Complex part molded line optimization design method based on multi-field node feature weighted aggregation

Abstract

The invention discloses a complex part molded line optimization design method based on multi-field node feature weighted aggregation. Comprising the following steps: firstly, performing finite element batch simulation according to a geometric design variable, a temperature and a contact surface friction coefficient of a to-be-optimized complex part molded line to obtain a training data set; then, constructing a depth map sampling aggregation neural network fusing space and temperature attention, then training the depth map sampling aggregation neural network fusing space and temperature attention by using the training data set, and obtaining a stress prediction model after training is completed; and finally, in combination with the stress prediction model, performing iterative optimization by taking minimization of the maximum equivalent stress of the complex part as an optimization target, and obtaining the optimal geometric parameters of the molded line of the current complex part in the target temperature interval. According to the method, rapid reconstruction of the stress distribution state of the complex part is achieved, the optimization speed is increased, meanwhile, the precision of an optimization result is fully guaranteed, and meanwhile calculation resources are greatly saved.

Description

Technical Field

[0001] The invention relates to an optimization design method for complex part profiles, and in particular to an optimization design method for complex part profiles based on weighted aggregation of multi-field node features. Background Art

[0002] In the industrial field, as the core components of industrial products such as mechanical and electronic equipment, aerospace, etc., the performance of complex parts directly affects the reliability, work efficiency and even service life of the entire equipment and system. As the most important structural feature of complex parts, their profile not only determines the shape of the parts themselves, but also directly affects the transmission performance and mechanical properties between parts.

[0003] In addition, complex parts are exposed to high temperature and high pressure for a long time and under the combined action of multiple loads. As the working area of ​​the parts, their profiles are prone to stress concentration due to frequent contact and extrusion. This stress concentration can easily lead to fatigue failure or even fracture of parts, which can lead to extremely serious safety accidents. Therefore, in order to effectively alleviate stress concentration and improve the stability and reliability of complex parts, it is very important to optimize the profile design of complex parts.

[0004] The optimization design of complex part profiles is to find a set of profile design variables that meet the requirements of the objective function under specific constraints, that is, the optimal solution. At the same time, since the input (geometric parameters) and output (stress and strain values) are implicitly expressed, it is impossible to simply obtain the optimal solution based on their gradient information. Therefore, an optimization algorithm is needed to find the optimal solution for the design problem.

[0005] At present, there are two main optimization design methods for complex part lines. One is to combine finite element analysis with optimization algorithms, regard finite element analysis software as a relationship model between design variables and objective functions, and drive finite element analysis software to perform iterative calculations and optimization through commercial optimization software or optimization control programs. With the help of optimization algorithms, the values ​​of design variables are continuously updated until convergence is achieved, thereby obtaining the optimal line parameter design values ​​of complex parts. However, solving the optimization problem of complex part lines involves complex processes such as modeling of the research object, thermo-solid coupling finite element simulation analysis, and post-processing of the results. A large amount of computing resources and time are required to achieve a complete analysis. Therefore, when establishing a complex part line optimization design model, it is often necessary to use approximate models of related indicators to replace the actual simulation process to ensure the smooth progress of the optimization process.

[0006] With the widespread application of data analysis, the line optimization design method based on the combination of surrogate models and optimization algorithms has gradually become the mainstream. That is, firstly, multiple sets of data are calculated through parametric modeling, and then a surrogate model is constructed to establish a mapping relationship from input to output. After that, the optimization is performed through an intelligent optimization algorithm. The commonly used surrogate models are response surface models, Kriging models, support vector regression, and neural networks. With the continuous development of deep learning technology, it has gradually become the mainstream to use various neural networks as surrogate models to achieve predictions from input to output. In the offline stage, samples are input to train the neural network model, and in the online stage, rapid predictions of test samples are achieved, which greatly improves the speed of optimization design and saves system overhead.

[0007] Unlike traditional graph neural networks that need to operate on the features of all graph nodes during model training, the Graph Sampling Aggregation Network (GraphSAGE) can randomly sample information from neighboring nodes by iteratively aggregating the features of neighboring nodes, and nonlinearly process this information with its own features through the aggregation function of each layer to form a new feature vector. In addition, GraphSAGE uses a small batch and multi-round training method to effectively solve the problem of solution failure caused by hardware computing power limitations, and reduces the dependence of traditional graph convolutional neural networks on full graph information. It is particularly suitable for processing large graph structures with a large number of nodes and complex adjacency relationships, so it is very suitable for predicting finite element mesh node features.

[0008] The stress field prediction of complex parts is to predict the stress values ​​of different grid nodes after given design variables (geometric parameters) to reconstruct the stress field. However, only establishing a relationship model from geometric parameters to stress field cannot reflect the adaptability and stability of the prediction results under variable working conditions. For complex parts, the working conditions under actual working conditions will inevitably change. Complex parts are often in a high temperature and high pressure working environment, so temperature, as the main working condition factor affecting its stress field distribution, needs to be included in the consideration of stress results. Therefore, taking into account the temperature field distribution that has a significant impact on the stress field distribution results, restoring the temperature value of each node and then forming a relationship model that considers the influence of multiple fields under variable temperature has important research significance for the optimization design of the profile.

[0009] At the same time, since the nodes of the finite element mesh are interconnected and influence each other, the traditional GraphSAGE algorithm is difficult to fully explore the interaction relationship between nodes and fully characterize them, resulting in unsatisfactory prediction and optimization results. In addition, since the contact areas of complex parts frequently experience friction and extrusion, and areas with large temperature differences are prone to concentration of thermal stress, more attention needs to be paid to the above areas. When building the graph neural network, the characteristics of the nodes in the area should also be fully considered. Otherwise, the accuracy of the relationship model will be affected due to the lack of reasonable allocation of attention to each area of ​​the complex parts, resulting in certain limitations in its optimization process. Summary of the invention

[0010] In order to solve the problems existing in the background technology, the present invention proposes a complex part line optimization design method based on weighted aggregation of multi-field node features. The method proposed in the present invention can overcome the shortcomings of the existing methods, taking into account the influence of the joint action of multiple fields including temperature field and structural fields such as stress field and strain field on stress prediction and optimization results, and realizing weighted aggregation of multi-field node features by introducing spatial attention coefficient and temperature attention coefficient, which can distinguish the importance of grid nodes, and pay higher attention to the grid node information of frequent contact parts and large temperature difference parts. At the same time, the regional maximum temperature and regional minimum temperature are introduced as input items of the model, taking into account the uncertainty brought by temperature to the results, and realizing the rapid prediction of stress and strain field distribution of complex parts in the variable temperature range. At the same time, the improved dung beetle optimization algorithm based on the optimal Latin hypercube is used for the optimization design of complex part line parameters, which enriches the diversity of the population and improves the accuracy of the optimization results.

[0011] To achieve the above object, the technical solution of the present invention is as follows:

[0012] 1. A complex part line optimization design method based on weighted aggregation of multi-field node features

[0013] S1: Perform finite element simulation based on the geometric design variables, temperature and contact surface friction coefficient of the complex part profile to be optimized, obtain the characteristic matrix, adjacency matrix, and equivalent stress and equivalent strain values ​​of each grid node, and then obtain the training data set; among which, the geometric design variables and contact surface friction coefficient constitute the geometric parameters;

[0014] S2: Construct a deep graph sampling aggregation neural network that integrates space and temperature attention, and then use the training data set to train the deep graph sampling aggregation neural network that integrates space and temperature attention. After the training is completed, a stress prediction model is obtained;

[0015] S3: Combined with the stress prediction model, iterative optimization is performed with the optimization goal of minimizing the maximum equivalent stress of complex parts to obtain the optimal geometric parameters of the current complex part profile within the target temperature range.

[0016] In S1, the temperature specifically refers to the maximum temperature and the minimum temperature in the target area.

[0017] In S2, the aggregation function in the deep graph sampling aggregation neural network that integrates spatial and temperature attention satisfies the following formula:

[0018]

[0019] in, is the jth feature formed by the aggregation of the central node e by its i neighboring nodes, (ne) i is the ith neighbor node of the central node e, L_att[(ne) i ] is the spatial attention coefficient, T_att[(ne) i ] is the temperature attention coefficient, W is the adjustable weight parameter, is the jth feature of the i-th neighbor node, softmax[] is the normalized exponential function, is the spatial distance between the ith neighbor node and the central node, is the temperature difference between the i-th neighbor node and the central node e, i = 1,…,n, n is the number of neighbor nodes of the central node e.

[0020] In S3, the network parameters of the stress prediction model are frozen and the stress prediction model is used as the fitness function in the optimization algorithm to obtain an improved optimization algorithm. The improved optimization algorithm iteratively updates the geometric parameters with minimizing the maximum equivalent stress of the complex part as the optimization goal until the optimal geometric parameters of the current complex part profile within the target temperature range are obtained.

[0021] The S3 is specifically:

[0022] Firstly, the initial parameters of the dung beetle algorithm are determined. Then, the optimal Latin hypercube function is used to initialize the distribution of the dung beetle population so that the initial population of dung beetles is evenly distributed in the entire optimization space, thereby obtaining the initialized individual position matrix of dung beetles. Next, the stress prediction model of the frozen network parameters is used as the fitness function of the dung beetle algorithm to obtain the improved dung beetle algorithm. Then, based on the initialized individual position matrix of dung beetles, the improved dung beetle algorithm is used to iteratively update the position of the dung beetle population until the convergence condition is finally met, thereby obtaining the optimal geometric parameters of the current complex part profile within the target temperature range.

[0023] 2. A complex part line optimization design system based on weighted aggregation of multi-field node features

[0024] The data set construction unit is used to perform finite element simulation according to the geometric design variables of the complex part profile to be optimized, the temperature and the contact surface friction coefficient, obtain the characteristic matrix, the adjacency matrix and the equivalent stress value and the equivalent strain value of each grid node, and obtain the training data set after repeated multiple times;

[0025] A stress prediction unit, used for storing a deep graph sampling aggregation neural network integrating spatial and temperature attention and training the deep graph sampling aggregation neural network integrating spatial and temperature attention using a training data set, thereby obtaining a stress prediction model;

[0026] The geometric parameter optimization unit is used to combine the stress prediction model to iteratively optimize the complex parts with the minimization of the equivalent stress as the optimization goal, and obtain the optimal geometric parameters of the current complex part profile within the target temperature range.

[0027] 3. A computer device

[0028] The device includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the complex part profile optimization design method based on weighted aggregation of multi-field node features are implemented.

[0029] 4. A computer-readable storage medium

[0030] The medium stores a computer program, which, when executed by a processor, implements the steps of the complex part profile optimization design method based on weighted aggregation of multi-field node features.

[0031] V. A computer program product

[0032] The product includes a computer program / instruction, which, when executed by a processor, implements the steps of the complex part profile optimization design method based on weighted aggregation of multi-field node features.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] 1) The influence of temperature and contact surface friction conditions on the stress field prediction results of complex parts is taken into account, and the stress field distribution prediction of complex part profile parameters under variable working conditions (temperature range, friction coefficient) is realized, and then the complex part profile optimization design considering multiple fields in the variable temperature range is carried out.

[0035] 2) The spatial attention coefficient and temperature attention coefficient were defined, and the GraphSAGE neural network was defined based on them, which realized the weighted aggregation of multi-field node features. Combined with the mechanism of finite element thermal-solid coupling analysis, the central node paid special attention to the grid nodes in the contact part of complex parts and the grid nodes with large temperature differences when aggregating the information of neighboring nodes, thus realizing the reasonable allocation of attention to the grid nodes, effectively aggregating the parts that have a significant impact on the stress results, and aggregating more feature information for important nodes in the network.

[0036] By considering the thermal-solid coupling mechanism of complex parts and using graph neural networks as proxy models to complete the optimization design, the stress distribution state of complex parts can be quickly reconstructed, which fully guarantees the accuracy of the optimization results while improving the optimization speed, and greatly saves computing resources.

[0037] 3) In order to solve the problem that DBO randomly generates the initial dung beetle population, resulting in uneven population distribution and slowing down the convergence of the algorithm, an improved dung beetle optimization algorithm (OLH-DBO) was proposed based on the optimal Latin hypercube. The distribution quality of the initial population in the search space was improved, and its global search capability was enhanced, making the optimization results more accurate and meeting the needs of solving complex part profile optimization design problems. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 This is the overall diagram of the model process for optimizing the profile parameters of the blade root groove.

[0039] Figure 2 It is a schematic diagram of the blade root groove profile and a schematic diagram of multi-field node characteristics.

[0040] Figure 3 Schematic diagram of the characteristic information of grid nodes.

[0041] Figure 4 Figure 2. The graph sampling aggregation network structure diagram for integrating spatial and temperature attention.

[0042] Figure 5 This is the flow chart of the OLH-DBO algorithm.

[0043] Figure 6 This is a comparison chart of the results before and after blade root groove optimization. DETAILED DESCRIPTION

[0044] The following takes the blade root groove of a compressor as a specific implementation object of the present invention and further describes the present invention:

[0045] The overall process diagram of the blade root groove profile optimization design method is as follows: Figure 1 shown.

[0046] S1: Perform finite element simulation according to the geometric design variables, temperature and contact surface friction coefficient of the blade root wheel groove profile to be optimized, that is, after parameterized batch generation, obtain the feature matrix, adjacency matrix and equivalent stress value and equivalent strain value of each grid node, wherein the feature matrix and adjacency matrix are used as the input of the graph neural network, and the equivalent stress value and equivalent strain value of each grid node are used as labels, thereby obtaining the training data set; wherein the geometric design variables and the contact surface friction coefficient constitute the geometric parameters;

[0047] S1 is specifically:

[0048] The research object is simplified, the influence of axial stress is ignored, and it is regarded as a stress analysis problem in a two-dimensional plane. The input of the data set of blade root groove geometric parameters-stress and strain consists of the outer arc R of the blade root groove Ex / mm, inner arc of blade root groove R In / mm, blade root groove contact section tooth width D / mm, blade root groove contact section tooth inclination angle α / °, blade root bottom width L / mm, and the maximum regional temperature t of the blade root groove max / ℃, minimum temperature of the area T min / ℃ and the contact surface friction coefficient μ, a total of 8 design variables, among which the maximum temperature and the minimum temperature are set because the temperature distribution state of the blade root groove is that the temperature gradually increases from the center of the bottom of the groove to the top boundary of the blade root. By determining the temperatures of these two parts, the temperature field distribution of the entire research object (blade root groove) can be determined; the equivalent stress σ (thermal stress and contact stress) and the equivalent strain ε (thermal strain and elastic strain) in the entire area of ​​the blade root groove are output. The value range of the design variables is shown in Table 1:

[0049] Table 1 is the range of design variables

[0050]

[0051] Table 2 is the range of constraint variables

[0052]

[0053] The learning task of the model is to establish a mapping relationship y=f(x) from x to y, where x is the input geometric parameter set and y is the output stress field and strain field distribution. is the stress field information, is the strain field information:

[0054] x={x1,x2,x3,x4,x5,x6,x7,x8}

[0055]

[0056] y=f(x)

[0057] The blade root wheel groove profile diagram and multi-field node feature diagram described based on the above geometric parameters and constraints are as follows Figure 2 shown.

[0058] Then, based on VBA and python scripts, the parametric design module of Ansys Workbench is automatically called to perform thermo-solid coupling finite element simulation calculations of blade root and wheel groove, and 1,000 sets of calculation results are automatically generated in batches, each of which contains about 2,000 mesh nodes. Among them, all mesh nodes use triangular meshes, and the mesh size is refined by local encryption for the contact part of the blade root and wheel groove. In addition, after the thermo-solid coupling analysis and calculation is completed, the coordinate information [x, y] and temperature field information T of all mesh nodes in the area need to be output, and the horizontal coordinate x, vertical coordinate y and node temperature T are used as input items of each graph convolution layer. Since the neural network used for stress and strain field prediction is the graph sampling aggregation neural network GraphSAGE, it is also necessary to construct the adjacency matrix of the mesh nodes to characterize the adjacency relationship between nodes.

[0059] The adjacency matrix is ​​a square matrix with M rows and M columns, denoted by A M×M Each node in the graph network is a mesh node divided during finite element calculation, and the edge information in the graph network indicates whether the two end nodes are connected. Based on this, the adjacency matrix of the mesh node is defined as follows:

[0060]

[0061] Among them, A i,j is the adjacency coefficient between the i-th node and the j-th node in the adjacency matrix.

[0062] Then, the 1000 sample sets were divided into training set, validation set and test set in the ratio of 7:1.5:1.5, with a total of 700 training sets, 150 validation sets and 150 test sets.

[0063] S2: Construct a deep graph sampling aggregation neural network (Latt&Tatt-SAGEConv) that integrates spatial and temperature attention. Then use the training data set to train the deep graph sampling aggregation neural network that integrates spatial and temperature attention. After the training is completed, the stress prediction model is obtained;

[0064] In GraphSAGE, taking the structure of triangular grid unit nodes as an example, each node is connected to four adjacent nodes. During the aggregation process, the central node aggregates the feature information of the four neighboring nodes and splices it with the original node information to form its own new node feature expression. The node feature information is as follows: Figure 3As shown in the figure, after updating the node information, it is used for downstream node feature prediction. The aggregation function of GraphSAGE can be expressed as:

[0065]

[0066] Among them, f (feature) is the feature of the node; F is the node feature after aggregation; e (element) is the node; ne (neighbor element) is the neighbor node; i is the number of neighbor nodes included in each aggregation; j is the number of features of the node / neighbor node; Agg[] is the aggregation function.

[0067] Different from the existing traditional aggregation methods such as mean aggregation and max-pooling aggregation, the aggregation function in the GraphSAGE graph network is optimized by introducing two adjustment coefficients, spatial attention weight and temperature attention weight, based on the spatial attention and temperature attention of grid nodes.

[0068] For spatial attention, special consideration is given to the mesh encryption nodes at the contact part between the blade root and the wheel groove, because they are in the frequent contact area between the two parts and are also the key area where the maximum equivalent stress may appear. The weight distribution of different neighbor nodes to the central node during the aggregation process is shown in Table 3.

[0069] Table 3 is the aggregation process parameter table based on spatial attention weights

[0070]

[0071] For spatial attention, the aggregation process is defined as follows:

[0072]

[0073] Among them, L_att[(ne) i ] is the spatial attention weight, which ranges from (0,1) and is defined as follows:

[0074]

[0075] In the formula The definition is as follows:

[0076]

[0077] Among them, e represents the central node, (ne) i represents the i-th neighbor node of e; x e ,y e Respectively represent the horizontal and vertical coordinates of the central node e, They respectively represent the horizontal and vertical coordinates of the i-th neighbor node of the central node e. is the i-th neighbor node (ne) i The j-th feature of , softmax[] is a normalized exponential function that aims to smooth the weights and amplify the importance of important neighbor nodes.

[0078] By defining the spatial attention coefficient, nodes with smaller grid node spacing can be given more weight each time they are aggregated, thereby gaining more attention. After multiple aggregation operations, the attention trend of the weights continues to develop toward the refined grid nodes, thereby aggregating more feature information of the blade root groove contact area.

[0079] For temperature attention, special consideration is given to the grid nodes with large temperature difference between the blade root and the wheel groove. According to the calculation formula of thermal stress, it can be seen that the temperature difference and thermal stress are positively correlated, that is, the larger the temperature difference of the node, the greater the probability of the maximum value of the equivalent stress. This area also requires special attention. The weight distribution of different neighbor nodes to the central node during the aggregation process is shown in Table 4.

[0080] Table 4 is a table of aggregation process parameters based on temperature attention weight

[0081]

[0082] Therefore, with regard to temperature attention, the polymerization process is defined as follows:

[0083]

[0084] Among them, T_att[(ne) i ] is the temperature attention weight, which ranges from (0,1) and is defined as follows:

[0085]

[0086] In the formula The definition is as follows:

[0087]

[0088] Among them, e represents the central node, (ne) i represents the i-th neighbor node of e; T e represents the temperature of the central node e, represents the temperature of the i-th neighbor node of the central node e, and || represents the absolute value.

[0089] By defining the temperature attention coefficient, the nodes with larger temperature difference can receive more attention during the aggregation process. After multiple aggregation operations, the attention trend of the weights will continue to develop toward the grid nodes with larger temperature gradients, thereby aggregating more feature information of locations prone to thermal stress concentration.

[0090] The aggregation process considering spatial attention and temperature attention is weighted to obtain the final aggregation function:

[0091]

[0092] Among them, W is an adjustable weight parameter, and its value range is (0,1).

[0093] The deep GraphSAGE neural network integrating spatial and temperature attention (Latt&Tatt-GraphSAGE) consists of an input layer, a multi-layer Latt&Tatt-SAGEConv graph convolution layer, and an output layer. The graph convolution module can be expressed as:

[0094]

[0095] The graph convolution module of each layer includes the aggregation function Latt&Tatt-Aggregate() and the concatenation function CONCAT(), and uses LeakyReLU as the activation function to increase nonlinear excitation and improve the efficiency of fitting. K is a trainable weight matrix. is the feature after neighbor nodes are aggregated, f e (j) is the central node feature. Compared with the ReLU function, it solves the problem of neuron death in the negative half of the function, and has the ability of the ReLU function to overcome the gradient vanishing and gradient exploding problems. At the same time, a dropout layer is added before the last output layer to avoid the occurrence of neural network overfitting. The dropout rate is set to 0.25. The structure of the network is as follows: Figure 4 shown.

[0096] After normalization, the data set is divided into training set, validation set and test set, and the network model is trained and tested.

[0097] When training GraphSAGE, not only the feature matrix but also the adjacency matrix needs to be input to fully express the correlation between nodes.

[0098] The initial parameters of the Latt&Tatt-GraphSAGE neural network are set as follows: epoch is 1000, batch size batch_size is 64, and initial learning rate is 0.001.

[0099] The stress and strain prediction in the present invention is a supervised learning task. In each training, the model takes the node feature graph and the adjacency matrix as input, performs forward propagation calculation based on the current network parameters, and captures the information flow between nodes through the propagation of multiple graph convolution layers. Then, the loss function is calculated to quantify the error between the output result and the label value, and the gradient of the deviation value relative to the model parameter is calculated by continuous back propagation, and the Adam optimizer is used to update the network parameters using these gradient information. By continuously iterating to reduce the loss function, the prediction model gradually learns how to predict unknown labels based on node features and the structure of the graph network.

[0100] In order to evaluate the prediction performance of the Latt&Tatt-GraphSAGE neural network model, the true value y of L test samples is evaluated i With the predicted value There are three indicators of error between them, including:

[0101] Root Mean Square Error (RMSE):

[0102]

[0103] The root mean square error refers to the square root of the ratio of the sum of the squares of the deviations between the predicted value and the true value to the total number of test samples. It is used to measure the difference between the predicted value and the true value. In theory, its value is a positive number, and the closer it is to 0, the better the prediction effect of the model.

[0104] Mean Absolute Error (MAE):

[0105]

[0106] The mean absolute error is obtained by calculating the ratio of the absolute value of the deviation between the predicted value and the true value to the total number of test samples. In theory, its value is a positive number, and the closer it is to 0, the better.

[0107] Coefficient of determination (R 2 ):

[0108]

[0109] Coefficient of determination R 2 It is a statistical indicator used to evaluate the goodness of fit of a regression model. 2 The value of is between (0,1), and R 2 The closer it is to 1, the better the sample fit of the model regression prediction.

[0110] S3: Combined with the stress prediction model, iterative optimization is performed with the minimization of the blade root groove equivalent stress as the optimization goal to obtain the optimal geometric parameters of the current blade root groove profile within the target temperature range.

[0111] S3 is specifically:

[0112] Minimizing the maximum equivalent stress is selected as the objective function of the optimization design. Considering the influence of multiple fields, the thermal stress and contact stress of the blade root groove are reduced as much as possible under the premise of satisfying all constraints. According to the design variables, constraints and objective functions, the mathematical model of the blade root groove optimization design is as follows. The specific meaning is to find a set of x in the range of 400 (x7) to 600℃ (x8), and minimize the maximum stress value in the stress field under the condition of satisfying the W(x) and H(x) constraint variables:

[0113] Find x = {x1,x2,x3,x4,x5,x6}

[0114]

[0115]

[0116] Among them, x is the blade root groove design variable, x1, x2, x3, x4, x5, x6, x7, x8 are the blade root groove contact section tooth width D / mm, blade root groove contact section tooth inclination angle α / °, blade root groove outer arc R Ex / mm, inner arc of blade root groove R In / mm, blade root bottom width L / mm, contact surface friction coefficient μ, regional minimum temperature T min / ℃, maximum temperature of blade root groove area T max / ℃, is the maximum stress value in the stress field, W(x) is the geometric constraint variable in the horizontal direction of the blade root groove, is the horizontal constraint limit, which is generally taken as 0.9 times of the initial design value. y(x) is the geometric constraint variable in the vertical direction of the blade root groove. It is the vertical constraint limit, which is generally taken as 0.9 times of the initial design value.

[0117] The expressions of constraint variables can be derived by using plane geometric relationships in the horizontal and vertical directions, as shown in Table 5:

[0118] Table 5 is the constraint variable expression

[0119]

[0120] At the same time, since the population of the dung beetle algorithm is initialized by random generation, the dung beetle population may gather in a certain area, which makes the algorithm easy to fall into local optimality and poor convergence accuracy during iterative search. In order to solve these problems, the optimal Latin hypercube is introduced to initialize the distribution of the dung beetle population to replace the original random initialization strategy, so that the dung beetle population is relatively evenly distributed in the entire optimization space, increasing the diversity of the initial population while avoiding the problem of the algorithm falling into the local optimality during the search process. The improved dung beetle initial function is as follows:

[0121] [x n ,y n ]=OLH(Maxnum,lb,ub)

[0122] Where ub is the search upper bound, lb is the search lower bound, Maxnum is the population size, [x n ,y n ] is the updated dung beetle individual position matrix.

[0123] The parameters of the dung beetle optimization algorithm are set as follows: the initial number of dung beetles is 100, the maximum number of iterations is 300, and the improved optimization algorithm is as follows: Figure 5 When the optimization process reaches the convergence condition, a set of optimal parameter combinations can be obtained as shown in Table 6:

[0124] Table 6 shows the geometric design variable values ​​before and after optimization

[0125]

[0126]

[0127] Comparison of results before and after blade root groove optimization Figure 6 As shown, the maximum equivalent stress in the blade root groove area is reduced from 202.22 MPa to 167.6 MPa, which is 17.1% lower than the initial configuration. It can be seen that the maximum equivalent stress in the blade root groove is significantly improved, which fully demonstrates the effectiveness and feasibility of the optimization design method proposed in the present invention.

[0128] The present invention also proposes a blade root wheel groove profile optimization design system based on multi-field node feature weighted aggregation, including:

[0129] A data set construction unit is used to perform finite element simulation according to the geometric design variables of the blade root wheel groove profile to be optimized, the temperature and the contact surface friction coefficient, obtain the characteristic matrix, the adjacency matrix and the equivalent stress value and the equivalent strain value of each grid node, and obtain the training data set after repeated multiple times;

[0130] A stress prediction unit, used for storing a deep graph sampling aggregation neural network integrating spatial and temperature attention and training the deep graph sampling aggregation neural network integrating spatial and temperature attention using a training data set, thereby obtaining a stress prediction model;

[0131] The geometric parameter optimization unit is used to combine the stress prediction model to iteratively optimize the blade root groove with the minimization of the blade root groove equivalent stress as the optimization goal, and obtain the optimal geometric parameters of the current blade root groove line within the target temperature range.

[0132] The present invention also proposes a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of a blade root wheel groove profile optimization design method based on weighted aggregation of multi-field node features are implemented.

[0133] The present invention also proposes a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a blade root wheel groove profile optimization design method based on weighted aggregation of multi-field node features.

[0134] The present invention also proposes a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of a blade root wheel groove profile optimization design method based on weighted aggregation of multi-field node features.

[0135] The above embodiments are used to illustrate the present invention rather than to limit the present invention. Any modification and change made to the present invention within the spirit of the present invention and the protection scope of the claims shall fall within the protection scope of the present invention.

Claims

1. A complex part line optimization design method based on multi-field node feature weighted aggregation, characterized in that: The following steps are involved: S1: Perform finite element simulation based on the geometric design variables, temperature and contact surface friction coefficient of the complex part profile to be optimized, obtain the characteristic matrix, adjacency matrix, and equivalent stress and equivalent strain values ​​of each grid node, and then obtain the training data set; among which, the geometric design variables and contact surface friction coefficient constitute the geometric parameters; S2: Construct a deep graph sampling aggregation neural network that integrates space and temperature attention, and then use the training data set to train the deep graph sampling aggregation neural network that integrates space and temperature attention. After the training is completed, a stress prediction model is obtained; S3: Combined with the stress prediction model, iterative optimization is performed with the optimization goal of minimizing the maximum equivalent stress of complex parts to obtain the optimal geometric parameters of the current complex part profile within the target temperature range.

2. The complex part profile optimization design method based on multi-field node feature weighted aggregation according to claim 1 is characterized in that: In S1, the temperature specifically refers to the maximum temperature and the minimum temperature in the target area.

3. The complex part profile optimization design method based on multi-field node feature weighted aggregation according to claim 1 is characterized in that: In S2, the aggregation function in the deep graph sampling aggregation neural network integrating spatial and temperature attention satisfies the following formula: in, is the jth feature formed by the aggregation of the central node e by its i neighboring nodes, (ne) i is the ith neighbor node of the central node e, L_att[(ne) i ] is the spatial attention coefficient, T_att[(ne) i ] is the temperature attention coefficient, W is the adjustable weight parameter, f (ne)i (j) is the jth feature of the i-th neighbor node, softmax[ ] is the normalized exponential function, is the spatial distance between the ith neighbor node and the central node, is the temperature difference between the i-th neighbor node and the central node e, i = 1,…,n, n is the number of neighbor nodes of the central node e.

4. The complex part profile optimization design method based on multi-field node feature weighted aggregation according to claim 1 is characterized in that: In S3, the network parameters of the stress prediction model are frozen and the stress prediction model is used as the fitness function in the optimization algorithm to obtain an improved optimization algorithm. The improved optimization algorithm iteratively updates the geometric parameters with minimizing the maximum equivalent stress of the complex part as the optimization goal until the optimal geometric parameters of the current complex part profile within the target temperature range are obtained.

5. The complex part profile optimization design method based on multi-field node feature weighted aggregation according to claim 1 is characterized in that: The S3 is specifically: Firstly, the initial parameters of the dung beetle algorithm are determined; then the optimal Latin hypercube function is used to initialize the distribution of the dung beetle population, so that the initial population of dung beetles is evenly distributed in the entire optimization space, thereby obtaining the initialized dung beetle individual position matrix; then, the stress prediction model of the frozen network parameters is used as the fitness function of the dung beetle algorithm to obtain the improved dung beetle algorithm; Then, based on the initialized individual position matrix of dung beetles, the improved dung beetle algorithm is used to iteratively update the position of the dung beetle population until the convergence condition is finally met, thereby obtaining the optimal geometric parameters of the current complex part profile within the target temperature range.

6. A complex part line optimization design system based on multi-field node feature weighted aggregation, characterized in that: include: The data set construction unit is used to perform finite element simulation according to the geometric design variables of the complex part profile to be optimized, the temperature and the contact surface friction coefficient, obtain the characteristic matrix, the adjacency matrix and the equivalent stress value and the equivalent strain value of each grid node, and obtain the training data set after repeated multiple times; A stress prediction unit, used for storing a deep graph sampling aggregation neural network integrating spatial and temperature attention and training the deep graph sampling aggregation neural network integrating spatial and temperature attention using a training data set, thereby obtaining a stress prediction model; The geometric parameter optimization unit is used to combine the stress prediction model to iteratively optimize the complex parts with the minimization of the equivalent stress as the optimization goal, and obtain the optimal geometric parameters of the current complex part profile within the target temperature range.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the complex part profile optimization design method based on weighted aggregation of multi-field node features as described in any one of claims 1 to 5 are implemented.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the complex part profile optimization design method based on weighted aggregation of multi-field node features as described in any one of claims 1 to 5 are implemented.

9. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instruction is executed by a processor, the steps of the complex part profile optimization design method based on weighted aggregation of multi-field node features as described in any one of claims 1 to 5 are implemented.

Images