is expected to achieve accurate control of the trace element Mg content in cast high-temperature alloys. Summary of the invention

[0005] In view of the problems in the prior art, the present invention provides a method for accurately controlling the content of trace element Mg in high-temperature alloys based on artificial neural networks. This method can accurately and stably control the trace element Mg in high-temperature alloys to a target content, thereby guiding the casting production of high-temperature alloys.

[0006] In order to achieve the above objectives, the present invention adopts the following technical solutions:

[0007] A method for accurately controlling the content of trace element Mg in a high-temperature alloy based on an artificial neural network comprises the following steps:

[0008] (1) Determine the influence of Mg addition and smelting process parameters on the content of trace element Mg in high-temperature alloys through experiments or field production data, and obtain a machine learning data set between smelting process parameters and Mg content. The smelting process parameters include argon filling pressure, argon filling time, low-temperature refining temperature, and low-temperature refining time;

[0009] (2) Normalize and standardize the machine learning dataset;

[0010] (3) An artificial neural network model was established. The model included three layers: input layer, hidden layer and output layer. The input layer was used to receive input signals, and its variables were the amount of Mg element added, argon filling pressure, argon filling time, low-temperature refining temperature and low-temperature refining time. The output layer was used to output signals, i.e., calculation results, and its variables were Mg content. The number of neural nodes in the hidden layer was calculated according to the formula Determine, where m represents the number of neurons in the input layer, n represents the number of neurons in the output layer, and a represents a constant between [1, 10];

[0011] (4) dividing the input and output machine learning data sets into a training set, a validation set, and a test set, and using the training set to train the artificial neural network. The training algorithms include the Levenberg-Marquardt method, the Bayesian regularization method, and the quantized conjugate gradient method. When the output error meets the set requirements, the training is stopped and a nonlinear regression model is obtained.

[0012] (5) using the validation set and the test set to perform validation tests on the nonlinear regression model to obtain a validation test error. When the validation test error is within a preset error range, the obtained nonlinear regression model is used as a Mg content prediction model;

[0013] (6) randomly setting a set of smelting process parameters, performing the same normalization process as step (2) on the smelting process parameters, and inputting the smelting process parameters into the Mg content prediction model obtained in step (5) to obtain the corresponding predicted content of the trace element Mg, performing actual smelting according to the randomly set smelting process parameters, and determining the actual content of Mg in the alloy, and comparing the predicted content of Mg with the actual content to determine the validity and accuracy of the Mg content prediction model;

[0014] (7) The target content of trace element Mg in the high-temperature alloy is set, the smelting process parameters, i.e., argon filling pressure, argon filling time, low-temperature refining temperature and low-temperature refining time, are fixed, the amount of Mg added in the alloy is adjusted, and the Mg content in the alloy is adjusted to the target content through the Mg content prediction model. The Mg-containing alloy is smelted according to the adjusted smelting process parameters, thereby achieving precise control of the trace element Mg content.

[0015] Furthermore, the above-mentioned method for accurately controlling the content of trace element Mg in a high-temperature alloy based on an artificial neural network, and its preferred embodiment is that in step (1), the Mg element addition amount ranges from 20 to 2000 ppm, the argon filling pressure ranges from 0 to 60 kPa, the argon filling time ranges from 1 to 15 min, the low-temperature refining temperature ranges from 1350 to 1450°C, and the low-temperature refining time ranges from 1 to 10 min. The method for determining the content of trace element Mg in the high-temperature alloy includes but is not limited to ICP-MS or ICP-AES.

[0016] Furthermore, the above-mentioned method for accurately controlling the content of trace element Mg in high-temperature alloys based on artificial neural networks has a preferred embodiment that, in step (2), the data normalization and standardization processing methods include Min-Max normalization, Z-Score normalization and nonlinear normalization methods.

[0017] Furthermore, the above-mentioned method for accurately controlling the content of trace element Mg in high-temperature alloys based on artificial neural networks has a preferred embodiment in which, in step (2), the data normalization and standardization processing method is Min-Max normalization, and the formula is:

[0018]

[0019] Where X is the original data, Xmax and Xmin are the maximum and minimum values ​​in the data set, respectively, and Xnorm is the normalized data.

[0020] Furthermore, the above-mentioned method for accurately controlling the content of trace element Mg in a high-temperature alloy based on an artificial neural network has a preferred embodiment in which, in step (3), the type of the artificial neural network is a Back-Propagation artificial neural network, the number of neurons in the input layer m is 5, the number of neurons in the output layer n is 1, the number of neurons in the hidden layer ranges from [4, 13], the input layer and the output layer are expressed by purelin functions, and the hidden layer is expressed by sigmoid functions;

[0021] Calculation process description: The number of neural nodes in the hidden layer is calculated according to the formula Determine, where m represents the number of neurons in the input layer, n represents the number of neurons in the output layer, and a represents a constant between [1, 10]. In the present invention, the number of neurons in the input layer m is 5, and the number of neurons in the output layer n is 1, so is approximately equal to 2.4, which is rounded to 3, so The value range is [4, 13].

[0022] Furthermore, in the above-mentioned method for accurately controlling the content of trace element Mg in high-temperature alloys based on artificial neural networks, a preferred embodiment is that the number of neurons in the hidden layer is 10.

[0023] Furthermore, the above-mentioned method for accurately controlling the content of trace element Mg in high-temperature alloys based on artificial neural networks has a preferred embodiment that, in step (4), the proportions of the training set, validation set and test set in the machine learning data set are 60-70%, 15-20% and 15-20% respectively, the training algorithm is the Levenberg-Marquardt method, the learning rate is 0.01, the number of iterations is 1000, and the training accuracy requirement is 0.001.

[0024] Furthermore, the above-mentioned method for accurately controlling the content of trace element Mg in high-temperature alloys based on artificial neural network has a preferred embodiment that, in step (4), the specific training process is:

[0025] a. Forward propagation: The training set data is input into the input layer of the Back-Propagation artificial neural network, and after the nonlinear transformation of the hidden layer, it reaches the output layer and outputs the result. If the output result does not match the expected output, the error is back propagated;

[0026] b. Back propagation: Use the error function to calculate the error between the output result and the expected output, propagate the output error back through the hidden layer to the input layer layer by layer, and distribute the error to all neurons in each layer so that the error decreases along the gradient direction. The error obtained in each layer is used as the basis for adjusting the weights and thresholds.

[0027] c. Repeat training: Repeat steps a and b until the preset convergence condition is reached, determine the weights and thresholds corresponding to the minimum input error, and complete the training.

[0028] Furthermore, the above-mentioned method for accurately controlling the content of trace element Mg in high-temperature alloys based on artificial neural networks has a preferred embodiment in which, in step (5), the preset error of the verification test is 0.001.

[0029] Furthermore, the above-mentioned method for accurately controlling the content of trace element Mg in high-temperature alloys based on artificial neural networks has a preferred embodiment that, in step (6), the randomly set smelting process parameters are within the process parameter range described in step (1), and the alloy is vacuum induction melted according to the set process, and the actual content of Mg element in the alloy is determined by ICP-MS or ICP-AES method. When the error between the predicted content and the actual content of the Mg content prediction model is within ±20ppm, the validity and accuracy of the prediction model is determined.

[0030] The design idea of ​​the present invention is:

[0031] The trace element Mg has an optimal content range in high-temperature alloys. If its content in the alloy is too low or too high, the optimal effect cannot be achieved. Therefore, the precise control of the trace element Mg content in high-temperature alloys is particularly important. However, due to the high saturated vapor pressure of the Mg element and its great tendency to volatilize, and the many factors that affect the volatilization rate of Mg during vacuum induction melting, it is difficult to achieve precise control of the Mg content. The traditional trial and error method to adjust the alloy melting process parameters is bound to cause a large waste of manpower and financial resources, greatly increasing the production and trial and error costs of high-temperature alloys.

[0032] With the development of computer hardware and machine learning algorithm models, the modeling and prediction of complex nonlinear systems can be realized by machine learning methods. Artificial neural network machine learning models are widely used in materials science research due to their simple algorithms and clear structures. At present, they have been successfully applied to the accurate prediction of material composition, structure and performance. Therefore, by using artificial neural network training to fit the complex nonlinear relationship between smelting process parameters and Mg element yield, a Mg content prediction model is established, which is expected to achieve accurate control of the trace element Mg content in cast high-temperature alloys. The method of the present invention can reduce the time and cost of exploring smelting process parameters, reduce the production cost of high-temperature alloy casting, and has important theoretical research significance and application value.

[0033] Advantages and beneficial effects of the present invention:

[0034] 1. The present invention is based on artificial neural network training to fit the complex nonlinear relationship between smelting process parameters and Mg element yield, and realizes precise control of trace element Mg content by establishing a Mg content prediction machine learning model, thereby avoiding the waste of manpower and resources caused by the traditional trial and error method, effectively improving the efficiency of casting high-temperature alloy production and the qualified rate of components, thereby greatly reducing the production cost of high-temperature alloys.

[0035] 2. The method of the present invention is reasonably designed and highly operable, and can provide method guidance for the precise control of the content of other easily burned elements such as Y, Ce, La, etc.

[0036] 3. The method of the present invention is simple and easy to implement, and is suitable for engineering application and promotion. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a flow chart of the scheme of the present invention;

[0038] Figure 2 This is a schematic diagram of the smelting process of high-temperature alloys with Mg added;

[0039] Figure 3 This is a structural diagram of the Back-Propagation artificial neural network model in an embodiment of the present invention;

[0040] Figure 4 It is a training set fitting regression graph in an embodiment of the present invention;

[0041] Figure 5 It is a validation set fitting regression graph in an embodiment of the present invention;

[0042] Figure 6 This is a test set fitting regression graph in an embodiment of the present invention. DETAILED DESCRIPTION

[0043] In the specific implementation process, the method of the present invention first determines the influence of smelting process parameters such as Mg element addition, argon filling pressure, argon filling time, low-temperature refining temperature, and low-temperature refining time on the trace element Mg content in the high-temperature alloy through experiments or on-site production data, and obtains a machine learning data set between input (smelting process parameters) and output (Mg content); then, the multi-factor nonlinear coupling relationship between the input and output is statistically analyzed through an artificial neural network machine learning algorithm, and a Mg content prediction model based on an artificial neural network is established, and the artificial neural network structure and related algorithms such as the number of neural network layers and the number of neurons are further optimized to improve the prediction accuracy of the content prediction model; finally, the optimized Mg content prediction model is used to predict the Mg content under certain process conditions, and then the alloy smelting process parameters are adjusted to adjust the Mg element in the alloy to the target content, thereby achieving precise control of the trace element Mg content.

[0044] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. Figure 1 As shown, the scheme flow of a method for accurately controlling the content of trace element Mg in high-temperature alloys based on artificial neural networks is as follows:

[0045] (1) According to Figure 2 The high-temperature alloy plus Mg smelting process shown in the figure is used to smelt the high-temperature alloy, and the influence of smelting process parameters such as Mg element addition, argon filling pressure, argon filling time, low-temperature refining temperature, and low-temperature refining time on the content of trace element Mg in the high-temperature alloy is determined, wherein the range of Mg element addition is 20-260ppm, the range of argon filling pressure is 0-60kPa, the range of argon filling time is 1-7min, the range of low-temperature refining temperature is 1370-1430℃, and the range of low-temperature refining time is 1-4min. The content of trace element Mg in the high-temperature alloy is determined by ICP-MS method, and a machine learning data set between smelting process parameters and Mg content is obtained;

[0046] (2) Normalize and standardize the machine learning data set using Min-Max normalization, with the formula:

[0047]

[0048] Where X is the original data, Xmax and Xmin are the maximum and minimum values ​​in the data set, respectively, and Xnorm is the normalized data;

[0049] (3) Establish Figure 3 The BP artificial neural network model shown in the figure includes three layers of networks: input layer, hidden layer and output layer. The input layer is used to receive input signals, the number of neurons is 5, and its variables are the amount of Mg element addition, argon filling pressure, argon filling time, low-temperature refining temperature, and low-temperature refining time; the output layer is used to output signals, i.e., calculation results, the number of neurons is 1, and its variable is the actual Mg content in the high-temperature alloy; the number of neurons in the hidden layer is 10. The input and output layers are expressed by purelin functions, and the hidden layer is expressed by sigmoid functions;

[0050] (4) The input and output machine learning data sets are divided into training set, validation set and test set, with the ratios of 60%, 20% and 20% respectively. The training set is used to train the artificial neural network. The training algorithm adopts the Levenberg-Marquardt method, the learning rate is set to 0.01, the number of iterations is set to 1000, and the training accuracy requirement is set to 0.001. The training is stopped when the output error meets the set requirements. The training set fitting regression graph is shown in the figure below: Figure 4 As shown, a nonlinear regression model is obtained;

[0051] (5) The validation set and test set were used to validate the nonlinear regression model. It was found that the validation test error was less than 0.001. The validation set fitting regression graph and the test set fitting regression graph were as follows: Figure 5 , 6 As shown, the obtained nonlinear regression model was used as the Mg content prediction model;

[0052] (6) Randomly set a set of smelting process parameters, specifically: Mg element addition amount is 167ppm, argon filling pressure is 40kPa, argon filling time is 5min, low temperature refining temperature is 1370, low temperature refining time is 1min, the smelting process parameters are normalized in the same way as step (2), and then input into the Mg content prediction model obtained in step (5), and the predicted content of trace element Mg is 112ppm. Actual smelting is carried out according to the set smelting process parameters, and the actual content of Mg in the alloy is determined by ICP-MS to be 97ppm. The difference between the predicted content of Mg and the actual content is 15ppm, thereby determining the effectiveness and accuracy of the Mg content prediction model;

[0053] (7) The target content of trace element Mg in the high-temperature alloy was set to 40, 70, and 100 ppm, respectively. The smelting process parameters such as argon filling pressure, argon filling time, low-temperature refining temperature, and low-temperature refining time were fixed. The amount of Mg element addition was determined by the Mg content prediction model. The Mg-containing alloy was smelted according to the adjusted smelting process parameters. The comparison between the target content of trace element Mg in the high-temperature alloy and the actual content of the two groups is shown in Table 1. It was found that the difference between the target content and the actual content was within the range of ±20 ppm, indicating that the method of the present invention can achieve precise control of the content of trace element Mg.

[0054] Table 1 Comparison of target content of trace element Mg in high temperature alloy and actual content of two groups

[0055]

[0056]

[0057] Based on the above results, it can be seen that the present invention can achieve precise control of the content of trace element Mg in high-temperature alloys by establishing an artificial neural network model to train and fit the complex nonlinear relationship between smelting process parameters and Mg element yield, which verifies the correctness and accuracy of the technical solution of the present invention, thereby providing a methodological guidance for formulating smelting process parameters for casting high-temperature alloys and controlling the content of easily burned elements in the alloy.

[0058] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. However, these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for accurately controlling the content of trace element Mg in high-temperature alloys based on artificial neural networks, characterized in that: The following steps are involved: (1) Determine the influence of Mg addition and smelting process parameters on the content of trace element Mg in high-temperature alloys through experiments or field production data, and obtain a machine learning data set between smelting process parameters and Mg content. The smelting process parameters include argon filling pressure, argon filling time, low-temperature refining temperature, and low-temperature refining time; (2) Normalize and standardize the machine learning dataset; (3) An artificial neural network model was established. The model included three layers: input layer, hidden layer and output layer. The input layer was used to receive input signals, and its variables were the amount of Mg element added, argon filling pressure, argon filling time, low-temperature refining temperature and low-temperature refining time. The output layer was used to output signals, i.e., calculation results, and its variables were Mg content. The number of neural nodes in the hidden layer was calculated according to the formula Determine, where m represents the number of neurons in the input layer, n represents the number of neurons in the output layer, and a represents a constant between [1, 10]; (4) dividing the input and output machine learning data sets into a training set, a validation set, and a test set, and using the training set to train the artificial neural network. The training algorithms include the Levenberg-Marquardt method, the Bayesian regularization method, and the quantized conjugate gradient method. When the output error meets the set requirements, the training is stopped and a nonlinear regression model is obtained. (5) using the validation set and the test set to perform validation tests on the nonlinear regression model to obtain a validation test error. When the validation test error is within a preset error range, the obtained nonlinear regression model is used as a Mg content prediction model; (6) randomly setting a set of smelting process parameters, performing the same normalization process as step (2) on the smelting process parameters, and inputting the smelting process parameters into the Mg content prediction model obtained in step (5) to obtain the corresponding predicted content of the trace element Mg, performing actual smelting according to the randomly set smelting process parameters, and determining the actual content of Mg in the alloy, and comparing the predicted content of Mg with the actual content to determine the validity and accuracy of the Mg content prediction model; (7) The target content of trace element Mg in the high-temperature alloy is set, the smelting process parameters, i.e., argon filling pressure, argon filling time, low-temperature refining temperature and low-temperature refining time, are fixed, the amount of Mg added in the alloy is adjusted, and the Mg content in the alloy is adjusted to the target content through the Mg content prediction model. The Mg-containing alloy is smelted according to the adjusted smelting process parameters, thereby achieving precise control of the trace element Mg content.

2. The method for accurately controlling the content of trace element Mg in a high-temperature alloy based on an artificial neural network according to claim 1, characterized in that: In step (1), the amount of Mg element added is in the range of 20-2000ppm, the argon filling pressure is in the range of 0-60kPa, the argon filling time is in the range of 1-15min, the low-temperature refining temperature is in the range of 1350-1450°C, the low-temperature refining time is in the range of 1-10min, and the method for determining the content of trace element Mg in the high-temperature alloy includes but is not limited to ICP-MS or ICP-AES.

3. The method for accurately controlling the content of trace element Mg in high-temperature alloys based on artificial neural network according to claim 1, characterized in that: In step (2), the data normalization and standardization processing methods include Min-Max normalization, Z-Score normalization and nonlinear normalization methods.

4. The method for accurately controlling the content of trace element Mg in high-temperature alloys based on artificial neural network according to claim 1, characterized in that: In step (2), the data normalization and standardization processing method is Min-Max normalization, and the formula is: Where X is the original data, Xmax and Xmin are the maximum and minimum values ​​in the data set, respectively, and Xnorm is the normalized data.

5. The method for accurately controlling the content of trace element Mg in high-temperature alloys based on artificial neural network according to claim 1, characterized in that: In step (3), the artificial neural network type is a Back-Propagation artificial neural network, the number of neurons in the input layer m is 5, the number of neurons in the output layer n is 1, the number of neurons in the hidden layer ranges from [4, 13], the input layer and the output layer are expressed using purelin functions, and the hidden layer is expressed using sigmoid functions.

6. The method for accurately controlling the content of trace element Mg in high-temperature alloys based on artificial neural network according to claim 5, characterized in that: The number of neurons in the hidden layer is 10.

7. The method for accurately controlling the content of trace element Mg in high-temperature alloys based on artificial neural network according to claim 1, characterized in that: In step (4), the proportions of the training set, validation set, and test set in the machine learning data set are 60-70%, 15-20%, and 15-20%, respectively. The training algorithm is the Levenberg-Marquardt method, the learning rate is 0.01, the number of iterations is 1000, and the training accuracy requirement is 0.

001.

8. The method for accurately controlling the content of trace element Mg in high-temperature alloys based on artificial neural network according to claim 1, characterized in that: In step (4), the specific training process is: a. Forward propagation: The training set data is input into the input layer of the Back-Propagation artificial neural network, and after the nonlinear transformation of the hidden layer, it reaches the output layer and outputs the result. If the output result does not match the expected output, the error is back propagated; b. Back propagation: Use the error function to calculate the error between the output result and the expected output, propagate the output error back through the hidden layer to the input layer layer by layer, and distribute the error to all neurons in each layer so that the error decreases along the gradient direction. The error obtained in each layer is used as the basis for adjusting the weights and thresholds. c. Repeat training: Repeat steps a and b until the preset convergence condition is reached, determine the weights and thresholds corresponding to the minimum input error, and complete the training.

9. The method for accurately controlling the content of trace element Mg in high-temperature alloys based on artificial neural network according to claim 1, characterized in that: In step (5), the verification test preset error is 0.

001.

10. The method for accurately controlling the content of trace element Mg in high temperature alloy based on artificial neural network according to claim 1, characterized in that: In step (6), the smelting process parameters are randomly set within the process parameter range described in step (1), and the alloy is vacuum induction melted according to the set process. The actual content of Mg element in the alloy is determined by ICP-MS or ICP-AES method. When the error between the predicted content and the actual content of the Mg content prediction model is within ±20ppm, the validity and accuracy of the prediction model are determined.

Images