Mechanical property prediction method and mechanical property prediction apparatus
The method and device improve mechanical property prediction by incorporating factors like temperature and solidification rate, using ensemble learning to enhance accuracy and reduce the number of explanatory variables, addressing the inefficiencies of existing technologies.
Patent Information
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- RYOBI
- Filing Date
- 2024-04-17
- Publication Date
- 2026-06-02
AI Technical Summary
Existing methods for predicting the mechanical properties of castings are time-consuming, costly, and do not adequately account for the influence of reinforcing elements dissolved after solidification or changes in mold temperature, requiring a high computational cost due to the need for multiple explanatory variables.
A method and device using a cavity in openable and closable molds that predict mechanical properties by considering factors like temperature, solidification rate, and wall thickness changes, employing ensemble learning to combine multiple learning models for improved accuracy with fewer explanatory variables.
Provides high-accuracy mechanical property predictions with a reduced number of variables, enhancing the efficiency and effectiveness of the prediction process.
Smart Images
Figure 0007869257000002 
Figure 0007869257000003 
Figure 0007869257000004
Abstract
Description
[Technical Field]
[0001] The present invention relates to a method for predicting mechanical properties and a device for predicting mechanical properties. [Background technology]
[0002] Traditionally, during the design phase of a casting, determining whether the casting possessed the desired mechanical properties involved repeatedly creating prototypes while adjusting casting conditions and material composition based on the designer's experience. However, this method was time-consuming and costly. Therefore, in recent years, objective methods that do not rely solely on the designer's experience have been developed to predict the mechanical properties of castings.
[0003] For example, Patent Document 1 below discloses a technique for predicting the mechanical properties of a casting by creating a mold model for CAE (Computer Aided Engineering) analysis, using this mold model to perform molten metal flow analysis and solidification analysis under predetermined casting conditions, calculating predetermined factors from the analysis results, and then performing a multiple regression analysis with the desired mechanical properties as the dependent variable, using these factors as explanatory variables. [Prior art documents] [Patent Documents]
[0004] [Patent Document 1] Japanese Patent Publication No. 2019-105592 [Overview of the Initiative] [Problems that the invention aims to solve]
[0005] However, the above technology does not take into account the amount of reinforcing elements dissolved in the casting after solidification, or the effect on mechanical properties due to changes in the thickness of the casting caused by changes in mold temperature, indicating room for improvement in the selection of explanatory variables. Furthermore, since the above technology requires at least three factors as explanatory variables, the computational cost of obtaining these variables is high, and a simpler prediction method was needed.
[0006] This invention has been made in view of the problems present in the prior art described above, and its objective is to provide a method for predicting the mechanical properties of a casting with high accuracy by considering various influences that occur in the casting. Furthermore, it aims to provide a method for easily obtaining good prediction results with a small number of explanatory variables. [Means for solving the problem]
[0007] The present invention will be described below. Reference numerals for the attached drawings are indicated in parentheses to facilitate understanding of the present invention; however, this does not mean that the present invention is limited to the illustrated forms.
[0008] The mechanical properties prediction method according to the present invention involves a cavity formed in a pair of openable and closable molds. Made of aluminum alloyA method for predicting the mechanical properties of a casting (C) formed by injecting molten metal and allowing the molten metal to solidify, comprising: an explanatory variable acquisition step (step S2) in which an explanatory variable is obtained by performing a predetermined analysis using a mold model for CAE analysis which divides the cavity into a plurality of elements, thereby obtaining explanatory variables which are information relating to the casting conditions of the casting (C) or the molten metal; and a mechanical property prediction step (step S4) in which the explanatory variable is input into a prediction model which calculates the correspondence between the explanatory variable and the target variable which is information relating to the mechanical properties of the casting (C), and predicts the target variable, wherein the explanatory variables are the temperature of the casting (C) at the time the casting (C) is removed from the mold, the temperature of the molten metal at the completion of filling, the solidification rate of the casting (C), and the amount of change in the wall thickness of the casting (C), and the prediction model is obtained from linear regression or ensemble learning which generates a single learning model by combining a plurality of learning models.
[0011] Furthermore, the mechanical properties prediction method according to the present invention can include factors related to the cleanliness of the molten metal in addition to the explanatory variables.
[0015] The mechanical properties prediction device (10) according to the present invention is a cavity formed in a pair of openable and closable molds. Made of aluminum alloy A mechanical property prediction device (10) that predicts the mechanical properties of a casting (C) formed by injecting molten metal and allowing the molten metal to solidify, comprising: an explanatory variable acquisition unit (22) that obtains explanatory variables which are information relating to the casting conditions of the casting (C) or the molten metal by performing a predetermined analysis using a mold model for CAE analysis which divides the cavity into a plurality of elements; and a mechanical property prediction unit (32) that inputs the explanatory variables into a prediction model which calculates the correspondence between the explanatory variables and the target variable which is information relating to the mechanical properties of the casting (C), and predicts the target variable, wherein the explanatory variables are the casting (C)Using the temperature of the casting (C) at the time of removal from the mold, the temperature of the molten metal at the completion of filling, the solidification rate of the casting (C), and the amount of change in the wall thickness of the casting (C), the prediction model is obtained from ensemble learning that combines linear regression or a plurality of learning models to generate one learning model.
Effect of the Invention
[0017] According to the present invention, by considering various influences occurring in the casting, a method for predicting the mechanical properties of the casting with high accuracy can be provided. In addition, a method for easily obtaining good prediction results with a small number of explanatory variables can be provided.
Brief Description of the Drawings
[0018] [Figure 1] It is a functional block diagram of the mechanical property prediction device according to the present embodiment. [Figure 2] It is a hardware configuration diagram of the mechanical property prediction device according to the present embodiment. [Figure 3] It is a flowchart of the mechanical property prediction method according to the present embodiment. [Figure 4] It is a schematic diagram of a casting used for obtaining predicted values and measured values in the mechanical property prediction method according to the example. [Figure 5] It is a graph showing the prediction accuracy of the target variable for each of the mechanical property prediction methods according to the conventional example and the present embodiment. In the figure, subfigure (a) is a graph showing the results according to the conventional example, and subfigure (b) in the figure is a graph showing the results according to the mechanical property prediction method according to the present embodiment. [Figure 6] It is a graph comparing the predicted value and the measured value of the target variable in Example 2. In the figure, subfigure (a) is a graph comparing the predicted value obtained by linear regression with the measured value, and subfigure (b) in the figure is a graph comparing the predicted value obtained by ensemble learning with the measured value. [Figure 7]The graphs in Example 3 compare the predicted and actual values of the target variable. Figure (a) shows a comparison of the predicted values obtained by linear regression with the actual values, and Figure (b) shows a comparison of the predicted values obtained by ensemble learning with the actual values. [Figure 8] The graphs in Example 4 compare the predicted and actual values of the target variable. Figure (a) shows a comparison of the predicted values obtained by linear regression with the actual values, and Figure (b) shows a comparison of the predicted values obtained by ensemble learning with the actual values. [Figure 9] The graphs in Example 5 compare the predicted and actual values of the target variable. Figure (a) shows a comparison of the predicted values obtained by linear regression with the actual values, and Figure (b) shows a comparison of the predicted values obtained by ensemble learning with the actual values. [Figure 10] The graphs in Example 6 compare the predicted and actual values of the target variable. Figure (a) shows a comparison of the predicted values obtained by linear regression with the actual values, and Figure (b) shows a comparison of the predicted values obtained by ensemble learning with the actual values. [Figure 11] This graph compares the predicted values obtained through ensemble learning with the actual values for the target variable in Example 7. [Figure 12] The graphs in Example 8 compare the predicted and actual values of the target variable. Figure (a) shows a comparison of the predicted values obtained by linear regression with the actual values, and Figure (b) shows a comparison of the predicted values obtained by ensemble learning with the actual values. [Figure 13] The graphs in Example 9 compare the predicted and actual values of the target variable. Figure (a) shows a comparison of the predicted values obtained by linear regression with the actual values, and Figure (b) shows a comparison of the predicted values obtained by ensemble learning with the actual values. [Figure 14] This graph shows the correlation between Mg and Si content and 0.2% yield strength. [Figure 15]This graph compares the predicted value obtained by linear regression with the actual value for the dependent variable in Example 10. [Modes for carrying out the invention]
[0019] Hereinafter, preferred embodiments for carrying out the present invention will be described with reference to the drawings. Note that the following embodiments are not intended to limit the invention as described in each claim, and not all combinations of features described in the embodiments are necessarily essential to the solution of the invention.
[0020] [Mechanical properties prediction device] First, the configuration of the mechanical property prediction device 10 according to this embodiment will be explained using Figures 1 and 2. Here, Figure 1 is a functional block diagram of the mechanical property prediction device 10 according to this embodiment, and Figure 2 is a hardware configuration diagram of the mechanical property prediction device 10 according to this embodiment.
[0021] As shown in Figure 1, the mechanical property prediction device 10 according to this embodiment comprises a CAE analysis unit 20 and a model prediction unit 30. The CAE analysis unit 20 is composed of a mold model creation unit 21 and an explanatory variable acquisition unit 22, while the model prediction unit 30 is composed of a prediction model acquisition unit 31 and a mechanical property prediction unit 32.
[0022] The mold model creation unit 21 creates a mold model, which is a mesh model for CAE analysis, by dividing the cavity formed in a pair of openable and closable molds into multiple elements, based on information regarding model creation conditions and casting conditions. Here, the model creation conditions include design data of the casting, such as 3D CAD (Computer-Aided Design) data. The casting conditions include information regarding the temperature and composition of the molten metal poured into the cavity. The division size and element shapes of the mold model can be determined arbitrarily.
[0023] The explanatory variable acquisition unit 22 acquires explanatory variables, including information about the casting conditions or molten metal of the casting, by performing a predetermined analysis using the mold model created by the mold model creation unit 21. In the mechanical property prediction device 10 according to this embodiment, predetermined explanatory variables are obtained by performing either or both of the solidification analysis and / or flow analysis of the molten metal. In addition to the solidification analysis and / or flow analysis of the molten metal, the analysis method can be freely selected according to the explanatory variables to be acquired.
[0024] Incidentally, it is known that the mechanical properties of cast products, including die-cast products, are greatly influenced by the thickness of the casting. Therefore, conventionally, the inventors have used "thickness of the casting" as an explanatory variable to easily predict the mechanical properties of castings, and have constructed a simple prediction model by fitting a straight line obtained by the least squares method to a scatter plot of mechanical property values plotted against "thickness of the casting." However, this prediction model had a large variability in the plot relative to the straight line, and there was room for improvement in prediction accuracy. Therefore, the inventors diligently conducted research to devise a new explanatory variable and found that accuracy could be improved by adopting "a factor related to the solid solubility of reinforcing elements contained in the casting after solidification" as an explanatory variable. Specifically, they used "cooling rate after solidification of the casting," and conceived that this index could be substituted with "the temperature of the casting at the time of removal from the mold."
[0025] Furthermore, the inventors anticipated that the accuracy of predicting the mechanical properties of a casting would improve by adding explanatory variables that show a correlation with the mechanical properties of the casting, in addition to the explanatory variables mentioned above. Based on this prediction, in this embodiment, one or more of the following were adopted as explanatory variables: "factors related to the amount of solid solution of reinforcing elements contained in the casting after solidification," "factors related to the growth of the solidification structure of the casting," "amount of change in wall thickness of the casting," "factors related to the cleanliness of the molten metal," and "solid phase ratio during the flow of the molten metal." Note that "amount of solid solution of reinforcing elements" includes information on the amount of reinforcing elements that dissolve into the base material of the casting. Also, "factors related to the growth of the solidification structure of the casting" are factors that represent the change in the size of the structure composed of the base material of the casting, and there is a correlation that the faster the solidification rate of the molten metal, the smaller the size of the structure becomes and the higher the strength.
[0026] Specific combinations of the above factors include using either or both of the following factors for "factors related to the amount of reinforcing elements dissolved in the casting after solidification": "the temperature of the casting at the time of removal from the mold" and / or "the temperature of the molten metal at the completion of filling"; using "solidification rate" for "factors related to the growth of the solidified structure after casting"; and using "contact time of the molten metal with air" for "factors related to the cleanliness of the molten metal."
[0027] The prediction model acquisition unit 31 acquires a prediction model that calculates the correspondence between the explanatory variables acquired by the explanatory variable acquisition unit 22 and the target variable, which is information regarding the mechanical properties of the casting. Here, the inventors conducted diligent research to further improve the prediction accuracy and came up with the idea of using machine learning methods, particularly ensemble learning, as the prediction model. Therefore, in the mechanical property prediction device 10 according to this embodiment, a machine learning model using linear regression or ensemble learning is adopted as the prediction model. Here, linear regression is a model in which the correspondence between explanatory variables and the target variable is modeled as a linear relationship as shown in the following formula (Equation 1). Note that X k is the kth explanatory variable, Y is the dependent variable, a k represents the coefficient of the k-th explanatory variable, and a0 represents a constant.
[0028]
number
[0029] Ensemble learning is a machine learning model that generates a single learning model by combining multiple learning models that have learned the correspondence between explanatory variables and the target variable. Here, the calculation of the correspondence between explanatory variables and the target variable in linear regression and ensemble learning was performed using experimental data accumulated in the past. In addition to past experimental data, it is also possible to calculate the correspondence by using, for example, publicly available external databases, or by combining these.
[0030] In this embodiment, ensemble learning uses a random forest, which combines multiple decision trees as learning models in parallel. However, the present invention can employ other ensemble learning models, as well as methods that combine multiple learning models in series.
[0031] The mechanical properties prediction unit 32 predicts the target variable by inputting the explanatory variables obtained by the explanatory variable acquisition unit 22 into the prediction model acquired by the prediction model acquisition unit 31. In this embodiment, one of the following was adopted as the target variable: "0.2% yield strength", "tensile strength", or "elongation at break".
[0032] Figure 2 shows a hardware configuration diagram of the mechanical property prediction device 10 according to this embodiment. The mechanical property prediction device 10 is physically implemented in a computer 40 and consists of an arithmetic unit 41, an input device 42, a main memory 43, an auxiliary memory 44, and an output device 45.
[0033] The operation of each part of the mechanical properties prediction device 10 is stored in program form in an auxiliary storage device 44, which consists of an HDD (Hard Disk Drive) and an SSD (Solid State Drive), etc. The arithmetic unit 41 consists of a CPU (Central Processing Unit) and a GPU (Graphics Processing Unit), etc., reads the program from the auxiliary storage device 44, expands it into the main storage device 43, which consists of RAM (Random Access Memory) and ROM (Read Only Memory), etc., and executes the processing of each part of the mechanical properties prediction device 10 according to the program. The input device 42 consists of a mouse and a keyboard, etc., and can input data related to model creation conditions and casting conditions to the mold model creation unit 21 of the mechanical properties prediction device 10. The output device 45 consists of a display, etc., and can display the prediction results calculated by the mechanical properties prediction unit 32. In this embodiment, the mechanical properties prediction device 10 is configured on one computer 40, but it may be configured on multiple computers.
[0034] The configuration of the mechanical property prediction device 10 according to this embodiment has been described above using Figures 1 and 2. Next, the mechanical property prediction method according to this embodiment will be described using Figure 3. Here, Figure 3 is a flowchart of the mechanical property prediction method according to this embodiment.
[0035] [Method for predicting mechanical properties] As shown in Figure 3, the mechanical property prediction method according to this embodiment consists of a mold model creation step (step S1), an explanatory variable acquisition step (step S2), a prediction model acquisition step (step S3), and a mechanical property prediction step (step S4), and each process is performed by the mechanical property prediction device 10.
[0036] In the mold model creation step (step S1), various conditions such as model creation conditions and casting conditions are input to the mold model creation unit 21. As a result, the mold model creation unit 21 creates a mesh model for CAE analysis, which is the mold model, based on the input conditions.
[0037] In the explanatory variable acquisition step (step S2), solidification analysis of the molten metal and molten metal flow analysis are performed based on the mold model and casting conditions created in the mold model creation step (step S1). This allows the desired explanatory variables to be obtained.
[0038] In the prediction model acquisition step (step S3), after the prediction model is acquired by the prediction model acquisition unit 31, the prediction model is stored in the mechanical properties prediction unit 32. Note that the prediction model acquisition step (step S3) does not need to be performed before the mechanical properties prediction step (step S4), and it can also be performed before the mold model creation step (step S1) or the explanatory variable acquisition step (step S2).
[0039] In the mechanical property prediction step (step S4), the explanatory variables obtained in the explanatory variable acquisition step (step S2) are input to the mechanical property prediction unit 32, and the predicted value of the target variable is output. After these steps, the mechanical property prediction method according to this embodiment is completed.
[0040] The mechanical property prediction method according to this embodiment has been described above with reference to Figure 3. Next, in order to verify the prediction accuracy of the predicted mechanical properties obtained by the mechanical property prediction method according to this embodiment, the measured values obtained by actually measuring the mechanical properties of the casting were compared with the predicted values. The following describes an example for verifying such prediction accuracy.
[0041] [Verification of prediction accuracy] First, the structure of the casting C used in the example will be explained using Figure 4. Here, Figure 4 is a schematic diagram of the casting C used to obtain predicted and measured values in the mechanical property prediction method according to the example.
[0042] Casting C is a die-cast product formed by injecting molten metal into a cavity formed in a pair of openable and closable molds and allowing the molten metal to solidify. It is made of an aluminum alloy. Casting C may be of any shape and may be made of any material.
[0043] The configuration of the casting C used to obtain predicted and measured values in the mechanical property prediction method according to this embodiment has been described above using Figure 4. Below, we will describe the results of verifying the prediction accuracy of the mechanical property prediction method for casting C while changing the explanatory variables and objective variables.
[0044] [Example 1] First, as Example 1, a comparison was made with a conventional example. Specifically, the prediction accuracy was compared for "thickness of the casting," which was a conventional explanatory variable used when simply predicting mechanical properties, and "temperature of the casting at the time of removal from the mold," which was an explanatory variable discovered by the inventors. In this comparison, "0.2% yield strength" was used as the target variable among the mechanical properties of casting C, and linear regression was used as the prediction model. The results are shown in Figure 5. Here, Figure 5 is a graph comparing the prediction accuracy of the target variable for the mechanical property prediction method of the conventional example and the present embodiment, with sub-figure (a) showing the results of the conventional example and sub-figure (b) showing the results of the mechanical property prediction method of the present embodiment.
[0045] The dashed lines in Figure 5 represent the linear function obtained by linear regression for each plot. The constants and coefficients of these linear functions were calculated using the least squares method, and the smaller the variability of the plots relative to the dashed line, the higher the prediction accuracy. Here, in order to examine the prediction accuracy of the "0.2% tolerance" for each explanatory variable in Figure 5, the coefficient of determination (R) of the dashed line for each plot is calculated. 2 When the coefficient of determination was calculated, the coefficient of determination when "thickness of the casting" was adopted as the explanatory variable was R 2 =0.3236, the coefficient of determination when "the temperature of the casting at the time of removal from the mold" is used as the explanatory variable is R 2 = 0.5893. Here, the coefficient of determination (R) 2 This value represents how well the line fits the data; the closer this value is to 1, the better the approximation of the data.
[0046] Based on the above, it was found that by adopting "factors related to the amount of reinforcing elements dissolved in the solidified casting" as an explanatory variable, that is, "the temperature of the casting at the time of removal from the mold," the mechanical properties of casting C can be easily predicted with better accuracy than conventional methods.
[0047] [Example 2] In Example 2, as in Example 1, "0.2% yield strength" was used as the dependent variable, and "factors related to the amount of solid solubility of strengthening elements contained in the solidified casting," i.e., "the temperature of the casting at the time of removal from the mold," were adopted as the independent variables. On the other hand, ensemble learning was adopted as the predictive model.
[0048] Figure 6 shows the results of comparing the actual value and the predicted value of "0.2% yield strength" under these conditions. Here, Figure 6 is a graph comparing the predicted value and the actual value of the objective variable in Example 2. In the figure, sub-graph (a) is a graph comparing the predicted value obtained by linear regression with the actual value, and sub-graph (b) is a graph comparing the predicted value obtained by ensemble learning with the actual value.
[0049] In FIG. 6, the measured values with respect to the predicted values are plotted respectively. Also, the straight line represents the correspondence when the predicted value and the measured value match, and the closer the plot is to the straight line, the higher the prediction accuracy. Here, in order to examine the scatter of the plots with respect to the straight line in FIG. 6, that is, the prediction accuracy with respect to the measured value, the coefficient of determination (R 2 ) of the straight line for each plot was calculated. As shown in Example 1, the coefficient of determination in linear regression was R 2 = 0.5893, and the coefficient of determination in ensemble learning was R 2 = 0.8910.
[0050] From the above, when "the temperature of the casting at the time of taking out the casting from the mold" is adopted as an explanatory variable, it was confirmed that the prediction accuracy is further improved by adopting a prediction model based on ensemble learning instead of linear regression in the prediction model.
[0051] [Example 3] In Example 3, while adopting the same objective variable and prediction model as in Example 2, as explanatory variables, "factors related to the solid solution amount of strengthening elements contained in the casting after solidification" and "factors related to the growth of the solidification structure of the casting" were used. Specifically, "solidification rate" was adopted as "a factor related to the growth of the solidification structure of the casting". Also, "the temperature of the casting at the time of taking out the casting from the mold" was adopted as "a factor related to the solid solution amount of strengthening elements contained in the casting after solidification".
[0052] The results of comparing the actual performance values and predicted values of "0.2% proof stress" under such conditions are shown in FIG. 7. Here, FIG. 7 is a graph comparing the predicted values and measured values of the objective variable in Example 3. Sub - figure (a) in the figure is a graph comparing the predicted values obtained by linear regression with the measured values, and sub - figure (b) in the figure is a graph comparing the predicted values obtained by ensemble learning with the measured values.
[0053] In order to examine the scatter of the plots with respect to the straight line in FIG. 7, that is, the prediction accuracy with respect to the measured value, the coefficient of determination (R 2When we calculated the coefficient of determination in linear regression, it was found that R 2 =0.7445, the coefficient of determination in ensemble learning is R 2 The result was 0.9118. From the above, it was found that by adding "factors related to the growth of the solidification structure of the casting," specifically "solidification rate," to the explanatory variables in Example 2, the prediction accuracy improved in all prediction models.
[0054] [Example 4] Next, in Example 4, the same objective variable and predictive model as in Example 3 were adopted, while the explanatory variables used were "factors related to the amount of solid solubility of reinforcing elements contained in the casting after solidification," "factors related to the growth of the solidification structure of the casting," and "change in wall thickness of the casting." Specifically, "temperature of the casting at the time of removal from the mold" and "temperature of the molten metal at the completion of filling" were adopted as "factors related to the growth of the solidification structure of the casting," and "solidification rate" was adopted as "factors related to the amount of solid solubility of reinforcing elements contained in the casting after solidification."
[0055] Figure 8 shows the results of comparing the actual value and the predicted value of "0.2% yield strength" under these conditions. Here, Figure 8 is a graph comparing the predicted value and the actual value of the objective variable in Example 4. In the figure, sub-graph (a) is a graph comparing the predicted value obtained by linear regression with the actual value, and sub-graph (b) is a graph comparing the predicted value obtained by ensemble learning with the actual value.
[0056] To examine the variability of the plots relative to the straight line in Figure 8, i.e., the prediction accuracy relative to the measured values, the coefficient of determination (R) of the straight line for each plot was calculated. 2 When we calculated the coefficient of determination in linear regression, it was found that R 2 =0.9147, the coefficient of determination in ensemble learning is R 2 The result was 0.952117. From the above, it was found that by adding "the temperature of the molten metal at the completion of filling" and "the change in the wall thickness of the casting" from the "factors related to the amount of solid solution of strengthening elements contained in the casting after solidification" to the explanatory variables in Example 3, the prediction accuracy was improved in all prediction models.
[0057] [Example 5] Next, in Example 5, the same explanatory variables and predictive model as in Example 4 were adopted, but "tensile strength" was adopted as the dependent variable.
[0058] Figure 9 shows the results of comparing the actual and predicted values of "tensile strength" under these conditions. Here, Figure 9 is a graph comparing the predicted and measured values of the objective variable in Example 5. In the figure, sub-graph (a) is a graph comparing the predicted values obtained by linear regression with the measured values, and sub-graph (b) is a graph comparing the predicted values obtained by ensemble learning with the measured values.
[0059] To examine the variability relative to the straight line in Figure 9, i.e., the prediction accuracy relative to the measured values, the coefficient of determination (R) of the straight line for the plotted prediction results is calculated. 2 When we calculated the coefficient of determination in linear regression, it was found that R 2 =0.32131, the coefficient of determination in ensemble learning is R 2 The result was 0.78045. Therefore, when "tensile strength" is used as the target variable, linear regression does not yield good prediction accuracy, but ensemble learning does.
[0060] [Example 6] Next, in Example 6, the same explanatory variables and predictive model as in Example 5 were adopted, but "elongation at break" was adopted as the dependent variable.
[0061] Figure 10 shows the results of comparing the actual and predicted values of "elongation at break" under these conditions. Here, Figure 10 is a graph comparing the predicted and measured values of the objective variable in Example 6. In the figure, sub-graph (a) is a graph comparing the predicted value obtained by linear regression with the measured value, and sub-graph (b) is a graph comparing the predicted value obtained by ensemble learning with the measured value.
[0062] To examine the variability of the plots relative to the straight line in Figure 10, i.e., the prediction accuracy relative to the measured values, the coefficient of determination (R) of the straight line for each plot was calculated. 2 When we calculated the coefficient of determination in linear regression, it was found that R 2 =0.23164, the coefficient of determination in ensemble learning is R 2 The result was 0.72687. Therefore, it was found that when "break elongation" is used as the target variable, the prediction accuracy is not good in linear regression, but the target variable can be predicted well in ensemble learning.
[0063] [Example 7] Next, in Example 7, the same objective variable and predictive model as in Example 6 were adopted, while the explanatory variables used were "factors related to the amount of solid solubility of reinforcing elements contained in the casting after solidification," "factors related to the growth of the solidification structure of the casting," "change in wall thickness of the casting," and "factors related to the cleanliness of the molten metal." Specifically, "temperature of the casting at the time of removal from the mold" and "temperature of the molten metal at the completion of filling" were adopted as "factors related to the amount of solid solubility of reinforcing elements contained in the casting after solidification," "solidification rate" as "factors related to the growth of the solidification structure of the casting," and "contact time with air in the molten metal" as "factors related to the cleanliness of the molten metal."
[0064] Figure 11 shows the results of comparing the actual and predicted values of "elongation at break" under these conditions. Here, Figure 11 is a graph comparing the predicted values obtained by ensemble learning with the measured values for the objective variable in Example 7.
[0065] To examine the variability of the plots relative to the straight line in Figure 11, i.e., the prediction accuracy relative to the measured values, the coefficient of determination (R) of the straight line for each plot was calculated. 2 When we calculated the coefficient of determination, it was found that R 2 The result was 0.8143. Therefore, it was found that adding a "factor related to the cleanliness of the molten metal," specifically "contact time with air in the molten metal," to the explanatory variables in Example 6 improved the prediction accuracy in the prediction model using ensemble learning.
[0066] [Example 8] Next, in Example 8, the same objective variable and predictive model as in Example 6 were adopted, while the explanatory variables used were "factors related to the amount of solid solution of reinforcing elements contained in the casting after solidification," "factors related to the growth of the solidification structure of the casting," "change in wall thickness of the casting," and "solid phase ratio during molten metal flow." Specifically, "temperature of the casting at the time of removal from the mold" and "temperature of the molten metal at completion of filling" were adopted as "factors related to the amount of solid solution of reinforcing elements contained in the casting after solidification," and "solidification rate" was adopted as "factors related to the growth of the solidification structure of the casting."
[0067] Figure 12 shows the results of comparing the actual and predicted values of "elongation at break" under these conditions. Here, Figure 12 is a graph comparing the predicted and measured values of the objective variable in Example 8. In the figure, sub-graph (a) is a graph comparing the predicted value obtained by linear regression with the measured value, and sub-graph (b) is a graph comparing the predicted value obtained by ensemble learning with the measured value.
[0068] To examine the variability of the plots relative to the straight line in Figure 12, i.e., the prediction accuracy relative to the measured values, the coefficient of determination (R) of the straight line for each plot was calculated. 2 When we calculated the coefficient of determination in linear regression, it was found that R 2 =0.7720, the coefficient of determination in ensemble learning is R 2 = 0.87732. From the above, it was found that adding "solid fraction during molten metal flow" to the explanatory variables in Example 6 improved the prediction accuracy in all prediction models. In particular, as shown in Example 6, when the prediction model was linear regression, the coefficient of determination before adding "solid fraction during molten metal flow" to the explanatory variables was R 2 While the result was previously 0.23164, adding this explanatory variable significantly improved the prediction accuracy.
[0069] [Example 9] From the results of Example 8, the inventors conceived that when the objective variable is "elongation at fracture," the "solid fraction during molten metal flow" is a particularly important explanatory variable. To verify this, Example 9 adopted the same objective variable and prediction model as Example 8, but used only the "solid fraction during molten metal flow" as the explanatory variable. Figure 13 shows the results of comparing the actual value and the predicted value of "elongation at fracture" under these conditions. Here, Figure 13 is a graph comparing the predicted value and the actual value of the objective variable in Example 9, with sub-figure (a) being a graph comparing the predicted value obtained by linear regression with the actual value, and sub-figure (b) being a graph comparing the predicted value obtained by ensemble learning with the actual value.
[0070] To examine the variability relative to the straight line in Figure 13, i.e., the prediction accuracy relative to the measured values, the coefficient of determination (R) of the straight line for the plotted prediction results is calculated. 2 When we calculated the coefficient of determination in linear regression, it was found that R 2 =0.6753, the coefficient of determination in ensemble learning is R 2 The result was 0.9081. From the above, it was found that when the dependent variable is "elongation at fracture", adopting only "solid fraction during molten metal flow" as an explanatory variable resulted in good prediction accuracy in all prediction models. In particular, in ensemble learning, it was found that adopting only "solid fraction during molten metal flow" improved prediction accuracy compared to adopting five explanatory variables: "temperature of the casting at the time of removal from the mold", "temperature of the molten metal at completion of filling", "solidification rate", "change in wall thickness of the casting", and "solid fraction during molten metal flow".
[0071] [Example 10] Furthermore, the inventors conceived the idea that "the amount of material components contained in the molten metal" is an important explanatory variable governing the solid solubility of the reinforcing elements. To verify this, they investigated the correlation between "the amount of material components contained in the molten metal" as the explanatory variable and "the 0.2% yield strength" as the dependent variable. The results are shown in Figures 14 and 15. Specifically, "the amount of material components contained in the molten metal" was defined as "the content of Mg and Si." Figure 14 is a graph showing the correlation between the content of Mg and Si and the 0.2% yield strength. Figure 15 is a graph comparing the predicted value obtained by linear regression with the measured value for the dependent variable in Example 10.
[0072] Referring to Figure 14, when the weight percentage concentration of Mg in the molten metal is kept constant, the 0.2% yield strength increases as the weight percentage concentration of Si in the molten metal increases. Also, when the weight percentage concentration of Si is kept constant, the 0.2% yield strength increases as the weight percentage concentration of Mg increases. Therefore, a correlation can be observed between the weight percentage concentrations of Mg and Si and the 0.2% yield strength. In other words, a correlation can be observed between the amount of material components contained in the molten metal and the mechanical properties of casting C. This is thought to be because as Mg and Si increase, the amount of solid solution such as Mg2Si in the aluminum base material constituting the molten metal increases, thereby improving the mechanical properties of casting C.
[0073] To examine the variability of the plots relative to the straight line in Figure 15, i.e., the prediction accuracy relative to the measured values, the coefficient of determination (R) of the straight line for each plot was calculated. 2 When we calculated R 2The result was 0.9284. From the above, it was found that when the dependent variable is "0.2% yield strength", good prediction accuracy can be obtained by using only "the amount of material components contained in the molten metal" as the explanatory variable. In Example 10, the mechanical properties of casting C were predicted using only "the amount of material components contained in the molten metal" as the explanatory variable, but the inventors have confirmed that the prediction accuracy can be further improved by combining it with other explanatory variables. For example, the prediction accuracy can be further improved by using one or all of the following as explanatory variables in addition to "the amount of material components contained in the molten metal": "the temperature of the casting at the time the casting is removed from the mold", "the temperature of the molten metal at the completion of filling", and "solidification rate".
[0074] Although preferred embodiments of the present invention have been described above, the technical scope of the present invention is not limited to the scope described in the above embodiments. Various modifications or improvements can be made to the above embodiments and examples.
[0075] For example, in the above embodiment, the mechanical properties of the casting C were predicted using up to five explanatory variables. However, the inventors have confirmed that good prediction results can be obtained even when using any combination of explanatory variables other than those shown in the above embodiment.
[0076] Furthermore, in the above embodiments, "0.2% yield strength," "tensile strength," and "elongation at break" were used as the objective variables, but other mechanical properties may also be used as objective variables. For example, various mechanical properties can be predicted as long as they show a relationship with the explanatory variables, such as "shear stress," "hardness," and "density."
[0077] Furthermore, while linear regression and ensemble learning were used as prediction models in this embodiment, other methods may also be used in the present invention. For example, any known prediction method such as neural networks, support vector machines, or decision trees can be used.
[0078] It is clear from the claims that such modified or improved forms may also fall within the technical scope of the present invention. [Explanation of symbols]
[0079] 10 Mechanical property prediction device, 20 CAE analysis unit, 21 Mold model creation unit, 22 Explanatory variable acquisition unit, 30 Model prediction unit, 31 Prediction model acquisition unit, 32 Mechanical property prediction unit, 40 Computer, 41 Calculation unit, 42 Input device, 43 Main memory, 44 Auxiliary memory, 45 Output device, C Casting.
Claims
1. A method for predicting the mechanical properties of a casting formed by pouring molten aluminum alloy into a cavity formed in a pair of openable and closable molds and allowing the molten metal to solidify, wherein the mechanical properties of the casting are predicted. An explanatory variable acquisition step involves performing a predetermined analysis using a mold model for CAE analysis which is divided into multiple elements, thereby obtaining explanatory variables that are information relating to the casting conditions of the casting or the molten metal. A mechanical properties prediction step involves inputting the explanatory variables into a prediction model that calculates the correspondence between the explanatory variables and the target variable, which is information regarding the mechanical properties of the casting, in order to predict the target variable. Includes, The explanatory variables used are the temperature of the casting at the time of removal from the mold, the temperature of the molten metal at the completion of filling, the solidification rate of the casting, and the change in wall thickness of the casting. The mechanical property prediction method is characterized in that the prediction model is obtained from linear regression or ensemble learning, which generates a single learning model by combining multiple learning models.
2. A method for predicting mechanical properties according to claim 1, A method for predicting mechanical properties, characterized by adding a factor related to the cleanliness of the molten metal to the explanatory variables.
3. A mechanical properties prediction device that predicts the mechanical properties of a casting formed by pouring molten aluminum alloy into a cavity formed in a pair of openable and closable molds and allowing the molten metal to solidify, An explanatory variable acquisition unit obtains explanatory variables, which are information relating to the casting conditions of the casting or the molten metal, by performing a predetermined analysis using a mold model for CAE analysis which is divided into multiple elements of the cavity. A mechanical properties prediction unit that inputs the explanatory variables into a prediction model that calculates the correspondence between the explanatory variables and the target variable, which is information regarding the mechanical properties of the casting, and predicts the target variable, Equipped with, The explanatory variables used are the temperature of the casting at the time of removal from the mold, the temperature of the molten metal at the completion of filling, the solidification rate of the casting, and the change in wall thickness of the casting. The mechanical property prediction device is characterized in that the prediction model is obtained from linear regression or ensemble learning, which generates a single learning model by combining multiple learning models.