JP2001101251A - Automatic designing device and design processing method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To realize an automatic designing device and a design processing method capable of automatically designing an optional functional form such as the shape of a lens by improvement of an artificial life type discovery system (S-system). SOLUTION: This device is constituted so that a functional solution is stably and efficiently obtained by performing uniform crossover to replace part or the whole of a constant factor between genes, simplifying the tree structure of a function and introducing mutation to replace part of the function with another function when an artificial life is reproduced by the artificial life type discovery system (S-system) to be used in the automatic designing device and the designing processing method.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an automatic design apparatus and a design processing method for enabling an arbitrary part having a predetermined function such as an optical lens and a filter to be designed by computer technology.

[0002]

2. Description of the Related Art In recent years, artificial life (A-
Life) is being actively researched. For example, research on establishment of an evolution model using artificial life, morphological growth process, evolutionary learning, parallel processing, accident repair, image generation by self-reproduction, and the like can be mentioned. These studies can be broadly divided into art-oriented studies, science-oriented studies, and engineering-oriented studies.

[0003] The art-oriented research is to create a work of art based on a recursive self-reproduction model and to develop a system for supporting the work. Science-oriented research
It establishes the computational logic of life. Engineering-oriented research develops systems that have the ability to evolve and adapt, and has been applied to various fields, such as shape design and scheduling by self-organization, insect robot design, and genetic algorithms ( Genetic Al
gorithm: hereinafter abbreviated as GA). GA, which is the basis of the present invention, is designed by devising how to give an evaluation function and how to express a solid.
Can solve combination problems. A framework in which GA is extended to a tree structure expression is referred to as Genetic Programming (hereinafter abbreviated as GP). GP is based on the concept of GA, but the difference is that GA mainly aims at optimization, whereas the scope of GP is limited to solving practical problems such as AI problems, robots, molecular biology, etc. Is a wide variety of problems (Ref. 1).

[0004] Several unknown functions have been discovered so far using GP. However, functions discovered using GP are very different from functions discovered by humans, and are very complicated. Is often. In addition, the solution is often unstable in GP. Therefore, the present inventors first,
A new function discovery system (S-system) that extends GP has been proposed (reference document 2). This S-system expresses functions as genes and introduces various analogies of organisms such as sexual reproduction into the algorithm of discovery, so this system is called "artificial life type discovery system".

[0005] The S-system has a feature that the solution is hard to diverge and is stable, and the function discovered in the S-system is simpler than the GP method. As a result of applying this S-system to various laws, it was found that this system has excellent function discovery ability. Then, as a result of applying this system to image recognition, it was possible to extract a plurality of different curves included in the image (Reference Document 3).

(Reference 1) Koza, J., Genetic Programm
ing: On the Programming of Computers by means of Na
tural Selection.MIT press.1992 (Reference 2) Serikawa, S. and Shimomura, T., “Prop
osal of a Systemof Function-Discovery Using a Bug
Type of ArtificialLife ”, IEE Trans. Jpn, Vol. 118-
C, No.2, pp.170-179, 1998. (Reference 3) Serikawa, Mikawa, Shimomura, "Extraction of Multiple Different Curves in Figures by Insect-type Artificial Life", 1997 IEEJ Kyushu Chapter United Conference, 246, p. 102, 1997.

[0007]

There have been several reports on lens design using a genetic algorithm (GA) (reference document 4). However, they are designed to optimize the numerical value of a lens. Yes, the shape (function) of the lens itself must be determined in advance, and there is a part that depends on the experience and intuition of the designer. On the other hand, there is a genetic programming (GP) method as a system for automatically finding a function adapted to a set input / output condition, but there is a problem that a solution is complicated and unstable. Accordingly, the present inventors have previously proposed an artificial life type discovery system (S-system) which is an extension of the genetic programming (GP) method.
Even in this system, there is a problem that a solution does not easily converge or a solution cannot be obtained depending on setting conditions.

The present invention provides an artificial life type discovery system (S-sy
The aim of the present invention is to realize an automatic design apparatus and a design processing method that enable practical automatic design for generating an arbitrary function form such as a lens shape by improving the stem.

(Reference 4) Fujimoto, Y., Nishiguchi,
M., Nomoto, K., Takahashi, R. and Tsutsui, S., “An ev
olutionary design for f-thetalenses ”, Parallel Pro
blem Solving from Nature -PPSNiv.The 4th Internat
ional Conference on Parallel Problem Solving from
Nature., Pp. 992-1001, 1996.

[0010]

The artificial life type discovery system (S-system) used in the automatic design device and the design processing method of the present invention exchanges constant elements between genes when reproducing artificial life. Improvements have been made by performing uniform crossovers, simplifying the function tree structure, and introducing mutations that replace some of the functions with other functions.
The outline is described below.

FIG. 3 shows a conceptual configuration of an artificial life insect function body 1 having a gene 2 in the body. In the artificial life type discovery system (S-system) of the present invention, many such insect function bodies 1 are used.

The gene 2 of the insect function body 1 is composed of an operator, a constant,
It is composed of variables as elements. A tree structure is used to represent functions as genes. The tree structure can be equated with the S-expression of LISP. 4 as an example, the function K ₁
· Sin (x) + y / shows a tree structure and S expression (K ₂ · x). Similarly, various functions can be expressed as genes.

In the artificial life type discovery system, reproduction is performed. The feature of this system is that the concept of sexual and asexual reproduction was introduced during reproduction. First, the same kind and different kinds will be described. Species with exactly the same genetic structure are defined as the same species, and species with different genetic structures are defined as different species. In the example shown in FIG. 5, insect 1 and insect 2 are of the same species. However, the insect 3 has a different genetic structure from other insects. Insect 3 is therefore heterologous to other insects. Sexual reproduction occurs between conspecifics. It is not performed between different kinds of people. Only the constant part of the gene changes during sexual reproduction. Functions and variables do not change. Therefore,
In the case of sexual reproduction, create a sequence (gene) consisting only of constants,
Cross over constants. In addition, an insect GA strategy is used to improve the local search ability, and arrays for minute changes in each constant are created and crossover is performed on them. If the same species does not exist, asexual reproduction produces two worms with exactly the same gene.

FIG. 1 shows an overall flow of an artificial life type discovery system (S-system) according to the present invention. Next, the operation of each step (1) to (9) in the flow will be described. (1) In the initial state, Pop functional bodies (hereinafter, abbreviated as "insects") having an arbitrary function are randomly generated. (2) Generation g is set to 1. (3) is set to 0 the internal energy E _p of all of the insects. Internal energy E _p represents the fitness function to the target, is used to evaluate the function. (4) The insect number p is set to 1, and the flow of (5) to (7) is repeated until p = Pop. (5) Each insect moves. “Migration” means slightly changing the value of a constant term in a gene. Here, a large number of gene sequences in which the value of the constant term is slightly changed are created. (6) The insect captures the observation data. Here, the "capture" substitutes observations insect gene (function) to calculate the internal energy (fitness) E _p of insects. Insects that are better adapted to the observed data have higher internal energies. This means that the internal energy is increased by insects eating the observation data. (7) When the observation data and the function of the insect match well (when the internal energy exceeds a certain threshold), it is regarded that the function has been found, and the program is terminated. (8) If the generation g has reached the final generation Max _g , the program ends. Otherwise, the descendant generation routine (9) is executed. (9) After performing sexual reproduction, asexual reproduction, and mutation in the offspring generation routine, add 1 to generation g to advance the current generation number, and return to (3).

FIG. 2 shows a flow of a descendant generation routine.
The outline is as follows. (A) Par insects are left in the next generation due to the generation gap. Here, the elite strategy was used. (B) Set the insect number p to 2, and repeat the flow of (c) to (g) until p = Pop-Par. (C) Select one insect according to the tournament strategy. (D) If the insect selected in (c) has the same species, execute (e), and if the insect does not exist, execute (f). (E) Two insects are generated with the insect having the highest energy among insects of the same species as the crossing partner. Next, proceed to (g). (F) Produce two insects that are exactly equivalent to the selected insect. (G) _Replace a part of the gene with another gene structure at the rate of mutation rate _Rmut .

Although the S-system of the present invention is an extended model of GP, the biggest difference between S-system and GP lies in the intersection. G
In P, since the crossover is performed between different functions, the shape of the function changes dramatically for each generation. For this reason, the schema is easily broken, and the solution is not stable. In addition, since the constant value cannot be changed slightly, the local search capability is lacking. For example, a distribution of a function having a strong nonlinearity greatly changes due to a slight difference in a constant value. In the case of GP crossover or mutation, the probability that the form of the function changes is much higher than the probability that only the constant value changes slightly. For this reason, the gene length is often very long or extremely short. In addition, the fitness is increased to some extent, but not higher. On the other hand, S-system uses the concept of sexual reproduction and insect movement, and the tendency for insect movement to decrease as the fitness increases steadily without breaking the schema allows for more detailed searches. .

The genetic structure of the insect does not change during reproduction. For this reason, diversity of insects cannot be maintained. Thus, mutations are used to maintain insect diversity. Mutations change a part of the gene with a certain probability, and a new species is born. The next generation will asexually augment the conspecifics since the newly born insects due to the mutation will not have conspecifics. The next generation can be sexually reproductive and evolve further. FIG. 6 shows an example in which a part of the function K ₁ · sin (x) + y / (K ₂ · x) is replaced by tan (x · y) by mutation.

Another feature of the S-system is that it limits the number of similar types. As generations progress, only one type of insect may dominate the majority of the population.
This is because the excellent species selectively survive. In such a case, even if an insect having a superior functional form emerges, it is difficult to evolve if the fitness of the insect is low. That is, the search is caught by the local optimum. Therefore, it is desirable to limit the number of the same kind in order to make the search as successful as possible.

[0019]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Embodiments of the present invention will be specifically described.

In order to improve the function search efficiency of the S-system,
The following improvements can be made. (1) Change from single-point crossover to uniform crossover In the S-system, crossover is performed only for sexual reproduction. As shown in FIG. 7A, only the constant portion changes due to crossover, so in the case of sexual reproduction, an array (gene) containing only constants is created and crossover is performed on this. This part has the same strategy as the real type GA shown in FIG. 7B. In addition, in order to improve the local search ability, an array of small changes in constants is created as shown in FIG.
Crossover is also performed for this. This is a strategy similar to that of insect GA. Generally, it is said that the search capability of the crossover operator in the real GA is efficient in the order of one-point crossover <two-point crossover <uniform crossover. Therefore, in this embodiment, the crossover method is changed from one-point crossover to uniform crossover. (2) Simplification of Gene When an initial population is generated or a mutation occurs, a partial tree such as K ₁ + K ₂ or sin (K ₁ ) may appear in some cases. However, all such subtrees are K
_It can be simplified to one constant of 1. There are also operations such as addition and multiplication in which the exchange rule holds. That is,
x + K ₁ and K ₁ + x are the same. In such a case, it is determined that the gene is heterogeneous because the gene type is different. This means that many insects with different genotypes but exactly the same function will breed. Therefore, insects having other functions are unlikely to appear, and the search is likely to fall into a local solution. In such a case, it is meaningless to set an upper limit on the number of inhabitants of the same species. Therefore,
Genes are simplified according to the following rules during initial population generation and after mutation. (A) As a result of the operation, when a lower level than a certain node can be represented by one constant, the node is replaced with a constant. The value of the constant is the operation result of the subtree below the node. (B) When the exchange rule is satisfied by an operation including a constant and a variable, the constant is rearranged in the left tree.

FIG. 9 shows an example of simplification. (A), (b)
It can be seen that applying the law of simplifies the function. Although there are many equivalent functions even when no constant term is included, such as x + sin (y) and sin (y) + x, many equivalent functions related to the constant term appear in the S-system. Then, in order to save the trouble of calculation, the above (a),
Only the rule of (b) was applied. (3) Extension of functions used Functions that can be used in the S-system are + (add),-(sub), and × (m
ul), ÷ (div), power (pow), √ (sqr), sin, cos
, Tan, and log, but this time, the lens function of the illumination function was considered, and the square wave, sawtooth wave, circular wave, and step function were expanded so that they could be expressed in a tree structure. These use sqw, stw, cir, and stp, respectively, as nonterminal symbols. These functions are defined as in the following Expression 1.

[0022]

(Equation 1)

Where a = x− [x / π] · π, b =
(K + 10) · π / 20, c = x− [x / 2] · x,
[X] is the largest integer not exceeding x.

Here, the constant range of the constant k is -10 to 10. The curves generated by these functions are shown in FIG.

In FIG. 10, the square wave and the sawtooth wave in the figure are both periodic waveforms having a period π and a maximum amplitude of 1. Also,
When k ≧ 0, W ≧ π / 2, and when k <0, W ≧ π /
It becomes 2. The period of the circular wave is 2, and the maximum amplitude is 1.
By changing arguments and coefficients, it is possible to create waveforms of various periods and amplitudes. For example, 2.0 ^*
By setting cir (0.5 ^* x), it is possible to change to a circular wave having twice the period and twice the amplitude. Designing the optical component Considering the incident light as input data and the light intensity near the image point as output data, and expressing the shape of the optical component as a function, the function (shape of the optical component) that matches the input / output is automatically Can be found. First, an example in which a convex lens is designed will be described.

When the S-system was used, the shape of the lens was changed so that light was condensed at one point even when the light source position and the focal position were changed or the number of light sources was increased. In addition, as an application to lighting equipment, we attempted to design a backlight shape that uniformly emits light. It is not easy for a person without specialized knowledge to intuitively discover such a backlight shape. The following example applies the S-system to the design of optical components. Optical Component to be Designed The optical component to be designed here is an ideal lens in which all light rays emitted from a point light source are converged at one point.
However, chromatic aberration is not considered. 2 for simplicity
Consider a dimensional model.

FIG. 11 shows the shape of the convex lens, the state of image formation, and the coordinate system used when searching for the lens shape. The position of the light source is (x _L , −y _L ), the coordinates of the imaging point are (x _f , y _f ), the distance from the light source to the lens is y _L , and the range of x is from x _min to x _max . The plane part of the convex lens is fixed by a certain straight line (y = 0), and the function of the convex lens is expressed as a bug gene. A region surrounded by this function and a fixed straight line is a lens portion. In this paper, the function Ss
Search automatically by ystem. If this method is extended to a three-dimensional model and the refractive index is changed for each wavelength of the light source, it can be extended to an actual three-dimensional model. Here, a convex lens that collects all the rays emitted from the point light source into one point is designed. In other words, the evaluation should be such that the fitness of the insect increases as more rays converge at one point. The Fresnel method is used to calculate the light transmission and the reaction direction.

One ray i emitted from the light source is y
= Y _f , the score of the ray is determined by the following equation (2).

[0029]

(Equation 2)

The score _i is maximum (= 1) when t _x = x _f
Is an upward convex curve. Here, t _x is such that the ray i is y =
This is the x value of the intersection when reaching y _f . max (a, b) is
A function that returns the larger value of a and b. _Li is y
= Y represents the intensity of the light ray on the _ff plane. Note that the light intensity of one light beam emitted from the light source is 1. The light intensity decreases with each reflection and refraction. This follows Fresnel's law. Thus, score _i is obtained for all rays that have reached the y = y _f plane. At this time, the fitness Fitness is defined as in the following Expression 3.

[0031]

(Equation 3)

[0032] Here, N is the, number of rays reaching the y = y _f plane, h is the total number of rays emitted from the light source. As seen from this equation, a strong light increases the fitness as gather x _f point. The fitness value takes a value of 0 to 1. However, since a part of the incident light is reflected by the lens surface, the fitness does not reach 1.

As described above, an insect having a function capable of collecting a larger number of rays at one point has a higher fitness. If we consider fitness as the internal energy gained by the insect, the insect is absorbing energy from the light. Therefore, the light beam can be regarded as an insect mochi. Optical Component Design Algorithm The flowchart of the optical component design algorithm is the same as the flowcharts shown in FIGS. The difference is that the function discovery algorithm uses the observed data as insect prey. Along with this, the fitness is changed as in the above equation (6). FIG. 12 shows a list of parameters used in the function search.
FIG. 13 shows a convex lens designed by S-system based on this condition. However, 5% for one incident light
Less than reflected and transmitted light is not depicted. Note that for clarity the degree of condensing showed light intensity distribution that has reached the y = y _f surface at the top of FIG. The gene in FIG. 13 is a sine function found in the 98th generation. Each parameter is needed to represent an arbitrary sine function. For this gene, k ₁ is an amplitude, k ₂ phase, k ₃ is periodic,
k ₄ contributes to the height. Since the initial population is generated completely at random, these parameters are not given beforehand, but are obtained during evolution. FIG. 14 shows the transition of the maximum fitness (the fitness value of the insect with the highest fitness in each generation) for each generation.

Looking at the transition of the maximum fitness, there is a place where the fitness changes drastically, especially in the first half generation.
Examining this part in detail revealed that the gene structure changed significantly and the type changed significantly in each generation. In other words, a useful partial structure is obtained in the gene by the mutation. The convex lens in FIG. 13 is easier to form an image if the lens is symmetric with respect to the optical axis. Further, in order to refract a light beam in an appropriate direction, an element such as a thickness of a lens is required. The useful substructures for insects are these elements. Insects that have acquired these beneficial substructures during evolution will certainly evolve. The reason is as follows. When an insect with a beneficial partial structure is born due to the mutation, the insect is asexually reproduced in the next generation. Sexual reproduction becomes possible when the number increases. Therefore, in subsequent generations, the function evolves further. As the generation progresses to some extent, the shape of the function is fixed to some extent, so sexual reproduction is the main, but the movement width of the insects becomes shorter and finer search is performed.

A total of 30 trials were performed. as a result,
A shape almost similar to FIG. 13 was obtained 26 times. However, the obtained function differs for each trial. Equation 4 below shows two of the obtained functions.

[0036]

(Equation 4)

Design of convex lens when light source is different from optical axis Next, an experiment was performed by changing the position of the light source. Each parameter used in the function search is the same as in FIG.
Same as 2. The light source coordinates were (−10, −20), and the emission angle was 20 °. FIG. 15 shows a convex lens designed by S-system based on this condition. It is not easy for us humans to come up with a lens shape as shown in FIG. 15 by combining these functions. The gene in this figure is the fifth
It is a gene found in seven generations. k ₁ is related to the amplitude, k ₂ is related to the phase, k ₃ and k ₄ and k ₅ are related to the period, and there are few redundant parts to represent the function having this complicated shape. However, as can be seen from the light intensity distribution at the top of the figure, the luminous flux density at the tail of the comet, referred to as coma, is low.
Most of the light emitted from the light source is concentrated at the image forming point, so that the adaptability is high. Lens design when there are two light sources Next, a lens that can simply collect light rays emitted from two point light sources into one point is designed. This is not to determine the need for a lens, but to determine whether the S-system can flexibly adapt to various environments. The parameters used in the function search are the same as those shown in FIG. 12 except for the position of the light source and the radiation angle. The coordinates of the two light sources are (-5, -20), (5, -20), and the emission angles are both 2
0 degrees.

The gene shown in FIG. 16 is a cosine function found in the 95th generation. After the 60th generation, many cosine functions passing near the light source began to be seen. As can be seen from the figure, it is not possible to form an image completely, but the luminous flux density around the image forming part is low and most of the light emitted from the light source is concentrated at the image forming point, so that the adaptability is high.

As described above, the possibility of the S-system has been searched through the lens design. However, there is a possibility that a system that can perform thinking instead of humans can be constructed depending on the application.

Effects of S-system Improvement The effects of S-system improvement will be discussed. (A) This model, (b) When the uniform crossover is returned to the one-point crossover,
The transition of fitness was examined in four cases where (c) the gene was not simplified and (d) the GP model was used. The design conditions at this time are the same as those at the time of lens design shown in FIG. The results are shown in FIG. This figure is (a)-
For each of (d), the transition of the maximum fitness 30 times is shown. From this figure, it can be seen that this model has the highest fitness and the highest rate of fitness increase in the early generations.

In the case of lens design, it is not possible to determine which function is the correct answer, and it is not possible to determine the generation until the correct answer is reached. Therefore, the average generation until reaching 95% of the maximum fitness was obtained. FIG. 18 shows the result. FIG.
8, it can be seen that the model (a) has the fastest evolution speed. These reasons will be described below.

First, the reason why the search ability of (b) is inferior to the search ability of (a) will be considered.

In the S-system, in the case of sexual reproduction, only constants are subject to crossover. Here, this model creates a sequence (gene) consisting only of constants for sexual reproduction and performs crossover on this. This performs the same operation as the crossover of the real type GA. In general, it is considered that the crossover of the real type GA has a higher search efficiency in the case of the uniform crossover than in the one-point crossover. In this case, the result was almost the same.

Next, the reason why the search capability of (c) is lower than the search capability of (a) will be described. One is that without gene simplification, there are many redundant parts and the search efficiency is low.

For example, k ₁ · sin (x + k ₂ ) and k ₁ · sin
_{k 2 · sin (x + k} 3 k4 -k 5) but is equivalent function, for the latter is often constant term, is poor search efficiency. Another reason is that without simplification, many species with equivalent functions will appear and they may occupy a large number. In such a case, even if insects having a better functional form than those species appear, if the insects have low fitness, it is difficult to evolve. That is, the search is caught by the local optimum. On the other hand, when simplification of a function is performed, since species having an equivalent function do not occupy a large number, various species can evolve even at the stage of advanced generation. From the above, taking the average of 30 trials,
It is probable that (a) evolved faster than (c) and the final fitness was higher.

Finally, comparing (d) with (a) to (c),
(d) has the lowest rate of fitness increase in the early generation,
The fitness of the last generation is also low. The reason is as follows. In GP, a schema is easily destroyed because crossover is performed between different types. In addition, since sexual reproduction and insects do not have the concept of migration, detailed search is difficult. For this reason, even if it is not the optimal function, the fitness is steadily increased from the early generation.
In addition, as the fitness level increases, the movement width of the insects tends to decrease, and a more detailed search can be performed. Thus, the S-system compensates for the lack of local search capability associated with GP search. Application to backlight design As an example of application to lighting, we tried to design a backlight for liquid crystal. As the screen size increases, the direct-type shape as shown in FIG. 19 increases in terms of light quantity. A diffusing agent is injected into the plastic plate (diffusion plate) 3 in FIG. A plurality of light sources 4 are arranged below the light source 4 to emit light to the upper surface. The problem at this time is that the thickness _{Tf of the} entire frame and the thickness T
_p . From weight and size point, T _f and T _p are both preferably thin. However, when T _f and T _p are reduced, the distribution of light radiated to the upper surface becomes light and dark. At present, in order to eliminate this, a mask pattern 5 for light shielding is provided on the lower surface of the plastic plate. Although the light is made uniform by the mask pattern, the amount of light emitted to the upper surface is reduced. In this case, by applying a shape to the lower surface of the plastic plate, it is desirable that the backlight which does not use the mask pattern and the diffusing agent has a uniform distribution of light emitted to the upper surface and a large amount of light. Therefore, the fitness of Expression (6) shown in Expression 3 is changed to Expression (7) shown in Expression 5 below.

[0047]

(Equation 5)

Here, N is y = y in the coordinate system of FIG.
The number of rays reaching the y _f plane, L ₁ is y = y _f surface light intensity of the reach each ray in, h is the total number of rays emitted from the light source. Therefore, fit _Intensity is an index representing the light intensity on the y = y _f plane. y = y The width in the x direction of the y _f plane (x _max −
The light intensity reaching into the small width when the x _min) was equally divided into M Lf _j, the average of Lf _j and Lf _j. Then, σ
_Distribution is an index that indicates the spread of the distribution, fit
_Distribution is the index representing the uniformity of distribution in the y = y _f plane. fit _Intensity and fit _Distribution are both 0
In the range of 1, the higher the numerical value, the higher the light intensity and the more uniform the distribution. The ratio between the two is adjusted by a constant c. In this example, c = 0.5 and M = 100.

FIG. 20 shows an example of the search result. This figure is
The coordinate system in FIG. 12 is defined as x _min = −200, x _max = 200, x _f =
0.1 and y _L = 15. The light source is y
= 8 placed at 50mm intervals on the y _L plane, and the emission angle of each light source is 14
0 degrees. Sqw, stw, st
p was added, and a constraint was provided so that the thickness of the lens would be 5 mm or less. The crossover rate is 0.5 and the mutation rate is 0.
The other conditions are the same as in FIG. From this figure, it can be seen that the distribution becomes almost uniform without using a diffusing agent or a shadow mask. However, in order to obtain a product having a completely uniform distribution, improvements such as placing a diffusion sheet on the upper surface of the plastic plate or injecting a diffusion agent into the plastic may be necessary. In addition, even in the direct type backlight shown in FIG. 14, a diffusion sheet is placed on the upper surface of a plastic plate in the product.

As described above, there is a possibility that the present invention can be applied to other shape designs by changing the adaptability in accordance with the light collecting conditions and setting appropriate constraints such as the lens thickness.

In the present method, insects having parts of various shapes are born during the evolution process. Although the original purpose is to develop a component having an optimal shape, it is also significant to generate various components. Humans could make further modifications to the application based on the various components discovered by the S-system, which could be a design tool. Application to other fields Basically, S-system can be regarded as an extended model of GP, so it seems to be applicable to the field where GP is currently applied. However, the S-system has the features of excellent local search ability and relatively short gene length (function length) to be discovered. Application is possible. [A] Information processing such as graphic recognition It is applicable to simultaneous extraction of a plurality of different shapes from image data and graphic recognition. Further, since the function length is relatively short, application to high-efficiency coding is possible. [B] Shape Design of Heat Conductive Object Since the shape can be represented by a mathematical formula, the system can estimate several shapes for obtaining the heat transfer characteristic desired by the designer, and can assist the designer in terms of ideas. [C] Shape design to obtain an effective flow [D] Analog electronic circuit design In the case of analog electronic circuits, the oscillation frequency and voltage waveform differ greatly due to slight differences in the values of resistors and capacitors.
It is not easy to find a desired circuit by combining parts by GP. On the other hand, this system is effective for circuit design because it has excellent local search ability based on the concept of sexual reproduction and insect migration. System Implementation The artificial life type discovery system takes the form of a program that runs on a computer. Therefore, the present invention is realized as an apparatus configuration of a computer that performs automatic design, and also as a design processing method that uses a computer. Further, it has a form of a storage medium in which a program of the artificial life type discovery system is stored. FIG.
1 shows a configuration of an apparatus of an embodiment of a computer equipped with this system.

In FIG. 21, a computer 6 has a CPU
10, an artificial life type discovery system 13 including a memory 11, an I / O controller 12, and the like.
The data is loaded from the DD 19 to the memory 11 and executed by the CPU 10 on condition that the input data 14 and the output data 15 are set in advance. Using the display 9 and the keyboard 16 or a pointing tool such as a mouse (not shown), the processing of the artificial life type discovery system 13 proceeds in an interactive manner.

The program of the artificial life type discovery system 13 can be stored in a portable storage medium such as the FD 20 or the CD-ROM 21 and can be read from the FDD 17 or the CD-ROM drive 18 into the memory 11 or the HDD 19. The artificial life type discovery system 13 is a host 8
By managing the library in advance and downloading it to the computer 6 via the network 7 when necessary,
Can be executable.

[0054]

The automatic design apparatus and design processing method using the artificial life type discovery system (S-system) improved by the present invention include lens design, fluid design, heat conduction design, electronic circuit design, lighting design, etc. In a wide range of design fields,
The function solution can be automatically found without requiring a high level of skill or a heavy burden on the user. In particular, since it is not necessary to determine the function form in advance, it is possible to design from a free viewpoint that is not restricted by human knowledge.

[Brief description of the drawings]

FIG. 1 is a diagram showing an overall flowchart of an artificial life type discovery system.

FIG. 2 is a diagram showing a flowchart of a descendant generation routine.

FIG. 3 is a diagram showing the concept of insect genes.

FIG. 4 is a diagram showing a description of a tree structure and LISP S-expressions.

FIG. 5 is a diagram showing differences in genes in reproduction between same species and different species.

FIG. 6: K ₁ · sin (x) + y / (K ₂ · x) due to mutation
FIG. 7 is a diagram showing an example in which a part of is replaced with tan (x · y).

FIG. 7 is a diagram showing crossing by sexual reproduction and crossing of real GA.

FIG. 8 is a diagram showing the crossover (uniform crossover) of insect GA.

FIG. 9 is a diagram showing an example of gene simplification.

FIG. 10 shows function curves sqw (x, k) and stw (x,
FIG. 7 is a diagram showing k), cir (x), and stp (x).

FIG. 11 is a diagram showing a coordinate system when a lens function is searched.

FIG. 12 is a diagram showing a list of parameters used for convex lens design.

FIG. 13 is a diagram showing a convex lens (1) designed by S-system.

FIG. 14 is a diagram showing the transition of the maximum fitness for each generation.

FIG. 15 is a diagram showing a convex lens (2) designed by S-system.

FIG. 16 is a diagram showing a lens design in the case of two light sources.

FIG. 17 is a diagram showing the average fitness of 30 trials for each generation.

FIG. 18 is a diagram showing an average generation of each model until 95% of the maximum fitness is reached.

FIG. 19 is a view showing the structure of a direct backlight.

FIG. 20 is a diagram showing a lens design of a direct backlight.

FIG. 21 is a diagram showing a configuration of an apparatus according to an embodiment of the present invention.

[Explanation of symbols]

 8: Host 9: Display 10: CPU 11: Memory 12: I / O controller 13: Artificial life type discovery system 14: Input data 15: Output data 16: Keyboard 17: FDD 18: CD-ROM drive 19: HDD 20: FD 21: CD

Claims (11)

[Claims] A function that satisfies the relationship between given input data and output data is generated by using an insect function body incorporating a gene consisting of an array having an operator, a constant, and a variable as elements. Is an automatic design device that automatically generates based on artificial life type discovery system by genetic programming method. The artificial life type discovery system is composed of a plurality of insect functions that incorporate genes with arbitrary functions at random in the initial state. Insect function body generating means for generating the body, function correcting means for slightly changing the function of the insect function body gene, and applying the above input data and output data to the insect function body to evaluate the fitness of the gene function Function generation means for generating a plurality of insect function bodies of the next generation from the plurality of insect function bodies; and a plurality of insect function bodies generated by the insect function body generation means. The input data and the output data are sequentially given, and the fitness of the function is evaluated by the function evaluation means while changing the function little by little by the function correction means. When the target function is obtained, the processing is terminated. Control means for generating a plurality of insect function bodies of the next generation by the descendant generation means if not obtained, and controlling the same processing repeatedly, wherein the descendant generation means at least between the insect function bodies of the current generation. An automatic design device characterized by generating a next generation of a new insect functional body by exchanging a part or all of the constant elements of each gene. 2. A new insect function body of the next generation by replacing a part or all of the functional elements of the gene of the insect function body of the current generation with another function. An automatic design device characterized by the following. 3. The insect function body gene according to claim 1 or 2, wherein when a lower node than a certain node is represented by one constant by operation, the node is replaced by a constant. An automatic design apparatus characterized by rearranging constants in the left tree and variables in the right tree when the law of exchange is satisfied by the operation of constants and variables. 4. An automatic design apparatus according to claim 1, wherein input data and output data given to the insect function body are light distributions, respectively, and the function is a lens shape. 5. An automatic design apparatus according to claim 1, wherein input data and output data given to the insect functional body are signal waveforms, respectively, and the function is a characteristic of a filter. 6. A function that satisfies the relationship between given input data and output data is generated using an insect function body incorporating a gene consisting of an array having an operator, a constant, and a variable as elements. Is a computer-aided design processing method based on an artificial life type discovery system based on a genetic programming method. The artificial life type discovery system has a plurality of genes each having an arbitrary function randomly incorporated in the initial state. An insect functioning body generating step for generating an insect functioning body; a function correcting step for slightly changing the function of the insect functioning body gene; providing the above input data and output data to the insect functioning body; And a progeny generation step of generating a plurality of insect function bodies of the next generation from the plurality of insect function bodies. To each functional unit, sequentially provide input data and output data, while changing little by little the function in the function modification step to evaluate the fitness of the function in the function evaluation phase, terminates if the function of the target is obtained,
If the target function is not obtained, a plurality of insect functions of the next generation are generated in the offspring generation stage, and the same processing is repeated. A design processing method characterized by generating a next-generation new insect function body by exchanging a part or all of the constant elements of the above. 7. A new offspring function of the next generation by replacing a part or all of the functional elements of the gene of the insect function of the current generation with another function in the offspring generation step. A design processing method characterized by the above-mentioned. 8. A gene according to claim 6 or 7, wherein the gene of the insect functioning body is described in a tree structure of elements, and when a node lower than a certain node is represented by one constant by operation, the node is replaced with a constant. A design processing method characterized by rearranging constants in a left tree and variables in a right tree when the law of exchange is satisfied by the operation of constants and variables. 9. A design processing method according to claim 6, wherein the input data and the output data given to the insect functional body are light distributions, respectively, and the function is a lens shape. 10. A design processing method according to claim 6, wherein input data and output data given to the insect function body are signal waveforms, respectively, and the function is a filter characteristic. 11. A function that satisfies the relationship between given input data and output data is generated by using an insect function body incorporating a gene consisting of an array of operators, constants, and variables. Generating a plurality of insect functional bodies, each of which incorporates a gene having an arbitrary function at random in an initial state, the method comprising: A step of slightly changing the function of the gene of the functional body; a step of providing the above-described input data and output data to the insect functional body to evaluate the fitness of the function of the gene; Generating a plurality of insect function bodies, input data and output data are sequentially given to each of the plurality of generated insect function bodies, and the function is slightly changed. The fitness of the function is evaluated while performing, and if the target function is obtained, the process is terminated.If the target function is not obtained, a plurality of insect functions of the next generation are generated, and the same processing is repeated. However, when generating the next generation of insect function bodies, at least some or all of the constant elements of each gene are exchanged between the current generation of insect function bodies to generate the next generation of new insect function bodies. A computer-readable storage medium storing a program adapted to be executed.