JP2007087098A - Optimization system, optimization method, optimization program and program medium - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an optimization system, capable of acquiring both a design space and the physical meaning of a relation between optimization design value and response model. <P>SOLUTION: The optimization system comprises a response model having responses of evaluation characteristic values to a plurality of combinations of design parameters, an arithmetic processing unit computing optimum values of the design parameters based on the evaluation characteristic values; and a display device displaying the arithmetic result computed by the arithmetic processing unit. The arithmetic processing unit includes a classification means classifying the plurality of combinations of design parameters according to similarity, a characteristic contour map forming means arranging the classified combinations on a plane and recording evaluation characteristic values corresponding to the combinations in this position, and a parameter contour map generation means arranging the classified combinations on a plane and recording the value of design parameter of one of the combinations in this position. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to an optimization technique for optimizing design parameters that achieve desired characteristics in a response model in which characteristics are determined based on a plurality of design parameters.

  Conventionally, Non-Patent Document 1 has been disclosed as a technique using a response phase model as an optimization method. In this system, structural analysis is performed based on design parameters created using an orthogonal table or random numbers, and the estimation formula is calculated by least square approximation for the number of characteristics to be evaluated from the relationship between the design parameters and the obtained response model. create. Next, design constraint conditions are set, and a design parameter that maximizes or minimizes all estimation formulas created under the constraint conditions is obtained. The design parameter obtained here is the optimum value to be obtained.

In this way, the relationship between the design parameter and the response model is estimated, and only the estimation formula needs to be handled until the optimum design value is obtained. Therefore, the input box (design parameter) and the output (optimum design value) are black boxed ( Mathematical processing in the system: solving simultaneous equations of estimation equations). Therefore, the merits of the optimization method using the response phase method can be enumerated as follows.
(1) It is possible to create an estimation formula from the relationship between a large number of design parameters and the response model, and the optimization calculation is efficient because only the estimation formula needs to be handled.
(2) Optimal design values can be obtained without knowing the design.
(3) Optimal design values can be obtained without knowledge of optimization methods using response phase methods.

  Non-Patent Document 2 is disclosed as a technique using a genetic algorithm as an optimization method. In this method, a structural analysis is performed based on a design parameter created using an orthogonal table or a random number, and a direction in which a maximum value or a minimum value (that is, an optimum value) is likely to exist is searched from the response model. Next, the structural analysis is repeated until the optimum value is obtained while changing the design parameters so that the response appears in the direction in which the optimum value exists. Here, a genetic algorithm is used to search for a direction in which an optimum value is likely to exist and to create a new design parameter.

Since changes in design parameters are made discretely using a genetic algorithm, the input parameters (design parameters) and outputs (optimal value design) are black boxed (discrete numerical processing in the system: design parameters based on the genetic algorithm) Can be discretized). Therefore, the merit of the optimization method using the genetic algorithm can be enumerated as follows.
(4) If it is left to the optimization system using a genetic algorithm, the optimum design value can be obtained.
(5) Optimal design values can be obtained without knowing the design.
(6) Optimal design values can be obtained without knowing an optimization system using a genetic algorithm.
Transaction of JSCES, Paper No. 20000019, Japan Society for Computational Engineering (issued May 24, 2000) Annual Meeting 2001 of the Japan Opportunity Association (2001.27.27-30)

  However, the above prior art has the following problems. That is, in any of the techniques described in Non-Patent Document 1 and Non-Patent Document 2, the validity of the accuracy of the estimation formula is unknown, and if the multi-modality is strong, the truly optimal design value I do not know what is required. Further, since the optimization calculation is black boxed (mathematical numerical processing), the optimal physical meaning (design meaning) is not known.

  The present invention has been made paying attention to the above-mentioned problems, and the object of the present invention is an optimization system capable of grasping both the design space and the physical meaning of the relationship between the optimized design value and the response model, the optimum An object is to provide an optimization method, an optimization program, and a program medium.

  In order to achieve the above object, the optimization system of the present invention calculates an optimum value of the design parameter based on a response model having a response of an evaluation characteristic value to a combination of a plurality of design parameters and the evaluation characteristic value. In an optimization system including an arithmetic processing device and a display device that displays an arithmetic result calculated by the arithmetic processing device, the arithmetic processing device classifies the combination of the plurality of design parameters according to the degree of similarity. Classifying means for arranging the classified combinations on a plane, characteristic contour map creating means for recording the evaluation characteristic values corresponding to the combinations of the positions, and the classified combinations arranged on a plane. Parameter contour map creating means for recording the value of one design parameter of the combination of the positions; Characterized in that it has a. Note that the response model refers to a system that changes an input based on some rule and outputs the input, and includes not only a structure but also an electric circuit, an economic model, and the like.

  Therefore, the state of the design space can be grasped by the characteristic contour map. Further, by displaying the characteristic contour map and the parameter contour map, it is possible to grasp the correlation between the evaluation characteristics and the design parameters, and thus the physical meaning.

  Hereinafter, the best mode for realizing the optimization system of the present invention will be described based on embodiments shown in the drawings.

  FIG. 1 is a system diagram illustrating an optimization system configuration according to the first embodiment. Connected to the arithmetic processing device 1 are a display device 2 (corresponding to display means) for displaying the arithmetic results of the arithmetic processing device 1 and a keyboard 3 for inputting information to the arithmetic processing device 1.

[Concept of optimization system of embodiment 1]
FIG. 2 is a flowchart showing an arithmetic processing structure in the arithmetic processing device 1. The optimization system according to the first embodiment is roughly composed of four components (A) to (D).

  The component (A) is a component that generates a design parameter using an orthogonal table or a random number and performs a structural analysis based on the generated design parameter. In this component, a response model model to be subjected to structural analysis is incorporated. This response model model is equipped with a simulator capable of outputting evaluation characteristics based on input design variables.

  Component (B) is a contour map group (corresponding to storage means) of at least one evaluation characteristic, and a characteristic contour map of the response model (corresponding to evaluation characteristic value) obtained as a result of structural analysis, and corresponding to it A pair of parameter contour maps of design parameters to be present exists for each evaluation characteristic (corresponding to characteristic contour map creating means and parameter contour map creating means).

  All response models (evaluation characteristic values) are classified by clustering (corresponding to the classification means), and as shown in FIG. 3, the classified group locations are displayed so as to be the same in all contour maps. The contour map does not have the concept of the vertical and horizontal axes, and is simply arranged based on the proximity (similarity) of the clusters. The black line in FIG. 3 is a cluster boundary and is common to all contour maps. Note that clustering is a known technique in which similar objects are grouped to form a cluster.

  In other words, a layout map in which a plurality of design parameters or evaluation characteristic values are arranged on a plane in accordance with the degree of similarity is prepared in advance, and the characteristic contour map is a record of the evaluation characteristics of the response model on this layout chart. A parameter contour map is obtained by recording design parameter values on the layout drawing.

  Component (C) is a component that superimposes the contour map of each evaluation characteristic created in component (B) (corresponds to the superimposing means) and searches the contour map for an area where the optimum value is likely to exist (corresponding to the extracting means). is there. Note that it is possible to search for a region that the user (designer) wants to know, not the region whose objective is the optimum value. The evaluation characteristics to be overlaid can be freely selected by the user (designer). In addition, by assigning a weight to a certain evaluation characteristic at the time of superimposition, it is possible to perform an optimum region search that places importance on the weighted evaluation characteristic.

  Component (D) is an optimization component that uses the conventional optimization method to obtain the optimum calculated value in the area where the optimum value obtained in Component (C) is likely to exist (the area that the user wants to know). .

  The components (A) and (D) are configured in the same way as the conventional optimization method, but components (B) and (C) are added to solve the problems of the conventional method. This is a feature of the first embodiment. In addition, each said component of Example 1 is integrated in the arithmetic processing unit 1 as an optimization program.

Here, the problems of the prior art will be described in detail. As described in the background art, the optimization system using the response phase method (see Non-Patent Document 1) has the following problems although there are merits (1) to (3).
(a) Since the accuracy of the estimation formula depends on the number of relations between the design parameters and the response model, a lot of structural analysis is required to improve the accuracy. Also, it is not known in advance how many structural analyzes are appropriate.
(b) It is not known whether the optimum design value is truly obtained by the braking of the estimation equation.
(c) Optimization calculation is black-boxed (mathematical numerical processing), but the effects shown in (2) and (3) above are obtained, but the optimal physical meaning (design meaning) is not known. .

In addition, the technique described in Non-Patent Document 2 has the following problems although there are merits (4) to (5).
(d) In the case of a phenomenon that exhibits very multimodality (strong non-linearity), a very large amount of structural analysis is required to obtain the optimum design value, which is inefficient.
(e) If the multimodality is very strong, there is a possibility of falling into LocalMinimum (or LocalMaximum), and there is a possibility of finding an answer that is not truly the optimum design value, so it is not known whether it is really optimum. This is because there is no means for confirming the entire design space. Note that LocalMinumum or LocalMaximum represents a local minimum value or maximum value, and is different from the true minimum value or maximum value.
(f) Since optimization calculation is black boxed (discrete numerical processing), the effects shown in (5) and (6) above are obtained, but the optimal physical meaning (design meaning) is not known. Absent.

That is, when the above problems are arranged, the following problems can be cited.
-It requires a lot of calculations to improve accuracy and is inefficient.
・ I can't figure out the entire design space, so I don't know if it's really optimal.
-Since the optimization process is black-boxed (mathematical and discrete), the physical meaning of the relationship between the optimal design value and the response model is unknown.

  Therefore, in the first embodiment, the entire design space is grasped using the characteristic contour map and the parameter contour map, and the physical meaning of the relationship between the optimum design value (or simply the design value) and the response model is found. .

[Application to actual design problems]
Next, the case where it applies to an actual design problem based on the concept of the said Example 1 is demonstrated.

(Determination of optimal spring-S / ABS specifications for sedan vehicles in light bottoming)
Applying the above concept to the optimization process when determining the optimal spring-S / ABS specification for a sedan vehicle in light bottoming driving will be explained. Note that light bottoming means that the vehicle nose-dives during braking, and “optimum spring-S / ABS specification” is a combination of the spring constant of the coil spring of the suspension and the damping characteristics of the shock absorber. Represents a decision. As a response model, a sedan vehicle model with a suspension was constructed, and each design variable was set.

The evaluation characteristics shown in Table 1 were taken up. The spring-S / ABS specifications to be determined are shown in Table 2. Therefore, the optimum spring-S / ABS specification to be obtained is the derivation of each parameter value shown in Table 2 that is the minimum after simultaneously satisfying all the evaluation characteristics of Table 1.

  FIG. 4 is a characteristic diagram showing the characteristics of the suspension. The horizontal axis represents the relationship between the piston speed and the vertical axis represents the damping force. The evaluation characteristic “Blow-by” used in Table 2 represents the piston speed when the slope of this relationship changes (point B). The “valve characteristic gradient” represents the gradient of the area expressed by a linear function (point C).

・ Component (A)
Design parameters are set between the upper and lower limits of each design parameter shown in Table 2 using random numbers. Here, a set of 100 design parameters was created. An analysis model is created based on the design parameters created here, light bottoming travel is calculated using a simulator, and a response model shown in Table 1 is obtained. Therefore, 100 response models are obtained for a set of 100 design parameters.

・ Component (B)
The 100 response models obtained by component (A) are classified using clustering based on the design parameters and displayed on the contour map. Here, six clusters (classified groups) were created. The result is shown in FIG. The black line in FIG. 5 is a cluster boundary and is common to all contour maps. As shown in FIG. 5, the contour map of the evaluation characteristics is displayed independently.

  Further, the contour map of each design parameter in the lower two columns in FIG. 5 is the same for each evaluation characteristic in the case of this problem. The contour map dots indicate that the smaller the density, the smaller the value, and the higher the density, the larger the value. The contour map has no concept of vertical and horizontal axes, and is simply arranged based on the proximity (similarity) of clusters.

・ Component (C)
The contour maps of the six evaluation characteristics created by the component (B) are overlapped, and an area where the optimum value is likely to exist is searched from the contour map. The result of overlaying the contour maps of the six evaluation characteristics is shown in FIG. From the result of FIG. 6, it can be seen that the optimum value, that is, the part where all are overlapped and becomes the smallest (the part where the density is low) is a region surrounded by a circle in FIG. It can be seen that the design parameter patterns existing in the circled area tend to have a certain width as shown in FIG.

  Therefore, in the next component (D), it is sufficient to create a model that falls within this range and perform optimization calculation. In this case, the region is determined by directly confirming the overlaid contour map. However, as shown in FIG. 8, the threshold value is set and the threshold value is increased or decreased to supplement the region where the optimum value is likely to exist. It is also possible to add.

・ Component (D)
In order to obtain the optimum design value in the region 9 where the optimum value is likely to exist (the user wants to know the region 9) obtained by the component (C), optimization using a conventional optimization method is performed. Here, the component (C A model that fits within the width of the design pattern obtained in step (1) is created and optimized, and the contour map of the optimum evaluation characteristics obtained in this way is shown in Fig. 9. The optimum design value at this time is The pattern shown by the thick solid line in FIG.

[Effects based on the above components]
Next, operations derived from the above components will be described.

(Function of contour map)
FIG. 11 shows a contour map of the evaluation characteristics of the front wheel installation load. The contour map of the load characteristic of the front wheel installation represents the entire design area as a contour diagram with quantitative values of evaluation characteristic values. Therefore, the user (designer) can confirm the magnitude of the evaluation characteristic value by viewing FIG. The optimum value of the evaluation characteristic is the minimum value or the maximum value in the contour map. In FIG. Since the contour map is in the above-described component (B) stage, a true optimum value is not always obtained. Therefore, in this stage, it is an “area where an optimum value is likely to exist”.

  Further, as shown in FIG. 3, all response models (evaluation characteristic values) are classified by clustering, and the locations of the classified groups are displayed so as to be the same in all contour maps. For example, as shown in FIG. 12, in order to reduce the front wheel installation load in the direction of the arrow, the design parameter may be changed in the same direction of the arrow. For example, “Fr is large → small on the blow-by extension side, Rr is small → large”, “Fr is small → large on the blow-by contraction side, and Rr remains large”, “Fr is large → small on the valve characteristic tilt expansion side From the contour map, “Rr is small → large”, “Fr is small → large, and Rr remains large on the valve characteristic tilt reduction side”.

  Since the contour map can also be grasped by a quantitative value, if the target evaluation characteristic value is set on the evaluation characteristic contour map, the value on the parameter contour map corresponding to the same place may be read.

(Overlapping effect)
As described in FIG. 6, by overlapping the contour maps of independent evaluation characteristics, the place where the value is small (density is low) is small (density is low), and the value is large (density is low). Places that are dark are large (dense), places that are not common become average, and places that are common are emphasized. Therefore, it can be seen that the region where the optimum value (minimum value or maximum value) is likely to exist is a place where the density of the contour map as a result of superimposition is low or a place where the density is high.

  For example, even when it is necessary to consider a trade-off relationship in which one characteristic is good and the other characteristic is deteriorated, there is a case where it is desired to consider a certain evaluation characteristic with particular emphasis. At this time, by assigning a weight to the value of the evaluation characteristic to be regarded as important, the density of the contour map becomes clearer, and the evaluation characteristic regarded as important when superimposed can be emphasized. In addition, what is necessary is just to judge the area | region where the optimal value is likely to exist from light and shade.

(Extraction of areas where optimal values are likely to exist)
It can be seen that the contour maps of the respective evaluation characteristics are overlapped to extract a region where an optimum value is likely to exist, and the design parameter pattern satisfying this region tends to have a certain width as shown in FIG. For example, in the case of the above spring-S / ABS specifications, the user (designer) knows what the current specifications tend to be, what the other vehicle's specifications tend to be, In many cases, the specifications of the developed vehicle are determined. In the optimization system as shown in the background art, the entire region could not be shown. Therefore, in such a case, an analysis model was created from the beginning and calculation was performed.

  On the other hand, in the optimization system of the first embodiment, as shown in FIG. 13, the entire area of the evaluation characteristics and the entire area of the design parameter are indicated by a contour map, and each contour map is all on the same design parameter pattern. Therefore, it is possible to instantaneously obtain design parameter patterns and trends in an area where an optimum value is likely to exist.

(Operation by comparison of contour maps)
FIG. 14 is a diagram showing the effect of comparison of contour maps. As shown in FIG. 14, when the contour line of each evaluation characteristic rising to the right is confirmed, in the case of the light bottoming shown in the first embodiment, the vertical displacement of the center of gravity increases to the right (high density), and the front wheels Installation load increases to the left (high density). That is, these relationships are clearly in a trade-off relationship. On the other hand, other evaluation characteristics do not change so much on the solid line. In this way, the tendency of each evaluation characteristic can be recognized by comparing the contour maps (corresponding to comparison means).

(Operation by comparison of characteristic contour map and parameter contour map)
FIG. 15 is a diagram showing a design pattern in each region classified in the characteristic contour map. As described above, in FIG. 15, the black line is a cluster boundary (a set of objects having similarities), and is common to all contour maps. Therefore, a design pattern having a maximum width and a minimum width among the design patterns existing in the black line area is displayed.

  In other words, each cluster is a set of similar design patterns, and a wide design pattern is a display of features of each cluster by a combination of design patterns.

  Using this relationship, for example, as shown in FIG. 16, when the characteristic is changed in the direction of the arrow on the characteristic contour map, the design pattern is (pattern 1) → (pattern 2) → (pattern 3). It can be understood that changes.

  FIG. 17 is a diagram in which design pattern changes when the above characteristics are changed are superimposed. As shown in FIG. 17, when the design patterns are overlapped and the changes in the design patterns are observed, the “blow-by extension side” and “blow-by contraction side” on the front side and rear side change from bottom to top. You can see how they are doing. That is, when the evaluation characteristics are desired to be changed by the arrows shown in FIG. 16, it can be realized by changing the “blow-by extension side” and the “blow-by reduction side” on the front side and rear side from the bottom to the top.

(Other examples)
In the first embodiment, all the characteristic contour maps are overlapped, and an area estimated to be the optimum value is extracted from the overlapped map, and the optimization process is performed. Alternatively, only the evaluation characteristics may be selected, and a design pattern in which only the evaluation characteristics selected by overlapping them are optimal may be extracted.

  In the first embodiment, the design pattern corresponding to the characteristic contour map is extracted. However, the characteristic value corresponding to the parameter contour map may be extracted, and the change of the characteristic value due to the change of the design pattern may be seen.

  Further, in the first embodiment, the specification determination of the sedan vehicle is shown as an example, but similar characteristic contour maps and parameter contour maps are created in other vehicle types (for example, one-box vehicles), and these contour maps are superimposed. Thus, it is possible to obtain a specification corresponding to the middle between the sedan vehicle and other vehicle types without calculating from the beginning.

  In the first embodiment, an example in which an optimization program is incorporated in the arithmetic processing unit 1 has been described. However, the optimization program may be distributed in the market using a program medium in which the optimization program is stored. Alternatively, the program may be stored in a server or the like, and the user may install the arithmetic processing apparatus 1 as appropriate by downloading or the like.

  In the first embodiment, the response model is applied to a system including a suspension coil spring and a shock absorber. However, the response model is not limited to this, and may be applied to an electric circuit, an economic model, a program, and the like. .

1 is a system diagram illustrating an optimization system configuration of Embodiment 1. FIG. 3 is a flowchart illustrating an arithmetic processing structure in the arithmetic processing apparatus according to the first embodiment. It is a figure showing the contour map of Example 1. FIG. 6 is a characteristic diagram illustrating characteristics of the suspension of Example 1. FIG. It is a figure showing the specific example of the contour map of Example 1. FIG. It is a figure showing the superimposition of the contour map of Example 1. FIG. It is a figure showing the relationship between the characteristic contour map of Example 1, and a design pattern. It is a figure showing the process which supplements the area | region where the optimal value of Example 1 seems to exist. FIG. 6 is a diagram showing an area of optimum evaluation characteristics on the superimposed contour map of Example 1. It is a figure showing the relationship between the design pattern of Example 1, and an optimal design value. The contour map of the evaluation characteristic of the front wheel installation load of Example 1 is shown. It is a figure showing the relationship between the front wheel installation load of Example 1, and a design parameter. It is a figure showing that the whole area | region of the evaluation characteristic of Example 1, and the whole area | region of a design parameter are shown by a contour map, and each contour map is displayed on the same design parameter pattern altogether. It is a figure showing the effect | action by the comparison of the contour map of Example 1. FIG. It is a figure showing the design pattern in each area | region classified within the characteristic contour map of Example 1. FIG. It is a figure showing the change on the characteristic contour map of Example 1, and the change of a design pattern. It is a figure showing the change of the design pattern of Example 1. FIG.

Explanation of symbols

1 arithmetic processing unit 2 display unit 3 keyboard

Claims (10)

A response model having an evaluation characteristic value response to a combination of a plurality of design parameters;
An arithmetic processing device that calculates an optimum value of the design parameter based on the evaluation characteristic value;
In an optimization system with
The arithmetic processing unit includes:
Classification means for classifying the combination of the plurality of design parameters according to the degree of similarity;
A characteristic contour map creating means for arranging the classified combinations on a plane and recording the evaluation characteristic values corresponding to the combinations of the positions;
A parameter contour map creating means for arranging the classified combinations on a plane and recording a value of one design parameter of the combinations at the positions;
An optimization system characterized by comprising: The optimization system of claim 1,
The response model has a plurality of evaluation characteristic values,
The arithmetic processing unit includes an overlapping unit that creates the characteristic contour map for each of the plurality of types of evaluation characteristic values and superimposes the characteristic contour maps at positions where the similarities are the same. system. The optimization system according to claim 1 or 2,
The said arithmetic processing apparatus has the extraction means which extracts the area | region where the candidate of an optimal value exists from the said characteristic contour map, The optimization system characterized by the above-mentioned. The optimization system according to any one of claims 1 to 3,
The response model has multiple types of evaluation characteristics;
The said arithmetic processing apparatus has the comparison means which produces the said characteristic contour map for every said several types of evaluation characteristic, and compares each characteristic contour map, The optimization system characterized by the above-mentioned. The optimization system according to any one of claims 1 to 4,
The optimization processing system includes a parameter pattern display unit that displays a pattern of the design parameter corresponding to a position of the characteristic contour map. The optimization system according to any one of claims 1 to 5,
The optimization processing system includes a characteristic pattern display unit that displays a pattern of the evaluation characteristic according to a position of the characteristic contour map. The optimization system according to any one of claims 1 to 6,
The response model has multiple types of evaluation characteristics;
The optimization system includes an accumulation unit that creates the characteristic contour map for each of the plurality of types of evaluation characteristics and accumulates the parameter contour map independently for each of the characteristic contour maps. When there is a response model having an evaluation characteristic response to a combination of a plurality of design parameters, an optimization method for calculating an optimal value of the design parameter based on the evaluation characteristic,
Classifying the combination of the plurality of design parameters according to the similarity,
Arranging the classified combinations on a plane, creating a characteristic contour map that records the evaluation characteristic values corresponding to the combination of the positions,
An optimization method characterized by arranging the classified combinations on a plane and creating a parameter contour map in which a value of one design parameter of the combinations of the positions is recorded. When there is a response model having an evaluation characteristic response to a combination of a plurality of design parameters, an optimization program for calculating an optimum value of the design parameter based on the evaluation characteristic,
Classifying the combination of the plurality of design parameters according to the similarity, and placing the classified combination on a plane;
Displaying an evaluation characteristic of the response model corresponding to the combination of the positions at the arranged position;
Displaying the value of one design parameter of the combination of the positions in the arranged arrangement;
An optimization program characterized by comprising   A program medium in which the optimization program according to claim 9 is stored.

Images