Method for predicting room temperature fracture toughness of Niobium-Silicon based alloys based on machine learning

The machine learning-based method accurately predicts niobium-silicon alloy fracture toughness, overcoming inefficiencies in existing methods by training with collected data and using advanced algorithms, enabling efficient alloy design.

KR103046756B1Active Publication Date: 2026-11-10FOUND FOR RES & BUSINESS SEOUL NAT UNIV OF SCI & TECH
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
KR1020230081669
Authority / Receiving Office
KR · KR
Patent Type
Patents
Current Assignee / Owner
Filing Date
2023-06-26
Publication Date
2026-11-10
Estimated Expiration
2043-06-26

AI Technical Summary

Technical Problem

Existing methods for predicting the room temperature fracture toughness of niobium-silicon alloys are inefficient and inaccurate, relying on time-consuming trial-and-error experiments and mechanical measurements, making it difficult to achieve an optimal balance between high-temperature strength and room-temperature fracture toughness.

Method used

A machine learning-based method that trains a system with pre-collected data to quantitatively predict the influence of various factors affecting room temperature fracture toughness using a machine learning algorithm with high accuracy, analyzing correlations through algorithms like Maximum Information Coefficient and Pearson's Correlation Coefficient, and employing models such as Random Forest, Linear Regression, Bayesian Ridge, and Support Vector Machine.

Benefits of technology

This approach eliminates the need for repetitive experimental verification, reduces development time and costs, and enables efficient design of niobium-silicon alloys with excellent room-temperature fracture toughness, even for new chemical compositions without additional experiments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 112026045253539-PAT00008_ABST
    Figure 112026045253539-PAT00008_ABST
Patent Text Reader

Abstract

The present invention relates to a machine learning-based method for predicting the room temperature fracture toughness of a niobium-silicon alloy. The machine learning-based method for predicting the room temperature fracture toughness of a niobium-silicon alloy according to the present invention comprises: a data collection unit collecting experimental data on the room temperature fracture toughness of a niobium-silicon alloy through literature; an input / output value setting unit setting pre-set alloy elements and room temperature fracture toughness as input and output values ​​for machine learning analysis of the collected data; a correlation analysis unit analyzing the correlation between the input and output values ​​to quantitatively predict and compare the influence of the input value on the output value; a learning / prediction unit learning and predicting the room temperature fracture toughness of the niobium-silicon alloy through a plurality of machine learning algorithms based on the input and output values ​​for which the correlation has been analyzed; and a fracture toughness prediction unit predicting the room temperature fracture toughness of the niobium-silicon alloy for an untrained dataset using a model learned with the algorithm that showed the highest accuracy.
Need to check novelty before this filing date? Find Prior Art

Description

Technology Field

[0001] The present invention relates to a machine learning-based method for predicting the room temperature fracture toughness of a niobium-silicon alloy, and more specifically, to a machine learning-based method for predicting the room temperature fracture toughness of a niobium-silicon alloy with a high level of accuracy by quantitatively predicting the relative influence of various factors affecting the room temperature fracture toughness of the niobium-silicon alloy when only sampled data is input into a machine learning system, and by applying a machine learning algorithm having the highest prediction accuracy. Background Technology

[0003] Superalloys refer to metals that possess high mechanical properties and creep characteristics in environments of approximately 700°C or higher. With the advancement of technology today, interest in environmentally sustainable development is growing, and high-melting-point superalloys used in extreme environments such as space, aviation, and nuclear power are receiving significant attention. Among these, niobium (Nb)-based alloys exhibit a high melting point (above 1750°C), low density (6.6–7.2 g / cm³), and excellent high-temperature strength, making them a candidate material capable of replacing existing nickel (Ni)-based superalloys. Their application in gas turbine blades and jet engine parts is anticipated.

[0004] In particular, niobium-silicon alloys based on niobium and silicon (Si) and containing various metallic elements exhibit excellent properties at high temperatures above 1350°C, leading to extensive research. Niobium-silicon alloys consist of brittle intermetallic compound phases (such as silicide, Nb5Si3, or Nb3Si) and niobium solid solutions (Nbss) that possess relatively high ductility. While niobium solid solutions are generally known to improve ductility and room-temperature fracture toughness, silicides provide excellent high-temperature strength, creep resistance, and oxidation resistance.

[0005] However, the discrepancy between room-temperature fracture toughness and high-temperature strength limits the commercialization of niobium-silicon alloys. Therefore, an appropriate balance between high-temperature strength and room-temperature fracture toughness is required for commercialization. Generally, the room-temperature fracture toughness of niobium-silicon alloys is calculated as a function of the silicide volume fraction, and several methods have been proposed to control this volume fraction; among these, alloying is considered one of the most important and effective methods.

[0006] Recently, much research has been focused on developing niobium-silicon alloys with excellent high-temperature strength and room-temperature fracture toughness by adding alloying elements such as zirconium (Zr), yttrium (Y), hafnium (Hf), aluminum (Al), tungsten (W), titanium (Ti), molybdenum (Mo), and chromium (Cr). However, the development of new niobium-silicon alloys with specific required properties is not always possible due to the trade-offs in the influence of each alloying element. Furthermore, as the number of added alloying elements increases, it becomes difficult to accurately determine the correlation between the influence of each element and mechanical properties. Numerous alloying studies conducted to date for the commercialization of niobium-silicon alloys have relied on time-consuming trial-and-error methods, making them highly inefficient in terms of time and development costs. Therefore, there is a need to establish an optimal alloy design model.

[0007] Meanwhile, Korean Published Patent Application No. 10-2022-0116974 (Patent Document 1) discloses a "method for measuring fracture toughness based on a double clip gauge considering the plastic rotation coefficient." Accordingly, the method for measuring fracture toughness based on a double clip gauge considering the plastic rotation coefficient comprises the steps of: preparing a test specimen with a notch formed according to a predetermined standard specification and placing it on a seating portion of a test device; installing a jig having a seating surface of a predetermined height on the upper part of the test specimen and installing a displacement measuring portion of the test device at a position adjacent to the entrance of the notch; and applying a repetitive tensile load of a certain amplitude to the upper part of the notch using a load application portion of the test device to form a pre-fatigue crack extending inward from the inner end of the notch. The method is characterized by comprising: a step of calculating the amount of displacement in the plastic region based on the tip of the pre-fatigue crack based on the result of measuring the amount of displacement at the entrance of the pre-fatigue crack by applying a quasi-static tensile load while the displacement measuring part is in contact with the seating surface of the jig; and a step of calculating the plastic rotation coefficient value and the fracture toughness prediction value for the plastic region, respectively, based on the calculated amount of displacement in the plastic region and the height difference between the seating surface and the top surface of the test specimen.

[0008] In the case of Patent Document 1 as described above, although there is an advantage in that plastic deformation can be measured more sensitively than in the conventional case by modifying the plastic rotation coefficient value that affects plastic deformation occurring at the pre-cracking tip of the test specimen when calculating the fracture toughness prediction value, this is a mechanism in which the plastic rotation coefficient value and the fracture toughness prediction value for the plastic region are each calculated by a mechanical measurement method using a test device and a test specimen, so it is difficult to quantitatively predict the relative influence of various factors affecting room temperature fracture toughness, and there is a limitation in that it is difficult to predict room temperature fracture toughness with a high level of accuracy as the plastic rotation coefficient value and the fracture toughness prediction value are each calculated based on mechanical measurement values.

[0009] In addition, since the trial-and-error method, which requires the direct use of the test device and test specimen, is applied, it has the disadvantage of being inefficient in terms of time and cost. Prior art literature

[0010] Korean Published Patent Application No. 10-2022-0116974 (August 23, 2022) The problem to be solved

[0011] The present invention was created by comprehensively considering the above matters, and aims to provide a machine learning-based method for predicting the room-temperature fracture toughness of a niobium-silicon alloy with a high level of accuracy by training a machine learning system with pre-collected data to quantitatively predict the relative influence of various factors affecting the room-temperature fracture toughness of the niobium-silicon alloy, and by applying a machine learning algorithm having the highest prediction accuracy among a plurality of machine learning algorithms. means of solving the problem

[0013] To achieve the above objective, the machine learning-based method for predicting room temperature fracture toughness of a niobium-silicon alloy according to the present invention comprises:

[0014] a) A step in which a data collection unit collects experimental data on the room temperature fracture toughness of niobium-silicon alloys through literature;

[0015] b) A step in which the input / output value setting unit sets the alloy elements and room temperature fracture toughness, which are pre-set for machine learning analysis of the collected data, as input and output values;

[0016] c) A step in which the correlation analysis unit analyzes the correlation between input and output values ​​to quantitatively predict and compare the influence of the input value on the output value;

[0017] d) a step in which a learning / prediction unit learns and predicts the room temperature fracture toughness of a niobium-silicon alloy through a plurality of machine learning algorithms based on the input / output values ​​for which the correlations have been analyzed; and

[0018] e) The fracture toughness prediction unit is characterized by including a step of predicting the room temperature fracture toughness of a niobium-silicon alloy for an untrained dataset using a model trained with the algorithm that showed the highest accuracy.

[0019] Here, in step a), the data related to the room temperature fracture toughness of the niobium-silicon alloy may include alloying elements and experimental room temperature fracture toughness.

[0020] In addition, when setting the alloy elements and room temperature fracture toughness preset in step b) above as input and output values, the input value may be set to 17 chemical composition variables including silicon (Si), titanium (Ti), chromium (Cr), aluminum (Al), hafnium (Hf), molybdenum (Mo), tungsten (W), tantalum (Ta), zirconium (Zr), yttrium (Y), boron (B), carbon (C), iron (Fe), cobalt (Co), vanadium (V), manganese (Mn), and rhenium (Re), and the output value may be room temperature fracture toughness (MPa·m 1 / 2 It can be set to ).

[0021] In addition, in step c) above, when the correlation analysis unit analyzes the correlation between the input and output values, the correlation between the input and output values ​​can be analyzed through the Maximum Information Coefficient (MIC) algorithm and the Pearson's Correlation Coefficient (PCC) algorithm.

[0022] In addition, in step d) above, the plurality of machine learning algorithms may include four machine learning algorithms: Random Forest, Linear Regression, Bayesian Ridge, and Support Vector Machine. Effects of the invention

[0024] According to the present invention, there is no need to repeatedly verify experimental results of room-temperature fracture toughness, which is time-consuming, and to design new alloys; furthermore, high prediction accuracy can be obtained with minimal input data, thereby offering the advantage of efficiently developing niobium-silicon alloys exhibiting excellent room-temperature fracture toughness in terms of cost and time.

[0025] Furthermore, even for niobium-silicon alloys with new chemical compositions for which data is unavailable, it is possible to predict the room-temperature fracture toughness by training the model with pre-collected data through various prediction algorithms without conducting experiments, and inputting the sampled data into the developed machine learning model. This offers the advantage of improving the economic feasibility of niobium-silicon alloys used as next-generation ultra-high temperature structural materials. Brief explanation of the drawing

[0027] FIG. 1 is a schematic diagram showing the configuration of a machine learning-based room temperature fracture toughness prediction system for niobium-silicon alloys employed for the implementation of the prediction method of the present invention. FIG. 2 is a flowchart illustrating the execution process of a machine learning-based method for predicting the room temperature fracture toughness of a niobium-silicon alloy according to the present invention. Figure 3 is a table showing the rank scores of input features from correlation analysis using the Pearson correlation coefficient and maximum information correlation coefficient algorithms. Figure 4 is a diagram showing the learning results of four machine learning models with different numbers of features according to the ranking of the maximum information correlation coefficient and Pearson correlation coefficient algorithm analysis. Figure 5 is a figure showing the actual value versus the predicted value of room temperature fracture toughness for unlearned experimental data using a machine learning model developed by the inventor of the present invention. Figure 6 is a table showing the optimal combination of algorithms based on the fact that the Random Forest and Support Vector Machine algorithms showed high accuracy. Figure 7 is a graph showing the results of the optimal combination of Random Forest and Support Vector Machine algorithms based on the table in Figure 6. Specific details for implementing the invention

[0028] Terms and words used in this specification and claims should not be interpreted as being limited to their ordinary or dictionary meanings, and should be interpreted in a meaning and concept consistent with the technical spirit of the invention, based on the principle that the inventor can appropriately define the concept of the terms to best describe his invention.

[0029] Throughout the specification, when a part is described as "comprising" a certain component, this means that, unless specifically stated otherwise, it does not exclude other components but may include additional components. Furthermore, terms such as "…part," "…unit," "module," and "device" as used in the specification refer to a unit that processes at least one function or operation, and this may be implemented in hardware, software, or a combination of hardware and software.

[0030] Embodiments of the present invention will be described in detail below with reference to the attached drawings.

[0031] FIG. 1 is a schematic diagram showing the configuration of a machine learning-based room temperature fracture toughness prediction system for niobium-silicon alloys employed for the implementation of the prediction method of the present invention.

[0032] Referring to FIG. 1, a machine learning-based room temperature fracture toughness prediction system (100) for a niobium-silicon alloy employed for the implementation of the prediction method according to the present invention comprises a data collection unit (110), an input / output value setting unit (120), a correlation analysis unit (130), a learning / prediction unit (140), a fracture toughness prediction unit (150), and a control unit (160).

[0033] The data collection unit (110) collects experimental data on the room temperature fracture toughness of niobium-silicon alloys through literature (e.g., existing patents, papers, and corporate reports, etc.) as part of collecting experimental data at room temperature (e.g., niobium-silicon alloys, based on room temperature fracture toughness).

[0034] The input / output value setting unit (120) sets the alloy elements and room temperature fracture toughness, which are pre-set for machine learning analysis of the collected data, as input and output values.

[0035] The correlation analysis unit (130) analyzes the correlation between the input and output values ​​to quantitatively predict and compare the influence of the input value on the output value.

[0036] The learning / prediction unit (140) learns and predicts the room temperature fracture toughness of the niobium-silicon alloy through a plurality of machine learning algorithms based on the input / output values ​​of the above-mentioned correlation analysis.

[0037] The fracture toughness prediction unit (150) predicts the room temperature fracture toughness of a niobium-silicon alloy for an untrained data set using a model trained with the algorithm that showed the highest accuracy.

[0038] The control unit (160) checks the status and controls the operation of the data collection unit (110), input / output value setting unit (120), correlation analysis unit (130), learning / prediction unit (140), and fracture toughness prediction unit (150). Such a control unit (160) may be composed of a microprocessor or a microcontroller.

[0039] Here, the machine learning-based room temperature fracture toughness prediction system (100) of a niobium-silicon alloy having the above configuration may further include a communication unit (170) for transmitting / receiving data or information through wired / wireless communication with an external system or terminal.

[0040] Here, the data collection unit (110), input / output value setting unit (120), correlation analysis unit (130), learning / prediction unit (140), fracture toughness prediction unit (150), control unit (160), and communication unit (170) as described above may also be integrated into a single module or package.

[0041] Additionally, depending on the case, the data collection unit (110), input / output value setting unit (120), correlation analysis unit (130), learning / prediction unit (140), resistance prediction unit (150), control unit (160), and communication unit (170) may be replaced by a single computer system (e.g., desktop PC, laptop PC, etc.).

[0042] Then, below, we will explain a machine learning-based method for predicting the room temperature fracture toughness of niobium-silicon alloys based on a machine learning-based system for predicting the room temperature fracture toughness of niobium-silicon alloys having the configuration described above.

[0043] FIG. 2 is a flowchart illustrating the execution process of a machine learning-based method for predicting the room temperature fracture toughness of a niobium-silicon alloy according to an embodiment of the present invention.

[0044] Referring to FIG. 2, the machine learning-based method for predicting room temperature fracture toughness of a niobium-silicon alloy according to the present invention first collects experimental data on room temperature fracture toughness of a niobium-silicon alloy through literature (e.g., existing patents, papers, and corporate reports, etc.) as part of collecting experimental data obtained at room temperature (e.g., niobium-silicon alloy, room temperature fracture toughness as a standard) (step S201). Here, the data related to the room temperature fracture toughness of the niobium-silicon alloy may include alloy elements and experimental room temperature fracture toughness.

[0045] When data related to the room temperature fracture toughness of the niobium-silicon alloy is collected in this way, the input / output value setting unit (120) sets the pre-set alloy elements and room temperature fracture toughness as input values ​​and output values ​​for machine learning analysis of the collected data (step S202). Here, in setting the pre-set alloy elements and room temperature fracture toughness as input values ​​and output values, the input values ​​may be set to 17 chemical composition variables of silicon (Si), titanium (Ti), chromium (Cr), aluminum (Al), hafnium (Hf), molybdenum (Mo), tungsten (W), tantalum (Ta), zirconium (Zr), yttrium (Y), boron (B), carbon (C), iron (Fe), cobalt (Co), vanadium (V), manganese (Mn), and rhenium (Re), and the output values ​​may be room temperature fracture toughness (MPa·m 1 / 2 It can be set to ).

[0046] When the input and output values ​​are set according to the above, the correlation analysis unit (130) analyzes the correlation between the input and output values ​​to quantitatively predict and compare the influence of the input value on the output value (step S203). Here, in analyzing the correlation between the input and output values, the correlation between the input and output values ​​can be analyzed through the Maximum Information Coefficient (MIC) algorithm and the Pearson's Correlation Coefficient (PCC) algorithm. Such correlation analysis will be explained again later.

[0047] Here, I will add a little more explanation regarding the maximum information correlation coefficient and the Pearson correlation coefficient algorithm.

[0048] While the Pearson correlation coefficient algorithm identifies only linear relationships between input and target features, the maximum information correlation coefficient algorithm can identify non-linear relationships in large-scale, high-dimensional datasets. Both the Pearson correlation coefficient approach and the maximum information correlation coefficient approach can provide rankings based on quantitative analysis to determine threshold scores for feature selection. If high-ranking features are already known and generally accepted, correlation analysis can verify previously established mechanisms.

[0049] When the correlation analysis of the input / output values ​​is completed as described above, the learning / prediction unit (140) learns and predicts the room temperature fracture toughness of the niobium-silicon alloy through a plurality of machine learning algorithms based on the input / output values ​​for which the correlation has been analyzed (step S204). Here, the plurality of machine learning algorithms may include four machine learning algorithms: Random Forest, Linear Regression, Bayesian Ridge, and Support Vector Machine. The learning and prediction of the room temperature fracture toughness of the niobium-silicon alloy will be explained again later.

[0050] Here, we will add a little more explanation regarding each of the machine learning algorithms mentioned above: Random Forest, Linear Regression, Bayesian Ridge, and Support Vector Machine.

[0051] Random Forest is an ensemble learning method used for classification and regression analysis, which derives classifications or averages from multiple decision trees constructed during the training process. Decision trees generate branches that best classify the response variable, whereas Random Forest selects each branch randomly when generating them. This Random Forest improves generalization performance by ensuring that the predictions of each tree are decored due to randomness. It also has advantages such as the ability to handle thousands of input variables without elimination, high accuracy and learning rates, and maintaining good performance even on noisy data.

[0052] Linear regression models the linear relationship between a single dependent variable and N explanatory variables. It assumes that the dependent variable is in a relationship that can be expressed as a linear combination of explanatory variables and linear regression coefficients, and measures the accuracy of the trained model using the least squares method. This linear regression has the advantages of being easily accessible, having a simple algorithm, and being easy to interpret.

[0053] Bayesian Ridge is an algorithm that utilizes conditional posterior distributions or predictive distributions for models, allowing for parameter estimation or prediction by inputting multiple different survey results into a single analysis model. Such Bayesian Ridge has the advantages of being able to study the posterior distribution of coefficient estimates, facilitating interpretation, and improving the flexibility of model design (e.g., changing the likelihood function, constructing complex hierarchical models).

[0054] Support Vector Machines are models that can be used for classification as well as regression, and are considered non-parametric techniques because they are based on kernel functions. These Support Vector Machines classify data by generating hyperplanes based on the principle of finding the optimal (maximizing margin) line for classifying data. In other words, Support Vector Machines are models that define decision boundaries, that is, baselines for classification. Compared to logistic regression or neural network algorithms, such Support Vector Machines are useful for handling more complex non-linear functions and have the advantage of preventing overfitting to training data.

[0055] When learning and prediction are completed through multiple machine learning algorithms in this way, the fracture toughness prediction unit (150) predicts the room temperature fracture toughness of a niobium-silicon alloy for an untrained dataset using a model trained with the algorithm that showed the highest accuracy among the multiple machine learning algorithms (i.e., four machine learning algorithms: Random Forest, Linear Regression, Bayesian Ridge, and Support Vector Machine) (e.g., Random Forest algorithm) (step S205). This will be further explained below.

[0056] Below, we will provide further explanation regarding the machine learning-based method for predicting the room-temperature fracture toughness of a niobium-silicon alloy according to the present invention as described above.

[0057] Figure 3 is a table showing the rank scores of input features from correlation analysis using the Pearson correlation coefficient and maximum information correlation coefficient algorithms.

[0058] As can be seen from the table in Figure 3, the five alloying elements with the largest values ​​in the maximum information correlation coefficient are aluminum (Al), hafnium (Hf), yttrium (Y), chromium (Cr), and titanium (Ti), and the five alloying elements with the largest values ​​in the absolute value of the Pearson correlation coefficient are yttrium (Y), hafnium (Hf), aluminum (Al), titanium (Ti), and molybdenum (Mo). Among these, molybdenum (Mo) and manganese (Mn) have a negative correlation with fracture toughness. As can be seen from the table in Figure 3, the measured values ​​of the maximum information correlation coefficient and the Pearson correlation coefficient range from 0.143 to 0.321 and from 0 to 0.471, respectively. Given that both values ​​are relatively small, it can be confirmed that there is not a large correlation between the alloying elements (input characteristics) and room temperature fracture toughness (output characteristics). This is because various factors, such as the phase constituting the niobium-silicon alloy or processing conditions, have a greater impact on fracture toughness than the influence of the alloying elements.

[0059] It is generally accepted that the fracture toughness of materials is improved by incorporating ductile reinforcements. Previous studies have reported that configuring ductile Nbss (matrix) phases and other phases in appropriate proportions is important for enhancing fracture toughness in niobium-silicon alloys. For this reason, the correlation between chemical composition and mechanical properties often does not appear as a regular distribution, resulting in small absolute values ​​for the maximum information correlation coefficient and the Pearson correlation coefficient. Both algorithmic analyses of the maximum information correlation coefficient and the Pearson correlation coefficient indicate that Al, Hf, Ti, and Y are the most powerful characteristics influencing the room-temperature fracture toughness of niobium-silicon alloys. Ti is well known to significantly improve the fracture toughness of Nb-based alloys composed of Nbss and intermetallic compounds. According to numerous studies, Al is considered an element that does not directly affect the room-temperature fracture toughness of niobium-silicon alloys. To increase the strength and oxidation resistance of niobium-silicon alloys, Al is typically added along with Ti.

[0060] Furthermore, it has been noted that Al enhances the positive effect of Ti on room-temperature fracture toughness by increasing the solid solubility of Ti in Nbss and Nb5Si3 phases. Therefore, since most of the collected Al data is combined with Ti, Al is expected to be a factor showing a high correlation in the analysis of the maximum information correlation coefficient and Pearson correlation coefficient algorithms. The addition of Hf to niobium-silicon alloys improves room-temperature fracture toughness by inducing a sophisticated eutectic structure that increases the degree of crack propagation deformation. The addition of Y (yttrium) improves fracture toughness in niobium-silicon alloys by forming the α-Nb5Si3 phase; it was found that α-Nb5Si3 exhibits higher fracture toughness than γ-Nb5Si because α-Nb5Si3 possesses higher unstable stacking fault energies and Peierls-Nabarros barrier energies. Therefore, the results of the correlation analysis in this study (invention) can provide reliable insights into the relationship between input and output characteristics, which can be used by other researchers as an important indicator when designing niobium-silicon alloys with excellent room-temperature fracture toughness.

[0061] In this invention, machine learning training is performed based on the correlation analysis described above. To develop a machine learning model with optimal accuracy among four machine learning algorithms, the performance of each model was evaluated by varying the number of input features.

[0062] In the present invention, four representative machine learning models (Random Forest (RF), Linear Regression (LR), Bayesian Ridge (BR), and Support Vector Machine (SVM)) were used to accurately predict the fracture toughness of niobium-silicon-based alloys.

[0063] Figure 4 is a diagram showing the learning results of four machine learning models with different numbers of features according to the ranking of the maximum information correlation coefficient and Pearson correlation coefficient algorithm analysis.

[0064] Referring to Fig. 4, the accuracy (R) of each machine learning model according to the number of upper features. 2 ) can be interpreted in relation to key input features. For example, the accuracy (R) of four machine learning models 2 ) increases when the three features (Al, Hf, Y) that have the greatest influence on room-temperature fracture toughness are input. As a result of machine learning, the Random Forest model including all 17 alloying elements showed the highest accuracy (R) among the input features used in the analysis of the maximum information correlation coefficient and Pearson correlation coefficient algorithms. 2 ...did ) ... The above results demonstrated that the Random Forest model exhibited the most optimized performance for predicting room-temperature fracture toughness, even with a small training dataset, among the four machine learning models developed by the inventors of the present invention. The Random Forest algorithm generally estimates missing data and achieves an acceptable level of accuracy (R 2 It is recognized as an effective method for maintaining ). In addition, it has the advantage of being able to process a relatively large number of input features simultaneously without deleting variables.

[0065] When all alloying elements are used as input features for prediction, the accuracy (R) of the Random Forest model 2 ) is the maximum value (R 2 : represents 0.51). R 2 Although a value of 0.51 is higher than that of the other three machine learning models developed by the inventors of the present invention, the prediction accuracy can be considered low. The aforementioned results may reflect that existing experimental alloy design datasets, which could be considered "massive," are in fact too small to allow for an effective predictive analysis approach. However, in the present invention, low accuracy (R 2 The value of ) does not mean that the prediction is meaningless. R 2The value indicates the extent to which the dependent variable is explained by the independent variable, so it does not reflect the accuracy of the causal relationship. Therefore, the R² of the Random Forest model 2 The fact that the value is 0.51 does not appear to suggest that the influence of alloying elements has little effect on room-temperature fracture toughness; rather, approximately 51% of the various factors affecting the target material properties may be attributed to chemical composition. Processing conditions, phase volume fraction, and morphology can be additional factors affecting room-temperature fracture toughness. In Figure 4, (a) shows the prediction accuracy (R) of the machine learning model based on the analysis of the maximum information correlation coefficient algorithm. 2 (b) is the prediction accuracy of the machine learning model based on the Pearson correlation coefficient algorithm analysis (R 2 Each represents ).

[0066] To obtain a preliminary impression of the performance of the developed machine learning model, room-temperature fracture toughness was predicted using 12 test datasets not included in the 126-dataset training dataset. Additionally, a validation strategy utilizing new datasets allows for the evaluation of uncertainties, such as overfitting, occurring between machine learning models. For prediction, the highest accuracy (R²) including all 17 input features was used. 2 A random forest model with ) was selected.

[0067] Figure 5 is a figure showing the actual value versus the predicted value of room temperature fracture toughness for unlearned experimental data using a machine learning model developed by the inventor of the present invention.

[0068] As shown in Figure 5, the predictions of the random forest model fall within the 90% confidence interval for this test dataset of 12 new niobium-silicon alloys. Unlike the random forest model, linear regression, Bayesian Ridge, and support vector machine models exhibit significant errors. These results demonstrate that the random forest model provides superior performance in predicting room temperature fracture toughness compared to other models. However, for this selected test dataset, a clear correlation between the predicted values ​​and the actual values ​​is not evident, which suggests limitations in the predictive capability for niobium-silicon alloys using the current input features.

[0069] Meanwhile, Figure 6 is a table showing the optimal combination of algorithms based on the high accuracy shown by the Random Forest and Support Vector Machine algorithms, and Figure 7 is a graph showing the result of the optimal combination of the Random Forest and Support Vector Machine algorithms based on the table of Figure 6.

[0070] Referring to Figures 6 and 7, as described above, when a room temperature fracture toughness prediction test was performed using four machine learning models (Random Forest (RF), Linear Regression (LR), Bayesian Ridge (BR), and Support Vector Machine (SVM)) for 17 input features (17 alloy elements), the Random Forest and Support Vector Machine algorithms showed high accuracy in both the analysis of maximum information correlation coefficients and Pearson correlation coefficients. Therefore, by combining the Random Forest and Support Vector Machine algorithms, an optimal algorithm for predicting room temperature fracture toughness can be derived as shown in Figures 6 and 7.

[0072] As described above, the machine learning-based method for predicting the room temperature fracture toughness of a niobium-silicon alloy according to the present invention quantitatively predicts the relative influence of various factors affecting the room temperature fracture toughness of the niobium-silicon alloy and applies a machine learning algorithm having the highest prediction accuracy among multiple machine learning algorithms. This eliminates the need to repeatedly verify experimental results of room temperature fracture toughness, which is time-consuming, and to design new alloys. Furthermore, since high prediction accuracy can be obtained with minimal input data, there is an advantage in being able to develop a niobium-silicon alloy exhibiting excellent room temperature fracture toughness efficiently in terms of cost and time.

[0073] Furthermore, even for niobium-silicon alloys with new chemical compositions for which data is unavailable, it is possible to predict the room-temperature fracture toughness by training the model with pre-collected data through various prediction algorithms without conducting experiments, and inputting the sampled data into the developed machine learning model. This offers the advantage of improving the economic feasibility of niobium-silicon alloys used as next-generation ultra-high temperature structural materials.

[0074] Although the present invention has been described in detail through preferred embodiments, the invention is not limited thereto, and it is obvious to those skilled in the art that various modifications and applications can be made within the scope of the technical concept of the invention. Accordingly, the true scope of protection of the present invention should be interpreted by the following claims, and all technical concepts within an equivalent scope should be interpreted as being included within the scope of rights of the present invention. Explanation of the symbols

[0075] 110: Data collection unit 120: Input / output value setting unit 130: Correlation Analysis Unit 140: Learning / Prediction Unit 150: Fracture toughness prediction unit 160: Control unit 170: Communications Department

Claims

Claim 1 a) a step in which a data collection unit collects experimental data on the room temperature fracture toughness of niobium-silicon alloys through literature; b) a step in which an input / output value setting unit sets pre-set alloy elements and room temperature fracture toughness as input and output values ​​for machine learning analysis of the collected data; c) a step in which a correlation analysis unit analyzes the correlation between the input and output values ​​to quantitatively predict and compare the influence of the input values ​​on the output values; d) a step in which a learning / prediction unit learns and predicts the room temperature fracture toughness of the niobium-silicon alloy through a plurality of machine learning algorithms based on the input and output values ​​for which the correlations have been analyzed; and e) a step of predicting the room temperature fracture toughness of a niobium-silicon alloy for an untrained dataset using a model trained with the algorithm that showed the highest accuracy in the fracture toughness prediction unit, wherein in step b), the pre-set alloy elements and the room temperature fracture toughness are set as input and output values, the input values ​​are set as chemical composition variables of molybdenum (Mo), tungsten (W), tantalum (Ta), zirconium (Zr), yttrium (Y), boron (B), carbon (C), iron (Fe), cobalt (Co), vanadium (V), manganese (Mn), and rhenium (Re), and in step c), additional factors affecting the output value (room temperature fracture toughness) include processing conditions, phase volume fraction, and shape, characterized by a machine learning-based method for predicting the room temperature fracture toughness of a niobium-silicon alloy. Claim 2 In claim 1, the room temperature fracture toughness data of the niobium-silicon alloy in step a) includes alloy elements and experimental room temperature fracture toughness, a machine learning-based method for predicting the room temperature fracture toughness of a niobium-silicon alloy. Claim 3 delete Claim 4 A machine learning-based method for predicting room temperature fracture toughness of a niobium-silicon alloy, characterized in that, in step c), the correlation analysis unit analyzes the correlation between input and output values ​​using a Maximum Information Coefficient (MIC) algorithm and a Pearson's Correlation Coefficient (PCC) algorithm. Claim 5 A machine learning-based method for predicting room temperature fracture toughness of a niobium-silicon alloy, wherein, in step d), the plurality of machine learning algorithms includes four machine learning algorithms: Random Forest, Linear Regression, Bayesian Ridge, and Support Vector Machine.

Citation Information

Patent Citations

  • Machine Learning to Accelerate Alloy Design

    US20200257933A1