Information processing method, information processing device, and computer program

Abstract

This invention provides an information processing method that can reduce the computation time required to optimize the design parameters of a design object compared to optimization calculations using genetic algorithms with CAE analysis such as the finite element method. [Solution] An information processing method for optimizing design parameters that define individual models representing the object to be designed using a genetic algorithm based on FEA calculation, wherein the number of generations is increased while reducing computation time and computation amount by using a surrogate model in a particular generation, or the number of individual models in each generation is increased by using a surrogate model.

Description

Technical Field

[0001] The present disclosure relates to an information processing method, an information processing apparatus, and a computer program.

Background Art

[0002] In product design, it is common to calculate the characteristics or behaviors of a simulation target model representing a design object such as a motor by CAE analysis such as the finite element method. For example, Patent Document 1 discloses a design method that enables easy and rapid design of a desired inductor even for a non-expert designer. Specifically, the design method according to Patent Document 1 is a method for designing a wound inductor using a computer, and includes steps of initializing variables describing the structure of the inductor, calculating the characteristics of the inductor having the structure described by the variables by magnetic field analysis based on pre-input attribute information of a core and a coil conductor, calculating an evaluation value using a predetermined evaluation function having at least inductor characteristics as parameters, determining whether the evaluation value satisfies a predetermined convergence condition, ending the process when the convergence condition is satisfied, and varying the values of the variables by a genetic algorithm when the convergence condition is not satisfied, and repeating the analysis process and the optimization process until the convergence condition is satisfied.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] However, in the optimization calculation using a genetic algorithm, there is a problem that it is necessary to repeatedly calculate (M + 1) × N (number of generations × number of individuals) times of CAE analysis, which requires a large amount of calculation time.

[0005] The purpose of this disclosure is to provide an information processing method, an information processing device, and a computer program that can reduce the computation time required for optimizing the design parameters of a design object compared to optimization calculations using genetic algorithms with CAE analysis such as the finite element method. [Means for solving the problem]

[0006] An information processing method according to one aspect of the present disclosure is an information processing method in which a computer optimizes design parameters that define individual models representing an object to be designed using a genetic algorithm, wherein the computer performs a generation step of generating multiple new individual models for the current generation by genetic processing based on a predetermined number of design parameters of individual models inherited from the previous generation; a calculation step of calculating response values ​​indicating the characteristics of the multiple individual models using a first calculation method; a ranking step of ranking the predetermined number of individual models inherited from the previous generation and the newly generated multiple individual models based on the calculated response values; and a selection step of selecting a predetermined number of top-ranked individual models to be inherited by the next generation based on the ranking results, wherein the generation step or the calculation step increases the number of generations or the number of individuals in each generation by calculating the response values ​​indicating the characteristics of the individual models using a second calculation method which has a faster calculation speed than the first calculation method.

[0007] An information processing device according to one aspect of the present disclosure is an information processing device comprising a calculation unit that performs a process of optimizing design parameters defining individual models representing an object to be designed using a genetic algorithm, wherein the calculation unit performs a generation step of generating multiple new individual models for the current generation by genetic processing based on a predetermined number of design parameters of individual models inherited from the previous generation; a calculation step of calculating response values ​​indicating the characteristics of the multiple individual models using a first calculation method; a ranking step of ranking the predetermined number of individual models inherited from the previous generation and the newly generated multiple individual models based on the calculated response values; and a selection step of selecting a predetermined number of top-ranked individual models to be inherited by the next generation based on the ranking results, wherein the generation step or the calculation step increases the number of generations or the number of individuals in each generation by calculating the response values ​​indicating the characteristics of the individual models using a second calculation method which has a faster calculation speed than the first calculation method.

[0008] A computer program according to one aspect of this disclosure is a computer program that causes a computer to perform information processing to optimize design parameters defining individual models representing an object to be designed using a genetic algorithm, wherein the computer is caused to perform a generation step of generating multiple new individual models for the current generation by genetic processing based on a predetermined number of design parameters of individual models inherited from the previous generation; a calculation step of calculating response values ​​indicating the characteristics of the multiple individual models using a first calculation method; a ranking step of ranking the predetermined number of individual models inherited from the previous generation and the newly generated multiple individual models based on the calculated response values; and a selection step of selecting a predetermined number of top-ranked individual models to be inherited by the next generation based on the ranking results, wherein the generation step or the calculation step increases the number of generations or the number of individuals in each generation by calculating the response values ​​indicating the characteristics of the individual models using a second calculation method which has a faster calculation speed than the first calculation method. [Effects of the Invention]

[0009] According to this disclosure, the computation time required to optimize the design parameters of the object can be reduced compared to optimization calculations using genetic algorithms with CAE analysis such as the finite element method. [Brief explanation of the drawing]

[0010] [Figure 1] This is a block diagram showing an example configuration of the information processing device 1 according to Embodiment 1. [Figure 2] This is a conceptual diagram illustrating the vertical optimization calculation method according to Embodiment 1. [Figure 3] This is a conceptual diagram showing the 0th generation in optimization calculations. [Figure 4] This is a conceptual diagram illustrating the first generation of vertical optimization calculations. [Figure 5] This is a conceptual diagram illustrating the second generation of vertical optimization calculations. [Figure 6] This is a conceptual diagram illustrating the third generation of vertical optimization calculations. [Figure 7] This is a flowchart showing the processing procedure for vertical optimization calculations. [Figure 8] This is a conceptual diagram illustrating the vertical optimization calculation for a modified example. [Figure 9] This is a conceptual diagram showing the horizontal optimization calculation method according to Embodiment 2. [Figure 10] This is a conceptual diagram showing the first generation of horizontal optimization calculations. [Figure 11] This is a conceptual diagram showing the mth generation in horizontal optimization calculations. [Figure 12] This is a flowchart showing the processing procedure for horizontal optimization calculations. [Figure 13] This graph shows the effect of the optimization calculation related to this disclosure. [Figure 14] This is a conceptual diagram showing the optimization calculation for a first modified example that combines vertical and horizontal methods. [Figure 15] This is a conceptual diagram showing the optimization calculation for a second modified example that combines vertical and horizontal methods. [Figure 16]It is a flowchart showing the main part of the optimization calculation process according to Embodiment 3. [Figure 17] It is a flowchart showing the main part of the optimization calculation process according to Embodiment 4. [Figure 18] It is a conceptual diagram showing the first generation in the vertical type optimization calculation according to Embodiment 5. <目标 <目标 [Figure 19] It is a conceptual diagram showing the third generation in the vertical type optimization calculation according to Embodiment 5. <目标 <目标 [Figure 20] It is a flowchart showing the main part of the optimization calculation process according to Embodiment 5. <目标 <目标 [Figure 21] It is a conceptual diagram showing the second generation in the vertical type optimization calculation according to Embodiment 6. <目标 <目标 [Figure 22] It is a conceptual diagram showing the third generation in the vertical type optimization calculation according to Embodiment 6. <目标 <目标

Mode for Carrying Out the Invention

[0011] <目标 An information processing method, an information processing apparatus, and a computer program according to an embodiment of the present disclosure will be described below with reference to the drawings. Note that the present disclosure is not limited to these examples, and is indicated by the claims, and is intended to include all modifications within the meaning and scope equivalent to the claims. In addition, at least a part of the embodiments described below may be arbitrarily combined. <目标 <目标

[0012] <目标 (Embodiment 1)<目标 FIG. 1 is a block diagram showing a configuration example of an information processing apparatus 1 according to Embodiment 1. The information processing apparatus 1 is a computer that implements the information processing method according to Embodiment 1, and includes an arithmetic unit 11, a display unit 12, an operation unit 13, and a storage unit 14. Each unit is connected by a bus. The information processing method according to Embodiment 1 is a method of optimizing design parameters that define a design object such as a motor using a genetic algorithm such as NSGA-II. Hereinafter, in the genetic algorithm, a model in which a design object is represented by design parameters is called an individual model. <目标 <目标

[0013] <目标 The information processing device 1 may be a standalone computer or a server connected to a network. Furthermore, the information processing device 1 may be a computer in an on-premises environment or a server or other computer in a cloud environment. The information processing device 1 may be configured with multiple computers for distributed processing, implemented with multiple virtual machines on a single server, or implemented using a cloud server.

[0014] The arithmetic unit 11 is a processor having arithmetic circuits such as a CPU (Central Processing Unit), multi-core CPU, GPU (Graphics Processing Unit), GPGPU (General-purpose computing on graphics processing units), TPU (Tensor Processing Unit), ASIC (Application Specific Integrated Circuit), FPGA (Field-Programmable Gate Array), and NPU (Neural Processing Unit), internal storage devices such as ROM (Read Only Memory) and RAM (Random Access Memory), I / O terminals, a timing unit, etc. The arithmetic unit 11 preferably has a configuration having multiple arithmetic cores. The arithmetic unit 11 implements the information processing method according to this embodiment 1 by executing a computer program (program product) 141 stored in the storage unit 14, which will be described later. Note that each functional unit of the information processing device 1 may be implemented in software, or some or all of them may be implemented in hardware.

[0015] The display unit 12 is, for example, a display device such as a liquid crystal panel or an organic EL (Electro-Luminescence) display.

[0016] The operation unit 13 is, for example, an input device such as a hardware keyboard, a pointing device, or a touch panel. The user of the information processing device 1 can input any information into the information processing device 1 using the operation unit 13. The operation unit 13 may be configured integrally with the display unit 12.

[0017] The storage unit 14 includes, for example, a main storage unit and an auxiliary storage unit. The main storage unit is a temporary storage area such as SRAM (Static Random Access Memory), DRAM (Dynamic Random Access Memory), or flash memory, and temporarily stores data necessary for the arithmetic unit 11 to perform arithmetic processing. The auxiliary storage unit is a storage device such as a hard disk or EEPROM (Electrically Erasable Programmable ROM). The storage unit 14 stores the computer program 141, CAE model 142, and surrogate model (SM) 143 executed by the arithmetic unit 11. The surrogate model 143 may be created in advance, or it may be created during the optimization process of design parameters related to the object being designed.

[0018] The CAE model 142 includes, for example, a 3D shape model such as 3D CAD data representing the shape of the object to be designed, and material properties of each part constituting the 3D shape model. Specifically, if the object to be designed is a motor, the CAE model 142 includes a 3D shape model such as 3D CAD data representing the shapes of multiple coils, a stator, and a rotor that constitute the motor, and data such as material properties, boundary conditions, and voltage conditions of each part constituting the 3D shape model. Material properties include magnetization properties, electrical properties, and mechanical properties. The various data defining the CAE model 142 include the design parameters to be optimized in this embodiment 1. The design parameters include dimensional parameters and topology parameters. By changing these design parameters, it is possible to generate CAE models 142 with various shapes and properties, i.e., individual models.

[0019] The calculation unit 11 calculates response values ​​related to the characteristics or behavior of the individual model by analyzing the CAE model 142 using a known calculation method (first calculation method) such as the finite element method (FEA) or the boundary element method. In other words, it simulates the characteristics or behavior of the individual model. For example, the calculation unit 11 performs a magnetic field analysis on the CAE model 142 of a motor. The magnetic field analysis is performed using a known magnetic field analysis simulator such as the finite element method or the boundary element method. The finite element method is a method that divides the rotor and stator of a motor, which have complex shapes and electromagnetic characteristics, into small regions (elements) with simple shapes and electromagnetic characteristics, and predicts the behavior or characteristics of the entire motor by approximately calculating the characteristics of each simplified element. Hereinafter, in this embodiment 1, the calculation process of response values ​​using the CAE model 142 as the individual model will be appropriately referred to as FEA calculation.

[0020] The surrogate model 143 is a machine learning model that outputs response values ​​regarding the characteristics or behavior of an individual model when design parameters defining the individual model of the object being designed are input. The surrogate model 143 includes, for example, a neural network (NN) that has been trained by machine learning. The surrogate model 143 has an input layer into which design parameters defining the individual model are input, an intermediate layer that extracts features of the individual model, and an output layer that outputs response values ​​regarding the characteristics or behavior of the individual model.

[0021] The calculation unit 11 calculates response values ​​related to the characteristics or behavior of the individual model using a calculation method (second calculation method) that utilizes the surrogate model 143. The calculation of response values ​​using the surrogate model 143 is performed more quickly than the FEA calculation. Hereinafter, in this embodiment 1, the calculation of response values ​​using the surrogate model 143 as the individual model will be referred to as the SM calculation as appropriate.

[0022] The surrogate model 143 is generated by performing machine learning using training data in a computer-based learning device. The training data includes numerous datasets that associate design parameters defining individual models representing the object under design with response values ​​related to the characteristics or behavior of the individual models. The method for creating the surrogate model 143 is as follows. The learning device inputs design parameters that define the individual model, recorded in the training data, into a pre-trained model whose weight coefficients have not yet been adjusted. The pre-trained model outputs a response value corresponding to the input design parameters. The learning device adjusts the parameters of the pre-trained model so that the error between the response value output by the pre-trained model and the response value associated with the input design parameters is minimized. For example, the weight coefficients are adjusted using backpropagation.

[0023] The learning device generates a surrogate model 143 by repeatedly performing the above process using multiple datasets included in the training data and adjusting the weight coefficients of the pre-trained model. The weight coefficients of the surrogate model 143 are stored in the storage unit 14 of the information processing device 1.

[0024] While a neural network was described as an example of surrogate model 143, other models such as SVR (Support Vector Regression), decision trees, and linear regression are also acceptable, and the method used is not restricted.

[0025] The auxiliary storage unit may be an external storage device connected to the information processing device 1. The computer program 141 and surrogate model 143 may be written to the storage unit 14 during the manufacturing stage of the information processing device 1, or the information processing device 1 may acquire them via communication from an external server and store them in the storage unit 14. The computer program 141 and surrogate model 143 may also be recorded in a readable manner on a recording medium 10 such as a magnetic disk, optical disk, or semiconductor memory, and a reader may read them from the recording medium 10 and store them in the storage unit 14.

[0026] <Information Processing Methods> This section explains optimization processing using a genetic algorithm. In motor design using magnetic field analysis, optimization calculations using GA (genetic algorithm) are frequently used. Optimization calculations are, for example, calculations to determine a motor design that improves the average torque of the motor and reduces iron loss. The design parameters of the individual model are, for example, values ​​that characterize the position (e.g., distance from the rotor center to the center of the permanent magnet) and shape (vertical and horizontal dimensions in the case of a rectangular cross-section) of the permanent magnets that make up the motor rotor, and the shape of the flux barrier. The response values ​​calculated by the CAE model 142 and surrogate model 143 of the individual model are, for example, the average torque and the iron loss value in the rotor.

[0027] Figure 2 is a conceptual diagram showing the vertical optimization calculation according to Embodiment 1. The information processing according to this embodiment aims to reduce the computation time and computational load required for optimizing design parameters by replacing the time-consuming FEA calculation with SM calculation, which can calculate response values ​​instantaneously, in the optimization calculation using a genetic algorithm. In particular, the information processing method according to Embodiment 1, as shown in Figure 2, aims to reduce computation time by calculating the response value in a specific generation using SM calculation. In other words, the information processing method according to Embodiment 1 makes it possible to increase the number of generations while suppressing the increase in the calculation time of the response value compared to when calculation is performed only with FEA calculation.

[0028] In Figure 2, the arrows indicate the computational processes for each generation in the genetic algorithm. Outlined arrows indicate the process of calculating the response value of the individual model using FEA calculation, while filled arrows indicate the process of calculating the response value of the individual model using SM calculation. The length of the arrows represents the computation time and the amount of computation. The left figure shows the computation time when calculating the response value of the individual model for each generation using only FEA calculation, and the right figure shows the computation time when the calculation of the response value in a specific generation is replaced with SM calculation.

[0029] Figure 3 is a conceptual diagram showing the 0th generation in the optimization calculation, Figure 4 is a conceptual diagram showing the 1st generation in the vertical optimization calculation, Figure 5 is a conceptual diagram showing the 2nd generation in the vertical optimization calculation, Figure 6 is a conceptual diagram showing the 3rd generation in the vertical optimization calculation, and Figure 7 is a flowchart showing the processing procedure related to the vertical optimization calculation.

[0030] First, the calculation unit 11 assigns 0 to the variable m (hereinafter referred to as the number of generations m) which indicates the generation (step S11). Next, the calculation unit 11 generates N individual models (initial individuals) of the initial generation, as shown in Figure 3 (step S12). N is an integer greater than or equal to 2. Specifically, the calculation unit 11 creates N design parameters that define each of the N individual models. Since each individual model is defined by the design parameters, more precisely, the calculation unit 11 generates N sets of design parameters. The design parameters for the initial generation (generation 0) are preferably generated using random numbers. The method for generating the design parameters for the initial individuals is not particularly limited.

[0031] Next, the calculation unit 11 calculates the FEA (Functional Energy Analysis) of the response value of each individual model based on the design parameters of each individual model generated (step S13). The calculation unit 11 then aggregates the design parameters of the initial individuals and the response values ​​of the N initial individuals calculated in the above process into the GA engine (step S14). The GA engine is a calculation module that optimizes the design parameters of individual models using a genetic algorithm.

[0032] Then, the calculation unit 11 uses the design parameters of each individual model and the response values ​​calculated in step S13 as training data to create a surrogate model 143 (step S15). If the surrogate model 143 has already been created, the training process for the surrogate model 143 may be omitted. Alternatively, the surrogate model 143 may be retrained. Furthermore, if the surrogate model 143 has already been created, the system may be configured to calculate the SM (Simulation Model) from the response values ​​of the individual models starting from generation 0.

[0033] After completing the processing in step S15, the arithmetic unit 11 determines whether or not it is a specific SM calculation generation (step S16). For example, the storage unit 14 stores data indicating which generations should have their response values ​​calculated by FEA calculation and which generations should have their response values ​​calculated by SM calculation, and the arithmetic unit 11 determines whether or not it is an SM calculation generation by calculating this data. Alternatively, the arithmetic unit 11 may determine that odd-numbered generations are SM calculation generations and even-numbered generations are FEA calculation generations. The method for determining whether or not it is an SM calculation generation is not particularly limited.

[0034] If it is determined that it is an SM calculation generation (step S16: YES), N individuals are generated for the current generation (step S17). Specifically, the calculation unit 11 generates N new individual models for the current generation through genetic processing such as crossover and mutation, based on the design parameters of the N individual models inherited from the previous generation. Crossover is an operation that generates an individual model with new design parameters by combining the design parameters of multiple individual models inherited from the previous generation. Mutation is an operation that generates an individual model with new design parameters by randomly changing some of the design parameters of multiple individual models inherited from the previous generation. The calculation unit 11 may generate new individual models by both crossover and mutation. In the first generation, as shown in Figure 4, the calculation unit 11 inherits the individual model of the previous generation, generation 0, as is, and generates N individual models based on the design parameters of that individual model. In the mth generation from the second generation onward, as shown in Figures 5 and 6, the calculation unit 11 inherits the N individual models selected from the (m-1)th generation and generates N individual models through genetic processing based on the design parameters of those individual models.

[0035] Next, as shown in Figures 4 and 6, the calculation unit 11 performs SM calculation on the response values ​​of the N individual models (child models) generated in step S16 using the surrogate model 143 (step S18). Then, the calculation unit 11 aggregates the generated N design parameters and the response values ​​of the individual models into the GA engine (step S19). The response values ​​of the N individual models inherited from the previous generation have already been calculated and are used as is.

[0036] If it is determined that it is not an SM calculation generation (step S16: NO), the calculation unit 11 generates N individuals in the current generation (step S20), and as shown in Figure 5, it performs FEA calculation on the response values ​​of the N individual models (child individuals) generated in step S16 using the CAE model 142 (step S21).

[0037] Furthermore, the calculation unit 11 aggregates the generated N design parameters and the response values ​​of the individual model in the GA engine (step S22), and uses the design parameters of the N individual models (child individuals) generated in step S20 and the response values ​​calculated in step S19 as training data to retrain the surrogate model 143 (step S23).

[0038] Then, the calculation unit 11 ranks the 2N individual models based on their response values ​​(step S24) and selects the top-ranked N individual models (step S25). For example, the arithmetic unit 11 classifies the 2N individual models into groups or individuals in order of proximity to the Pareto front using a non-superiority sort in the NSGA-II algorithm, and assigns a rank to each classified group or individual. The smaller the rank number of a group or individual, the higher the rank (higher evaluation), i.e., the closer the group or individual is to the Pareto front. As the rank number increases, the rank decreases (lower evaluation). For example, when designing a motor to improve the average torque and reduce iron loss, the larger the average torque and the smaller the iron loss, the closer it is to the Pareto front and the smaller the rank number. Note that there may be multiple individuals belonging to a given rank number, or there may be only one. Also, the number of individuals belonging to each rank differs depending on the rank. The number of ranks obtained by classifying the 2N individuals differs depending on the design parameters of the 2N individual models. Then, of the 2N individual models, N individual models that are far from the Pareto front and have large rank numbers are eliminated, and N individual models that are close to the Pareto front and have small rank numbers are passed on to the next generation. If selecting individuals in order from smallest rank number would result in more than N individuals being selected, a congestion sort is performed to select some of the highly congested individuals so that the number of selected individuals is N.

[0039] The arithmetic unit 11 then determines whether the current generation (m generation) is the M generation, which is set as the number of iterations for the optimization calculation (step S26). If it is determined that the current generation is not the M generation (step S26: NO), the arithmetic unit 11 increments the generation number m by 1 (step S27) and returns to step S16. If it is determined that the generation is the M generation (step S26: YES), the arithmetic unit 11 finishes processing.

[0040] According to the information processing method, etc., of Embodiment 1 configured in this way, by using a vertical approach to replace FEA calculations in some generations with SM calculations, the computation time required for optimizing the design parameters of the object of design can be reduced compared to optimization calculations performed entirely by FEA.

[0041] Figure 8 is a conceptual diagram showing the vertical optimization calculation according to a modified example. Similar to Figure 2, Figure 8 shows the computational processing for each generation in the genetic algorithm using arrows. While Figures 2 to 6 describe an example where FEA calculation generations and SM calculation generations alternate, the calculation unit 11 may also optimize the design parameters by repeating the SM calculation generation a predetermined number of times, as shown in Figure 8. In other words, the calculation unit 11 may perform SM calculations with a bias towards specific generations, rather than performing SM calculations without bias across generations 0 to M. For example, the calculation unit 11 may optimize the design parameters by performing SM calculations at a high cycle rate in the first half of a generation.

[0042] In this embodiment 1, a motor was described as the simulation target and magnetic field analysis was described as the simulation method using CAE model 142. However, the simulation target and simulation content are not particularly limited. For example, the information processing method according to this embodiment can be applied to the optimization of the magnetic circuit of a solenoid valve or the magnetic circuit of a plastic magnet. Furthermore, the simulation method may be thermal analysis, electric field analysis, structural analysis, fluid analysis, etc., or it may be an analysis process that combines multiple analysis processes, such as magnetic field analysis and structural analysis.

[0043] Furthermore, while examples of the first computation method include FEA calculation using CAE model 142 and an example of the second computation method include calculation using machine learning model surrogate model 143, the content of the first and second computation methods is not particularly limited. As long as the computational precision of the second computation method is lower than that of the first computation method, and the computational speed of the second computation method is faster than that of the first computation method, the content of the first and second computation methods is not particularly limited. For example, any model such as CAE model 142, SVR (Support Vector Regression), decision tree, linear regression, or neural network can be used as the first and second computation methods.

[0044] (Embodiment 2) The information processing device 1 according to Embodiment 2 differs from that of Embodiment 1 in the application of SM calculation. The other configurations and processes of the information processing device 1 are the same as those of the information processing device 1 according to Embodiment 1, so the same reference numerals are used for the same parts, and detailed explanations are omitted.

[0045] Figure 9 is a conceptual diagram showing the horizontal optimization calculation according to Embodiment 2. Similar to Figure 2, Figure 9 shows the computational processing for each generation in the genetic algorithm using arrows. The black arrows indicate the process of calculating the response values ​​of the individual models using SM calculations; their elongated shape indicates that the number of individuals for which response values ​​are calculated is greater than the usual N.

[0046] When there are many design parameters for the object being designed, it is necessary to increase the number of individuals in each generation to obtain a convergent solution, which leads to the problem of increased computation time and computational load. The information processing method according to this embodiment 2 generates N × k (k > 1) individual models (child individuals) as described later, performs SM calculation on the response value of each individual, selects the N individual models with the highest evaluation according to the purpose of design parameter optimization, and performs FEA calculation on the selected N individual models with the highest evaluation, thereby enabling the convergent solution of the design parameters of the individual models to be obtained faster. Furthermore, when the number of generations to be calculated and the number of offspring are the same, the computation time is not necessarily shorter compared to optimizing the design parameters using only FEA calculations, but the time to obtain a converged solution is shorter. In other words, it can be expected that the number of generations required to obtain a converged solution can be reduced.

[0047] Figure 10 is a conceptual diagram showing the first generation in the horizontal optimization calculation, Figure 11 is a conceptual diagram showing the mth generation in the horizontal optimization calculation, and Figure 12 is a flowchart showing the processing procedure related to the horizontal optimization calculation. Note that the 0th generation is the same as that shown in Figure 3 of Embodiment 1.

[0048] First, the calculation unit 11 performs processes such as generating N initial individuals and calculating the FEA of the response value, similar to steps S11 to S15 of Embodiment 1 (steps S31 to S35).

[0049] Next, the calculation unit 11 generates N × k individuals in the current generation, as shown in Figure 10 (step S36). k is a value greater than 1, and here, it generates more individual models than the number of individuals N that should be generated in the current generation.

[0050] Then, the calculation unit 11 performs SM calculation on the response values ​​of the N × k individual models (child individuals) generated in step S36 using the surrogate model 143 (step S37).

[0051] Next, the calculation unit 11 selects N individual models with the highest evaluation from among the N × k individual models based on the response values ​​of the N × k individual models, according to the purpose of design parameter optimization (step S38). The purpose of design parameter optimization is, for example, to improve the average torque of the motor, which is the object of design, and to suppress iron loss. For example, it is possible to select N individual models using an evaluation function in which the evaluation value increases as the average torque increases and the iron loss decreases. In this case, the calculation unit 11 passes the average torque and iron loss, which are the response values ​​of the individual models, as arguments to the evaluation function to calculate the evaluation value, and selects N individual models with the highest evaluation value that have a large calculated evaluation value.

[0052] The calculation unit 11 performs FEA calculation on the response values ​​of the N individual models (child models) selected in step S38 using the CAE model 142 (step S39), and aggregates the design parameters and response values ​​of the N individual models (step S40). Then, the calculation unit 11 uses the aggregated design parameters and response values ​​of the individual models (child models) as training data to retrain the surrogate model 143 (step S41).

[0053] Next, the calculation unit 11 performs processes such as ranking and elimination of the generated individual models, similar to steps S24 to S27 of Embodiment 1 (steps S42 to S45). For the second generation and beyond, as shown in Figure 11, the N individual models selected from the previous generation are inherited, and the generation, ranking, and elimination processes for offspring individuals are performed in the same manner as for the first generation.

[0054] According to the information processing method, etc., of Embodiment 2 configured in this manner, the computation time required for optimizing the design parameters of the object to be designed can be shortened, similar to Embodiment 1. In particular, according to Embodiment 2, even when there are many design parameters for the object to be designed, the number of individuals in each generation can be increased while suppressing computation time, and a converged solution for the design parameters can be obtained more quickly.

[0055] Figure 13 is a graph showing the effect of the optimization calculations related to this disclosure. The horizontal axis represents the number of FEA calculations, and the vertical axis represents the hypervolume index of the Pareto front. The number of FEA calculations corresponds to the time required to optimize the design parameters. The hypervolume index is the volume of the region enclosed by the so-called Pareto front and the reference point, and it is an index that becomes a stationary value when it converges to the Pareto front. When the design parameters that define the individual model are optimized to a convergent solution by the multi-objective genetic algorithm, the hypervolume index becomes a stationary value.

[0056] As can be seen from Figure 13, by incorporating SM calculations, the optimal solution for design parameters can be obtained more quickly compared to performing the optimization process using only FEA calculations. Depending on the characteristics of the object being designed, the number of design parameters, etc., it is advisable to try each method and select the appropriate one as needed.

[0057] While Embodiments 1 and 2 have described examples of performing SM calculations using the vertical and horizontal methods, respectively, the calculation unit 11 may also perform design parameter optimization processing by combining the vertical and horizontal methods.

[0058] Figure 14 is a conceptual diagram showing the optimization calculation for the first modified example combining the vertical and horizontal methods, and Figure 15 is a conceptual diagram showing the optimization calculation for the second modified example combining the vertical and horizontal methods. For example, as shown in Figure 14, the calculation unit 11 can be configured to calculate the response values ​​of individual models in each generation using vertical SM calculations in the first half of the generation, and to calculate the response values ​​using horizontal SM calculations in the second half of the generation. Conversely, the arithmetic unit 11 can be configured to calculate the response values ​​of individual models in each generation using horizontal SM calculations in the first half of the generation, and to calculate the response values ​​using vertical SM calculations in the second half of the generation. Furthermore, the calculation unit 11 may arbitrarily combine the vertical SM calculation shown in Figures 2 and 8 with the horizontal SM calculation shown in Figure 9 to optimize the design parameters of the individual model. Furthermore, the calculation unit 11 may be configured to alternately or randomly select and execute vertical SM calculations, horizontal SM calculations, and FEA calculations.

[0059] (Embodiment 3) The information processing device 1 according to Embodiment 3 differs from that of Embodiment 1 in that it avoids the problem of individual models having shape-defective design parameters remaining after SM calculation. The other configurations and processes of the information processing device 1 are the same as those of the information processing device 1 according to Embodiment 1, so the same reference numerals are used for the same parts, and detailed explanations are omitted.

[0060] When the calculation unit 11 generates child individuals, it may set design parameters that result in geometric breakdown as the model of the object to be designed. In other words, an individual model that cannot function as a design object may be generated. If an FEA calculation is attempted using the design parameters of such an individual model, no response value will be calculated because of the geometric breakdown. However, when using the surrogate model 143, a plausible response value is calculated. If a response value for a higher evaluation is calculated, such an individual model will persist in the future despite its geometric breakdown, which may cause problems in the optimization calculation.

[0061] To solve this problem, in Embodiment 3, the calculation unit 11 determines whether the shape created by the design parameters of the child individual is valid, and prevents individual models with broken shapes from being passed on to subsequent generations. The specific processing procedure is described below.

[0062] Figure 16 is a flowchart showing the main parts of the optimization calculation process according to Embodiment 3. For example, in step S24 of Embodiment 1, the calculation unit 11 performs the following process. The calculation unit 11 verifies the validity of the design parameters of the individual whose response value was calculated by SM calculation (step S241).

[0063] For example, the calculation unit 11 determines whether or not the edges that constitute the shape represented by the design parameters of the individual model under verification intersect. If the edges intersect, the calculation unit 11 determines that the shape is broken.

[0064] Furthermore, the calculation unit 11 attempts to perform an FEA calculation using the design parameters of the individual model to be verified. If the FEA calculation can be performed up to a certain point, it determines that the shape of the object to be designed is valid. Specifically, if the calculation unit 11 can continue the FEA calculation for a predetermined time or longer, it determines that the shape is not broken, and if a calculation error occurs midway through, it determines that the shape is broken.

[0065] Next, the calculation unit 11 applies a penalty to the response value of the individual model that has been determined to have a shape defect (step S242). In other words, the calculation unit 11 changes the response value so that the evaluation of the individual model decreases.

[0066] According to the information processing method of Embodiment 3 configured in this way, by processing individual models with deformed shapes so that they do not remain for future generations, the design parameters of individual models can be optimized with greater accuracy.

[0067] Next, the calculation unit 11 ranks the 2N individual models based on the response values ​​of the 2N individual models after verification (step S243), and then completes the process.

[0068] In the above embodiment, an example of applying a penalty to the response value was described, but it is also possible to configure the model to exclude the individual model in the current generation as soon as it is determined that the shape is distorted.

[0069] Furthermore, although an example of executing the above process in step S24 shown in Figure 7 has been described, it is also possible to configure the system to execute the processes of steps S241 and S242 in step S18.

[0070] Furthermore, the horizontal optimization process shown in Embodiment 2 may also be configured to apply the processing of Embodiment 3 to determine whether or not shape distortion occurs. For example, the process may be configured to execute steps S241 and S242 in step S37.

[0071] (Embodiment 4) The information processing device 1 according to Embodiment 4 differs from that of Embodiment 1 in that it re-evaluates existing individual models when the surrogate model 143 is retrained. The other configurations and processes of the information processing device 1 are the same as those of the information processing device 1 according to Embodiment 1, so the same reference numerals are used for the same parts, and detailed explanations are omitted.

[0072] In both the vertical and horizontal approaches, the accuracy of the surrogate model 143 tends to improve with each generation. This is due to the increase in training data with each generation. Therefore, the response values ​​calculated using the SM in the first half of the generation are expected to be larger than those calculated using the SM in the second half.

[0073] Therefore, in this embodiment 4, in order to reduce the error in the response values ​​calculated by SM calculation, the response values ​​of individual models calculated in previous generations are re-evaluated using the latest surrogate model 143. This is expected to improve the accuracy of the optimization calculation. The specific processing procedure is described below.

[0074] Figure 17 is a flowchart showing the main parts of the optimization calculation process according to Embodiment 4. The calculation unit 11 performs the following process, for example, in step S23 of Embodiment 1. The calculation unit 11 retrains the surrogate model 143 as described in step S23 (step S231).

[0075] Next, the arithmetic unit 11 identifies the individual models from among the existing individual models in the current generation for which response values ​​were calculated using the surrogate model 143 before retraining (step S232). The response values ​​of the identified individual models are recalculated using the latest surrogate model 143 after retraining (step S233). In other words, the arithmetic unit 11 replaces the response values ​​calculated by SM in past generations with response values ​​calculated by SM using the latest surrogate model 143.

[0076] According to the information processing method of Embodiment 4 configured in this way, each time the surrogate model 143 is retrained, the response values ​​of the individual model calculated in previous generations are replaced with response values ​​calculated using the latest surrogate model 143, thereby enabling more accurate optimization of the individual model's design parameters. Furthermore, since the calculation of response values ​​using the surrogate model 143 is performed instantaneously, it does not affect the computation time of the optimization calculation.

[0077] In the above embodiment, an example was described in which the response value is re-evaluated in step S23, but the calculation unit 11 may be configured to re-evaluate the response value at any time. Also, although an example applied to the vertical method embodiment 1 was described, the response value may be configured to be re-evaluated in the horizontal method embodiment 2 as well. For example, in step S41 shown in Figure 12, the response value may be re-evaluated and replaced when the surrogate model 143 is retrained.

[0078] (Embodiment 5) The information processing device 1 according to Embodiment 5 differs from Embodiment 1 in that it performs individual number expansion in the vertical optimization process as in Embodiment 2. The other configurations and processes of the information processing device 1 are the same as those of the information processing device 1 according to Embodiment 1, so the same reference numerals are used for the same parts, and detailed explanations are omitted.

[0079] Figure 18 is a conceptual diagram showing the first generation in the vertical optimization calculation according to Embodiment 5, Figure 19 is a conceptual diagram showing the third generation in the vertical optimization calculation according to Embodiment 5, and Figure 20 is a flowchart showing the main parts of the optimization calculation process according to Embodiment 5. Note that the 0th and 2nd generations are the same as those shown in Figures 3 and 5 of Embodiment 1.

[0080] First, the calculation unit 11 performs processes such as generating N initial individuals and calculating the FEA of the response value, similar to steps S11 to S15 of Embodiment 1 (steps S51 to S55).

[0081] Next, the calculation unit 11 determines whether or not it is an SM calculation generation (step S56). If it is determined to be an SM calculation generation (step S56: YES), the calculation unit 11 generates N × k individuals in the current generation, as shown in Figures 18 and 19 (step S57). k is a value greater than 1, and in this case, it generates more individual models than the number of individuals N that should be generated in the current generation.

[0082] Then, the calculation unit 11 performs SM calculation on the response values ​​of the N × k individual models (child individuals) generated in step S57 using the surrogate model 143 (step S58).

[0083] Next, the calculation unit 11 selects the N best-performing individual models from among the N×k individual models based on the response values ​​of the N×k individual models, according to the purpose of optimizing the design parameters (step S59), and aggregates them (step S60).

[0084] On the other hand, if it is determined that it is an FEA calculation generation (step S56: NO), the calculation unit 11 generates N individuals in the current generation (step S61). Then, the calculation unit 11 performs FEA calculation on the response values ​​of the N individual models (child individuals) generated in step S61 using the CAE model 142 (step S62), and aggregates the design parameters and response values ​​of the individual models (step S63). Meanwhile, the calculation unit 11 uses the aggregated design parameters of each individual model and the FEA-calculated response values ​​as training data to retrain the surrogate model 143 (step S64).

[0085] Having completed the processing in step S60 or step S64, the arithmetic unit 11 performs the same processing as in steps S24 to S27 (steps S65 to S68).

[0086] The above processing details will be explained using Figures 18 and 19. In the first generation, the calculation unit 11 generates N × k individuals in the current generation based on the design parameters of the individual model inherited from the 0th generation, as shown in Figure 18.

[0087] The calculation unit 11 then performs a SM calculation on the response values ​​of the generated N × k individual models (child models), and based on the response values ​​of the N × k individual models, selects the N individual models with the highest evaluation according to the design parameter optimization objective. Next, the calculation unit 11 evaluates and ranks the N child models that have been SM calculated and selected, and the N individual models inherited from the 0th generation, and selects the N individual models with the highest rank. The N individual models selected in this way are then passed on to the next generation.

[0088] In the second generation, the calculation unit 11 creates second-generation child individual models in the same manner as in Embodiment 1, calculates response values ​​by FEM calculation, aggregates design parameters and response values, ranks the individuals, and selects N individual models. The N individual models selected in the second generation include individuals whose response values ​​were calculated using FEA and individuals whose response values ​​were calculated using SM.

[0089] In the third generation, the calculation unit 11 generates N × k individuals in the current generation based on the design parameters of the individual models inherited from the second generation, as shown in Figure 19, similar to the first generation. The calculation unit 11 then performs a SM calculation on the response values ​​of the generated N × k individual models (child individuals) and selects the N individual models with the highest evaluation according to the design parameter optimization objective based on the response values ​​of the N × k individual models. Next, the calculation unit 11 evaluates and ranks the N child individuals that have been SM calculated and selected, along with the N individual models inherited from the second generation, and selects the N individual models with the highest rank.

[0090] Similarly, generations in which response values ​​are calculated using SM calculations and generations in which response values ​​are calculated using FEA calculations are repeated, thereby optimizing the design parameters of the individual model.

[0091] According to the information processing method, etc., of Embodiment 5 configured in this way, even in a vertical approach, by increasing the number of individuals generated in each generation, a convergence solution for the design parameters can be obtained more quickly.

[0092] Furthermore, the population expansion described in Embodiment 5 may also be applied to modified examples relating to any combination of the vertical and horizontal methods described above.

[0093] (Embodiment 6) The information processing device 1 according to Embodiment 6 differs from Embodiment 1 in how it handles individual models inherited from the previous generation in the vertical FEA calculation generation. The other configurations and processes of the information processing device 1 are the same as those of the information processing device 1 according to Embodiment 1, so the same reference numerals are used for the same parts, and detailed explanations are omitted.

[0094] Figure 21 is a conceptual diagram showing the second generation in the vertical optimization calculation according to Embodiment 6, and Figure 22 is a conceptual diagram showing the third generation in the vertical optimization calculation according to Embodiment 6. Note that the 0th and 1st generations are the same as those shown in Figures 3 and 4 of Embodiment 1.

[0095] The arithmetic unit 11 of the information processing device 1 performs the same processing as in steps S11 to S15 of Embodiment 1 to create a 0th generation individual model and aggregate the response values. Also, similar to Embodiment 1, the arithmetic unit 11 creates a 1st generation child individual model, calculates the response value by SM calculation, aggregates the design parameters and response values, ranks the individuals, and selects N individual models. The N individual models selected in the 1st generation include individuals whose response values ​​were calculated by FEA and individuals whose response values ​​were calculated by SM.

[0096] In the second generation, the calculation unit 11 inherits the N individual models selected from the first generation, generates N individual models (offspring individuals) through genetic processing based on the design parameters of the individual models, and calculates the response values ​​of the generated offspring individuals using FEA.

[0097] As shown in Figure 21, the calculation unit 11 selects individuals from among the N individual models inherited from the first generation for which the response value has been calculated using FEA. The X shown in Figure 21 is the number of individual models for which the response value has been calculated using SM.

[0098] In Embodiment 6, the (NX) individuals whose response values ​​have been calculated using FEA and the N individuals whose response values ​​have been calculated using FEA in the second generation are the targets of selection. The calculation unit 11 ranks the (2N-X) individuals whose response values ​​have been calculated using FEA and selects the top-ranked N individual models. In this way, the response values ​​of the N individual models selected through selection are all calculated using FEA and are carried over to the next generation. Meanwhile, the calculation unit 11 uses the design parameters and response values ​​of (2N-X) individual models, whose response values ​​have been calculated using FEA, as training data to retrain the surrogate model 143.

[0099] Next, as shown in Figure 22, the calculation unit 11 creates third-generation child individual models for the N individual models inherited from the second generation in the same manner as the first generation, calculates response values ​​using SM calculation, aggregates design parameters and response values, ranks the individuals, and selects the N individual models through selection. In the third generation, the N individual models selected through selection include individuals whose response values ​​were calculated using FEA and individuals whose response values ​​were calculated using SM, just like in the first generation.

[0100] Similarly, generations in which response values ​​are calculated using SM calculations and generations in which response values ​​are calculated using FEA calculations are repeated, thereby optimizing the design parameters of the individual model.

[0101] According to the information processing method of Embodiment 6 configured in this way, in the FEA calculation generation, the response values ​​of all selected and eliminated individual models are calculated using FEA. Therefore, it is possible to prevent the problem of unsuitable individual models remaining in later generations. For example, if an individual model has inappropriate design parameters, such as a shape breakdown, but the SM-calculated response value is highly evaluated, such an individual model will remain in later generations. Embodiment 6 can resolve this problem. [Explanation of Symbols]

[0102] 1: Information Processing Device 10: Recording media 11: Arithmetic section 12:Display section 13:Operation section 14: Storage section 141: Computer Programs 142: CAE Model 143: Surrogate Model

Claims

1. An information processing method in which a computer optimizes design parameters that define an individual model representing an object to be designed, using a genetic algorithm, The aforementioned computer, A generation step in which multiple new individual models of the current generation are generated by genetic processing based on the design parameters of a predetermined number of individual models inherited from the previous generation, A calculation step in which response values ​​representing the characteristics of multiple individual models are calculated using the first calculation method, A ranking step in which a predetermined number of individual models inherited from the previous generation and several newly generated individual models are ranked based on the calculated response values, Based on the ranking results, a selection step is taken to select a predetermined number of top-ranked individual models to be passed on to the next generation. Execute, The generation step or the calculation step is The number of generations or the number of individuals in each generation is increased by calculating the response values ​​that represent the characteristics of the individual model using a second calculation method, which has a faster calculation speed than the first calculation method. Information processing methods.

2. The first calculation method is a method for calculating response values ​​using CAE analysis including the finite element method. The second computation method is a method that uses a machine learning model to calculate the response value. The information processing method according to claim 1.

3. The calculation step is, The system includes a surrogate operation step that calculates the response values ​​of multiple newly generated individual models in the current generation using a second operation method. The information processing method according to claim 1 or claim 2.

4. The generation step is, The steps include generating N × k (k > 1) new individual models for the current generation based on a predetermined number of individual model design parameters inherited from the previous generation, and The steps include: calculating the response values ​​of the newly generated current generation N × k individual models using a second calculation method; The process involves selecting N top-performing individual models from among the N×k individual models based on the response values ​​of N×k individual models, according to the objective of optimizing the design parameters. Includes, The calculation step is, The response values ​​of the N selected top-ranking individual models are calculated using the first calculation method. The information processing method according to claim 1 or claim 2.

5. The aforementioned computer, The second calculation method involves determining whether the shape represented by the design parameters of the individual model, for which the response value was calculated, is broken or not. The steps include: applying a penalty to the response value of an individual model with a broken shape, or excluding an individual model with a broken shape from the current generation; The information processing method according to claim 1 or claim 2, which performs the following.

6. The aforementioned computer, The steps include: retraining the machine learning model using the design parameters of multiple individual models and the response values ​​of multiple individual models calculated by the first calculation method as training data; If, before retraining the aforementioned machine learning model, there is an individual model for which the response value has already been calculated using the second calculation method, the response value of that individual model is recalculated using the machine learning model after retraining. The information processing method according to claim 2, which performs the following:

7. The generation step is, The steps include generating N × k (k > 1) new individual models for the current generation based on a predetermined number of individual model design parameters inherited from the previous generation, and Equipped with, The calculation step is, The steps involve calculating the response values ​​of the current generation's new N × k individual models using a second calculation method, The process involves selecting N top-performing individual models from among the N×k individual models based on the response values ​​of N×k individual models, according to the objective of optimizing the design parameters. The information processing method according to claim 1 or claim 2, comprising:

8. An information processing device comprising a calculation unit that performs a process to optimize design parameters that define an individual model representing an object to be designed, using a genetic algorithm, The aforementioned arithmetic unit, A generation step in which multiple new individual models of the current generation are generated by genetic processing based on the design parameters of a predetermined number of individual models inherited from the previous generation, A calculation step in which response values ​​representing the characteristics of multiple individual models are calculated using the first calculation method, A ranking step in which a predetermined number of individual models inherited from the previous generation and several newly generated individual models are ranked based on the calculated response values, Based on the ranking results, a selection step is taken to select a predetermined number of top-ranked individual models to be passed on to the next generation. It is configured to execute, The generation step or the calculation step is The number of generations or the number of individuals in each generation is increased by calculating the response values ​​that represent the characteristics of the individual model using a second calculation method, which has a faster calculation speed than the first calculation method. Information processing device.

9. A computer program that causes a computer to perform information processing to optimize design parameters that define an individual model representing the object to be designed, using a genetic algorithm, To the aforementioned computer, A generation step in which multiple new individual models of the current generation are generated by genetic processing based on the design parameters of a predetermined number of individual models inherited from the previous generation, A calculation step in which response values ​​representing the characteristics of multiple individual models are calculated using the first calculation method, A ranking step in which a predetermined number of individual models inherited from the previous generation and several newly generated individual models are ranked based on the calculated response values, Based on the ranking results, a selection step is taken to select a predetermined number of top-ranked individual models to be passed on to the next generation. Make it run, The generation step or the calculation step is The number of generations or the number of individuals in each generation is increased by calculating the response values ​​that represent the characteristics of the individual model using a second calculation method, which has a faster calculation speed than the first calculation method. Computer program.

Images