JP2020064479A - Search device - Google Patents

Abstract

To provide a search device capable of efficiently searching an applicable range expressing a range of design variables satisfying various performance conditions and/or constrain conditions.SOLUTION: A search device calculates a first function for minimizing an evaluation index in which a violation amount violating prescribed conditions is taken into account, for determining whether or not an applicable range is present. The search device uses prescribed models (a kriging model and a Gaussian process), for calculating prediction values at calculation candidate points representing candidates at following calculation points within an important design space and a value expressing fluctuations in the prediction values. The search device uses a second function expressing a likelihood for whether or not the calculation candidate points are within the applicable range, for selecting a following calculation point from a collection of the calculation candidate points.SELECTED DRAWING: Figure 2

Description

  The present invention relates to a search device that searches for a satisfaction range that represents a range of variables (design variables) that satisfy a predetermined condition.

  2. Description of the Related Art Conventionally, there has been proposed a technique of searching for a value of a design variable satisfying various performance conditions and / or constraint conditions at the design stage of a structure such as a vehicle (for example, see Patent Document 1).

JP-A-11-120215

  In the initial design stage of the structure, in order to secure the degree of freedom of design in the detailed design stage, which is a later stage, find out the range of establishment that represents the range of design variables that satisfy various performance conditions and / or constraint conditions. Is important. However, the device described in Patent Document 1 (hereinafter, referred to as “conventional device”) is a device that searches for an optimal solution (value of design variable) at one point that satisfies the constraint condition. When the conventional device is used, even if the operator wants to change the design of the structure in the detailed design stage, it is not known whether the changed design satisfies various performance conditions and / or constraint conditions. There is.

  On the other hand, in recent years, development of simulation technology has progressed. Although it may be possible to search for the above-mentioned establishment range using a simulation technique, at present, there is no guideline for efficient search. Therefore, a huge amount of time and resources are required. Therefore, there is a demand for a technique for efficiently searching the established range with limited time and resources.

  The present invention has been made to solve the above problems, and one of the objects thereof is to efficiently search for a valid range representing a range of design variables satisfying various performance conditions and / or constraint conditions. It is to provide a search device capable of

A search device of the present invention (hereinafter, also referred to as “device of the present invention”) uses a plurality of design variables and at least one of the plurality of design variables as an input value, and according to the input value. At least one evaluation function that outputs an output value is defined, and when a predetermined condition to be satisfied with respect to the output value of the evaluation function is set, a satisfaction range representing a range of the design variable that satisfies the predetermined condition Is a device (10) for searching for.
The device of the present invention is
A first function that defines a violation amount (J _i (x)) that violates the predetermined condition in each of the evaluation functions and minimizes an evaluation index (J ₀ (x)) that takes all the violation amounts into consideration. First calculation means (11, step 210, step 215) for determining whether or not the established range exists by calculating (Equation (4));
When it is determined by the first calculating means that the established range exists (step 215: Yes), a key design space that is a space defined by two or three variables of the plurality of design variables is set. 2 calculation means (11, step 225),
A third function of calculating the first function (Equation (5)) for a search data set, which is a set of calculation points in the important design space, and determining whether the calculation points are within the established range. Calculation means (11, step 235),
Fourth calculation means for calculating a predicted value of the value of the first function of a calculation candidate point which is a candidate for the next calculation point in the important design space and a value representing variation of the predicted value using a predetermined model. (11, step 240), the value of the first function is not calculated in the predetermined model when the value of the first function at each calculation point included in the search data set is given. Fourth calculation means, which is a model (Kriging model, Gaussian process) capable of calculating the predicted value of the value of the first function at another calculation point that is and calculating the value representing the variation of the predicted value.
A second value that includes, as a variable, the predicted value of the value of the first function at the calculation candidate point and a value that represents the dispersion of the predicted value, and that represents the probability of whether the calculation candidate point falls within the established range. Using the function (Equation (6)), the process of selecting the next calculation point from the set of calculation candidate points and adding the next calculation point to the search data set is performed under a predetermined end condition ( Fifth calculation means (11, step 245, step 250, step 260) repeatedly executed until the expression (8) is satisfied,
Display means (19, step 255) for displaying information indicating the established range in the important design space;
Equipped with.

  According to the device of the present invention, the amount of violation that violates a predetermined condition is defined in each of the evaluation functions, and one evaluation index in consideration of all the amounts of violation is used. Therefore, the device of the present invention can easily determine whether or not the established range exists by using one evaluation index even when a plurality of performance conditions and / or constraint conditions exist. Further, an important design space defined by 2 or 3 variables of the plurality of design variables is set. Therefore, the device of the present invention can efficiently search the established range in the design space (important design space) in which the operator pays attention. In addition, the device of the present invention uses a predetermined model to calculate the predicted value of the value of the first function at another calculation point (calculation candidate point) in which the value of the first function has not been calculated in the important design space. Calculate and calculate a value that represents the variation in the predicted value. Then, the device of the present invention uses the second function that includes the predicted value and the value indicating the variation of the predicted value as variables and that represents the probability of whether or not the calculation candidate point is within the established range, and calculates the next calculation point. Select. In this way, the next calculation point is selected according to the probability of whether or not the calculation candidate point falls within the established range, so that the established range can be searched efficiently.

  In the above description, in order to help understanding of the invention, the reference numerals used in the embodiments are attached in parentheses to the configurations of the invention corresponding to the embodiments. It is not limited to the embodiment defined by.

It is a block diagram which shows the hardware constitutions of the search apparatus (henceforth "this embodiment apparatus") which concerns on embodiment of this invention. 6 is a flowchart showing a “search routine” executed by the CPU of the present embodiment. It is a figure which shows an example of the data set for a search input in step 230 of the routine of FIG. It is a figure which shows an example of the figure displayed on a display screen in step 255 of the routine of FIG. It is a figure explaining a suspension beam model and a design variable defined in the suspension beam model. 6 is a diagram showing an example of a diagram displayed on a display screen in step 255 when the routine of FIG. 2 is executed for the suspension beam model of FIG. 7 is a diagram showing another example of the diagram displayed on the display screen in step 255 when the routine of FIG. 2 is executed for the suspension beam model of FIG. 5.

<Structure of the present embodiment>
As shown in FIG. 1, the search device 10 includes a calculation unit 11, an input unit 12, and an output unit 13. The arithmetic unit 11 includes a CPU 14, a RAM 15, a ROM 16, a hard disk (HDD) 17, and an I / O interface 18.

  The ROM 16 and / or the HDD 17 previously stores various data (data of “design variables, evaluation functions, mathematical formulas, and the like” to be described later) necessary for searching the established range. Various data may be input to the search device 10 via a network. Further, the ROM 16 stores instructions (programs, routines) executed by the CPU 14. The CPU 14 realizes various functions described below by executing the instruction.

  The arithmetic unit 11 is connected to the input unit 12 and the output unit 13 via the I / O interface 18. The input unit 12 is a device that receives various requests from workers, and includes a keyboard and a mouse. Therefore, the operator can use the input unit 12 to input various information when executing the routine of FIG. 2 described later. The output unit 13 has a display screen 19 which is visible to the operator.

<Outline of search processing in the present embodiment>
First, the outline of the operation of the present embodiment will be described. A plurality of design variables and a plurality of evaluation functions are defined in the design stage of a target structure (for example, a vehicle). At least one evaluation function may be defined.

  The evaluation function takes at least one of the plurality of design variables as an input value and outputs an output value (response value) according to the input value. In the evaluation function, a predetermined condition to be satisfied regarding the output value is set. The predetermined condition is a condition corresponding to the performance condition or constraint condition of the target structure (vehicle). For example, in each of the evaluation functions, the upper limit value and / or the lower limit value is set as the predetermined condition.

  In such a case, it is difficult to study on a desk the range of design variables (that is, the established range) in which a predetermined condition is satisfied in each of all evaluation functions. On the other hand, the present embodiment defines a violation amount that violates a predetermined condition in each of the evaluation functions, and uses one evaluation index in consideration of all the violation amounts, so that the predetermined condition is satisfied in each of the evaluation functions. It can be easily determined whether or not is satisfied.

  Here, if the above evaluation index is calculated for all design variables, the amount of calculation becomes enormous and it is not appropriate from the viewpoint of efficiency. Therefore, in the present embodiment, a design space (hereinafter referred to as “important design space”) that the operator pays attention to is set. The important design space is a space defined by 2 or 3 variables of the plurality of design variables. The present embodied system searches for an established range in the important design space.

  Further, the present embodied system selects the next calculation point by using the active learning method when searching for the established range in the important design space. In addition, in order to clarify the established range with the minimum amount of calculation, the present embodiment selects the following from a set of predetermined calculation point candidates (hereinafter referred to as “calculation candidate points”). Select the calculation point of. With the above-described configuration, the device of the present embodiment can efficiently search for the established range.

<Specific contents of search processing>
Next, the specific content of the search process in the present embodiment will be described.

を設計変数とする。更に、

を構造物（車両）の性能を評価する評価関数とする。 1. Definition of Evaluation Index First, an evaluation index for evaluating the establishment of a plurality of evaluation functions will be described.

Is a design variable. Furthermore,

Is an evaluation function for evaluating the performance of the structure (vehicle).

ここで、

は、評価関数ｆ_ｉの出力値に対する下限値を表す。

は、評価関数ｆ_ｉの出力値に対する上限値を表す。

は、正規化のために使用される所定の値である。 For each evaluation function f _i , constraints (conditions) on the upper limit value and the lower limit value are set for the output value. When a certain design variable x is given, the violation amount that violates the upper limit value and the lower limit value of the evaluation function f _i is defined by the following expression (1).

here,

Represents a lower limit value for the output value of the evaluation function f _i .

Represents an upper limit value for the output value of the evaluation function f _i .

Is a predetermined value used for normalization.

If offense against J _{i (x)} is a negative value, the design variables x, the evaluation will satisfy the constraints of the function f _i (i.e., while the output value of the evaluation function f _i is the upper limit value and the lower limit value It is in.). We define an index for judging constraints whether satisfies the (upper and lower limit values) for all of the evaluation function f _i by the following equation (2) when one design variable x is given.

Hereinafter, J ₀ (x) is referred to as “evaluation index”. If the evaluation index J _{0 (x)} is a negative value, the design variables x will satisfy the constraints of all the evaluation function f _i (upper and lower limit values).

ここで、

は、違反量Ｊ_ｉ（ｘ）が寄与する度合いをコントロールするパラメータであり、正の定数である。 Note that the max operation of Expression (2) may cause discontinuity of the function. In order to reduce such a discontinuity problem, the following expression (3) may be used as an evaluation index.

here,

Is a parameter that controls the degree of contribution of the violation amount J _i (x), and is a positive constant.

ここで、

は、設計変数と対象となる評価関数との間の関係を記述する不等式制約であり、

は、設計変数と対象となる評価関数との間の関係を記述する等式制約である。
以降において、評価指標Ｊ_０（ｘ）を最小化する（４）式の関数を「第１関数」と称呼する場合がある。 2. Optimization Problem of Evaluation Index Next, consider an optimization problem that minimizes the evaluation index J ₀ (x). This optimization problem is defined by the following equation (4).

here,

Is an inequality constraint that describes the relationship between the design variables and the target evaluation function,

Is an equality constraint that describes the relationship between the design variables and the evaluation function of interest.
Hereinafter, the function of Expression (4) that minimizes the evaluation index J ₀ (x) may be referred to as a “first function”.

By calculating the first function of Expression (4), it is possible to determine whether or not the constraints (upper limit value and lower limit value) of all evaluation functions f _i are satisfied when a certain design variable x is given. . As described above, when the evaluation index J ₀ (x) is a negative value, the design variable x given at that time satisfies the constraints (upper limit value and lower limit value) of all evaluation functions f _i , that is, It means within the range. Therefore, by calculating the first function of the equation (4), it is possible to determine whether the established range exists.

ここで、

は、重要設計空間として選択された設計変数のインデックスベクトルである。

は、重要設計空間の関心点として設定された変数ｘ_ｋの定数ベクトルである。なお、（５）式のｙ（ａ）＝０は、成立範囲の境界を表す。 Now, consider computing the first function of equation (4) for all design variables. In this case, the amount of calculation becomes huge, which is not appropriate from the viewpoint of efficiency. Therefore, as described above, the design space (important design space) that the worker pays attention to is set, and the established range is searched in the important design space. An optimization problem that minimizes the evaluation index J ₀ (x) in such an important design space is defined by the following equation (5).

here,

Is the index vector of the design variable selected as the important design space.

Is a constant vector of a variable x _k set as an interest point in the important design space. Note that y (a) = 0 in the equation (5) represents the boundary of the established range.

3. Selection of the Next Calculation Point Using the Active Learning Method Next, a method of automatically selecting the minimum calculation points necessary for clarifying the established range using the active learning method will be described.

  The present embodiment uses a kriging model to predict the distribution of y (a) in equation (5). By using the Kriging model, when the calculation point where the value of y (a) of the equation (5) is known in the important design space is given, the value of y (a) of the equation (5) is not calculated. It is possible to calculate the predicted value of the value of y (a) at another calculation point and also calculate the uncertainty (variation) of the predicted value. Since the Kriging model is a known model, detailed description is omitted.

Further, the present embodiment selects the next calculation point by using the active learning method based on the EI (Expected Improvement) function. The EI function is known and is defined by the following equation (6). The EI function may be referred to as a "second function".

は、正規累積分布関数（normal cumulative distribution function）であり、

は、正規確率密度関数（normal probability density function）である。
ｙ_ｐｒｅｄ（ａ）は、クリギングモデルにより求められた「（５）式のｙ（ａ）の予測値」である。更に、クリギングモデルを用いて、ｙ（ａ）の予測値の平均二乗誤差（ＭＳＥ：Mean Squared Error）ｓ^２（ａ）を求めることができる。従って、（６）式のｓ（ａ）は、ｙ（ａ）の予測値のばらつきを表す値であり、上記のｓ^２（ａ）から求めることができる。 here,

Is the normal cumulative distribution function,

Is a normal probability density function.
y _pred (a) is the “predicted value of y (a) in the equation (5)” obtained by the Kriging model. Further, using the Kriging model, the mean squared error (MSE: Mean Squared Error) s ² (a) of the predicted value of y (a) can be obtained. Therefore, s (a) in the equation (6) is a value representing the variation in the predicted value of y (a), and can be obtained from s ² (a) above.

の下において、計算点の候補（計算候補点）の集合の中から、「（６）式のＥＩ関数が最大となる点」を次の計算点として選択する。次の計算点を

とすると、以下の（７）式により次の計算点を求めることができる。

ここで、ｄ_０は、現在の計算点から最も近い計算点までの最小距離である。（７）式において、Ａは、既知の計算点の集合として、以下のように定義される。

ここで、

は、計算点の候補（計算候補点）の集合である。 This device has the following conditions

In the following, from the set of calculation point candidates (calculation candidate points), “the point at which the EI function of equation (6) is the maximum” is selected as the next calculation point. Next calculation point

Then, the following calculation point can be obtained by the following equation (7).

Here, d ₀ is the minimum distance from the current calculation point to the closest calculation point. In Expression (7), A is defined as follows as a set of known calculation points.

here,

Is a set of calculation point candidates (calculation candidate points).

ここで、

は、予測の不確実性をコントロールするパラメータである。
以降において、（８）式は単に「終了条件」と称呼される場合がある。 The process of selecting the next calculation point by the formula (7) is repeatedly executed until the condition of the following formula (8) is satisfied.

here,

Is a parameter that controls the uncertainty of the prediction.
Hereinafter, the expression (8) may be simply referred to as “end condition”.

<Specific operation of the present embodiment>
The specific operation of the present embodiment will be described. The CPU 14 (simply referred to as "CPU") of the arithmetic unit 11 is adapted to execute the "search routine" shown in FIG.

  The inventor of the present application prepared a simple evaluation function f1 having multiple peaks. The evaluation function f1 is a function of two variables and one response. When two design variables x1 and x2 are input to the evaluation function f1, the evaluation function f1 outputs one response value. An upper limit value and a lower limit value are set for each of the design variables x1 and x2. Furthermore, the upper limit value of the response value of the evaluation function f1 is set to a predetermined value (“-4.0” in this example). In the following, a flow for obtaining "the range of the design variables x1 and x2 in which the response value of the evaluation function f1 is a predetermined value (-4.0) or less" as the "established range" will be described using the search routine of FIG. .

  The CPU starts the routine of FIG. 2 from step 200, sequentially performs the processes of step 205 and step 210 below, and then proceeds to step 215.

  Step 205: The CPU reads the initial data set previously stored in the ROM 16 or the HDD 17 into the RAM 15. The initial data set is a data set of a combination of design variables x1 and x2. The worker may input the initial data set to the search device 10 using the input unit 12. Hereinafter, in a two-dimensional space in which the first axis is the design variable x1 and the second axis is the design variable x2, the point defined by the combination of the design variables x1 and x2 is referred to as the “design point”.

Step 210: The CPU calculates an optimization problem that minimizes the evaluation index J ₀ (x) for each design point included in the initial data set. That is, the CPU calculates the equation (4) for each design point.

Next, when the CPU proceeds to step 215, the CPU determines whether or not the established range exists. Specifically, the CPU determines whether or not there is a design point within the established range among the design points included in the initial data set. The CPU determines that the established range exists when the value calculated by the equation (4) (the minimum value of the evaluation index J ₀ (x)) is a negative value.

  It is assumed that there is no design point within the established range among the design points included in the initial data set (that is, the value calculated by the equation (4) does not become a negative value). In this case, the CPU makes a “No” determination at step 215 to proceed to step 220. As described above, when it is determined “No” in step 215, it is highly possible that the established range itself does not exist. That is, under the condition of the initial design variable x1 (upper limit value and lower limit value), the condition of the initial design variable x2 (upper limit value and lower limit value), and the condition of the initial evaluation function f1 (upper limit value), it is established. It is likely that the range does not exist. Therefore, the operator needs to review the conditions of the design variable x1, the conditions of the design variable x2, and / or the conditions of the evaluation function f1. When the CPU proceeds to step 220, it displays the value calculated by the equation (4) for each design point on the display screen 19. The worker may change the condition (upper limit value and lower limit value) of the design variable x1 and / or the condition (upper limit value and lower limit value) of the design variable x2 with reference to the value displayed on the display screen 19. . When the condition of the evaluation function f1 and / or the condition of the design variable x2 is changed, the operator prepares a new data set of design points (combination of the design variables x1 and x2) according to the change, and A new data set is input to the search device 10. Therefore, the CPU sequentially executes the processes of step 210 and step 215 for the new data set of the design point. Further, the worker may change the condition (upper limit value) of the evaluation function f1. In this case, the CPU sequentially executes the processes of step 210 and step 215 under the condition (upper limit value) of the changed evaluation function f1.

  On the other hand, it is assumed that there are design points within the established range among the design points included in the initial data set (that is, the value calculated by the equation (4) becomes a negative value). In this case, the CPU determines “Yes” in step 215, sequentially performs the processes of steps 225 to 245 described below, and then proceeds to step 250.

  Step 225: The worker uses the input unit 12 to set the important design space. Specifically, the worker selects two or three design variables to be noted from the plurality of design variables. The important design space is a two-dimensional or three-dimensional space defined by the design variables selected by the operator. In this example, the operator selects the design variable x1 and the design variable x2. Therefore, the important design space is a two-dimensional space defined by the design variables x1 and x2.

  Step 230: The operator uses the input unit 12 to input a data set for performing a search in the important design space. This data set is called a "search data set". The search data set is a data set of calculation points (points on a two-dimensional space defined by the design variables x1 and x2) in the important design space. As shown in FIG. 3, the calculation points included in the search data set satisfy the condition of the design variable x1 (upper limit value and lower limit value) and the condition of the design variable x2 (upper limit value and lower limit value), and the important design It is evenly distributed in the space (so that the distances between the calculation points are evenly spaced). Note that the search data set is not limited to the example of FIG. 3, and may be appropriately set according to the target design variable and / or evaluation function.

Step 235: The CPU calculates an optimization problem that minimizes the evaluation index J ₀ (x) for the search data set. That is, the CPU calculates the equation (5) for each calculation point of the search data set. Further, the CPU determines whether each calculation point of the search data set is within the established range. When the value of y (a) in the equation (5) is a negative value, the CPU determines that the calculation point is within the established range.

Step 240: The CPU uses the kriging model for the set of calculation point candidates (calculation candidate points), and predicts the predicted value y _pred (a) of y (a) in Expression (5) and the dispersion of the predicted values. Calculate the value s (a) representing In this example, 20 levels are set at predetermined intervals on each axis of the design variables x1 and x2, and 400 calculation candidate points are prepared.

Step 245: The CPU calculates the equation (7) using the predicted value y _pred (a) of y (a) and the value s (a) representing the variation of the predicted value, and selects the next calculation point. To do. That is, the CPU selects the point having the maximum EI function from the set of calculation candidate points as the next calculation point.

  Next, when the CPU proceeds to step 250, it determines whether or not the end condition is satisfied. Specifically, the CPU determines whether the condition of expression (8) is satisfied.

  If the condition of Expression (8) is not satisfied, the CPU makes a “No” determination at step 250 to proceed to step 260, and adds the next calculation point obtained at step 245 to the search data set. . Next, the CPU returns to step 235, calculates the equation (5) with respect to the added calculation point, and thereafter executes the processing of steps 240 to 250 in order. The CPU repeatedly executes such processing until the end condition is satisfied.

  On the other hand, when the condition of the expression (8) is satisfied, the CPU makes a “Yes” determination at step 250 to proceed to step 255, and displays on the display screen 19 information indicating the established range in the important design space. FIG. 4 is a diagram displayed on the display screen 19 in this example. The CPU displays the established range in the important design space on the display screen 19 by enclosing the range of calculation points where the value of y (a) calculated in step 235 is a negative value with a solid line. After that, the CPU proceeds to step 295 to end the present routine.

  In the present example, the device of the present embodiment was able to obtain the established range (the established range shown in FIG. 4) that matches the correct answer by 64 calculations. Therefore, the apparatus according to the present embodiment can reduce the calculation amount by about 84% as compared with the case of calculating all calculation candidate points (400 points). As described above, the present embodiment was able to efficiently search the range (establishment range) of the design variables x1 and x2 that satisfies the condition (upper limit value) of the evaluation function f1.

<Application to vehicle model>
Next, an example in which the present embodiment is applied to a vehicle model will be described. The inventor of the present application prepared a suspension beam model shown in FIG. In this model, the design variables shown in Table 1 below were defined.

Furthermore, the evaluation functions shown in Table 2 below were defined. In addition, "-" in the lower limit value column represents that there is no lower limit value.

Using the apparatus of this embodiment, the routine of FIG. 2 was executed for the above suspension beam model. The operator selects the design variable K _x and the design variable K _y in step 225. Therefore, the important design space is a two-dimensional space defined by the design variable K _x and the design variable K _y .

Further, in this example, 20 levels are set at predetermined intervals on each axis of the design variable K _x and the design variable K _y , and 400 calculation candidate points are prepared.

FIG. 6 is a diagram displayed on the display screen 19 in step 255. Of the calculation points in the important design space (two-dimensional space defined by the design variables K _x and K _y ), the calculation points within the established range are displayed as black circles and the calculation points outside the established range are displayed as white circles. To be done. As described above, instead of or in addition to the solid line indicating the established range, information indicating whether or not each calculation point is within the established range may be displayed in the diagram on the display screen 19.

  Further, as shown in FIG. 6, a region may be divided for each range of the value of y (a) in the expression (5), and a different color may be displayed for each region. In the example of FIG. 6, the smaller the value of y (a), the lighter the color. Thereby, the operator can also grasp which calculation point does not satisfy the condition (upper limit value and lower limit value) of the evaluation function to the extent outside the established range.

  The device of the present embodiment was able to obtain the established range (the established range shown in FIG. 6) that matches the correct answer by 61 calculations. Therefore, the apparatus according to the present embodiment was able to reduce the calculation amount by about 85% as compared with the case of calculating all calculation candidate points (400 points).

Note that the apparatus of the present embodiment may display the diagram shown in FIG. 7 on the display screen 19 in step 255. In the example of FIG. 7, which evaluation function at each calculation point is dominant to the evaluation index J ₀ (x) (that is, which evaluation function contributes most to the evaluation index J ₀ (x))? Information is displayed. For example, the calculation point indicated by the upward triangle mark means that the violation amount selected by the max operation of Expression (2) is the violation amount of the evaluation function J ₁₅ . With this configuration, it is possible to easily grasp at which calculation point in the important design space the constraint of which evaluation function is violated.

  Although the search device according to the embodiment has been described above, the present invention is not limited to the above embodiment, and various modifications can be made without departing from the object of the present invention.

  The model used in step 240 is not limited to the Kriging model. As an alternative model to the Kriging model, “when the value of the first function at each calculation point included in the search data set is given, the first value at another calculation point where the value of the first function is uncalculated A model capable of calculating the predicted value of the value of the function and calculating the value indicating the variation of the predicted value may be adopted. Even in this case, the same processing as that of the above-described present embodiment can be performed. For example, a Gaussian process may be used to calculate the predicted value of y (a) and a value representing the variation in the predicted value.

  The function used to select the next calculation point in step 245 is not limited to the EI function. As an alternative function of the EI function, “probability of whether or not the calculation candidate point is within the established range, including the predicted value of the value of the first function at the calculation candidate point and the value representing the variation of the predicted value as variables May be adopted. Even in this case, the same processing as that of the above-described present embodiment can be performed. For example, a function similar to the normal cumulative distribution function on the right side of the first term of Expression (6) may be adopted. The CPU may select the calculation candidate point that maximizes the value of the function as the next calculation point.

  When three design variables are selected in step 225, the important design space is a three-dimensional space. In this case, in step 255, the space showing the established range in the three-dimensional space may be displayed. According to this, the worker can grasp the established range in the three-dimensional space.

  Although the example of the vehicle model has been described as the structure, the present invention is not limited to this, and the present embodiment can be applied to the design stage of another structure, for example.

10: search device, 11: arithmetic unit, 12: input unit, 13: output unit, 14: CPU, 15: RAM, 16: ROM, 17: hard disk (HDD), 18: I / O interface, 19: display screen .

Claims (1)

A plurality of design variables and at least one evaluation function that outputs at least one of the plurality of design variables as an input value and outputs an output value according to the input value are defined, and the output value of the evaluation function is satisfied. When a predetermined condition to be set is set, a search device for searching the established range representing the range of the design variable satisfying the predetermined condition,
Whether or not the established range exists by defining a violation amount that violates the predetermined condition in each of the evaluation functions and calculating a first function that minimizes the evaluation index in consideration of all the violation amounts. A first calculation means for determining whether
Second calculating means for setting an important design space, which is a space defined by two or three variables of the plurality of design variables, when the first calculating means determines that the established range exists;
Third calculating means for calculating the first function for a search data set, which is a set of calculation points in the important design space, and determining whether or not the calculation points are within the established range;
Fourth calculation means for calculating a predicted value of the value of the first function of a calculation candidate point which is a candidate for the next calculation point in the important design space and a value representing variation of the predicted value using a predetermined model. The predetermined model is a different calculation point in which the value of the first function is uncalculated when the value of the first function at each calculation point included in the search data set is given. And a fourth calculation means, which is a model capable of calculating the predicted value of the value of the first function and calculating the value representing the dispersion of the predicted value.
A second value that includes, as a variable, the predicted value of the value of the first function at the calculation candidate point and a value that represents the dispersion of the predicted value, and that represents the probability of whether the calculation candidate point falls within the established range. Using a function, the process of selecting the next calculation point from the set of calculation candidate points and adding the next calculation point to the search data set is repeatedly executed until a predetermined end condition is satisfied. A fifth calculating means,
Display means for displaying information indicating the established range in the important design space;
A search device comprising.

Images