Electromechanical product design method and system based on deep learning of part database

Abstract

The invention relates to the field of mechanical and electrical product design, and provides a mechanical and electrical product design method and system based on deep learning of a part database, and the method comprises the following steps: building a part database; establishing an initial hybrid design optimization model, and training and testing the initial hybrid design optimization model through the part database to obtain a hybrid design optimization model; obtaining mechanical and electrical product design demands, obtaining a plurality of demand part model index data corresponding to the mechanical and electrical product design demands based on the mechanical and electrical product design demands and the part assembly demand analysis model, and inputting the plurality of demand part model index data into the hybrid design optimization model, outputting a plurality of mechanical and electrical product 3D models corresponding to the mechanical and electrical product design requirements; and producing a corresponding mechanical and electrical product according to the mechanical and electrical product 3D model. A part database including part model index data, a part assembly scheme and a mechanical and electrical product 3D model is established, and a hybrid design optimization model is established by using a deep learning network, so that the mechanical and electrical product 3D model meeting the requirement can be automatically generated according to the mechanical and electrical product design requirement given by a user; and intelligent design of mechanical and electrical products based on the full-life-cycle data of the products is realized.

Description

Technical Field

[0001] The present invention relates to the field of electromechanical product design, and in particular to an electromechanical product design method and system based on deep learning of a parts database. Background Art

[0002] Electromechanical products are important basic equipment for national economic construction, such as construction machinery and high-speed trains. Their adaptability and intelligence have always been the focus and difficulty of the manufacturing industry. With the rapid development of technologies such as the Internet of Things, CPS, big data, and artificial intelligence, it has become possible to collect, analyze, and mine more and more product life cycle related data and provide them to designers. Among them, the large amount of data generated during the service of the product contains more accurate and objective information. How to effectively use the product's usage data and operation data to bring breakthroughs in the research and application of the field of electromechanical product design has become the main problem to be solved at present.

[0003] In the existing technology, the digital twin is used to establish a perceptual connection between the physical entity and the virtual model, and to realize feedback control and closed-loop iterative optimization under the fusion of multi-source heterogeneous information-physical data, so as to implement the working condition data and information feedback of the electromechanical product and realize closed-loop design. However, the digital twin model often finds it difficult to well capture the performance change law of the product throughout its life cycle, especially the fault evolution process that changes with the use time. It can only simulate the design based on the existing measured data, lacks the exploration of the law between product design and performance, and does not have the ability to predict the design. Summary of the invention

[0004] In view of this, the present invention proposes a mechatronic product design method and system based on deep learning of component database. By establishing a component database including component model indicator data, component assembly schemes and mechatronic product 3D models, and using a deep learning network to establish a hybrid design optimization model, it is possible to automatically generate a mechatronic product 3D model that meets the mechatronic product design requirements given by the user. This solves the problem in the prior art that the digital twin model is difficult to capture the performance change laws of the product throughout its life cycle, and is unable to effectively use the product usage and operation data to provide support for the design, thereby realizing the intelligent design of mechatronic products based on the product's full life cycle data.

[0005] The technical solution of the present invention is implemented as follows: On the one hand, the present invention provides a method for designing electromechanical products based on deep learning of a parts database, comprising the following steps:

[0006] S1, establishing a parts database, wherein the parts database includes a plurality of parts model index data, a plurality of parts assembly schemes and a plurality of electromechanical product 3D models. When the types and quantities of parts are determined, all the parts model index data and any combination of parts assembly schemes correspond to a unique electromechanical product 3D model;

[0007] S2, establishing an initial hybrid design optimization model, training and testing the initial hybrid design optimization model through a parts database, and obtaining a hybrid design optimization model;

[0008] S3, obtaining electromechanical product design requirements, obtaining multiple demand component model index data corresponding to the electromechanical product design requirements based on the electromechanical product design requirements and the component assembly requirement analysis model, inputting the multiple demand component model index data into the hybrid design optimization model, and outputting multiple electromechanical product 3D models corresponding to the electromechanical product design requirements;

[0009] S4, producing a corresponding electromechanical product according to the electromechanical product 3D model.

[0010] Based on the above technical solution, preferably, step S2 includes:

[0011] The initial hybrid design optimization model includes an input layer, a first sub-network layer, a second sub-network layer, a feature fusion layer, an assembly optimization layer, and an output layer. The input layer is connected to the first sub-network layer and the second sub-network layer, respectively. The first sub-network layer and the second sub-network layer are connected to the feature fusion layer, respectively. The feature fusion layer is connected to the assembly optimization layer, and the assembly optimization layer is connected to the output layer.

[0012] The input layer is used to receive input features; the first sub-network layer is used to extract the timing performance change characteristics of components; the second sub-network layer is used to extract the structural features of components; the feature fusion layer is used to fuse the timing performance change characteristics of components and the structural features of components to obtain fused features; the assembly optimization layer is used to apply multiple component assembly schemes to the fused features to obtain multiple corresponding 3D models of electromechanical products.

[0013] On the basis of the above technical solution, preferably, the first sub-network layer adopts a long short-term memory network, the first sub-network layer includes an input gate, a forget gate, a unit state, an output gate and a hidden state, and the characteristic function expression of the component timing performance change of the first sub-network layer is:

[0014] m r =Sigmoid(α1·[h t-1 ,x t ]+τ1);

[0015] m y =Sigmoid(α2·[h t-1 ,x t ]+τ2);

[0016] d t =m y ⊙dt-1 +m r ⊙tanh(α3[h t-1 ,x t ]+τ3);

[0017] m c =Sigmoid(α4·[h t-1 ,x t ]+τ4);

[0018] h t =m c ⊙tanh(d t );

[0019]

[0020] F s =h n ;

[0021] Among them, m r 、m y 、m c are the outputs of the input gate, forget gate, and output gate respectively, d t is the cell state at the current time step, h t is the hidden state of the current time step, h t-1 is the hidden state of the previous time step, h n is the hidden state of the last time step, x t is the input of the current time step, α1, α2, α3, α4 are the weight matrices of the input gate, forget gate, unit state update and output gate, τ1, τ2, τ3, τ4 are the bias vectors of the input gate, forget gate, unit state update and output gate, respectively. s is the input feature of the input layer, ⊙ is the Hadamard product, x n is the input of the last time step, F s is the output of the first sub-network layer.

[0022] On the basis of the above technical solution, preferably, the second sub-network layer adopts a convolutional neural network, including a convolutional layer, a sigmoid activation layer and a pooling layer, and the component structure characteristic function expression of the second sub-network layer is:

[0023]

[0024] Among them, F j is the output of the second sub-network layer, Pooling(·) is the pooling function, exp(·) is the exponential function, T s is the input feature of the input layer, ω1 is the weight matrix of the component structure feature, and δ1 is the bias vector of the component structure feature.

[0025] On the basis of the above technical solution, preferably, the feature fusion function expression of the feature fusion layer is:

[0026] F r =λF s +(1-λ)F j ;

[0027] Among them, F r is the fusion feature, λ is the fusion weight of the first sub-network layer, F s is the output of the first sub-network layer, F j is the output of the second sub-network layer.

[0028] On the basis of the above technical solution, preferably, the assembly optimization layer includes a fully connected layer, and the function expression of the output of the assembly optimization layer is:

[0029]

[0030] in, A collection of 3D models of electromechanical products. ys (·) is the assembly optimization feature parameter mapping function, F r For fusion features;

[0031] f ys (·)Yes|F r | and the mapping function of the 3D model of the electromechanical product, where |F r |Corresponding to multiple 3D models of electromechanical products.

[0032] On the basis of the above technical solution, preferably, step S3 includes:

[0033] The component assembly demand analysis model includes a demand input layer, a demand analysis layer, a component model layer and a component data output layer, wherein the demand input layer is connected to the demand analysis layer, the demand analysis layer is connected to the component model layer, and the component model layer is connected to the component data output layer;

[0034] The demand input layer is used to input the design requirements of electromechanical products, and the design requirements of electromechanical products include the size, functional requirements and service life of electromechanical products;

[0035] The demand analysis layer is used to determine the types of components required for the electromechanical product design according to the electromechanical product design requirements;

[0036] The component model layer is used to search for multiple required component model index data of the required components from the component database according to the required component type;

[0037] The component data output layer connection is used to output the plurality of required component model indicator data.

[0038] On the basis of the above technical solution, preferably, step S1 includes:

[0039] The component model index data includes component geometric properties, component material properties and component timing performance change properties, and the component assembly plan includes assembly sequence and combination method.

[0040] Based on the above technical solution, preferably, step S4 includes:

[0041] According to the 3D model of the electromechanical product, the corresponding rapid manufacturing technology is selected, the manufacturing process flow is formulated, the corresponding electromechanical product is produced, and appearance inspection and performance testing are carried out. The rapid manufacturing technology includes 3D printing, CNC machining, numerical control machining and injection molding.

[0042] On the other hand, the present invention also provides a mechatronic product design system based on deep learning of a parts database, the system comprising:

[0043] A database establishment module is used to establish a parts database, wherein the parts database includes a plurality of parts model index data, a plurality of parts assembly schemes and a plurality of electromechanical product 3D models. When the types and quantities of parts are determined, all the parts model index data and any combination of parts assembly schemes correspond to a unique electromechanical product 3D model;

[0044] A hybrid design optimization module is used to establish an initial hybrid design optimization model, train and test the initial hybrid design optimization model through a parts database, and obtain a hybrid design optimization model;

[0045] A product demand analysis module is used to obtain electromechanical product design requirements, obtain multiple demand component model index data corresponding to the electromechanical product design requirements based on the electromechanical product design requirements and the component assembly demand analysis model, input the multiple demand component model index data into the hybrid design optimization model, and output multiple electromechanical product 3D models corresponding to the electromechanical product design requirements;

[0046] The electromechanical product generation module is used to produce the corresponding electromechanical product according to the electromechanical product 3D model.

[0047] The electromechanical product design method and system based on deep learning of parts database of the present invention have the following beneficial effects compared with the prior art:

[0048] (1) By establishing a component database that includes component model index data, component assembly schemes, and 3D models of electromechanical products, and using a deep learning network to establish a hybrid design optimization model, it is possible to automatically generate a 3D model of an electromechanical product that meets the electromechanical product design requirements given by the user, thus realizing the intelligent design of electromechanical products based on the product's full life cycle data;

[0049] (2) The long short-term memory network is used in the first sub-network layer to effectively process the temporal performance change characteristics of the components, extract key information and update the hidden state, and obtain the temporal performance change characteristics of the components to reflect the performance change trend of the components over time;

[0050] (3) Using a convolutional neural network in the second sub-network layer, the structural features of the parts are effectively processed, key information is extracted and the structural features are summarized to obtain the structural features of the parts, which are used to reflect the static characteristics of the parts;

[0051] (4) By assembling and optimizing the feature parameter mapping function, the fused features are mapped to a collection of 3D models of electromechanical products, thereby realizing the output of multiple 3D models of electromechanical products. During the design process, multiple possible product design schemes can be obtained, providing designers with more choices. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0053] Figure 1 This is a flow chart of the electromechanical product design method based on deep learning of parts database of the present invention;

[0054] Figure 2 This is a structural diagram of the electromechanical product design system based on deep learning of parts database of the present invention. DETAILED DESCRIPTION

[0055] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0056] See also Figure 1The present invention provides a method for designing electromechanical products based on deep learning of a parts database, comprising the following steps:

[0057] S1, establishing a parts database, wherein the parts database includes a plurality of parts model index data, a plurality of parts assembly schemes and a plurality of electromechanical product 3D models. When the types and quantities of parts are determined, all the parts model index data and any combination of parts assembly schemes correspond to a unique electromechanical product 3D model;

[0058] S2, establishing an initial hybrid design optimization model, training and testing the initial hybrid design optimization model through a parts database, and obtaining a hybrid design optimization model;

[0059] S3, obtaining electromechanical product design requirements, obtaining multiple demand component model index data corresponding to the electromechanical product design requirements based on the electromechanical product design requirements and the component assembly requirement analysis model, inputting the multiple demand component model index data into the hybrid design optimization model, and outputting multiple electromechanical product 3D models corresponding to the electromechanical product design requirements;

[0060] S4, producing a corresponding electromechanical product according to the electromechanical product 3D model.

[0061] Specifically, the present embodiment provides a method for designing electromechanical products based on deep learning of a parts database, establishes a comprehensive parts database, including part model indicator data, assembly schemes, and 3D models of electromechanical products, and realizes systematic management of part information.

[0062] A hybrid design optimization model was established through deep learning, which can automatically generate 3D models of electromechanical products that meet product design requirements.

[0063] By utilizing component data throughout the product life cycle, data-driven intelligent design is achieved, solving the problem that it is difficult to capture the laws of product performance changes in existing technologies.

[0064] The automation of the entire process from design requirements to physical model production improves the efficiency and pertinence of electromechanical product design.

[0065] A method for designing electromechanical products based on deep learning of a parts database in this embodiment realizes intelligent design of electromechanical products based on product life cycle data by establishing a parts database and a hybrid design optimization model, thereby improving design efficiency and pertinence and solving the shortcomings of the prior art.

[0066] Step S1 includes:

[0067] The component model index data includes component geometric properties, component material properties and component timing performance change properties, and the component assembly plan includes assembly sequence and combination method.

[0068] Specifically, step S1 helps to optimize product design and improve product characteristics by comprehensively describing the geometric properties, material properties and timing performance change properties of parts, as well as the assembly sequence and combination methods, including the geometric properties, material properties and timing performance change properties of parts, which comprehensively reflect the physical properties and time-varying performance of parts. At the same time, it records the assembly information of parts, including the assembly sequence and combination method, and generates a unique corresponding 3D model of the electromechanical product, which has not only static geometric material properties, but also dynamic timing performance changes, covering the entire life cycle data of the electromechanical product.

[0069] The component geometric properties are used to describe the shape, size, geometric features, etc. of the component, including but not limited to length, width, height, curvature, angle, etc.;

[0070] The component material properties are used to describe the type of material used in the component, including but not limited to physical and chemical properties such as density, strength, hardness, thermal conductivity, and the manufacturing process and characteristics of the material.

[0071] The component timing performance change attribute is used to describe the performance characteristics of the component that change over time, including but not limited to the dynamic performance parameters such as the running speed change, temperature change, pressure change, and current change of the component;

[0072] In a specific embodiment, the timing performance change attributes of the components include the changes in the speed, current, and power of the motor over time, the changes in the strength, hardness, and fracture toughness of metal components over time, and the changes in the hardness, elasticity, and wear resistance of rubber and plastic over time and other dynamic performance parameters.

[0073] Step S2 includes:

[0074] The initial hybrid design optimization model includes an input layer, a first sub-network layer, a second sub-network layer, a feature fusion layer, an assembly optimization layer, and an output layer. The input layer is connected to the first sub-network layer and the second sub-network layer, respectively. The first sub-network layer and the second sub-network layer are connected to the feature fusion layer, respectively. The feature fusion layer is connected to the assembly optimization layer, and the assembly optimization layer is connected to the output layer.

[0075] The input layer is used to receive input features; the first sub-network layer is used to extract the timing performance change characteristics of components; the second sub-network layer is used to extract the structural features of components; the feature fusion layer is used to fuse the timing performance change characteristics of components and the structural features of components to obtain fused features; the assembly optimization layer is used to apply multiple component assembly schemes to the fused features to obtain multiple corresponding 3D models of electromechanical products.

[0076] Specifically, step S2 extracts the timing performance change characteristics and structural characteristics of the components through the first sub-network layer and the second sub-network layer respectively, thereby realizing multi-level feature extraction, which helps to more comprehensively describe the characteristics of the components.

[0077] The feature fusion layer is used to fuse the time series performance change features and structural features to obtain fusion features, thereby comprehensively considering the characteristics of different parts and components and improving the overall performance of the model. Through the assembly optimization layer, multiple component assembly schemes are applied to the fusion features to obtain multiple corresponding 3D models of electromechanical products. Through this hybrid design optimization model, multiple corresponding 3D models of electromechanical products can be automatically generated, improving design efficiency and accuracy.

[0078] In a specific embodiment, all generated electromechanical product 3D models may be ranked by performance, and a preset number of electromechanical product 3D models may be selected, thereby optimizing and comparing different assembly schemes, thereby improving design flexibility and diversity.

[0079] Step S2 effectively extracts and fuses features through a hybrid design optimization model, and applies multiple assembly schemes for optimization, thereby achieving comprehensive optimization and automatic generation of electromechanical product design and improving design accuracy and efficiency.

[0080] The first sub-network layer adopts a long short-term memory network. The first sub-network layer includes an input gate, a forget gate, a unit state, an output gate, and a hidden state. The characteristic function expression of the component timing performance change of the first sub-network layer is:

[0081] m r =Sigmoid(α1·[h t-1 ,x t ]+τ1);

[0082] m y =Sigmoid(α2·[h t-1 ,x t ]+τ2);

[0083] d t =m y ⊙d t-1 +m r ⊙tanh(α3[ht-1 ,x t ]+τ3);

[0084] m c =Sigmoid(α4·[h t-1 ,x t ]+τ4);

[0085] h t =m c ⊙tanh(d t );

[0086]

[0087] F s =h n ;

[0088] Among them, m r 、m y 、m c are the outputs of the input gate, forget gate, and output gate respectively, d t is the cell state at the current time step, h t is the hidden state of the current time step, h t-1 is the hidden state of the previous time step, h n is the hidden state of the last time step, x t is the input of the current time step, α1, α2, α3, α4 are the weight matrices of the input gate, forget gate, unit state update and output gate, τ1, τ2, τ3, τ4 are the bias vectors of the input gate, forget gate, unit state update and output gate, respectively. s is the input feature of the input layer, ⊙ is the Hadamard product, x n is the input of the last time step, F s is the output of the first sub-network layer.

[0089] Specifically, the first sub-network layer adopts a long short-term memory network, which is suitable for processing sequence data, can effectively capture the timing performance change characteristics of components, and is helpful for modeling and analyzing the dynamic characteristics of components.

[0090] The first sub-network layer includes structures such as an input gate, a forget gate, a unit state, an output gate, and a hidden state, which can effectively control the input, forgetting, and output of information, and realize the accurate extraction and processing of the timing performance change characteristics of components.

[0091] By updating the unit state and hidden state, the key characteristics of the timing performance changes of components can be effectively captured, providing important information support for subsequent feature fusion and assembly optimization.

[0092] The first sub-network layer of this embodiment adopts a long short-term memory network, which can effectively process the timing performance change characteristics of components, extract key information and update hidden states, and obtain the timing performance change characteristics of components, which are used to reflect the performance change trend of components over time.

[0093] The second sub-network layer adopts a convolutional neural network, including a convolutional layer, a sigmoid activation layer and a pooling layer. The component structure characteristic function expression of the second sub-network layer is:

[0094]

[0095] Among them, F j is the output of the second sub-network layer, Pooling(·) is the pooling function, exp(·) is the exponential function, T s is the input feature of the input layer, ω1 is the weight matrix of the component structure feature, and δ1 is the bias vector of the component structure feature.

[0096] Specifically, the second sub-network layer adopts a convolutional neural network, which is suitable for processing images and structured data, can effectively extract the structural features of parts, and helps to model and analyze the static characteristics of parts.

[0097] The second sub-network layer includes a convolution layer, a sigmoid activation layer and a pooling layer. The convolution layer is used to extract features, the activation layer is used to introduce nonlinearity, and the pooling layer is used to reduce dimensionality and retain key information, so as to achieve effective processing and extraction of component structural features.

[0098] Through operations such as convolution, activation, and pooling, the structural features of parts can be effectively captured, providing important information support for subsequent feature fusion and assembly optimization.

[0099] The second sub-network layer of this embodiment adopts a convolutional neural network, which can process the structural features of parts, extract key information and summarize the structural features, and obtain the structural features of parts to reflect the static characteristics of parts.

[0100] The feature fusion function expression of the feature fusion layer is:

[0101] F r =λF s +(1-λ)F j ;

[0102] Among them, F r is the fusion feature, λ is the fusion weight of the first sub-network layer, F s is the output of the first sub-network layer, F j is the output of the second sub-network layer.

[0103] Specifically, the feature fusion layer of this embodiment fuses the outputs of the first sub-network layer and the second sub-network layer through a feature fusion function to obtain fused features. This fusion can comprehensively consider the features extracted by different sub-networks, making the final feature representation more comprehensive and rich. The design of the feature fusion layer can effectively fuse the features extracted by different sub-networks and improve the feature expression capability.

[0104] The assembly optimization layer includes a fully connected layer, and the function expression of the output of the assembly optimization layer is:

[0105]

[0106] in, A collection of 3D models of electromechanical products. ys (·) is the assembly optimization feature parameter mapping function, F r For fusion features;

[0107] f ys (·)Yes|F r | and the mapping function of the 3D model of the electromechanical product, where |F r |Corresponding to multiple 3D models of electromechanical products.

[0108] Specifically, the assembly optimization layer of this embodiment maps the fusion features to a set of 3D models of electromechanical products by assembling the optimization feature parameter mapping function, thereby realizing the output of multiple 3D models of electromechanical products. Multiple possible product design solutions can be obtained during the design process, providing designers with more options.

[0109] Step S3 includes:

[0110] The component assembly demand analysis model includes a demand input layer, a demand analysis layer, a component model layer and a component data output layer, wherein the demand input layer is connected to the demand analysis layer, the demand analysis layer is connected to the component model layer, and the component model layer is connected to the component data output layer;

[0111] The demand input layer is used to input the design requirements of electromechanical products, and the design requirements of electromechanical products include the size, functional requirements and service life of electromechanical products;

[0112] The demand analysis layer is used to determine the types of components required for the electromechanical product design according to the electromechanical product design requirements;

[0113] The component model layer is used to search for multiple required component model index data of the required components from the component database according to the required component type;

[0114] The component data output layer connection is used to output the plurality of required component model indicator data.

[0115] Specifically, step S3 of this embodiment converts the user's electromechanical product design requirements into specific required component model indicator data through the component assembly requirement analysis model.

[0116] The component assembly demand analysis model includes a demand input layer, a demand analysis layer, a component model layer and a component data output layer.

[0117] The demand input layer is used to receive the design requirements of electromechanical products, including information on dimensions, functional requirements, and service life, to ensure that all necessary design parameters are taken into account during the design process.

[0118] The demand analysis layer can accurately determine the required component types based on the input electromechanical product design requirements, and is used to clarify the types and functions of the required components during the design process.

[0119] The component model layer searches for multiple required component model indicator data of the required components from the component database according to the determined component type, so as to help designers obtain the detailed information and characteristics of the required components.

[0120] The component data output layer is used to output the index data of multiple required component models that have been searched. These data are used to input into the hybrid design optimization model trained in the previous step to help generate multiple 3D models of electromechanical products that meet the design requirements.

[0121] Step S4 includes:

[0122] According to the 3D model of the electromechanical product, the corresponding rapid manufacturing technology is selected, the manufacturing process flow is formulated, the corresponding electromechanical product is produced, and appearance inspection and performance testing are carried out. The rapid manufacturing technology includes 3D printing, CNC machining, numerical control machining and injection molding.

[0123] Specifically, step S4 of this embodiment selects applicable rapid manufacturing technology, including 3D printing, CNC machining, numerical control machining and injection molding, according to the 3D model of the electromechanical product. According to the designed 3D model of the electromechanical product, the most suitable manufacturing technology can be selected to realize the production of the product.

[0124] After selecting the rapid manufacturing technology, the corresponding manufacturing process will be formulated, including determining the production steps, process parameters, material selection, etc., to ensure that the product can be produced according to the design requirements.

[0125] According to the established manufacturing process, the production line will begin to manufacture the corresponding electromechanical products. Through rapid manufacturing technology, the designed 3D model can be converted into an actual product prototype.

[0126] Finally, the electromechanical products produced are subjected to appearance inspection and performance testing to ensure that the appearance quality of the products meets the requirements and that the performance indicators meet the design standards, thereby verifying the quality and reliability of the products.

[0127] See also Figure 2 Based on the above method embodiments, the present invention also provides a mechatronic product design system based on deep learning of a parts database, the system comprising:

[0128] A database establishment module is used to establish a parts database, wherein the parts database includes a plurality of parts model index data, a plurality of parts assembly schemes and a plurality of electromechanical product 3D models. When the types and quantities of parts are determined, all the parts model index data and any combination of parts assembly schemes correspond to a unique electromechanical product 3D model;

[0129] A hybrid design optimization module is used to establish an initial hybrid design optimization model, train and test the initial hybrid design optimization model through a parts database, and obtain a hybrid design optimization model;

[0130] A product demand analysis module is used to obtain electromechanical product design requirements, obtain multiple demand component model index data corresponding to the electromechanical product design requirements based on the electromechanical product design requirements and the component assembly demand analysis model, input the multiple demand component model index data into the hybrid design optimization model, and output multiple electromechanical product 3D models corresponding to the electromechanical product design requirements;

[0131] The electromechanical product generation module is used to produce the corresponding electromechanical product according to the electromechanical product 3D model.

[0132] The above system embodiments and method embodiments correspond one to one, and for a brief description of the system embodiments, please refer to the method embodiments.

[0133] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for designing electromechanical products based on deep learning of component database, characterized in that: The following steps are involved: S1, establishing a parts database, wherein the parts database includes a plurality of parts model index data, a plurality of parts assembly schemes and a plurality of electromechanical product 3D models. When the types and quantities of parts are determined, all the parts model index data and any combination of parts assembly schemes correspond to a unique electromechanical product 3D model; S2, establishing an initial hybrid design optimization model, training and testing the initial hybrid design optimization model through a parts database, and obtaining a hybrid design optimization model; S3, obtaining electromechanical product design requirements, obtaining multiple demand component model index data corresponding to the electromechanical product design requirements based on the electromechanical product design requirements and the component assembly requirement analysis model, inputting the multiple demand component model index data into the hybrid design optimization model, and outputting multiple electromechanical product 3D models corresponding to the electromechanical product design requirements; S4, producing a corresponding electromechanical product according to the electromechanical product 3D model.

2. The electromechanical product design method based on deep learning of parts database according to claim 1, characterized in that: Step S2 includes: The initial hybrid design optimization model includes an input layer, a first sub-network layer, a second sub-network layer, a feature fusion layer, an assembly optimization layer, and an output layer. The input layer is connected to the first sub-network layer and the second sub-network layer, respectively. The first sub-network layer and the second sub-network layer are connected to the feature fusion layer, respectively. The feature fusion layer is connected to the assembly optimization layer, and the assembly optimization layer is connected to the output layer. The input layer is used to receive input features; the first sub-network layer is used to extract the timing performance change characteristics of components; the second sub-network layer is used to extract the structural features of components; the feature fusion layer is used to fuse the timing performance change characteristics of components and the structural features of components to obtain fused features; the assembly optimization layer is used to apply multiple component assembly schemes to the fused features to obtain multiple corresponding 3D models of electromechanical products.

3. The electromechanical product design method based on deep learning of parts database according to claim 2, characterized in that: The first sub-network layer adopts a long short-term memory network. The first sub-network layer includes an input gate, a forget gate, a unit state, an output gate, and a hidden state. The characteristic function expression of the component timing performance change of the first sub-network layer is: m r =Sigmoid(α1·[h t-1 ,x t ]+τ1); m y =Sigmoid(α2·[h t-1 ,x t ]+τ2); d t =m y ⊙d t-1 +m r ⊙tanh(α3[h t-1 ,x t ]+τ3); m c =Sigmoid(α4·[h t-1 ,x t ]+τ4); h t =m c ⊙tanh(d t ); F s =h n ; Among them, m r 、m y 、m c are the outputs of the input gate, forget gate, and output gate respectively, d t is the cell state at the current time step, h t is the hidden state of the current time step, h t-1 is the hidden state of the previous time step, h n is the hidden state of the last time step, x t is the input of the current time step, α1, α2, α3, α4 are the weight matrices of the input gate, forget gate, unit state update and output gate, τ1, τ2, τ3, τ4 are the bias vectors of the input gate, forget gate, unit state update and output gate, respectively. s is the input feature of the input layer, ⊙ is the Hadamard product, x n is the input of the last time step, F s is the output of the first sub-network layer.

4. The electromechanical product design method based on deep learning of parts database according to claim 3, characterized in that: The second sub-network layer adopts a convolutional neural network, including a convolutional layer, a sigmoid activation layer and a pooling layer. The component structure characteristic function expression of the second sub-network layer is: Among them, F j is the output of the second sub-network layer, Pooling(·) is the pooling function, exp(·) is the exponential function, T s is the input feature of the input layer, ω1 is the weight matrix of the component structure feature, and δ1 is the bias vector of the component structure feature.

5. The electromechanical product design method based on deep learning of parts database according to claim 4, characterized in that: The feature fusion function expression of the feature fusion layer is: F r =λF s +(1-λ)F j ; Among them, F r is the fusion feature, λ is the fusion weight of the first sub-network layer, F s is the output of the first sub-network layer, F j is the output of the second sub-network layer.

6. The electromechanical product design method based on deep learning of parts database according to claim 5, characterized in that: The assembly optimization layer includes a fully connected layer, and the function expression of the output of the assembly optimization layer is: in, A collection of 3D models of electromechanical products. ys (·) is the assembly optimization feature parameter mapping function, F r For fusion features; f ys (·)Yes|F r | and the mapping function of the 3D model of the electromechanical product, where |F r |Corresponding to multiple 3D models of electromechanical products.

7. The electromechanical product design method based on deep learning of parts database according to claim 1, characterized in that: Step S3 includes: The component assembly demand analysis model includes a demand input layer, a demand analysis layer, a component model layer and a component data output layer, wherein the demand input layer is connected to the demand analysis layer, the demand analysis layer is connected to the component model layer, and the component model layer is connected to the component data output layer; The demand input layer is used to input the design requirements of electromechanical products, and the design requirements of electromechanical products include the size, functional requirements and service life of electromechanical products; The demand analysis layer is used to determine the types of components required for the electromechanical product design according to the electromechanical product design requirements; The component model layer is used to search for multiple required component model index data of the required components from the component database according to the required component type; The component data output layer connection is used to output the plurality of required component model indicator data.

8. The electromechanical product design method based on deep learning of parts database according to claim 1, characterized in that: Step S1 includes: The component model index data includes component geometric properties, component material properties and component timing performance change properties, and the component assembly plan includes assembly sequence and combination method.

9. The electromechanical product design method based on deep learning of parts database according to claim 1, characterized in that: Step S4 includes: According to the 3D model of the electromechanical product, the corresponding rapid manufacturing technology is selected, the manufacturing process flow is formulated, the corresponding electromechanical product is produced, and appearance inspection and performance testing are carried out. The rapid manufacturing technology includes 3D printing, CNC machining, numerical control machining and injection molding.

10. A mechatronic product design system based on deep learning of parts database, characterized in that: The system comprises: A database establishment module is used to establish a parts database, wherein the parts database includes a plurality of parts model index data, a plurality of parts assembly schemes and a plurality of electromechanical product 3D models. When the types and quantities of parts are determined, all the parts model index data and any combination of parts assembly schemes correspond to a unique electromechanical product 3D model; A hybrid design optimization module is used to establish an initial hybrid design optimization model, train and test the initial hybrid design optimization model through a parts database, and obtain a hybrid design optimization model; A product demand analysis module is used to obtain electromechanical product design requirements, obtain multiple demand component model index data corresponding to the electromechanical product design requirements based on the electromechanical product design requirements and the component assembly demand analysis model, input the multiple demand component model index data into the hybrid design optimization model, and output multiple electromechanical product 3D models corresponding to the electromechanical product design requirements; The electromechanical product generation module is used to produce the corresponding electromechanical product according to the electromechanical product 3D model.

Images