The demand for product design has risen as new technologies such as big data and cyber-physical systems (CPSs) emerge. Digital product design entails employing modern digital technology to complete the product design process. In the contemporary world, the trend of consumers’ preference for customized design products is rising exponentially. To meet this demand, the product evolutionary design approach is used in this paper. It is based on user preferences by realizing the personalized intelligent design of products. For instance, if one takes the example of table lamps, based on the method of parametric modeling, the parameters of shape and material are used as product genes for chromosome coding. In our proposed approach, the multiobjective bee colony interactive evolutionary algorithm (MBCIEA) is used to build the table lamp evolution system by relying on the color harmony theory. To evolve the design scheme, the user choice is employed as the fitness value. The findings of a comparative experiment between this technique and the existing schemes suggest that this method may let users engage more effectively in collaborative design and achieve a good customized design scheme.

1. Introduction

Advanced digital technologies such as geometry modeling, kinematic and dynamic simulation, multidisciplinary coupling, virtual assembly, virtual reality (VR), multiobjective optimization (MOO), and human-computer interface are used to complete the product design process. Although there is no common definition for customized design, its core meaning is that a customized product is created to meet the customer's unique and varied needs as fast and inexpensively as feasible. Many academics have investIGAted the technique and essential technologies of product design. With the advent of the fourth industrial revolution, emerging user groups have increasingly strong demands for product modeling. Under the development trend of the times, the design and production modes are developing towards personalization and intellectualization. At present, product design and manufacturing are dominated by enterprises. The specific process is implemented by designers and consumers only accept the design results. How to establish the relationship between consumers' personalized needs and product design and production is a subject worthy of research [1, 2].

In terms of production mode, the maturity of 3D printing technology makes the manufacturing mode socialized and personalized. However, in terms of design mode, the effective intelligent means of personalized design is still missing. The design and performance of product appearance still depend on the designer. It is being helped using computer-aided design software. However, it is difficult to realize efficient customization for users’ personalized needs. It is of great research significance to provide consumers with the means to participate in design rather than regular design schemes. It reduces the design threshold and lets users participate in the design process. Thus, it helps build a personalized design platform for user collaborative design [35].

An effective personalized intelligent design platform needs to face the following research problems:(i)How to read users’ personalized demands and turn users’ subjective aesthetic images into objective design schemes(ii)How to reduce the threshold of participation in the design and ensure that users without design experience can get a satisfactory design scheme by participating in collaborative design(iii)How to get the output of a complete design scheme, including the design of product modeling and color matching

As a result, developing an intelligent design platform to enable the development of industrial goods is critical. In this approach, digital design or goods will progress toward intelligence and customization. This paper proposes a personalized intelligent design method. Through the innovation of system construction method and algorithm, a personalized intelligent design system meeting the above three requirements is established.

The rest of the research paper is structured as follows. Section 2 will explain all the related work that has been done in this research. It is being followed by Section 3 which explains the construction of the product evolution system. Section 4 and Section 5 explain the multiobjective interactive bee colony evolutionary algorithm and color control. Finally, Section 6 and Section 7 explain the experiment, analysis, and concluding remarks. The explanation is as follows.

Product gene is the use of genetic engineering principles in the realm of custom product creation. Gero and Roseman were the first to propose the notion of a product gene [6, 7]. Researchers have improved the product gene building and population optimization approaches in recent years by broadening the meaning of the product gene notion. Early research focused on the two-dimensional design scheme of products or icons [8, 9]. With the deepening of research, researchers began to take the three-dimensional modeling of products with a simple structure as an example to carry out the intelligent optimization design of three-dimensional models [10, 11]. Some researchers also took genetic networks as the starting point [12]. This kind of research is mainly based on an interactive genetic algorithm (IGA). It takes the user score as the evolution basis to realize the evolution of the design scheme according to the user preference.

In addition to the optimization of product modeling design, some researchers also turn their goal toward product color matching [13, 14]. In addition to the color scheme evolution based on user evaluation, the color evolution research based on color harmony theory has gradually emerged. It pays more attention to the improvement of color matching beauty [1516].

In recent years, GAN has also been widely used in the design field because of its excellent image generation ability. Some researchers have proposed 3D-GAN based on GAN and applied it to the intelligent design of the 3D model. The point cloud model built using current technology, however, lacks information due to its large 3D convolution computation needs, as illustrated in Figure 1. Furthermore, this technology has stringent training data requirements, and the resulting product picture is constrained by learning examples, making it impossible to innovate [17, 18].

In the generation of the three-dimensional model, the intelligent generation based on the product gene has more details, higher authenticity, and faster generation speed. It is a better choice for the application. Starting from product gene, combined with design principles and product features, this paper establishes the expression of product gene covering rich features. The evolutionary process of interactive genetic algorithms is optimized to better adapt to the diversified characteristics of personalized aesthetic needs.

3. Construction of Product Evolution System

To reflect the feasibility of personalized intelligent design based on product genes, a table lamp, a product with moderate morphological complexity, is selected as the research object. It is a common product, which has the characteristics of unified function and diverse appearance. It is conducive to reflecting the users’ personalized demands for the product.

3.1. Table Lamp Gene Construction

To construct the lamp gene, first, we take the gene coding as input for parametric modeling.

3.1.1. Gene Extraction

To meet the personalized customization needs of different customers in the market segment, the goal of product gene construction is to make the gene code compile having many kinds of lamps as possible. Therefore, before constructing the gene, we need to understand the appearance characteristics of lamps and find an appropriate parametric modeling method [19].

First, we investigate and analyze the table lamps on the market. All kinds of table lamps can be summarized into three basic units: lamp base, lamp pole, and light source. The feature types are subdivided based on these three units. The features regulated by each gene in the product gene chain are totally responsible for the appearance of the automatically manufactured product. Automatic parametric modeling is realized by writing MEL (Maya embedded language) script command by mapping the relationship between genes and traits. To make the table lamp population cover the basic modeling characteristics of the table lamp on the market, the table lamp gene is divided into characteristic genes and parameter genes. The parameter gene controls the size details of each part, such as length, width, height, cross-section radius, and turning position. It is shown in Table 1.

3.1.2. Gene Construction

The method of combining genes into particular biological qualities in biological genetics is analogous to the process of gene coding to parametric modeling. The system requires the designer to formulate a set of rules to inform the modeling program which part of the parametric modeling corresponds to each gene in the gene coding as a parameter. When compiling rules, in order to improve the efficiency of gene expression, according to the control mode of genes on traits, the characters were divided into independent characters and dependent characters (see Table 2).

Figure 2 shows a table lamp model and its wireframe model generated based on system automation on the left, and the main top view of the wireframe model of the base on the first floor of the model on the right. Some dimensions are marked in the figure, among which R0, 1, and T1 ∼ T5 are independent traits. They are controlled by one locus. It represents the radius and position height of the corresponding section circle of the model. The radius and height of other section circles are dependent properties. For instance, the radius of other section circles of the first layer base is R2 ∼ 4, and the first layer base is obtained by the lofting operation of these section circles. R4 is controlled by three genes of “base convergence c”, “incremental parameter d”, and “bottom radius R1”. The value of c is - 1, 1, or 0. If the value is 1, the cross-section size converges upward. If it is - 1, otherwise, if it is 0, the cross-section size remains unchanged. The value of d is 0 ∼ 1. It represents the difference between 4 and 1. The calculation of R4 can be expressed as formula (1). R2 and 3 are the radius of the auxiliary section circle, and its existence is controlled by “base softness s”. When the value of s is 2, it denotes the presence of two auxiliary sections. It has the ability to shape a curved side curve. When s is set to 1, an auxiliary section is created. It has the ability to shape a curved side curve. There is no auxiliary section when s is set to 0, and the side is linear.

The key parameters of the material part of the model include RGB value, metallicity, roughness, high luminosity, IOR value, and transparency. Each parameter is represented by a linear gene locus. 40 gene loci are used to represent the possible basic materials and colors in the table lamp. In addition, some special materials and colors are expressed by characteristic genes. For example, it is difficult to obtain white materials by random values of basic color genes. Only when the R, G, and B values of colors are close to 0 at the same time can the materials appear white. This probability is very small as white is a common color in product design. As a result, a characteristic gene of “if it is white” must be defined. The basic color gene is disregarded when the value of this gene is set to be within a particular range. The material color should be changed directly to white [20].

Based on the above theory, the parameter gene and characteristic gene of the product are constructed, and the constraint conditions and topological rules are established. The shape of the lamp is determined by the value of 197 genes. The distribution of genes is shown in Table 3.

3.1.3. Gene Coding

The abstract expression of gene coding is shown in Figure 3. One chromosome represents the complete gene coding of an individual. It is composed of 197 genes. Each gene is composed of 5 binary gene points, and each chromosome has a total of 985 gene points.

It is necessary to determine the value range of each gene in the model. The threshold value of the parameter gene is determined by the extreme value of the product size in the market research. For example, the threshold value of the circle radius of the lowest cross section of the circular base of the table lamp is 4.5cm–25 cm, and the value of the gene is defined as the floating-point number in the range of (4.5, 25.0). The threshold value of the characteristic gene is determined by the number of characteristic types, and the gene parameter is an integer. Each number represents a certain type or characteristic of different parts. The following is the formula for converting binary gene point coding to gene value: M indicates the location of the gene point, D represents the value of the gene point, and the characteristic gene requires rounding after conversion.

The program can randomly generate multiple arrays as gene codes and take the random gene codes as input for parametric modeling to establish the initial population. It runs on an ordinary computer with 8 GB (GIGA Bytes) memory and an MX350 graphics card. It takes a total of 15 seconds to generate 60 lamp design schemes. The rendered results are shown in Figure 4.

3.2. Evolutionary System Construction

In the process of solving the problem of a personalized design scheme, the evaluation of the design scheme depends on the user’s subjective aesthetics. When the fitness function is difficult to be objectively defined, IGA is an appropriate choice. Its main feature is that the individual fitness value is determined by the user’s interactive evaluation rather than the calculation formula. The algorithm is suitable for dealing with the optimization problems of implicit or fuzzy evaluation indexes, especially for decision-making problems with emotional elements such as emotion, intuition, and preference. Taking the user’s score as the evolutionary basis, let users participate in the process of product personalized customization design and ensure that the design results meet the users’ subjective needs [21].

The IGA algorithm flow is shown in Figure 5. The user scores each individual in the population according to the individual’s subjective preference. The system defines the individual fitness value as the value scored by the user for the individual and carries out the evolutionary operation after scoring. The main evolutionary operations include selection, crossover, and variation. Among them, the selection operation uses roulette rules according to the fitness value of each individual in the population. Some excellent individuals are selected from the population as parents for reproduction. The crossover procedure couples the selected people at random and swaps some chromosomes between them according to the crossover rate. With the likelihood of mutation rate, the mutation operation modifies the gene values of some loci [22].

4. Multiobjective Interactive Bee Colony Evolutionary Algorithm

Relevant studies have revealed some defects in the design of the IGA application. They mainly include the following:(i)The evolution process is long and the fatigue degree of subjects is high(ii)In the later stage of evolution, the design scheme is easy to fit too fast and the evolution efficiency is slow. Furthermore, it is demonstrated that the user's personalized aesthetics is a multiobjective problem rather than a single-objective problem for the challenge of personalized product design.Aiming at the above problems, a feasible algorithm optimization method is proposed [23, 24]

4.1. Cluster Structure

A population is a collection of all the same species in a region. A cluster is a subset of the population and a collection of closely related individuals living together. For example, bees in the same hive are a cluster. The purpose of MBCIEA is based on the cluster structure.

Individuals in the cluster usually come from a common ancestor and have similar gene composition. Therefore, the chromosome similarity between individuals is taken as the basis for cluster grouping. Formula (3) is the calculation formula of chromosome similarity. Among them, k is the gene weight, and the weight is 1 by default. The more key characteristic genes have a greater impact on the lamp shape, the weight is also greater, and d stands for the identity of the corresponding position of the binary coding of the gene. If the gene is the same, it is 1, and if the gene is different, it is 0. m stands for the position of the gene point in the gene . When the value of R is greater than a certain value, it is considered that two individuals are similar and belong to a group [25].

With the iteration of the population, there will be an obvious cluster structure in the later stage. Individuals with high fitness in the population have more opportunities to enter inheritance, and there will be more genes inherited from this individual in the offspring population. Because these offspring inherit their excellent genes, their fitness values are often high. It also has competitive advantages. After several generations of accumulation, more and more individuals will inherit the excellent genes of this ancestor. This is a sign that the population is gradually maturing and approaching the user satisfaction solution, but the emergence of cluster structure has brought some adverse effects:

⁃With the continuous expansion of excellent clusters, it is easy to reproduce among individuals in the cluster. Because the genetic differences among individuals in the cluster are very small, the result of mutual reproduction will be to produce an individual with very similar genes, resulting in the gradual convergence of genes in the later cluster and affecting the evolutionary efficiency.

⁃With the continuous expansion of excellent clusters, the probability of individuals outside the cluster entering heredity is greatly reduced. Evolution will revolve around the largest cluster. Even if individuals outside the cluster have excellent genes in the process of evolution, they will be swallowed up by the advantages of large clusters and are difficult to further evolve. The exploration of users’ personalized aesthetics is the solution to multiobjective problems. The evolution process needs to take into account multiple clusters, rather than be limited to the evolution of a single cluster [26].

To solve the above problems, a Multiobjective Interactive Bee Colony Evolutionary Algorithm (MBCIEA) is proposed by combining the idea of multiobjective optimization with Bee Evolutionary Genetic Algorithm and interactive genetic algorithm [27].

4.2. Algorithm Principle and Flow

According to the laws of nature, bees will strengthen their memory of nectar in the process of collecting nectar. Therefore, bees in a bee colony often collect nectar from only one flower. A bee colony represents a cluster. Suppose that bees in a bee colony will only pollinate one orchid, and in an environment with two kinds of orchids, bees in two bee colonies will perform their respective duties. If orchid A grows in a humid environment, and orchid B is exposed to the sun, bee colony A needs to evolve to adapt to humidity, and bee colony B needs to evolve to adapt to the sun. The evolution of bee colony A and bee colony B will converge on two different goals.

For the evolution of personalized products, bees, the main body of evolution, also represent products, and different orchids represent different aesthetic preferences of a user. The process of seeking corresponding products to meet different aesthetic perspectives according to different angles of a person’s aesthetic preferences is also like the process of a continuous evolution of bees in each hive aiming at different orchids.

The algorithm flow is shown in Figure 6. The basic idea is to cluster the first-generation population according to gene similarity to form several bee colonies of different sizes. In each colony, the individual with the highest fitness is selected as the queen and the other individuals as the drone. Colony A sends the queen to reproduce with the drones in other colonies. It should be noted that only the queen whose fitness value is higher than the average fitness of the population of the generation has the opportunity to be sent as a representative. The overall size of the population is constant, and there is competition in each colony. The better the queen's fitness value in the colony, the more possibilities for her genes to enter the inheritance, increasing the number of people with comparable outstanding genes in the following generation. As a result, if the queen has good genes, the colony may grow.

The evolutionary process shown in Figure 6 can be described as follows:(i)Step 1. The user scores all individuals in the P (T) generation population and takes the score value as the fitness value(ii)Step 2. Read the fitness value. If the evolution termination condition is met, the algorithm outputs the result and stops running; otherwise, it continues(iii)Step 3. Divide the bee population into multiple bee colonies according to gene similarity(iv)Step 4. According to the fitness of individuals in each colony, the optimal individual in each colony is counted as a queen bee and the rest as a drone(v)Step 5. Select a queen bee and use the roulette rules based on fitness to select n drones from other drones. The higher the fitness of the queen bee, the greater the value of N(vi)Step 6. The queen bee performs crossover and mutation operations with n drones to generate N newborn individuals(vii)Step 7. Repeat steps 5 and 6 to make all queens complete the selection(viii)Step 8. Obtain t+ 1 generation population P (T + 1), t = t + 1(ix)Step 9. Turn to step 1

Compared with IGA, MBCIEA has three main characteristics:(i)Each pair of parents originally contains queen bees (the optimal individual of the cluster)(ii)The queen bees in each cluster only mate with the drones outside the cluster(iii)The queen bees will be selected regardless of the size of the cluster

The first feature strengthens the mining ability of the genetic algorithm. The offspring mainly rely on the cross operation and the optimal individual of the cluster. Of course, this also increases the possibility of premature convergence of the algorithm. The second feature breaks the limitation of the later search space of the genetic algorithm and improves the exploration ability of the genetic algorithm by mating with drones in different bee colonies. The third feature makes that when the advantages of large bee colonies are expanded in the later stage if there are queens with high fitness in other small bee colonies. These three characteristics speed up the evolution efficiency of the algorithm and prevent premature fitting. At the same time, the single-target search is transformed into a multitarget search.

5. Color Control

In order to build a completely personalized design system, color harmony theory is introduced to realize the coevolution of the model and material.

5.1. Color Harmony Theory

The difficulty of coevolution of model and color is that when the product model changes with evolution, the number, position, size, and other factors affecting the color matching effect will inevitably change. At this time, the effect of the same color matching scheme will also change. Therefore, by introducing the color harmony theory, the relationship between material and model is established. In order to realize the coevolution of the two, color harmony evaluation is a theoretical method for evaluating the beauty of color matching. Moon and Spencer [15] used the Munsell HVC color system to give the evaluation formula of color harmony. It can calculate the harmony degree of color combination. The calculation formula of color harmony is as follows: the O represents the color order factor and C represents color complexity [28].

The relationship between the order factor and the change of single-color attributes has been tested through experiments, and the data results are consistent with the results obtained in the early theory, which proves the reliability of the theory. This theoretical system is applied to the color matching evolution system to develop the color control evolution in the direction of high beauty. Firstly, these discrete values are fitted by cubic spline interpolation in the program. The curve after fitting the discrete data is shown in Figure 7. Then, the interpolation in three dimensions is obtained from the gene and substituted into the fitted function to obtain the value of the order factor [16].

5.2. Color Evolution Rules

For product color matching, the advantages and disadvantages of color matching cannot be simply evaluated according to several colors themselves. In addition to the changing relationship of color itself, its position in the product, the area occupied, the interval between colors, and other factors also have a decisive impact on the matching results. Zhang Quan’s research has proved that color matching is not only affected by the attributes of color itself. At the same time, it is also affected by such factors as color block area, spacing between color blocks, contact length between color blocks, and color block position. The influence factors of each factor affecting color harmony are shown in Figure 8 [29].

According [25] to the size of the influence factor, combined with the specific conditions of each part of the color block in the offspring model, the color harmony intensity between each two-color block is weighted. The higher the color harmony intensity between the two-color blocks, the greater the influence of the color matching between the two-color blocks. The color genes of the parents are replaced to take out the required number of color genes, the color harmony degree of each color scheme is obtained by weighted calculation, and the group with the highest total score of color harmony degree is selected.

For example, if a child’s lamp model has three-color blocks, it is necessary to take out three-color blocks from the parent’s color blocks and give them to the child. There must be mutual spacing, area, and position relationships among the three-color blocks of the child. The influence factor of color harmony can be regarded as the weighted sum of these factors, and S is the area parameter, representing the ratio of color block area to the total area. Formula (6) is the calculation formula of area parameters; d is the distance parameter, representing the distance between two-color blocks; p is the position parameter, representing the distance between the color block and the visual center. Ks, kd, and kp represent the weight of three factors, respectively. Formula (7) is the calculation formula of the influence factor of color harmony.

The calculation formula of color harmony degree of color matching relationship composed of multicolor blocks is shown as

For all color permutations and combinations in the parent, we substitute them into the corresponding parameters of the offspring model, calculate a group of color matching schemes with the highest color harmony, and give them to the offspring. Figure 9 depicts the result. The male and female parents on the left and the offspring formed via cross inheritance on the right are represented by the two models on the left. Both male and female features can be seen in the form of the progeny. The color matching on the offspring model also comes from the parents, and the optimal color matching scheme is given to the offspring through program calculation and selection.

6. Experiment and Analysis

In this section, the experimental methods, data and results, and evolutionary analysis and discussion are explained. This will help us analyze the experimental values as well as the end analysis of the experiments performed. The explanation is as follows.

6.1. Experiment

A total of 12 subjects were invited to participate in the experiment. It includes 7 males and 5 females, aged between 20 and 55 years. Among them, 6 subjects were college students or workers with design experience, and the rest had no design experience. Subjects with design experience could better provide analysis and suggestions for the experimental process while the subjects without design experience could verify that the system had no design knowledge threshold.

Subjects are required to score each table lamp design scheme according to subjective aesthetic preference in the evolutionary system as the individual fitness value. The score adopts the 5-point scale method. 1 represents very dissatisfied, 2 is dissatisfied, 3 represents the average, 4 represents satisfied, and 5 represents very satisfied. After scoring all individuals, subjects need to score the diversity of the population in the range of 1–5. The higher the score, the richer the diversity and the population will evolve to the next generation. Each individual was required to undertake a series of tests in both the classic IGA and MBCIEA evolutionary systems (hereinafter referred to as “IGA system” and “MBCIEA system”).

Previous experiments found that in the later stage of evolution, the average fitness value of the population increased significantly slowly after it reached 3.5 with evolution. At this time, the next-generation population generally contains more than six individuals with 5 points, which means that the population has evolved and matured. In the interview, the subjects thought that at this time, the design scheme in line with personalized aesthetics can be selected from the population. Therefore, the termination conditions of evolution are set as follows:(i)The average fitness value reaches 3.5(ii)The population contains more than 6 individuals with 5 points.

After one of the conditions is met, the evolution is terminated, and the subjects select the most satisfactory design scheme from the final population as the evolution result. In order to avoid excessive fatigue and inaccurate evaluation caused by too long experimental time, the maximum number of evolution times is set to 15. If the termination conditions are not met after evolution to 15 generations, the experiment is automatically terminated

6.2. Experimental Data and Results

The experimental data are divided into four groups according to Table 4. Figure 10 contains the basic data of 12 subjects completing the experiment in the two systems. Among them, subjects 1–6 are those without design experience, and the rest are those with design experience.

In the IGA system, the number of evolutions used by the subjects to complete the evolution is more than 10, of which two inexperienced subjects have not reached the target conditions after evolving to 15 generations. In the MBCIEA system, the subjects generally complete the evolution earlier, and the number of evolutions used is 7 to 11. The average number of evolutions required by the subjects to complete the evolution in the IGA system is 12.7, and the average number of evolutions required in the MBCIEA system is 9.2.

6.3. Evolutionary Analysis and Discussion

According to the evolution trend table in Figure 10, groups A and B are compared with groups C and D. From the overall trend of average fitness, the average fitness of the IGA system increases slowly in the early and late stages, while the average fitness of MBCIEA system increases steadily. From the overall trend of population diversity, the diversity of the IGA system decreases gradually with evolution, while the diversity of the MBCIEA system decreases rapidly in the early stage. It drops to about 3.5 and tends to be stable.

According to the analysis of data, in the first three generations of evolution, compared with the IGA system, the population diversity and average fitness value of the MBCIEA system show a faster change trend, indicating that its evolutionary efficiency is higher in the early stage. In the interview, four subjects mentioned that in the early evolution of the IGA system, the previous generation gives low score design scheme types, which will still appear in the next generation. This is the primary explanation for early evolution's inefficiency. In two systems, Figure 11 demonstrates how a subject has progressed to the third-generation population. The population of the IGA system does not form a cluster structure, while there are seven small-scale clusters in the population of the MBCIEA system. At the same time, the average fitness value of MBCIEA rises faster, which proves that the preference choices of subjects in the first two generations are better inherited.

In the middle and later stages of evolution, the population diversity of the IGA system still maintained a downward trend, while the population diversity of the MBCIEA system tended to be stable, and the fitness value maintained a rapid increase rate, indicating that the population maintained a healthy evolutionary trend in the middle and later stages. Figure 12 shows the population of a subject after the evolution of the two systems to the target conditions. Among them, the scheme in the green box is the most satisfactory scheme selected by the subjects in the population. Compared with the two populations, the final population of the IGA system converges in one direction and tends to assimilate seriously while the final population of the MBCIEA system converges to multiple targets and forms multiple medium-sized clusters. The subjects believe that most of the target populations of the IGA system have too high weight duplication. The emergence of these repeated individuals is of little significance, while the target population of the MBCIEA system contains more design types and has more choice space. It is found that the design scheme selected in the blue box in B is very similar to the final satisfactory scheme of subjects in A, indicating that the evolution direction of the IGA system is only an aesthetic angle of subjects. MBCIEA system can explore users’ personalized aesthetic preferences from multiple angles, to more effectively generate design schemes in line with users’ preferences.

According to the factor of whether the subjects have design experience or not, we compare the experimental data in Figure 10. In the IGA system, those with design experience complete evolution faster than those without design experience, and the average difference in evolution times is 2 generations, while the MBCIEA system reduces the difference between the two to 0.7 generations. Combined with interview and evolution process analysis, it is found that those with design experience have more clear goals in the scoring process. Among them, two subjects said that in the early stage of evolution, there were excellent design schemes that only need fine-tuning in detail. The evolution may be finished faster if the scheme may be actively updated throughout the evolution process while the evaluation of inexperienced individuals on each design scheme is based on their present subjective feelings, which is susceptible to the preceding generation's high-score modeling. When the next generation appears in a similar form, they found a better type and gave them high scores. The MBCIEA can efficiently cluster the new type and rapidly expand its advantages by relying on the advantages of the queen bee. The IGA will continue the previous advantages. It is difficult for new high-score types to accumulate advantages. It is proven that the MBCIEA can better reduce the design threshold and improve the use feeling of those without design experience.

Compared with the IGA system using a traditional evolutionary algorithm, the MBCIEA system consumes fewer evolutionary times. Subjects agree that their evolutionary process has a better experience, and most of them say that the evolutionary results of the system are more in line with personalized aesthetics. The disadvantage is that some subjects still express that they are tired of scoring in the evolutionary process. The subject of fatigue is still a problem to be solved in the interactive evolutionary method.

7. Conclusion

This paper explores the intelligent design method of personalized products, constructs the design evolution system based on product genes, and improves the evolution algorithm. The lamp evolution system is built using the MBCIEA and the color harmony theory. The fitness value represents the user's desire, which is utilized to evolve the design scheme. According to the findings of comparison trials between this approach and the traditional way, this method allows users to better engage in collaborative design and obtain satisfying individualized design schemes. Experiments show the advantages of the method so that users without design experience can participate in collaborative design. This method can also be applied to the personalized design of other products, which is helpful to promote the personalized and intelligent development of the design mode.

Data Availability

The datasets used and analyzed during the current study are available from the corresponding author upon reasonable request.

Conflicts of Interest

The authors declare that they have no conflicts of interest.


This research was funded by the National Natural Science Foundation of China, grant no. 62002321.