Multi-performance optimization design device and multi-performance optimization design method

Abstract

The present invention provides a multi-performance optimization design device and a multi-performance optimization design method. The multi-performance optimization design device performs the following processing, that is, fills in the discrete observed values obtained for each of a plurality of performances through simulation, respectively, so as to output continuous predicted values and prediction errors for each of the plurality of performances, and based on the predicted values and the prediction errors, calculates a plurality of calculation points for searching the regions where each of the plurality of performances can be executed for each of the plurality of performances, calculates the probability distributions where each of the plurality of performances can be executed for each of the plurality of performances by using the calculated plurality of calculation points, and outputs the total product of the probability distributions calculated for each of the plurality of performances as the multi-performance executable region.

Description

Technical Field

[0001] The present disclosure relates to a multi-performance optimization design device related to the design of a structure such as a vehicle body, and a multi-performance optimization design method. Background Art

[0002] In the design of a structure, establishing a design method that can optimize multiple performance characteristics (multi-performance) such as strength, rigidity, weight reduction, and vibration suppression, which may be contradictory depending on the situation, has been set as one of the important issues, and a method of optimizing each of the multi-performance characteristics simultaneously in parallel through computer simulation is being studied.

[0003] In Japanese Patent Laid-Open No. 2010-55466, an invention of a product optimization design support system is disclosed. The product optimization design support system can analyze and evaluate changes in evaluation indices of a design required for a product, such as performance and cost, by using a design method for obtaining the setting cardinality of a design solution as a set related to multi-performance while considering various uncertainties. Summary of the Invention

[0004] Problems to be Solved by the Invention

[0005] However, in the technology of Japanese Patent Laid-Open No. 2010-55466, as the dimension of the problems involved in the study of multi-performance increases, the search space grows exponentially, so there is a problem that the computational cost for calculating the executable region of multi-performance becomes huge.

[0006] In addition, in the technology of Japanese Patent Laid-Open No. 2010-55466, there are problems as follows: when the number of valid solutions that meet the conditions is small, it becomes difficult to obtain information related to the boundary of the executable region of multi-performance, and sampling of new variables related to additional multi-performance conditions also becomes difficult.

[0007] An object of the present disclosure is to efficiently obtain the executable region of multi-performance.

[0008] Means for Solving the Problems

[0009] The first method is a multi-performance optimization design device, which includes: a memory; a processor connected to the memory, and the processor is configured to fill in the discrete observation values obtained by simulation for each of the multiple performances respectively, so as to output continuous predicted values and prediction errors for each of the multiple performances, calculate multiple calculation points for searching the area where each of the multiple performances can be executed based on the predicted values and the prediction errors, use the calculated multiple calculation points to calculate the probability distribution where each of the multiple performances can be executed, and output the total product of the probability distributions calculated for each of the multiple performances as the multi-performance executable area.

[0010] Since the multi-performance optimization design device of the first method fills in the discrete observation values obtained by simulation to output continuous predicted values, in the simulation, it is not necessary to derive a large number of observation values. As a result, the computing cost in the simulation can be suppressed.

[0011] The multi-performance optimization design device of the first method can efficiently calculate the calculation points for searching the multi-performance executable area based on the calculated predicted values and prediction errors.

[0012] The multi-performance optimization design device of the first method can express the executable area in the form of a probability distribution, so that it can be expressed by continuous values with the achievability of each performance as a scale.

[0013] The second method is that in the multi-performance optimization design device of the first method, the processor is configured to, when a new performance constraint occurs, fill in the discrete observation values obtained by simulation for the new performance, so as to output continuous predicted values and prediction errors for the new performance, calculate multiple calculation points for searching the area where the new performance can be executed based on the predicted values and prediction errors of the new performance, calculate the probability distribution where the new performance can be executed by using the multiple calculation points for searching the area where the new performance can be executed, and output the product of the total product of the probability distributions calculated for each of the multiple performances and the probability distribution where the new performance holds as the new multi-performance executable area.

[0014] Even when a new performance constraint occurs, the multi-performance optimization design device of the second method can update the multi-performance executable area by multiplying the probability distribution related to the new performance constraint by the probability distribution representing the current multi-performance executable area.

[0015] The third method is that in the multi-performance optimization design device of the first method or the second method, the processor is configured to set the point where the product of a first acquisition function related to the search inside the executable region and a second acquisition function related to the search near the boundary between the executable region and the non-executable region becomes the maximum as the calculation point.

[0016] The multi-performance optimization design device of the third method can search the boundary and the inside of the executable region simultaneously, thereby calculating the calculation point.

[0017] The fourth method is that in the multi-performance optimization design device of any one of the first to third methods, the processor is configured to end the calculation of the calculation point when the ratio of the region where the calculation point has not been calculated to the entire region involved in the design is less than a predetermined threshold.

[0018] The multi-performance optimization design device of the fourth method can set a calculation point sufficient to determine a multi-performance execution region in the region involved in the design.

[0019] The fifth method is a multi-performance optimization design method, which performs the following processing through a processor, that is, fills in the discrete observed values obtained by simulation for each of the multiple performances respectively, thereby outputting continuous predicted values and prediction errors for each of the multiple performances, calculates multiple calculation points for searching the regions where each of the multiple performances can be executed based on the predicted values and the prediction errors, uses the calculated multiple calculation points to calculate the probability distribution where each of the multiple performances can be executed for each of the multiple performances, and outputs the total product of the probability distributions calculated for each of the multiple performances as the multi-performance executable region.

[0020] Since the multi-performance optimization design method of the fifth method fills in the discrete observed values obtained by simulation and outputs continuous predicted values, a large number of observed values do not need to be derived in the simulation. As a result, the operation cost in the simulation can be suppressed.

[0021] The multi-performance optimization design method of the fifth method can efficiently calculate the calculation points for searching the multi-performance executable region based on the calculated predicted values and prediction errors.

[0022] The multi-performance optimization design method of the fifth method can express the executable region in the form of a probability distribution, and thus can be expressed by continuous values with the achievability of each performance as a scale.

[0023] The sixth method is as follows. In the multi-performance optimization design method of the fifth method, when a new performance limit occurs, the discrete observed values obtained through simulation for the new performance are filled, so as to output a continuous predicted value and a prediction error for the new performance. Based on the predicted value and the prediction error of the new performance, a plurality of calculation points for searching the region where the new performance can be executed are calculated. Using the plurality of calculation points for searching the region where the new performance can be executed, the probability distribution where the new performance can be executed is calculated. The total product of the probability distributions calculated for each of the plurality of performances and the product of the probability distribution where the new performance holds are output as the new multi-performance executable region.

[0024] Even when a new performance limit occurs, the multi-performance optimization design method of the sixth method can update the multi-performance executable region by multiplying the probability distribution related to the new performance limit by the probability distribution representing the current multi-performance executable region.

[0025] The seventh method is as follows. In the multi-performance optimization design method of the fifth method or the sixth method, the point where the product of the first acquisition function related to the search inside the executable region and the second acquisition function related to the search near the boundary between the executable region and the non-executable region becomes the maximum is set as the calculation point.

[0026] The multi-performance optimization design method of the seventh method can calculate the calculation point by searching the boundary and the inside of the executable region simultaneously.

[0027] The eighth method is as follows. In any one of the multi-performance optimization design methods from the fifth method to the seventh method, when the ratio of the region where the calculation point has not been calculated to the entire region related to the design is less than a predetermined threshold, the calculation of the calculation point is ended.

[0028] The multi-performance optimization design method of the eighth method can set calculation points sufficient to determine the multi-performance execution region in the region related to the design. The first method to the eighth method can also be implemented in the form of a computer-readable recording medium.

[0029] Advantages of the Invention

[0030] According to the present disclosure, the executable region of multi-performance can be efficiently obtained. Brief Description of the Drawings

[0031] Figure 1 It is a block diagram showing an example of the structure of the multi-performance optimization design device according to the embodiment.

[0032] Figure 2It is a functional block diagram of the CPU of the multi-performance optimization design device related to the embodiment.

[0033] Figure 3 It is a schematic diagram showing the hierarchical structure of design and verification in the multi-performance optimization design related to the embodiment.

[0034] Figure 4 It is a conceptual diagram of the system design in the embodiment.

[0035] Figure 5 It is a flowchart showing an example of the derivation process of the executable region of machine learning related to the embodiment.

[0036] Figure 6 It is a schematic diagram showing an example of the regression of a one-dimensional function using Gaussian process regression.

[0037] Figure 7 It is an example of the probability distribution of the constraint conditions derived by Gaussian process regression.

[0038] Figure 8 It is for showing the acquisition function a PoF (x) is a schematic diagram showing an example of the calculation result obtained.

[0039] Figure 9 It is for showing the acquisition function a ES (x) is a schematic diagram showing an example of the calculation result obtained.

[0040] Figure 10 It is for showing the hybrid acquisition function a PoF-ES (x) is a schematic diagram showing an example of the calculation result obtained.

[0041] Figure 11 It is an explanatory diagram showing the concept of joint probability distribution calculation. Detailed implementation mode

[0042] Hereinafter, using Figure 1 , the multi-performance optimization design device and the multi-performance optimization design method related to the present embodiment will be described. Figure 1 It is a block diagram showing an example of the specific structure of the multi-performance optimization design device 10 related to the embodiment of the present invention.

[0043] The multi-performance optimization design device 10 is configured to include a computer 30. The computer 30 includes a CPU (Central Processing Unit) 32 as an example of a hardware processor, a ROM (Read Only Memory) 34 as an example of a memory, a RAM (Random Access Memory) 36, and an input / output port 38. As an example, preferably, the computer 30 is a model such as an engineer workstation or a supercomputer that can execute advanced arithmetic processing at high speed.

[0044] In the computer 30, the CPU 32, the ROM 34, the RAM 36, and the input / output port 38 are connected to each other via various buses such as an address bus, a data bus, and a control bus. To the input / output port 38, a display 40, a mouse 42, a keyboard 44, a hard disk (HDD) 46, and a disk drive 50 that reads information from various disks (e.g., CD-ROM or DVD, etc.) 48 are respectively connected as various input / output devices.

[0045] In addition, a network 52 is connected to the input / output port 38 and is configured to be able to exchange information with various devices connected to the network 52. In the present embodiment, it is configured that a data server 56 connected to a database (DB) 54 is connected to the network 52 and information can be exchanged with the DB 54.

[0046] In the DB 54, data related to multi-performance optimization design and the like are stored in advance. The storage of information in the DB 54 can be performed either by the computer 30 or the data server 56, or by other devices connected to the network 52.

[0047] Although in the present embodiment, a structure in which data related to multi-performance optimization design and the like are stored in the DB 54 connected to the data server 56 has been described, it is also possible to adopt a method in which the information of the DB 54 is stored in the HDD 46 built in the computer 30 or an external storage device such as an external hard disk.

[0048] In the HDD 46 of the computer 30, a multi-performance optimization design program for multi-performance optimization design is installed. In the present embodiment, the multi-performance optimization design program is loaded and executed by the CPU 32, thereby performing multi-performance optimization design. In addition, the CPU 32 causes the display 40 to display the processing result obtained by the multi-performance optimization design program. Further, for the multi-performance optimization design program of the present embodiment, there are several methods of installing it in the computer 30. For example, the multi-performance optimization design program is stored together with the setup program in a non-temporary computer-readable recording medium such as a CD-ROM (Compact Disc Read-Only Memory) or a DVD (Digital Video Disc), and the disk such as the CD-ROM or the DVD is placed in the disk drive 50, and by executing the setup program for the CPU 32, the multi-performance optimization design program is installed in the HDD 46 which is an example of a non-temporary computer-readable recording medium or a memory. Alternatively, the following method may be adopted, that is, by communicating with other information processors connected to the computer 30 via a public telephone line or the network 52, the multi-performance optimization design program is installed in the HDD 46.

[0049] Figure 2 FIG. is a functional block diagram of the CPU 32 of the multi-performance optimization design device 10. Various functions realized by loading and executing the multi-performance optimization design program by the CPU 32 of the multi-performance optimization design device 10 will be described. The multi-performance optimization design program includes a simulation function of obtaining observed values for each of a plurality of performances by simulation, an observed value filling function of filling the obtained discrete observed values to output continuous predicted values and prediction errors for each of the plurality of performances, a calculation point calculation function of calculating a plurality of calculation points for searching regions where each of the plurality of performances can be executed, a probability distribution calculation function of calculating a probability distribution where each of the plurality of performances can be executed through the plurality of calculation points, and a multi-performance executable region output function of outputting the total product of the probability distributions calculated for each of the plurality of performances as a multi-performance executable region. By executing the multi-performance optimization design program having the above respective functions by the CPU 32, the CPU 32 functions as Figure 2 a simulation unit 72, an observed value filling unit 74, a calculation point calculation unit 76, a probability distribution calculation unit 78, and a multi-performance executable region output unit 80 as shown.

[0050] Figure 3A schematic diagram showing a hierarchical structure of design and verification in the multi-performance optimization design according to this embodiment. In the design of a structure such as a vehicle, the design of the system as a whole of the structure, the design of the subsystem as the lower structure of the structure, and the design of structural elements such as components as the lower structure of the subsystem are respectively required. In the multi-performance optimization design, an executable region is derived, and the executable region is a region in which various types of performance, such as strength, rigidity, weight reduction, and vibration suppression, which are different from each other and may be opposite to each other depending on the situation, can be optimized respectively in each of the system, subsystem, and structural elements. The appropriateness of the executable region optimized in a multi-performance manner is confirmed through verification. When it is determined through verification that the executable region is inappropriate, redesign is performed. The result of the redesign is verified again, and the appropriateness of the executable region is judged.

[0051] This design and verification are respectively executed in the structural elements, subsystems, and systems. In order to derive an executable region in the multi-performance optimization design, generally, the following order is implemented. First, a model between the performance-related variables and the response of the variables is defined. Next, candidates for the executable region are obtained by random sampling on the previously defined model. Then, according to the obtained results, specimens that satisfy the response constraint conditions are extracted. Although this step is relatively simple in principle, when the dimension of the design variables increases, the space to be searched will increase exponentially, and as a result, there is a problem that the calculation cost becomes huge.

[0052] In this embodiment, a SetBase Concurrent design method using machine learning (Active learning) is used to derive a multi-performance executable region. By using machine learning, it is possible to suppress the exponential increase in the calculation cost.

[0053] In addition, in this embodiment, by expressing the executable region in the form of a probability distribution, it is possible to independently obtain the executable region of each performance in the multi-performance, and by multiplying the executable regions of the respective performances, it is possible to easily derive the multi-performance executable region. In addition, even when new constraint conditions are arranged, it is possible to independently derive the probability distribution related to the new constraint conditions and multiply the derived probability distribution by the above-mentioned multi-performance executable region to derive the multi-performance executable region considering the new constraint conditions.

[0054] Figure 4 A conceptual diagram of the system design in this embodiment. In Figure 4In (1), in the initial stage of design, the executable regions that effectively satisfy the limiting conditions of the respective performances of Performance 1, Performance 2, and Performance 3 are derived as probability distributions Pr(C i (x)) (i = 1, 2, 3). Since Pr(C i (x)) in this embodiment is the probability related to the realization of each performance i, it becomes as follows. C i (x) is a Boolean function with x as a variable.

[0055] 0 ≤ Pr(C i (x)) ≤ 1

[0056] Figure 4 In (2), an example of an expression obtained from the probability of the multi-performance executable region is shown. As described above, since the probability distributions respectively realized by Performance 1 to 3 are Pr(C i (x)), the multi-performance executable region, which is the region where Performance 1 to 3 are simultaneously realized, is represented by the total product of the probability distributions related to each performance.

[0057] Figure 4 In (3), a case is shown where an additional request for specifications is made when transferring to production after the product is designed. Pr(C new (x)) is derived as the probability distribution for realizing the new restrictions related to the additional request.

[0058] Figure 4 In (4), a case of updating the multi-performance executable region according to the additional request is shown. As described above, since the probability distribution related to the new restrictions is Pr(C new (x)), the multi-performance executable region can be updated by multiplying Pr(C new (x)) by the multi-performance executable region derived in (2). Figure 4

[0059] Figure 5 This is an example of a flowchart related to the derivation of the executable region of machine learning according to this embodiment. In step 400, the conditions for realizing multiple performances are input. As an example, the conditions input in step 400 are the constraint functions gi(x) for each performance in the following multiple performances. The subscript i in the constraint function is the index for each performance in the multiple performances, and K is the number of performances constituting the multiple performances.

[0060] g i (x) ≤ 0, i ∈ {1, 2,..., K}

[0061] ​In addition, the probability of being able to execute each performance among multiple performances is represented by the following formula (1). As described above, Ci(x) in the left side of formula (1) is a Boolean function with x as a variable. δi in the right side of formula (1) is a small positive value representing an allowable error.

[0062]

[0063] In addition, the region where performances i (i = 1, 2, …, K) are simultaneously achieved, that is, the multi-performance executable region, is represented by the following formula (2) as the total product of probabilities related to each performance.

[0064]

[0065] In step 402, an experimental plan is generated, which is used to derive the region of multi-performance y such as strength, rigidity, weight reduction, and vibration suppression with respect to the variable x such as the position of the structure or the moment of inertia acting on the structure. In step 404, the design defined according to the experimental plan is evaluated by simulation such as CAE (Computer Aided Engineering). Through simulation such as CAE, for example, the y value corresponding to the variable x is discretely derived as an observed value.

[0066] In step 406, prediction of the observed value y and filling based on the prediction are performed. In the present embodiment, a method such as Gaussian process regression is used, that is, a method that can fill and predict the observed value y by considering the correlation between the observed value y and the variable x. Gaussian process regression generally determines the correlation between the variable x and the observed value y through a Gaussian distribution, and is characterized in that it can not only randomly and continuously fill the discrete observed value y, but also calculate the prediction error.

[0067] Figure 6 It is a schematic diagram showing an example of regression of a one-dimensional function using Gaussian process regression. In Figure 6 a curve 102 assuming the case where the observed values 100A, 100B, 100C, 100D, 100F are continuous is shown. As Figure 6 shown, the curve 102 becomes a continuous function corresponding to the variable x. Since it is a continuous function, not only can discrete data be filled and predicted, but also differentiation can be performed with the variable x. In Figure 6 a region of prediction error 106 exists around the curve 102. Regarding the region of the prediction error 106, it becomes narrower when the reliability of the predicted value shown by the curve 102 is higher, but becomes wider when the reliability is lower. In Figure 6In this example, the inequality limit value 104 for y = 0 is shown. In the present embodiment, the region where the response y is less than the inequality limit value 104 is defined as the executable region.

[0068] In the present embodiment, the cumulative density distribution (CDF) with respect to the variable x is calculated using the filling of discrete data obtained by Gaussian process regression, the prediction error 106, and the inequality limit value 104.

[0069] Figure 7 This is an example of the probability distribution of the constraint conditions derived by Gaussian process regression. In Figure 7 the horizontal axis represents the variable x, and the vertical axis represents the value of the cumulative density distribution CDF. Figure 7 The cumulative density distribution 110 represents the probability when the value of y in Figure 6 becomes less than or equal to the inequality limit value 104. When using the cumulative density distribution Φ, as expressed by the following formula (1A), the above formula (1) calculates the probability distribution Pr(Ci(x)) with respect to the variable x according to the cumulative density distribution Φ. In the following formula, b is a value related to the lower boundary 134 described later. In addition, σ(x) in the following formula is the prediction deviation calculated during the calculation of Gaussian process regression, and μ(x) is the prediction average. Although calculating the probability distribution for all points in the region (design space) related to the design defined by variables x such as the position of the structure or the moment of inertia acting on the structure through formula (1A) would result in a huge computational cost and is not realistic, in the present embodiment, by sequentially calculating the calculation points for searching the multi-performance executable region using the acquisition function described later, the information of these calculation points is used to update the teacher data of machine learning. Moreover, based on the updated teacher data, the probability distribution where multi-performance holds in the design space is calculated.

[0070]

[0071] In the present embodiment, using the results (predicted values, prediction errors) of Gaussian process regression, the calculation points for searching the multi-performance executable region in the design space are calculated. The calculation points can be obtained in the following manner.

[0072] In step 408, the calculation of the acquisition function is performed. In the present embodiment, two acquisition functions with different properties are defined based on the results of Gaussian process regression, and each acquisition function is used to evenly search for the calculation points.

[0073] One of the acquisition functions is a PoF (Probability of Feasibility) PoF (x).a PoF (x) is used to search for computation points inside a multi-performance executable region. Figure 8 To indicate that a PoF (x) is a schematic diagram of an example of the calculation results obtained. Figure 8 , an executable area 120 is shown surrounded by an upper boundary 124 relative to a non-executable area 122, and a non-executable area 132 exists inside the executable area 120 via a lower boundary 134. Figure 8 In FIG. 1 , executable computation points 126 are indicated by black circles, and unexecutable computation points 130 are indicated by white triangles. Figure 8 As shown, a PoF (x) is suitable for searching inside the executable area 120 rather than near the upper boundary 124 or the lower boundary 134. PoF (x) is represented by the following formula (3).

[0074]

[0075] In formula (3), Φ is the cumulative density distribution, a is a value related to the upper boundary 124, and b is a value related to the lower boundary 134. In addition, in formula (3), σ2(x) is the prediction variance calculated in the calculation process of Gaussian process regression in order to search for the executable area under the condition of a<y<b, σ(x) is the prediction deviation, and μ(x) is the prediction mean. In this embodiment, by replacing a shown in formula (3) with PoF (x) is set to the maximum, thus generating a new calculation point.

[0076] Another acquisition function is a ES (Entropy Search) ES (x).a ES (x) is used to search for a calculation point near the boundary (upper boundary 124, lower boundary 134) between the executable area 120 and the non-executable areas 122, 132. Figure 9 To indicate that a ES A schematic diagram of an example of the calculation results obtained by (x). Figure 9 There are executable computation points 126 and inexecutable computation points 128 near the upper boundary 124 and the lower boundary 134, so a ES (x) is suitable for searching near the upper boundary 124 or the lower boundary 134. ES(x) is represented by the following formula (4). In formula (4), H(p(f(x))) is the entropy (Shannon information content).

[0077]

[0078] Furthermore, in the above formula (4), the following is the censored Gaussian distribution for which the entropy can be analytically calculated.

[0079]

[0080]

[0081]

[0082] In the present embodiment, by setting a ES (x) shown in formula (4) to be maximum, new calculation points are generated.

[0083] Although the above two are the acquisition functions related to the present embodiment, when using two different acquisition functions separately, individual processing needs to be performed for each function. In the present embodiment, the acquisition function a PoF-ES (x) represented by the following formula (5) is used to calculate the calculation points for searching the executable region.

[0084] a POF-ES (x) = a PoF (x) · a ES (x)…(5)

[0085] The right side of formula (5) is the product of a PoF (x) and a ES (x). In the present embodiment, the acquisition function a PoF-ES (x) shown in formula (5) is called the mixed acquisition function.

[0086] The following formula (6) is the formula for generating new calculation points (i.e., the variable x). As shown in formula (6), the new calculation point x new is calculated as the point where the mixed acquisition function a PoF-ES (x) is maximum.

[0087]

[0088] By setting the mixed acquisition function a PoF-ES (x) to be maximum, it is possible to simultaneously set each of the acquisition functions a PoF (x), a ES (x) to be maximum, and there is no need to separately process the acquisition functions a PoF (x), a ES(x) Perform operations individually.

[0089] In step 410, the teacher data is updated by adding the newly calculated points obtained using the above formula (6) to the teacher data of the machine learning. Figure 10 To represent the calculation result obtained by the mixed acquisition function a PoF-ES (x) is a schematic diagram showing an example of the calculation result. In Figure 10 , not only inside the executable region 120, but also near the upper boundary 124 and the lower boundary 134, there are executable calculation points 126. Based on the updated teacher data, the executable region 120 composed of the executable calculation points 126 is shown in the form of a probability distribution Pr(Ci(x)) that satisfies the constraint conditions of the performance i (i = 1, 2,..., K) by the above formula (1A).

[0090] In step 412, it is judged whether the end condition of the process is satisfied. The end conditions of step 412 are defined by the following formulas (7), (8), and (9) respectively, and the convergence judgment of the calculation is performed using the end condition shown in formula (9). Formula (9) represents the ratio of the region where it is not possible to sufficiently perform the establishment / non - establishment judgment with respect to the entire region, that is, the ratio of the region where the calculation points are not calculated to the region involved in the design. δk in formula (8) is a small positive value representing the allowable error. In addition, ε on the right side of formula (9) is a threshold representing the end condition and is a small positive value. When the end condition of the process is satisfied in step 412, the step is transferred to step 414. When the end condition of the process is not satisfied in step 412, the step is transferred to step 404 to continue the calculation of new calculation points.

[0091]

[0092]

[0093]

[0094] In step 414, the output of the model representing the executable region is implemented. If the executable regions of each performance can be obtained as the probability distribution Pr(Ci(x)), then as shown in the above formula (2), it is possible to easily obtain the multi - performance executable region that satisfies all performance constraints as the joint probability distribution. The following shows formula (2) again.

[0095]

[0096] Figure 11 It is an explanatory diagram showing the concept of the joint probability distribution calculation shown in formula (2). Figure 11In the case of deriving a multi-performance executable region when three performances of performance 1 to 3 are given, and Figure 4 It is an explanatory diagram that combines the descriptions of (1) and (2) of Figure 4 . By calculating the joint probability distribution of each probability distribution, the multi-performance executable region can be obtained. In addition, for a region determined not to be a multi-performance executable region, it is possible to determine which restriction is the cause.

[0097] In step 414, after outputting the model representing the executable region, the Figure 5 processing shown ends.

[0098] As described above, according to the present embodiment, since discrete observed values obtained by simulation are filled in and continuous predicted values are output, in simulation, it is not necessary to derive a large number of observed values. As a result, the computational cost in simulation can be suppressed. In addition, Gaussian process regression can calculate prediction errors in addition to predicted values, and calculates calculation points for searching for a multi-performance executable region in sequence based on the calculated predicted values and prediction errors.

[0099] By using the hybrid acquisition function a PoF-ES (x) that can search the boundary and the inside of the executable region simultaneously, calculation points can be calculated efficiently, and the obtained calculation points are added to the teacher data of machine learning. Then, based on the updated teacher data, the probability distribution of each performance executable at each calculation point is calculated.

[0100] By using the teacher data obtained through the efficient operation achieved by the hybrid acquisition function a PoF-ES (x) shown in the present embodiment, the computational cost can be significantly reduced compared to methods such as random sampling.

[0101] In addition, in the present embodiment, by expressing the executable region in the form of a probability distribution, it is possible to express it as a continuous value with the achievability of each performance as a scale. Conventionally, since the executable region has been processed by binary representation of whether it holds or not, it has been difficult to obtain design pointers when a valid solution has not been obtained. In the present embodiment, for example, design pointers such as additional calculation of regions with high probability and reduction of calculation cost of regions with low probability can be easily obtained.

[0102] In addition, since a large number of points of valid solutions have been collected to infer the executable region (surface), a large number of calculation points are required to predict the boundary forming the surface. However, in the present embodiment, since the executable region is directly modeled, the executable region can be directly inferred.

[0103] Moreover, in the present embodiment, even when a new performance limit occurs, the multi-performance executable region can be updated by multiplying the probability distribution related to the new performance limit by the probability distribution representing the current multi-performance executable region.

[0104] In addition, "acquisition function a PoF (x)" corresponds to the "first acquisition function", "acquisition function a ES (x)" corresponds to the "second acquisition function", and "ε" corresponds to the "predetermined threshold".

Claims

1. A multi-performance optimization design device, comprising: A memory; A processor, which is connected to the memory, The processor is configured to, Generate a continuous function using Gaussian process regression for each of the discrete observed values obtained by simulation and for each of multiple performances with respect to the position of the structure, so as to output a continuous predicted value and a prediction error for each of the multiple performances, Based on the predicted value and the prediction error, calculate multiple calculation points for searching an executable region for each of the multiple performances, Set the executable region as a region where multiple types of performances, namely strength, rigidity, weight reduction, and vibration suppression, which are different from each other and may be opposite to each other depending on the situation, are respectively optimized, Using the calculated multiple calculation points, calculate a probability distribution for each of the multiple performances that can execute each of the multiple performances, Output the total product of the probability distributions calculated for each of the multiple performances as a multi-performance executable region that satisfies the constraints of all performances, Set the point where the product of a first acquisition function that is the feasibility probability regarding the search inside the executable region and a second acquisition function that is regarding the search near the upper boundary between the executable region and the first non-executable region and near the lower boundary between the executable region and the second non-executable region is maximized as the calculation point. The first non-executable region exists outside the executable region across the upper boundary, and the second non-executable region exists inside the executable region across the lower boundary.

2. The multi-performance optimization design device according to claim 1, wherein, The processor is configured to, In the case where new performance constraints are generated, Fill in the discrete observed values obtained by simulation and for the new performance, so as to output a continuous predicted value and a prediction error for the new performance, Calculate multiple calculation points for searching an executable region for the new performance based on the predicted value and the prediction error of the new performance, Calculate a probability distribution for the new performance that can execute the new performance using the multiple calculation points for searching an executable region for the new performance, Output the product of the total product of the probability distributions calculated for each of the multiple performances and the probability distribution where the new performance holds as a new multi-performance executable region.

3. The multi-performance optimization design device according to claim 1 or 2, wherein, The processor is configured to, In the case where the ratio of the region where the calculation point is not calculated to the entire region involved in the design is less than a predetermined threshold, end the calculation of the calculation point.

4. A multi-performance optimization design method, which is executed by a processor as follows, Generate a continuous function using Gaussian process regression for each of the discrete observations obtained through simulation for each of multiple performances with respect to the position relative to the structure, thereby outputting a continuous predicted value and a prediction error for each of the multiple performances. Based on the predicted value and the prediction error, calculate multiple calculation points for searching an executable region capable of executing each of the multiple performances for each of the multiple performances. Set the executable region as a region where multiple types of performances, such as strength, rigidity, weight reduction, and vibration suppression, which are different from each other and may be opposite to each other depending on the situation, are respectively optimized. Using the calculated multiple calculation points, calculate a probability distribution capable of executing each of the multiple performances for each of the multiple performances. Output the total product of the probability distributions calculated for each of the multiple performances as a multi-performance executable region that satisfies the limitations of all performances. Set the point at which the product of a first acquisition function, which is a feasibility probability related to the search inside the executable region, and a second acquisition function, which is related to the search near the upper boundary between the executable region and the first non-executable region and near the lower boundary between the executable region and the second non-executable region, is maximized as the calculation point. The first non-executable region exists outside the executable region across the upper boundary, and the second non-executable region exists inside the executable region across the lower boundary.

5. The multi-performance optimization design method according to claim 4, wherein When a new performance limitation is generated, Interpolate the discrete observations obtained through simulation for the new performance, thereby outputting a continuous predicted value and a prediction error for the new performance. Calculate multiple calculation points for searching a region capable of executing the new performance based on the predicted value and the prediction error of the new performance. Calculate a probability distribution capable of executing the new performance using the multiple calculation points for searching a region capable of executing the new performance. Output the product of the total product of the probability distributions calculated for each of the multiple performances and the probability distribution where the new performance holds as a new multi-performance executable region.

6. The multi-performance optimization design method according to claim 4 or 5, wherein When the ratio of the region where the calculation point is not calculated to the entire region involved in the design is less than a predetermined threshold, end the calculation of the calculation point.

7. A computer-readable recording medium that records a program for causing a computer to execute a multi-performance optimization design process, the multi-performance optimization design process including the following processes, namely, Generate a continuous function using Gaussian process regression for each of the discrete observations obtained through simulation for each of multiple performances with respect to the position relative to the structure, thereby outputting a continuous predicted value and a prediction error for each of the multiple performances. Based on the predicted value and the prediction error, a plurality of calculation points for searching an executable region capable of executing each of the plurality of performances are calculated for each of the plurality of performances. The executable region is set as a region in which performances that are different in a plurality of categories including strength, rigidity, weight reduction, and vibration suppression and are opposite to each other according to circumstances are respectively optimized. Using the calculated plurality of calculation points, a probability distribution capable of executing each of the plurality of performances is calculated for each of the plurality of performances. The total product of the probability distributions calculated for each of the plurality of performances is output as a multi-performance executable region that satisfies the limitations of all performances. The point at which the product of a first acquisition function that is a feasibility probability related to the search inside the executable region and a second acquisition function that is related to the search near the upper boundary between the executable region and the first non-executable region and near the lower boundary between the executable region and the second non-executable region becomes the maximum is set as the calculation point. The first non-executable region exists outside the executable region across the upper boundary, and the second non-executable region exists inside the executable region across the lower boundary.

8. The computer-readable recording medium according to claim 7, wherein the multi-performance optimization design process includes the following process, that is, when a new performance limitation is generated, the discrete observed values obtained by simulation for the new performance are filled to output a continuous predicted value and a prediction error for the new performance, based on the predicted value and the prediction error of the new performance, a plurality of calculation points for searching an executable region capable of executing the new performance are calculated, using the plurality of calculation points for searching an executable region capable of executing the new performance, a probability distribution capable of executing the new performance is calculated, the product of the total product of the probability distributions calculated for each of the plurality of performances and the probability distribution in which the new performance holds is output as a new multi-performance executable region.

9. The computer-readable recording medium according to claim 7 or 8, wherein the multi-performance optimization design process includes the following process, that is, when the ratio of the region where the calculation point is not calculated to the entire region related to the design is less than a predetermined threshold, the calculation of the calculation point is ended.

Images