Prediction device, prediction method, and prediction program

Abstract

The present invention makes it easier to understand the degree of influence exerted on the phase fraction in association with a change in the material composition. This prediction device: comprises a storage unit that stores a trained model having been trained by using learning data in which the material composition of a material to be learned is associated with the phase fraction of the material to be learned at each temperature in a prescribed temperature section; partially differentiates, when predicting the phase fraction of a material to be predicted at each of the temperatures in the prescribed temperature section by inputting the material composition of the material to be predicted to the trained model, values predicted for a predesignated phase at the respective temperatures by using the material composition of the material to be predicted; and displays, in a display mode according to the values obtained by the partial differentiation, respective regions identified on the basis of components and temperatures in a heat map in which the components of the material to be predicted are disposed on a first axis and the temperatures in the prescribed temperature section are disposed on a second axis.

Description

Prediction device, prediction method, and prediction program

[0001] The present disclosure relates to a prediction device, a prediction method, and a prediction program.

[0002] In the design of materials such as alloys, it is important to calculate the phase fraction in a thermodynamic equilibrium state over a specified temperature range. Therefore, surrogate models that predict the phase fraction based on material composition information (in the case of alloys, the weight percentage of added elements) are being developed.

[0003] Here, the phase fraction in a given temperature range correlates with the material properties. Therefore, when designing materials using the above model, it is important for the designer to understand in advance how to change the material composition to approach the desired phase fraction (in the case of an alloy, which phase in which temperature range can be affected by increasing or decreasing the weight percentage of which added element).

[0004] International Publication No. 2020 / 090617

[0005] In contrast to this, for example, by sequentially increasing or decreasing the weight percent of all additive elements, predicting the phase fraction within a predetermined temperature range, and comparing the predicted phase fraction with the phase fraction before the change, it is possible to grasp in advance the degree of influence of the weight percent of each additive element on the phase fraction. However, with such a method, in the case of an alloy or the like containing a large number of additive elements, the designer's workload increases and it is difficult for the designer to grasp the degree of influence on the phase fraction.

[0006] The present disclosure aims to make it easier to understand the degree of influence on phase fractions that accompanies changes in material composition.

[0007] A prediction device according to a first aspect of the present disclosure has a memory unit that stores a trained model that has been trained using training data in which the material composition of a material to be trained is associated with the phase fraction of the material to be trained at each temperature within a predetermined temperature range, and is configured to predict the phase fraction at the (i+1)th temperature using the phase fractions predicted by the trained model for temperatures up to the i-th temperature (i is an integer greater than or equal to 1) within the predetermined temperature range; when the phase fraction of the material to be predicted at each temperature within the predetermined temperature range is predicted by inputting the material composition of the material to be predicted into the trained model, the predicted value for a pre-specified phase at each temperature is partially differentiated using the material composition of the material to be predicted; and in a heat map in which each component of the material to be predicted is arranged on a first axis and each temperature within the predetermined temperature range is arranged on a second axis, each area identified based on each component and each temperature is displayed in a display manner corresponding to the value obtained by the partial differentiation.

[0008] A second aspect of the present disclosure is a prediction device according to the first aspect, wherein the trained model is applied with an architecture capable of processing time series data, which is data for each predetermined time interval, and the trained model predicts the phase fraction for each predetermined temperature interval based on the material composition of the material to be predicted.

[0009] A third aspect of the present disclosure is the prediction device according to the second aspect, wherein the trained model is any one of RNN, Bidirectional RNN, Seq2Seq, Seq2Seq with Attention mechanism, GRU, LSTM, or Transformer.

[0010] A fourth aspect of the present disclosure is the prediction device according to the third aspect, wherein the trained model has an encoder unit that outputs a feature quantity in response to input of a material composition of a material to be predicted, and a decoder unit that predicts the phase fraction at the (i+1)th temperature in response to input of the output feature quantity and the predicted phase fractions at the temperatures up to the i-th temperature.

[0011] A fifth aspect of the present disclosure is the prediction device according to any one of the first to fourth aspects, wherein the phase fraction is a phase fraction in a thermodynamic equilibrium state.

[0012] A sixth aspect of the present disclosure is the prediction device according to any one of the first to fifth aspects, wherein, when predicting the phase fraction at the (i+1)th temperature, the predicted value for the pre-specified phase is partially differentiated using a material composition of the material to be predicted, and each region in the heat map identified based on each component and the (i+1)th temperature is displayed in a display mode corresponding to the value obtained by the partial differentiation.

[0013] A prediction method according to a seventh aspect of the present disclosure is a prediction method in which a computer executes a trained model that has been trained using training data in which the material composition of a material to be trained is associated with the phase fraction of the material to be trained at each temperature within a predetermined temperature interval, and that is configured to predict the phase fraction at the (i+1)th temperature using the phase fractions predicted by the trained model for temperatures up to the i-th temperature (i is an integer greater than or equal to 1) within the predetermined temperature interval, wherein the computer executes the following steps: inputting the material composition of the material to be predicted into the trained model to predict the phase fraction of the material to be predicted at each temperature within the predetermined temperature interval, and partially differentiating the predicted value for a pre-specified phase at each temperature using the material composition of the material to be predicted; and displaying each region identified based on each component and each temperature in a heat map in which each component of the material to be predicted is arranged on a first axis and each temperature within the predetermined temperature interval is arranged on a second axis, in a display manner corresponding to the value obtained by the partial differentiation.

[0014] A prediction program according to an eighth aspect of the present disclosure is a prediction program for causing a computer to execute a trained model, the trained model being trained using training data in which the material composition of a training target material and the phase fraction of the training target material at each temperature within a predetermined temperature range are associated, and configured to predict the phase fraction at the (i+1)th temperature using the phase fractions predicted by the trained model for temperatures up to the i-th temperature (i is an integer greater than or equal to 1) within the predetermined temperature range. The prediction program causes the computer to execute the following steps: inputting the material composition of the training target material into the trained model to predict the phase fraction of the training target material at each temperature within the predetermined temperature range, and partially differentiating the predicted value for a pre-specified phase at each temperature using the material composition of the training target material; and displaying, in a heat map in which each component of the training target material is arranged on a first axis and each temperature within the predetermined temperature range is arranged on a second axis, each area identified based on each component and each temperature in a display manner corresponding to the value obtained by the partial differentiation.

[0015] According to the present disclosure, it is possible to easily grasp the degree of influence on the phase fraction due to a change in material composition.

[0016] FIG. 1 is a diagram illustrating an example of a system configuration of a prediction system and an example of a functional configuration of a learning device and a prediction device. FIG. 2 is a diagram illustrating an example of a hardware configuration of a learning device and a prediction device. FIG. 3 is a diagram illustrating a configuration of learning data and specific examples of input data and correct answer data. FIG. 4 is a diagram illustrating an example of a functional configuration of a learning data generation unit. FIG. 5 is a diagram illustrating an example of a functional configuration of a learning unit. FIG. 6 is a first diagram illustrating an example of an operation of the learning unit. FIG. 7 is a second diagram illustrating an example of an operation of the learning unit. FIG. 8 is a diagram illustrating an example of a method for calculating a loss by a loss function calculation unit. FIG. 9 is a flowchart illustrating the flow of a learning process. FIG. 10 is a diagram illustrating an example of a functional configuration of a prediction unit. FIG. 11A is a first diagram illustrating an example of an operation of the prediction unit. FIG. 11B is a second diagram illustrating an example of an operation of the prediction unit. FIG. 11C is a third diagram illustrating an example of an operation of the prediction unit. FIG. 12 is a flowchart illustrating the flow of a prediction process. FIG. 13 is a diagram illustrating an example of a predicted phase fraction. FIG. 14 is a diagram illustrating an example of a heat map. FIG. 15A is a first diagram illustrating an example of a verification result. Fig. 15B is a second diagram showing an example of the verification result. Fig. 15C is a third diagram showing an example of the verification result. Fig. 15D is a fourth diagram showing an example of the verification result. Fig. 15E is a fifth diagram showing an example of the verification result. Fig. 15F is a sixth diagram showing an example of the verification result. Fig. 15G is a seventh diagram showing an example of the verification result. Fig. 15H is an eighth diagram showing an example of the verification result.

[0017] Hereinafter, each embodiment will be described with reference to the accompanying drawings. In this specification and drawings, components having substantially the same functional configurations are designated by the same reference numerals, and redundant description will be omitted.

[0018] First Embodiment <System Configuration of Prediction System, Learning Device, and Functional Configuration of Prediction Device> First, a system configuration of a prediction system including a prediction device according to the first embodiment, and functional configurations of the learning device and prediction device included in the prediction system will be described. In this embodiment, the prediction system is a system that predicts phase fractions (phase fractions in a thermodynamic equilibrium state; hereinafter simply referred to as "phase fractions") at each temperature within a predetermined temperature interval based on material composition information.

[0019] 1 is a diagram illustrating an example of a system configuration of a prediction system and an example of a functional configuration of a learning device and a prediction device. As shown in FIG. 1, the prediction system 100 includes a learning device 110 and a prediction device 120.

[0020] A learning program is installed in the learning device 110, and by executing the program, the learning device 110 functions as a learning data generation unit 111 and a learning unit 112.

[0021] The learning data generation unit 111 generates learning data for training a prediction model (surrogate model) and stores the data in the learning data storage unit 113. In this embodiment, the learning data for training a prediction model is stored in the learning data storage unit 113 in association with each other, including: a plurality of different combinations of material composition information; and phase fractions at each temperature within a predetermined temperature range for materials having each of the material composition information.

[0022] The learning unit 112 reads out learning data from the learning data storage unit 113 and uses the read learning data to train the prediction model. The learning unit 112 trains the prediction model so that the output data when the material composition information stored in the learning data is input to the prediction model approaches the phase fraction at each temperature within a predetermined temperature interval, which is stored in association with the learning data. In this way, the learning unit 112 generates a trained prediction model. The learning unit 112 also stores the generated trained prediction model in the memory unit of the prediction unit 122 of the prediction device 120.

[0023] In this embodiment, the learning unit 112 applies, as a prediction model, the following architectures (architectures generally used for natural language processing, etc.) that can process time-series data, which is data at predetermined time intervals: Seq2Seq with Attention mechanism, or Transformer. As a result, according to this prediction model, by inputting material composition information stored in the learning data, it is possible to sequentially output output data corresponding to the phase fraction at each temperature within a predetermined temperature interval.

[0024] Here, "sequentially outputting output data corresponding to the phase fraction at each temperature" means that the prediction model outputs output data corresponding to the phase fraction at the (i+1)th temperature using the correct answer data stored in the learning data for the first i temperatures, where i is an integer equal to or greater than 1.

[0025] In this way, the learning unit 112 has a configuration for sequentially outputting output data corresponding to the phase fraction at each temperature (a configuration for processing the output data corresponding to the phase fraction at each temperature as time-series data), which allows the learning unit 112 to perform learning that reflects the output data corresponding to the phase fraction in adjacent temperature ranges.

[0026] A prediction program is installed in the prediction device 120 , and by executing this program, the prediction device 120 functions as a material composition input unit 121 , a prediction unit 122 , and a display unit 123 .

[0027] When material composition information of the material to be predicted is input, the material composition input unit 121 accepts it and notifies the prediction unit 122 of it.

[0028] The prediction unit 122 stores in the storage unit the trained prediction model trained by the learning unit 112. The prediction unit 122 also reads out the trained prediction model from the storage unit and inputs the material composition information notified by the material composition input unit 121 into the trained prediction model, thereby sequentially predicting the phase fraction at each temperature within a predetermined temperature interval. In other words, the prediction unit 122 sequentially predicts the phase fraction within a predetermined temperature interval for each predetermined temperature interval.

[0029] Here, "predicting the phase fraction sequentially at predetermined temperature intervals" means, for example, that the trained prediction model predicts the phase fraction at the (i+1)th temperature using the phase fractions predicted by the trained prediction model for temperatures up to the i-th temperature (i is an integer greater than or equal to 1).

[0030] As described above, in this embodiment, the prediction unit 122 has a configuration for sequentially predicting the phase fraction at each temperature (i.e., a configuration for processing the phase fraction at each temperature as time-series data using a recurrent network). This allows the prediction unit 122 to perform predictions that reflect the predicted data of the phase fraction in adjacent temperature ranges. As a result, a decrease in the accuracy of the phase fraction prediction can be suppressed compared to, for example, when a multilayer neural network is applied (when the predicted results of the phase fraction in adjacent temperature ranges are not reflected). In other words, the prediction unit 122 can improve the prediction accuracy when predicting the phase fraction over a predetermined temperature range based on the material composition.

[0031] Furthermore, when the prediction unit 122 predicts the phase fraction within a predetermined temperature interval in sequence for each predetermined temperature interval, it performs a process of partially differentiating the predicted value for a pre-specified phase at each temperature using the material composition of the material to be predicted. As a result, the prediction unit 122 can calculate the degree of change (amount of change) of the pre-specified phase at each temperature when the material composition is changed by a small amount.

[0032] Furthermore, in a heat map in which the horizontal axis (first axis) represents each component of the material to be predicted, and the vertical axis (second axis) represents each temperature within a predetermined temperature range, the prediction unit 122 colors each area identified based on each component and each temperature according to the value obtained by the partial differentiation.

[0033] The display unit 123 displays the phase fraction at each temperature within a predetermined temperature interval predicted by the prediction unit 122. The display unit 123 displays the phase fraction at each temperature within a predetermined temperature interval predicted by the prediction unit 122 in different colors for each phase. However, the phase fraction at each temperature within a predetermined temperature interval predicted by the prediction unit 122 may be displayed using different line types for each phase, such as a solid line, a dotted line, or a dashed line.

[0034] The display unit 123 also displays the heat map generated by the prediction unit 122. This allows the material designer to easily understand, based on the displayed heat map, how to change the material composition to bring the phase fraction closer to a desired phase fraction at each temperature within the displayed predetermined temperature range.

[0035] In this way, according to the prediction device 120 of the first embodiment, when a material designer predicts a phase fraction based on material composition information, the designer can easily grasp, based on the heat map, the degree of impact on the phase fraction due to a change in the material composition.

[0036] <Hardware Configuration of Learning Device and Prediction Device> Next, we will explain the hardware configuration of the learning device 110 and the prediction device 120. Note that the learning device 110 and the prediction device 120 have similar hardware configurations, so here, the hardware configurations of the learning device 110 and the prediction device 120 will be collectively explained using Figure 2.

[0037] 2 is a diagram showing an example of the hardware configuration of the learning device and the prediction device. As shown in Fig. 2, the learning device 110 and the prediction device 120 each include a processor 201, a memory 202, an auxiliary storage device 203, an I / F (Interface) device 204, a communication device 205, and a drive device 206. The hardware components of the learning device 110 and the prediction device 120 are connected to each other via a bus 207.

[0038] The processor 201 has various computing devices such as a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), etc. The processor 201 reads various programs (for example, a learning program, a prediction program, etc.) into the memory 202 and executes them.

[0039] The memory 202 has a main storage device such as a read-only memory (ROM) or a random access memory (RAM). The processor 201 and the memory 202 form a so-called computer, and the processor 201 executes various programs read onto the memory 202, causing the computer to realize the various functions described above.

[0040] The auxiliary storage device 203 stores various programs and various data used when the various programs are executed by the processor 201. For example, the learning data storage unit 113 is realized in the auxiliary storage device 203. Alternatively, a storage unit that stores a trained prediction model is realized in the auxiliary storage device 203.

[0041] The I / F device 204 is a connection device that connects to an operation device 211, which is an example of a user interface device, and a display device 212. The communication device 205 is a communication device for communicating with an external device (not shown) via a network.

[0042] The drive device 206 is a device for loading a recording medium 213. The recording medium 213 here includes media that record information optically, electrically, or magnetically, such as a CD-ROM, a flexible disk, a magneto-optical disk, etc. The recording medium 213 may also include semiconductor memory that records information electrically, such as a ROM, a flash memory, etc.

[0043] The various programs to be installed in the auxiliary storage device 203 are installed, for example, by setting the distributed recording medium 213 in the drive device 206 and reading the various programs recorded on the recording medium 213 by the drive device 206. Alternatively, the various programs to be installed in the auxiliary storage device 203 may be installed by being downloaded from a network via the communication device 205.

[0044] <Configuration of Training Data, Input Data, and Supervisory Data> Next, the configuration of training data and specific examples of input data and supervisory data included in the training data will be described. Fig. 3 is a diagram showing the configuration of training data and specific examples of input data and supervisory data.

[0045] As shown in FIG. 3, the learning data 300 includes information items such as "input data" and "correct answer data."

[0046] The "input data" stores different combinations of material composition information, such as "material composition information 1," "material composition information 2," etc. In Fig. 3, reference numeral 301 indicates a specific example of "material composition information 1," which represents an example of the weight percentage of each component (each added element Si, Fe, Cu, Mn, Mg, Cr, Zn, Ti) that constitutes "6013," which is the name of a known aluminum alloy standard.

[0047] Similarly, reference numeral 302 indicates a specific example of "material composition information 2," which indicates the weight percentage of each component (each added element: Si, Fe, Cu, Mn, Mg, Cr, Zn, Ti) that constitutes "6060," a known aluminum alloy standard designation.

[0048] The "correct answer data" stores "phase fraction 1," "phase fraction 2," etc., which are phase fractions at each temperature in a predetermined temperature range (100°C to 800°C in the example of Fig. 3), associated with "material composition information 1," "material composition information 2," etc. In Fig. 3, reference numeral 311 indicates a specific example of "phase fraction 1," and reference numeral 312 indicates a specific example of "phase fraction 2," with the horizontal axis representing temperature and the vertical axis representing phase fraction.

[0049] Here, reference numerals 311 and 312 are results obtained by operating a high-cost simulator with high prediction accuracy for a long period of time, and in this embodiment, these results are used as correct answer data.

[0050] The simulator referred to here refers to software that calculates the phase fraction at each temperature within a predetermined temperature interval through thermodynamic equilibrium calculations, and includes, for example, software such as CaTCalc (registered trademark) and MatCalc (registered trademark). Alternatively, the simulator may include software such as Termosuite (registered trademark), FactStage (registered trademark), and Pandat (registered trademark). Alternatively, the simulator may include software such as MALT2 (registered trademark), Thermo-Calc (registered trademark), and OpenCalchad (registered trademark).

[0051] <Details of Each Unit of the Learning Device> Next, the details of each unit of the learning device 110 (the learning data generation unit 111 and the learning unit 112) will be described.

[0052] (1) Functional Configuration of the Learning Data Generator First, a description will be given of the functional configuration of the learning data generator 111. Fig. 4 is a diagram showing an example of the functional configuration of the learning data generator.

[0053] As shown in FIG. 4, the learning data generation unit 111 includes an element information input unit 401 , a combination determination unit 402 , a simulation unit 403 , and a storage control unit 404 .

[0054] The element information input unit 401 receives input of, for example, the types of additive elements to be added when producing a specific alloy. The element information input unit 401 also receives input of upper and lower limits of the amount (wt %) of each additive element.

[0055] The combination determination unit 402 determines multiple combinations of the amounts of each additive element added by randomly selecting the amount of each additive element added under the constraints of the upper and lower limits for each input additive element. Note that the combinations of the amounts of each additive element added may be created in predetermined increments between the upper and lower limits. The combination determination unit 402 also notifies the simulation unit 403 of material composition information indicated by the multiple combinations determined.

[0056] The simulation unit 403 calculates the phase fraction at each temperature within a predetermined temperature range by operating the simulator described above for each of the plurality of pieces of material composition information notified by the combination determination unit 402. The simulation unit 403 also notifies the storage control unit 404 of the plurality of pieces of material composition information and the corresponding phase fraction at each temperature within the predetermined temperature range.

[0057] The storage control unit 404 generates learning data in which the multiple pieces of material composition information notified by the simulation unit 403 are used as "input data" and the phase fractions at each temperature within the corresponding specified temperature range are used as "correct data," and stores the data in the learning data storage unit 113.

[0058] (2) Functional Configuration of the Learning Unit Next, a description will be given of the functional configuration of the learning unit 112. Fig. 5 is a diagram showing an example of the functional configuration of the learning unit.

[0059] As shown in FIG. 5 , the learning unit 112 includes a prediction model (surrogate model) and a loss function calculation unit 503 .

[0060] As described above, the prediction model applies Seq2Seq or Transformer with an attention mechanism, and includes an encoder unit 501 and a decoder unit 502 .

[0061] The encoder unit 501 receives "material composition information 1," "material composition information 2," . . . stored in the "input data" of the training data 300, and outputs feature quantities.

[0062] The decoder unit 502 receives the feature values ​​output by the encoder unit 501 and sequentially outputs output data. Specifically, the decoder unit 502 reads "phase fraction 1," "phase fraction 2," ..., etc. stored in the "correct data" of the training data 300. The decoder unit 502 then uses the phase fractions at temperatures up to the i-th temperature (i.e., the correct data at temperatures up to the i-th temperature) to output output data at the (i+1)-th temperature. Thus, for example, the decoder unit 502 can sequentially output output data corresponding to the phase fractions at each temperature within a predetermined temperature interval based on the feature values ​​output by the encoder unit 501 and the "correct data" for "phase fraction 1" when "material composition information 1" is input. Similarly, for example, the decoder unit 502 can sequentially output output data corresponding to the phase fractions at each temperature within a predetermined temperature interval based on the feature values ​​output by the encoder unit 501 and the "correct data" for "phase fraction 2" when "material composition information 2" is input.

[0063] It is assumed that a softmax function is used in the output layer of the decoder unit 502. As a result, the multiple values ​​included in the output data of the decoder unit 502 (i.e., values ​​indicating the proportions of each phase) fall within the range of 0 to 1, and the sum of the multiple values ​​(i.e., the sum of the values ​​indicating the proportions of each phase) is "1."

[0064] In this way, by using a softmax function in the output layer of the decoder unit 502, output data corresponding to the phase fraction at each temperature within a specified temperature range can be output sequentially without performing any additional calculations.

[0065] The loss function calculation unit 503 reads out “phase fraction 1,” “phase fraction 2,” etc. stored in the “correct data” of the learning data 300, and compares it with the output data at the (i+1)th temperature output by the decoder unit 502.

[0066] Furthermore, the loss function calculation unit 503 calculates multiple types of losses during comparison and calculates an integrated loss by weighting and adding the calculated multiple types of losses. Furthermore, the loss function calculation unit 503 updates the model parameters of the encoder unit 501 and the decoder unit 502 based on the calculated integrated loss. This generates a trained prediction model (including a trained encoder unit and a trained decoder unit).

[0067] (3) Operation Example 1 of the Learning Unit Next, an operation example of the learning unit 112 (mainly an operation example of the prediction model among the prediction model and the loss function calculation unit) will be described. Fig. 6 is a first diagram showing an operation example of the learning unit.

[0068] 6A illustrates a state in which the "input data" of the training data 300 is read and input to the encoder unit 501, resulting in the output of features from the encoder unit 501. FIG. 6A also illustrates a state in which a start signal is input to the decoder unit 502, resulting in the decoder unit 502 outputting output data. In the example of FIG. 6A, the output data output by the decoder unit 502 corresponds to the phase fraction at 800°C. The output data output by the decoder unit 502 (here, the output data corresponding to the phase fraction at 800°C) is notified to the loss function calculation unit 503. The start signal is a signal for starting the operation of the decoder unit 502. In this embodiment, an arbitrary value (e.g., "000000") different from the output data output from the decoder unit 502 is input to the decoder unit 502 as the start signal.

[0069] 6(b) shows a state in which the feature quantities output from the encoder unit 501 are input to the decoder unit 502, and the phase fraction at 800°C (correct data at 800°C) is input to the decoder unit 502, resulting in the decoder unit 502 outputting output data. In the example of FIG. 6(b), the output data output by the decoder unit 502 corresponds to the phase fraction at 790°C. The output data output from the decoder unit 502 (here, the output data corresponding to the phase fraction at 790°C) is notified to the loss function calculation unit 503.

[0070] 6(c) shows how the decoder unit 502 outputs output data after the feature quantities output from the encoder unit 501 are input to the decoder unit 502 and the phase fractions up to 790°C (correct data up to 790°C) are input to the decoder unit 502. However, due to space limitations, only the correct data at 790°C is shown in FIG. 6(c) (in reality, the correct data at 800°C and the correct data at 790°C are weighted and input to the decoder unit 502).

[0071] 6(c), the output data output by the decoder unit 502 corresponds to the phase fraction at 780° C. The output data output by the decoder unit 502 (here, the output data corresponding to the phase fraction at 780° C.) is notified to the loss function calculation unit 503.

[0072] 6(d) shows how the decoder unit 502 outputs output data after the feature values ​​output from the encoder unit 501 are input to the decoder unit 502 and the phase fractions up to 120° C. (correct data up to 120° C.) are input to the decoder unit 502. However, due to space limitations, only the correct data at 120° C. is shown in FIG. 6(d) (in reality, each correct data from 800° C. to 120° C. is weighted and input to the decoder unit 502).

[0073] 6(d), the output data output by the decoder unit 502 corresponds to the phase fraction at 110° C. The output data output by the decoder unit 502 (here, the output data corresponding to the phase fraction at 110° C.) is notified to the loss function calculation unit 503.

[0074] 6(e) shows how the decoder unit 502 outputs output data after the feature values ​​output from the encoder unit 501 are input to the decoder unit 502 and the phase fractions up to 110°C (correct data up to 110°C) are input to the decoder unit 502. However, due to space limitations, only the correct data at 110°C is shown in FIG. 6(e) (in reality, each correct data from 800°C to 110°C is weighted and input to the decoder unit 502).

[0075] 6( e), the output data output by the decoder unit 502 corresponds to the phase fraction at 100° C. The output data output by the decoder unit 502 (here, the output data corresponding to the phase fraction at 100° C.) is notified to the loss function calculation unit 503.

[0076] (4) Operation Example 2 of the Learning Unit Next, an operation example of the learning unit 112 (mainly an operation example of the loss function calculation unit 503 among the prediction model and loss function calculation units) will be described. FIG. 7 is a second diagram showing an operation example of the learning unit. As shown in FIG. 7, the loss function calculation unit 503 reads out "correct data" from the learning data 300. The example in FIG. 7 shows how the following phase fractions at each temperature within a predetermined temperature range included in "phase fraction 1" are read out: Phase fraction at 800°C (correct data), Phase fraction at 790°C (correct data), ... Phase fraction at 100°C (correct data).

[0077] 7 shows an example in which output data is sequentially output from the decoder unit 502 based on the feature quantities output from the encoder unit 501 in response to input of "material composition information 1," and the loss function calculation unit 503 stores the following: Phase fraction at 800°C (output data), Phase fraction at 790°C (output data), ... Phase fraction at 100°C (output data).

[0078] 7 shows how the loss function calculation unit 503 calculates the integrated loss by comparing the phase fraction (correct data) at each temperature from 100°C to 800°C with the phase fraction (output data) at each temperature from 100°C to 800°C. Also, the example of FIG. 7 shows how the loss function calculation unit 503 updates the model parameters of the encoder unit 501 and the decoder unit 502 based on the calculated integrated loss. The functional configuration of the loss function calculation unit 503 for calculating the integrated loss will be described in detail below.

[0079] (5) Details of the Functional Configuration of the Loss Function Calculation Unit Fig. 8 is a diagram showing an example of a method for calculating losses by the loss function calculation unit. As shown in Fig. 8, the loss function calculation unit 503 has the following functions for calculating multiple types of losses: a phase fraction error calculation unit 801 at generation / disappearance temperatures, a phase fraction error calculation unit 802, a phase fraction error (logarithmic value) calculation unit 803, a phase fraction error (differential value) calculation unit 804, a cross entropy error calculation unit 805, and a weighted addition unit 806.

[0080] The phase fraction error calculation unit 801 at generation / disappearance temperatures compares the phase fraction (output data) at each temperature from 100°C to 800°C (correct data) for the temperatures at which the curves for each phase fraction become zero among the correct data of the learning data. The phase fraction error calculation unit 801 at generation / disappearance temperatures then adds up the errors, for all phases, between the phase fractions identified based on the output data and the phase fractions identified based on the learning data at the temperatures at which each phase identified based on the learning data is generated or disappeared. The phase fraction error calculation unit 801 at generation / disappearance temperatures then outputs the addition result (first addition result).

[0081] The phase fraction error calculation unit 802 compares the phase fraction (correct data) at each temperature from 100° C. to 800° C. with the phase fraction (output data) at each temperature from 100° C. to 800° C. As a result, the phase fraction error calculation unit 802 adds up the phase fraction errors at each temperature for a predetermined temperature interval and outputs the addition result (second addition result).

[0082] The phase fraction error (logarithmic value) calculation unit 803 compares the phase fraction (correct data) at each temperature from 100° C. to 800° C. with the phase fraction (output data) at each temperature from 100° C. to 800° C. As a result, the phase fraction error (logarithmic value) calculation unit 803 adds up the errors in the logarithmic values ​​of the phase fractions at each temperature within a predetermined temperature range, for the predetermined temperature range, and outputs the addition result (third addition result).

[0083] The phase fraction error (difference value) calculation unit 804 compares the phase fraction (correct data) at each temperature from 100° C. to 800° C. with the phase fraction (output data) at each temperature from 100° C. to 800° C. As a result, the phase fraction error (difference value) calculation unit 804 adds up the errors in the phase fraction difference values ​​between adjacent temperatures for a predetermined temperature interval, and outputs the addition result (fourth addition result).

[0084] The cross-entropy error calculation unit 805 compares the phase fraction (correct data) at each temperature from 100° C. to 800° C. with the phase fraction (output data) at each temperature from 100° C. to 800° C. As a result, the cross-entropy error calculation unit 805 adds up the errors in the phase fraction ratios at each temperature within a predetermined temperature range for the predetermined temperature range, and outputs the addition result (fifth addition result).

[0085] The weighted addition unit 806 calculates the integrated loss by weighting and adding the addition results (the first addition result to the fifth addition result) output from the phase fraction error calculation unit 801 to the cross entropy error calculation unit 805 at the generation / disappearance temperatures.

[0086] In this way, by calculating and weighting multiple types of losses, the loss function calculation unit 503 can process the loss between the output data and the correct data in a multifaceted manner. As a result, the learning unit 112 can appropriately update the model parameters when learning the prediction model.

[0087] <Learning Process Flow> Next, a description will be given of the flow of the learning process performed by the learning device 110. Fig. 9 is a flowchart showing the flow of the learning process.

[0088] In step S901, the learning data generation unit 111 receives input of element information, such as the type of additive element to be added when generating a specific alloy, and the upper and lower limits of the amount (by weight) of each additive element.

[0089] In step S902, the learning data generation unit 111 randomly selects the amount of each additive element added subject to the constraints of upper and lower limits, thereby determining multiple combinations of the amount of each additive element added (wt %).

[0090] In step S903, the learning data generating unit 111 operates a simulator for each of the material composition information indicated by the determined combinations, and calculates the phase fraction at each temperature within a predetermined temperature range.

[0091] In step S904 , the learning data generation unit 111 generates learning data and stores it in the learning data storage unit 113 .

[0092] In step S 905 , the learning unit 112 inputs a start signal to the decoder unit 502 .

[0093] In step S906, the learning unit 112 reads out the "input data" of the learning data and inputs it into the prediction model.

[0094] In step S907, the learning unit 112 sequentially inputs the phase fractions of the “correct data” of the learning data at temperatures up to the i-th temperature (i.e., correct data at temperatures up to the i-th temperature), and sequentially outputs the (i+1)-th output data.

[0095] In step S908, the learning unit 112 determines whether all output data corresponding to the phase fraction at each temperature within the predetermined temperature range have been output. If it is determined in step S908 that there is a temperature for which output data corresponding to the phase fraction has not been output (NO in step S908), the process returns to step S907.

[0096] On the other hand, if it is determined in step S908 that all output data corresponding to the phase fractions at each temperature within the predetermined temperature range have been output (YES in step S908), the process proceeds to step S909.

[0097] In step S909, the learning unit 112 reads out the "correct data" from the learning data and compares it with the output data.

[0098] In step S910, the learning unit 112 calculates multiple types of losses and calculates an integrated loss by weighting and adding the calculated multiple types of losses. The learning unit 112 also updates the model parameters of the encoder unit 501 and the decoder unit 502 based on the calculated integrated loss.

[0099] In step S911, the learning unit 112 determines whether to continue the learning process. If it is determined in step S911 that the learning process should be continued (YES in step S911), the process returns to step S906, and the same process is performed by reading the next "input data" of the learning data.

[0100] On the other hand, if it is determined in step S911 that the learning process is to be ended (NO in step S911), the process proceeds to step S912.

[0101] In step S912, the learning unit 112 stores the learned encoder unit and the learned decoder unit in the storage unit of the prediction device 120 as a learned prediction model.

[0102] <Details of Each Unit of the Prediction Device> Next, each unit of the prediction device 120 (here, the prediction unit 122) will be described in detail.

[0103] (1) Functional Configuration of the Prediction Unit First, a description will be given of the functional configuration of the prediction unit 122. Fig. 10 is a diagram illustrating an example of the functional configuration of the prediction unit.

[0104] 10, the prediction unit 122 includes a trained encoder unit 1001 and a trained decoder unit 1002 that have been trained by the training unit 112. The trained encoder unit 1001 and the trained decoder unit 1002 form a trained prediction model.

[0105] The trained encoder unit 1001 receives material composition information (material composition information of the material to be predicted) notified by the material composition input unit 121, calculates features, and outputs the calculated features to the trained decoder unit 1002.

[0106] The trained decoder unit 1002 receives the feature values ​​output from the trained encoder unit 1001 and sequentially outputs predicted data. Specifically, the trained decoder unit 1002 uses the predicted data at temperatures up to the i-th temperature to output predicted data at the (i+1)-th temperature. This enables the trained decoder unit 1002 to predict the phase fraction at each temperature within a predetermined temperature range based on the feature values ​​output from the trained encoder unit 1001 when material composition information of the material to be predicted is input.

[0107] 10, the prediction unit 122 further includes a phase designation unit 1003 and a sensitivity characteristic calculation unit 1004. The phase designation unit 1003 and the sensitivity characteristic calculation unit 1004 form an input sensitivity analysis unit.

[0108] The phase designation unit 1003 receives input of a phase designated by a material designer. The phase designated by the material designer refers to a phase for which the material designer wants to know in advance how much the phase will change at each temperature when the material composition is changed. The phase designation unit 1003 notifies the sensitivity characteristic calculation unit 1004 of the phase for which input has been received.

[0109] When the sensitivity characteristic calculation unit 1004 outputs the predicted data at the (i+1)th temperature, it partially differentiates the predicted value for the phase notified by the phase designation unit 1003 using the material composition information notified by the material composition input unit 121.

[0110] Specifically, the sensitivity characteristic calculation unit 1004 derives "a value obtained by partially differentiating the value at the i+1th temperature predicted for the specified phase using material composition information" by using: - partial differential values ​​at each layer calculated by processing each layer in the trained encoder unit 1001 from when material composition information is input until the feature is output; and - partial differential values ​​at each layer calculated by processing each layer in the trained decoder unit 1002 from when a feature is input until the value predicted for the specified phase at the i+1th temperature is output.

[0111] In addition, the sensitivity characteristic calculation unit 1004 notifies the display unit 123 (not shown in FIG. 10 ) of “the value obtained by partially differentiating the value at the (i+1)th temperature predicted for the specified phase using the material composition information.”

[0112] The display unit 123 colors the heat map in accordance with the value obtained by partially differentiating the region of the heat map that is specified based on each component (each added element) of the material to be predicted and the (i+1)th temperature. As a result, the display unit 123 performs these processes over a predetermined temperature range, thereby generating and displaying a heat map for the predetermined temperature range.

[0113] (2) Example of Operation of the Prediction Unit Next, an example of operation of the prediction unit 122 will be described. Figures 11A to 11C are first to third diagrams showing an example of operation of the prediction unit. Of these, (a) of Figure 11A shows how material composition information of the material to be predicted is input to the trained encoder unit 1001, resulting in feature values ​​being output from the trained encoder unit 1001. Also, (a) of Figure 11A shows how the feature values ​​output from the trained encoder unit 1001 are input to the trained decoder unit 1002, and how a start signal is input to the trained decoder unit 1002, resulting in predicted data (phase fraction at 800°C) being output.

[0114] Also, (a) of FIG. 11A shows how, when prediction data (phase fraction at 800°C) is output, "a value (partial differential value) obtained by partially differentiating the predicted value for the specified phase using the material composition information of the material to be predicted" is derived at 800°C.

[0115] 11A (b) shows a state in which the feature output from the trained encoder unit 1001 is input to the trained decoder unit 1002, and the predicted data (phase fraction at 800°C) is input to the trained decoder unit 1002. As a result, the trained decoder unit 1002 outputs the predicted data (phase fraction at 790°C).

[0116] Also, (b) of FIG. 11A shows how, when the prediction data (phase fraction at 790°C) is output, the "value (partial differential value) obtained by partially differentiating the predicted value for the specified phase using the material composition information of the material to be predicted" is derived at 790°C.

[0117] 11B(c) shows how the feature values ​​output from the trained encoder unit 1001 are input to the trained decoder unit 1002, and how the predicted data (phase fraction up to 790°C) is input to the trained decoder unit 1002. Note that due to space limitations, only the phase fraction at 790°C is shown in FIG. 11B(c) (in reality, the phase fractions at 800°C and 790°C are weighted and input to the trained decoder unit 1002). As a result, the trained decoder unit 1002 outputs the predicted data (phase fraction at 780°C).

[0118] Also, (c) of FIG. 11B shows how, when the prediction data (phase fraction at 780°C) is output, the "value (partial differential value) obtained by partially differentiating the predicted value for the specified phase using the material composition information of the material to be predicted" is derived at 780°C.

[0119] 11B(d) shows how the feature values ​​output from the trained encoder unit 1001 are input to the trained decoder unit 1002, and how the predicted data (phase fractions up to 120°C) are input to the trained decoder unit 1002. Note that due to space limitations, only the phase fraction at 120°C is shown in FIG. 11B(d) (in reality, each phase fraction from 800°C to 120°C is weighted and input to the trained decoder unit 1002). As a result, the trained decoder unit 1002 outputs the predicted data (phase fraction at 110°C).

[0120] Also, (d) of FIG. 11B shows how, when the predicted data (phase fraction at 110°C) is output, the "value (partial differential value) obtained by partially differentiating the predicted value for the specified phase at 110°C using the material composition information of the material to be predicted" is derived.

[0121] FIG. 11C (e) shows a state in which the feature values ​​output from the trained encoder unit 1001 are input to the trained decoder unit 1002, and the predicted data (phase fractions up to 110°C) is also input to the trained decoder unit 1002. Note that due to space limitations, FIG. 11C (e) only shows the phase fraction at 110°C (in reality, each phase fraction from 800°C to 110°C is weighted and input to the trained decoder unit 1002). As a result, the trained decoder unit 1002 outputs the predicted data (phase fraction at 100°C). Note that the example of FIG. 11C (e) also shows the predicted data (phase fraction at 100°C) output from the trained decoder unit 1002 upon input of the feature values ​​and the predicted data (phase fractions up to 110°C).

[0122] Also, (e) of FIG. 11C shows how, when the predicted data (phase fraction at 100°C) is output, the "value (partial differential value) obtained by partially differentiating the predicted value for the specified phase at 100°C using the material composition information of the material to be predicted" is derived.

[0123] <Flow of Prediction Process> Next, a description will be given of the flow of prediction process by the prediction device 120. Fig. 12 is a flowchart showing the flow of prediction process.

[0124] In step S1201, the material composition input unit 121 receives input of material composition information of the material to be predicted.

[0125] In step S1202, the phase designation unit 1003 accepts the designation of a phase.

[0126] In step S1203, the prediction unit 122 receives a start signal and inputs the input material composition information into the trained prediction model to predict the phase fraction at each temperature within a predetermined temperature range.

[0127] In step S1204, the sensitivity characteristic calculation unit 1004 derives a partial differential value obtained by partially differentiating the predicted value for the specified phase at each temperature within a predetermined temperature range using the material composition information.

[0128] In step S1205, the display unit 123 displays the predicted phase fraction at each temperature, and also displays a heat map for the specified phase.

[0129] <Example of predicted phase fraction> Next, a description will be given of a specific example of the phase fraction at each temperature within a predetermined temperature interval predicted by the prediction unit 122. Fig. 13 is a diagram showing an example of the predicted phase fraction, and shows predicted data when the weight percentages of each additive element constituting "6013," which is the name of a known aluminum alloy standard (however, in this case, the weight percentages of the minimum composition), are input as material composition information of the material to be predicted.

[0130] In this embodiment, when evaluating the prediction accuracy, the weight percentage (here, the weight percentage of the minimum composition) of each additive element constituting "6013," a known aluminum alloy standard, was used as input, and the output when the above-mentioned simulator was operated was used as the correct data. In addition, the mean squared logarithmic error (MSLE) loss was used to calculate the loss from the correct data. In the case of Figure 13, the MSLE loss between the predicted data and the correct data was 2.45 x 10 －５ This is what happened.

[0131] In this way, when predicting the phase fraction over a predetermined temperature range based on material composition information, the prediction device 120 according to this embodiment can improve prediction accuracy by: - ​​processing the phase fraction at each temperature as time-series data using a recurrent network; and - processing losses in a multifaceted manner using multiple types of loss functions.

[0132] <Example of Heat Map> Next, a specific example of a heat map displayed by the display unit 123 will be described. Fig. 14 is a diagram showing an example of a heat map, in which the designer has specified "Phase="Al １５ (Mn, Fe) ３ Si ２ " is specified.

[0133] 14, in a heat map 1400, the horizontal axis represents each component (each added element) of the material to be predicted. In the case of Fig. 14, the horizontal axis represents each component (each added element) of the material to be predicted, which are the added elements constituting "6013," a known aluminum alloy standard designation, namely, Si, Fe, Cu, Mn, Mg, Cr, Zn, and Ti.

[0134] 14, in the heat map 1400, the vertical axis indicates temperatures within a predetermined temperature range of 100 to 800°C. In FIG. 14, reference numeral 1410 indicates an example of a color scheme according to the partial differential value derived by the prediction unit 122. In the example of FIG. 14, a lighter color indicates a greater degree of influence on the specified phase (a larger partial differential value), and a darker color indicates a smaller degree of influence on the specified phase (a smaller partial differential value).

[0135] According to the example of FIG. 14, the material designer can see that: When the weight percentage of the added element = "Fe" is increased, the phase = "Al" is formed at around 580°C. １５ (Mn, Fe) ３ Si ２ " increases significantly. When the weight percentage of the added element = "Mn" is increased, the phase = "Al" increases from 100 ° C to 580 ° C. １５ (Mn, Fe) ３ Si ２ " always increases, and even if the weight percentage of the added element = "Si" is increased, the phase = "Al １５ (Mn, Fe) ３ Si ２ It is easy to understand which temperature range of a specified phase can be affected by increasing or decreasing the weight percentage of an additive element, such as "has little effect on the temperature range of a specified phase."

[0136] In this way, by configuring the system so that the degree of impact on the phase fraction due to a change in material composition can be grasped in advance, material designers can perform efficient material design. Furthermore, with the above configuration, processing can be performed in parallel with the output of prediction data (prediction of phase fraction). Therefore, compared to a method in which, for example, the weight percentage of all additive elements is sequentially increased or decreased, the phase fraction within a predetermined temperature range is predicted, and the prediction result is compared with the phase fraction before the change, the calculation load of the prediction device 120 can be significantly reduced.

[0137] <Verification of Heat Map> Next, the results of verification of the calculation accuracy of the heat map 1400 shown in Fig. 14 in the following manner will be described. - Material to be predicted: aluminum alloy (alloy standard "6013") - Added elements whose weight percentages are changed: Si, Fe, Cu, Mn, Mg, Cr, Zn, Ti - Method of changing the weight percentages: increasing the weight percentage by 0.05 from the minimum composition weight percentage (Si = 0.6, Fe = 0.01, Cu = 0.6, Mn = 0.2, Mg = 0.8, Cr = 0.01, Zn = 0.01, Ti = 0.01, Zr = 0, B = 0, Bi = 0, Pb = 0, Al = 97.76) - Prescribed temperature range: 100°C to 800°C - Designated phase: Al １５ (Mn, Fe) ３ Si ２ , ・Calculation method of data (difference value) used for verification: Al at each temperature before and after change １５ (Mn, Fe) ３ Si ２ The value indicating the ratio is calculated using a simulator (Thermo-Calc (registered trademark)), and the difference value before and after the change is calculated.

[0138] 15A to 15H are first to eighth figures showing an example of the verification results. Among them, Fig. 15A shows the calculated difference before and after the change when the weight percent of Si is increased by a predetermined weight percent (=0.05) from the minimum weight percent (=0.6) and the weight percent of Al is decreased by a predetermined weight percent (=0.05). １５ (Mn, Fe) ３ Si ２The difference value of the value indicating the ratio of Al at each temperature before and after the change is calculated using the partial differential value. １５ (Mn, Fe) ３ Si ２ The difference between the values ​​indicating the ratio of 1 to 2 is calculated using a simulator (Thermo-Calc (registered trademark)) and the values ​​of 2 to 3 are shown (however, the range of the vertical axis is narrower in (b)).

[0139] Similarly, FIG. 15B shows the calculated difference before and after the change when the weight percent of Fe is increased by a predetermined weight percent (=0.05) from the weight percent of the minimum composition (=0.01) and the weight percent of Al is decreased by a predetermined weight percent (=0.05). (a) shows the difference between the weight percent of Al at each temperature before and after the change. １５ (Mn, Fe) ３ Si ２ The difference value of the value indicating the ratio of Al at each temperature before and after the change is calculated using the partial differential value. １５ (Mn, Fe) ３ Si ２ The difference between the values ​​indicating the ratio of 1 to 2 is calculated using a simulator (Thermo-Calc (registered trademark)) and the values ​​of 2 to 3 are shown (however, the range of the vertical axis is narrower in (b)).

[0140] Comparing the graph transition from 100°C to 800°C shown in Fig. 15B(a) with the graph transition from 100°C to 800°C shown in Fig. 15B(b), it can be said that the trends are generally the same. For example, the trend of the difference value increasing around 580°C is consistent.

[0141] Similarly, FIG. 15C shows the calculated difference before and after the change when the weight percent of Cu is increased by a predetermined weight percent (=0.05) from the minimum weight percent (=0.6) and the weight percent of Al is decreased by a predetermined weight percent (=0.05). (a) shows the difference between the values ​​of Al at each temperature before and after the change. １５ (Mn, Fe) ３ Si ２ The difference value of the value indicating the ratio of Al at each temperature before and after the change is calculated using the partial differential value. １５ (Mn, Fe) ３ Si ２The difference between the values ​​indicating the ratio of 1 to 2 is calculated using a simulator (Thermo-Calc (registered trademark)) and the values ​​of 2 to 3 are shown (however, the range of the vertical axis is narrower in (b)).

[0142] Similarly, FIG. 15D shows the calculated difference before and after the change when the weight percent of Mn is increased by a predetermined weight percent (=0.05) from the minimum weight percent (=0.2) and the weight percent of Al is decreased by a predetermined weight percent (=0.05). (a) shows the Al at each temperature before and after the change. １５ (Mn, Fe) ３ Si ２ The difference value of the value indicating the ratio of Al at each temperature before and after the change is calculated using the partial differential value. １５ (Mn, Fe) ３ Si ２ The difference between the values ​​indicating the ratio of 1 to 2 is calculated using a simulator (Thermo-Calc (registered trademark)) and the values ​​of 2 to 3 are shown (however, the range of the vertical axis is narrower in (b)).

[0143] Comparing the transition of the graph from 100°C to 800°C shown in (a) of FIG. 15D with the transition of the graph from 100°C to 800°C shown in (b) of FIG. 15D, it can be said that the trends are generally the same. For example, over the range from 100°C to 580°C, １５ (Mn, Fe) ３ Si ２ The trend is consistent with an increase in the value indicating the proportion of

[0144] Similarly, (a) in FIG. 15E shows the calculated difference before and after the change when the weight percent of Mg is increased by a predetermined weight percent (=0.05) from the minimum weight percent (=0.8) and the weight percent of Al is decreased by a predetermined weight percent (=0.05). (a) shows the Al at each temperature before and after the change. １５ (Mn, Fe) ３ Si ２ The difference value of the value indicating the ratio of Al at each temperature before and after the change is calculated using the partial differential value. １５ (Mn, Fe) ３ Si ２The partial differential values ​​of the values ​​indicating the ratio of , and are calculated using a simulator (Thermo-Calc (registered trademark)) (however, the range of the vertical axis is narrower in (b)).

[0145] Similarly, FIG. 15F shows the calculated difference before and after the change when the weight percent of Cr is increased by a predetermined weight percent (=0.05) from the minimum weight percent (=0.01) and the weight percent of Al is decreased by a predetermined weight percent (=0.05). (a) shows the difference between the weight percent of Al at each temperature before and after the change. １５ (Mn, Fe) ３ Si ２ The difference value of the value indicating the ratio of Al at each temperature before and after the change is calculated using the partial differential value. １５ (Mn, Fe) ３ Si ２ The difference between the values ​​indicating the ratio of 1 to 2 is calculated using a simulator (Thermo-Calc (registered trademark)) and the values ​​of 2 to 3 are shown (however, the range of the vertical axis is narrower in (b)).

[0146] Similarly, FIG. 15G shows the calculated difference before and after the change when the weight percent of Zn is increased by a predetermined weight percent (=0.05) from the minimum weight percent (=0.01) and the weight percent of Al is decreased by a predetermined weight percent (=0.05). (a) shows the difference between the Al content and the Al content at each temperature before and after the change. １５ (Mn, Fe) ３ Si ２ The difference value of the value indicating the ratio of Al at each temperature before and after the change is calculated using the partial differential value. １５ (Mn, Fe) ３ Si ２ The difference between the values ​​indicating the ratio of 1 to 2 is calculated using a simulator (Thermo-Calc (registered trademark)) and the values ​​of 2 to 3 are shown (however, the range of the vertical axis is narrower in (b)).

[0147] Similarly, FIG. 15H shows the calculated difference before and after the change when the weight percent of Ti is increased by a predetermined weight percent (=0.05) from the weight percent of the minimum composition (=0.01) and the weight percent of Al is decreased by a predetermined weight percent (=0.05). (a) shows the difference between the weight percent of Al at each temperature before and after the change.１５ (Mn, Fe) ３ Si ２ The difference value of the value indicating the ratio of Al at each temperature before and after the change is calculated using the partial differential value. １５ (Mn, Fe) ３ Si ２ The difference between the values ​​indicating the ratio of 1 to 2 is calculated using a simulator (Thermo-Calc (registered trademark)) and the values ​​of 2 to 3 are shown (however, the range of the vertical axis is narrower in (b)).

[0148] As described above, it can be said that the heat map 1400 shown in FIG. 14 reproduces the tendency for the additional elements (for example, Fe, Mn) that show a large difference before and after the change.

[0149] <Summary> As is clear from the above description, the prediction device 120 according to the first embodiment: has a memory unit that stores a trained prediction model trained using training data in which the material composition of a training target material is associated with the phase fraction of the training target material at each temperature within a predetermined temperature interval; inputs the material composition of a training target material into the trained prediction model to predict the phase fraction of the training target material at each temperature within a predetermined temperature interval, and then partially differentiates the predicted value for a pre-specified phase at each temperature using the material composition of the training target material; and displays each region identified based on each component and each temperature in a color scheme corresponding to the value obtained by partial differentiation in a heat map in which each component of the training target material is plotted on a first axis and each temperature within a predetermined temperature interval is plotted on a second axis.

[0150] As a result, according to the first embodiment, it is possible to easily grasp the degree of influence on the phase fraction due to a change in material composition.

[0151] Second Embodiment In the first embodiment, the color scheme of the heat map is the color scheme indicated by reference numeral 1410 in Fig. 14, but a color scheme other than the color scheme indicated by reference numeral 1410 may be used. Furthermore, the heat map may be expressed in a display mode other than the color scheme.

[0152] In the first embodiment, the calculated partial differential values ​​are represented using a heat map so that the designer can easily understand which temperature range the phase fraction will be affected by increasing or decreasing the weight percentage of an additional element. However, the method of representing the calculated partial differential values ​​is not limited to a heat map. For example, if it is desired to understand which temperature in a predetermined temperature range has the greatest degree of influence, a graph showing the transition of the degree of influence due to temperature differences may be displayed instead of a heat map.

[0153] In the first embodiment, the alloy composition indicating the weight percentage of the additive elements (other metallic elements or non-metallic elements) relative to the metallic elements contained in the alloy is described as a specific example of the material composition, but the material composition is not limited to the alloy composition. For example, the material composition may be a chemical composition indicating the weight percentage of each chemical component contained in a substance other than the alloy.

[0154] In the first embodiment, either Seq2Seq with an Attention mechanism or a Transformer is described as being applied as the trained prediction model. However, the trained prediction model is not limited to these, and other architectures may be applied as long as they are capable of processing time-series data, which is data at predetermined time intervals. Specifically, a recurrent neural network (RNN), a bidirectional RNN, Seq2Seq, or the like may be applied. Alternatively, a gated recurrent unit (GRU), a long short-term memory (LSTM), or the like may be applied.

[0155] Furthermore, in the first embodiment, the learning device 110 and the prediction device 120 are configured as separate entities, but the learning device 110 and the prediction device 120 may be configured as an integrated device.

[0156] The present invention is not limited to the configurations described in the above embodiments, but may be combined with other elements, etc. These aspects can be changed without departing from the spirit of the present invention, and can be appropriately determined depending on the application form.

[0157] This application claims priority based on Japanese Patent Application No. 2022-211219, filed on December 28, 2022, the entire contents of which are incorporated herein by reference.

[0158] REFERENCE SIGNS LIST 100: Prediction system 110: Learning device 111: Learning data generation unit 112: Learning unit 120: Prediction device 121: Material composition input unit 122: Prediction unit 123: Display unit 300: Learning data 401: Element information input unit 402: Combination determination unit 403: Simulation unit 404: Storage control unit 501: Encoder unit 502: Decoder unit 503: Loss function calculation unit 801: Phase fraction error calculation unit at formation / disappearance temperature 802: Phase fraction error calculation unit 803: Phase fraction error (logarithmic value) calculation unit 804: Phase fraction error (difference value) calculation unit 805: Cross entropy error calculation unit 806: Weighted addition unit 1001: Learned encoder unit 1002: Learned decoder unit 1003: Phase designation unit 1004: Sensitivity characteristic calculation unit 1400: Heat map

Claims

1. a trained model trained using training data in which a material composition of a training target material is associated with a phase fraction of the training target material at each temperature within a predetermined temperature interval, the trained model having a memory unit configured to predict a phase fraction at an (i+1)th temperature using phase fractions predicted by the trained model for temperatures up to an i-th temperature (i is an integer equal to or greater than 1) within the predetermined temperature interval; When a phase fraction of the material to be predicted at each temperature within the predetermined temperature interval is predicted by inputting a material composition of the material to be predicted into the trained model, a predicted value for a pre-specified phase at each temperature is partially differentiated using the material composition of the material to be predicted; A prediction device which displays, in a heat map in which each component of the material to be predicted is placed on a first axis and each temperature within the specified temperature range is placed on a second axis, each area specified based on each component and each temperature is displayed in a display manner corresponding to the value obtained by performing the partial differentiation.

2. The prediction device of claim 1, wherein the trained model is applied with an architecture capable of processing time series data, which is data for each predetermined time interval, and the trained model predicts phase fraction for each predetermined temperature interval based on the material composition of the material to be predicted.

3. The prediction device according to claim 2 , wherein the trained model is any one of an RNN, a Bidirectional RNN, Seq2Seq, a Seq2Seq with an Attention mechanism, a GRU, a LSTM, or a Transformer.

4. The trained model is an encoder unit that outputs feature quantities by inputting a material composition of a material to be predicted; a decoder unit which receives the output feature amount and the predicted phase fractions at the first to the i-th temperatures and predicts the phase fraction at the i+1-th temperature; The prediction device according to claim 3 , further comprising:

5. The prediction device according to claim 1 , wherein the phase fraction is a phase fraction in a thermodynamic equilibrium state.

6. When predicting the phase fraction at the (i+1)th temperature, the predicted value for the pre-specified phase is partially differentiated using a material composition of the material to be predicted; The prediction device according to claim 1 , wherein in the heat map, each area identified based on each of the components and the (i+1)th temperature is displayed in a display manner according to a value obtained by performing the partial differentiation.

7. A prediction method in which a computer executes a trained model trained using training data in which a material composition of a training target material is associated with a phase fraction at each temperature of the training target material within a predetermined temperature interval, the trained model being configured to predict a phase fraction at an (i+1)th temperature using phase fractions predicted by the trained model for temperatures up to an (i)th temperature (i is an integer equal to or greater than 1) within the predetermined temperature interval, The computer, a step of inputting a material composition of a material to be predicted into the trained model to predict a phase fraction at each temperature of the material to be predicted within the predetermined temperature interval, partially differentiating a predicted value for a pre-specified phase at each temperature using the material composition of the material to be predicted; a step of displaying, in a heat map in which each component of the material to be predicted is arranged on a first axis and each temperature within the predetermined temperature range is arranged on a second axis, each region specified based on each component and each temperature in a display mode corresponding to a value obtained by performing the partial differentiation; A forecasting method to perform.

8. A prediction program for causing a computer to execute a trained model trained using training data in which a material composition of a training target material is associated with a phase fraction at each temperature of the training target material within a predetermined temperature interval, the trained model being configured to predict a phase fraction at an (i+1)th temperature using phase fractions predicted by the trained model for temperatures up to an (i)th temperature (i is an integer equal to or greater than 1) within the predetermined temperature interval, The computer includes: a step of inputting a material composition of a material to be predicted into the trained model to predict a phase fraction at each temperature of the material to be predicted within the predetermined temperature interval, partially differentiating a predicted value for a pre-specified phase at each temperature using the material composition of the material to be predicted; a step of displaying, in a heat map in which each component of the material to be predicted is arranged on a first axis and each temperature within the predetermined temperature range is arranged on a second axis, each region specified based on each component and each temperature in a display mode corresponding to a value obtained by performing the partial differentiation; A prediction program for executing the above.