Machine learning and case reasoning hybrid-driven mechanical product design method

Abstract

The invention discloses a machine learning and case-based reasoning hybrid-driven mechanical product design method, which comprises the following steps: firstly, constructing an alternative case set, and then constructing a product demand-product parameter correlation matrix and a product function clustering relation matrix based on domain knowledge; decomposing a new mechanical product design task into a plurality of parallel subtasks, and constructing a clustering model of the subtasks by utilizing the clustering relationship; solving the clustering model by adopting a genetic algorithm to obtain a solved product function clustering relation matrix; on the basis of values of clustering relations in the product function clustering relation matrix, product parameters with the same clustering relation serve as a subtask; obtaining a demand parameter associated with each sub-task through the matrix, sorting the sub-tasks, obtaining an available case set through demand parameter retrieval, constructing a classification model, and training a neural network model; and for each subtask, obtaining a design result of the subtask through the classification model and the neural network model.

Description

Technical Field

[0001] The present invention relates to the field of mechanical product design, and in particular to a mechanical product design system driven by a hybrid of machine learning and case reasoning. Background Art

[0002] The development process of mechanical products mainly includes steps such as demand analysis, scheme design, detailed design, parts production and processing, assembly, testing, production and use, and maintenance. In the detailed design stage, designers use computers and other auxiliary tools to complete the overall design of mechanical products and parts design, and generate design instructions, drawings, and factory data. Product design requires scheme design, detailed design, and verification of calculation results based on design requirements. It is an important part of mechanical product development and has a significant impact on product quality, manufacturing costs, and product life.

[0003] The increasing globalization of the market has given rise to the need to accelerate the development of new products. Design methods that mainly rely on human engineering knowledge and auxiliary tools can no longer meet the needs. Therefore, it is necessary to explore lower-cost and more efficient mechanical product design methods. With the development of technology, many classic mechanical product design methods have emerged with the development and application of design concepts and tools, such as optimization design, computer-aided design, industrial design, value engineering and modular design, etc. These methods have played a significant role in assisting designers to analyze, optimize and combine solutions. Although many intelligent design methods have emerged to improve design efficiency and reduce design costs, there are still many problems that have not been solved: for example, due to the relatively small number of case samples of mechanical products, the application of learning algorithms in mechanical product design is restricted; fixed knowledge template-driven design is difficult to maintain and upgrade the knowledge base; case-based reasoning design focuses on providing similar solutions to target problems, but the adjustment of specific parameters needs to be completed based on experience and knowledge. Summary of the invention

[0004] The present invention provides a mechanical product design method driven by a hybrid of machine learning and case reasoning to overcome the problems existing in existing intelligent design methods.

[0005] In order to achieve the above tasks, the present invention adopts the following technical solutions:

[0006] A mechanical product design method driven by a combination of machine learning and case-based reasoning includes the following steps:

[0007] Step 1: Based on the completed mechanical product design task, the demand parameters and product parameters are obtained and digitized, and the digitized demand parameters and product parameters are used as an alternative case to construct an alternative case set;

[0008] Step 2: Based on the knowledge of domain experts and designers, a product demand-product parameter correlation matrix and a product function clustering relationship matrix are constructed; the product demand-product parameter correlation matrix is ​​used to characterize the causal correlation between different demand parameters and product parameters, as well as the serial correlation between different product parameters; the product function clustering relationship matrix is ​​used to characterize the clustering relationship between product parameters;

[0009] Step 3, decomposing the new mechanical product design task into multiple parallel subtasks, and using the clustering relationship to build a clustering model of the subtasks; using a genetic algorithm to solve the clustering model to obtain a solved product function clustering relationship matrix; based on the clustering relationship value in the product function clustering relationship matrix, taking product parameters with the same clustering relationship as a subtask;

[0010] Step 4: Based on the causal correlation in the product requirement-product parameter correlation matrix, determine which requirement parameters are related to the product parameters in the decomposed subtasks, so as to obtain the requirement parameters associated with each subtask; based on the sequence correlation between the product parameters, determine the completion order of the subtasks;

[0011] Step 5: Based on the requirement parameters associated with the subtasks, use similarity as a search condition to filter out available cases from the candidate case set and construct an available case set;

[0012] Step 6: Use the available case set to build a classification model and train a neural network model; for each subtask, obtain the design result of the subtask through the classification model and the neural network model.

[0013] Furthermore, the requirement parameters are used to characterize the requirement items of the design task, and the product parameters are used to characterize the corresponding product attributes;

[0014] Both requirement parameters and product parameters include numerical parameters and coded parameters. Numerical parameters are simply parameters represented by continuous numerical values, while coded parameters are parameters that cannot be represented by continuous numerical values. When performing numerical processing, numerical parameters do not need to be encoded, while coded parameters need to be encoded.

[0015] Furthermore, the product demand-product parameter correlation matrix is ​​expressed as follows:

[0016]

[0017] In the product demand-product parameter correlation matrix M rp In, P r i is the i-th demand parameter, P p j is the jth product parameter; different demand parameters and product parameters P ri and P p j The causal relationship between ij (i=1,2,...,n;j=1,2,...,m) represents P p i The implementation must depend on P p j , n represents the number of demand parameters, m represents the number of product parameters; different product parameters P p k and P p j The serial correlation between kj (k=1,2,...,m;j=n,n+1,...,n+m) is used to represent P p k and P p j The sequence relationship, that is, P p k The realization of P p j The premise of realization; among them, a ij and a kj The value of is given by domain experts and designers, and is represented by professional design knowledge.

[0018] Furthermore, the product function clustering relationship matrix is ​​expressed as follows:

[0019]

[0020] In the matrix M p In the product parameter P p j′ (j′=1,2,...,m) and P p j The clustering relationship b between (j=1,2,...,m) jj′ It can be obtained by analyzing three relationships, including functional relationship, geometric relationship and physical relationship. The functional relationship represents P p j′ and P p j They are the key to completing a certain function; the geometric relationship indicates P p j′ and P p j Adjacent in the geometry of the mechanical product; physical relationship indicates P p j′ and P p j There may be flows of material, information, or energy between them;

[0021] Each bjj′ The specific values ​​are as follows:

[0022]

[0023] Where A(x) is the two product parameters P provided by expert x according to the t-th relationship. p j′ and P p j The evaluation value of the relationship between * (t) is the average evaluation value of knowledge under the tth relationship, F is the number of experts who provide this evaluation value; M is the number of relationships.

[0024] Furthermore, the clustering model of the subtask is expressed as follows:

[0025]

[0026]

[0027] Where MaxA in and MinA a are the objectives of subtask decomposition, which respectively represent maximizing the degree of aggregation within the design subtasks and minimizing the degree of association between the design subtasks; C(a) is the number of subtasks into which the design task is decomposed; w j′j is an aggregation operator, if P p j′ and P p j In the same subtask, w ij =1; otherwise, w ij =0, m represents the number of product parameters.

[0028] Furthermore, the method of selecting available cases from the candidate case set by using similarity as a search condition and constructing the available case set includes:

[0029] (1) Construct the similarity between subtasks and alternative cases:

[0030]

[0031]

[0032] S i =S in +w*S ic

[0033] In the formula, y j1 is the normalized value of the jth requirement parameter associated with the subtask, y ji is the normalized value of the jth demand parameter in the ith alternative case, k 1is the number of numerical parameters in the subtask associated requirement parameters; k 2 is the number of encoded parameters in the requirement parameters associated with the subtask; is the encoding parameter in the jth requirement parameter associated with the subtask; is the encoding parameter of the jth requirement parameter in the i-th alternative case; S in Indicates the similarity between the design problem and the alternative cases in terms of matching of numerical factors; S ic represents the similarity between the design problem and the alternative case in terms of encoding parameter matching; w represents the coefficient, S i represents the similarity between the design problem and the i-th alternative case;

[0034] (2) In the set of alternative cases, select i The value of S is retrieved i The value is used to sort the alternative cases in the alternative case set, and the first H alternative cases are used as available cases to construct the available case set; H is a preset number.

[0035] Furthermore, the use of the available case set to construct a classification model and train a neural network model includes:

[0036] For the requirement parameters and product parameters of each available case in the available case set, respectively extract the numerical parameters and coding parameters in the requirement parameters and product parameters;

[0037] Using the coding parameters in the demand parameters of the available cases and the coding parameters in the product parameters, a classification model is constructed based on the KNN algorithm to obtain the classification relationship between the demand parameters and the product parameters;

[0038] A data set is constructed using the numerical parameters in the demand parameters and the numerical parameters in the product parameters of the available cases, and a neural network is trained to obtain a neural network model for characterizing the fitting relationship between the demand parameters and the corresponding product parameters.

[0039] Furthermore, for each subtask, the classification model and the neural network model are used to obtain the design results of the subtask, including:

[0040] For each subtask, the encoding parameters in the requirement parameters associated with the subtask are input into the classification model to obtain the corresponding product parameters; the numerical parameters in the requirement parameters associated with the subtask are input into the neural network model to obtain the corresponding product parameters; the design result of the subtask is the set of two output product parameters.

[0041] Compared with the prior art, the present invention has the following technical features:

[0042] 1. The present invention quantifies the relationship between demand parameters and design parameters, and on this basis applies the GA method to decompose the design problem into multiple parallel small-scale sub-tasks; the present invention can not only increase the data sample, but also reduce the data dimension, and is more suitable for the design of mechanical products with smaller samples.

[0043] 2. The present invention proposes a method of combining machine learning with product design knowledge to intelligently implement the scheme and detailed design of mechanical product design. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a schematic diagram of the process of the present invention.

[0045] Figure 2 It is a schematic diagram of the design process of requirements and products in the process of association analysis and task decomposition of the present invention. DETAILED DESCRIPTION

[0046] In view of the problem that there are few actual mechanical product case samples, the present invention provides a mechanical product design method driven by a hybrid of machine learning and case reasoning. Figure 1 , including the following steps:

[0047] Step 1: Based on the completed mechanical product design task, obtain the demand parameters and product parameters and digitize them, take the digitized demand parameters and product parameters as an alternative case, and build an alternative case set.

[0048] Among them, the design task of mechanical products is the process of designing corresponding product attributes (including functions, performance, etc.) to meet the requirement items based on the requirement items; in this scheme, the requirement items are called requirement parameters, and the product attribute values ​​are called product parameters; both the requirement parameters and the product parameters contain numerical parameters (parameters that can be represented by continuous numerical values, taking cutting tool design as an example, such as the strength and hardness of the blade), as well as coded parameters (parameters that cannot be represented by continuous numerical values, such as the clamping method of the blade, which can be selected from screw clamping, hydraulic clamping and pressure plate clamping).

[0049] When numerical processing is performed on demand parameters and product parameters, the numerical parameters do not need to be encoded, but the coded parameters need to be encoded, as shown below:

[0050] C k =(x 1 ,x 2 ,...,x i ,...x nk )

[0051] In the formula, C k is the encoding of the kth encoding parameter; x iis the i-th case; nk is all possible cases; for example, if nk is 4, and the case belongs to 3; then, C k =(0010).

[0052] Step 2: Express the knowledge of domain experts and designers in a formulaic form, so as to reasonably decompose the design task into multiple parallel subtasks and form a product requirement-product parameter correlation matrix M. rp and product function clustering relationship matrix M p :

[0053]

[0054]

[0055] In the matrix M rp In, P r i is the i-th demand parameter, P p j is the jth product parameter; different demand parameters and product parameters P r i and P p j The causal relationship between ij (i=1,2,...,n;j=1,2,...,m) represents P p i The implementation must depend on P p j , n represents the number of demand parameters, m represents the number of product parameters; different product parameters P p k and P p j The serial correlation between kj (k=1,2,...,m;j=n,n+1,...,n+m) is used to represent P p k and P p j The sequence relationship, that is, P p k The realization of P p j The premise of realization; among them, a ij and a kj The value of is given by domain experts and designers, and is represented by professional design knowledge.

[0056] In the matrix M p In the product parameter P p j′ (j′=1,2,...,m) and P pj The clustering relationship b between (j=1,2,...,m) jj′ It can be obtained by analyzing three relationships (functional relationship, geometric relationship and physical relationship). The functional relationship represents P p j′ and P p j They are the key to completing a certain function; the geometric relationship indicates P p j′ and P p j Adjacent in the geometry of the mechanical product; physical relationship indicates P p j′ and P p j There may be a flow of material, information or energy between them.

[0057] M p Each b jj′ The specific values ​​are as follows:

[0058]

[0059]

[0060] Where A(x) is the two product parameters (P p j′ and P p j ) The set of A(x) values ​​given by different experts is a piece of knowledge; K * (t) is the average evaluation value of knowledge under the t-th relationship, F is the number of experts who provide this evaluation value; M is the number of relationships; b jj′ is the comprehensive evaluation value of knowledge under M relational standards, representing product parameter P p j′ and P p j The clustering relationship between them.

[0061] Product design is a complex process that requires consideration of multiple factors, including user needs, product features, and the relationships between them. Figure 2 A design process with n demand parameters and m product parameters is given as an example. rp It is a matrix used to represent the causal relationship between demand parameters and product parameters, as well as the sequence relationship between product parameters; P r i is the i-th demand parameter; P p jis the jth product parameter, n and m are the number of requirements and product parameters respectively.

[0062] Step 3: decompose the new mechanical product design task into multiple parallel subtasks, and use the clustering relationship to build a clustering model of the subtasks; use a genetic algorithm to solve the clustering model to obtain the solved product function clustering relationship matrix M p Based on M p The clustering relationship in b jj′ The value of the clustering relationship b jj′ The same product parameters as a subtask.

[0063] With the same number of alternative cases, compared with the overall design task, decomposing the task into smaller subtasks can reduce the constraints of case retrieval, thereby increasing data samples, reducing data dimensions, and improving design efficiency.

[0064] According to the product function clustering relationship matrix M provided above p , whose goal is to maximize the aggregation within the design subtasks while minimizing the correlation between the design subtasks; after subtask decomposition, the clustering model of the subtasks is as follows:

[0065]

[0066]

[0067] Where MaxA in and MinA a are the objectives of subtask decomposition, which respectively represent maximizing the degree of aggregation within the design subtasks and minimizing the degree of association between the design subtasks; C(a) is the number of subtasks into which the design task is decomposed; w j′j is an aggregation operator, if P p j′ and P p j In the same subtask, w ij =1; otherwise, w ij =0; Because a product parameter cannot be concentrated in different subtasks, it must meet the following constraints:

[0068] if w j′j =1&&w j′j″ =1

[0069] w jj″ =1

[0070] Among them, j″=1,2,...,m and j″≠j.

[0071] Taking C(a) = 2 and four product parameter classifications as an example, the Mp As shown below:

[0072]

[0073] According to M p The clustering relationship b jj′ , and set their P to 1 p 1 , P p 2 and P p 4 As a group, P is 0 p 3 As a group, there are two subtasks in total.

[0074] Taking turning tool design as an example, its required parameters mainly include workpiece material, cutting range, cutting accuracy and cutting state, etc., and product parameters mainly include tool material, tool geometry (front angle, back angle and tool tip arc radius, etc.) and tool clamping method. Through the task decomposition of the invented method, the design of external cylindrical turning tools is finally decomposed into four subtasks. That is, tool material design based on workpiece material and cutting range, tool geometry design based on workpiece material, tool material and cutting range, tool cutting groove chip design based on workpiece material, tool material and cutting range, and tool clamping method design based on workpiece material and cutting state.

[0075] Step 4: Based on the product demand-product parameter correlation matrix M rp The causal correlation in M ​​can determine which demand parameters are related to the product parameters in the decomposed subtasks, so as to obtain the demand parameters associated with each subtask; rp The serial correlation between product parameters in the process can be used to determine the order in which subtasks are completed; it can also be used to determine the impact of the design results of a subtask on the design of the next subtask.

[0076] Step 5: Based on the requirement parameters associated with the subtasks, use similarity as a search condition to filter out available cases from the candidate case set and construct an available case set, as follows:

[0077] (1) Construct the similarity between the subtask and the alternative case as follows:

[0078]

[0079]

[0080] S i =S in +w*S ic

[0081] In the formula, y j1is the normalized value of the jth requirement parameter associated with the subtask, y ji is the normalized value of the jth demand parameter in the ith alternative case, k 1 is the number of numerical parameters in the subtask associated requirement parameters; k 2 is the number of encoded parameters in the requirement parameters associated with the subtask; is the encoding parameter in the jth requirement parameter associated with the subtask; is the encoding parameter of the jth requirement parameter in the i-th alternative case; S in Indicates the similarity between the design problem and the alternative cases in terms of matching of numerical factors; S ic represents the similarity between the design problem and the alternative case in terms of encoding parameter matching; w represents the similarity between the design problem and the alternative case in terms of encoding parameter matching; in and S ic The magnitude of the additional coefficient. i Represents the similarity between the design problem and the i-th alternative case.

[0082] (2) In the set of alternative cases, select i The specific principle is based on the value of S i The value of S is used to sort the cases in the candidate case set; i A set of candidate cases is sorted by value, and the first H candidate cases are taken as available cases to construct a set of available cases.

[0083] Step 6: Use the available case set to build a classification model based on the KNN algorithm and train the neural network model; for each subtask, obtain the design result of the subtask through the classification model and the neural network model. The specific instructions are as follows:

[0084] (1) for each available case in the available case set, extract the numerical parameters and coding parameters in the required parameters and product parameters respectively;

[0085] Using the coding parameters in the demand parameters of the available cases and the coding parameters in the product parameters, a classification model is constructed based on the KNN algorithm to obtain the classification relationship between the demand parameters and the product parameters;

[0086] A data set is constructed using the numerical parameters in the demand parameters and the numerical parameters in the product parameters of the available cases, and a neural network is trained to obtain a neural network model for characterizing the fitting relationship between the demand parameters and the corresponding product parameters.

[0087] (2) For each subtask, the coding parameters in the requirement parameters associated with the subtask are input into the classification model to obtain the corresponding product parameters; the numerical parameters in the requirement parameters associated with the subtask are input into the neural network model to obtain the corresponding product parameters; the design result of the subtask is the set of two output product parameters.

[0088] The design results of each subtask are obtained in the order of the subtasks, thereby completing the new mechanical product design task; then the design results of each subtask can be three-dimensionalized using 3D modeling software to obtain the final product design 3D model.

[0089] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A mechanical product design method driven by a combination of machine learning and case-based reasoning, characterized in that: The following steps are involved: Step 1: Based on the completed mechanical product design task, the demand parameters and product parameters are obtained and digitized, and the digitized demand parameters and product parameters are used as an alternative case to construct an alternative case set; Step 2: Based on the knowledge of domain experts and designers, a product demand-product parameter correlation matrix and a product function clustering relationship matrix are constructed; the product demand-product parameter correlation matrix is ​​used to characterize the causal correlation between different demand parameters and product parameters, as well as the serial correlation between different product parameters; the product function clustering relationship matrix is ​​used to characterize the clustering relationship between product parameters; Step 3, decomposing the new mechanical product design task into multiple parallel subtasks, and using the clustering relationship to build a clustering model of the subtasks; using a genetic algorithm to solve the clustering model to obtain a solved product function clustering relationship matrix; based on the clustering relationship value in the product function clustering relationship matrix, taking product parameters with the same clustering relationship as a subtask; Step 4: Based on the causal correlation in the product requirement-product parameter correlation matrix, determine which requirement parameters are related to the product parameters in the decomposed subtasks, so as to obtain the requirement parameters associated with each subtask; based on the sequence correlation between the product parameters, determine the completion order of the subtasks; Step 5: Based on the requirement parameters associated with the subtasks, use similarity as a search condition to filter out available cases from the candidate case set and construct an available case set; Step 6: Use the available case set to build a classification model and train a neural network model; for each subtask, obtain the design result of the subtask through the classification model and the neural network model.

2. The mechanical product design method driven by a hybrid of machine learning and case-based reasoning according to claim 1, characterized in that: The requirement parameters are used to characterize the requirement items of the design task, and the product parameters are used to characterize the corresponding product attributes; Both requirement parameters and product parameters include numerical parameters and coded parameters. Numerical parameters are simply parameters represented by continuous numerical values, while coded parameters are parameters that cannot be represented by continuous numerical values. When performing numerical processing, numerical parameters do not need to be encoded, while coded parameters need to be encoded.

3. The mechanical product design method driven by hybrid machine learning and case-based reasoning according to claim 1, characterized in that: The product demand-product parameter correlation matrix is ​​expressed as follows: In the product demand-product parameter correlation matrix M rp In, P r i is the i-th demand parameter, P p j is the jth product parameter; different demand parameters and product parameters P r i and P p j The causal relationship between ij (i=1,2,...,n;j=1,2,...,m) represents P p i The implementation must depend on P p j , n represents the number of demand parameters, m represents the number of product parameters; different product parameters P p k and P p j The serial correlation between kj (k=1,2,...,m;j=n,n+1,...,n+m) is used to represent P p k and P p j The sequence relationship, that is, P p k The realization of P p j The premise of realization; among them, a ij and a kj The value of is given by domain experts and designers, and is represented by professional design knowledge.

4. The mechanical product design method driven by hybrid machine learning and case-based reasoning according to claim 1, characterized in that: The product function clustering relationship matrix is ​​expressed as follows: In the matrix M p In the product parameter P p j′ (j′=1,2,...,m) and P p j The clustering relationship b between (j=1,2,...,m) jj′ It can be obtained by analyzing three relationships, including functional relationship, geometric relationship and physical relationship. The functional relationship represents P p j′ and P p j They are the key to completing a certain function; the geometric relationship indicates P p j′ and P p j Adjacent in the geometry of the mechanical product; physical relationship indicates P p j′ and P p j There may be flows of material, information, or energy between them; Each b jj′ The specific values ​​are as follows: Where A(x) is the two product parameters P provided by expert x according to the t-th relationship. p j′ and P p j The estimated value of the relationship between * (t) is the average evaluation value of knowledge under the tth relationship, F is the number of experts who provide this evaluation value; M is the number of relationships.

5. The mechanical product design method driven by hybrid machine learning and case-based reasoning according to claim 1, characterized in that: The clustering model of the subtask is expressed as follows: Where MaxA in and MinA a are the objectives of subtask decomposition, which respectively represent maximizing the degree of aggregation within the design subtasks and minimizing the degree of association between the design subtasks; C(a) is the number of subtasks into which the design task is decomposed; w j′j is an aggregation operator, if P p j′ and P p j In the same subtask, w ij =1; Otherwise, w ij =0, m represents the number of product parameters.

6. The mechanical product design method driven by hybrid machine learning and case-based reasoning according to claim 1, characterized in that: The method of selecting available cases from the candidate case set by using similarity as a search condition and constructing the available case set includes: (1) Construct the similarity between subtasks and alternative cases: S i =S in +w*S ic In the formula, y j1 is the normalized value of the jth requirement parameter associated with the subtask, y ji is the normalized value of the jth requirement parameter in the ith alternative case, k1 is the number of numerical parameters in the requirement parameters associated with the subtask; k2 is the number of coded parameters in the requirement parameters associated with the subtask; is the encoding parameter in the jth requirement parameter associated with the subtask; is the encoding parameter of the jth requirement parameter in the i-th alternative case; S in Indicates the similarity between the design problem and the alternative cases in terms of matching of numerical factors; S ic represents the similarity between the design problem and the alternative case in terms of encoding parameter matching; w represents the coefficient, S i represents the similarity between the design problem and the i-th alternative case; (2) In the set of alternative cases, select i The value of S is retrieved i The value is used to sort the alternative cases in the alternative case set, and the first H alternative cases are used as available cases to construct the available case set; H is a preset number.

7. The mechanical product design method driven by hybrid machine learning and case-based reasoning according to claim 1, characterized in that: The method of using the available case set to construct a classification model and train a neural network model includes: For the requirement parameters and product parameters of each available case in the available case set, respectively extract the numerical parameters and coding parameters in the requirement parameters and product parameters; Using the coding parameters in the demand parameters of the available cases and the coding parameters in the product parameters, a classification model is constructed based on the KNN algorithm to obtain the classification relationship between the demand parameters and the product parameters; A data set is constructed using the numerical parameters in the demand parameters and the numerical parameters in the product parameters of the available cases, and a neural network is trained to obtain a neural network model for characterizing the fitting relationship between the demand parameters and the corresponding product parameters.

8. The mechanical product design method driven by a hybrid of machine learning and case-based reasoning according to claim 1, characterized in that: For each subtask, the classification model and the neural network model are used to obtain the design results of the subtask, including: For each subtask, the encoding parameters in the requirement parameters associated with the subtask are input into the classification model to obtain the corresponding product parameters; the numerical parameters in the requirement parameters associated with the subtask are input into the neural network model to obtain the corresponding product parameters; the design result of the subtask is the set of two output product parameters.

Images