Information Processing Method, Information Processing Apparatus, and Computer Program

Abstract

Provided is an information processing method capable of shortening the calculation time required for optimizing design parameters of a design object, as compared with an optimization calculation by a genetic algorithm using CAE analysis such as the finite element method. 【Solution means】An information processing method for optimizing design parameters that define an individual model representing a design object by a genetic algorithm based on FEA calculation, which increases the number of generations while suppressing the calculation time and calculation amount by using a surrogate model in a specific generation, or increases the number of individual models in each generation 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 that describe the structure of the inductor, calculating the characteristics of the inductor having the structure described by the variables based on the attribute information of the core and coil conductor input in advance by magnetic field analysis, calculating an evaluation value using a predetermined evaluation function having at least the inductor characteristics as parameters, determining whether the evaluation value satisfies a predetermined convergence condition, ending the process if the convergence condition is satisfied, and varying the value of the variable by a genetic algorithm if 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] An object of the present disclosure is to provide an information processing method, an information processing apparatus, and a computer program capable of shortening the calculation time required for optimizing design parameters of a design object compared to optimization calculation by a genetic algorithm using CAE analysis such as the finite element method.

Means for Solving the Problems

[0006] An information processing method according to an aspect of the present disclosure is an information processing method in which a computer optimizes design parameters defining an individual model representing a design object using a genetic algorithm. The computer includes: a generation step of generating a plurality of new individual models of the current generation by genetic processing based on design parameters of a predetermined number of individual models inherited from the previous generation; a calculation step of calculating response values indicating characteristics of the plurality of individual models by a first calculation method; a ranking step of ranking the predetermined number of individual models inherited from the previous generation and the plurality of newly generated individual models based on the calculated response values; and a selection step of selecting a predetermined number of individual models of the upper rank to be inherited to the next generation based on the ranking result. In the generation step or the calculation step, the number of generations or the number of individuals in each generation is increased by calculating a response value indicating the characteristics of the individual model by a second calculation method having a faster calculation speed than the first calculation method.

[0007] An information processing apparatus according to an aspect of the present disclosure is an information processing apparatus including an arithmetic unit that executes a process of optimizing design parameters that define an individual model representing a design object using a genetic algorithm. The arithmetic unit includes: a generation step of generating a plurality of new individual models in the current generation by genetic processing based on design parameters of a predetermined number of individual models inherited from the previous generation; an arithmetic step of calculating a response value indicating characteristics of the plurality of individual models by a first arithmetic method; a ranking step of ranking a predetermined number of individual models inherited from the previous generation and the plurality of newly generated individual models based on the calculated response value; and a selection step of selecting a predetermined number of individual models of higher ranks to be inherited to the next generation based on the ranking result. In the generation step or the arithmetic step, the number of generations or the number of individuals in each generation is increased by calculating a response value indicating characteristics of the individual model by a second arithmetic method having a faster arithmetic speed than the first arithmetic method.

[0008] A computer program according to an aspect of the present disclosure is a computer program that causes a computer to execute information processing for optimizing design parameters that define an individual model representing a design object using a genetic algorithm. The computer is caused to execute: a generation step of generating a plurality of new individual models in the current generation by genetic processing based on design parameters of a predetermined number of individual models inherited from the previous generation; an arithmetic step of calculating a response value indicating characteristics of the plurality of individual models by a first arithmetic method; a ranking step of ranking a predetermined number of individual models inherited from the previous generation and the plurality of newly generated individual models based on the calculated response value; and a selection step of selecting a predetermined number of individual models of higher ranks to be inherited to the next generation based on the ranking result. In the generation step or the arithmetic step, the number of generations or the number of individuals in each generation is increased by calculating a response value indicating characteristics of the individual model by a second arithmetic method having a faster arithmetic speed than the first arithmetic method.

Advantages of the Invention

[0009] According to the present disclosure, compared with the optimization calculation by a genetic algorithm using CAE analysis such as the finite element method, the calculation time required for optimizing the design parameters of a design object can be shortened.

Brief Description of the Drawings

[0010]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Figure 13

Figure 14

Figure 15

Figure 16

Figure 17

Figure 18

Figure 19

Figure 20

Figure 21

Figure 22

Embodiments for Carrying Out the Invention

[0011] 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 defined by the claims, and is intended to include all modifications within the meaning and scope equivalent to the claims. Also, at least a part of the embodiments described below may be arbitrarily combined.

[0012] (Embodiment 1) 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.

[0013] Note that the information processing apparatus 1 may be a stand-alone computer or a server device connected to a network. Also, the information processing apparatus 1 may be a computer in an on-premises environment or a computer such as a server in a cloud environment. The information processing apparatus 1 may be configured by a plurality of computers for distributed processing, may be realized by a plurality of virtual machines provided in one server, or may be realized using a cloud server.

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

[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 an input device such as a hardware keyboard, a pointing device, or a touch panel. A user of the information processing apparatus 1 can input arbitrary information into the information processing apparatus 1 using the operation unit 13. Note that the operation unit 13 may be integrally configured with the display unit 12.

[0017] The storage unit 14 has, 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 execute 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 a computer program 141, a CAE model 142, and a surrogate model (SM) 143 that are executed by the arithmetic unit 11. The surrogate model 143 may be created in advance or may be created during the optimization process of the design parameters related to the object to be designed.

[0018] The CAE model 142 includes, for example, a three-dimensional shape model such as three-dimensional CAD data representing the shape of the object to be designed, and material characteristics of each part constituting the three-dimensional shape model. Specifically, when the object to be designed is a motor, the CAE model 142 includes a three-dimensional shape model such as three-dimensional CAD data representing the shapes of a plurality of coils, a stator, and a rotor constituting the motor, and data such as material characteristics, boundary conditions, and voltage conditions of each part constituting the three-dimensional shape model. Examples of the material characteristics include magnetization characteristics, electrical characteristics, and mechanical characteristics. The various data defining the CAE model 142 include the design parameters to be optimized in the first embodiment. The design parameters include dimensional parameters and parameters indicating topology. By changing the design parameters, CAE models 142 having various shapes and characteristics, that is, individual models, can be generated.

[0019] The calculation unit 11 calculates response values related to the characteristics or behaviors of the individual model by analyzing the CAE model 142 using a known calculation method (the first calculation method) such as the finite element method (FEA: Finite Element Analysis) or the boundary element method. That is, it simulates the characteristics or behaviors of the individual model. For example, the calculation unit 11 performs magnetic field analysis on the CAE model 142 of the 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 of predicting the behavior or characteristics of the entire motor by dividing the rotor and stator of the motor, which have complex shapes and electromagnetic characteristics, into small regions (elements) with simple shapes and electromagnetic characteristics and approximately calculating the characteristics of each simplified element. Hereinafter, in the first embodiment 1, the calculation process of the response value using the CAE model 142 as the individual model is appropriately referred to as FEA calculation.

[0020] The surrogate model 143 is a machine learning model that outputs response values related to the characteristics or behaviors of the individual model when the design parameters that define the individual model of the design object described above are input. The surrogate model 143 includes, for example, a trained neural network (NN: Neural Network) by machine learning. The surrogate model 143 has an input layer to which the design parameters that define the individual model are input, an intermediate layer that extracts the feature amounts of the individual model, and an output layer that outputs response values related to the characteristics or behaviors of the individual model.

[0021] The calculation unit 11 calculates response values related to the characteristics or behaviors of the individual model by a calculation method (the second calculation method) using the surrogate model 143. The calculation process of the response value using the surrogate model 143 is performed more quickly than the FEA calculation. Hereinafter, in the first embodiment 1, the calculation process of the response value using the surrogate model 143 as the individual model is appropriately referred to as SM calculation.

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

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

[0024] Although a neural network has been described as an example of the surrogate model 143, a model using SVR (support vector regression), decision tree, linear regression, etc. may also be used, and the method is not limited.

[0025] Note that the auxiliary storage unit may be an external storage device connected to the information processing device 1. The computer program 141 and the surrogate model 143 may be written in the storage unit 14 at the manufacturing stage of the information processing device 1, or the information processing device 1 may acquire through communication what is distributed by an external server and store it in the storage unit 14. The computer program 141 and the surrogate model 143 may be in a mode recorded on a recording medium 10 such as a magnetic disk, an optical disk, or a semiconductor memory so as to be readable, and may be read by a reading device from the recording medium 10 and stored in the storage unit 14.

[0026] <Information Processing Method> An optimization process using a genetic algorithm will be described. In motor design and the like using magnetic field analysis, an optimization calculation using a GA (genetic algorithm) is often used. The optimization calculation is, for example, a calculation for calculating a motor design plan that improves the average torque of the motor and suppresses iron loss. The design parameters of the individual model are, for example, the position of the permanent magnet constituting the rotor of the motor (for example, the distance from the center of the rotor to the center of the permanent magnet) and the shape (in the case of a rectangular cross-section, the longitudinal dimension and the transverse dimension), and the values characterizing the shape of the flux barrier. The response values calculated by the CAE model 142 and the surrogate model 143 of the individual model are, for example, the average value of torque and the iron loss value in the rotor.

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

[0028] In FIG. 2, the arrows indicate the calculation processes of each generation in the genetic algorithm. The white arrows indicate the process of calculating the response value of the individual model by FEA calculation, and the black arrows indicate the process of calculating the response value of the individual model by SM calculation. The length of the arrow represents the magnitude of the calculation time and calculation amount. The left figure shows the calculation time when calculating the response value of the individual model of each generation using only FEA calculation, and the right figure shows the calculation time when replacing the response value calculation in a specific generation with SM calculation.

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

[0030] First, the arithmetic unit 11 substitutes 0 into a variable m indicating the generation (hereinafter, generation number m) (step S11). Next, as shown in FIG. 3, the arithmetic unit 11 generates N individual models (initial individuals) of the initial generation (step S12). N is an integer of 2 or more. Specifically, the arithmetic 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 specifically, the arithmetic unit 11 generates N sets of design parameters. The design parameters of the initial generation (0th generation) may be generated by random numbers. The method for generating the design parameters of the initial individuals is not particularly limited.

[0031] Next, the arithmetic unit 11 performs FEA calculations on the response values of each generated individual model based on the design parameters of each individual model (step S13), and the arithmetic unit 11 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 an arithmetic module that optimizes the design parameters of the individual model using a genetic algorithm.

[0032] Then, the arithmetic unit 11 creates a surrogate model 143 using the design parameters of each individual model and the response values calculated in step S13 as training data (step S15). Note that if the surrogate model 143 has already been created in advance, the learning process of the surrogate model 143 may be omitted. Also, the surrogate model 143 may be re-learned. Further, if the surrogate model 143 has already been created in advance, it may be configured to calculate the response value of the individual model from the 0th generation by SM.

[0033] After finishing the process of step S15, the arithmetic unit 11 determines whether it is a specific SM calculation generation (step S16). For example, the storage unit 14 stores data indicating the generation for which the response value should be calculated by FEA calculation and the generation for which the response value should be calculated by SM calculation, and the arithmetic unit 11 determines whether it is an SM calculation generation by calculating the data. Also, the arithmetic unit 11 may determine that odd generations are SM calculation generations and even generations are FEA calculation generations. The method for determining whether 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 in the current generation are generated (step S17). Specifically, based on the design parameters of the N individual models inherited from the previous generation, the arithmetic unit 11 generates new N individual models in the current generation through genetic processes such as crossover and mutation. Crossover is an operation of generating an individual model with new design parameters by combining the design parameters of a plurality of individual models inherited from the previous generation. Mutation is an operation of generating an individual model with new design parameters by randomly changing a part of the design parameters of a plurality of individual models inherited from the previous generation. The arithmetic unit 11 may generate new individual models through both operations of crossover and mutation. In the first generation, as shown in FIG. 4, the arithmetic unit 11 directly inherits the individual models of the 0th generation, which is the previous generation, and generates N individual models based on the design parameters of the individual models. In the mth generation after the second generation, as shown in FIGS. 5 and 6, the arithmetic unit 11 inherits the N individual models selected by elimination from the (m - 1)th generation and generates N individual models through genetic processes based on the design parameters of the individual models.

[0035] Next, as shown in FIGS. 4 and 6, the calculation unit 11 calculates the SM using the response values of the N individual models (child individuals) generated in step S16 with the surrogate model 143 (step S18). Then, the calculation unit 11 aggregates the generated N design parameters and the response value of the individual model to the GA engine (step S19). Since the response values of the N individual models inherited from the previous generation have already been calculated, they are used as they are.

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

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

[0038] Then, the calculation unit 11 ranks the 2N individual models based on the response values of the 2N individual models (step S24), and selects the N individual models with the higher ranks (step S25). For example, the arithmetic unit 11 classifies 2N individual models in order from the one closer to the Pareto front by non-dominated sorting in the NSGA-II algorithm into populations or individuals, and assigns a rank to each classified population or individual. The smaller the rank number of a population or an individual, the higher the rank (higher evaluation), that is, the closer the population or individual is to the Pareto front. As the rank number increases, the rank becomes lower (lower evaluation). For example, when performing a motor design to improve the average torque of a motor as the design object and suppress 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 the number of individuals belonging to a certain rank number may be plural or single. Also, the number of individuals belonging to each rank differs depending on the rank. The number of ranks obtained by classifying 2N individuals differs depending on the design parameters of the 2N individual models. Then, among the 2N individual models, N individual models that are far from the Pareto front and have a large rank number are eliminated, and N individual models that are close to the Pareto front and have a small rank number are carried over to the next generation. When selecting individuals in order from the one with the smallest rank number and the number of selected individuals exceeds N, by performing crowding sort, some individuals with a high degree of crowding are selected so that the number of selected individuals becomes N.

[0039] Then, the arithmetic unit 11 determines whether the current generation (m-th generation) is the M generations set as the number of repetitions of the optimization calculation (step S26). When it is determined that the current generation is not the M generations (step S26: NO), the arithmetic unit 11 increments the generation number m by 1 (step S27) and returns the process to step S16. When it is determined that the generation is the M generations (step S26: YES), the arithmetic unit 11 ends the process.

[0040] According to the information processing method and the like according to the first embodiment configured as described above, by vertically substituting the FEA calculation in some generations with the SM calculation, the calculation time required for optimizing the design parameters of the design object can be shortened compared to the optimization calculation performed entirely by FEA.

[0041] FIG. 8 is a conceptual diagram showing an optimization calculation of the vertical method according to a modified example. FIG. 8, in the same manner as FIG. 2, shows the arithmetic processing of each generation in the genetic algorithm by arrows. In the optimization processing shown in FIGS. 2 to 6, an example in which the FEA calculation generation and the SM calculation generation appear alternately has been described. However, as shown in FIG. 8, the arithmetic unit 11 may repeat the SM calculation generation a predetermined number of times to optimize the design parameters. That is, the arithmetic unit 11 may execute the SM calculation with a bias toward a specific generation instead of performing the SM calculation evenly over generations 0 to M. For example, the arithmetic unit 11 may optimize the design parameters by executing the SM calculation at a high cycle in the first half of the generations.

[0042] In addition, in the first embodiment, although the magnetic field analysis has been described as a simulation method using the motor and the CAE model 142 as the simulation target, the simulation target and the simulation content are not particularly limited. For example, the information processing method according to the present embodiment can be applied to the optimization of the magnetic circuit of the solenoid valve and the optimization of the magnetic circuit in the plastic magnet. Further, the simulation method may be a thermal analysis, an electric field analysis, a structural analysis, a fluid analysis, etc., or may be an analysis process combining a plurality of analysis processes, for example, a magnetic field analysis and a structural analysis.

[0043] In addition, as an example of the first arithmetic method, the FEA calculation using the CAE model 142 and, as an example of the second arithmetic method, the calculation using the surrogate model 143 which is a machine learning model have been illustrated. However, the contents of the first arithmetic method and the second arithmetic method are not particularly limited. As long as the arithmetic accuracy of the second arithmetic method is lower than that of the first arithmetic method and the arithmetic speed of the second arithmetic method is faster than that of the first arithmetic method, the contents of the first and second arithmetic methods are not particularly limited. For example, as the first arithmetic method and the second arithmetic method, any models such as the CAE model 142, SVR (support vector regression), decision tree, linear regression, and neural network can be used.

[0044] (Embodiment 2) The information processing apparatus 1 according to Embodiment 2 is different from that of Embodiment 1 in the application location of SM calculation. Since the other configurations and processes of the information processing apparatus 1 are the same as those of the information processing apparatus 1 according to Embodiment 1, the same reference numerals are assigned to the same parts, and detailed descriptions thereof are omitted.

[0045] FIG. 9 is a conceptual diagram showing the optimization calculation of the horizontal method according to Embodiment 2. Similar to FIG. 2, FIG. 9 shows the arithmetic processing of each generation in the genetic algorithm by arrows. The black arrows indicate the process of calculating the response value of the individual model by SM calculation, and the horizontal length indicates that the number of individuals for calculating the response value is larger than the normal N.

[0046] When there are many design parameters of the object to be designed, it is necessary to increase the number of individuals in each generation in order to obtain a converged solution, and there is a problem that the calculation time and the amount of calculation increase. The information processing method according to the present Embodiment 2 generates N×k (k>1) individual models (child individuals) as described later, calculates the response value of each individual by SM calculation, selects N individual models with higher evaluation according to the purpose of design parameter optimization, and performs FEA calculation on the selected N individual models with higher evaluation, thereby making it possible to obtain a converged solution of the design parameters of the individual model more quickly. Note that when the number of generations to be calculated is the same and the number of child individuals is also the same, the calculation time is not shorter than that when optimizing the design parameters only by FEA calculation, but the time until a converged solution is obtained is shorter. That is, it can be expected to reduce the number of generations until a converged solution is obtained.

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

[0048] First, similar to steps S11 to S15 of Embodiment 1, the arithmetic unit 11 executes processes such as generation of N initial individuals and FEA calculation of response values (steps S31 to S35).

[0049] Next, as shown in FIG. 10, the arithmetic unit 11 generates N×k individuals in the current generation (step S36). k is a value greater than 1, and here, an individual model with a number of individuals larger than the N individuals to be generated in the current generation is generated.

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

[0051] Next, based on the response values of the N×k individual models, the arithmetic unit 11 selects N top - evaluated individual models according to the design parameter optimization objective from among the N×k individual models (step S38). The design parameter optimization objective is, for example, to improve the average torque of the motor, which is the object of design, and suppress iron loss. For example, it is advisable to select N individual models using an evaluation function in which the evaluation value increases as the average torque is larger and the iron loss is smaller. In this case, the arithmetic unit 11 passes the average torque and iron loss, which are the response values of the individuals, as arguments to the evaluation function to calculate the evaluation value, and selects N top - evaluated individual models with a large calculated evaluation value.

[0052] The arithmetic unit 11 performs FEA calculation on the response values of the N individual models (child individuals) 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 arithmetic unit 11 retrains the surrogate model 143 using the aggregated design parameters and response values of the individual models (child individuals) as training data (step S41).

[0053] Next, the arithmetic unit 11 executes processes such as ranking and elimination selection of the generated individual models in the same manner as in steps S24 to S27 of the first embodiment (steps S42 to S45). Note that from the second generation onward, as shown in FIG. 11, the N individual models selected by elimination from the previous generation are inherited, and the generation, ranking, and elimination selection processes of child individuals are executed in the same manner as in the first generation.

[0054] According to the information processing method and the like according to the second embodiment configured as described above, similar to the first embodiment, the calculation time required for optimizing the design parameters of the design object can be shortened. In particular, according to the second embodiment, even when there are many design parameters of the design object, the number of individuals in each generation can be increased while suppressing the calculation time, and a convergent solution of the design parameters can be obtained more quickly.

[0055] FIG. 13 is a graph showing the effect of the optimization calculation according to the present 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 for optimizing the design parameters. The hypervolume index is the volume of the region surrounded by the so-called Pareto front and the reference point, and is an index that becomes a steady value when converging to the Pareto front. When the design parameters defining the individual model are optimized to the convergent solution by the multi-objective genetic algorithm, the hypervolume index becomes a steady value.

[0056] As can be seen from FIG. 13, by incorporating the SM calculation, the optimal solution of the design parameters can be obtained more quickly than when performing the optimization process only by the FEA calculation. Depending on the characteristics of the design object, the number of design parameters, etc., each method may be tried and an appropriate method may be selected as appropriate.

[0057] Although the examples of executing the SM calculation in the vertical method and the horizontal method have been described in the first embodiment and the second embodiment so far, the arithmetic unit 11 may execute the optimization process of the design parameters by combining the vertical method and the horizontal method.

[0058] FIG. 14 is a conceptual diagram showing an optimization calculation according to a first modified example combining a vertical method and a horizontal method, and FIG. 15 is a conceptual diagram showing an optimization calculation according to a second modified example combining a vertical method and a horizontal method. For example, as shown in FIG. 14, the calculation unit 11 can be configured to calculate the response value of the individual model in each generation using the vertical SM calculation in the first half of the generations, and calculate the response value using the horizontal SM calculation in the second half of the generations. Conversely, the calculation unit 11 can be configured to calculate the response value of the individual model in each generation using the horizontal SM calculation in the first half of the generations, and calculate the response value using the vertical SM calculation in the second half of the generations. Further, the calculation unit 11 may arbitrarily combine the vertical SM calculation shown in FIGS. 2 and 8 and the horizontal SM calculation shown in FIG. 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 the vertical SM calculation, the horizontal SM calculation, and the FEA calculation.

[0059] (Embodiment 3) The information processing apparatus 1 according to Embodiment 3 differs from that of Embodiment 1 in that it avoids the defect that an individual model having a shape-breaking design parameter remains due to the SM calculation. Since the other configurations and processes of the information processing apparatus 1 are the same as those of the information processing apparatus 1 according to Embodiment 1, the same reference numerals are given to the same parts, and detailed descriptions thereof are omitted.

[0060] When generating a child individual, the calculation unit 11 may set a design parameter with a shape break as a model of the object to be designed. That is, an individual model that cannot be established as the object to be designed may be generated. Even if an attempt is made to perform FEA calculation using the design parameters of such an individual model, since the shape is broken, the response value cannot be calculated. However, when the surrogate model 143 is used, a seemingly correct response value is calculated. If a response value of a higher evaluation is calculated, such an individual model may remain until later generations despite the shape break, and there is a risk of a problem in the optimization calculation.

[0061] To solve such a problem, in Embodiment 3, it is determined whether the shape created by the design parameters of the child individuals generated by the calculation unit 11 is valid, and the individual models with shape breakdown are not passed on to the subsequent generations. Hereinafter, specific processing procedures will be described.

[0062] FIG. 16 is a flowchart showing the main part of the optimization calculation process according to Embodiment 3. The calculation unit 11 executes the following processing, for example, in step S24 of Embodiment 1. The calculation unit 11 verifies the validity of the design parameters of the individuals for which the response values have been calculated by SM calculation (step S241).

[0063] For example, the calculation unit 11 determines whether the sides constituting the shape represented by the design parameters of the individual model to be verified intersect. If the sides intersect, the calculation unit 11 determines that the shape has broken down.

[0064] In addition, the calculation unit 11 attempts FEA calculation using the design parameters of the individual model to be verified, and if the FEA calculation can be executed halfway, it is determined that the shape can be established as the shape of the object to be designed. Specifically, when the calculation unit 11 can continue the FEA calculation for a predetermined time or more, it determines that the shape has not broken down, and when a calculation error occurs halfway, it determines that the shape has broken down.

[0065] Next, the calculation unit 11 assigns a penalty to the response value of the individual model determined to have a shape breakdown (step S242). That is, the calculation unit 11 changes the response value so that the evaluation of the individual model decreases.

[0066] According to the information processing method and the like according to Embodiment 3 configured as described above, by processing so that the individual models with shape breakdown do not remain in the later generations, the design parameters of the individual models can be optimized more accurately.

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

[0068] In the above embodiment, an example of applying a penalty to the response value has been described. However, when it is determined that the shape is broken, it may be configured to exclude the individual model in the current generation.

[0069] Also, although an example of executing the above process in step S24 shown in FIG. 7 has been described, in step S18, the processes of step S241 and step S242 may be configured to be executed.

[0070] Furthermore, also in the horizontal optimization process shown in Embodiment 2, the process of this Embodiment 3 may be applied to determine the presence or absence of shape breakage. For example, in step S37, the processes of step S241 and step S242 may be configured to be executed.

[0071] (Embodiment 4) The information processing apparatus 1 according to Embodiment 4 is different from Embodiment 1 in that when the surrogate model 143 is re-learned, the existing individual models are re-evaluated. Since the other configurations and processes of the information processing apparatus 1 are the same as those of the information processing apparatus 1 according to Embodiment 1, the same reference numerals are given to the same parts, and detailed descriptions are omitted.

[0072] In both the vertical method and the horizontal method, the accuracy of the surrogate model 143 tends to improve as generations pass. This is due to the increase in training data as generations pass. Therefore, it is predicted that the response value calculated by SM in the first half of the generation is larger than the response value calculated by SM in the second half of the generation.

[0073] Therefore, in the fourth embodiment, in order to reduce the error of the response value calculated by the SM calculation, the response value of the individual model calculated in the past generations is re-evaluated using the latest surrogate model 143. As a result, it is expected to improve the accuracy of the optimization calculation. The following describes the specific processing procedure.

[0074] FIG. 17 is a flowchart showing the main part of the optimization calculation process according to the fourth embodiment. The calculation unit 11 executes the following processes in step S23 of the first embodiment, for example. The calculation unit 11 re-learns the surrogate model 143 (step S231) as described in step S23.

[0075] Next, the calculation unit 11 identifies the individual models among the individual models existing in the current generation for which the response value was calculated using the surrogate model 143 before re-learning (step S232). The response value of the identified individual model is recalculated using the latest surrogate model 143 after re-learning (step S233). That is, the calculation unit 11 replaces the response value calculated by the SM in the past generations with the response value calculated by the SM using the latest surrogate model 143.

[0076] According to the information processing method and the like according to the fourth embodiment configured as described above, each time the surrogate model 143 is re-learned, the response value of the individual model calculated in the past generations is replaced with the response value calculated using the latest surrogate model 143, so that the design parameters of the individual model can be optimized more accurately. In addition, since the calculation process of the response value using the surrogate model 143 is performed instantaneously, it does not affect the calculation time of the optimization calculation.

[0077] In the above-described embodiment, an example in which the response value is re-evaluated in step S23 has been described. However, the arithmetic unit 11 may be configured to re-evaluate the response value at an arbitrary timing. Further, although an example of applying to the first embodiment of the vertical method has been described, it may be configured to re-evaluate the response value in the second embodiment of the horizontal method. For example, in step S41 shown in FIG. 12, when the surrogate model 143 is re-learned, the response value may be re-evaluated and replaced.

[0078] (Embodiment 5) The information processing apparatus 1 according to the fifth embodiment is different from the first embodiment in that in the vertical method optimization process, the population size is expanded as in the second embodiment. Since the other configurations and processes of the information processing apparatus 1 are the same as those of the information processing apparatus 1 according to the first embodiment, the same reference numerals are given to the same parts, and detailed descriptions thereof are omitted.

[0079] FIG. 18 is a conceptual diagram showing the first generation in the vertical method optimization calculation according to the fifth embodiment, FIG. 19 is a conceptual diagram showing the third generation in the vertical method optimization calculation according to the fifth embodiment, and FIG. 20 is a flowchart showing the main part of the optimization calculation process according to the fifth embodiment. Note that the 0th generation and the 2nd generation are the same as those shown in FIGS. 3 and 5 of the first embodiment.

[0080] First, the arithmetic unit 11 executes processes such as generation of N initial individuals and FEA calculation of response values, in the same manner as steps S11 to S15 of the first embodiment (steps S51 to S55).

[0081] Next, the arithmetic unit 11 determines whether it is the SM calculation generation (step S56). If it is determined that it is the SM calculation generation (step S56: YES), the arithmetic unit 11 generates N×k individuals in the current generation as shown in FIGS. 18 and 19 (step S57). k is a value greater than 1, and here, an individual model with a larger number of individuals than the N individuals to be generated in the current generation is generated.

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

[0083] Next, the arithmetic unit 11 selects (step S59) and aggregates (step S60) the top N evaluated individual models according to the purpose of design parameter optimization from among the N×k individual models based on the response values of the N×k individual models.

[0084] On the other hand, when it is determined that it is the FEA calculation generation (step S56: NO), the arithmetic unit 11 generates N individuals in the current generation (step S61). Then, the arithmetic 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). On the other hand, the arithmetic unit 11 retrains the surrogate model 143 using the design parameters of each aggregated individual model and the response values obtained by the FEA calculation as training data (step S64).

[0085] After finishing the process of step S60 or step S64, the arithmetic unit 11 executes the same processes as steps S24 to S27 (steps S65 to S68).

[0086] The above-described processing content will be described with reference to FIGS. 18 and 19. In the first generation, as shown in FIG. 18, the arithmetic unit 11 generates N×k individuals in the current generation based on the design parameters of the individual models inherited from the 0th generation.

[0087] Then, the calculation unit 11 calculates the SM of the response values of the generated N×k individual models (child individuals), and selects N top-ranked individual models 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 for which the SM has been calculated and the N individual models inherited from the 0th generation, and selects the N individual models with the top ranks. The N individual models thus selected by elimination are inherited to the next generation.

[0088] In the second generation, in the same manner as in Embodiment 1, the calculation unit 11 creates the child individual models of the second generation, calculates the response values by FEM calculation, aggregates the design parameters and the response values, ranks the individuals, and selects N individual models by elimination. Among the N individual models selected by elimination in the second generation, there are a mixture of individuals for which the response values have been calculated by FEA calculation and individuals for which the response values have been calculated by SM calculation.

[0089] In the third generation, in the same manner as in the first generation, as shown in FIG. 19, based on the design parameters of the individual models inherited from the second generation, the calculation unit 11 generates N×k individuals in the current generation. Then, the calculation unit 11 calculates the SM of the response values of the generated N×k individual models (child individuals), and selects N top-ranked individual models 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 for which the SM has been calculated and the N individual models inherited from the second generation, and selects the N individual models with the top ranks.

[0090] Thereafter, in the same manner, the generations in which the response values are calculated by SM calculation and the generations in which the response values are calculated by FEA calculation are repeated, and the design parameters of the individual models are optimized.

[0091] According to the information processing method and the like according to Embodiment 5 configured as described above, even in the vertical method, by increasing the number of individuals generated in each generation, it is possible to obtain a convergent solution of the design parameters more quickly.

[0092] Note that the population expansion described in Embodiment 5 may be applied to a modification according to any combination of the vertical method and the horizontal method described above.

[0093] (Embodiment 6) In the vertical FEA calculation generation according to Embodiment 6, the handling of the individual models inherited from the previous generation is different from that in Embodiment 1. Since the other configurations and processes of the information processing apparatus 1 are the same as those of the information processing apparatus 1 according to Embodiment 1, the same reference numerals are given to the same parts, and detailed descriptions thereof are omitted.

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

[0095] The calculation unit 11 of the information processing apparatus 1 executes the same processes as steps S11 to S15 of Embodiment 1, creates an individual model of the 0th generation, and aggregates response values. Further, the calculation unit 11 creates a child individual model of the 1st generation in the same manner as in Embodiment 1, calculates a response value by SM calculation, aggregates design parameters and response values, ranks individuals, and selects N individual models by elimination selection. Among the N individual models selected by elimination selection in the 1st generation, there are a mixture of individuals for which the response value has been FEA calculated and individuals for which the response value has been SM calculated.

[0096] In the 2nd generation, the calculation unit 11 inherits the N individual models selected by elimination selection from the 1st generation, generates N individual models (child individuals) by genetic processing based on the design parameters of the individual models, and calculates the response values of the generated child individuals by FEA calculation.

[0097] As shown in FIG. 21, the calculation unit 11 selects, from among the N individual models inherited from the 1st generation, the individuals for which the response value has been FEA calculated. The X shown in FIG. 21 is the number of individual models for which the response value has been SM calculated.

[0098] In Embodiment 6, the (N-X) individuals for which response values have been calculated by FEA and the N individuals for which response values have been calculated by FEA in the second generation are subject to elimination selection. The calculation unit 11 ranks the (2N-X) individuals for which response values have been calculated by FEA and selects the N individual models with the highest ranks. In this way, the response values of the N individual models selected by elimination selection are all calculated by FEA and are carried over to the next generation. On the other hand, the calculation unit 11 uses the design parameters and response values of the (2N-X) individual models for which response values have been calculated by FEA as training data and retrains the surrogate model 143.

[0099] Next, as shown in FIG. 22, for the N individual models inherited from the second generation, the calculation unit 11 creates child individual models of the third generation in the same manner as in the first generation, calculates response values by SM calculation, aggregates the design parameters and response values, ranks the individuals, and selects N individual models by elimination selection. Among the N individual models selected by elimination selection in the third generation, as in the first generation, there are a mixture of individuals for which response values have been calculated by FEA and individuals for which response values have been calculated by SM calculation.

[0100] Thereafter, in the same manner, the generations in which response values are calculated by SM calculation and the generations in which response values are calculated by FEA calculation are repeated, and the design parameters of the individual models are optimized.

[0101] According to the information processing method and the like according to Embodiment 6 configured as described above, in the FEA calculation generation, the response values of all the individual models selected by elimination selection are calculated by FEA. Therefore, it is possible to prevent the problem that inappropriate individual models remain until later generations. For example, when the evaluation of the response value calculated by SM is high even though the design parameters of the individual model are inappropriate, such as being geometrically flawed, such an individual model will remain until later generations. According to Embodiment 6, such a problem can be solved.

Explanation of Signs

[0102] 1: Information processing apparatus 10: Recording medium 11: Calculation unit 12: Display unit 13: Operation unit 14: Memory unit 141: Computer program 142: CAE model 143: Surrogate model

Claims

1. An information processing method for a computer to optimize design parameters defining an individual model representing a design object using a genetic algorithm, comprising: the computer:[[]] a generation step of generating a plurality of new individual models of the current generation by genetic processing based on design parameters of a predetermined number of individual models inherited from the previous generation; a calculation step of calculating response values indicating characteristics of the plurality of individual models by a first calculation method; a ranking step of ranking a predetermined number of individual models inherited from the previous generation and the plurality of newly generated individual models based on the calculated response values; a selection step of selecting a predetermined number of individual models of the upper ranks to be inherited to the next generation based on the ranking result and executing, the generation step:[[]] a step of generating N×k (k>1) new individual models of the current generation based on design parameters of a predetermined number of individual models inherited from the previous generation; a step of calculating response values of the N×k individual models of the newly generated current generation by a second calculation method; a step of selecting N individual models with upper evaluations according to the purpose of design parameter optimization from the N×k individual models based on the response values of the N×k individual models and including, the calculation step:[[]] calculating response values of the selected N individual models with upper evaluations by a first calculation method information processing method.

2. An information processing method for a computer to optimize design parameters defining an individual model representing a design object using a genetic algorithm, comprising: the computer:[[]] a generation step of generating a plurality of new individual models of the current generation by genetic processing based on design parameters of a predetermined number of individual models inherited from the previous generation; a calculation step of calculating response values indicating characteristics of the plurality of individual models by a first calculation method; a ranking step of ranking a predetermined number of individual models inherited from the previous generation and the plurality of newly generated individual models based on the calculated response values; a selection step of selecting a predetermined number of individual models of the upper ranks to be inherited to the next generation based on the ranking result and executing, the generation step:[[]] comprising a step of generating N×k (k>1) new individual models of the current generation based on design parameters of a predetermined number of individual models inherited from the previous generation, the calculation step:[[]] When new N individual models of the current generation are generated, calculating the response values of the N individual models by a first calculation method; When new N×k individual models of the current generation are generated, calculating the response values of the new N×k individual models of the current generation by a second calculation method; Based on the response values of the N×k individual models, selecting N individual models with top evaluations according to the purpose of design parameter optimization from among the N×k individual models comprising an information processing method.

3. The first calculation method is a method of calculating response values by CAE analysis including the finite element method, The second calculation method is a method of calculating response values using a machine learning model The information processing method according to claim 1 or claim 2.

4. The computer using the design parameters of a plurality of individual models and the response values of the plurality of individual models calculated by the first calculation method as training data to relearn the machine learning model; before the relearning of the machine learning model, if there is an individual model for which the response value has been calculated by the second calculation method, recalculating the response value of the individual model using the machine learning model after the relearning The information processing method according to claim 3, which executes.

5. The computer determining whether the shape represented by the design parameters of the individual model for which the response value has been calculated by the second calculation method is defective; imposing a penalty on the response value of the individual model with a defective shape, or excluding the individual model with a defective shape from the current generation The information processing method according to claim 1 or claim 2, which executes.

6. An information processing apparatus including an arithmetic unit that executes a process of optimizing design parameters that define an individual model representing a design object using a genetic algorithm, The arithmetic unit a generation step of generating a plurality of new individual models of the current generation by genetic processing based on the design parameters of a predetermined number of individual models inherited from the previous generation; an arithmetic step of calculating response values indicating the characteristics of a plurality of individual models by a first calculation method; a ranking step of ranking a predetermined number of individual models inherited from the previous generation and a plurality of newly generated individual models based on the calculated response values; a selection step of selecting a predetermined number of individual models with top ranks to be inherited to the next generation based on the ranking result is configured to execute In the generation step, the arithmetic unit generates new N × k (k > 1) individual models of the current generation based on the design parameters of a predetermined number of individual models inherited from the previous generation, calculates the response values of the newly generated N × k individual models of the current generation by a second arithmetic method, selects N top - evaluated individual models according to the purpose of design parameter optimization from the N × k individual models based on the response values of the N × k individual models and executes, In the arithmetic step, the arithmetic unit calculates the response values of the selected N top - evaluated individual models by a first arithmetic method Information processing apparatus. **Claim 7** An information processing apparatus including an arithmetic unit that executes a process of optimizing design parameters that define an individual model representing a design object using a genetic algorithm, wherein the arithmetic unit performs a generation step of generating new plural individual models of the current generation by genetic processing based on the design parameters of a predetermined number of individual models inherited from the previous generation, an arithmetic step of calculating response values indicating characteristics of the plural individual models by a first arithmetic method, a ranking step of ranking a predetermined number of individual models inherited from the previous generation and the newly generated plural 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 to the next generation based on the ranking result and is configured to execute, In the generation step, the arithmetic unit generates new N × k (k > 1) individual models of the current generation based on the design parameters of a predetermined number of individual models inherited from the previous generation, In the arithmetic step, the arithmetic unit when new N individual models of the current generation are generated, calculates the response values of the N individual models by a first arithmetic method, when new N × k individual models of the current generation are generated, calculates the response values of the new N × k individual models of the current generation by a second arithmetic method, and selects N top - evaluated individual models according to the purpose of design parameter optimization from the N × k individual models based on the response values of the N × k individual models Information processing apparatus that executes. **Claim 8** A computer program that causes a computer to execute information processing for optimizing design parameters that define an individual model representing a design object using a genetic algorithm to cause the computer to a generation step of generating a plurality of new individual models of the current generation by genetic processing based on design parameters of a predetermined number of individual models inherited from the previous generation; a calculation step of calculating response values indicating characteristics of the plurality of individual models by a first calculation method; a ranking step of ranking a predetermined number of individual models inherited from the previous generation and the plurality of newly generated individual models based on the calculated response values; a selection step of selecting a predetermined number of individual models of the upper ranks to be inherited to the next generation based on the ranking result to execute, the generation step includes a step of generating N×k (k>1) new individual models of the current generation based on design parameters of a predetermined number of individual models inherited from the previous generation; a step of calculating response values of the N×k individual models of the newly generated current generation by a second calculation method; a step of selecting, from the N×k individual models, N individual models with upper evaluations according to the purpose of design parameter optimization based on the response values of the N×k individual models and the calculation step calculates the response values of the selected N individual models with upper evaluations by a first calculation method A computer program. [

9. ] A computer program for causing a computer to execute information processing for optimizing design parameters defining an individual model representing a design object using a genetic algorithm, to cause the computer to a generation step of generating a plurality of new individual models of the current generation by genetic processing based on design parameters of a predetermined number of individual models inherited from the previous generation; a calculation step of calculating response values indicating characteristics of the plurality of individual models by a first calculation method; a ranking step of ranking a predetermined number of individual models inherited from the previous generation and the plurality of newly generated individual models based on the calculated response values; a selection step of selecting a predetermined number of individual models of the upper ranks to be inherited to the next generation based on the ranking result to execute, the generation step includes a step of generating N×k (k>1) new individual models of the current generation based on design parameters of a predetermined number of individual models inherited from the previous generation, the calculation step when N new individual models of the current generation are generated, a step of calculating the response values of the N individual models by a first calculation method When a new N×k individual model of the current generation is generated, calculating the response value of the new N×k individual models of the current generation by means of a second calculation method; selecting, based on the response values of the N×k individual models, N individual models with top evaluations according to the purpose of design parameter optimization from among the N×k individual models; and comprising a computer program.

Images