Composition output device, composition output method and composition output program

Abstract

To output information on a composition from a wider composition target.SOLUTION: A composition output device 1 includes a repeat part 11 for calculating an evaluation value on the basis of a plurality of characteristics of a composition of each of a plurality of compositions, and repeating and executing calculation extraction processing for extracting one or more compositions in which the calculated evaluation value satisfies prescribed reference, the repeat part 11 generating a plurality of compositions based on at least one of the one or more compositions extracted in the calculation extraction processing to execute the next calculation extraction processing about the plurality of generated compositions during repeating, and an output part 12 for outputting information on at least one of the one or more compositions extracted by the repeat part 11. The composition may be a glass fiber composition. The plurality of characteristics may contain a spinning characteristic and two or more characteristics other than the spinning characteristic.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] One aspect of the present disclosure relates to a composition output device, a composition output method, and a composition output program that output information about a composition extracted from a plurality of compositions. [Background technology]

[0002] Patent Document 1 below discloses a physical property prediction system that, when a glass composition and specifications for each of a plurality of physical properties are input, predicts a value indicating the degree to which the plurality of physical properties satisfy the specifications as a predicted value, including uncertainty, based on the learning results of learning the relationship between the glass composition and the plurality of physical properties. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent Publication No. 2022-065466 Summary of the Invention [Problem to be solved by the invention]

[0004] The above-mentioned physical property prediction system determines a composition that satisfies specifications from the composition of the glass to be predicted. In other words, the system targets a predetermined composition of the glass to be predicted, and the target is limited. Therefore, it is desirable to output information on compositions from a wider range of target compositions. [Means for solving the problem]

[0005] A composition output device according to one aspect of the present disclosure includes an iterator that repeatedly executes a calculation / extraction process to calculate an evaluation value for each of a plurality of compositions based on a plurality of characteristics of the composition and extract one or more compositions for which the calculated evaluation value satisfies a predetermined standard, the iterator generating a plurality of compositions based on at least one of the one or more compositions extracted in the previous calculation / extraction process and executing a next calculation / extraction process for the generated plurality of compositions, and an output unit that outputs information about at least one of the one or more compositions extracted by the iterator. In this aspect, the calculation / extraction process is repeated to execute the next calculation / extraction process for the plurality of compositions generated based on at least one of the one or more compositions extracted in the previous calculation / extraction process. This makes it possible to output information about compositions from a wider range of target compositions.

[0006] In the composition output device according to one aspect of the present disclosure, the composition may be a glass fiber composition. In such an aspect, information regarding the glass fiber composition from a plurality of glass fiber compositions can be output.

[0007] In the composition output device according to one aspect of the present disclosure, the plurality of characteristics may include spinning characteristics and two or more characteristics other than the spinning characteristics. In such an aspect, since the evaluation value can be calculated based on the spinning characteristics and two or more characteristics other than the spinning characteristics, it is possible to calculate, for example, a more accurate evaluation value.

[0008] In the composition output device according to one aspect of the present disclosure, each of the generated compositions may be one in which the proportion of at least one component is changed for any of the one or more compositions extracted in the previous calculation / extraction process. In such an aspect, it is possible to search for a composition with a higher evaluation value, for example.

[0009] In the composition output device according to one aspect of the present disclosure, the iterator may stop the iteration when the highest evaluation value calculated in the calculation-extraction process does not change from the highest evaluation value calculated in the calculation-extraction process preceding the calculation-extraction process. In such an aspect, for example, the iteration can be stopped when the evaluation value remains high, thereby reducing the processing amount and processing time.

[0010] In the composition output device according to one aspect of the present disclosure, a model may be used when calculating the evaluation value based on the plurality of characteristics, and the model may reflect experimental results for the composition indicated by the information output by the output unit. In such an aspect, for example, the evaluation value is calculated using a model that reflects actual experimental results for the composition indicated by the output information, so that a more accurate evaluation value that is in line with actual conditions can be calculated.

[0011] In the composition output device according to one aspect of the present disclosure, the iterator may perform the calculation and extraction process multiple times, using the same weights in the same iteration when calculating the evaluation value based on the multiple characteristics, and using different weights in different iterations. In such an aspect, since multiple iterations using different weights are performed, it is possible to output information about compositions from a wider range of compositions and based on evaluation values ​​using different calculation criteria, for example.

[0012] In the composition output device according to one aspect of the present disclosure, the output unit may output information about one or more compositions for which the evaluation value calculated by the iterator satisfies a criterion set by a user. In such an aspect, for example, information about a composition desired by the user can be output.

[0013] A composition output device according to one aspect of the present disclosure can also be described as follows. [1] an iterative unit that repeatedly executes a calculation and extraction process that calculates an evaluation value for each of a plurality of compositions based on a plurality of characteristics of the composition and extracts one or more compositions whose calculated evaluation value satisfies a predetermined standard, wherein, during the repetition, the iterative unit generates a plurality of compositions based on at least one of the one or more compositions extracted in the previous calculation and extraction process, and executes a next calculation and extraction process for the generated plurality of compositions; an output unit that outputs information about at least one of the one or more compositions extracted by the iterative unit; A composition output device comprising: [2] The composition is a glass fiber composition. [1] The composition output device according to the present invention. [3] The plurality of properties include spinning properties and two or more properties other than spinning properties, The composition output device according to [1] or [2]. [4] Each of the generated compositions is obtained by changing the ratio of at least one component of one or more of the compositions extracted in the previous calculation and extraction process. The composition output device according to any one of [1] to [3]. [5] the iterating unit stops the iteration when the highest evaluation value calculated in the calculation and extraction process does not vary from the highest evaluation value calculated in the calculation and extraction process preceding the calculation and extraction process. The composition output device according to any one of [1] to [4]. [6] a model is used when calculating the evaluation value based on the plurality of characteristics; The model reflects experimental results for the composition indicated by the information output by the output unit. The composition output device according to any one of [1] to [5]. [7] the iterator performs a plurality of iterations of the calculation and extraction process, and uses the same weights in the same iteration when calculating the evaluation value based on the plurality of characteristics, and uses different weights in different iterations. The composition output device according to any one of [1] to [6]. [8] the output unit outputs information about one or more compositions whose evaluation values ​​calculated by the iterative unit satisfy a criterion set by a user. The composition output device according to any one of [1] to [7]. [Effects of the Invention]

[0014] According to one aspect of the present disclosure, information about compositions from a broader range of compositions can be output. [Brief explanation of the drawings]

[0015] [Figure 1] FIG. 2 is a diagram illustrating an example of a functional configuration of a composition output device according to an embodiment. [Figure 2] FIG. 2 is a diagram illustrating an example of the hardware configuration of a computer used in the composition output device according to the embodiment. [Figure 3] FIG. 2 is a diagram showing the configuration of a composition output program according to an embodiment. [Figure 4] 4 is a flowchart showing an example of a process executed by the composition output device according to the embodiment. [Figure 5] FIG. 10 is a diagram showing an example table of data (part 1) relating to a plurality of glass fiber compositions in the first round of one trial. [Figure 6] FIG. 10 is a diagram showing an example table of data (part 2) relating to a plurality of glass fiber compositions in the first round of one trial. [Figure 7] FIG. 10 is a diagram showing an example table of data (part 3) relating to a plurality of glass fiber compositions in the first round of one trial. [Figure 8] FIG. 10 is a diagram showing an example table of data (part 4) relating to a plurality of glass fiber compositions in the first round of one trial. [Figure 9] FIG. 10 is a diagram showing an example of a table of data relating to glass fiber composition extracted in the first round of one trial. [Figure 10]FIG. 10 is a diagram showing an example table of data (part 1) relating to a plurality of glass fiber compositions in the second round of one trial. [Figure 11] FIG. 10 is a diagram showing an example table of data (part 2) relating to a plurality of glass fiber compositions in the second round of the first trial. [Figure 12] FIG. 10 is a diagram showing an example table of data (part 3) relating to a plurality of glass fiber compositions in the second round of the first trial. [Figure 13] FIG. 10 is a diagram showing an example table of data (part 4) relating to a plurality of glass fiber compositions in the second round of the first trial. [Figure 14] FIG. 10 is a diagram showing an example of a table of data relating to glass fiber composition extracted in the second round of one trial. [Figure 15] FIG. 10 is a diagram showing an example table of data (part 1) relating to a plurality of glass fiber compositions in the first round of the second trial. [Figure 16] FIG. 10 is a diagram showing an example table of data (part 2) relating to a plurality of glass fiber compositions in the first round of the second trial. [Figure 17] FIG. 10 is a diagram showing an example table of data (part 3) relating to a plurality of glass fiber compositions in the first round of the second trial. [Figure 18] FIG. 10 is a diagram showing an example table of data (part 4) relating to a plurality of glass fiber compositions in the first round of the second trial. [Figure 19] FIG. 10 is a diagram showing an example of a table of data relating to glass fiber composition extracted in the first round of two tries. [Figure 20] FIG. 10 is a diagram showing an example table of data (part 1) relating to a plurality of glass fiber compositions in the second round of the second trial. [Figure 21] FIG. 10 is a diagram showing an example table of data (part 2) relating to a plurality of glass fiber compositions in the second round of the second trial. [Figure 22] FIG. 10 is a diagram showing an example table of data (part 3) relating to a plurality of glass fiber compositions in the second round of the second trial. [Figure 23] FIG. 10 is a diagram showing an example table of data (part 4) relating to a plurality of glass fiber compositions in the second round of the second trial. [Figure 24]FIG. 10 is a diagram showing an example of a table of data relating to glass fiber composition extracted in the second round of two tries. [Figure 25] FIG. 10 is a diagram showing an example (part 1) of plotting the results of principal component analysis of a comparative example and a target example. [Figure 26] FIG. 10 is a diagram showing a plot example (part 2) of the results of principal component analysis of a comparative example and a target example. [Figure 27] FIG. 10 is a diagram showing an example of plotting the results of principal component analysis of the old model and the new model. DETAILED DESCRIPTION OF THE INVENTION

[0016] Hereinafter, embodiments of the present disclosure will be described in detail with reference to the drawings. In the description of the drawings, the same elements are designated by the same reference numerals, and duplicate explanations will be omitted. Furthermore, the embodiments of the present disclosure in the following description are specific examples of the present invention, and the present invention is not limited to these embodiments unless otherwise specified to limit the present invention.

[0017] 1 is a diagram showing an example of the functional configuration of a composition output device 1 (composition output device) according to an embodiment. The composition output device 1 is a computer device that outputs information about a composition extracted from a plurality of compositions.

[0018] As shown in FIG. 1, the composition output device 1 includes a storage unit 10, a repeating unit 11 (repeating unit), and an output unit 12 (output unit).

[0019] Each functional block of the composition output device 1 is assumed to function within the composition output device 1, but is not limited to this. For example, some of the functional blocks of the composition output device 1 may function within a computer device different from the composition output device 1, and connected to the composition output device 1 via a network, while appropriately sending and receiving information with the composition output device 1. Furthermore, some functional blocks of the composition output device 1 may be omitted, multiple functional blocks may be integrated into one functional block, or one functional block may be separated into multiple functional blocks.

[0020] FIG. 2 is a diagram showing an example of the hardware configuration of a computer used in the composition output device 1. As shown in FIG. 2, the composition output device 1 is physically configured as a computer system including a CPU (Central Processing Unit) 100, which is a central processing unit (processor), a RAM (Random Access Memory) 101 and a ROM (Read Only Memory) 102, which are main storage devices, an input / output device 103 such as a keyboard, a microphone, and a display, a communication module 104, which is a data transmission / reception device, and an auxiliary storage device 105 such as a hard disk and an SSD (Solid State Drive). The CPU 100, the RAM 101, the ROM 102, the input / output device 103, the communication module 104, and the auxiliary storage device 105 may each be configured with multiple components. The functions of each functional block shown in FIG. 1 are realized by loading predetermined computer software onto hardware such as the CPU 100 and RAM 101 shown in FIG. 2, which operates the input / output device 103 and the communication module 104 under the control of the CPU 100 and reads and writes data from and to the RAM 101 and the auxiliary storage device 105.

[0021] Fig. 3 is a diagram showing the configuration of a composition output program P1 according to an embodiment. The composition output program P1 is a computer program for causing a computer to execute a series of processes by the composition output device 1. As shown in Fig. 3, the composition output program P1 is stored in a program storage area formed in the auxiliary storage device 105.

[0022] The composition output program P1 is configured with a storage module P10, a repetition module P11, and an output module P12. The functions realized by executing the storage module P10, the repetition module P11, and the output module P12 are similar to the functions of the storage unit 10, the repetition unit 11, and the output unit 12 of the composition output device 1 described above, respectively.

[0023] The composition output program P1 may be configured so that a part or all of it is transmitted via a transmission medium such as a communication line, and is received and stored (including installed) by another device. Furthermore, each module of the composition output program P1 may be installed on one of multiple computers, rather than on a single computer. In this case, the series of processes of the composition output program P1 described above are performed by a computer system consisting of the multiple computers.

[0024] Hereinafter, each function of the composition output device 1 shown in FIG. 1 will be described.

[0025] The storage unit 10 stores any information that is used or output in the processing of the composition output device 1. The storage unit 10 may store information calculated by each function of the composition output device 1. The information stored by the storage unit 10 may be referenced by each function of the composition output device 1 as appropriate.

[0026] The iterative unit 11 repeatedly executes a calculation and extraction process that calculates an evaluation value for each of a plurality of compositions based on a plurality of characteristics of the composition and extracts one or more compositions whose calculated evaluation values ​​satisfy a predetermined standard. During the iteration, the iterative unit 11 generates a plurality of compositions based on at least one of the one or more compositions extracted in the previous calculation and extraction process, and executes a next calculation and extraction process for the generated plurality of compositions.

[0027] The composition may be any composition (components constituting the compound etc. and the proportions of the amounts thereof), and may be a glass fiber composition.

[0028] The multiple properties may include spinning properties and two or more properties other than spinning properties. Examples of properties other than spinning properties include elastic modulus, tensile strength, CTE (Coefficient of Thermal Expansion), dielectric constant, dielectric loss tangent, and refractive index. Properties other than spinning properties may include properties for which the higher the value, the better (e.g., elastic modulus and tensile strength), properties for which the lower the value, the better (e.g., CTE and dielectric loss tangent), and properties for which a high value may be desirable or a low value may be desirable depending on the purpose (e.g., dielectric constant and refractive index). When a property for which a low value is desirable is included, the weight described below is a negative value.

[0029] Each of the generated multiple compositions may be obtained by changing the ratio of at least one component of one or more compositions extracted in the previous calculation and extraction process. The change in ratio may be a change below a predetermined threshold. The change may be random. The generated multiple compositions may include a composition that is the same as one or more compositions extracted in the previous calculation and extraction process.

[0030] The iterative unit 11 may stop the iteration when the highest evaluation value calculated in the calculation / extraction process does not change from the highest evaluation value calculated in the calculation / extraction process immediately preceding the calculation / extraction process. "No change" may mean that the rate of change is within a predetermined rate.

[0031] A model may be used when calculating the evaluation value based on multiple characteristics, and the model may reflect experimental results for the composition indicated by the information output by the output unit 12. The model may be a mathematical model or a trained model in machine learning. The experiment for the composition may be a chemical experiment using an actual composition.

[0032] The iteration unit 11 may perform multiple iterations (tries) of the calculation and extraction process, and may use the same weight when calculating evaluation values ​​based on multiple characteristics within the same iteration (trie), and may use different weights in different iterations (tries).

[0033] The output unit 12 outputs (suggests to the user) information relating to at least one of the one or more compositions extracted by the iterative unit 11.

[0034] When the iteration unit 11 performs multiple iterations (tries) of the calculation and extraction process, the output unit 12 may output information regarding at least one of the one or more compositions extracted by the iteration unit 11 in each iteration (trie).

[0035] The output unit 12 may output information about one or more compositions whose evaluation values ​​calculated by the iterative unit 11 satisfy a criterion set by a user. The criterion set by the user may be stored in advance by the storage unit 10.

[0036] Hereinafter, the repeating unit 11 and the output unit 12 will be described in detail with reference to FIG. 4 and FIGS. 5 to 24.

[0037] FIG. 4 is a flowchart showing an example of a process (composition output method) executed by the composition output device 1. First, the iterative unit 11 sets "1" as the trie T (step S1). Next, the iterative unit 11 sets a weight W (step S2). W may be set to a predetermined value or may be set randomly. Next, the iterative unit 11 sets "1" as the cycle L (step S3). Note that one cycle corresponds to one calculation and extraction process.

[0038] Next, the iterative unit 11 sets multiple compositions (step S3). If it is the first time in a cycle (L is "1"), new multiple compositions are set. The new multiple compositions may be set to predetermined ones or may be set randomly. New multiple compositions may also be set for the second or subsequent cycles. If it is the second or subsequent try (if T is "2" or more) and it is the first time in a cycle (L is "1"), the same multiple compositions that were set for the first time in the cycle (L is "1") in the first try (T is "1") may also be set.

[0039] Next, the iterative unit 11 calculates the calculated value of the evaluation characteristic for each of the multiple compositions set in step S3 (step S5). Here, the evaluation characteristic refers to a characteristic other than the spinning characteristic. The calculated value may be a numerical value indicating whether the evaluation characteristic is good or bad. The calculation method for the calculated value may be a method using existing technology, such as using a trained model.

[0040] Next, the iterative unit 11 calculates the calculated values ​​of the spinning characteristics for each of the multiple compositions set in step S3 (step S6). The calculated values ​​may be numerical values ​​indicating the quality of the spinning characteristics. The calculation method for the calculated values ​​may be a method using existing technology, such as using a trained model.

[0041] Next, the iterative unit 11 calculates a score for each of the multiple compositions set in step S3 based on W (step S7). More specifically, the iterative unit 11 calculates a score based on the calculated values ​​of the evaluation properties calculated in step S5, the calculated values ​​of the spinning properties calculated in step S6, and W. More specifically, the iterative unit 11 calculates a score by weighting the calculated values ​​of the evaluation properties calculated in step S5 and the calculated values ​​of the spinning properties calculated in step S6 by W. In step S7, the iterative unit 11 further calculates the highest score among the calculated scores. In step S7, the iterative unit 11 may extract one or more compositions whose scores satisfy a predetermined standard.

[0042] Next, the iterative unit 11 determines whether L is "1" (L==1) and whether the highest score calculated in step S7 is different from the previous highest score using an OR condition (step S8). Here, the previous highest score is a variable set in step S9 described later, and is irrelevant when L is "1".

[0043] If it is determined in step S8 that L is "1" or if it is determined that the highest score calculated in step S7 varies from the previous highest score (S8: YES), the iterative unit 11 sets the highest score calculated in step S7 as the previous highest score (step S9). Next, the iterative unit 11 increments L (plus 1) (step S10) and returns to S4 (performs the next round). Note that when returning to S4, the iterative unit 11 may generate multiple compositions based on at least one of the one or more compositions extracted in step S7 of the previous (immediately preceding) round, and set the generated multiple compositions in S4.

[0044] If it is determined in step S8 that L is not "1" and that the highest score calculated in step S7 has not changed from the previous highest score (S8: NO), the iterative unit 11 sets the highest score as the previous highest score (step S11). Next, the iterative unit 11 determines whether T is smaller than N (step S12).

[0045] If it is determined in step S12 that T is smaller than N (S12: YES), the iterative unit 11 increments T (plus 1) (step S13) and returns to S2 (performs the next try). Note that when returning to S2, the W set in S2 may be different from the W set in the previous try.

[0046] If it is determined in step S12 that T is not smaller than N (S12: NO), the iterative unit 11 outputs information about the composition with the highest previous score for each T (or the composition extracted in step S7) (step S14).

[0047] Examples of data that are set or calculated in each process in the flowchart shown in FIG. 4 will be described with reference to FIGS.

[0048] 5 to 8 are diagrams showing examples of tables of data relating to a plurality of glass fiber compositions in one try (T is "1") and the first round (L is "1").

[0049] Fig. 5 shows composition 1-1, composition 1-2, composition 1-3, ..., composition 1-N as multiple compositions (initial input group) set in step S4 of Fig. 4. As shown in Fig. 5, for example, composition 1-1 is composed of 50% SiO2, 35% Al2O3, and 15% MgO.

[0050] 6 shows the calculated values ​​of the evaluation characteristics (evaluation characteristic 1, evaluation characteristic 2) calculated in step S5 of FIG. 4. Evaluation characteristic 1 and evaluation characteristic 2 for each of the multiple compositions may be calculated using a model (e.g., a trained statistical model). Note that the higher the evaluation characteristic 1, the higher the evaluation, and the lower the evaluation characteristic 2, the higher the evaluation.

[0051] Fig. 7 shows the calculated values ​​of the spinning characteristics calculated in step S6 of Fig. 4. In the example shown in Fig. 7, the spinning characteristics (operating temperature range (1000P temperature - liquidus temperature)) are calculated using a trained statistical model, and are evaluated as "1" (constant) if the temperature is -50°C or higher, and as "0" if the temperature is less than -50°C (parameter specific to glass fiber). Note that continuous values ​​may be used for the spinning characteristics.

[0052] Fig. 8 shows the scores (including the calculation formula) calculated in step S7 of Fig. 4. In the example shown in Fig. 8, when T is "1", the weight w11 for evaluation property 1 is set to "0.1", and the weight w12 for evaluation property 2 is set to "-1". For example, the score of composition 1-1 is calculated by summing the product of the calculated value of evaluation property 1, "85", and w11, "0.1", and the product of the calculated value of evaluation property 2, "2.0", and w12, "-1", and then multiplying this sum by the calculated value of the spinning property, "0", to obtain the product "0".

[0053] FIG. 9 is a diagram showing an example table of data related to glass fiber compositions extracted in the first round of one trial. In FIG. 9, only composition 1-2, which has the highest score of "6.2" among the multiple compositions shown in FIG. 8, is extracted and shown. In this way, the composition with the highest score (only one in the case of FIG. 9) may be extracted (selected). Also, the composition with the highest score ("SiO2 55%, Al2O3 30%, MgO 15%) may be recorded as the highest composition in the previous round.

[0054] 10 to 13 are diagrams showing examples of tables of data relating to a plurality of glass fiber compositions in the first try (T is "1") and the second round (L is "2").

[0055] The compositions shown in FIG. 10 were produced based on composition 1-2 shown in FIG. 9. For example, composition 2-1 in FIG. 10 is identical to composition 1-2 in FIG. 9. Composition 2-2 in FIG. 10 is obtained by increasing the SiO2 ratio by 1% and decreasing the Al2O3 ratio by 1% in composition 1-2 in FIG. 9. Composition 2-3 in FIG. 10 is obtained by increasing the SiO2 ratio by 1% and decreasing the MgO ratio by 1% in composition 1-2 in FIG. 9. Composition 2-N in FIG. 10 is obtained by decreasing the SiO2 ratio by 5%, increasing the Al2O3 ratio by 3%, and increasing the MgO ratio by 2% in composition 1-2 in FIG. 9.

[0056] 11 to 13 are similar to FIGS. 6 to 8, respectively, and therefore the description thereof will be omitted.

[0057] FIG. 14 shows an example table of data related to glass fiber compositions extracted in the second round of the first trial. In FIG. 14, only composition 2-2, which has the highest score of "6.3" among the multiple compositions shown in FIG. 13, is extracted and shown. It is also possible to extract two compositions, including composition 2-3, which also has the highest score. FIG. 14 also includes composition 1-2, which is the highest composition in the previous round. In FIG. 14, the highest score of this round (composition 2-2) can be compared with the highest composition in the previous round (composition 1-2). If the composition with the highest score is the same as the highest composition in the previous round, the trial can be terminated and this composition can be recorded as the final composition in the trial. If the composition with the highest score is different from the highest composition in the previous round, the highest composition in the previous round can be replaced with the composition with the highest score and the next round can be performed.

[0058] 15 to 18 are diagrams showing examples of tables of data relating to a plurality of glass fiber compositions in the first round (L is "1") of the second try (T is "2").

[0059] In FIG. 15, the same compositions as those in FIG. 5 for the first round of the first trial are shown as the multiple compositions set in step S4 of FIG.

[0060] 16 to 18 are similar to FIGS. 6 to 8, respectively, and therefore will not be described. However, in the second try, the weight w21 for evaluation characteristic 1 is set to "0.2," and the weight w22 for evaluation characteristic 2 is set to "-1." Therefore, the values ​​of the score column in FIG. 18 are different from the values ​​of the score column in FIG. 8.

[0061] Fig. 19 is a diagram showing an example of a table of data related to glass fiber compositions extracted in the first round of two trials. Fig. 19 shows only composition 1-2, which has the highest score of "14.2" among the multiple compositions shown in Fig. 18. The composition with the highest score may be recorded as the highest composition in the previous round.

[0062] 20 to 23 are diagrams showing example tables of data relating to a plurality of glass fiber compositions in the second try (T is "2") and the second round (L is "2"). Since FIGS. 20 to 23 are similar to FIGS. 10 to 13, respectively, their explanations are omitted. However, as described above, in the second try, the weight w21 for evaluation characteristic 1 is set to "0.2" and the weight w22 for evaluation characteristic 2 is set to "-1." Therefore, the values ​​of the score column in FIG. 23 differ from the values ​​of the score column in FIG. 13.

[0063] FIG. 24 is a diagram showing an example of a table of data related to glass fiber compositions extracted in the second round of the second trial. In FIG. 24, only composition 2-2, which has the highest score of "14.4" among the multiple compositions shown in FIG. 23, is extracted and shown. In addition, composition 1-2, which was the highest composition in the previous round, is also included in FIG. 14. The rest is the same as in FIG. 14.

[0064] 25 to 27, we will explain the results of actual experiments using the composition output device 1. The experimental results that did not use the composition output device 1 are referred to as "comparative examples," and the experimental results that used the composition output device 1 are referred to as "control examples."

[0065] Fig. 25 is a diagram showing a plot example (part 1) of the results of principal component analysis of the comparative example and the control example. In the plot example shown in Fig. 25, principal component analysis was performed on the entire composition output for each of the comparative example and the control example, and the first principal component is plotted on the horizontal axis and the second principal component is plotted on the vertical axis.

[0066] FIG. 26 is a diagram showing a plot example (part 2) of the results of principal component analysis of the comparative example and the target example. The plot example shown in FIG. 26 shows the plot examples shown in FIG. 25 with CTE prediction values ​​of less than 3. The plot example shown in FIG. 26 shows that the range of the target example is wide. This indicates that the composition output device 1 has covered the compositions proposed in the comparative example, while also finding composition systems that were not found in the comparative example.

[0067] FIG. 27 shows an example plot of the results of principal component analysis of the old model and the new model. As described above, the new model is a model that reflects the experimental results (experimental data) for the composition indicated by the information output by the output unit 12. On the other hand, the old model is a model that does not reflect the experimental results. According to the example plot shown in FIG. 27, the compositions output by the old model all have small predicted CTE values ​​(dark fill (small numerical values ​​indicated in the legend)), but when actually measured, there are cases where vitrification does not occur or the CTE is higher than predicted. On the other hand, the new model, which was created by repeating the experiment, picks out the areas with low CTE among the dense regions of the old model, and also concentrates the compositions in other composition systems.

[0068] Next, the effects of the composition output device 1 will be described.

[0069] The composition output device 1 includes an iterative unit 11 that repeatedly executes a calculation and extraction process to calculate an evaluation value for each of a plurality of compositions based on a plurality of characteristics of the composition and extract one or more compositions whose calculated evaluation value satisfies a predetermined standard, the iterative unit 11 generating a plurality of compositions based on at least one of the one or more compositions extracted in the previous calculation and extraction process and executing a next calculation and extraction process for the generated plurality of compositions, and an output unit 12 that outputs information about at least one of the one or more compositions extracted by the iterative unit 11. With this configuration, the calculation and extraction process is repeated to execute a next calculation and extraction process for the plurality of compositions generated based on at least one of the one or more compositions extracted in the previous calculation and extraction process. This makes it possible to output information about compositions from a wider range of target compositions.

[0070] The composition may be a glass fiber composition in the composition output device 1. With this configuration, it is possible to output information about the glass fiber composition from a plurality of glass fiber compositions.

[0071] In the composition output device 1, the multiple characteristics may include spinning characteristics and two or more characteristics other than the spinning characteristics. With this configuration, the evaluation value can be calculated based on the spinning characteristics and two or more characteristics other than the spinning characteristics, so that, for example, a more accurate evaluation value can be calculated.

[0072] Each of the multiple compositions generated by the composition output device 1 may be one in which the ratio of at least one component is changed from one or more compositions extracted in the previous calculation and extraction process. This configuration makes it possible to search for a composition with a higher evaluation value, for example.

[0073] In the composition output device 1, the iterative unit 11 may stop the iteration when the highest evaluation value calculated in the calculation and extraction process does not change from the highest evaluation value calculated in the calculation and extraction process preceding the calculation and extraction process. With this configuration, it is possible to stop the iteration when the evaluation value remains high, for example, thereby reducing the processing amount and processing time.

[0074] In the composition output device 1, a model may be used when calculating an evaluation value based on a plurality of characteristics, and the model may reflect experimental results for the composition indicated by the information output by the output unit 12. With this configuration, for example, the evaluation value is calculated using a model that reflects actual experimental results for the composition indicated by the output information, so that a more accurate evaluation value that is in line with the actual situation can be calculated.

[0075] In the composition output device 1, the iterative unit 11 may perform multiple iterations of the calculation and extraction process, using the same weight when calculating evaluation values ​​based on multiple characteristics within the same iteration, and different weights in different iterations. With this configuration, multiple iterations using different weights are performed, making it possible to output information about compositions from a wider range of compositions and based on evaluation values ​​using different calculation criteria, for example.

[0076] In the composition output device 1, the output unit 12 may output information about one or more compositions whose evaluation values ​​calculated by the iterating unit 11 satisfy a criterion set by a user. With this configuration, it is possible to output information about a composition desired by the user, for example.

[0077] Materials informatics (MI) is known as one of the methods for supporting materials creation. MI is a method for supporting materials creation using informatics techniques, and generally involves aggregating data on the structure and properties of materials and using machine learning to search for materials with new structures or properties.

[0078] However, the application of MI to glass fiber compositions is difficult for the following reasons. · There are many components (explanatory variables) that affect the characteristics. - There is little accumulated experience (compared to the number of components), and the training data is insufficient. - There is a bias in the physical property data measured depending on the type of composition.

[0079] Therefore, when MI is applied to glass fiber composition, especially when dealing with simultaneous optimization of multiple physical properties, the predictive accuracy of statistical learning models is insufficient and they tend to fall into local solutions.

[0080] Furthermore, in order to improve the prediction accuracy of the statistical learning model, it is possible to conduct experiments on the composition predicted by the statistical learning model and add experimental data. However, it is not easy to turn glass fiber compositions into glass fibers and conduct experiments on them, and it is necessary to add experimental data in a small amount to produce a large effect.

[0081] The composition output device 1 can search a wide range avoiding local solutions when simultaneously optimizing multiple properties, and propose a limited number of glass fiber compositions that are suitable for experimental verification to improve the performance of the statistical learning model, although not limited to these.

[0082] The method for proposing a glass fiber composition (composition output method) may be comprised of the following steps. Step (1): Generate a plurality of initial input populations (glass fiber compositions), Step (2): For multiple composition information, two or more evaluation target properties other than the spinning properties of each composition (e.g., elastic modulus, tensile strength, CTE, dielectric constant, dielectric loss tangent, refractive index) are calculated using a statistical learning model, and a calculated value A of each evaluation property is obtained. Step (3): Calculate the spinning properties of each composition using a statistical learning model to obtain a calculated value B; Step (4): Multiply the weighted average of the calculated value A by the calculated value B (for example, B × (W11 × A1 + W12 × A2)) to obtain an evaluation value for each composition (obtaining step); Step (5): Select multiple compositions with high evaluation values, and record the one with the highest evaluation value as the previous highest value (setting step). Step (6): Generate a new input population of multiple related compositions (m compositions) centered on the composition selected in the previous step (generation step); Step (7): Repeat steps (2) to (5) for the generated top-related compositions. Step (8): If the highest evaluation value does not change (within ~%) from the previous highest value, stop the first phase process (except for the first time). Step (9): In the above acquisition step, the weight for calculating the weighted average of the calculated value A is changed, and steps (2) to (8) are repeated N-1 times; Step (10): The final compositions of the first phase to the Nth phase are output (output step).

[0083] For the glass fiber composition proposed by the above-mentioned proposal method, the evaluation target properties and spinning properties may be experimented, the experimental values ​​may be added to the existing learning data, the statistical learning model may be updated, and the updated statistical learning model may be used to make a proposal by the above-mentioned proposal method. Also, in the composition search, a glass fiber composition that satisfies target values ​​or threshold values ​​or more for multiple properties set by the user may be proposed.

[0084] The method for proposing a glass fiber composition (composition output method) may be comprised of the following steps. [Phase 0] Step (1): Generate a plurality of initial input populations (glass fiber compositions). [Phase 1] Step (2): For the plurality of composition information, two or more evaluation target properties (the property of elastic modulus or tensile strength is designated as A1, and the CTE and dielectric property are designated as A2) other than the spinning properties of each of the plurality of compositions are calculated using a statistical learning model to obtain a calculated value A; Step (3): Calculate the spinning properties of each composition using a statistical learning model to obtain a calculated value B; Step (4): Obtain the evaluation value of each composition by B × (W11 × A1 + W12 × A2) (obtaining step); Step (5): Select one or more compositions with high evaluation values, and record the one with the highest evaluation value among them as the previous highest value (setting step); Step (6): Generate a new input population of multiple related compositions (m compositions) centered on the composition selected in the previous step (generation step); Step (7): Repeat steps (2) to (5) for the generated top-related compositions. Step (8): If the highest evaluation value does not change (within ~%) from the previous highest value, stop the first phase process (except for the first time). Step (9): After changing the weighting W of the evaluation characteristic (in this example, at least one of Wn1 and Wn2), steps (2) to (8) are repeated N-1 times. Step (10): The final compositions of the first phase to the Nth phase are output (output step).

[0085] The composition output device 1 can output information about a variety of predicted compositions by evaluating various indices while changing weights (parameters). The composition output device 1 can output information about better compositions by repeating the process of outputting information about compositions for actual experiments, conducting experiments, reflecting the experimental results in a model, and then outputting information about the compositions. The composition output device 1 can solve the problem of falling into a local solution by setting the multiple compositions generated during iteration. [Explanation of symbols]

[0086] 1...composition output device, 10...storage unit, 11...repetition unit, 12...output unit, 100...CPU, 101...RAM, 102...ROM, 103...input / output device, 104...communication module, 105...auxiliary storage device, P1...composition output program, P10...storage module, P11...repetition module, P12...output module.

Claims

1. an iterative unit that repeatedly executes a calculation and extraction process that calculates an evaluation value for each of a plurality of compositions based on a plurality of characteristics of the composition and extracts one or more compositions whose calculated evaluation value satisfies a predetermined standard, wherein, during the repetition, the iterative unit generates a plurality of compositions based on at least one of the one or more compositions extracted in the previous calculation and extraction process and executes a next calculation and extraction process for the generated plurality of compositions; an output unit that outputs information regarding at least one of the one or more compositions extracted by the iterative unit; A composition output device comprising:

2. The composition is a glass fiber composition. The composition output device according to claim 1 .

3. The plurality of properties include spinning properties and two or more properties other than spinning properties, The composition output device according to claim 2 .

4. Each of the generated compositions is obtained by changing the ratio of at least one component of one or more of the compositions extracted in the previous calculation and extraction process. The composition output device according to claim 1 .

5. the iterating unit stops the iteration when the highest evaluation value calculated in the calculation and extraction process does not vary from the highest evaluation value calculated in the calculation and extraction process preceding the calculation and extraction process. The composition output device according to claim 1 .

6. a model is used when calculating the evaluation value based on the plurality of characteristics; The model reflects experimental results for the composition indicated by the information output by the output unit. The composition output device according to claim 1 .

7. the iterator performs a plurality of iterations of the calculation and extraction process, and uses the same weights in the same iteration when calculating the evaluation value based on the plurality of characteristics, and uses different weights in different iterations. The composition output device according to claim 1 .

8. the output unit outputs information about one or more compositions whose evaluation values ​​calculated by the iterative unit satisfy a criterion set by a user. The composition output device according to claim 1 .

9. 1. A computer-implemented composition output method, comprising: an iterative step of repeatedly executing a calculation and extraction process in which an evaluation value is calculated for each of a plurality of compositions based on a plurality of characteristics of the composition, and one or more compositions for which the calculated evaluation value satisfies a predetermined standard are extracted, wherein, during the repetition, a plurality of compositions are generated based on at least one of the one or more compositions extracted in the previous calculation and extraction process, and a next calculation and extraction process is executed for the generated plurality of compositions; an output step of outputting information regarding at least one of the one or more compositions extracted in the iterative step; A composition output method including:

10. Computer, an iterative unit that repeatedly executes a calculation and extraction process that calculates an evaluation value for each of a plurality of compositions based on a plurality of characteristics of the composition and extracts one or more compositions whose calculated evaluation value satisfies a predetermined standard, wherein, during the repetition, the iterative unit generates a plurality of compositions based on at least one of the one or more compositions extracted in the previous calculation and extraction process and executes a next calculation and extraction process for the generated plurality of compositions; an output unit that outputs information regarding at least one of the one or more compositions extracted by the iterative unit; A composition output program to function as.

Images