A priori design method for adjustable properties of lattice materials
Through deep learning and hierarchical progressive strategies combined with finite element simulation and neural network training, a heterogeneous topological association of lattice materials is constructed, which solves the problem that traditional design methods are difficult to achieve controllable performance of lattice materials, and realizes lattice-oriented lattice Controllable design of material properties to meet different performance needs.
Patent Information
- Application Number
- CN202410682071.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-05-29
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2044-05-29
AI Technical Summary
Traditional lattice material design methods are difficult to achieve adjustable design for lattice material performance and cannot effectively meet different performance needs.
Deep learning and hierarchical progressive strategies are adopted to construct heterogeneous topological association and optimization algorithms of lattice cell cells, combined with finite element simulation and neural network training, controllable prior design of lattice material performance is achieved.
It realizes a controllable design of lattice material performance, can meet different performance requirements and design requirements, replaces the traditional spatial topology combination method of different curved surfaces of the rod, and improves the controllability of the design.
Smart Images

Figure CN118645186B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the field of lattice material design in mechanical structures, and in particular to a method for a priori design of lattice material performance controllable. Background Art
[0002] As a new type of artificially designed microscopic structure, the lattice structure has better performance than traditional materials. Because of its excellent specific stiffness, specific strength and energy absorption characteristics, it is widely used in industry, life, emerging industries, aerospace and other fields.
[0003] In the design of lattice material structures, traditional design methods are mostly a posteriori design, which generally relies on the spatial topological combination of different rod surfaces to further a posteriori lattice material performance. This method is difficult to achieve adjustable design for the performance requirements of lattice materials. To address this problem, a method for adjustable a priori design of lattice material performance is solved using deep learning and a hierarchical progressive strategy. Summary of the invention
[0004] The technical problem to be solved by the present invention is to provide a method for a priori design of adjustable performance of lattice materials, which takes the expected adjustable performance target of lattice cells as the design starting point, and uses the heterogeneous topological association of cells with differentiated dominant performance to construct the lattice material design space and its performance adjustability; further constructs the effectiveness of the optimal design space and lattice material performance through hierarchical progressive strategic design; and finally constructs a controllable a priori design of lattice material performance based on deep learning analysis of the mapping mechanism between the lattice material design space and its performance.
[0005] In order to solve the above technical problems, the present invention provides a method for a priori design of adjustable properties of lattice materials, comprising the following steps:
[0006] Step 1, selecting multiple types of basic lattice structures with different mechanical behaviors according to the stretching-dominated and bending-dominated characteristics of the lattice structure;
[0007] Step 2: Use Abaqus to perform parametric modeling design on the selected basic lattice structure;
[0008] Lattice placement process: construct a 2×2×2 two-layer lattice structure finite element model, arbitrarily select the basic lattice structure in step 1 and place it at eight positions in the two-layer lattice structure finite element model to obtain a two-layer lattice structure, and record the array formed by arranging the basic lattice structures at the eight positions in the order of type;
[0009] Establishment of the output index corresponding to the array: A uniaxial compression model is constructed based on the two-layer lattice structure finite element model, and numerical simulation is performed to obtain the change curve of force and displacement corresponding to the current array. The maximum bearing capacity in the change curve is used as the first output index of the current array, and the change curve is integrated, and the integration result is used as the second output index of the current array;
[0010] Step 3, repeat the dot placement process in step 2 to obtain an array set. Assume that there are N1 arrays in the obtained array set, preferably N1 is greater than or equal to N / 5 and less than or equal to N / 2, N is the total array set containing all dot placement possibilities, and take the N1 arrays as input, and the first output index and the second output index corresponding to each array to construct a sample set;
[0011] Step 4: Build a deep neural network model, set the number of hidden layers and neurons, neural network layer activation function, learning rate, normalization method and neural network training algorithm, neural network training and single training call sample size; and use the sample set constructed in step 3 to train the built deep neural network model;
[0012] Step 5: Build an optimization model, and use the optimization model to optimize and solve the trained deep neural network model to obtain the optimal target unit that meets the design requirements;
[0013] Step 6: Based on the idea of hierarchical progressive strategy, the preferred target unit obtained in step 5 is regarded as the basic lattice structure in step 1, and steps 2 to 5 are repeated to stack and construct a multi-layer stacked lattice structure, and the preferred target unit corresponding to the multi-layer stacked lattice structure is obtained;
[0014] The number of stacking times is determined according to the target number of layers, and the preferred target unit corresponding to the multi-layer stacked lattice structure obtained by the last stacking is used as the optimal lattice structure model to complete the controllable a priori design of the lattice material performance.
[0015] Furthermore, in step 1, the selection is based on the Maxwell formula in three-dimensional space:
[0016] M=s-3n+6
[0017] Where M is the Maxwell number, s is the number of pillars in the cell, and n is the number of nodes in the cell. When M is less than 0, it is a basic lattice structure dominated by bending. When M is greater than or equal to 0, it is a basic lattice structure dominated by stretching.
[0018] Furthermore, in step 2, a two-layer lattice structure finite element model is first constructed, parameters of the lattice structure are set, including length, cross-sectional shape and cross-sectional size, and physical properties such as Young's modulus and density are assigned;
[0019] The finite element model of the lattice structure is simplified in the form of beam elements, and the beam section direction is specified to be perpendicular to the beam direction.
[0020] Furthermore, among the basic lattice structures placed at eight positions, a coordinate system is established with a corner point of one of the basic lattice structures as the origin, and the three sides connected to this corner point are the positive directions of the X, Y, and Z axes respectively. The second basic lattice structure is closely placed in the positive x direction of the first basic lattice structure, the third basic lattice structure is closely placed in the positive y direction of the second basic lattice structure, the fourth basic lattice structure is closely placed in the positive y direction of the first lattice structure, the fifth basic lattice structure is closely placed in the positive z direction of the first basic lattice structure, the sixth basic lattice structure is closely placed in the positive x direction of the fifth basic lattice structure, the seventh basic lattice structure is closely placed in the positive y direction of the sixth basic lattice structure, and the eighth basic lattice structure is closely placed in the positive y direction of the fifth basic lattice structure, finally forming a two-layer lattice structure of 2×2×2.
[0021] Furthermore, the models of the pressure plate and the load-bearing plate are constructed, and the length, width and thickness parameters of the pressure plate and the load-bearing plate are set and set as rigid bodies; the two-layer lattice structure finite element model and the rigid body model are assembled to construct a uniaxial compression model; and the contact and contact properties between the pressure plate, the load-bearing plate and the two-layer lattice structure finite element model in the rigid body model are set;
[0022] According to the compression requirements, the model boundary conditions are set: including setting the load-bearing plate to be completely fixed and giving the movement displacement of the pressure plate; meshing the uniaxial compression model and performing numerical simulation.
[0023] Furthermore, in step 3, the sample set is selected by using Python instructions to select a certain number of arrays from the array set as inputs of the sample set, and the number of arrays selected is less than one-half of the number of arrays in the array set;
[0024] Production of sample set data: Take the selected array as input, use Python instructions, generate the corresponding model file according to the type of basic lattice structure at each position in the two-layer lattice structure of 2×2×2, and submit the finite element analysis model file to Abaqus for batch finite element simulation calculation;
[0025] Extract and organize data to form a sample set: Use Python instructions to extract data from the finite element simulation calculation results in step 3.2, and store each input and its corresponding two outputs in the same file to form a sample set.
[0026] Furthermore, in step 4, the sample set is divided into a training set and a test set according to 90% and 10%, and the average of the sum of squared errors is used as the error evaluation indicator of deep learning; when the accuracy of the sample set is insufficient, the number of arrays in the array set is increased and step 4 is repeated.
[0027] Furthermore, by adjusting the number of hidden layers and neurons, neural network layer activation function, learning rate, normalization method and neural network training algorithm, neural network training and single training call sample size, the error of the average value is reduced to within the standard, and the trained neural network is saved.
[0028] Furthermore, the optimizer in the optimization model is the SGD optimizer, the optimization function is the trained deep neural network model, the optimization variable is the type of the basic lattice structure at eight positions in the two-layer lattice structure, the optimization target is the maximum output of the neural network within the value range, and the optimization constraint is the sum of the masses of the basic lattice structures at eight positions in the two-layer lattice structure. The initial point of the optimization variable, the optimizer learning rate and the optimization training rounds are set.
[0029] Furthermore, based on the optimization model, the optimal lattice structure is solved according to different design requirements; the design requirements include optimal maximum bearing capacity, optimal cushioning and shock absorption, and two performance indicators achieving overall optimality according to their respective weight requirements.
[0030] Beneficial effects of the present invention:
[0031] The traditional design method of spatial topological combination of different rod surfaces is replaced by the heterogeneous topological association of cells with differentiated dominant performance. This can largely solve the problem that traditional design methods are difficult to achieve controllable a priori design. By combining deep learning and optimization algorithms into the design process, it can achieve controllable design for different performance requirements of lattice materials and different requirements for the same performance. BRIEF DESCRIPTION OF THE DRAWINGS
[0032] Figure 1 It is a schematic diagram of the process of the method for a priori design of adjustable properties of lattice materials of the present invention;
[0033] Figure 2 It is a dot matrix A model diagram of the present invention;
[0034] Figure 3 It is a dot matrix B model diagram of the present invention;
[0035] Figure 4 It is a dot matrix C model diagram of the present invention;
[0036] Figure 5 It is a uniaxial compression model diagram of the present invention;
[0037] Figure 6This is the dot matrix A model diagram of the present invention with the best bearing capacity;
[0038] Figure 7 The present invention Figure 6 Force-displacement diagram of the midpoint matrix A;
[0039] Figure 8 It is the most suitable dot matrix B model diagram for a given performance value of the present invention. DETAILED DESCRIPTION
[0040] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments so that those skilled in the art can better understand the present invention and implement it, but the embodiments are not intended to limit the present invention.
[0041] Reference Figure 1 As shown, an embodiment of the method for prior design of adjustable lattice material properties of the present invention is specifically a method for prior design of adjustable lattice material properties solved by deep learning and hierarchical progressive strategy, including the following contents:
[0042] First, according to the stretch-dominant and bending-dominant characteristics of the lattice structure, basic lattice structures with different mechanical behaviors are selected respectively. The selection and judgment basis is: Maxwell formula in three-dimensional space:
[0043] M=s-3n+6
[0044] Where M is the Maxwell number, s is the number of pillars in the cell, and n is the number of nodes in the cell. When M is less than 0, it is a bending-dominated lattice structure, and when M is greater than or equal to 0, it is a stretching-dominated lattice structure.
[0045] Select basic lattice structures with different mechanical behaviors, such as Figures 2 to 4 As shown, lattice A, lattice B, and lattice C are selected as three types of basic lattice structures.
[0046] Then Abaqus is used to perform parametric modeling design on the basic lattice structure selected above; the details include the following:
[0047] Build a lattice structure finite element model, set the parameters of the lattice structure finite element model, including side length of 8, circular cross-section shape, 0.5mm cross-section radius, Young's modulus of 1987.68MPa and density of 1.1E-09t / mm2; then simplify the lattice structure model in the form of beam elements, and specify that the beam section direction is perpendicular to the beam direction; use Python to calculate the vertical vector of each beam in the lattice structure finite element model and assign the beam section direction;
[0048] Build the rigid body model of the pressure plate and the load-bearing plate, set their length, width, thickness and other parameters and set them as rigid bodies. Figure 5 As shown, RP-1 and RP-2 are the rigid reference points of the pressure plate and the load-bearing plate respectively;
[0049] Then, according to lattice A, lattice B, and lattice C, eight structures are selected in any combination, such as lattice A, lattice A, lattice A, A, lattice A, lattice A lattice B, and lattice B. A corner point of one of the basic lattice structures is taken as the origin, and the three sides connected to this corner point are the positive directions of the X, Y, and Z axes, respectively. A coordinate system is established, and the second basic lattice structure is closely placed in the positive x direction of the first basic lattice structure, the third basic lattice structure is closely placed in the positive y direction of the second, and the fourth basic lattice structure is closely placed in the positive y direction of the first basic lattice structure. The fifth basic lattice structure is closely placed in the positive z direction of the first basic lattice structure, the sixth basic lattice structure is closely placed in the positive x direction of the fifth basic lattice structure, the seventh basic lattice structure is closely placed in the positive y direction of the sixth basic lattice structure, and the eighth basic lattice structure is closely placed in the positive y direction of the fifth basic lattice structure. The lattice placement process is completed, and finally a 2×2×2 two-layer lattice structure finite element model is formed. Arrange the eight basic lattice structure types in order to form an array as input, such as Figure 5 As shown, the input at this time is [1, 1, 1, 1, 1, 2, 2];
[0050] The above-mentioned 2×2×2 two-layer lattice structure finite element model and rigid body model are assembled to construct a uniaxial compression model; and the contact and contact properties between the pressure plate and the bearing plate and the two-layer lattice structure finite element model are set; the contact between the two plates and the two-layer lattice structure finite element model is set to "surface-to-surface" contact, the normal contact property adopts "hard" contact, the tangential contact property adopts "penalty", and the friction coefficient is assigned to 0.3. According to the working conditions, the boundary conditions of the uniaxial compression model are set, including setting the bearing plate to be completely fixed, setting the pressure plate to a fixed displacement of 8mm and preventing torsion, limiting the degrees of freedom of the four sides of the bottom surface, the lower plate is the bearing plate, and the upper plate is the pressure plate, such as Figure 5 shown.
[0051] According to the uniaxial compression model and accuracy requirements, the uniaxial compression model is meshed and numerically simulated to obtain the force and displacement change curves. The maximum bearing capacity in the curve is taken as the first output indicator, and the pressure exerted by the pressure plate on the two-layer lattice structure and the pressure plate displacement change curve are integrated as the performance indicator of buffering and shock absorption, and the integration result is taken as the second output indicator.
[0052] Finally, an array and the first output indicator and the second output indicator corresponding to the array are obtained.
[0053] Based on the selected dot matrix A, dot matrix B, and dot matrix C, there are 3 to the 8th power of array possibilities, that is, 6561 array inputs. If all arrays are used to construct the sample set, the amount of data will be large, and the meaning of reverse solution will be lost. Therefore, the dot matrix placement process (the process of randomly selecting combinations) is repeated to obtain an array set. Assume that there are N1 arrays in the obtained array set, preferably N1 is greater than or equal to N / 5 and less than or equal to N / 2, and N is the total array set containing all dot matrix placement possibilities (that is, a set of 6561 arrays). Take N1 arrays as input, and the first output index and the second output index corresponding to each array to construct the sample set, specifically:
[0054] First, for the selection of sample sets, the dot placement process is repeated 1000 times through Python instructions in 6561 possible array inputs to obtain 1000 arrays (which can be adjusted according to the calculation results). The 1000 arrays are used as the input of the sample set, effectively reducing the amount of calculation;
[0055] The sample set data is prepared by using Python instructions to input the sample set, generate the corresponding model file according to the type of basic lattice structure selected at each position in the two-layer lattice structure, and submit the finite element analysis model file to Abaqus to realize batch finite element simulation calculation;
[0056] Then, the data are extracted and organized to form a sample set. Through Python instructions, data is extracted from the finite element simulation results. The maximum bearing capacity corresponding to each input is used as the first output indicator. The pressure exerted by the pressure plate on the two-layer lattice structure and the change curve of the pressure plate displacement are integrated, and the integration result is used as the second output indicator. Each of the above inputs and its corresponding two outputs are stored in the same file as a sample set. The construction of the sample set is completed.
[0057] Then build a deep neural network model, set the number of hidden layers and neurons, neural network layer activation function, learning rate, normalization method and neural network training algorithm, neural network training and single training call sample size; use the sample set to train the built neural network to obtain a trained deep neural network model; specifically: divide the sample set into training set and test set according to 90% and 10%, and use the mean square error (MSE) as the error evaluation index of deep learning. At this time, if it is found that the sample set accuracy is insufficient, add repeated dot placement process to obtain 1500 arrays, and retrain with the sample set formed by 1500 arrays. By adjusting the number of hidden layers and neurons, neural network layer activation function, learning rate, normalization method and neural network training algorithm, neural network training and single training call sample size, the MSE error is reduced to an appropriate standard, and the trained neural network is saved.
[0058] Then, an optimization model is constructed, including: the optimizer is the SGD optimizer, the optimization function is the trained neural network, the optimization variables are the basic lattice types at eight positions in a two-layer lattice structure finite element model of 2×2×2, the optimization objective is to find the maximum and minimum values of the neural network output within the value range, the optimization constraint is the total mass of the basic lattice types at eight positions, and the initial point of the optimization variable, the learning rate of the optimizer, and the number of optimization training rounds are set.
[0059] Based on the component optimization model, according to different lattice structure design requirements: the maximum bearing capacity is optimal, the buffering and shock absorption is optimal, and the two performance indicators reach the overall optimum according to their respective weight requirements; the optimal lattice structures are obtained by solving respectively, and the optimization principle is:
[0060] x = argmax x (λ1f(x) + λ2g(x))
[0061] where x is the type of the basic lattice structure at eight positions, f(x) is the neural network trained based on the maximum bearing capacity performance data, g(x) is the neural network trained based on the buffering and shock absorption performance data, and λ1 and λ2 are the relative importance degrees of the two performance indicators respectively. When λ1 is 1 and λ2 is 0, it means that when only considering the optimal maximum bearing capacity performance, the value of x obtained by solving under the mass constraint condition; when λ1 is 0 and λ2 is 1, it means that when only considering the optimal buffering and shock absorption performance, the value of x obtained by solving under the mass constraint condition; when λ1 is m (0 < m < 1) and λ2 is (1 - m), it means that the obtained basic lattice structure x makes both the maximum bearing capacity f(x) and the buffering and shock absorption g(x) reach the relatively optimal values under the condition of meeting the constraint conditions. As Figure 6 and Figure 7 shown, it means that when λ1 is 1 and λ2 is 0, only considering the optimal maximum bearing capacity performance, the value result of x obtained by solving under the mass constraint condition, and at this time x is [1, 1, 1, 1, 1, 1, 2, 2];
[0062] When the lattice structure performance design requirement is a specific value rather than the optimal value, through the optimization model combined with the theory of inverse solution neural network, the lattice structure combination that best meets the requirements can be obtained, and the theoretical principle is:
[0063] x = argmin x |f(x) - y|
[0064] Among them, x is the type of the basic lattice structure at eight positions, y is the performance index corresponding to the basic lattice structure at eight positions, and f(x) is the neural network trained based on the maximum bearing capacity performance data and the cushioning and shock absorption performance data. When the absolute value of the error between the output value of the neural network and the given value is the smallest, x is the input value corresponding to y, such as Figure 8 As shown in the figure, when the given value of the maximum bearing capacity is 600N, the optimization result of x is [3, 3, 3, 2, 3, 2, 2, 2], and the value of f(x) is 595.65N.
[0065] Based on the idea of layered progressive strategy, the combination of the two-layer lattice structure of 2×2×2 that best meets the requirements after the above optimization is taken as the optimal target unit, and the optimal target unit is regarded as the basic lattice structure. The basic lattice structure is further stacked to form a 4×4×4 four-layer lattice structure finite element model. That is, a multi-layer stacked lattice structure is constructed, and the preferred target unit corresponding to the four-layer lattice structure finite element model is obtained. The preferred target unit is used as the optimal lattice structure model, and finally the controllable a priori design of lattice material performance is completed.
[0066] Based on the above method, if the target number of layers is an eight-layer lattice structure finite element model of 8×8×8, the preferred target unit corresponding to the four-layer lattice structure finite element model will continue to be used as the basic lattice structure, and then stacked and calculated again, and finally the preferred target unit corresponding to the eight-layer lattice structure finite element model will be used as the optimal lattice structure model.
[0067] The above method is mainly based on the design requirements, first verifying the performance of the lattice structure, based on the basic lattice structure with performance differences, and adjusting the different types of basic lattice structures at various positions to achieve the adjustable design of the overall lattice structure performance, thereby obtaining the optimal lattice structure that meets the design requirements.
[0068] This application takes the expected adjustable performance target of lattice materials as the design starting point, and constructs the lattice material design space and its performance adjustability based on the heterogeneous topological association of the cells with differentiated dominant performance; further constructs the effectiveness of the optimal design space and the lattice material performance through hierarchical progressive strategic design; and introduces optimization algorithms combined with deep learning to analyze the mapping mechanism between the lattice material design space and its performance, and finally constructs a controllable a priori design of lattice material performance.
[0069] The above-described embodiments are only preferred embodiments for fully illustrating the present invention, and the protection scope of the present invention is not limited thereto. Equivalent substitutions or changes made by those skilled in the art based on the present invention are within the protection scope of the present invention. The protection scope of the present invention shall be subject to the claims.
Claims
1. A method for designing lattice material properties with adjustable a priori features, characterized in that: The following steps are involved: Step 1, selecting multiple types of basic lattice structures with different mechanical behaviors according to the stretching-dominated and bending-dominated characteristics of the lattice structure; Step 2: Use Abaqus to perform parametric modeling design on the selected basic lattice structure; Lattice placement process: construct a 2×2×2 two-layer lattice structure finite element model, arbitrarily select the basic lattice structure in step 1 and place it at eight positions in the two-layer lattice structure finite element model to obtain a two-layer lattice structure, and record the array formed by arranging the basic lattice structures at the eight positions in the order of type; Establishment of the output index corresponding to the array: A uniaxial compression model is constructed based on the two-layer lattice structure finite element model, and numerical simulation is performed to obtain the change curve of force and displacement corresponding to the current array. The maximum bearing capacity in the change curve is used as the first output index of the current array, and the change curve is integrated, and the integration result is used as the second output index of the current array; Step 3, repeat the dot placement process in step 2 to obtain an array set. Assume that there are N1 arrays in the obtained array set, N1 is greater than or equal to N / 5 and less than or equal to N / 2, N is the total array set containing all dot placement possibilities, take the N1 arrays as input, and construct a sample set with the first output index and the second output index corresponding to each array; Step 4: Build a deep neural network model, set the number of hidden layers and neurons, neural network layer activation function, learning rate, normalization method and neural network training algorithm, neural network training and single training call sample size; And use the sample set constructed in step 3 to train the built deep neural network model; Step 5: Build an optimization model and use it to optimize and solve the trained deep neural network model to obtain the target unit that meets the design requirements; Step 6: Based on the idea of hierarchical progressive strategy, the target unit obtained in step 5 is regarded as the basic lattice structure in step 1, and steps 2 to 5 are repeated to stack and construct a multi-layer stacked lattice structure, and the target unit corresponding to the multi-layer stacked lattice structure is obtained; Determine the stacking times according to the target number of layers, take the target unit corresponding to the multi-layer stacked lattice structure obtained by the last stacking as the optimal lattice structure model, and complete the controllable a priori design of the lattice material performance; In step 1, the selection is based on the Maxwell formula in three-dimensional space: ; Where M is the Maxwell number, s is the number of pillars in the cell, and n is the number of nodes in the cell. When M is less than 0, it is a basic lattice structure dominated by bending. When M is greater than or equal to 0, it is a basic lattice structure dominated by stretching. In step 2, a two-layer lattice structure finite element model is first built, the parameters of the lattice structure are set, including length, cross-sectional shape and cross-sectional size, and the physical properties of Young's modulus and density are assigned; The finite element model of the lattice structure is simplified in the form of beam elements, and the beam section direction is specified to be perpendicular to the beam direction; Among the basic lattice structures placed at eight positions, a corner point of one of the basic lattice structures is taken as the origin, and the three sides connected to the corner point are respectively the positive directions of the X, Y, and Z axes, to establish a coordinate system, the second basic lattice structure is closely placed in the positive x direction of the first basic lattice structure, the third basic lattice structure is closely placed in the positive y direction of the second basic lattice structure, the fourth basic lattice structure is closely placed in the positive y direction of the first lattice structure, the fifth basic lattice structure is closely placed in the positive z direction of the first basic lattice structure, the sixth basic lattice structure is closely placed in the positive x direction of the fifth basic lattice structure, the seventh basic lattice structure is closely placed in the positive y direction of the sixth basic lattice structure, and the eighth basic lattice structure is closely placed in the positive y direction of the fifth basic lattice structure, finally forming a two-layer lattice structure of 2×2×2; Build models of the pressure plate and the load-bearing plate, set the length, width and thickness parameters of the pressure plate and the load-bearing plate and set them as rigid bodies; assemble the two-layer lattice structure finite element model and the rigid body model to construct a uniaxial compression model; And set the contact and contact properties between the pressure plate, the load-bearing plate and the two-layer lattice structure finite element model in the rigid body model; According to the compression requirements, the model boundary conditions are set: including setting the load-bearing plate to be completely fixed and giving the movement displacement of the pressure plate; meshing the uniaxial compression model and performing numerical simulation.
2. The method for designing lattice material properties with adjustable performance according to claim 1, characterized in that: In step 3, the sample set data is prepared: the selected array is used as input, and Python instructions are used to generate the corresponding model file according to the type of the basic lattice structure at each position in the two-layer lattice structure of 2×2×2, and the finite element analysis model file is submitted to Abaqus for batch finite element simulation calculation; Extract and organize data to form a sample set: Use Python instructions to extract data from finite element simulation calculation results, store each input and its corresponding two outputs in the same file to form a sample set.
3. The method for designing lattice material properties with adjustable performance according to claim 1, characterized in that: In step 4, the sample set is divided into a training set and a test set according to 90% and 10%, and the average of the sum of squared errors is used as the error evaluation indicator of deep learning; when the accuracy of the sample set is insufficient, the number of arrays in the array set is increased and step 4 is repeated.
4. The method for designing lattice material properties with adjustable performance according to claim 3, characterized in that: By adjusting the number of hidden layers and neurons, neural network layer activation function, learning rate, normalization method and neural network training algorithm, neural network training and single training call sample size, the error of the average value is reduced to within the standard, and the trained neural network is saved.
5. The method for designing lattice material properties with adjustable performance according to claim 1, characterized in that: The optimizer in the optimization model is the SGD optimizer, the optimization function is the trained deep neural network model, the optimization variable is the type of the basic lattice structure at eight positions in the two-layer lattice structure, the optimization target is the maximum output of the neural network within the value range, and the optimization constraint is the sum of the masses of the basic lattice structures at eight positions in the two-layer lattice structure. The initial point of the optimization variable, the optimizer learning rate and the optimization training rounds are set.
6. The method for designing lattice material properties with adjustable performance according to claim 5, characterized in that: Based on the optimization model, the optimal lattice structure is solved according to different design requirements; the design requirements include optimal maximum bearing capacity, optimal cushioning and shock absorption, and two performance indicators achieving overall optimality according to their respective weight requirements.
