Prediction device, prediction method, and prediction program

Abstract

And the influence degree on the phase volume fraction during material component change is easy to master. A prediction device is provided with a storage unit that stores a learned model obtained by learning using learning data in which a material component of a material to be learned is associated with a phase content at each temperature of the material to be learned within a predetermined temperature range, when a phase content at each temperature of a material to be predicted within the predetermined temperature range is predicted by inputting a material component of the material to be predicted into the learned model, a value obtained by predicting a pre-specified phase at each temperature is partially differentiated using the material component of the material to be predicted. In a heat map in which each component of the material to be predicted is disposed on a first axis and each temperature within the predetermined temperature range is disposed on a second axis, each region determined on the basis of each component and each temperature is displayed in a display manner corresponding to a value obtained by performing the partial differentiation.

Description

Technical Field

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

[0002] In the design of materials such as alloys, it is very important to calculate the phase fraction (i.e., the ratio of each phase) in the thermodynamic equilibrium state within a predetermined temperature range. An alternative model is currently being developed that can predict the phase fraction based on material composition information (in the case of alloys, the weight percentage of added elements).

[0003] Here, the phase content within a predetermined temperature range is related to the material properties. Therefore, when using the above model for material design, it is very important for designers to know in advance how to change the material composition to get closer to the desired phase content (in the case of alloys, increasing or decreasing the weight percentage of any added element will affect which phase within which temperature range) for efficient material design.

[0004] [Reference document]

[0005] [Patent document]

[0006] [Patent Document 1] International Publication No. 2020 / 090617 Summary of the invention

[0007] [Technical issues to be solved]

[0008] For this purpose, for example, if the following work is performed, that is, for all added elements, while the weight percentage is sequentially increased or decreased, the phase content (note: the Japanese original is "phase fraction") within a predetermined temperature range is predicted and compared with the phase content before the change, the degree of influence of the weight percentage of each added element on the phase content can be grasped in advance. However, when this method is adopted, there is also the problem that, for alloys with a large number of added elements, it will increase the workload of designers and make it difficult for designers to grasp the degree of influence on the phase content.

[0009] The purpose of the present disclosure is to make it easier to understand the degree of influence on the phase content when changing the material composition.

[0010] [Technical solution]

[0011] A prediction device according to a first aspect of the present disclosure includes:

[0012] a storage unit storing a learned model, wherein the learned model is a model learned by using learning data, wherein the learning data is data in which a material composition of a learning target material and a phase content at each temperature of the learning target material within a predetermined temperature range are associated, wherein the learned model is configured to predict a phase content at an i+1th temperature using phase contents predicted by the learned model at temperatures up to an i-th temperature (i is an integer greater than or equal to 1) within the predetermined temperature range,

[0013] in,

[0014] When the phase content at each temperature of the prediction target material within the predetermined temperature range is predicted by inputting the material composition of the prediction target material into the learned model, a value obtained by predicting a pre-specified phase at each temperature is partially differentiated using the material composition of the prediction target material.

[0015] In a heat map in which each component of the material of the prediction object is arranged on the first axis and each temperature within the predetermined temperature range is arranged on the second axis, each region determined based on the components and the temperatures is displayed in a display method corresponding to the value obtained by performing the partial differentiation.

[0016] A second aspect of the present disclosure is the prediction device according to the first aspect, wherein:

[0017] The learned model uses a framework capable of processing data at each predetermined time interval, ie, time series data, and the learned model predicts the phase content at each predetermined temperature interval based on the material composition of the material of the prediction object.

[0018] A third aspect of the present disclosure is the prediction device according to the second aspect, wherein:

[0019] The learned model is any one of RNN, Bidirectional RNN, Seq2Seq, Seq2Seq with Attention mechanism, GRU, LSTM and Transformer.

[0020] A fourth aspect of the present disclosure is the prediction device according to the third aspect, wherein:

[0021] The learned model has:

[0022] an encoder unit that outputs a feature value by inputting a material component of the material of the prediction target; and

[0023] The decoder unit predicts the phase content at the (i+1)th temperature by inputting the outputted feature amount and the predicted phase content at the temperatures up to the i-th temperature.

[0024] A fifth aspect of the present disclosure is the prediction device according to any one of the first to fourth aspects, wherein:

[0025] The phase content is the phase content in a thermodynamic equilibrium state.

[0026] A sixth aspect of the present disclosure is the prediction device according to any one of the first to fifth aspects, wherein:

[0027] When predicting the phase content at the (i+1)th temperature, a value obtained by predicting the predetermined phase is partially differentiated using the material composition of the material to be predicted,

[0028] In the heat map, each region determined based on the components and the (i+1)th temperature is displayed in a display manner corresponding to the value obtained by performing the partial differentiation.

[0029] The prediction method of the seventh aspect of the present disclosure is a method for executing a learned model by a computer, wherein the learned model is a model obtained by learning using learning data, the learning data being learning data in which the material composition of a material of a learning object and the phase contents at each temperature of the material of the learning object within a predetermined temperature range are associated, and the learned model is configured to predict the phase content at the i+1th temperature using the phase contents predicted by the learned model for temperatures up to the i-th temperature (i is an integer greater than 1) within the predetermined temperature range,

[0030] The computer executes:

[0031] When predicting the phase content at each temperature of the material to be predicted within the predetermined temperature range by inputting the material composition of the material to be predicted into the learned model, a step of partially differentiating a value obtained by predicting a pre-specified phase at each temperature using the material composition of the material to be predicted; and

[0032] In a heat map in which each component of the material of the prediction object is arranged on the first axis and each temperature within the predetermined temperature range is arranged on the second axis, a step of displaying each area determined based on the each component and the each temperature in a display method corresponding to the value obtained by performing the partial differentiation.

[0033] The prediction program of the eighth aspect of the present disclosure is a prediction program for causing a computer to execute a learned model, wherein the learned model is a model obtained by learning learning data, the learning data being data in which the material composition of a material of a learning object and the phase contents at each temperature of the material of the learning object within a predetermined temperature range are associated, and the learned model is configured to predict the phase content at the i+1th temperature using the phase contents predicted by the learned model for temperatures up to the i-th temperature (i is an integer greater than 1) within the predetermined temperature range,

[0034] The computer is caused to execute:

[0035] When predicting the phase content at each temperature of the material to be predicted within the predetermined temperature range by inputting the material composition of the material to be predicted into the learned model, a step of partially differentiating a value obtained by predicting a pre-specified phase at each temperature using the material composition of the material to be predicted; and

[0036] In a heat map in which each component of the material of the prediction object is arranged on the first axis and each temperature within the predetermined temperature range is arranged on the second axis, a step of displaying each area determined based on the each component and the each temperature in a display method corresponding to the value obtained by performing the partial differentiation.

[0037] [Beneficial Effects]

[0038] According to the present disclosure, it is possible to more easily understand the degree of influence on the phase content when the material composition changes. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] [ Figure 1 ] Figure 1 It is a diagram showing 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.

[0040] [ Figure 2 ] Figure 2 This is a diagram showing an example of the hardware configuration of a learning device and a prediction device.

[0041] [ Figure 3 ] Figure 3 This is a diagram showing a specific example of the structure of learning data and input data and correct data.

[0042] [ Figure 4 ] Figure 4 This is a diagram showing an example of the functional configuration of a learning data generating unit.

[0043] [ Figure 5 ] Figure 5 This is a diagram showing an example of the functional configuration of the learning unit.

[0044] [ Figure 6 ] Figure 6 FIG. 1 is a first diagram showing an operation example of the learning unit.

[0045] [ Figure 7 ] Figure 7 FIG. 2 is a second diagram showing an operation example of the learning unit.

[0046] [ Figure 8 ] Figure 8 This is a diagram showing an example of a method for calculating a loss performed by a loss function calculation unit.

[0047] [ Fig. 9 ] Fig. 9 It is a flowchart which shows the flow of a learning process.

[0048] [ Fig.10 ] Fig.10 This is a diagram showing an example of the functional configuration of the prediction unit.

[0049] [ Fig.11A ] Fig.11A FIG. 1 is a first diagram showing an operation example of the prediction unit.

[0050] [ Fig. 11B ] Fig. 11B FIG. 2 is a second diagram showing an operation example of the prediction unit.

[0051] [ Fig. 11C ] Fig. 11C FIG3 is a third diagram showing an operation example of the prediction unit.

[0052] [ Fig.12 ] Fig.12 It is a flowchart which shows the flow of prediction processing.

[0053] [ Fig.13 ] Fig.13 This is a diagram showing an example of the predicted phase contents.

[0054] [ Fig.14 ] Fig.14 This is a diagram showing an example of a heat map.

[0055] [ Fig.15A ] Fig.15A FIG1 is a first diagram showing an example of the verification result.

[0056] [ Fig. 15B ] Fig. 15B FIG. 2 is a second diagram showing an example of the verification result.

[0057] [ Fig. 15C ] Fig. 15C FIG3 is a third diagram showing an example of the verification result.

[0058] [ Fig.15D ] Fig.15D FIG. 4 is a fourth diagram showing an example of the verification result.

[0059] [ Fig.15E ] Fig.15E FIG5 is a diagram showing an example of the verification result.

[0060] [ Fig.15F ] Fig.15F FIG6 is a sixth diagram showing an example of the verification result.

[0061] [ Figure 15G ] Figure 15G FIG. 7 is a diagram showing an example of the verification result.

[0062] [ Fig.15H ] Fig.15H FIG8 is an example of a verification result. DETAILED DESCRIPTION

[0063] It should be noted that in this specification and the drawings, components having substantially the same functional configuration are given the same reference numerals (reference numerals), and repeated descriptions are omitted.

[0064] [First embodiment]

[0065] <System Configuration of Prediction System, Learning Device, and Functional Configuration of Prediction Device>

[0066] First, the system configuration of the prediction system including the prediction device of the first embodiment and the functional configuration of the learning (also called "training") device and the prediction device of the prediction system are described. In this embodiment, the prediction system is a system that predicts the phase content (phase content in a thermodynamic equilibrium state, hereinafter referred to as "phase content") at each temperature within a predetermined temperature range based on material composition information.

[0067] Figure 1 FIG. 1 is a diagram showing 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. Figure 1 As shown, the prediction system 100 includes a learning device 110 and a prediction device 120 .

[0068] A learning program is installed in the learning device 110 , and by executing the program, the learning device 110 functions as a learning data generating unit 111 and a learning unit 112 .

[0069] The learning data generating unit 111 generates learning data for learning the prediction model (substitute model) and stores it in the learning data storing unit 113. In the present embodiment, the learning data for learning the prediction model is a plurality of material component information and data in different combinations (i.e., different combinations) stored in the learning data storing unit 113 in association with each other.

[0070] Phase content at each temperature within a predetermined temperature range of a material having each material composition information.

[0071] The learning unit 112 reads the learning data from the learning data storage unit 113, and uses the read learning data to learn the prediction model. The learning unit 112 learns the prediction model in such a way that the output data when the material composition information stored in the learning data is input into the prediction model is close to the phase content at each temperature within the predetermined temperature range stored in association with the learning data. Based on this, the learning unit 112 generates a learned (also called "learned") prediction model. In addition, the learning unit 112 also stores the generated learned prediction model in the storage unit of the prediction unit 122 of the prediction device 120.

[0072] It should be noted that, in this embodiment, as a prediction model, the learning unit 112 applies a framework that can process data at each predetermined time interval, that is, time series data (generally, a framework used for natural language processing, etc.), that is,

[0073] Seq2Seq with Attention mechanism or

[0074] Transformer.

[0075] According to this, by inputting the material composition information stored in the learning data, the prediction model can sequentially output output data corresponding to the phase content at each temperature within a predetermined temperature range.

[0076] Here, "outputting output data corresponding to the phase content at each temperature in sequence" means, for example, that the prediction model uses the correct data (ground truth data) stored in the learning data for the temperatures up to the i-th temperature to output output data corresponding to the phase content at the i+1-th temperature. It should be noted that i is an integer greater than 1.

[0077] In this way, the learning unit 112 has a configuration that sequentially outputs output data corresponding to the phase content at each temperature (i.e., a configuration that processes the output data corresponding to the phase content at each temperature as time series data). Thus, the learning unit 112 can perform learning that reflects the output data corresponding to the phase content in adjacent temperature regions.

[0078] The prediction device 120 has a prediction program installed therein, and by executing the program, the prediction device 120 functions as a material component input unit 121 , a prediction unit 122 , and a display unit 123 .

[0079] When material component information of a material to be predicted is input, the material component input unit 121 receives the information and notifies the prediction unit 122 of the information.

[0080] The prediction unit 122 stores the learned prediction model learned by the learning unit 112 in the storage unit. In addition, the prediction unit 122 reads the learned prediction model from the storage unit and inputs the material composition information notified from the material composition input unit 121 into the learned prediction model, thereby sequentially predicting the phase content at each temperature within the predetermined temperature range. That is, the prediction unit 122 sequentially predicts the phase content within the predetermined temperature range for each predetermined temperature interval (section).

[0081] Here, "predicting the phase content sequentially at each predetermined temperature interval" means, for example, that the learned prediction model uses the phase contents predicted by the learned prediction model for temperatures up to the i-th temperature (i is an integer greater than 1) to predict the phase content at the i+1-th temperature.

[0082] Thus, in the present embodiment, the prediction unit 122 has a configuration for sequentially predicting the phase content at each temperature (i.e., a configuration for processing the phase content at each temperature as time series data through a recursive network). Accordingly, the prediction unit 122 can predict the prediction data reflecting the phase content in the adjacent temperature range. For this reason, for example, compared with the case where a multilayer neural network is applied (the case where the prediction result of the phase content in the adjacent temperature range is not reflected), the decrease in the prediction accuracy of the phase content can be suppressed. That is, the prediction unit 122 can improve the prediction accuracy when predicting the phase content in a predetermined temperature range based on the material composition.

[0083] Furthermore, when predicting the phase content within the predetermined temperature range in sequence at each predetermined temperature interval, the prediction unit 122 performs partial differentiation processing on the value obtained by predicting the pre-specified phase at each temperature using the material composition of the prediction target material. Accordingly, when the material composition changes slightly, the prediction unit 122 can calculate the extent to which the pre-specified phase changes at each temperature (the amount of change).

[0084] In addition, the prediction unit 122

[0085] · Arrange the components of the material to be predicted on the horizontal axis (first axis) and

[0086] Each temperature within a predetermined temperature range is arranged on a heat map on the vertical axis (second axis), and each region determined based on each component and each temperature is colored according to the value obtained by performing partial differentiation.

[0087] The display unit 123 displays the phase content at each temperature within the predetermined temperature range predicted by the prediction unit 122. It should be noted that the display unit 123 displays the phase content at each temperature within the predetermined temperature range predicted by the prediction unit 122 in a color-coded manner for each phase. However, the phase content at each temperature within the predetermined temperature range predicted by the prediction unit 122 may be displayed by changing the line type such as a solid line, a dotted line, or a dashed line for each phase.

[0088] In addition, the display unit 123 also displays the heat map generated by the prediction unit 122. Thus, a designer who designs materials can easily understand how to change the material composition to approach the desired phase content at each temperature within the displayed predetermined temperature range based on the displayed heat map.

[0089] As described above, according to the prediction device 120 of the first embodiment, when predicting the phase content based on the material composition information, a designer who designs materials can easily understand the degree of influence on the phase content when changing the material composition based on the heat map.

[0090] <Hardware Configuration of Learning Device and Prediction Device>

[0091] Next, the hardware configuration of the learning device 110 and the prediction device 120 is described. It should be noted that the learning device 110 and the prediction device 120 have similar hardware configurations, so the hardware configuration of the learning device 110 and the prediction device 120 is described in detail here. Figure 2 The hardware configurations of the learning device 110 and the prediction device 120 will be collectively described.

[0092] Figure 2 FIG. 1 is a diagram showing an example of the hardware configuration of a learning device and a prediction device. Figure 2 As shown, the learning device 110 and the prediction device 120 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. It should be noted that the hardware of the learning device 110 and the prediction device 120 are connected to each other via a bus 207.

[0093] The processor 201 includes various computing devices such as a CPU (Central Processing Unit) and a GPU (Graphics Processing Unit). The processor 201 reads various programs (for example, a learning program, a prediction program, etc.) into the memory 202 and executes them.

[0094] The memory 202 includes a main storage device such as a ROM (Read Only Memory) or a RAM (Random Access Memory). The processor 201 and the memory 202 form a so-called computer, and the processor 201 executes various programs read into the memory 202 , and the computer realizes the various functions described above.

[0095] The auxiliary storage device 203 stores various programs and various data used when the processor 201 executes the various programs. For example, the learning data storage unit 113 is implemented in the auxiliary storage device 203. Alternatively, the storage unit storing the learned prediction model is implemented in the auxiliary storage device 203.

[0096] The I / F device 204 is a connection device connected to an operation device 211 as 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.

[0097] The drive device 206 is a device for setting the storage medium 213. The storage medium 213 mentioned here includes a medium that stores information in an optical, electrical or magnetic manner, such as a CD-ROM, a floppy disk, a magneto-optical disk, etc. In addition, the storage medium 213 may also include a semiconductor memory that stores information in an electrical manner, such as a ROM, a flash memory, etc.

[0098] It should be noted that, for example, the various programs installed in the auxiliary storage device 203 can be installed by placing the distributed storage medium 213 into the drive device 206, and the drive device 206 reads the various programs stored in the storage medium 213 for installation. Alternatively, the various programs installed in the auxiliary storage device 203 can also be installed by downloading the various programs from the network via the communication device 205 for installation.

[0099] <Learning data structure, input data, and correct data>

[0100] Next, a specific example of the structure of the learning data and the input data and correct data included in the learning data will be described. Figure 3 This is a diagram showing a specific example of the structure of learning data and input data and correct data.

[0101] like Figure 3 As shown, the learning data 300 includes “input data” and “correct data” as information items.

[0102] The “input data” stores different combinations of material composition information, namely, “material composition information 1”, “material composition information 2”, etc. Figure 3 , label 301 represents a specific example of "material composition information 1", which shows an example of the weight percentage of each component (each added element Si, Fe, Cu, Mn, Mg, Cr, Zn and Ti) constituting the known aluminum alloy standard named "6013".

[0103] Similarly, label 302 represents a specific example of "material composition information 2", which shows the weight percentage of each component (the added elements Si, Fe, Cu, Mn, Mg, Cr, Zn and Ti) that constitutes the known aluminum alloy standard named "6060".

[0104] The "correct data" stores the expected temperature ranges associated with "material composition information 1", "material composition information 2", etc. ( Figure 3 In the example, it is the phase content at each temperature within the range of 100°C to 800°C, namely "phase content 1", "phase content 2", etc. Figure 3 In the figure, reference numeral 311 indicates a specific example of "phase content 1", and reference numeral 312 indicates a specific example of "phase content 2". In both cases, the horizontal axis indicates temperature, and the vertical axis indicates phase content.

[0105] Here, reference numerals 311 and 312 are results obtained by operating a costly simulator having high prediction accuracy for a long period of time, and in this embodiment, these results are used as correct data.

[0106] It should be noted that the simulator referred to herein is software that calculates the phase content at each temperature within a predetermined temperature range by thermodynamic equilibrium calculation, 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 OpenCalphad (registered trademark).

[0107] <Details of each unit (functional unit) of the learning device>

[0108] Next, each functional unit (learning data generating unit 111 and learning unit 112 ) of the learning device 110 will be described in detail.

[0109] (1) Functional structure of the learning data generation unit

[0110] First, the functional configuration of the learning data generating unit 111 will be described. Figure 4 This is a diagram showing an example of the functional configuration of a learning data generating unit.

[0111] like Figure 4 As shown, the learning data generating unit 111 includes an element information input unit 401 , a combination specifying unit 402 , a simulation unit 403 , and a storage (saving) control unit 404 .

[0112] The element information input unit 401 receives input of the type of additional element to be added when generating a specific alloy, for example, and also receives input of the upper and lower limits of the added amount (weight percentage) of each additional element.

[0113] The combination determination unit 402 randomly selects the addition amount of each additional element obtained by receiving the input under the constraints of the upper limit value and the lower limit value, thereby determining multiple combinations of the addition amount of each additional element. It should be noted that the combination of the addition amount of each additional element can be generated based on a predetermined width (step length) between the upper limit value and the lower limit value. In addition, the combination determination unit 402 also notifies the simulation unit 403 of the material composition information represented by the determined multiple combinations.

[0114] The simulation unit 403 calculates the phase content at each temperature within the predetermined temperature range by operating the simulator for each of the plurality of material composition information notified from the combination determination unit 402. In addition, the simulation unit 403 notifies the storage control unit 404 of the plurality of material composition information and the corresponding phase content at each temperature within the predetermined temperature range.

[0115] The storage control unit 404 generates learning data using the plurality of material component information notified from the simulation unit 403 as “input data” and the phase contents at each temperature within the corresponding predetermined temperature range as “correct data”, and stores the data in the learning data storage unit 113 .

[0116] (2) Functional structure of the learning department

[0117] Next, the functional configuration of the learning unit 112 will be described. Figure 5 This is a diagram showing an example of the functional configuration of the learning unit.

[0118] like Figure 5 As shown, the learning unit 112 has a prediction model (substitute model) and a loss function calculation unit 503.

[0119] As described above, the prediction model uses Seq2Seq or Transformer with an Attention mechanism and includes an encoder part 501 and a decoder part 502 .

[0120] The encoder unit 501 outputs a feature value by inputting “material component information 1 ”, “material component information 2 ”, etc. stored in “input data” of the learning data 300 .

[0121] The decoder unit 502 sequentially outputs output data by inputting the feature quantity output from the encoder unit 501. Specifically, the decoder unit 502 reads "phase content 1", "phase content 2", etc. stored in the "correct data" of the learning data 300. Then, the decoder unit 502 uses the phase content at the temperature up to the i-th temperature (that is, the correct data at the temperature up to the i-th temperature) to output the output data at the i+1-th temperature. Accordingly, the decoder unit 502 can sequentially output output data corresponding to the phase content at each temperature within a predetermined temperature range based on the feature quantity output from the encoder unit 501 by inputting "material component information 1" and the "correct data" of "phase content 1". Similarly, the decoder unit 502 can sequentially output output data corresponding to the phase content at each temperature within a predetermined temperature range based on the feature quantity output from the encoder unit 501 by inputting "material component information 2" and the "correct data" of "phase content 2".

[0122] It should be noted that the Softmax function is used in the output layer of the decoder unit 502. Accordingly, the multiple values ​​(i.e., the values ​​representing the ratios of the phases) included in the output data of the decoder unit 502 are in the range of 0 to 1, and the sum of the multiple values ​​(i.e., the sum of the values ​​representing the ratios of the phases) is "1".

[0123] In this way, by adopting the Softmax function in the output layer of the decoder unit 502, output data corresponding to the phase content at each temperature within the predetermined temperature range can be sequentially output without performing additional calculations.

[0124] The loss function calculation unit 503 reads “phase content ratio 1”, “phase content ratio 2”, etc. stored in the “correct data” of the learning data 300 , and compares them with the output data at the i+1th temperature output by the decoder unit 502 .

[0125] In addition, when performing the comparison, the loss function calculation unit 503 calculates multiple types of losses and performs weighted summation of the calculated multiple types of losses to calculate the comprehensive loss. In addition, the loss function calculation unit 503 also updates the model parameters of the encoder unit 501 and the decoder unit 502 according to the calculated comprehensive loss. Based on this, a learned prediction model (including a learned encoder unit and a learned decoder unit) is generated.

[0126] (3) Learning unit operation example 1

[0127] Next, an operation example of the learning unit 112 (mainly, an operation example of the prediction model in the prediction model and loss function calculation unit) is described. Figure 6 FIG. 1 is a first diagram showing an operation example of the learning unit.

[0128] in, Figure 6 (a) shows a situation where the feature quantity is output from the encoder unit 501 by reading the "input data" of the learning data 300 and inputting it to the encoder unit 501. In addition, Figure 6 (a) also shows that the decoder unit 502 outputs the output data by inputting the start signal to the decoder unit 502. Figure 6 In the case of the example of (a), the output data output by the decoder unit 502 is equivalent to the phase content at 800°C. The output data output from the decoder unit 502 (here, the output data equivalent to the phase content at 800°C) is notified to the loss function calculation unit 503. It should be noted that the start signal is a signal for causing the decoder unit 502 to start operating. In this embodiment, an arbitrary value different from the output data output from the decoder unit 502 (for example, "000000", etc.) is input to the decoder unit 502 as a start signal.

[0129] Figure 6 (b) shows a situation where the decoder unit 502 outputs output data by inputting the feature quantity output from the encoder unit 501 to the decoder unit 502 and inputting the phase content at 800° C. (correct data at 800° C.) to the decoder unit 502 . Figure 6 In the case of example (b), the output data outputted from the decoder unit 502 corresponds to the phase content at 790° C. It should be noted that the output data outputted from the decoder unit 502 (here, the output data corresponding to the phase content at 790° C.) is notified to the loss function calculation unit 503 .

[0130] Figure 6 (c) shows a situation where the decoder 502 outputs output data by inputting the feature quantity output from the encoder 501 to the decoder 502 and inputting the phase content up to 790°C (correct data up to 790°C) to the decoder 502. However, Figure 6 In (c), due to space limitations, only the correct data at 790° C. are recorded (in fact, the correct data at 800° C. and the correct data at 790° C. are also input into the decoder unit 502 in a weighted manner).

[0131] in addition, Figure 6 In the case of example (c), the output data outputted from the decoder unit 502 corresponds to the phase content at 780° C. It should be noted that the output data outputted from the decoder unit 502 (here, the output data corresponding to the phase content at 780° C.) is notified to the loss function calculation unit 503 .

[0132] Figure 6 (d) shows a situation where the decoder 502 outputs output data by inputting the feature quantity output from the encoder 501 to the decoder 502 and inputting the phase content up to 120°C (correct data up to 120°C) to the decoder 502. However, Figure 6 In (d), due to space limitations, only the correct data at 120° C. are recorded (in fact, the correct data at 800° C. to 120° C. are also input into the decoder unit 502 in a weighted manner).

[0133] in addition, Figure 6 In the case of example (d), the output data outputted from the decoder unit 502 corresponds to the phase content ratio at 110° C. It should be noted that the output data outputted from the decoder unit 502 (here, the output data corresponding to the phase content ratio at 110° C.) is notified to the loss function calculation unit 503 .

[0134] Figure 6 (e) shows a situation where the decoder 502 outputs output data by inputting the feature quantity output from the encoder 501 to the decoder 502 and inputting the phase content up to 110°C (correct data up to 110°C) to the decoder 502. However, Figure 6 In (e), due to space limitations, only the correct data at 110° C. are recorded (in fact, the correct data at 800° C. to 110° C. are also input into the decoder unit 502 in a weighted manner).

[0135] in addition, Figure 6 In the case of the example (e), the output data outputted from the decoder unit 502 corresponds to the phase content ratio at 100° C. It should be noted that the output data outputted from the decoder unit 502 (here, the output data corresponding to the phase content ratio at 100° C.) is notified to the loss function calculation unit 503 .

[0136] (4) Learning unit operation example 2

[0137] Next, an operation example of the learning unit 112 (mainly an operation example of the loss function calculation unit 503 in the prediction model and loss function calculation unit) is described. Figure 7 FIG. 2 is a second diagram showing an operation example of the learning unit. Figure 7As shown, the loss function calculation unit 503 reads the “correct data” of the learning data 300 . Figure 7 In the example shown in FIG. 1 , the phase content at each temperature within the predetermined temperature range included in “phase content 1” is

[0138] Phase content at 800°C (correct data),

[0139] ·Phase content at 790℃ (correct data), ···, and

[0140] Phase content at 100°C (correct data)

[0141] Reading situation.

[0142] also, Figure 7 In the example, it is also shown that the result of output data sequentially output from the decoder unit 502 based on the feature quantity output from the encoder unit 501 by inputting "material composition information 1", that is, the loss function calculation unit 503 calculates the material composition information 1 in the loss function calculation unit 503.

[0143] Phase content at 800°C (output data),

[0144] ·Phase content at 790°C (output data), ···, and

[0145] Phase content at 100°C (output data)

[0146] Keep the situation.

[0147] in addition, Figure 7 The example also shows that the loss function calculation unit 503 compares the phase content (correct data) at each temperature of 100°C to 800°C with the phase content (output data) at each temperature of 100°C to 800°C, and calculates the comprehensive loss. In addition, Figure 7 The example also shows that the loss function calculation unit 503 updates the model parameters of the encoder unit 501 and the decoder unit 502 according to the calculated comprehensive loss. It should be noted that the functional structure of the loss function calculation unit 503 for calculating the comprehensive loss will be described in detail below.

[0148] (5) Details of the functional structure of the loss function calculation unit

[0149] Figure 8 is a diagram showing an example of a method for calculating a loss performed by a loss function calculation unit. Figure 8 As shown, as a function for calculating various types of losses, the loss function calculation unit 503 has

[0150] Phase content error calculation unit 801 at the generation / disappearance temperature,

[0151] Phase content error calculation unit 802,

[0152] Phase content error (logarithmic value) calculation unit 803,

[0153] Phase content error (difference value) calculation unit 804,

[0154] Cross entropy error calculation unit 805, and

[0155] A weighted summation unit 806 .

[0156] The phase content error calculation unit 801 at the generation / disappearance temperature compares the phase content at each temperature of 100°C to 800°C (correct data) with the phase content at each temperature of 100°C to 800°C (output data) for the temperature at which the curve of each phase content in the correct data of the learning data becomes 0. Accordingly, the phase content error calculation unit 801 at the generation / disappearance temperature adds (adds) the errors between the phase content determined based on the output data and the phase content determined based on the learning data at the generation or disappearance temperature of each phase determined based on the learning data. In addition, the phase content error calculation unit 801 at the generation / disappearance temperature also outputs the addition result (first addition result).

[0157] The phase content error calculation unit 802 compares the phase content (correct data) at each temperature of 100° C. to 800° C. with the phase content (output data) at each temperature of 100° C. to 800° C. Based on this, the phase content error calculation unit 802 adds the errors of the phase content at each temperature within the predetermined temperature range and outputs the addition result (second addition result).

[0158] The phase content error (logarithmic value) calculation unit 803 compares the phase content (correct data) at each temperature of 100° C. to 800° C. with the phase content (output data) at each temperature of 100° C. to 800° C. Based on this, the phase content error (logarithmic value) calculation unit 803 adds the errors of the logarithmic values ​​of the phase content at each temperature within the predetermined temperature range and outputs the addition result (third addition result).

[0159] The phase content error (difference value) calculation unit 804 compares the phase content (correct data) at each temperature of 100° C. to 800° C. with the phase content (output data) at each temperature of 100° C. to 800° C. Based on this, the phase content error (difference value) calculation unit 804 adds the errors of the difference values ​​of the phase content between adjacent temperatures within a predetermined temperature range, and outputs the addition result (fourth addition result).

[0160] The cross entropy error calculation unit 805 compares the phase content (correct data) at each temperature of 100° C. to 800° C. with the phase content (output data) at each temperature of 100° C. to 800° C. Based on this, the cross entropy error calculation unit 805 adds the errors of the ratios of the phase content at each temperature within the predetermined temperature range and outputs the addition result (the fifth addition result).

[0161] The weighted summation unit 806 performs weighted summation on the addition results (the first addition result to the fifth addition result) output from the phase content error calculation unit 801 to the cross entropy error calculation unit 805 at the generation / disappearance temperature, thereby calculating the comprehensive loss.

[0162] Thus, by calculating multiple types of losses and performing weighted summation, the loss function calculation unit 503 can process the loss between the output data and the correct data in various aspects. Therefore, when learning the prediction model, the learning unit 112 can appropriately update the model parameters.

[0163] <Flow of learning process>

[0164] Next, the flow of the learning process of the learning device 110 will be described. Fig. 9 It is a flowchart which shows the flow of a learning process.

[0165] In step S901 , the learning data generating unit 111 receives input of the type of additional element to be added when generating a specific alloy and the upper limit and lower limit of the addition amount (weight percentage) of each additional element as element information.

[0166] In step S902 , the learning data generating unit 111 randomly selects the amount of each additional element under the constraints of the upper limit value and the lower limit value, thereby determining a plurality of combinations of the amount of each additional element (percent by weight).

[0167] In step S903 , the learning data generating unit 111 causes the simulator to operate for each of the material component information represented by the determined plurality of combinations, and calculates the phase content at each temperature within a predetermined temperature range.

[0168] In step S904 , the learning data generating unit 111 generates learning data and stores the learning data in the learning data storing unit 113 .

[0169] In step S905 , the learning unit 112 inputs a start signal to the decoder unit 502 .

[0170] In step S906 , the learning unit 112 reads “input data” of the learning data, and inputs it into the prediction model.

[0171] In step S907 , the learning unit 112 sequentially inputs the phase contents at temperatures up to the i-th temperature of the “correct data” of the learning data (ie, the correct data at temperatures up to the i-th temperature), thereby sequentially outputting i+1-th output data.

[0172] In step S908, the learning unit 112 determines whether all output data corresponding to the phase content at each temperature within the predetermined temperature range have been output. In step S908, if it is determined that there is still a temperature for which output data corresponding to the phase content has not been output (in the case of NO in step S908), the process returns to step S907.

[0173] On the other hand, in step S908, when it is determined that all output data corresponding to the phase contents at the respective temperatures within the predetermined temperature range have been output (if YES in step S908), the process proceeds to step S909.

[0174] In step S909 , the learning unit 112 reads the “correct data” of the learning data and compares it with the output data.

[0175] In step S910, the learning unit 112 calculates multiple types of losses, and performs weighted summation on the calculated multiple types of losses to calculate the comprehensive loss. In addition, the learning unit 112 also updates the model parameters of the encoder unit 501 and the decoder unit 502 according to the calculated comprehensive loss.

[0176] In step S911, the learning unit 112 determines whether to continue the learning process. In step S911, if it is determined that the learning process is to continue (if YES in step S911), the process returns to step S906, reads the next "input data" of the learning data, and performs similar processing.

[0177] On the other hand, in step S911, when it is determined that the learning process is to be terminated (in the case of NO in step S911), the process proceeds to step S912.

[0178] In step S912 , the learning unit 112 regards the learned encoder unit and the learned decoder unit as a learned prediction model, and stores them in the storage unit of the prediction device 120 .

[0179] <Details of Each Part (Functional Part) of the Prediction Device>

[0180] Next, the details of each functional unit (here, the prediction unit 122) of the prediction device 120 will be described.

[0181] (1) Functional structure of the prediction unit

[0182] First, the functional configuration of the prediction unit 122 will be described. Fig.10 This is a diagram showing an example of the functional configuration of the prediction unit.

[0183] like Fig.10 As shown, the prediction unit 122 includes a learned encoder unit 1001 and a learned decoder unit 1002 obtained by learning by the learning unit 112. The learned encoder unit 1001 and the learned decoder unit 1002 form a learned prediction model.

[0184] The learned encoder unit 1001 calculates a feature quantity by inputting the material component information (material component information of a prediction target material) notified from the material component input unit 121 , and outputs the calculated feature quantity to the learned decoder unit 1002 .

[0185] The learned decoder unit 1002 receives the feature quantity output from the learned encoder unit 1001, and sequentially outputs prediction data. Specifically, the learned decoder unit 1002 uses the prediction data at the i-th temperature to output the prediction data at the i+1-th temperature. Thus, the learned decoder unit 1002 can predict the phase content at each temperature within a predetermined temperature range based on the feature quantity output from the learned encoder unit 1001 by inputting the material composition information of the material to be predicted.

[0186] In addition, if Fig.10 As shown, the prediction unit 122 further includes a phase specifying unit 1003 and a sensitivity characteristic calculating unit 1004. The phase specifying unit 1003 and the sensitivity characteristic calculating unit 1004 form an input sensitivity analyzing unit.

[0187] The phase designation unit 1003 receives an input of a phase designated by a designer who designs materials. The phase designated by the designer who designs materials is a phase that the designer who designs materials wants to know in advance how much it 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 obtained by receiving the input.

[0188] When outputting the predicted data at the (i+1)th temperature, the sensitivity characteristic calculation section 1004 partially differentiates the value obtained by predicting the phase notified from the phase specification section 1003 using the material component information notified from the material component input section 121 .

[0189] Specifically, the sensitivity characteristic calculation unit 1004 uses

[0190] The partial differential values ​​of each layer calculated by processing in each layer of the learned encoder unit 1001 during the period from when the material component information is input to when the feature value is output, and

[0191] ·During the period from when the feature value is input to when the value obtained by predicting the specified phase (also referred to as the "specified phase") at the i+1th temperature is output, the "value obtained by partially differentiating the value at the i+1th temperature obtained by predicting the specified phase using material composition information" is derived by using the partial differential values ​​of each layer calculated by processing each layer in the learned decoder unit 1002.

[0192] Furthermore, the sensitivity characteristic calculation unit 1004 notifies the display unit 123 of “the value obtained by partially differentiating the value at the (i+1)th temperature obtained by predicting the specified phase using the material composition information” ( Fig.10 (not shown in the figure).

[0193] It should be noted that the display unit 123 colors the area determined based on the components (additive elements) of the material of the prediction object and the i+1th temperature in the heat map according to the value obtained by partial differentiation. Accordingly, the display unit 123 can generate and display the heat map of the predetermined temperature range by performing these processes within the predetermined temperature range.

[0194] (2) Example of operation of the prediction unit

[0195] Next, an operation example of the prediction unit 122 will be described. Figures 11A to 11C FIG. 1 to FIG. 3 are diagrams showing examples of the operation of the prediction unit. Fig.11A (a) of FIG. 1 shows a case where the material composition information of the material to be predicted is input into the learned encoder unit 1001 and the feature quantity is output from the learned encoder unit 1001. Fig.11A (a) also shows a case where the feature quantity output from the learned encoder unit 1001 is input to the learned decoder unit 1002 and a start signal is input to the learned decoder unit 1002 to output predicted data (phase content at 800° C.).

[0196] in addition, Fig.11A (a) also shows that when the predicted data (phase content at 800°C) is output, the situation where "the value obtained by partially differentiating the value obtained by predicting the specified phase using the material composition information of the material of the prediction object (partial differential value)" is derived at a temperature of 800°C.

[0197] Fig.11A (b) shows a situation where the feature quantity output from the learned encoder unit 1001 is input to the learned decoder unit 1002 and the prediction data (phase content at 800°C) is input to the learned decoder unit 1002. Based on this, the learned decoder unit 1002 outputs the prediction data (phase content at 790°C).

[0198] also, Fig.11A (b) also shows that when the predicted data (phase content at 790°C) is output, the "value (partial differential value) obtained by partially differentiating the value obtained by predicting the specified phase using the material composition information of the material of the prediction object" is derived at a temperature of 790°C.

[0199] Fig. 11B (c) shows a situation where the feature quantity output from the learned encoder unit 1001 is input to the learned decoder unit 1002 and the prediction data (phase content up to 790° C.) is input to the learned decoder unit 1002. It should be noted that Fig. 11B In (c), due to space limitations, only the phase content at 790°C is recorded (in fact, the phase content at 800°C and the phase content at 790°C are also input to the learned decoder unit 1002 in a weighted manner). Based on this, the learned decoder unit 1002 outputs predicted data (phase content at 780°C).

[0200] also, Fig. 11B (c) also shows that when the predicted data (phase content at 780°C) is output, the situation where "the value obtained by partially differentiating the value obtained by predicting the specified phase using the material composition information of the material of the prediction object (partial differential value)" is derived at a temperature of 780°C.

[0201] Fig. 11B (d) of FIG. 1 shows a case where the feature quantity output from the learned encoder unit 1001 is input to the learned decoder unit 1002 and the prediction data (phase content up to 120° C.) is input to the learned decoder unit 1002. It should be noted that Fig. 11B In (d), due to space limitations, only the phase content at 120°C is described (actually, each phase content at 800°C to 120°C is also input into the learned decoder unit 1002 in a weighted manner). Based on this, the learned decoder unit 1002 outputs prediction data (phase content at 110°C).

[0202] also, Fig. 11B (d) also shows that when the predicted data (phase content at 110°C) is output, the situation where "the value obtained by partially differentiating the value obtained by predicting the specified phase using the material composition information of the material of the prediction object (partial differential value)" is derived at a temperature of 110°C.

[0203] Fig. 11C (e) shows a situation where the feature quantity output from the learned encoder unit 1001 is input to the learned decoder unit 1002 and the prediction data (phase content up to 110° C.) is input to the learned decoder unit 1002. It should be noted that Fig. 11CIn (e), due to space limitations, only the phase content at 110°C is recorded (in fact, each phase content at 800°C to 110°C is also input into the learned decoder unit 1002 in a weighted manner). Based on this, the learned decoder unit 1002 outputs the predicted data (phase content at 100°C). It should be noted that Fig. 11C The example of (e) also shows that the feature amount and the prediction data (phase content ratio up to 110° C.) are input and the prediction data (phase content ratio at 100° C.) is output from the learned decoder unit 1002 .

[0204] also, Fig. 11C (e) also shows that when the predicted data (phase content at 100°C) is output, the situation where "the value obtained by partially differentiating the value obtained by predicting the specified phase using the material composition information of the material of the prediction object (partial differential value)" is derived at a temperature of 100°C.

[0205] <Flow of prediction processing>

[0206] Next, the flow of the prediction process executed by the prediction device 120 will be described. Fig.12 It is a flowchart which shows the flow of prediction processing.

[0207] In step S1201 , the material component input unit 121 receives input of material component information of a prediction target material.

[0208] In step S1202, the phase specifying unit 1003 receives the specification of the phase.

[0209] In step S1203 , the prediction unit 122 receives the start signal and inputs the input material composition information into the learned prediction model, thereby predicting the phase content at each temperature within a predetermined temperature range.

[0210] In step S1204 , the sensitivity characteristic calculation unit 1004 derives “a partial differential value obtained by partially differentiating a value obtained by predicting a designated phase at each temperature within a predetermined temperature range by using the material component information”.

[0211] In step S1205, the display unit 123 displays the predicted phase contents at each temperature. In addition, the display unit 123 also displays a heat map of the designated phase.

[0212] <An example of predicted phase content>

[0213] Next, a specific example of the phase content at each temperature within the predetermined temperature range predicted by the prediction unit 122 will be described. Fig.13This is a diagram showing an example of predicted phase contents, showing predicted data when the material composition information of the material to be predicted is input as the weight percentage of each added element in the known aluminum alloy standard named "6013" (however, here it is the weight percentage of the minimum component).

[0214] It should be noted that, in the present embodiment, when evaluating (assessing) the prediction accuracy, the weight percentage of each added element constituting the known aluminum alloy standard named "6013" (however, here, the weight percentage of the minimum component) is used as input, and the output when the above-mentioned simulator is operated is used as correct data. In addition, the loss between the correct data and the mean square logarithmic error (MSLE: Mean Squared Logarithmic Error) loss is used. Fig.13 In the case of , the MSLE loss between the predicted data and the correct data is 2.45×10 -5 .

[0215] Thus, when the phase content within a predetermined temperature range is predicted based on the material composition information, by: treating the phase content at each temperature as time series data and processing it using a recursive network, and by using multiple types of loss functions to process the loss in various aspects,

[0216] The prediction device 120 of this embodiment can improve prediction accuracy.

[0217] <Example of a heat map>

[0218] Next, a specific example of the 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 a phase = "Al" is specified by a designer. 15 Thermal map of (Mn,Fe)3Si2”.

[0219] like Fig.14 As shown, in the heat map 1400, the horizontal axis represents each component (each added element) of the material to be predicted. Fig.14 In the case of , the horizontal axis is arranged with the various components (additive elements) of the material as the prediction object, which constitute the known aluminum alloy standard named "6013", namely Si, Fe, Cu, Mn, Mg, Cr, Zn and Ti.

[0220] In addition, if Fig.14 As shown, in the heat map 1400, the vertical axis is configured with each temperature within a predetermined temperature range, that is, 100 to 800°C. It should be noted that Fig.14 , reference numeral 1410 denotes an example of color arrangement corresponding to the partial differential value derived by the prediction unit 122 . Fig.14The example shows that the lighter the color, the greater the influence on the specified phase (the larger the partial differential value), and the darker the color, the smaller the influence on the specified phase (the smaller the partial differential value).

[0221] according to Fig.14 For example, designers working on material design can

[0222] If the weight percentage of the added element = "Fe" is increased, the phase = "Al 15 (Mn,Fe)3Si2” increases significantly near 580℃.

[0223] If the weight percentage of the added element = "Mn" is increased, the phase = "Al 15 (Mn,Fe)3Si2” always increases between 100℃ and 580℃, and

[0224] Even if the weight percentage of the added element = "Si" is increased, the phase = "Al 15 The influence of (Mn,Fe)3Si2" is also relatively small, so it is easy to understand which temperature range of the specified phase will be affected by increasing or decreasing the weight percentage of any added element.

[0225] In this way, by configuring the system so that the degree of influence on the phase content when the material composition changes can be grasped in advance, the designer who designs the material can perform efficient material design. It should be noted that by adopting the above configuration, the processing can also be performed in parallel with the output of the prediction data (prediction of the phase content). For this reason, compared with a method that performs an operation such as "for all the added elements, while increasing or decreasing the weight percentage in sequence, predicting the phase content within a predetermined temperature range and comparing it with the phase content before the change", the calculation load of the prediction device 120 can be greatly reduced.

[0226] <Verification of heat map>

[0227] Next, the following description is given Fig.14 The calculation accuracy of the heat map 1400 shown is verified and the verification results are obtained.

[0228] ·Material for prediction: Aluminum alloy (alloy standard "6013")

[0229] Additive elements whose weight percentage is changed: Si, Fe, Cu, Mn, Mg, Cr, Zn and Ti

[0230] Method of changing weight percentage: Add 0.05 weight percentage to the weight percentage of the smallest component (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 and Al=97.76)

[0231] ·Predetermined temperature range: 100℃~800℃

[0232] Specified phase: Al 15 (Mn,Fe)3Si2

[0233] · Calculation method of data (difference value) used in verification: Al at each temperature before and after the change was calculated using a simulator (Thermo-Calc (registered trademark)) 15 The value of the ratio of (Mn,Fe)3Si2 and the difference before and after the change are calculated

[0234] Figures 15A to 15H Figures 1 to 8 show an example of the verification result. Fig.15A The difference before and after the change is calculated when the weight percentage of Si is increased by a predetermined weight percentage (=0.05) from the weight percentage of the minimum component (=0.6) and the weight percentage of Al is decreased by a predetermined weight percentage (=0.05).

[0235] (a) Calculate the Al at each temperature before and after the change based on the above partial differential value 15 The difference in the value of the ratio of (Mn,Fe)3Si2, and

[0236] (b) The Al at each temperature before and after the change was calculated using a simulator (Thermo-Calc (registered trademark)). 15 The case of the difference in the value of the ratio of (Mn,Fe)3Si2 (here, the range of the axis in (b) is narrow).

[0237] Similarly, Fig. 15B The graphs are for calculating the difference before and after the change when the weight percentage of Fe is increased by a predetermined weight percentage (=0.05) from the weight percentage of the minimum component (=0.01) and the weight percentage of Al is decreased by a predetermined weight percentage (=0.05). (a) The difference between the weight percentage of Al at each temperature before and after the change is calculated based on the partial differential values. 15 The difference in the value of the ratio of (Mn,Fe)3Si2, and

[0238] (b) The Al at each temperature before and after the change was calculated using a simulator (Thermo-Calc (registered trademark)). 15 The case of the difference in the value of the ratio of (Mn,Fe)3Si2 (here, the range of the vertical axis in (b) is narrow).

[0239] Will Fig. 15B The transition of the graph at 100℃~800℃ shown in (a) and Fig. 15B Comparing the transitions of the graphs at 100°C to 800°C shown in (b), it can be seen that the trends are basically the same. For example, around 580°C, the difference value tends to increase.

[0240] Similarly, Fig. 15C The difference before and after the change is calculated when the weight percentage of Cu is increased from the weight percentage of the minimum component (=0.6) by a predetermined weight percentage (=0.05) and the weight percentage of Al is reduced by a predetermined weight percentage (=0.05).

[0241] (a) Calculate the Al at each temperature before and after the change based on the above partial differential value 15 The difference in the value of the ratio of (Mn,Fe)3Si2, and

[0242] (b) The Al at each temperature before and after the change was calculated using a simulator (Thermo-Calc (registered trademark)). 15 The case of the difference in the value of the ratio of (Mn,Fe)3Si2 (here, the range of the vertical axis in (b) is narrow).

[0243] Similarly, Fig.15D The graph is a diagram showing the difference before and after the change when the weight percentage of Mn is increased by a predetermined weight percentage (=0.05) from the weight percentage of the smallest component (=0.2) and the weight percentage of Al is decreased by a predetermined weight percentage (=0.05).

[0244] (a) Calculate the Al at each temperature before and after the change based on the above partial differential value 15 The difference in the value of the ratio of (Mn,Fe)3Si2, and

[0245] (b) The Al at each temperature before and after the change was calculated using a simulator (Thermo-Calc (registered trademark)). 15 The case of the difference in the value of the ratio of (Mn,Fe)3Si2 (here, the range of the vertical axis in (b) is narrow).

[0246] Will Fig.15D The transition of the graph at 100℃~800℃ shown in (a) and Fig.15D The trend is basically the same as that of the graphs in (b) at 100°C to 800°C. For example, in the range of 100°C to 580°C, there are 15 The value of the ratio of (Mn,Fe)3Si2 increases in this tendency.

[0247] Similarly, Fig.15E (a) is a diagram showing the difference before and after the change when the weight percentage of Mg is increased from the weight percentage of the minimum component (=0.8) by a predetermined weight percentage (=0.05) and the weight percentage of Al is decreased by a predetermined weight percentage (=0.05), and the difference between the weight percentage of Mg and the weight percentage of Al is calculated.

[0248] (a) Calculate the Al at each temperature before and after the change based on the above partial differential value 15 The difference in the value of the ratio of (Mn,Fe)3Si2, and

[0249] (b) The Al at each temperature before and after the change was calculated using a simulator (Thermo-Calc (registered trademark)). 15 The case of the partial differential value of the ratio of (Mn,Fe)3Si2 (here, the range of the vertical axis in (b) is narrow).

[0250] Similarly, Fig.15F The figure shows the difference before and after the change when the weight percentage of Cr is increased by a predetermined weight percentage (=0.05) from the weight percentage of the minimum component (=0.01) and the weight percentage of Al is decreased by a predetermined weight percentage (=0.05).

[0251] (a) Calculate the Al at each temperature before and after the change based on the above partial differential value 15 The difference in the value of the ratio of (Mn,Fe)3Si2, and

[0252] (b) The Al at each temperature before and after the change was calculated using a simulator (Thermo-Calc (registered trademark)). 15 The case of the difference in the value of the ratio of (Mn,Fe)3Si2 (here, the range of the vertical axis in (b) is narrow).

[0253] Similarly, Figure 15GThe difference before and after the change is calculated when the weight percentage of Zn is increased from the weight percentage of the minimum component (=0.01) to a predetermined weight percentage (=0.05) and the weight percentage of Al is decreased to a predetermined weight percentage (=0.05).

[0254] (a) Calculate the Al at each temperature before and after the change based on the above partial differential value 15 The difference in the value of the ratio of (Mn,Fe)3Si2, and

[0255] (b) The Al at each temperature before and after the change was calculated using a simulator (Thermo-Calc (registered trademark)). 15 The case of the difference in the value of the ratio of (Mn,Fe)3Si2 (here, the range of the vertical axis in (b) is narrow).

[0256] Similarly, Fig.15H The difference before and after the change is calculated when the weight percentage of Ti is increased by a predetermined weight percentage (=0.05) from the weight percentage of the minimum component (=0.01) and the weight percentage of Al is decreased by a predetermined weight percentage (=0.05).

[0257] (a) Calculate the Al at each temperature before and after the change based on the above partial differential value 15 The difference in the value of the ratio of (Mn,Fe)3Si2, and

[0258] (b) The Al at each temperature before and after the change was calculated using a simulator (Thermo-Calc (registered trademark)). 15 The case of the difference in the value of the ratio of (Mn,Fe)3Si2 (here, the range of the vertical axis in (b) is narrow).

[0259] From the above, we can see that Fig.14 The illustrated heat map 1400 reproduces the trend of added elements (eg, Fe and Mn) with larger differences before and after the change.

[0260] <Conclusion>

[0261] As can be clearly seen from the above description, the prediction device 120 of the first embodiment

[0262] a storage unit for storing a learned prediction model, the learned prediction model being a model obtained by learning using learning data, the learning data being data in which a material composition of a learning target material and a phase content at each temperature of the learning target material within a predetermined temperature range are associated with each other;

[0263] When the phase content at each temperature of the prediction target material within a predetermined temperature range is predicted by inputting the material composition of the prediction target material into the learned prediction model, a value obtained by predicting a pre-specified phase at each temperature is partially differentiated using the material composition of the prediction target material,

[0264] In a heat map in which each component of a prediction target material is arranged on a first axis and each temperature within a predetermined temperature range is arranged on a second axis, each region determined based on each component and each temperature is displayed with a color corresponding to a value obtained by partial differentiation.

[0265] According to the first embodiment, the degree of influence on the phase content when the material composition changes can be easily grasped.

[0266] [Second embodiment]

[0267] In the first embodiment, the color scheme of the heat map is Fig.14 However, it may be a color scheme other than the color scheme shown in the label 1410. In addition, a display method other than the color scheme may be used to display the heat map.

[0268] In addition, in the first embodiment, the following case is described, that is, by using a heat map to represent the calculated partial differential value, it is easy for the designer to understand in which temperature range the phase content will be affected by increasing or decreasing the weight percentage of any added element. However, the method of representing the calculated partial differential value is not limited to the heat map. For example, in the case where it is desired to understand at which temperature within a predetermined temperature range the degree of influence is greater, a graph showing the transition of the degree of influence caused by the temperature difference may be displayed instead of the heat map.

[0269] In addition, in the first embodiment, as a specific example of material composition, the alloy composition representing the weight percentage of the added element (other metal element or non-metal element) relative to the metal element contained in the alloy is described, but the material composition is not limited to the alloy composition. For example, it may also be a chemical composition representing the weight percentage of each chemical component contained in a substance other than the alloy.

[0270] In addition, in the first embodiment described above, as a learned prediction model, a model using either Seq2Seq or Transformer with an Attention mechanism is described. However, the learned prediction model is not limited thereto, and other architectures may be applied as long as they are architectures that can process data at each predetermined time interval, i.e., time series data. Specifically, RNN (Recurrent Neural Network), Bidirectional RNN, Seq2Seq, etc. may also be applied. Alternatively, GRU (Gated Recurrent Unit), LSTM (Long Short Term Memory), etc. may also be applied.

[0271] Furthermore, in the above-mentioned first embodiment, the learning device 110 and the prediction device 120 are separate structures, but the learning device 110 and the prediction device 120 may be an integrated structure.

[0272] It should be noted that the present invention is not limited to the configurations listed in the above embodiments and the combination with other elements, etc. These can be changed within the scope not departing from the gist of the present invention and can be appropriately determined according to the application form.

[0273] This application claims priority based on Japanese Patent Application No. 2022-211219 filed on December 28, 2022, and all contents of the Japanese Patent Application are incorporated herein by reference.

[0274] [Explanation of Reference Numerals]

[0275] 100: Prediction System

[0276] 110: Learning device

[0277] 111: Learning data generation unit

[0278] 112: Learning Department

[0279] 120: Prediction Device

[0280] 121: Material composition input department

[0281] 122: Forecasting Department

[0282] 123: Display unit

[0283] 300: Learning Data

[0284] 401: Element information input

[0285] 402: Combination determination unit

[0286] 403: Simulation Department

[0287] 404: Storage control unit

[0288] 501: Encoder unit

[0289] 502: Decoder

[0290] 503: Loss function calculation unit

[0291] 801: Phase content error calculation section at the formation / disappearance temperature

[0292] 802: Phase content error calculation section

[0293] 803: Phase content error (logarithmic value) calculation part

[0294] 804: Phase content error (difference value) calculation part

[0295] 805: Cross entropy error calculation unit

[0296] 806: Weighted summation unit

[0297] 1001: The encoder part has been learned

[0298] 1002: The decoder has been learned

[0299] 1003: Phase Designation Department

[0300] 1004: Sensitivity characteristic calculation unit

[0301] 1400: Heat map.

Claims

1. A prediction device, comprising: A storage unit stores a learned model, wherein the learned model is a model obtained by learning using learning data, wherein the learning data is data in which a material composition of a material of a learning object and a phase content at each temperature of the material of the learning object within a predetermined temperature range are associated, and the learned model is configured to predict the phase content at the i+1th temperature using the phase content predicted by the learned model for temperatures up to the i-th temperature (i is an integer greater than 1) within the predetermined temperature range, wherein When the phase content at each temperature of the prediction target material within the predetermined temperature range is predicted by inputting the material composition of the prediction target material into the learned model, a value obtained by predicting a pre-specified phase at each temperature is partially differentiated using the material composition of the prediction target material. In a heat map in which each component of the material of the prediction object is arranged on the first axis and each temperature within the predetermined temperature range is arranged on the second axis, each region determined based on the components and the temperatures is displayed in a display method corresponding to the value obtained by performing the partial differentiation.

2. The prediction device according to claim 1, wherein: The learned model uses a framework capable of processing data at each predetermined time interval, ie, time series data, and the learned model predicts the phase content at each predetermined temperature interval based on the material composition of the material of the prediction object.

3. The prediction device according to claim 2, wherein: The learned model is any one of RNN, Bidirectional RNN, Seq2Seq, Seq2Seq with Attention mechanism, GRU, LSTM and Transformer.

4. The prediction device according to claim 3, wherein: The learned model has: an encoder unit that outputs a feature value by inputting a material component of the material of the prediction target; and The decoder unit predicts the phase content at the (i+1)th temperature by inputting the outputted feature amount and the predicted phase content at the temperatures up to the i-th temperature.

5. The prediction device according to any one of claims 1 to 4, wherein: The phase content is the phase content in a thermodynamic equilibrium state.

6. The prediction device according to any one of claims 1 to 5, wherein: When predicting the phase content at the (i+1)th temperature, a value obtained by predicting the predetermined phase is partially differentiated using the material composition of the material to be predicted, In the heat map, each region determined based on the components and the (i+1)th temperature is displayed in a display manner corresponding to the value obtained by performing the partial differentiation.

7. A prediction method for executing a learned model by a computer, wherein: The learned model is a model obtained by learning using learning data, wherein the learning data is data in which a material composition of a material to be learned and a phase content at each temperature of the material to be learned within a predetermined temperature range are associated, and the learned model is configured to predict the phase content at the i+1th temperature using the phase content predicted by the learned 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 computer executes: When predicting the phase content at each temperature of the material to be predicted within the predetermined temperature range by inputting the material composition of the material to be predicted into the learned model, a step of partially differentiating a value obtained by predicting a pre-specified phase at each temperature using the material composition of the material to be predicted; and In a heat map in which each component of the material of the prediction object is arranged on the first axis and each temperature within the predetermined temperature range is arranged on the second axis, a step of displaying each area determined based on the each component and the each temperature in a display method corresponding to the value obtained by performing the partial differentiation.

8. A prediction program for causing a computer to execute a learned model, wherein: The learned model is a model obtained by learning using learning data, wherein the learning data is data in which a material composition of a material to be learned and a phase content at each temperature of the material to be learned within a predetermined temperature range are associated, and the learned model is configured to predict the phase content at the i+1th temperature using the phase content predicted by the learned 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 computer is caused to execute: When predicting the phase content at each temperature of the material to be predicted within the predetermined temperature range by inputting the material composition of the material to be predicted into the learned model, a step of partially differentiating a value obtained by predicting a pre-specified phase at each temperature using the material composition of the material to be predicted; and In a heat map in which each component of the material of the prediction object is arranged on the first axis and each temperature within the predetermined temperature range is arranged on the second axis, a step of displaying each area determined based on the each component and the each temperature in a display method corresponding to the value obtained by performing the partial differentiation.

Images