Prediction result analysis device, analysis method, and analysis program

Abstract

The degree of influence on the phase fraction due to the change in the material composition can be easily grasped. A prediction result analysis device includes: a storage unit that stores a trained model trained using training data in which a material composition of a material to be trained and a phase fraction are associated; an intermediate composition generation unit that generates a material composition of the intermediate material; a prediction unit that, when predicting a phase fraction by inputting the material composition of the plurality of intermediate materials into the trained model, calculates a partial differential value by performing partial differential on a predicted value of a specified phase using the material composition of the intermediate materials for each temperature; an integral gradient calculation unit that calculates an integral gradient for each temperature and each component by integrating the calculated partial differential value for each temperature for each component included in the material composition of the intermediate material; and a display unit that displays, in a heat map in which each component is set on a first axis and each temperature is set on a second axis, each region determined on the basis of each component and each temperature in a display manner corresponding to the integral gradient.

Description

Technical Field

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

[0002] In material design, it is very important to calculate the phase fraction in a thermodynamic equilibrium state within a specific temperature range, and we are developing proxy models that predict the phase fraction based on material composition information. As an example, we are developing a proxy model that predicts the phase fraction based on the content of added elements in the alloy.

[0003] Here, the phase fraction in a specific temperature range is related to the alloy properties. Therefore, when the designer uses the above model to design the alloy, he should first understand that increasing or decreasing the content of any added element will affect any phase in any temperature range, so that the desired phase fraction can be approached, which is very important for efficient alloy design.

[0004] Patent Document 1: International Publication No. 2020 / 090617 Summary of the invention

[0005] <Problems to be Solved by the Invention>

[0006] In contrast, for example, if a designer performs a task of predicting the phase fractions in a specific temperature range while sequentially increasing or decreasing the contents of all added elements, and then compares each predicted phase fraction with the phase fraction before the change, the designer can grasp in advance the degree of influence of the contents of each added element on the phase fraction by performing such a task.

[0007] However, for alloys with a large amount of added elements, such work is cumbersome for designers. In addition, there is also a problem that it is difficult for designers to understand the degree of influence of the content of each added element on the phase fraction by simply comparing the phase fractions.

[0008] An object of the present invention is to enable a designer to easily understand the degree of influence on the phase fraction caused by a change in material composition.

[0009] <Methods used to solve the problem>

[0010] A prediction result analysis device according to a first aspect of the present invention comprises:

[0011] a storage unit storing a trained model trained using training data, wherein the material composition of a material to be trained in the training data is associated with the phase fraction of the material to be trained at each temperature in a specific temperature range, the trained model predicting the phase fraction at an i+1th temperature using the phase fraction predicted by the trained model for each temperature up to an i-th temperature in the specific temperature range, wherein i is an integer greater than or equal to 1;

[0012] an intermediate composition generating unit that divides the material composition of the material to be predicted and the material composition of the material to be a baseline to generate a plurality of material compositions of intermediate materials;

[0013] a prediction unit that, when predicting the phase fraction of each of the plurality of intermediate materials in the specific temperature range at each temperature by inputting the material composition of each of the plurality of intermediate materials into the trained model, calculates a partial differential value for each temperature by partially differentiating a value predicted for a pre-specified phase using the material composition of each of the plurality of intermediate materials;

[0014] an integrated gradient calculation unit that calculates an integrated gradient for each temperature and each component by integrating the calculated partial differential value for each temperature according to each component included in the material composition of the plurality of intermediate materials; and

[0015] A display unit displays, in a heat map in which each component is arranged on a first axis and each temperature is arranged on a second axis, each region determined based on the each component and each temperature in a display mode corresponding to the integrated gradient.

[0016] A second aspect of the present invention is a prediction result analysis device according to the first aspect, wherein:

[0017] The integrated gradient calculation unit calculates the integrated gradient for each temperature and each component by integrating the calculated partial differential values ​​for each component included in the material composition of the plurality of intermediate materials for each temperature and averaging the calculated partial differential values.

[0018] A third aspect of the present invention is a prediction result analysis device according to the second aspect, wherein:

[0019] When integrating the calculated partial differential values ​​for each component included in the material composition of the plurality of intermediate materials at each temperature, the integrated gradient calculation unit integrates using composition weights calculated in advance for each of the plurality of intermediate materials.

[0020] A fourth aspect of the present invention is a prediction result analysis device according to the third aspect, wherein:

[0021] The integrated gradient calculation unit calculates the composition weight based on a difference between a material composition of each of the plurality of intermediate materials and a material composition of the prediction target material.

[0022] A fifth aspect of the present invention is a prediction result analysis device according to any one of the first to fourth aspects, wherein:

[0023] The intermediate composition generating unit generates the material composition of the plurality of intermediate materials based on the material composition of each of the plurality of materials serving as the baselines,

[0024] The integrated gradient calculation unit calculates the integrated gradient at each temperature and each component for the material composition of each of the plurality of materials serving as the baselines,

[0025] The display unit displays the heat map according to a display mode corresponding to an averaged integrated gradient obtained by averaging a plurality of the integrated gradients for each temperature and each component.

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

[0027] The trained model adopts a framework capable of processing data at specific time intervals, namely, time series data. The trained model predicts the phase fraction at specific temperature intervals based on the material composition of the material as the prediction object.

[0028] A seventh aspect of the present invention is a prediction result analysis device according to the sixth aspect, wherein:

[0029] The trained model is any one of RNN, bidirectional RNN, Seq2Seq, Seq2Seq with attention mechanism, GRU, LSTM and Transformer.

[0030] An eighth aspect of the present invention is a prediction result analysis device according to the seventh aspect, wherein:

[0031] The trained model has:

[0032] an encoder unit that outputs a feature quantity by inputting a material composition of a material to be predicted; and

[0033] The decoder unit predicts the phase fraction at the (i+1)th temperature by receiving the output feature amount and the predicted phase fractions at each temperature up to the i-th temperature.

[0034] A ninth aspect of the present invention is a prediction result analysis device according to any one of the first to eighth aspects, wherein:

[0035] The phase fraction is the phase fraction in a thermodynamic equilibrium state.

[0036] The analysis method of the tenth aspect of the present invention is an analysis method in which a trained model trained using training data is executed by a computer, wherein the material composition of a material as a training object in the training data is associated with the phase fraction of the material as the training object at each temperature in a specific temperature range, and the trained model uses the phase fraction predicted by the trained model for each temperature up to the i-th temperature in the specific temperature range to predict the phase fraction at the i+1-th temperature, wherein i is an integer greater than 1,

[0037] In the analysis method, the computer executes the following steps:

[0038] dividing the material composition of the material to be predicted and the material composition of the material as the baseline to generate material compositions of a plurality of intermediate materials;

[0039] When predicting the phase fraction of each of the plurality of intermediate materials at each temperature within the specific temperature range by inputting the material compositions of the plurality of intermediate materials into the trained model, for each temperature, a partial differential value is calculated by partially differentiating a value predicted for a pre-specified phase using the material composition of each of the plurality of intermediate materials;

[0040] calculating an integrated gradient of each temperature and each component by integrating the calculated partial differential value for each component included in the material composition of the plurality of intermediate materials for each temperature; and

[0041] In a heat map in which each component is arranged on a first axis and each temperature is arranged on a second axis, each region determined based on the each component and each temperature is displayed in a display mode corresponding to the integrated gradient.

[0042] The prediction program of the 11th aspect of the present invention is a prediction program for causing a computer to execute a trained model trained using training data, wherein the material composition of a material as a training object in the training data is associated with the phase fraction of the material as the training object at each temperature in a specific temperature range, and the trained model uses the phase fraction predicted by the trained model for each temperature up to the i-th temperature in the specific temperature range to predict the phase fraction at the i+1-th temperature, wherein i is an integer greater than 1,

[0043] The analysis program is used to cause the computer to execute the following steps:

[0044] dividing the material composition of the material to be predicted and the material composition of the material as the baseline to generate material compositions of a plurality of intermediate materials;

[0045] When predicting the phase fraction of each of the plurality of intermediate materials at each temperature within the specific temperature range by inputting the material compositions of the plurality of intermediate materials into the trained model, for each temperature, a partial differential value is calculated by partially differentiating a value predicted for a pre-specified phase using the material composition of each of the plurality of intermediate materials;

[0046] calculating an integrated gradient of each temperature and each component by integrating the calculated partial differential value for each component included in the material composition of the plurality of intermediate materials for each temperature; and

[0047] In a heat map in which each component is arranged on a first axis and each temperature is arranged on a second axis, each region determined based on the each component and each temperature is displayed in a display mode corresponding to the integrated gradient.

[0048] <Effects of the Invention>

[0049] According to the present invention, a designer can easily understand the degree of influence on the phase fraction caused by a change in material composition. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] 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 training device, a prediction device, and a prediction result analysis device.

[0051] Figure 2 This is a diagram showing an example of the hardware configuration of a training device, a prediction device, and a prediction result analysis device.

[0052] Figure 3 It is a diagram showing a specific example of the structure of training data, input data, and correct answer data.

[0053] Figure 4 This is a diagram showing an example of the functional structure of the training data generating unit.

[0054] Figure 5 This is a diagram showing an example of the functional structure of a training unit.

[0055] Figure 6 FIG. 1 is a first diagram showing an operation example of the training unit.

[0056] Figure 7 FIG. 2 is a second diagram showing an example of the operation of the training unit.

[0057] Figure 8This is a diagram showing an example of a method for calculating a total loss using a loss function calculation unit.

[0058] Fig. 9 It is a flowchart which shows the flow of a training process.

[0059] Fig.10 FIG1 is a first diagram showing an example of a functional structure of a prediction unit.

[0060] Fig.11 FIG1 is a first diagram showing an example of the operation of the prediction unit.

[0061] Fig.12 It is a flowchart which shows the flow of prediction processing.

[0062] Fig.13 It is a diagram for explaining the prediction accuracy of the phase fraction.

[0063] Fig.14 FIG1 is a first diagram showing a specific example of processing by the prediction result analysis device.

[0064] Fig.15 FIG. 2 is a second diagram showing an example of a functional structure of a prediction unit.

[0065] Fig.16A FIG. 2 is a second diagram showing an example of the operation of the prediction unit.

[0066] Fig. 16B FIG3 is a third diagram showing an example of the operation of the prediction unit.

[0067] Fig. 16C FIG. 4 is a fourth diagram showing an example of the operation of the prediction unit.

[0068] Fig.17 FIG. 2 is a second diagram showing a specific example of processing by the prediction result analysis device.

[0069] Fig.18 This is a flowchart showing the flow of prediction result analysis processing.

[0070] Fig.19 This is a flowchart showing the flow of baseline composition information input processing.

[0071] Fig. 20 : is a flowchart showing the flow of the integrated gradient calculation process.

[0072] Fig.21 This is a diagram showing an example of the material composition of the material to be predicted and the predicted phase fraction used in the prediction result analysis process.

[0073] Fig. 22 This is a diagram showing an example of a heat map displayed in a display format based on an integrated gradient. DETAILED DESCRIPTION

[0074] Hereinafter, each embodiment will be described with reference to the drawings. It should be noted that in the specification and the drawings, components having substantially the same functional structure are denoted by the same reference numerals, and repeated descriptions are omitted.

[0075] [First embodiment]

[0076] <System structure of prediction system, functional structure of training device, prediction device and prediction result analysis device>

[0077] First, the system structure of the prediction system including the prediction result analysis device of the first embodiment, and the functional structure of the training device, the prediction device and the prediction result analysis device of the prediction system are described. In this embodiment, the prediction system is a system that predicts the phase fraction at each temperature in a specific temperature range based on material composition information. The phase fraction mentioned here is the phase fraction in the thermodynamic equilibrium state, and the phase fraction in the thermodynamic equilibrium state is referred to as "phase fraction" hereinafter.

[0078] 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 training device, a prediction device, and a prediction result analysis device. Figure 1 As shown, the prediction system 100 includes a training device 110 , a prediction device 120 and a prediction result analysis device 130 .

[0079] A training program is installed in the training device 110 , and by executing the program, the training device 110 functions as a training data generating unit 111 and a training unit 112 .

[0080] After generating training data for training the prediction model, the training data generating unit 111 stores the generated training data in the training data storage unit 113. It should be noted that the prediction model referred to here is a surrogate model that replaces the previous model with machine learning. In this embodiment, the training data used for training the prediction model is stored in the training data storage unit 113.

[0081] · Multiple material composition information with different combinations of the content ratios of each component,

[0082] · For materials with their own material composition information, the phase fraction at each temperature within a specific temperature range,

[0083] The two are stored in association.

[0084] After the training unit 112 reads the training data from the training data storage unit 113, it uses the read training data to train the prediction model. The training unit 112 trains the prediction model so that the output data output by inputting the material composition information stored in the training data into the prediction model is close to the phase fraction at each temperature in the specific temperature range stored in association with the training data. Thus, the training unit 112 generates a trained prediction model. The training unit 112 stores the generated trained prediction model in the storage unit of the prediction unit 122 of the prediction device 120 and the storage unit of the prediction unit 134 of the prediction result analysis device 130.

[0085] It should be noted that, in this embodiment, the training unit 112 uses the following architecture as the prediction model:

[0086] Seq2Seq with attention mechanism,

[0087] Transformer,

[0088] These frameworks are capable of processing data at specific time intervals, i.e., time series data, and are generally used in the fields of natural language processing, etc. Therefore, according to the prediction model, by inputting the material composition information stored in the training data, it is possible to sequentially output output data corresponding to the phase fraction at each temperature within a specific temperature range.

[0089] Here, “outputting output data corresponding to the phase fraction at each temperature in sequence” means, for example, that the prediction model outputs output data corresponding to the phase fraction at the i+1th temperature by using the correct answer data at each temperature up to the i-th temperature. It should be noted that i is an integer greater than 1.

[0090] In this way, the training unit 112 has a structure that sequentially outputs output data corresponding to the phase fraction at each temperature. That is, the training unit 112 has a structure that processes the output data corresponding to the phase fraction at each temperature as time series data. Thus, according to the training unit 112, training that reflects the output data corresponding to the phase fraction between adjacent temperature regions can be performed.

[0091] A prediction program is installed in the prediction device 120 , and by executing the program, the prediction device 120 functions as a material composition input unit 121 , a prediction unit 122 , and a display unit 123 .

[0092] The material composition input unit 121 receives input of material composition information of a material to be predicted, and notifies the prediction unit 122 of the information.

[0093] The prediction unit 122 stores the trained prediction model trained by the training unit 112 in the storage unit. After reading the trained prediction model from the storage unit, the prediction unit 122 inputs the material composition information notified from the material composition input unit 121 into the trained prediction model, thereby sequentially predicting the phase fraction at each temperature in the specific temperature interval. That is, the prediction unit 122 sequentially predicts the phase fraction in the specific temperature interval according to the specific temperature interval.

[0094] Here, "predicting the phase fraction sequentially at specific temperature intervals" means, for example, that by using the phase fraction predicted by the trained prediction model for each temperature up to the i-th temperature (i is an integer greater than 1), the trained prediction model predicts the phase fraction at the i+1-th temperature.

[0095] Thus, in the present embodiment, the prediction unit 122 has a structure for predicting the phase fraction at each temperature in sequence. That is, the prediction unit 122 has a structure for processing the phase fraction at each temperature as time series data using a recursive network. Thus, according to the prediction unit 122, it is possible to predict the prediction data reflecting the phase fraction between adjacent temperature regions. Therefore, for example, compared with the case where a multilayer neural network is applied (i.e., the case where a network that does not reflect the prediction result of the phase fraction between adjacent temperature regions is applied), the decrease in the prediction accuracy of the phase fraction can be suppressed. In other words, according to the prediction unit 122, the prediction accuracy when predicting the phase fraction throughout a specific temperature range based on the material composition can be improved.

[0096] The display unit 123 displays the phase fraction at each temperature in the specific temperature range predicted by the prediction unit 122. It should be noted that, in the display unit 123, the phase fraction at each temperature in the specific temperature range predicted by the prediction unit 122 is displayed by color classification for each phase. However, the phase fraction at each temperature in the specific temperature range predicted by the prediction unit 122 may also be displayed by changing the type of line such as a solid line, a dotted line, a dashed line, etc. for each phase.

[0097] An analysis program is installed in the prediction result analysis device 130. When the program is executed, the prediction result analysis device 130 functions as a material composition input unit 131, a baseline composition input unit 132, an intermediate composition generation unit 133, a prediction unit 134, an integrated gradient calculation unit 135, and a display unit 136.

[0098] The material composition input unit 131 receives input of material composition information of the material to be predicted, and notifies the intermediate composition generation unit 133 of the information. It should be noted that the material composition information of the material to be predicted input to the material composition input unit 131 is the same as the material composition information of the material to be predicted input to the material composition input unit 121 of the prediction device 120.

[0099] The baseline composition input unit 132 receives input of material composition information of a material serving as a baseline, i.e., baseline composition information, and notifies the intermediate composition generation unit 133 of the information. The baseline composition information is material composition information used to calculate gradients when visualizing prediction evidence using the integrated gradients method.

[0100] For example, it is preferable to use representative material composition information of the material to be predicted, or material composition information including the upper limit or lower limit of the composition of the material to be predicted, as the baseline composition information. Specifically, in the case of 6000 series aluminum alloys, there are A6061, A6063, etc. as JIS standards, and each standard stipulates the upper limit and lower limit of the composition of the added elements. In this case, it is preferable to use the material composition information including the upper limit or lower limit of the added element in the standard to which the material to be predicted belongs as the baseline composition information.

[0101] Furthermore, when focusing on a specific precipitated phase, material composition information known to produce the specific precipitated phase may also be used as baseline composition information.

[0102] By using multiple baseline composition information, the degree of influence on the phase fraction can be appropriately evaluated even at a phase fraction saturated with a specific added element.

[0103] The intermediate composition generation unit 133 generates material composition information of the intermediate material, that is, intermediate composition information, based on the material composition information of the material as the prediction object notified from the material composition input unit 131 and the baseline composition information notified from the baseline composition input unit 132. Specifically, the intermediate composition generation unit 133 generates the intermediate composition information by, for example, dividing the material composition information of the material as the prediction object and the baseline composition information by the number of steps = M. Thus, according to the intermediate composition generation unit 133, M pieces of intermediate composition information gradually approaching the baseline composition information can be generated based on the material composition information as the prediction object.

[0104] The intermediate composition generating unit 133 sequentially inputs the material composition information of the generated plurality of intermediate materials, that is, M pieces of intermediate composition information, to the predicting unit 134 .

[0105] The prediction unit 134 stores the trained prediction model trained by the training unit 112 in the storage unit. After reading the trained prediction model from the storage unit, the prediction unit 134 inputs the M intermediate composition information notified from the intermediate composition generation unit 133 into the trained prediction model. Thus, the prediction unit 134 sequentially predicts the phase fraction at each temperature in the specific temperature interval. That is, the prediction unit 122 sequentially predicts the phase fraction in the specific temperature interval according to the specific temperature interval for the M intermediate composition information.

[0106] When predicting the phase fraction in a specific temperature interval in sequence according to the specific temperature interval, the prediction unit 134 uses the M intermediate composition information to perform partial differentiation on the value predicted for the pre-specified phase at each temperature. Thus, according to the prediction unit 134, by changing each of the M intermediate composition information by a small amount, it is possible to calculate a "partial differential value" indicating the degree of change of the pre-specified phase for each temperature.

[0107] The integrated gradient calculation unit 135 calculates the integrated gradient for each temperature and each component by integrating the partial differential value of each of the M intermediate composition information for each temperature and each component.

[0108] In the heat map (where each component is set on the horizontal axis and each temperature is set on the vertical axis), the display unit 136 colors each area determined based on each component and each temperature in a display method corresponding to the integrated gradient. It should be noted that the horizontal axis is an example of the first axis and the vertical axis is an example of the second axis. The display unit 136 displays the heat map colored in a display method corresponding to the integrated gradient.

[0109] Therefore, the designer of the material design can easily understand how to change the material composition so as to approach the desired phase fraction with respect to the phase fraction displayed by the prediction device 120 based on the heat map displayed by the prediction result analysis device 130 .

[0110] That is, according to the prediction result analysis device 130, when a designer of a material design predicts the phase fraction based on the material composition information, the designer can easily grasp the degree of influence of changing the material composition on the phase fraction based on the heat map.

[0111] <Hardware Structure of Training Device, Prediction Device, and Prediction Result Analysis Device>

[0112] Next, the hardware structure of the training device 110, the prediction device 120, and the prediction result analysis device 130 is described. It should be noted that since the training device 110, the prediction device 120, and the prediction result analysis device 130 have the same hardware structure, Figure 2 The hardware structures of the training device 110 , the prediction device 120 , and the prediction result analysis device 130 are summarized and described.

[0113] Figure 2 FIG. 1 is a diagram showing an example of the hardware configuration of a training device, a prediction device, and a prediction result analysis device. Figure 2 As shown, the training device 110, the prediction device 120 and the prediction result analysis device 130 include a processor 201, a memory 202, an auxiliary storage device 203, an interface device 204, a communication device 205 and a driver device 206. It should be noted that the hardware of the training device 110, the prediction device 120 and the prediction result analysis device 130 are connected to each other via a bus 207.

[0114] 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 to the memory 202 and executes them. The various programs mentioned here include, for example, training programs, prediction programs, analysis programs, and the like.

[0115] The memory 202 includes a main storage device such as a ROM (Read Only Memory) and a RAM (Random Access Memory). The processor 201 and the memory 202 form a so-called computer. The computer realizes the above-mentioned various functions by executing various programs read by the processor 201 to the memory 202.

[0116] The auxiliary storage device 203 stores various programs and various data used when the processor 201 executes the various programs. For example, the training data storage unit 113 is implemented in the auxiliary storage device 203. Alternatively, a storage unit storing a trained prediction model is implemented in the auxiliary storage device 203.

[0117] The interface 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.

[0118] The drive device 206 is a device for placing the recording medium 213. The recording medium 213 includes a medium that optically, electrically or magnetically records information such as a CD-ROM, a floppy disk and a magneto-optical disk. The recording medium 213 may include a medium such as a semiconductor memory that electrically records information such as a ROM, a flash memory, etc.

[0119] It should be noted that the various programs installed in the auxiliary storage device 203 are installed, for example, by setting the distributed recording medium 213 in the drive device 206 and having the drive device 206 read the various programs recorded in the recording medium 213. Alternatively, the various programs installed in the auxiliary storage device 203 may be downloaded from the network via the communication device 205.

[0120] <Structure of training data, input data, and correct answer data>

[0121] Next, a specific example of the structure of training data and input data and correct answer data included in the training data will be described. Figure 3 It is a diagram showing a specific example of the structure of training data and input data and correct answer data.

[0122] like Figure 3 As shown, the training data 300 includes "input data" and "correct answer data" as information items.

[0123] The "input data" stores "material composition information 1", "material composition information 2", etc., which are material composition information with different combinations of the content ratios of the components. Figure 3 , figure mark 301 represents a specific example of "material composition information 1", and represents a specific example of the components constituting "6013" which is the name of a known aluminum alloy standard, specifically, the content of each added element Si, Fe, Cu, Mn, Mg, Cr, Zn and Ti.

[0124] Similarly, figure mark 302 represents a specific example of "material composition information 2" and represents the various components constituting "6060" which is the name of a known aluminum alloy standard, specifically, the contents of each added element Si, Fe, Cu, Mn, Mg, Cr, Zn and Ti.

[0125] In the "correct answer data", the phase fractions "phase fraction 1", "phase fraction 2", etc. at each temperature in a specific temperature range corresponding to "material composition information 1", "material composition information 2", etc. are stored. Figure 3 In the example of , the specific temperature range is 100° C. to 800° C., and 100° C. and 800° C. are both included in the specific temperature range.

[0126] exist Figure 3 In the figure, reference numeral 311 indicates a specific example of "phase fraction 1", and reference numeral 312 indicates a specific example of "phase fraction 2". In all the specific examples, the horizontal axis indicates temperature, and the vertical axis indicates phase fraction.

[0127] Here, reference numerals 311 and 312 are results obtained by running a high-cost simulator with high prediction accuracy for a long period of time, and in the present embodiment, the results are used as correct answer data.

[0128] It should be noted that the simulator mentioned here refers to software that calculates the phase fraction at each temperature in a specific temperature range by thermodynamic equilibrium calculation. Specifically, the simulator includes software such as CaTCalc and MatCalc. Alternatively, the simulator may include software such as Termosuite, FactStage, and Pandat. Alternatively, the simulator may include software such as MALT2, Thermo-Calc, and OpenCalphad.

[0129] <Details of each unit of the training device>

[0130] Next, details of the training data generating unit 111 and the training unit 112 as each unit of the training device 110 will be described.

[0131] (1) Functional structure of training data generation unit

[0132] First, the functional configuration of the training data generating unit 111 will be described. Figure 4 This is a diagram showing an example of the functional structure of the training data generating unit.

[0133] like Figure 4 As shown, the training data generating unit 111 includes an element information input unit 401 , a combination determining unit 402 , a simulation unit 403 , and a storage control unit 404 .

[0134] The element information input unit 401 receives input of each component included in the material composition information on the type of additive element added when producing a specific alloy, for example. The element information input unit 401 receives input of the upper limit value and the lower limit value of the content rate of each additive element received input.

[0135] The combination determination unit 402 determines a plurality of combinations of the content of each added element by randomly selecting the content of each added element received as input under the constraints of the upper limit value and the lower limit value. It should be noted that the combination of the content of each added element can be created by setting the content between the upper limit value and the lower limit value using a specific step size. The combination determination unit 402 notifies the simulation unit 403 of the material composition information represented by the determined plurality of combinations.

[0136] The simulation unit 403 calculates the phase fraction at each temperature in the specific temperature range by operating the simulator for each of the plurality of material composition information notified from the combination determination unit 402. The simulation unit 403 notifies the storage control unit 404 of the plurality of material composition information and the corresponding phase fraction at each temperature in the specific temperature range.

[0137] The storage control unit 404 generates training data using the plurality of material composition information notified from the simulation unit 403 as “input data” and the phase fraction at each temperature in the corresponding specific temperature range as “correct answer data”. The storage control unit 404 stores the generated training data in the training data storage unit 113.

[0138] (2) Functional structure of training units

[0139] Next, the functional structure of the training unit 112 will be described. Figure 5 This is a diagram showing an example of the functional structure of a training unit.

[0140] like Figure 5 As shown, the training unit 112 has a prediction model and a loss function calculation unit 503. It should be noted that the prediction model is a proxy model that uses machine learning to replace the previous model. As mentioned above, the prediction model uses Seq2Seq or Transformer with an attention mechanism, including an encoder unit 501 and a decoder unit 502.

[0141] Encoder section 501 receives input of “material composition information 1”, “material composition information 2”, etc. stored in “input data” of training data 300 , and outputs a feature value.

[0142] The decoder unit 502 receives the feature quantity output by the encoder unit 501, and outputs the output data in sequence. Specifically, the decoder unit 502 reads out the "phase fraction 1", "phase fraction 2", etc. stored in the "correct answer data" of the training data 300. Then, the decoder unit 502 uses the phase fraction at each temperature up to the i-th temperature (i.e., the correct answer data at each temperature up to the i-th temperature) to output the output data of the i+1-th temperature. Thus, the decoder unit 502 can output the output data of the i+1-th temperature based on, for example,

[0143] The feature quantity output from the encoder unit 501 by inputting "material composition information 1",

[0144] · "Correct answer data" of "Phase fraction 1",

[0145] The output data corresponding to the phase fraction at each temperature within a specific temperature range is outputted in sequence. Similarly, the decoder unit 502 can be based on, for example,

[0146] The feature quantity output from the encoder unit 501 by inputting "material composition information 2",

[0147] · "Correct answer data" for "Phase fraction 2"

[0148] Output data corresponding to the phase fraction at each temperature within a specific temperature range is output sequentially.

[0149] It should be noted that the Softmax function (normalized exponential function) is used for the output layer of the decoder unit 502. Thus, the multiple values ​​included in the output data of the decoder unit 502, that is, the values ​​representing each ratio, have a range of greater than 0 and less than 1, and the sum of the multiple values, that is, the sum of the values ​​representing each ratio is "1".

[0150] In this way, by using the Softmax function in the output layer of the decoder unit 502, output data corresponding to the phase fraction at each temperature within a specific temperature range can be output in sequence without performing additional calculations.

[0151] The loss function calculation unit 503 reads out “phase fraction 1”, “phase fraction 2”, etc. stored in the “correct answer data” of the training data 300. The loss function calculation unit 503 compares the read “phase fraction 1”, “phase fraction 2”, etc. with the output data at the i+1th temperature outputted from the decoder unit 502.

[0152] In contrast, the loss function calculation unit 503 calculates multiple losses, and then performs weighted addition on the calculated multiple losses to calculate the comprehensive loss. In addition, 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. Thus, a trained prediction model is generated. It should be noted that the trained prediction model includes a trained encoder unit and a trained decoder unit.

[0153] (3) Example 1 of the training unit's actions

[0154] Next, an operation example of the training unit 112 will be described. Specifically, among the prediction model and the loss function calculation unit, an operation example of the prediction model will be mainly described. Figure 6 FIG. 1 is a first diagram showing an operation example of the training unit.

[0155] in, Figure 6 (a) shows a state where “input data” of training data 300 is read out and input to encoder section 501 , and feature quantities are output from encoder section 501 . Figure 6 (a) shows a situation where the feature quantity output from the encoder unit 501 is input to the decoder unit 502, and a start signal is input and output data is output from the decoder unit 502. Figure 6 In the example of (a), the output data output by the decoder unit 502 is equivalent to the phase fraction at 800°C. The output data output from the decoder unit 502, specifically, the output data equivalent to the phase fraction at 800°C, is notified to the loss function calculation unit 503. It should be noted that the start signal is a signal for starting the operation of the decoder unit 502. In the present embodiment, an arbitrary value different from the output data output from the decoder unit 502 is input to the decoder unit 502 as the start signal. An arbitrary value different from the output data can be, for example, a value such as "000000".

[0156] Figure 6 (b) shows a case 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 correct solution data at 800°C to the decoder unit 502. Figure 6 In the example of (b), the output data outputted from decoder section 502 corresponds to the phase fraction at 790° C. The output data outputted from decoder section 502 , specifically, the output data corresponding to the phase fraction at 790° C., is notified to loss function calculation section 503 .

[0157] Figure 6 (c) shows a case 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 correct solution data of the phase fraction at 790°C to the decoder unit 502. Figure 6 In (c), due to space constraints, only the correct answer data at 790° C. are described. In reality, the correct answer data at 800° C. and the correct answer data at 790° C. are input to the decoder unit 502 with weights added thereto.

[0158] exist Figure 6 In the example of (c), the output data outputted from the decoder unit 502 corresponds to the phase fraction at 780° C. The output data outputted from the decoder unit 502 , particularly the output data corresponding to the phase fraction at 780° C., is notified to the loss function calculation unit 503 .

[0159] Figure 6 (d) shows a case where the decoder unit 502 outputs output data by inputting the feature amount output from the encoder unit 501 to the decoder unit 502 and inputting the correct solution data having a phase fraction up to 120°C to the decoder unit 502. Figure 6 In (d) of FIG. 5 , due to space constraints, only the correct answer data at 120° C. are described. In practice, the correct answer data at 800° C. to 120° C. are weighted and input to the decoder unit 502 .

[0160] exist Figure 6 In the example of (d), the output data output by decoder section 502 corresponds to the phase fraction at 110° C. The output data output from decoder section 502 , specifically, the output data corresponding to the phase fraction at 110° C., is notified to loss function calculation section 503 .

[0161] Figure 6 (e) shows a case 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 correct solution data of the phase fraction up to 110°C to the decoder unit 502. Figure 6 In (e) of FIG. 5 , due to space constraints, only the correct answer data at 110° C. are described. In reality, each correct answer data from 800° C. to 110° C. is input to the decoder unit 502 with weights added thereto.

[0162] exist Figure 6 In the example of (e), the output data outputted from decoder section 502 corresponds to the phase fraction at 100° C. The output data outputted from decoder section 502 , particularly the output data corresponding to the phase fraction at 100° C., is notified to loss function calculation section 503 .

[0163] (4) Training unit action example 2

[0164] Next, an operation example of the training unit 112 will be described. Specifically, among the prediction model and loss function calculation units, an operation example of the loss function calculation unit 503 will be mainly described. Figure 7 FIG. 2 is a second diagram showing an example of the operation of the training unit. Figure 7 As shown, the loss function calculation unit 503 reads the “correct answer data” of the training data 300 . Figure 7 The example shows that the phase fraction at each temperature in the specific temperature range contained in "Phase Fraction 1" is read out, that is:

[0165] · Phase fraction at 800°C as correct answer data,

[0166] ·The phase fraction at 790°C as the correct answer data,

[0167] · Phase fraction at 100°C as correct answer data,

[0168] situation.

[0169] exist Figure 7 The example represents the following situation, in which, based on the feature quantity output from the encoder unit 501 when the "material composition information 1" is input, the output data is sequentially output from the decoder unit 502, and as a result, in the loss function calculation unit 503,

[0170] · Phase fraction at 800°C as output data,

[0171] · Phase fraction at 790°C as output data, ...

[0172] · Phase fraction at 100°C as output data,

[0173] is maintained.

[0174] Figure 7 The example represents the following situation, in which the loss function calculation unit 503 calculates the comprehensive loss after comparing the phase fraction at each temperature of 100°C to 800°C as the correct solution data with the phase fraction at each temperature of 100°C to 800°C as the output data. Figure 7 The example 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 comprehensive loss. It should be noted that the functional structure of the loss function calculation unit 503 for calculating the comprehensive loss is described in detail below.

[0175] (5) Details of the functional structure of the loss function calculation unit

[0176] Figure 8 FIG. 1 is a diagram showing an example of a method for calculating a comprehensive loss using a loss function calculation unit. Figure 8 As shown, the loss function calculation unit 503 has the following functions as functions for calculating various losses:

[0177] Phase fraction error calculation unit 801 at generation temperature and disappearance temperature,

[0178] Phase fraction error calculation unit 802,

[0179] Phase ratio error (logarithmic value) calculation unit 803,

[0180] Phase ratio error (difference) calculation unit 804,

[0181] Cross entropy error calculation unit 805,

[0182] Weighted addition unit 806.

[0183] The phase fraction error calculation unit 801 at the generation temperature and the disappearance temperature determines the temperature at which the curve of each phase fraction in the correct solution data of the training data is 0. The phase fraction error calculation unit 801 at the generation temperature and the disappearance temperature compares the phase fraction at each temperature of 100°C to 800°C as the correct solution data and the phase fraction at each temperature of 100°C to 800°C as the output data at a specific temperature. Thus, the phase fraction error calculation unit 801 at the generation temperature and the disappearance temperature performs a full phase addition of the error between the phase fraction at the temperature at which each phase is generated or disappears determined based on the training data and the phase fraction determined based on the training data. The phase fraction error calculation unit 801 at the generation temperature and the disappearance temperature outputs the first addition result after the full phase addition.

[0184] The phase fraction error calculation unit 802 compares the phase fraction at each temperature from 100° C. to 800° C. as the correct answer data with the phase fraction at each temperature from 100° C. to 800° C. as the output data. Thus, the phase fraction error calculation unit 802 outputs a second addition result obtained by adding the phase fraction errors at each temperature in a specific temperature range.

[0185] The phase fraction error (logarithmic value) calculation unit 803 compares the phase fraction at each temperature of 100° C. to 800° C. as the correct answer data with the phase fraction at each temperature of 100° C. to 800° C. as the output data. Thus, the phase fraction error (logarithmic value) calculation unit 803 outputs a third addition result obtained by adding the errors of the logarithmic values ​​of the phase fraction at each temperature in a specific temperature range in the specific temperature range.

[0186] The phase fraction error (difference) calculation unit 804 compares the phase fraction at each temperature of 100° C. to 800° C. as the correct answer data with the phase fraction at each temperature of 100° C. to 800° C. as the output data. Thus, the phase fraction error (difference) calculation unit 804 outputs a fourth addition result obtained by adding the errors of the differences in the phase fractions between adjacent temperatures in a specific temperature range.

[0187] The cross entropy error calculation unit 805 compares the phase fraction at each temperature from 100° C. to 800° C. as the correct answer data with the phase fraction at each temperature from 100° C. to 800° C. as the output data. Thus, the cross entropy error calculation unit 805 outputs a fifth addition result obtained by adding the errors of the ratios of the phase fractions at each temperature in a specific temperature interval in the specific temperature interval.

[0188] The weighted addition unit 806 calculates the comprehensive loss by weighted addition of the first to fifth addition results output from each unit from the phase fraction error calculation unit 801 at the generation temperature and the disappearance temperature to the cross entropy error calculation unit 805.

[0189] Thus, by adopting a structure of calculating multiple losses and weighted addition thereof, the loss function calculation unit 503 can handle the loss between the output data and the correct answer data in various aspects. As a result, in the training unit 112, when training the prediction model, the model parameters can be appropriately updated.

[0190] <Flow of training process>

[0191] Next, the flow of the training process by the training device 110 will be described. Fig. 9 It is a flowchart which shows the flow of a training process.

[0192] In step S901 , the training data generating unit 111 receives, as input of element information, the type of additional element added when generating a specific alloy and the upper limit value and the lower limit value of the content rate of each additional element.

[0193] In step S902 , the training data generating unit 111 randomly selects the content rate of each additional element based on the constraints of the upper limit value and the lower limit value, thereby determining a plurality of combinations of the content rate of each additional element.

[0194] In step S903 , the training data generating unit 111 operates the simulator for each of the material composition information represented by the determined plurality of combinations, and calculates the phase fraction at each temperature within the specific temperature range.

[0195] In step S904 , after generating the training data, training data generating section 111 stores the generated training data in training data storage section 113 .

[0196] In step S905 , the training unit 112 inputs a start signal to the decoder unit 502 .

[0197] In step S906 , after reading the “input data” of the training data, the training unit 112 inputs the read “input data” into the prediction model.

[0198] In step S907, the training unit 112 sequentially outputs the (i+1)th output data by sequentially inputting the phase fractions at each temperature up to the i-th temperature of the “correct answer data” of the training data, that is, the correct answer data at each temperature up to the i-th temperature.

[0199] In step S908, the training unit 112 determines whether all output data corresponding to the phase fraction at each temperature in the specific temperature range has been output. If it is determined in step S908 that there is a temperature at which output data corresponding to the phase fraction has not been output (in the case of "No" in step S908), the process returns to step S907.

[0200] On the other hand, when it is determined in step S908 that all output data corresponding to the phase fraction at each temperature within the specific temperature range has been output (in the case of YES in step S908 ), the process proceeds to step S909 .

[0201] In step S909 , after reading out the “correct answer data” of the training data, the training unit 112 compares the read out “correct answer data” with the output data.

[0202] In step S910, the training unit 112 calculates multiple losses and then calculates a comprehensive loss by weighted addition of the calculated multiple losses. The training unit 112 updates the model parameters of the encoder unit 501 and the decoder unit 502 based on the calculated comprehensive loss.

[0203] In step S911, the training unit 112 determines whether to continue the training process. When it is determined in step S911 that the training process is to be continued (in the case of "yes" in step S911), the process returns to step S906, and the same process as above is performed by reading the next "input data" of the training data.

[0204] On the other hand, when it is determined in step S911 that the training process is finished (in the case of NO in step S911 ), the process proceeds to step S912 .

[0205] In step S912 , the training unit 112 stores the trained encoder unit and the trained decoder unit as a trained prediction model in the storage unit of the prediction device 120 and the storage unit of the prediction result analysis device 130 .

[0206] <Details of each unit of the prediction device>

[0207] Next, the prediction unit 122 of the prediction device 120 will be described in detail.

[0208] (1) Functional structure of prediction unit

[0209] First, the functional structure of the prediction unit 122 will be described. Fig.10 FIG1 is a first diagram showing an example of a functional structure of a prediction unit.

[0210] like Fig.10 As shown, 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.

[0211] The trained encoder unit 1001 calculates a feature amount by inputting material composition information of a material as a prediction target notified from the material composition input unit 121 , and outputs the calculated feature amount to the trained decoder unit 1002 .

[0212] The trained decoder unit 1002 sequentially outputs prediction data by inputting the feature quantity output from the trained encoder unit 1001. Specifically, the trained decoder unit 1002 uses the prediction data up to the i-th temperature to output the prediction data of the i+1-th temperature. Therefore, the trained decoder unit 1002 can predict the relative score at each temperature in a specific temperature range based on the feature quantity output from the trained encoder unit 1001 by inputting the material composition information of the material as the prediction object.

[0213] (2) Example of operation of prediction unit

[0214] Next, an example of the operation of the prediction unit 122 will be described. Fig.11 FIG. 1 is a first diagram showing an example of the operation of the prediction unit. Fig.11 (a) shows a case where the feature quantity is output from the trained encoder unit 1001 by inputting material composition information of the material to be predicted into the trained encoder unit 1001 . Fig.11 (a) shows a case where the phase fraction at 800° C. is output by inputting the feature amount output from the trained encoder unit 1001 to the trained decoder unit 1002 and inputting the start signal to the trained decoder unit 1002 .

[0215] Fig.11 (b) shows a case where the feature quantity output from the trained encoder unit 1001 is input to the trained decoder unit 1002, and the phase fraction at 800° C. as prediction data is input to the trained decoder unit 1002. Therefore, the trained decoder unit 1002 outputs the phase fraction at 790° C. as prediction data.

[0216] Fig.11 (c) shows the following situation, in which the feature quantity output from the trained encoder unit 1001 is input to the trained decoder unit 1002, and the phase fraction up to 790°C as prediction data is input to the trained decoder unit 1002. It should be noted that Fig.11 In (c), due to paper space constraints, only the phase fraction at 790°C is recorded. In fact, the phase fraction at 800°C and the phase fraction at 790°C are weighted and input to the trained decoder unit 1002. As a result, the trained decoder unit 1002 outputs the phase fraction at 780°C as prediction data.

[0217] Fig.11 (d) shows the following situation, in which the feature quantity output from the trained encoder unit 1001 is input to the trained decoder unit 1002, and the phase fraction of 120°C as the prediction data is input to the trained decoder unit 1002. It should be noted that Fig.11 In (d), due to the paper condition, only the phase fraction at 120° C. is recorded. In fact, the phase fractions from 800° C. to 120° C. are weighted and input to the decoder unit 1002. Thus, the trained decoder unit 1002 outputs the phase fraction at 110° C. as prediction data.

[0218] Fig.11 (e) shows the following situation, in which the feature quantity output from the trained encoder unit 1001 is input to the trained decoder unit 1002, and the phase fraction of 110°C as the prediction data is input to the trained decoder unit 1002. It should be noted that Fig.11 In (e), due to paper space constraints, only the phase fraction at 110°C is recorded. In fact, the phase fractions from 800°C to 110°C are weighted and input to the trained decoder unit 1002. As a result, the trained decoder unit 1002 outputs the phase fraction at 100°C as prediction data. It should be noted that Fig.11 In the example of (e), the phase fraction at 100° C. outputted from the trained decoder unit 1002 as prediction data by inputting the feature quantity and the phase fraction up to 110° C. as prediction data is also shown.

[0219] <Flow of prediction processing>

[0220] Next, the flow of prediction processing by the prediction device 120 will be described. Fig.12 It is a flowchart which shows the flow of prediction processing.

[0221] In step S1201 , the material composition input unit 121 receives input of material composition information of a material to be predicted.

[0222] In step S1202 , the prediction unit 122 receives the start signal and inputs the input material composition information into the trained prediction model to predict the phase fraction at each temperature within a specific temperature range.

[0223] In step S1203 , the display unit 123 displays the predicted phase fraction at each temperature.

[0224] <Prediction accuracy of phase fraction>

[0225] Next, the prediction accuracy of the phase fraction at each temperature in the specific temperature interval predicted by the prediction unit 122 will be described. Fig.13 It is a diagram for explaining the prediction accuracy of the phase fraction. Fig.13 (a) to (c) of FIG. 1 show prediction data when the content of each added element constituting the name of "6013", which is a known aluminum alloy standard, is input as material composition information of the prediction object. It should be noted that in this embodiment, when evaluating the prediction accuracy, the prediction accuracy is calculated using Figure 3 The correct answer data is shown in the reference numeral 311. In calculating the loss with the correct answer data, the mean square logarithmic error (MSLE: MeansSquared Logarithmic Error) loss is used.

[0226] Fig.13 (a) shows the prediction data when a fully integrated multi-layer neural network is used in the prediction model as a comparative example. Fig.13 In the case of (a), the MSLE loss compared with the correct answer data is 3.79×10 -4 .

[0227] Fig.13 (b) shows the prediction data when Seq2Seq with attention mechanism is used in the prediction model. Fig.13 In the case of (b), the MSLE loss compared with the correct answer data is 5.10×10 -5 .

[0228] Fig.13 (c) shows the prediction data when Transformer is used in the prediction model. Fig.13 In the case of (c), the MSLE loss compared with the correct answer data is 2.45×10 -5 .

[0229] As described above, when a fully coupled multilayer neural network is applied as a prediction model for realizing phase fraction calculation at low cost, the result is that the estimation accuracy is low in some temperature regions (for example, from 400° C. to 600° C.).

[0230] On the other hand, when Seq2Seq with attention mechanism is applied as a prediction model to realize phase fraction calculation at low cost, the loss can be greatly improved. And when Transformer is applied, the phase fraction at the same level as the correct data can be reproduced.

[0231] That is, when predicting the phase fraction over a specific temperature range based on material composition information,

[0232] Using a recurrent network, the phase fraction at each temperature is processed as time series data, and

[0233] Use multiple loss functions to handle losses in many ways.

[0234] This can improve prediction accuracy.

[0235] <Details of prediction result analysis device>

[0236] Next, as details of the prediction result analysis device 130, a specific example of the processing of the material composition input unit 131, the baseline composition input unit 132, the intermediate composition generation unit 133, the integrated gradient calculation unit 135, and the functional structure and operation example of the prediction unit 134 are explained.

[0237] (1) Specific example of processing of material composition input unit, baseline composition input unit, and intermediate composition generation unit

[0238] First, among the units included in the prediction result analysis device 130 , a specific example of the processing of the material composition input unit 131 , the baseline composition input unit 132 , and the intermediate composition generation unit 133 will be described. Fig.14 FIG1 is a first diagram showing a specific example of processing by the prediction result analysis device.

[0239] like Fig.14 As shown, the material composition input unit 131 receives the input of "material composition information X" and uses it as the material composition information of the material to be predicted. The baseline composition input unit 132 receives the input of the number "N" of baseline composition information and receives the input of multiple baseline composition information, specifically N baseline composition information, namely "baseline composition information 1" to "baseline composition information N".

[0240] The intermediate composition generation unit 133 generates M pieces of intermediate composition information by dividing the material composition information of the material as the prediction target and each of the N pieces of baseline composition information by a predetermined number of steps=M. Fig.14 The example represents the following situation:

[0241] By dividing the "material composition information X" and the "baseline composition information 1" by the number of steps = M, "intermediate composition information Xto1_1" to "intermediate composition information Xto1_M" are generated;

[0242] By dividing the "material composition information X" and the "baseline composition information 2" by the number of steps = M, "intermediate composition information Xto2_1" to "intermediate composition information Xto2_M" are generated;

[0243] …

[0244] By dividing the “material composition information X” and the “baseline composition information N” by the number of steps = M, “intermediate composition information XtoN_1” to “intermediate composition information XtoN_M” are generated.

[0245] (2) Functional structure of prediction unit

[0246] Next, the functional configuration of the prediction unit 134 will be described in detail. Fig.15 FIG. 2 is a second diagram showing an example of a functional structure of a prediction unit.

[0247] like Fig.15 As shown, the prediction unit 134 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.

[0248] The functions of the trained encoder unit 1001 and the trained decoder unit 1002 are the same as those of the decoder unit 1001, except that the intermediate composition information is input instead of the material composition information as the prediction object. Fig.10 The functions of the trained encoder unit 1001 and the trained decoder unit 1002 are the same. Therefore, the description is omitted here.

[0249] like Fig.15 As shown, the prediction unit 134 further includes a phase specifying unit 1501 and a sensitivity characteristic calculating unit 1502. The phase specifying unit 1501 and the sensitivity characteristic calculating unit 1502 form an input sensitivity analyzing unit.

[0250] The phase specifying unit 1501 receives input of the phase of the analysis object specified by the material designer. The phase of the analysis object specified by the material designer is the phase whose degree of change when the material designer changes the components included in the material composition at each temperature is expected to be known in advance. The phase specifying unit 1501 notifies the sensitivity characteristic calculating unit 1502 of the phase of the analysis object whose input is received.

[0251] When the predicted data at the (i+1)th temperature is output, the sensitivity characteristic calculation unit 1502 partially differentiates the value predicted for the phase notified from the phase specification unit 1501 using the intermediate composition information notified from the intermediate composition generation unit 133 to derive a partial differential value.

[0252] More specifically, the sensitivity characteristic calculation unit 1502 calculates the sensitivity characteristic by using the following partial differential value:

[0253] · Partial differential values ​​in each layer, which are calculated by processing in each layer of the trained encoder unit 1001 during the period from input of intermediate component information to output of feature quantities; and

[0254] The partial differential value in each layer is calculated by processing in each layer of the trained decoder unit 1002 from the input feature quantity until the output of the predicted value for the specified phase at the i+1th temperature,

[0255] The “partial differential value obtained by partially differentiating the value at the (i+1)th temperature predicted for the phase of the designated analysis target by using the intermediate composition information” is derived.

[0256] Furthermore, the sensitivity characteristic calculation unit 1502 notifies the integrated gradient calculation unit 135 of “a partial differential value obtained by partially differentiating the value of the (i+1)th temperature predicted for the phase of the designated analysis target by using the intermediate composition information”.

[0257] (3) Example of operation of prediction unit

[0258] Next, a specific example of the processing of the prediction unit 134 is described. FIG. 16A to FIG. 16C FIG. 2 to FIG. 4 are examples of operations of the prediction unit. Fig.16A (a) represents the following situation, in which intermediate composition information generated based on material composition information of a material as a prediction object and baseline composition information is input to the trained encoder unit 1001, thereby outputting a feature quantity from the trained encoder unit 1001. Fig.16A (a) represents a case where a feature quantity output from the trained encoder unit 1001 is input to the trained decoder unit 1002, and a start signal is input to the trained decoder unit 1002, thereby outputting a phase fraction at 800°C.

[0259] also, Fig.16A (a) represents the case where, when the phase fraction at 800°C is output as prediction data, "a partial differential value obtained by partially differentiating the prediction value for the specified phase using the intermediate composition information" is derived for each component at 800°C.

[0260] Fig.16A (b) represents the following situation, in which the feature quantity output from the trained encoder unit 1001 is input to the trained decoder unit 1002, and the phase fraction at 800°C as prediction data is input to the trained decoder unit 1002. Thus, the trained decoder unit 1002 outputs the phase fraction at 790°C as prediction data.

[0261] also, Fig.16A(b) represents the case where, when the phase fraction at 790°C is output as prediction data, "a partial differential value obtained by partially differentiating the prediction value for the specified phase using the intermediate composition information" is derived for each component at 790°C.

[0262] Fig. 16B (a) represents the following situation, in which the feature quantity output from the trained encoder unit 1001 is input to the trained decoder unit 1002, and the phase fraction up to 790°C as prediction data is input to the trained decoder unit 1002. It should be noted that Fig. 16B In (a), due to the paper situation, only the phase fraction at 790°C is recorded. In fact, the phase fraction at 800°C and the phase fraction at 790°C are weighted and input to the trained decoder unit 1002. As a result, the trained decoder unit 1002 outputs the phase fraction at 780°C as prediction data.

[0263] also, Fig. 16B (a) represents the case where, when the phase fraction at 780°C is output as prediction data, "a partial differential value obtained by partially differentiating the prediction value for the specified phase using the intermediate composition information" is derived for each component at 780°C.

[0264] Fig. 16B (b) represents the case where the feature quantity output from the trained encoder unit 1001 is input to the trained decoder unit 1002, and the phase fraction up to 120°C as prediction data is input to the trained decoder unit 1002. Fig. 16B In (b), due to paper space constraints, only the phase fraction at 120°C is recorded. In fact, the phase fractions from 800°C to 120°C are weighted and input to the trained decoder unit 1002. As a result, the trained decoder unit 1002 outputs the phase fraction at 110°C as prediction data.

[0265] also, Fig. 16B (b) represents the case where, when the phase fraction at 110° C. is output as prediction data, “a partial differential value obtained by partially differentiating the prediction value for the specified phase using the intermediate composition information” is derived for each component at 110° C.

[0266] Fig. 16C represents a case where the feature quantity output from the trained encoder unit 1001 is input to the trained decoder unit 1002, and the phase fraction up to 110°C as prediction data is input to the trained decoder unit 1002. Fig. 16CIn the example, due to the paper condition, only the phase fraction at 110°C is recorded. In fact, the phase fractions from 800°C to 110°C are weighted and input to the trained decoder unit 1002. Therefore, the trained decoder unit 1002 outputs the phase fraction at 100°C as the prediction data. In addition, Fig. 16C In the example of , the phase fraction at 100° C. outputted from the trained decoder unit 1002 as prediction data by inputting the feature quantity and the phase fraction up to 110° C. as prediction data is also shown.

[0267] also, Fig. 16C This indicates a case where, when the phase fraction at 100° C. is output as prediction data, “a partial differential value obtained by partially differentiating a prediction value for a designated phase using intermediate composition information” is derived for each component at 100° C.

[0268] (4) Specific example of processing by the prediction unit and the integrated gradient calculation unit

[0269] Next, a specific example of the processing of the prediction unit 134 and the integrated gradient calculation unit 135 among the units included in the prediction result analysis device 130 will be described. Fig.17 FIG. 2 is a second diagram showing a specific example of processing by the prediction result analysis device.

[0270] like Fig.17 As shown, the prediction unit 134 sequentially predicts M phase fractions at each temperature in a specific temperature range by inputting M intermediate composition information notified from the intermediate composition generation unit 133 into the trained prediction model. In addition, the prediction unit 134 derives M partial differential values ​​for each temperature and each component for the phase to be analyzed.

[0271] according to Fig.17 For example, prediction unit 134:

[0272] By inputting M intermediate composition information of "intermediate composition information Xto1_1" to "intermediate composition information Xto1_M", M phase fractions in the case of "baseline composition information 1" are predicted, and M partial differential values ​​are derived for each temperature and each component for the phase to be analyzed;

[0273] By inputting M intermediate composition information, namely, "intermediate composition information Xto2_1" to "intermediate composition information Xto2_M", M phase fractions in the case of "baseline composition information 2" are predicted, and M partial differential values ​​are derived for each temperature and each component for the phase to be analyzed; ...

[0274] By inputting M intermediate composition information "intermediate composition information XtoN_1" to "intermediate composition information XtoN_M", M phase fractions in the case of "baseline composition information N" are predicted, and M partial differential values ​​are derived for each temperature and each component for the phase to be analyzed.

[0275] like Fig.17 As shown, the integrated gradient calculation unit 135 calculates the integrated gradient of each temperature and each component using the M partial differential values ​​derived by the prediction unit 134 for each temperature and each component.

[0276] exist Fig.17 Example representation: Integral gradient calculation unit 135

[0277] · Obtaining the partial differential values ​​of each temperature and each component derived from the "intermediate composition information Xto1_M";

[0278] · A case where partial differential values ​​of each temperature and each component derived from "intermediate composition information Xto1_M-1" are obtained;

[0279] …

[0280] · A case where partial differential values ​​of each temperature and each component derived from “intermediate composition information Xto1_1” are obtained;

[0281] A case where the integral gradient 1710_1 is calculated by averaging after integrating the component weights of M partial differential values ​​calculated in advance for each temperature and each component.

[0282] Likewise, in Fig.17 In the example, it is indicated that: the integrated gradient calculation unit 135

[0283] A case where partial differential values ​​of each temperature and each component derived from the "intermediate composition information Xto2_M" are obtained;

[0284] · Obtaining the partial differential values ​​of each temperature and each component derived from the "intermediate composition information Xto2_M-1";

[0285] …

[0286] · Obtaining the partial differential values ​​of each temperature and each component derived from the "intermediate composition information Xto2_1";

[0287] A case where the integral gradient 1710_2 is calculated by averaging after integrating the component weights of M partial differential values ​​calculated in advance for each temperature and each component.

[0288] Likewise, in Fig.17 In the example: Integral gradient calculation unit 135

[0289] · Obtaining partial differential values ​​of each temperature and each component derived from "intermediate composition information XtoN_M";

[0290] · Obtaining the partial differential values ​​of each temperature and each component derived from the "intermediate composition information XtoN_M-1";

[0291] …

[0292] · A case where partial differential values ​​of each temperature and each component derived from “intermediate composition information XtoN_1” are obtained;

[0293] For each temperature and each component, M partial differential values ​​are integrated using the composition weight calculated in advance, and then the integrated gradient 1710_N is calculated by averaging.

[0294] also, Fig.17 The example represents a case where the integrated gradient calculation unit 135 calculates a final integrated gradient, ie, an averaged integrated gradient, for a phase to be analyzed by calculating an average value of a plurality of integrated gradients 1710_1 to 1710_N for each temperature and each component.

[0295] <Flow of forecast result analysis and processing>

[0296] Next, the flow of the prediction result analysis process performed by the prediction result analysis device 130 will be described. Fig.18 This is a flowchart showing the flow of prediction result analysis processing.

[0297] In step S1801 , the material composition input unit 131 receives input of material composition information of a material to be predicted.

[0298] In step S1802, the baseline composition input unit 132 accepts input of baseline composition information. A detailed flowchart of the baseline composition information input process will be described later.

[0299] In step S1803 , the prediction unit 134 receives input of the phase of the analysis target.

[0300] In step S1804, the intermediate composition generation unit 133 generates intermediate composition information based on the material composition information and the baseline composition information of the material to be predicted. The prediction unit 134 predicts the phase fraction based on the generated intermediate composition information, and derives a partial differential value for each temperature and each component of the phase to be analyzed. The integral gradient calculation unit 135 integrates the derived partial differential value for each temperature and each component, thereby calculating the integral gradient for each temperature and each component. A detailed flowchart of the integral gradient calculation process performed by the intermediate composition generation unit 133, the prediction unit 134, and the integral gradient calculation unit 135 will be described later.

[0301] In step S1805 , the display unit 136 displays a heat map of the integrated gradient for the phase serving as the analysis target.

[0302] Therefore, according to the prediction result analysis device 130, the degree of influence of the content rate of each added element on the phase fraction can be grasped only by inputting a plurality of pieces of baseline composition information and the phase to be analyzed.

[0303] In other words, it is not necessary to sequentially increase or decrease the contents of all the added elements while predicting the phase fraction in a specific temperature range, and it is easy to grasp the degree of influence of the content of each added element on the phase fraction.

[0304] <Detailed flow of baseline composition information input processing>

[0305] Next, the detailed flow of the baseline composition information input process (step S1802) will be described. Fig.19 This is a flowchart showing the flow of baseline composition information input processing.

[0306] In step S1901 , the base line composition input unit 132 accepts a setting regarding the number (=N) of base line composition information from a designer of a material design.

[0307] In step S1902 , the base line composition input unit 132 sets a counter i that counts the number of base line composition information to “1”.

[0308] In step S1903, the baseline composition input unit 132 accepts input of i-th baseline composition information from the designer of the material design.

[0309] In step S1904 , the baseline composition input unit 132 determines whether the counter i has reached N. When it is determined in step S1904 that the counter i has not reached N (in the case of NO in step S1904 ), the process proceeds to step S1905 .

[0310] In step S1905, the baseline composition input unit 132 increments the counter i, and then returns to step S1903.

[0311] On the other hand, when it is determined in step S1905 that the counter i has reached N (in the case of YES in step S1905 ), the base line composition information input process is terminated.

[0312] <Detailed flow of integrated gradient calculation processing>

[0313] Next, the detailed flow of the integrated gradient calculation process (step S1804) is described. Fig. 20 : is a flowchart showing the flow of the integrated gradient calculation process.

[0314] In step S2001 , the intermediate composition generation unit 133 sets a counter i that counts the number of baseline composition information to “1”.

[0315] In step S2002 , the intermediate composition generation unit 133 generates M pieces of intermediate composition information by dividing the material composition information of the prediction target material and the i-th baseline composition information by the number of steps=M.

[0316] In step S2003 , the intermediate composition generation unit 133 sets a counter j that counts the number of generated intermediate composition information to “1”.

[0317] In step S2004, the intermediate composition generation unit 133 calculates the composition weight of the j-th intermediate composition information. The intermediate composition generation unit 133 calculates the composition weight based on the difference between the material composition information of the material as the prediction target and the j-th intermediate composition information.

[0318] In step S2005 , the prediction unit 134 predicts the phase fraction at each temperature within a specific temperature range by inputting the j-th intermediate composition information into the trained prediction model.

[0319] In step S2006 , the prediction unit 134 performs partial differentiation on the phase to be analyzed using the j-th intermediate composition information, thereby deriving partial differential values ​​of each temperature and each component.

[0320] In step S2007 , the integrated gradient calculation unit 135 corrects the partial differential values ​​derived for each temperature and each component respectively using the weights.

[0321] In step S2008 , the intermediate composition generating unit 133 determines whether the counter j counting the number of intermediate composition information has reached M. When it is determined in step S2008 that the counter j has not reached M (in the case of NO in step S2008 ), the process proceeds to step S2009 .

[0322] In step S2009 , the intermediate composition generating unit 133 increments the counter j, and returns to step S2004 .

[0323] On the other hand, when it is determined in step S2008 that the counter j has reached M (in the case of YES in step S2008 ), the process proceeds to step S2010 .

[0324] In step S2010, the integrated gradient calculation unit 135 integrates the M corrected partial differential values ​​derived for each temperature and each component, and then averages them by dividing them by M. Therefore, the integrated gradient calculation unit 135 calculates the integrated gradient for each temperature and each component corresponding to the i-th baseline composition information.

[0325] In step S2011 , the intermediate composition generating unit 133 determines whether the counter i counting the number of baseline composition information has reached N. When it is determined in step S2011 that the counter i has not reached N (in the case of NO in step S2011 ), the process proceeds to step S2012 .

[0326] In step S2012 , the intermediate composition generating unit 133 increments the counter i, and then returns to step S2002 .

[0327] On the other hand, when it is determined in step S2011 that the counter i has reached N (in the case of YES in step S2011 ), the process proceeds to step S2013 .

[0328] In step S2013, the N integrated gradients of each temperature and each component are integrated for each temperature and each component and then averaged by dividing by N. Therefore, the integrated gradient calculation unit 135 can calculate the final integrated gradient, that is, the averaged integrated gradient, for the material composition information of the material to be predicted.

[0329] <Example of heat map>

[0330] Next, we will refer to Fig.21 and Fig. 22 The prediction result analysis process ( Fig.18 ) is a specific example of a heat map of the integrated gradient calculated. Fig.21 This is a diagram showing an example of the material composition of the prediction target and the predicted phase fraction used in the prediction result analysis process. Fig. 22 This is a diagram showing an example of a heat map displayed in a display format corresponding to the integrated gradient.

[0331] like Fig.21As shown in FIG. 1 , when the material composition information to be predicted is the material composition information indicated by reference numeral 2110, the prediction unit 122 predicts reference numeral 2120 as the phase fraction at each temperature from 100° C. to 800° C. On the other hand, if the Al 15 (Mn,Fe)3Si2 is set as the phase to be analyzed, and the prediction result analysis device 130 calculates the thermal map of the integrated gradient, and then, for example, Fig. 22 Heat maps shown in (a) and (b).

[0332] It should be noted that Fig. 22 A and Fig. 22 B shows the heat map when "1" is set as the number of baseline composition information (=N). Fig. 22 A and Fig. 22 The baseline composition information entered in B is different from each other.

[0333] By reference Fig. 22 It can be understood that if the additive elements (Si, Fe, Cu, Mn, Mg, Cr, Zn, Ti) contained in the material to be predicted increase or decrease,

[0334] Each added element has an effect on the phase to be analyzed, namely Al 15 The influence degree of (Mn,Fe)3Si2,

[0335] ·At which temperature is the effect?

[0336] As described above, according to the prediction result analysis device 130 , since the heat map is displayed in a display format corresponding to the integrated gradient calculated for each temperature and each component, the designer can easily understand the degree of influence of the content of each added element on the phase fraction.

[0337] That is, compared with the case of comparing the phase fractions before and after the content is changed, it is easier for the designer to understand the degree of influence of the content of each added element on the phase fraction.

[0338] exist Fig. 22 In the examples of (a) and (b), the number of baseline composition information is set to "1". However, by further increasing the number of baseline composition information, the degree of influence of each added element on the phase of the analysis target can be calculated with higher accuracy.

[0339] In addition, although Fig. 22 In the examples (a) and (b), the number of steps = M is not mentioned, but by further increasing the number of steps = M, the degree of influence of each added element on the phase to be analyzed can be calculated with higher accuracy.

[0340] Summary

[0341] It can be clearly seen from the above description that the prediction result analysis device 130 of the first embodiment:

[0342] A storage unit is provided for storing a trained model trained using training data, wherein the material composition of a training target material and the phase fraction of the training target material at each temperature within a specific temperature range are associated with each other;

[0343] · The material composition of the material to be predicted and the material composition of the material to be used as the baseline are divided to generate the material compositions of a plurality of intermediate materials;

[0344] Predicting the phase fractions of multiple intermediate materials at each temperature within a specific temperature range by inputting the material compositions of the multiple intermediate materials into the trained model;

[0345] When predicting the phase fraction, partial differential values ​​are calculated by partially differentiating the values ​​predicted for the pre-specified phases using the material compositions of the plurality of intermediate materials for each temperature;

[0346] Calculating an integrated gradient for each temperature and each component by integrating the calculated partial differential value for each component included in the material composition of the plurality of intermediate materials for each temperature; and

[0347] In a heat map in which each component is placed on a first axis and each temperature is placed on a second axis, each region determined based on each component and each temperature is displayed in a display format corresponding to an integrated gradient.

[0348] Therefore, according to the first embodiment, a designer can easily understand the degree of influence on the phase fraction caused by a change in the material composition.

[0349] [Second embodiment]

[0350] In the first embodiment, the color scheme of the heat map is Fig. 22 The color schemes shown in (a) and (b) of the heat map can also be Fig. 22 Alternatively, the heat map may be displayed in a color scheme other than the color schemes shown in (a) and (b).

[0351] In the first embodiment described above, the case of expressing the calculated integral gradient using a heat map is described so that the designer can easily understand which added element content is increased or decreased to affect the phase fraction in which temperature range. However, the method of expressing the calculated integral gradient is not limited to a heat map. For example, in the case where it is desired to understand which temperature in a specific temperature range has a greater degree of influence, a graph showing the transition of the degree of influence caused by the temperature difference may be displayed instead of a heat map.

[0352] In the first embodiment described above, as a specific example of the material composition, the alloy composition indicating the content of the additional element, specifically other metal elements or non-metal elements, 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 be a chemical composition indicating the content of each chemical component contained in a substance other than the alloy.

[0353] In the first embodiment described above, as a trained prediction model, the case of using any one of the architectures of Seq2Seq and Transformer with an attention mechanism is described. However, the trained prediction model is not limited to models using these architectures, and may also be a model using other architectures, as long as it is an architecture that can process data at specific time intervals, i.e., time series data. Specifically, the trained prediction model may also be a model using architectures such as RNN (Recurrent Neural Network), bidirectional RNN, Seq2Seq, etc. Optionally, the trained prediction model may also be a model using architectures such as GRU (Gated Recurrent Unit), LSTM (Long Short Term Memory), etc.

[0354] In the first embodiment, the training device 110, the prediction device 120, and the prediction result analysis device 130 are configured as separate devices, but the training device 110, the prediction device 120, and the prediction result analysis device 130 may be configured as one device. Alternatively, any two of the training device 110, the prediction device 120, and the prediction result analysis device 130 may be configured as one device.

[0355] It should be noted that the present invention is not limited to the structures listed in the above embodiments and the combinations with other elements shown here. These points can be changed within the scope of the present invention and can be appropriately specified according to the application form.

[0356] This application claims priority based on Japanese invention patent application No. 2023-038608 filed on March 13, 2023, and all contents of the Japanese patent application are incorporated herein by reference.

[0357] Description of Reference Numerals

[0358] 100: Prediction System

[0359] 110: Training device

[0360] 111: Training data generation unit

[0361] 112: Training Unit

[0362] 120: Prediction device

[0363] 121: Material composition input unit

[0364] 122: Prediction unit

[0365] 123: Display unit

[0366] 130: Prediction result analysis device

[0367] 131: Material composition input unit

[0368] 132: Baseline composition input unit

[0369] 133: Intermediate composition generation unit

[0370] 134: Prediction unit

[0371] 135: Integral gradient calculation unit

[0372] 136: Display unit

[0373] 300: training data

[0374] 401: Element information input unit

[0375] 402: Combination decision unit

[0376] 403: Simulation unit

[0377] 404: Storage Control Unit

[0378] 501: Encoder unit

[0379] 502: Decoder unit

[0380] 503: Loss function calculation unit

[0381] 801: Phase fraction error calculation unit at generation temperature / disappearance temperature

[0382] 802: Phase ratio error calculation unit

[0383] 803: Phase ratio error (logarithmic value) calculation unit

[0384] 804: Phase ratio error (difference value) calculation unit

[0385] 805: Cross entropy error calculation unit

[0386] 806: Weighted addition unit

[0387] 1001: Trained encoder unit

[0388] 1002: Trained decoder unit

[0389] 1501: Phase designation unit

[0390] 1502: Sensitivity characteristic calculation unit.

Claims

1. A prediction result analysis device, comprising: a storage unit storing a trained model trained using training data, wherein the material composition of a material to be trained in the training data is associated with the phase fraction of the material to be trained at each temperature in a specific temperature range, the trained model predicting the phase fraction at an i+1th temperature using the phase fraction predicted by the trained model for each temperature up to an i-th temperature in the specific temperature range, wherein i is an integer greater than or equal to 1; an intermediate composition generating unit that divides the material composition of the material to be predicted and the material composition of the material to be a baseline to generate a plurality of material compositions of intermediate materials; a prediction unit that, when predicting the phase fraction of each of the plurality of intermediate materials in the specific temperature range at each temperature by inputting the material composition of each of the plurality of intermediate materials into the trained model, calculates a partial differential value for each temperature by partially differentiating a value predicted for a pre-specified phase using the material composition of each of the plurality of intermediate materials; an integrated gradient calculation unit that calculates an integrated gradient for each temperature and each component by integrating the calculated partial differential value for each temperature according to each component included in the material composition of the plurality of intermediate materials; and A display unit displays, in a heat map in which each component is arranged on a first axis and each temperature is arranged on a second axis, each region determined based on the each component and each temperature in a display mode corresponding to the integrated gradient.

2. The prediction result analysis device according to claim 1, wherein: The integrated gradient calculation unit calculates the integrated gradient for each temperature and each component by integrating the calculated partial differential values ​​for each component included in the material composition of the plurality of intermediate materials for each temperature and averaging the calculated partial differential values.

3. The prediction result analysis device according to claim 2, wherein: When integrating the calculated partial differential values ​​for each component included in the material composition of the plurality of intermediate materials at each temperature, the integrated gradient calculation unit integrates using composition weights calculated in advance for each of the plurality of intermediate materials.

4. The prediction result analysis device according to claim 3, wherein: The integrated gradient calculation unit calculates the composition weight based on a difference between a material composition of each of the plurality of intermediate materials and a material composition of the prediction target material.

5. The prediction result analysis device according to any one of claims 1 to 4, wherein: The intermediate composition generating unit generates the material composition of the plurality of intermediate materials based on the material composition of each of the plurality of materials serving as the baselines, The integrated gradient calculation unit calculates the integrated gradient at each temperature and each component for the material composition of each of the plurality of materials serving as the baselines, The display unit displays the heat map according to a display mode corresponding to an averaged integrated gradient obtained by averaging a plurality of the integrated gradients for each temperature and each component.

6. The prediction result analysis device according to any one of claims 1 to 5, wherein: The trained model adopts a framework capable of processing data at specific time intervals, namely, time series data. The trained model predicts the phase fraction at specific temperature intervals based on the material composition of the material as the prediction object.

7. The prediction result analysis device according to claim 6, wherein: The trained model is any one of RNN, bidirectional RNN, Seq2Seq, Seq2Seq with attention mechanism, GRU, LSTM and Transformer.

8. The prediction result analysis device according to claim 7, wherein: The trained model has: The encoder unit outputs a feature quantity by inputting a material composition of a material to be predicted; as well as The decoder unit predicts the phase fraction at the (i+1)th temperature by receiving the output feature amount and the predicted phase fractions at each temperature up to the i-th temperature.

9. The prediction result analysis device according to any one of claims 1 to 8, wherein: The phase fraction is the phase fraction in a thermodynamic equilibrium state.

10. An analysis method, wherein a computer executes a trained model trained using training data, wherein the material composition of a training object material in the training data is associated with the phase fraction of the training object material at each temperature in a specific temperature range, and the trained model predicts the phase fraction at the i+1th temperature using the phase fraction predicted by the trained model for each temperature up to the i-th temperature in the specific temperature range, wherein: i is an integer greater than 1, In the analysis method, the computer executes the following steps: dividing the material composition of the material to be predicted and the material composition of the material as the baseline to generate material compositions of a plurality of intermediate materials; When predicting the phase fraction of each of the plurality of intermediate materials at each temperature within the specific temperature range by inputting the material compositions of the plurality of intermediate materials into the trained model, for each temperature, a partial differential value is calculated by partially differentiating a value predicted for a pre-specified phase using the material composition of each of the plurality of intermediate materials; calculating an integrated gradient of each temperature and each component by integrating the calculated partial differential value for each component included in the material composition of the plurality of intermediate materials for each temperature; as well as In a heat map in which each component is arranged on a first axis and each temperature is arranged on a second axis, each region determined based on the each component and each temperature is displayed in a display mode corresponding to the integrated gradient.

11. An analysis program for causing a computer to execute a prediction program of a trained model trained using training data, wherein the material composition of a training object material in the training data is associated with the phase fraction of the training object material at each temperature in a specific temperature range, and the trained model predicts the phase fraction at the i+1th temperature using the phase fraction predicted by the trained model for each temperature up to the i-th temperature in the specific temperature range, wherein: i is an integer greater than 1, The analysis program is used to cause the computer to execute the following steps: dividing the material composition of the material to be predicted and the material composition of the material as the baseline to generate material compositions of a plurality of intermediate materials; When predicting the phase fraction of each of the plurality of intermediate materials at each temperature within the specific temperature range by inputting the material compositions of the plurality of intermediate materials into the trained model, for each temperature, a partial differential value is calculated by partially differentiating a value predicted for a pre-specified phase using the material composition of each of the plurality of intermediate materials; calculating an integrated gradient of each temperature and each component by integrating the calculated partial differential value for each component included in the material composition of the plurality of intermediate materials for each temperature; as well as In a heat map in which each component is arranged on a first axis and each temperature is arranged on a second axis, each region determined based on the each component and each temperature is displayed in a display mode corresponding to the integrated gradient.

Images