Training data generation device, training data generation method, and training data generation program

Abstract

The present invention reduces imbalance in simulation patterns when generating training data for constructing a surrogate model. This training data generation device generates training data for constructing a surrogate model, and has: a determination unit that randomly determines each of an initial value of input data and a final value of the input data; a path generation unit that generates a path by randomly allocating, to each step in a path from the determined initial value to the determined final value, the amount of change that is possible in said step; and a training data generation unit that generates training data including the determined initial value and final value, and the amount of change allocated to each step in the generated path.

Description

Technical Field

[0001] The present disclosure relates to a learning data generation device, a learning data generation method, and a learning data generation program.

Background Art

[0002] In recent years, as an alternative to physical simulation, the use of predictive processing using a surrogate model has been promoted. A surrogate model refers to, for example, a model constructed by learning simulation results when performing physical simulation using a simulation device that reproduces a manufacturing process. By performing predictive processing using the surrogate model, the calculation cost can be significantly reduced as compared with the case of performing physical simulation using a simulation device.

[0003] Here, when performing physical simulation and generating learning data for the purpose of constructing a surrogate model, for example, it is important to perform physical simulation under various simulation patterns generated based on the random walk theory or the like from the viewpoint of the prediction accuracy of the surrogate model.

Prior Art Documents

Non-Patent Documents

[0004]

Non-Patent Document 1

Summary of the Invention

[0005] However, when generating simulation patterns based on random walk theory, there is a problem in that as the number of simulation patterns increases, newly generated simulation patterns tend to be biased towards the average value of the simulation patterns generated up to that point.

[0006] This disclosure reduces the bias in simulation patterns when generating training data for building surrogate models. [Means for solving the problem]

[0007] The first aspect of this disclosure is, A training data generator that generates training data for constructing a surrogate model, A determination unit randomly determines the initial value of the input data and the final value of the input data, respectively. A path generation unit generates a path by randomly assigning to each step the possible amount of change that each step can take in the path from the determined initial value to the final value, The system includes a learning data generation unit that generates learning data including the determined initial and final values ​​and the amount of change assigned to each step of the generated path.

[0008] A second aspect of this disclosure is a learning data generation apparatus as described in the first aspect, The system further includes an extraction unit that extracts paths from the generated paths such that the total change obtained by adding the change amounts assigned to each step is equal to the difference between the determined initial value and the final value. The aforementioned training data generation unit is: Training data is generated that includes the determined initial and final values, and the amount of change assigned to each step of the extracted path.

[0009] A third aspect of this disclosure is a learning data generation apparatus as described in the first aspect, The system further includes a first setting unit that accepts an initial range of the input data and a final range of the input data, respectively. The determination unit randomly determines the initial value of the input data and the final value of the input data, respectively, within the range received by the first setting unit.

[0010] A fourth aspect of this disclosure is a learning data generation apparatus as described in the first aspect, It further has a second setting unit that accepts the number of steps, The route generation unit randomly assigns the possible change amounts for each step to each of the number of steps received by the second setting unit.

[0011] A fifth aspect of this disclosure is a learning data generation device as described in the first aspect, The system further includes a third setting unit that accepts the possible amounts of change in each of the aforementioned steps, The route generation unit randomly assigns the change amounts received by the third setting unit to each step.

[0012] A sixth aspect of this disclosure is a learning data generation device as described in the second aspect, The system further includes an acquisition unit that acquires the simulation results when the determined initial and final values ​​and the change amounts assigned to each step of the extracted path are input to a simulation device as input data. The aforementioned training data generation unit is: Training data is generated that includes the determined initial and final values, the amount of change assigned to each step of the extracted path, and the simulation results.

[0013] The training data generation method relating to the seventh aspect of this disclosure is: The computer in the training data generation device, which generates training data for building surrogate models, A step of randomly determining an initial value of input data and a final value of the input data, respectively; A step of generating a path by randomly assigning possible change amounts for each step in the path from the determined initial value to the final value to each step; A step of generating learning data including the determined initial value and final value and the change amounts assigned to each step of the generated path is executed.

[0014] The learning data generation program according to the eighth aspect of the present disclosure Causes a computer of a learning data generation device that generates learning data for constructing a surrogate model To execute a step of randomly determining an initial value of input data and a final value of the input data, respectively; A step of generating a path by randomly assigning possible change amounts for each step in the path from the determined initial value to the final value to each step; A step of generating learning data including the determined initial value and final value and the change amounts assigned to each step of the generated path.

Advantages of the Invention

[0015] According to the present disclosure, when generating learning data for constructing a surrogate model, it is possible to reduce the bias of the simulation pattern.

Brief Description of the Drawings

[0016] [Figure 1] FIG. 1 is a diagram showing an example of the system configuration of a learning system. [Figure 2] FIG. 2 is a diagram showing an example of the hardware configuration of a learning data generation device. [Figure 3] FIG. 3 is a diagram showing a specific example of time-series data of various simulation patterns. [Figure 4] FIG. 4 is a diagram showing an example of the detailed functional configuration of a generation unit. [Figure 5] Figure 5 is an example of a flowchart showing the flow of the training data generation process. [Figure 6] Figure 6 is the first figure showing a specific example of training data. [Figure 7] Figure 7 shows an example of the distribution of input data for the training data generated by the training data generation process. [Figure 8] Figure 8 shows an example of a path generated by the training data generation process. [Figure 9] Figure 9 is the first figure showing an example of the prediction accuracy of a trained model trained using the training data generated by the training data generation process. [Figure 10] Figure 10 is a second figure showing a specific example of training data. [Figure 11] Figure 11 is the second figure showing an example of the prediction accuracy of a trained model trained using the training data generated by the training data generation process. [Modes for carrying out the invention]

[0017] Each embodiment will be described below with reference to the attached drawings. In this specification and the drawings, components having substantially the same functional configuration are denoted by the same reference numerals, and redundant descriptions will be omitted.

[0018] [First Embodiment] <System Configuration of the Learning System> First, the system configuration of the learning system including the learning data generation device according to the first embodiment will be described. Figure 1 is a diagram showing an example of the system configuration of the learning system.

[0019] As shown in Figure 1, the learning system 100 includes a simulation device 110, a learning data generation device 120, and a learning device 130.

[0020] The simulation device 110 is a device that replicates the manufacturing device 10. Prerequisites for replicating the manufacturing device 10 (e.g., the structure, size, and materials of the device) are set for the simulation device 110. Once the prerequisites are set, the simulation device 110, which replicates the manufacturing device 10, is then given various manufacturing conditions (e.g., temperature, pressure, flow rate, and other physical quantities controlled by the manufacturing device 10 during manufacturing), and a physical simulation is executed.

[0021] In the first embodiment, the manufacturing conditions input to the simulation device 110 are assumed to be time-series data, and include time-series data for various simulation patterns. Specifically, the time-series data for various simulation patterns include: • Time-series data of various simulation patterns where the initial and final values ​​of manufacturing conditions differ from each other. • Time-series data of various simulation patterns where the initial and final values ​​of manufacturing conditions are the same, but the intermediate paths are different. It includes.

[0022] The training data generation device 120 has a training data generation program installed on it, and when this program is executed, the training data generation device 120 functions as a generation unit 121.

[0023] The generation unit 121 generates training data for each precondition set for the simulation device 110.

[0024] Specifically, the generation unit 121 generates time-series data of various simulation patterns as manufacturing conditions under each precondition. The generation unit 121 then inputs the generated time-series data of various simulation patterns into the simulation device 110, which has corresponding preconditions set, to execute a physical simulation and obtain the simulation results. Furthermore, the generation unit 121 generates training data for each precondition by associating the time-series data of various simulation patterns with the respective simulation results for each precondition. The generation unit 121 then stores the generated training data in the training data storage unit 122.

[0025] A learning program is installed on the learning device 130, and when this program is executed, the learning device 130 functions as a learning unit 131.

[0026] The learning unit 131 reads training data from the training data storage unit 122 and uses the read training data to train the model, thereby constructing a surrogate model, which is a pre-trained model.

[0027] <Hardware configuration of the training data generation device> Next, the hardware configuration of the training data generation device 120 will be described. Figure 2 shows an example of the hardware configuration of the training data generation device. As shown in Figure 2, the training data generation device 120 has a processor 201, memory 202, auxiliary storage device 203, I / F (Interface) device 204, communication device 205, and drive device 206. Each piece of hardware in the training data generation device 120 is interconnected via a bus 207.

[0028] The processor 201 has various computing devices such as a CPU (Central Processing Unit) and a GPU (Graphics Processing Unit). The processor 201 executes various programs (for example, training data generation programs) by reading them into memory 202.

[0029] Memory 202 has main memory devices such as ROM (Read Only Memory) and RAM (Random Access Memory). The processor 201 and memory 202 form a so-called computer, and the computer realizes various functions by the processor 201 executing various programs read from memory 202.

[0030] The auxiliary storage device 203 stores various programs and various data used when those programs are executed by the processor 201. For example, the learning data storage unit 122 is implemented in the auxiliary storage device 203.

[0031] The I / F device 204 is a connection device for connecting an operating device 211 and a display device 212, which are examples of user interface devices. The communication device 205 is a communication device for communicating with external devices via a network (not shown).

[0032] The drive device 206 is a device for setting the recording medium 213. The recording medium 213 here includes media that record information optically, electrically, or magnetically, such as CD-ROMs, flexible disks, and magneto-optical disks. The recording medium 213 may also include semiconductor memory that records information electrically, such as ROMs and flash memory.

[0033] The various programs to be installed on the auxiliary storage device 203 are installed, for example, when the distributed recording medium 213 is set in the drive device 206 and the various programs recorded on the recording medium 213 are read by the drive device 206. Alternatively, the various programs to be installed on the auxiliary storage device 203 may be installed when they are downloaded from the network via the communication device 205.

[0034] In the example shown in Figure 2, the hardware configuration of the learning data generation device 120, one of the devices constituting the learning system 100, is shown. However, the other devices constituting the learning system 100, the simulation device 110 and the learning device 130, also have the same hardware configuration as the learning data generation device 120. Therefore, the hardware configurations of the simulation device 110 and the learning device 130 will not be explained.

[0035] <Specific examples of time-series data> Next, we will describe specific examples of time-series data of various simulation patterns generated as manufacturing conditions by the generation unit 121 of the learning data generation device 120. Figure 3 shows specific examples of time-series data of various simulation patterns.

[0036] In Figure 3, the horizontal axis represents the step, and the vertical axis represents the manufacturing conditions. A step is the time interval obtained by dividing the time interval from the start to the end of the simulation by the simulation device 110 by a predetermined number (referred to as the number of steps), and refers to the minimum time interval when changing the manufacturing conditions over time.

[0037] The example in Figure 3 shows how time-series data for three simulation patterns with 23 steps are generated as manufacturing conditions. Of these, the time-series data shown by symbols 301 and 302 have the same combination of initial and final values ​​for the manufacturing conditions (initial value 1, final value 1), but their intermediate paths are different. The time-series data shown by symbol 303 has a different combination of initial and final values ​​for the manufacturing conditions (initial value 2, final value 2) than the time-series data shown by symbol 301, but their intermediate paths are the same.

[0038] Furthermore, if we assume that the length of the double arrow in Figure 3 represents the change in manufacturing conditions per step = +1, then the example in Figure 3 shows that the change in manufacturing conditions per step includes three types of changes: ±0, +1, and +2. Also, all examples in Figure 3 show that the total change in time series data = 11. However, the change in manufacturing conditions per step is not limited to three types. Also, the change in manufacturing conditions per step may include not only positive values ​​but also negative values. Furthermore, the total change in time series data is not limited to 11. In addition, the number of steps is not limited to 23.

[0039] If we let M be the number of initial manufacturing conditions, S be the number of steps, and n be the number of different types of changes in manufacturing conditions per step, The number of time-series data patterns that can be generated as manufacturing conditions = M × n S This is how it works. The training data generation device 120 generates time series data randomly, and then extracts the time series data that reaches a predetermined final value for each initial value to generate training data (details will be described later).

[0040] <Functional configuration of the generation unit> Next, the details of the functional configuration of the generation unit 121 will be described. Figure 4 is a diagram showing an example of the detailed functional configuration of the generation unit.

[0041] As shown in Figure 4, the generation unit 121 includes a step count setting unit 411, a change amount setting unit 412, an initial value range setting unit 413, a final value range setting unit 414, a determination unit 420, a path generation unit 430, and a difference calculation unit 440. The generation unit 121 also includes an extraction unit 450, a simulation result acquisition unit 460, and a training data generation unit 470.

[0042] The step count setting unit 411 is an example of a second setting unit, and it receives the step count entered by the user and notifies the route generation unit 430.

[0043] The change amount setting unit 412 is an example of a third setting unit, and it receives the possible change amounts of the manufacturing conditions per step, which are input by the user, and notifies the path generation unit 430.

[0044] The initial value range setting unit 413 receives the maximum and minimum possible initial values ​​entered by the user and notifies the determination unit 420. The final value range setting unit 414 receives the maximum and minimum possible final values ​​entered by the user and notifies the determination unit 420. Note that the initial value range setting unit 413 and the final value range setting unit 414 are examples of the first setting unit.

[0045] The determination unit 420 determines a random value as the initial value within the range of the maximum and minimum initial values ​​notified by the initial value range setting unit 413, and notifies the difference calculation unit 440.

[0046] Furthermore, the determination unit 420 determines a random value as the final value within the range between the maximum and minimum values ​​of the final value notified by the final value range setting unit 414, and notifies the difference calculation unit 440 of this.

[0047] The route generation unit 430 generates various routes by randomly assigning the change amounts notified by the change amount setting unit 412 to each step of the step count notified by the step count setting unit 411, and notifies the extraction unit 450 of these routes. The route generation unit 430 also calculates the total change amount for each of the various routes by adding up the change amounts assigned to each step, and notifies the extraction unit 450 of this total change amount.

[0048] The difference calculation unit 440 calculates the difference between the initial value and the final value notified by the determination unit 420 and notifies the extraction unit 450.

[0049] The extraction unit 450 extracts from the various paths notified by the path generation unit 430 the path whose total change is equal to the difference value notified by the difference calculation unit 440. The extraction unit 450 also notifies the simulation device 110 and the learning data generation unit 470 of the time series data including the extracted path and its corresponding initial and final values.

[0050] As a result, the simulation device 110 performs physical simulations on the time-series data of various simulation patterns generated in the generation unit 121.

[0051] The simulation result acquisition unit 460 acquires simulation results notified from the simulation device 110 in response to the extraction unit 450 notifying the simulation device 110 of time-series data of various simulation patterns. The simulation result acquisition unit 460 also notifies the training data generation unit 470 of the acquired simulation results.

[0052] The training data generation unit 470 generates training data by associating the simulation results notified by the simulation result acquisition unit 460 with time-series data of various simulation patterns notified by the extraction unit 450. The training data generation unit 470 stores the generated training data in the training data storage unit 122.

[0053] <Workflow for generating training data> Next, the flow of the training data generation process by the generation unit 121 of the training data generation device 120 will be explained. Figure 5 is an example of a flowchart showing the flow of the training data generation process.

[0054] In step S501, the learning data generation device 120 receives the maximum and minimum possible initial values ​​and the maximum and minimum possible final values, which are input by the user.

[0055] In step S502, the learning data generation device 120 receives the number of steps and the possible changes in the manufacturing conditions per step, which are input by the user.

[0056] In step S503, the training data generator 120 randomly determines an initial value within the range of the maximum and minimum initial values. The training data generator 120 also randomly determines a final value within the range of the maximum and minimum final values. Furthermore, the training data generator 120 calculates the difference between the randomly determined initial value and the final value.

[0057] In step S504, the learning data generation device 120 generates various paths by randomly assigning the received change amounts to each step of the received number of steps.

[0058] In step S505, the training data generation device 120 extracts time-series data of the paths whose total change is equal to the difference value from among the various paths that have been generated.

[0059] In step S506, the learning data generation device 120 determines whether a predetermined number of time-series data for various simulation patterns have been generated for each precondition. If it is determined in step S506 that a predetermined number of time-series data for various simulation patterns have not been generated (i.e., the answer in step S506 is NO), the process returns to step S503. In this case, steps S503 to S505 are executed for other randomly determined initial and final values ​​to generate new time-series data for various simulation patterns.

[0060] On the other hand, if it is determined in step S506 that a predetermined number of time-series data for various simulation patterns have been generated (if the answer in step S506 is YES), the process proceeds to step S507.

[0061] In step S507, the learning data generation device 120 notifies the simulation device 110 of a predetermined number of time-series data of various simulation patterns. In response to the notification of the predetermined number of time-series data of various simulation patterns, the learning data generation device 120 obtains simulation results from the simulation device 110. The learning data generation device 120 generates learning data by associating the obtained simulation results with the time-series data of various simulation patterns and stores it in the learning data storage unit 122.

[0062] In this way, by randomly determining the initial and final values ​​and then randomly generating paths, the training data generation device 120 can generate training data that includes time-series data of simulation patterns with less bias.

[0063] <Specific examples of training data> Next, we will describe a specific example of training data generated by the training data generation device 120 when the training data generation process shown in Figure 5 is executed. Figure 6 is the first figure showing a specific example of training data.

[0064] For the sake of simplicity, when generating the 600 training data points, • Number of steps = 5 • Possible variations in manufacturing conditions per step: -1, -0.5, ±0, +0.5, +1 • Maximum and minimum initial values: ±0 • Maximum and minimum final values: 5, -5 That's what I decided.

[0065] Furthermore, for the sake of simplicity, instead of obtaining simulation results, the final values ​​were used as the ground truth data for the 600 training data points.

[0066] As shown in Figure 6, the training data 600 includes "input data" and "correct answer data" as information items. Furthermore, the "input data" includes "manufacturing condition ID", "initial value", and "step 1" to "step 5" as information items.

[0067] The "Manufacturing Condition ID" stores an identifier that identifies each of the time-series data for the various simulation patterns generated.

[0068] The "initial value" field stores an initial value randomly determined within the range of the maximum and minimum possible initial values. In this case, since ±0 is set as the maximum and minimum initial values, "0" is stored in the "initial value" field.

[0069] Steps 1 through 5 store a randomly assigned amount of change from the possible changes in manufacturing conditions per step.

[0070] On the other hand, the "correct answer data" further includes a "final value" as an information item. In this case, the "final value" stores the value obtained by adding the total change amount, which is the sum of the change amounts assigned to each step, to the initial value.

[0071] <Distribution of training data> Next, we will explain the distribution of time-series data of various simulation patterns, which are generated by the training data generation device 120 when the training data generation process shown in Figure 5 is executed, and stored in the "input data" of the training data 600. Figure 7 shows an example of the distribution of input data for the training data generated by the training data generation process.

[0072] In Figure 7(a), the horizontal axis represents the possible changes in manufacturing conditions per step, and the vertical axis represents the frequency of the changes assigned to each step. Note that in the example in Figure 7(a), for comparison purposes, • In the case of time-series data of various simulation patterns generated based on random walk theory, • In the case of time-series data of various simulation patterns generated by the training data generation process shown in Figure 5, This is shown in repeated form.

[0073] As shown in Figure 7(a), in the case of time series data of various simulation patterns generated based on random walk theory, the frequency of assignment of the change amounts to each step is equal. On the other hand, in the case of time series data of various simulation patterns generated by the training data generation process shown in Figure 5, the frequency of assignment of the change amounts to each step is not equal, with -1 or 1 being assigned more frequently than ±0.

[0074] On the other hand, in Figure 7(b), the horizontal axis represents the total change, and the vertical axis represents the frequency of the total change. Note that the example in Figure 7(b) is also shown for comparison. • In the case of time-series data of various simulation patterns generated based on random walk theory, • In the case of time-series data of various simulation patterns generated by the training data generation process shown in Figure 5, This is shown in repeated form.

[0075] As shown in Figure 7(b), in the case of time-series data of various simulation patterns generated based on random walk theory, the frequency of total change being ±0 is highest, and the frequency of total change being -5 or +5 is lowest. In other words, as the number of simulation patterns increases, the newly generated simulation patterns tend to be biased towards the average value of the simulation patterns generated up to that point.

[0076] On the other hand, in the case of time-series data of various simulation patterns generated by the training data generation process shown in Figure 5, the frequency of the total change is uniform.

[0077] Thus, the time series data of various simulation patterns generated by the training data generation process shown in Figure 5 can be said to be less biased, or more evenly distributed, time series data.

[0078] <Specific examples of routes> Next, we will explain a specific example of a path generated by the training data generation device 120 when the training data generation process shown in Figure 5 is executed. Figure 8 is a diagram showing an example of a path generated by the training data generation process. In Figure 8, the horizontal axis represents the steps, and the vertical axis represents the total change up to each step.

[0079] Of these, Figure 8(a) shows time-series data of various simulation patterns generated based on random walk theory. On the other hand, Figure 8(b) shows time-series data of various simulation patterns generated by the training data generation process shown in Figure 5.

[0080] As is clear from comparing Figure 8(a) and Figure 8(b), in the case of time series data of various simulation patterns generated based on random walk theory, paths that pass around ±0 in total change are generated frequently. On the other hand, in the case of time series data of various simulation patterns generated by the training data generation process shown in Figure 5, the paths are generated evenly distributed without bias towards paths that pass around ±0 in total change.

[0081] <Regarding the prediction accuracy of pre-trained models> Next, we will explain the prediction accuracy of a pre-trained surrogate model when the pre-trained model is constructed by training the pre-trained data generated by the pre-trained data generation device 120 using the pre-trained data generation process. In the following, the pre-trained model is a 4-layer perceptron, which is a neural network constructed using the neural network library Keras.

[0082] Figure 9 is the first figure showing an example of the prediction accuracy of a trained model trained using the training data generated by the training data generation process.

[0083] Of these, (a-1) to (a-3) in Figure 9 are, • In the training data set of 600, 500 time-series data points were generated with manufacturing condition IDs from 1 to 500. • Using the 500 generated time-series data, train a predetermined model and construct a trained model. • For the trained model that has been constructed, When time series data with a correct answer of -5.0 is input, the predicted data is obtained. Predicted data when time series data where the correct answer = 0 is input. When time series data with a ground truth value of 5.0 is input, the predicted data is obtained. Calculate, The graphs shown are as follows. The left side of each graph shows, as a comparison, the predicted data predicted by a trained model when 500 time series data points were generated based on random walk theory.

[0084] Similarly, (b-1) to (b-3) in Figure 9 are, • In the training data set 600, 1000 time-series data points were generated with manufacturing condition IDs from 1 to 1000. • Using the 1000 generated time-series data, train a predetermined model and construct a trained model. • For the trained model that has been constructed, When time series data with a correct answer of -5.0 is input, the predicted data is obtained. Predicted data when time series data where the correct answer = 0 is input. Predicted data when time series data with a ground truth value of 5.0 is input as input data. Calculate, The graphs shown are as follows. The left side of each graph shows, as a comparison, the predicted data predicted by a trained model when 1000 time series data points were generated based on random walk theory.

[0085] Similarly, (c-1) to (c-3) in Figure 9 are, • In the training data 600, 5000 time-series data points were generated with manufacturing condition IDs from 1 to 5000. • Using the 5000 generated time-series data, train a predetermined model and construct a trained model. • For the trained model that has been constructed, When time series data with a correct answer of -5.0 is input, the predicted data is obtained. Predicted data when time series data where the correct answer = 0 is input. Predicted data when time series data with a ground truth value of 5.0 is input as input data. Calculate, The graphs shown are as follows. The left side of each graph shows, as a comparison, the predicted data predicted by a trained model when 5000 time series data points were generated based on random walk theory.

[0086] Similarly, (d-1) to (d-3) in Figure 9 are, • In the training data 600, 10,000 time-series data points were generated with manufacturing condition IDs from 1 to 10,000. • Using the 10,000 generated time-series data, train a predetermined model and construct a trained model. • For the trained model that has been constructed, When time series data with a correct answer of -5.0 is input, the predicted data is obtained. Predicted data when time series data where the correct answer = 0 is input. Predicted data when time series data with a ground truth value of 5.0 is input as input data. Calculate, The graphs shown are as follows. The left side of each graph shows, as a comparison, the predicted data predicted by a trained model when 10,000 time series data points were generated based on random walk theory.

[0087] Thus, when comparing a trained model constructed using training data generated by the training data generation process shown in Figure 5 with a trained model constructed using training data generated based on random walk theory, • A trained model constructed using the training data generated by the training data generation process shown in Figure 5 can achieve higher prediction accuracy when predicting values ​​far from the mean, regardless of the amount of training data. As the amount of training data increases, the prediction accuracy when predicting values ​​close to the mean becomes similar.

[0088] In other words, it was shown that reducing the bias in the time-series data of the simulation patterns can improve prediction accuracy.

[0089] <Summary> As is clear from the above description, the training data generation device 120 according to the first embodiment is a device that generates training data for constructing a surrogate model, The initial value and final value of the input data are determined randomly. The path is generated by randomly assigning to each step the possible amount of change that can occur in the path from the determined initial value to the final value. • Generate training data that includes the determined initial and final values, and the amount of change assigned to each step of the generated path.

[0090] Thus, the learning data generation device 120 according to the first embodiment is configured to randomly determine the initial and final values ​​and then randomly generate a path. As a result, the learning data generation device 120 according to the first embodiment can generate learning data that includes time-series data of simulation patterns with less bias than learning data generated based on random walk theory.

[0091] In other words, according to the first embodiment, when generating training data for building a surrogate model, the bias in the simulation patterns can be reduced.

[0092] [Second Embodiment] In the first embodiment described above, the case in which the final values ​​are used as the ground truth data for training was explained when verifying the prediction accuracy of the trained model. In other words, the prediction accuracy of the trained model was verified assuming that the ground truth data has a linear relationship with the input data. In contrast, in the second embodiment, the prediction accuracy of the trained model is verified assuming that the ground truth data has a nonlinear relationship with the input data.

[0093] Figure 10 is the second figure showing a concrete example of training data. The difference from the training data 600 shown in Figure 6 is that instead of storing the final value as the ground truth data, the cubed value of the final value is stored.

[0094] Figure 11 is the second figure showing an example of the prediction accuracy of a trained model trained using the training data generated by the training data generation process.

[0095] (a-1) to (a-3) in Figure 11 are, • In the training data set of 1000, generate 5000 time-series data points with manufacturing condition IDs from 1 to 5000. • Using the 5000 generated time-series data, train a predetermined model and construct a trained model. • For the trained model that has been constructed, When time series data where the correct answer = -125 is input, the predicted data is obtained. Predicted data when time series data where the correct answer = 0 is input. Predicted data when time series data with a correct answer value of 125 is input. Calculate, The graphs shown are as follows. The left side of each graph shows, as a comparison, the predicted data predicted by a trained model when 5000 time series data points were generated based on random walk theory.

[0096] Thus, according to the second embodiment, even if the training data has a nonlinear relationship with the input data, it is possible to construct a trained model with high prediction accuracy.

[0097] [Other embodiments] In the first embodiment described above, the manufacturing conditions of the manufacturing apparatus 10 were used as input data for the training data. However, the input data for the training data is not limited to the manufacturing conditions of the manufacturing apparatus 10, but may be other physical quantities input to the simulation apparatus 110.

[0098] Furthermore, in the first embodiment described above, the user inputs the number of steps, the possible amount of change in manufacturing conditions per step, the possible maximum and minimum values ​​of the initial value, and the possible maximum and minimum values ​​of the final value. However, any of these values ​​may be pre-set in the learning data generation device 120.

[0099] Furthermore, in the first embodiment described above, when inputting the range of initial values, the system is configured to input the maximum and minimum possible values ​​of the initial values, but the method of inputting the range of initial values ​​is not limited to this. Similarly, in the first embodiment described above, when inputting the range of final values, the system is configured to input the maximum and minimum possible values ​​of the final values, but the method of inputting the range of final values ​​is not limited to this.

[0100] Furthermore, although the details of the method for randomly determining the initial and final values ​​were not mentioned in the first embodiment described above, for example, the initial and final values ​​may be determined randomly using uniform random numbers. Alternatively, the initial and final values ​​may be determined randomly using random numbers that follow a predetermined distribution. In any case, the initial and final values ​​are determined to be variable.

[0101] It should be noted that the present invention is not limited to the configurations shown in the above embodiments, including combinations with other elements. These aspects can be modified without departing from the spirit of the present invention and can be appropriately determined according to their application.

[0102] This application claims priority based on Japanese Patent Application No. 2023-176970, filed on 12 October 2023, which is incorporated herein by reference to the entire contents of the said Japanese Patent Application. [Explanation of symbols]

[0103] 10: Manufacturing equipment 100: Learning System 110: Simulation device 120: Training data generation device 121 :Generation part 130: Learning device 131: Learning Department 411: Step count setting section 412: Change amount setting unit 413: Initial value range setting section 414: Final value range setting section 420: Decision Section 430: Route generation unit 440: Difference calculation part 450:Extraction part 460: Simulation result acquisition unit 470: Training data generation unit 600: Training data 1000: Training data

Claims

1. A training data generator that generates training data for constructing a surrogate model, A determination unit randomly determines the initial value of the input data and the final value of the input data, respectively. A path generation unit generates a path by randomly assigning to each step the possible amount of change that each step can take in the path from the determined initial value to the final value, A learning data generation unit generates learning data that includes the determined initial and final values ​​and the amount of change assigned to each step of the generated path. A training data generation device having [a certain feature].

2. The system further includes an extraction unit that extracts paths from the generated paths such that the total change obtained by adding the change amounts assigned to each step is equal to the difference between the determined initial value and the final value. The aforementioned training data generation unit is: The learning data generation device according to claim 1, which generates learning data including the determined initial value and final value and the amount of change assigned to each step of the extracted path.

3. The system further includes a first setting unit that accepts an initial range of the input data and a final range of the input data, respectively. The learning data generation device according to claim 1, wherein the determination unit randomly determines the initial value of the input data and the final value of the input data, respectively, within the range received by the first setting unit.

4. It further has a second setting unit that accepts the number of steps, The route generation unit randomly assigns to each of the number of steps received by the second setting unit the possible amount of change for each step. The learning data generation device according to claim 1.

5. The system further includes a third setting unit that accepts the possible amounts of change in each of the aforementioned steps, The learning data generation device according to claim 1, wherein the path generation unit randomly assigns the amount of change received by the third setting unit to each step.

6. The system further includes an acquisition unit that acquires the simulation results when the determined initial and final values ​​and the change amounts assigned to each step of the extracted path are input to a simulation device as input data. The aforementioned training data generation unit is: The learning data generation device according to claim 2, which generates learning data including the determined initial and final values, the amount of change assigned to each step of the extracted path, and the simulation results.

7. The computer in the training data generation device, which generates training data for building surrogate models, A step of randomly determining the initial value of the input data and the final value of the input data, The process of generating a path involves randomly assigning to each step the possible amount of change that each step can take in the path from the determined initial value to the final value, A step of generating training data that includes the determined initial and final values ​​and the amount of change assigned to each step of the generated path. A method for generating training data to execute this.

8. The computer of the training data generation device, which generates training data for building surrogate models, A step of randomly determining the initial value of the input data and the final value of the input data, The process of generating a path involves randomly assigning to each step the possible amount of change that each step can take in the path from the determined initial value to the final value, A step of generating training data that includes the determined initial and final values ​​and the amount of change assigned to each step of the generated path. A program for generating training data to run the program.

Images