Method and system for reconstructing blade root wheel groove stress field of turbine blade of gas turbine

Abstract

The invention discloses a gas turbine blade root wheel groove stress field reconstruction method and system, and relates to the technical field of stress field reconstruction, and the method comprises the following steps: collecting a plurality of sets of design variables, a plurality of sets of measuring point data and graph data corresponding to a plurality of finite element models under different molded line design schemes; performing finite element calculation on the finite element model to obtain a plurality of first stress fields and a plurality of second stress fields; the multiple sets of design variables and the multiple sets of measuring point data serve as input, the multiple first stress fields serve as output, and a convolutional neural network is trained; and training the graph convolutional neural network by taking the plurality of groups of design variables, the plurality of groups of measurement point data and the plurality of groups of unstructured grid data as inputs and the plurality of second stress fields as outputs. According to the method, finite element calculation consuming long time is not needed in the application stage, real-time simulation can be achieved, for different network structures, the most skilled data forms can be processed, and the stress field reconstruction efficiency under the blade root wheel groove profile design is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of stress field reconstruction, and in particular to a method and system for reconstructing the stress field of a blade root groove of a gas turbine blade. Background Art

[0002] Gas turbines, as core components of modern energy and power systems, are widely used in power generation, aviation propulsion, and marine propulsion. Turbine blades are key components of gas turbines, and their performance directly impacts their efficiency, reliability, and service life. Operating in extreme environments of high temperature, high pressure, and high speed, turbine blades are subject to complex aerodynamic, thermal, and centrifugal loads. Therefore, the design and manufacture of turbine blades must meet extremely high technical requirements.

[0003] The blade root groove is the key connection between the turbine blade and the disk, and its profile design has a decisive influence on the stress distribution at the root of the turbine blade. The complex loads borne by the turbine blade during operation are transmitted to the disk through the blade root groove. If the profile design of the blade root groove is unreasonable, it will lead to uneven stress distribution, which in turn causes stress concentration. Stress concentration will not only significantly reduce the fatigue life of the blade, but may also cause cracks or even fractures, seriously affecting the safety and reliability of the gas turbine. Therefore, one of the core goals of the blade root groove profile design is to optimize the profile shape to make the stress distribution more uniform and reduce the stress peak, thereby improving the fatigue resistance and overall strength of the blade. In addition, reasonable stress distribution can also reduce local plastic deformation and creep damage, thereby extending the service life of the blade.

[0004] In actual design, it is usually necessary to carry out profile design through means such as finite element analysis, and then accurately obtain the stress distribution under different profile design schemes, and use this as a basis for iterative optimization. Although finite element calculations are highly accurate, profile design usually requires a large number of finite element calculations, which are time-consuming. Especially in complex three-dimensional models and multi-stress field coupling analysis, the calculation time may be as long as several hours or even days. Secondly, profile design optimization usually requires multiple iterations, and each iteration requires re-finite element calculations, which further increases the design cycle and calculation costs. Therefore, profile design requires a large number of finite element calculations, which greatly increases the time cost of stress field reconstruction under different profiles of the blade root groove. Summary of the Invention

[0005] Based on the defects of the above-mentioned existing technologies, the present invention provides a method and system for reconstructing the stress field of the blade root groove of a gas turbine turbine blade, which solves the problem that the existing profile design requires a large amount of finite element calculations, which greatly increases the time cost of reconstructing the stress field under different profiles of the blade root groove.

[0006] The present invention adopts the following technical solutions:

[0007] In a first aspect, the present invention provides a method for reconstructing a stress field of a blade root groove of a gas turbine blade, comprising the following steps:

[0008] Construct multiple finite element models of the gas turbine blade root groove under different profile design schemes, and collect multiple sets of design variables, multiple sets of measurement point data, and image data corresponding to the multiple finite element models. In the finite element models, a structured mesh is used for the blade root groove profile area, and an unstructured mesh is used for the internal area of ​​the blade root.

[0009] Performing finite element calculations on the multiple finite element models to obtain multiple first stress fields corresponding to the structured grids and multiple second stress fields corresponding to the unstructured grids;

[0010] Constructing a stress field reconstruction model, the stress field reconstruction model includes a parallel convolutional neural network and a GMM graph convolutional neural network; training the convolutional neural network with multiple sets of design variables and multiple sets of measurement point data as input and multiple first stress fields as output; training the graph convolutional neural network with multiple sets of design variables, multiple sets of measurement point data and graph data as input and multiple second stress fields as output;

[0011] The design variables, measurement point data and map data of the blade root groove of the gas turbine turbine blade are obtained in real time, and are input into the trained stress field reconstruction model to obtain the reconstructed blade root groove stress field.

[0012] Preferably, the convolutional neural network includes 1 input layer, 3 upsampling layers and 1 output layer, the input layer includes 2 fully connected layers, each upsampling layer includes 1 interpolation layer, 1 batch normalization layer and 2 first convolution layers, and the output layer is 1 second convolution layer;

[0013] The training of the convolutional neural network specifically includes the following steps:

[0014] Combine the design variables and the measurement point data to obtain the input set;

[0015] The input layer linearly transforms and reshapes the input set to obtain the feature map tensor;

[0016] The feature map tensor is interpolated, batch normalized, and convolved once through the upsampling layer to obtain the output feature map;

[0017] The output feature map is convolved twice through the output layer to obtain the reconstructed first stress field.

[0018] Preferably, the graph data is specifically as follows:

[0019] D = {A', B', X};

[0020] Where D is the graph data, A' and B' are the adjacency matrix and degree matrix of the unstructured grid, and X is the information matrix of the unstructured grid nodes.

[0021] Preferably, the GMM graph convolutional neural network includes 1 input layer, 5 intermediate layers and 1 output layer, the input layer and the intermediate layers all use Gaussian deviation linear unit as the activation function, and the output layer does not use an activation function;

[0022] The training of the graph convolutional neural network specifically includes the following steps:

[0023] Weighting of design variables, measurement point data and graph data through the input layer;

[0024] The nonlinearity is added to the weighted data through the intermediate layer to obtain the reconstructed second stress field;

[0025] The reconstructed second stress field is output through the output layer.

[0026] Preferably, when training a convolutional neural network and a graph convolutional neural network, the network parameters are optimized by minimizing a sub-loss function, and the sub-loss function is specifically as follows:

[0027] L T =L F +L S +L gx +L gy ;

[0028] Where, L T is the sub-loss function, L F is the loss function of the global stress field, L S is the loss function of the field maximum, L gx and L gy is the loss function of the global stress field gradient in the x-direction and the y-direction.

[0029] Preferably, the stress field reconstruction model is trained as a whole through a total loss function, and the total loss function is specifically as follows:

[0030] L C =L T,1 +L T,2 +L′;

[0031] Where, L C is the total loss function, L T,1 With L T,2 is the sub-loss function of the convolutional neural network and the GMM graph convolutional neural network, and L′ is the loss function of the convolutional neural network and the GMM graph convolutional neural network for the predicted values ​​of the common nodes of the structured grid and the unstructured grid.

[0032] Preferably, the measurement point data includes temperature data and displacement data.

[0033] In a second aspect, the present invention provides a gas turbine blade root groove stress field reconstruction device, comprising:

[0034] The acquisition module is used to construct multiple finite element models of the gas turbine blade root groove under different profile design schemes and collect multiple sets of design variables, multiple sets of measurement point data, and image data corresponding to the multiple finite element models. In the finite element model, the blade root groove profile area uses a structured grid, and the internal area of ​​the blade root uses an unstructured grid.

[0035] A calculation module, configured to perform finite element calculations on the multiple finite element models to obtain multiple first stress fields corresponding to the structured grids and multiple second stress fields corresponding to the unstructured grids;

[0036] A training module is used to construct a stress field reconstruction model, wherein the stress field reconstruction model includes a parallel convolutional neural network and a GMM graph convolutional neural network; the convolutional neural network is trained with multiple sets of design variables and multiple sets of measurement point data as input and multiple first stress fields as output; the graph convolutional neural network is trained with multiple sets of design variables, multiple sets of measurement point data and graph data as input and multiple second stress fields as output;

[0037] The reconstruction module is used to obtain the design variables, measurement point data and map data of the blade root groove of the gas turbine turbine blade in real time, and input them into the trained stress field reconstruction model to obtain the reconstructed blade root groove stress field.

[0038] Compared with the prior art, the at least one technical solution adopted by the present invention can achieve the following beneficial effects:

[0039] The present invention first collects multiple sets of design variables, multiple sets of measurement point data, and multiple sets of graph data under different profile design schemes, and performs finite element calculations on the different profile design schemes, obtaining multiple first stress fields corresponding to structured grids and multiple second stress fields corresponding to unstructured grids. The present invention adopts a parallel dual-branch design, uses CNN branches to efficiently process structured grid data, and uses GCN branches to accurately process unstructured grid data. A stress field reconstruction model is proposed, and the stress field reconstruction model is trained using corresponding training data. This method does not require time-consuming finite element calculations in the application stage, can achieve real-time simulation, and can process the data form that each network structure is best at for different network structures, thereby improving the efficiency of stress field reconstruction under blade root wheel groove profile design. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. 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 any creative work.

[0041] Figure 1 It is a stress field reconstruction model based on graph combination convolutional neural network of the present invention;

[0042] Figure 2 The unstructured grid of the present invention and its corresponding undirected graph;

[0043] Figure 3 The figure shows the change curve of the loss function of GCCNN during the training process under different graph convolution models of the present invention;

[0044] in, Figure 3 (a): loss function value on the training set, Figure 3 (b): loss function value on the test set;

[0045] Figure 4 It is the field reconstruction deviation of GCCNN under different graph convolution models of the present invention;

[0046] in, Figure 4 (a): Maximum absolute deviation of the field, Figure 4 (b): Field mean absolute deviation map;

[0047] Figure 5 This is a comparison diagram of the field reconstruction deviation between GCCNN and GCN in the grid encryption layer of the present invention;

[0048] in, Figure 5 (a): Maximum absolute deviation of the field, Figure 5 (b): Field mean absolute deviation map;

[0049] Figure 6 The loss function change curve and local magnification diagram of the GCCNN of the present invention during the training process;

[0050] in, Figure 6 (a): Loss function change curve, Figure 6 (b): Local enlarged view;

[0051] Figure 7 : The turbine blade root wheel groove line diagrams of 6 typical samples of the present invention;

[0052] in, Figure 7 (a): Blade root profile, Figure 7 (b): wheel groove profile;

[0053] Figure 8 This is the GCCNN displacement field reconstruction result of the present invention;

[0054] in, Figure 8 of (a): u x displacement field, Figure 8 (b): u y displacement field, Figure 8 (c): Points u on the root wheel groove line x The distribution curve of Figure 8 (d): u at each point of the blade root wheel groove line y The distribution curve of

[0055] Figure 9 is the GCCNN strain ε of the present invention x , stress σ x Component field reconstruction results;

[0056] in, Figure 9 of (a): u x strain field, Figure 9 (b): σ x stress field, Figure 9 (c): ε of each point on the root wheel groove line x The distribution curve of Figure 9 (d): σ at each point on the root wheel groove line x The distribution curve of

[0057] Figure 10 is the GCCNN strain ε of the present invention y , stress σ y Component field reconstruction results;

[0058] in, Figure 10 (a): ε y strain field, Figure 10 (b): σ y stress field, Figure 10 (c): ε of each point on the root wheel groove line y The distribution curve of Figure 10 (d): σ at each point on the root wheel groove line y The distribution curve of

[0059] Figure 11 is the GCCNN strain γ of the present invention xy , stress τ xy Component field reconstruction results;

[0060] in, Figure 11 (a): γ xy strain field, Figure 11 (b): τxy stress field, Figure 11 (c): γ at each point on the root wheel groove line xy The distribution curve of Figure 11 (d): τ at each point on the root wheel groove line xy The distribution curve of

[0061] Figure 12 is the equivalent stress reconstruction result of sample T1 of the present invention;

[0062] in, Figure 12 (a): Equivalent stress field, Figure 12 (b): Distribution curve of equivalent stress at each point on the root wheel groove line;

[0063] Figure 13 This is the equivalent stress reconstruction result of sample T2 of the present invention;

[0064] in, Figure 13 (a): Equivalent stress field, Figure 13 (b): Distribution curve of equivalent stress at each point on the root wheel groove line;

[0065] Figure 14 This is the equivalent stress reconstruction result of sample T2 of the present invention;

[0066] in, Figure 14 (a): Equivalent stress field, Figure 14 (b): Distribution curve of equivalent stress at each point on the root wheel groove line;

[0067] Figure 15 This is the equivalent stress reconstruction result of sample T2 of the present invention;

[0068] in, Figure 15 (a): Equivalent stress field, Figure 15 (b): Distribution curve of equivalent stress at each point on the root wheel groove line;

[0069] Figure 16 This is the equivalent stress reconstruction result of sample T2 of the present invention;

[0070] in, Figure 16 (a): Equivalent stress field, Figure 16 (b): Distribution curve of equivalent stress at each point on the root wheel groove line;

[0071] Figure 17 This is the equivalent stress reconstruction result of sample T2 of the present invention;

[0072] in, Figure 17 (a): Equivalent stress field, Figure 17 (b): Distribution curve of equivalent stress at each point on the root wheel groove line;

[0073] Figure 18 The maximum equivalent stress prediction accuracy evaluation of the turbine blade root groove of the present invention is performed;

[0074] in, Figure 18 (a): Maximum equivalent stress prediction result, Figure 18 (b): distribution of relative deviation of maximum equivalent stress prediction;

[0075] Figure 19 This is a schematic diagram of the position of the profile parameters of the present invention;

[0076] in, Figure 19 (a): Blade root profile parameters, Figure 19 (b): Wheel groove profile parameters. DETAILED DESCRIPTION

[0077] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. 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 making creative efforts are within the scope of protection of the present invention.

[0078] In finite element analysis and computational mechanics, meshes are the basic tools for discretizing computational domains, and can be divided into structured meshes and unstructured meshes according to their geometric arrangement. A structured mesh is composed of regularly arranged units (such as quadrilaterals or hexahedrons), and the mesh nodes have clear topological relationships. The calculation is efficient and accurate, but it is difficult to handle complex geometries and is suitable for convolution operations; an unstructured mesh is composed of irregular units (such as triangles or tetrahedrons), and the connection relationship between mesh nodes is described by an adjacency list (non-coordinate index). It can accurately fit complex shapes but has a high computational cost and is suitable for graph convolution operations. Therefore, the present invention proposes a method for reconstructing the stress field of a gas turbine blade root groove, comprising the following steps:

[0079] S1: Construct multiple finite element models of the gas turbine blade root groove under different profile design schemes, and collect multiple sets of design variables, multiple sets of measurement point data, and image data corresponding to the multiple finite element models. In the finite element models, a structured mesh is used for the blade root groove profile area, and an unstructured mesh is used for the internal area of ​​the blade root.

[0080] The Latin hypercube method is used to generate a large amount of design variable data within a certain range. The required variables for parameterizing the profile are used to determine the profile shape.

[0081] Let the generated sample points be {x (1) ,x (2),...,x (N)}, x is the sampling parameter sample, N is the number of sample points, each point d is the dimension, and each dimension represents a parameter type. Ensure that the projection of each dimension i satisfy:

[0082]

[0083] in, is the i-th dimension sample point in the k-th sample, is the lower bound of the i-th dimension, that is, the minimum value of the sample point, Δx i is the length of the i-th dimension subspace, calculated as: p i (k) is a randomly shuffled interval index, p i (k) For different k, they are not repeated and are used to divide the subintervals of the i-th dimension. The interval is defined as: Each subinterval corresponds to an index, from 1 to N. That is, when p i (k) =1, it is the first subinterval; when p i (k) = N, it is the last subinterval. N groups of intervals in all dimensions are combined to generate N sample points.

[0084] The data preparation stage is mainly responsible for parsing the input information and constructing regular grid data suitable for CNN and graph structure data suitable for GCN.

[0085] (1) Design variable x d :Use Latin hypercube method to calculate the design variable x under multiple schemes d ∈R m Take samples.

[0086] (2) Measurement point data x s :Can directly read the measurement point data x s ∈R n The measurement point data is the data generated by the measurement sensors obtained at key locations, including temperature data and vibration displacement data.

[0087] (3) Graph data D: used for GCN calculation.

[0088] S2: Perform finite element calculations on the finite element models under different profile design schemes to obtain multiple first stress fields corresponding to the structured grid and multiple second stress fields corresponding to the unstructured grid.

[0089] S3: Construct a stress field reconstruction model, which includes a parallel convolutional neural network and a GMM graph convolutional neural network. The convolutional neural network is trained using multiple sets of design variables and multiple sets of measurement point data as input and multiple first stress fields as output. The graph convolutional neural network is trained using multiple sets of design variables, multiple sets of measurement point data, and graph data as input and multiple second stress fields as output.

[0090] In order to combine the advantages of convolutional neural networks for structured data and graph convolutional neural networks for unstructured data processing, the present invention combines these two neural networks and establishes a graph combined convolutional neural network (GCCNN) for processing finite element mesh data containing hybrid structures for strength analysis, thereby realizing stress field reconstruction. The overall framework of the model is as follows Figure 1 shown.

[0091] GCCNN consists of a convolutional neural network and a graph convolutional neural network in parallel. The convolutional neural network is used to reconstruct the stress field of the external encryption layer composed of structured grids, and the graph convolutional neural network is used to reconstruct the stress field of the internal unstructured grid. The convolutional neural network mainly includes several convolutional layers, pooling layers, fully connected layers, batch normalization layers, and activation functions. The input is a low-order tensor composed of design variables or local measurement point data. The graph convolutional neural network consists of several graph convolutional layers and activation functions. The input is graph data composed of design variables, local measurement point data, and node connection relationships. The output of both is the calculation results of the stress field of the corresponding finite element grid.

[0092] The input of the convolutional neural network is the design variable x d ∈R m And measurement point data x s ∈R n The set X is as follows:

[0093]

[0094] Among them, n is the dimension of the measurement points, and m is the dimension of the design variables.

[0095] The network consists of an input layer (fully connected layer), an upsampling layer, and an output layer.

[0096] The input layer consists of two fully connected layers:

[0097] Z (1) =W (1) X+b (1) ;

[0098] Z (2) =W (2) X+b(2) ;

[0099] Among them, W (1) and W (2) are the weight matrices of the fully connected layer, b (1) and b (2) is the bias term, Z (1) is the linear transformation output of the first fully connected layer, Z (2) is the linear transformation output of the second fully connected layer.

[0100] Then, for Z (2) Reshape it to match the spatial dimensions that the CNN can handle:

[0101] X (0) =reshape(Z (2) ,H0,W0,C0);

[0102] Among them, H0, W0, C0 are the feature dimensions after initialization, X (0) is the feature map tensor.

[0103] In convolutional neural networks, high-order tensor data and vectors can be equivalently converted to each other after being reshaped. If the problem to be solved is in the form of dimensionality reduction at the data structure level, such as evaluating flow performance indicators from flow field results, evaluating the stress on the bearing surface from the stress field, etc., the fully connected layer in this scenario mainly undertakes the dimensionality reduction task of the vector after the high-order tensor data is reshaped; when dealing with problems in the form of increased dimensionality, such as predicting the stress field from design parameters, the input low-dimensional vector forms a high-dimensional vector after being increased in dimensionality by the fully connected layer, and then a high-order tensor number can be generated through reshaping. These are two important application scenarios of fully connected layers in convolutional neural networks.

[0104] Each upsampling layer consists of 1 interpolation layer, 1 batch normalization layer and 2 convolutional layers. The convolution kernel size of the convolution layer is 3×3, the stride is 1, and the rectified linear unit is used as the activation function.

[0105] The function of the interpolation layer is to upsample the input tensor, that is, to use the data of adjacent position elements to perform interpolation operations, thereby increasing the length and width of the input tensor. The interpolation formula is as follows:

[0106]

[0107] Where I(·) is the interpolation operator. Bicubic interpolation is used here. This method uses 16 adjacent elements in the input tensor for interpolation. Since more elements are included, smoother operation results can be obtained. In the first upsampling layer, l = 0, Then proceed down and output X (1)Then in the second upsampling layer, Cycle in sequence.

[0108] For each point

[0109]

[0110] Where X(i, j) is the integer coordinate point on the original image, i and j are the integer coordinates of the discrete grid, and the weight w(i, j) is calculated according to the cubic convolution kernel:

[0111]

[0112] Wherein, t=xi represents the offset of the pixel point.

[0113] During the training of a convolutional neural network, if the distribution of each batch of data varies, the neural network will learn these differences during each iteration, which will greatly increase the training cost of the network. To accelerate network convergence and improve network generalization, the batch normalization layer is applied to the construction of deep neural networks. By introducing two parameters to be learned, the batch normalization layer performs a linear normalization (minimum-maximum normalization) on the input tensor across different channels, namely:

[0114]

[0115] in, is the normalized feature map, is the input feature map after interpolation, X max is the maximum value in the current batch, X min The minimum value in the current batch.

[0116] Unlike fully connected layers, each node in a convolutional layer is only connected to a subset of nodes in the previous layer. Furthermore, as the number of layers increases, the network's receptive field gradually expands, allowing deeper networks to learn non-local features. These are the two main characteristics of convolutional neural networks: local connectivity and parameter sharing. Therefore, convolution operations can better process the temporal and spatial characteristics of input data.

[0117] This contains two layers of 3×3 convolution and activation function RELU:

[0118] First layer:

[0119]

[0120] Second layer:

[0121]

[0122] Among them, * is the convolution operator.

[0123] Assuming there are L upsampling layers, the final output of the upsampling layer is X (L) , X (L) =X (l+k) .

[0124] The output layer is a convolutional layer with a convolution kernel size of 1×1 and a stride of 1. No activation function is used. The mathematical expression is as follows:

[0125] U J =W out *X (L) +b out ;

[0126] Among them, W out ∈R 1×1×C'×1 is the 1×1 convolution kernel weight parameter, c∈{1,2,...,C'} is the channel index, X (L) ∈R H′×W′×C′ is the output feature map of the last convolution of the upsampling layer, b out is the bias term of 1×1 convolution, U J The stress field results predicted by convolutional neural network.

[0127] CNN uses regular grid data to extract and reconstruct stress field information through upsampling and convolution operations.

[0128] (1) Dimensionality is increased through the fully connected layer to convert the input design variables into the initial feature map format suitable for CNN;

[0129] (2) Use interpolation upsampling to restore the spatial resolution of the feature map layer by layer.

[0130] (3) Use batch normalization to ensure feature stability and adaptability to data of different scales.

[0131] (4) Deep features are extracted through convolutional layers and activation functions ReLU, and each layer matches higher resolution information.

[0132] (5) Finally, the stress field prediction on the regular grid is generated through the convolutional layer.

[0133] GCN uses the GMM graph convolution model, which includes 1 input layer, 5 intermediate layers and 1 output layer. The input layer and the intermediate layers use Gaussian deviation linear unit as the activation function, and the output layer does not use an activation function.

[0134] First, let's introduce how to construct the adjacency matrix:

[0135] Usually, a graph can be described by a pair of G = {V, E}, where V is the vertex set, E is the edge set, and the vertex is represented by v i∈V, edge is represented by e i,j =(v i ,v j )∈E, called v i ,v j are neighbor vertices. Figure 2 Shown is an example of an unstructured grid and its corresponding undirected graph, where the nodes of the grid correspond to the vertices of the graph and the edges of the grid correspond to the edges of the graph.

[0136] When describing the connection relationship between vertices, the adjacency matrix A and the degree matrix B can be used. The adjacency matrix A is used to store information about whether vertices are connected. The degree matrix B is a diagonal matrix used to store the number of neighboring vertices of a vertex. The calculation method is as follows:

[0137]

[0138] Since mesh nodes in the finite element method are connected by cells, the calculation method of the connection relationship is different from that of the graph. Not only are the nodes at both ends of the edge considered as neighbors, but all nodes in the same cell are also considered as neighbors. Taking node p8 and its corresponding vertex v8 as an example, the adjacency matrix element a of vertex v8 can be calculated. 8,j , degree matrix element b 8,8 and the adjacency matrix element a′ of node p8 8,j′ , degree matrix element b′ 8,8 for:

[0139] a 8,j =1,j=3,7,9,13;b 8,8 =4

[0140] a′ 8,j′ =1,j′=2,3,5,7,9,10,11,12,13,14; b′ 8,8 =10;

[0141] Therefore, under graph theory, unstructured grid data (graph data) can be described as:

[0142] D = {A', B', X};

[0143] Where A' and B' are the adjacency matrix and degree matrix of the unstructured grid, and X is the information matrix of the unstructured grid node, which contains the node space coordinates and the result data of the finite element analysis.

[0144] Then introduce the calculation formula of all graph convolution layers (GMM-GCN):

[0145] GMM graph convolution is a special form of GCN that propagates information through weighted Gaussian mixture models. Its general form is as follows:

[0146]

[0147] Where K is the number of Gaussian mixture components (i.e., using K different adjacency matrices A' k ), γ k is the weight coefficient (learnable parameter) in the Gaussian mixture model, A' k is the normalized adjacency matrix of the kth Gaussian component, is the graph convolution weight matrix corresponding to the kth Gaussian component, b (l) is the bias term. The calculation formula of the normalized adjacency matrix is:

[0148]

[0149] Where B' k is the degree matrix of the kth Gaussian component.

[0150] The input of the graph convolutional neural network is the design variable x d ∈R m , measuring point data x s ∈R n And the collection of graph data D:

[0151]

[0152] The first layer of graph convolution is expressed as:

[0153]

[0154] The activation function is Gaussian bias linear unit:

[0155]

[0156] Where σ 2 It is an adjustable parameter that controls the scale of Gaussian activation.

[0157] Middle layer:

[0158] Each hidden layer is calculated using GMM-GCN:

[0159]

[0160] Each layer uses Gaussian deviation linear unit as the activation function, which makes the information propagation nonlinear.

[0161] Output layer:

[0162]

[0163] The output layer does not use an activation function, namely:

[0164] X(7) =Z (7) ;

[0165] Therefore, the final stress field prediction result is:

[0166] U T =X (7) ;

[0167] GCN propagates information through the adjacency matrix to predict stress fields on irregular grids (such as finite element grids).

[0168] (1) GMM-GCN (Gaussian mixture graph convolution) is used for information propagation, and multiple neighboring matrices are used to capture information of different ranges.

[0169] (2) Perform graph convolution calculation layer by layer.

[0170] (3) Gaussian deviation activation function is used to increase nonlinear expression ability and improve prediction accuracy.

[0171] (4) Finally, the stress field prediction results on the irregular grid are obtained.

[0172] Loss calculation and optimization.

[0173] (1) Comprehensively consider the loss function L of the global stress field f , the loss function L of the field maximum s , the loss function L of the global stress field gradient in the x-direction and y-direction gx ,L gy , the loss function of the convolutional neural network is L T,1 .

[0174] (2) Comprehensively consider the loss function L of the global stress field f , the loss function L of the field maximum s , the loss function L of the global stress field gradient in the x-direction and y-direction gx ,L gy , the loss function of the graph convolutional neural network is L T,2 .

[0175] (3) Comprehensively consider the loss function L of the convolutional neural network T,1 , the loss function L of the graph convolutional neural network T,2 And the loss function L′ formed by the two networks for the predicted values ​​of the common nodes of the structured grid and the unstructured grid, the total loss function L of GCCNN is obtained C , to realize the parallel connection of convolutional neural network and graph convolutional neural network.

[0176] Loss function:

[0177] In this paper, all tasks are regression tasks, and the mean square error (MSE) is mainly used as the loss function, which is expressed as:

[0178]

[0179] Where N is the sample size of the training samples in the dataset, with U i is the output and actual value of the neural network under the i-th sample, M i is the number of nodes of the i-th sample.

[0180] Considering the emphasis in structural strength calculation, the MSE of the field maximum value and the MSE of the global stress field gradient are additionally added to the loss function. The total loss function is:

[0181] L T =L F +L S +L gx +L gy ;

[0182] Where, L f is the loss function of the global stress field, expressed as:

[0183]

[0184] L s is the loss function of the field maximum value, and its expression is:

[0185]

[0186] in, is the maximum output of the neural network under the i-th sample, maxU i is the maximum actual value under the i-th sample;

[0187] L gx ,L gy is the loss function of the global stress field gradient in the x-direction and the y-direction, and its expression is:

[0188]

[0189] in, and is the global stress field gradient in the x and y directions of the output of the neural network under the i-th sample, and is the actual global stress field gradient in the x and y directions under the i-th sample;

[0190] In GCCNN, a combined loss function is constructed to achieve the parallel connection between convolutional neural network and graph convolutional neural network, namely:

[0191] L C =L T,1 +L T,2 +L′;

[0192] Where, L T,1 With L T,2 Convolutional neural network and graph convolutional neural network in formula L T The total network loss function calculated below is L′, which is the loss function composed of the predicted values ​​of the two networks for the common nodes of the structured grid and the unstructured grid. The expression is:

[0193]

[0194] in, and They are the predicted values ​​of the convolutional neural network and graph convolutional neural network at common nodes, respectively.

[0195] During the optimization process, the training of GCCNN is completed by back propagation of the combined loss function.

[0196] Compared with separate convolutional neural networks and graph convolutional neural networks, GCCNN enables the two networks to be trained simultaneously through a combination of loss functions. This not only effectively improves the accuracy and smoothness of the prediction results in the common area, but also reduces the possibility of overfitting to a certain extent by increasing the complexity of the model.

[0197] Compared with separate convolutional neural networks and graph convolutional neural networks, GCCNN enables the two networks to be trained simultaneously through a combination of loss functions. This not only effectively improves the accuracy and smoothness of the prediction results in the common area, but also reduces the possibility of overfitting to a certain extent by increasing the complexity of the model.

[0198] S4: Real-time acquisition of the design variables, measurement point data, and image data of the blade root groove of the gas turbine turbine, and inputting them into the trained stress field reconstruction model to obtain the reconstructed blade root groove stress field.

[0199] In order to verify the accuracy of the GCCNN method in reconstructing the stress field of the blade root groove of a gas turbine turbine blade, a simulation verification was carried out in the present invention, which mainly included the selection and verification of the graph convolution layer, the establishment of the stress field reconstruction model, the blade root groove profile example, the reconstruction example of the displacement, strain and stress components, and the analysis of the equivalent stress reconstruction results.

[0200] Validation of graph convolutional layer selection.

[0201] In the above description, the present invention directly specifies the graph convolution layer as Gaussian mixture graph convolution (GMM). In order to select the optimal graph convolution model in the field reconstruction task, the present invention compares the performance of four models in the stress and strain field reconstruction task under the same parameter settings: frequency domain graph convolution ChebNet, spatial domain graph convolution Graph Sample and Aggregate (GraphSAGE), Gaussian Mixture Model (GMM) and Spline Convolution (SplineConv). The loss function change curves of all models during the training process are shown in Figure 2. Figure 3 As shown in Figure 2, the blade root groove dataset is used as the training and test dataset for the model. The curve represents the loss function value of the model on the training set, and the scatter plot represents the loss function value of the model on the test set.

[0202] After training, the prediction performance of displacement, strain and stress of all models on the test set was further compared, with the maximum absolute deviation and the average absolute deviation as indicators. The results are as follows: Figure 4 As shown in the figure. Consistent with the loss function of the model on the validation set after training, the frequency domain graph convolution model Chebnet performs the worst, while the spatial domain graph convolution model GMM performs the best. In addition, even though SplineConv exhibits significant overfitting during training, its performance is still superior to other graph convolution models except GMM. In summary, the present invention selects GMM as the graph convolution model for GCCNN.

[0203] Since structured data can be regarded as a simplified form of unstructured data, structured grid data can also be described in the form of a graph. The blade groove stress field reconstruction task can be completed using only GCN. Therefore, the stress field reconstruction performance of GCCNN and GCN in the structured grid encryption area at the geometric boundary of the blade root groove is evaluated. The field reconstruction deviations of displacement, strain and stress on the test set are as follows: Figure 5 As shown in the figure, it can be seen that for each parameter, the maximum absolute deviation and the average absolute deviation of the GCCNN model are generally smaller than those of the GCN model alone. Therefore, the GCCNN method proposed in this invention has higher prediction accuracy and better effect than the GCN method alone.

[0204] Establishment of stress field reconstruction model.

[0205] CNN consists of one input layer, three upsampling layers, and one output layer. Neither the input nor the output layer uses an activation function, and the upsampling layer uses a rectified linear unit as the activation function. GCN uses a GMM graph convolutional model, consisting of seven graph convolutional layers, including one input layer, one output layer, and five intermediate layers. The input and intermediate layers use Gaussian deviation linear units as the activation function, and the output layer does not use an activation function. Table 1 shows the detailed parameters of the model.

[0206] It should be noted that the 10 × 247 size of the CNN upsampling layer 3 and the output layer in Table 1 is designed to correspond to the number of mesh nodes encrypted at the boundary of the turbine blade root groove profile line; the input size of the GCN is 14, which includes the x and y coordinates of the nodes and 12 design variables; the output size is 8, which includes the two displacement components, three strain components, and three stress components of the blade root groove.

[0207] Table 1 GCCNN model parameters

[0208]

[0209] Using the Latin hypercube method, we sampled within the range of blade root groove profile design parameters. We established a turbine blade root groove dataset consisting of profile samples that met the constraints. The dataset contained 6,000 samples, divided into a training set (70%, 4,200 samples) and a test set (30%, 1,800 samples) in a 7:3 ratio. This dataset was used to build the stress field reconstruction model. The sampling range is shown in Table 2.

[0210] Table 2 Value ranges of turbine blade root groove parameter design variables

[0211]

[0212] Among them, the schematic diagram of the position of the profile parameters is as follows Figure 19As shown. In order to avoid the situation where the blade root shape is too wide or too long due to unreasonable values ​​of design variables, the relevant parameters of the model were constrained during the research process, namely the overall width of the profile L1 / mm and the overall height of the profile L2, which are the boxed parameters in the figure. The overall width of the profile should be smaller than the width of the blade root midbody, and the increase in the overall height of the profile should not exceed 20%. The workstation configuration for training and testing the model is Intel Core i7-12700K CPU and Nvidia RTX A6000 (48GB) GPU, and the turbine blade root groove dataset is used as the training and test dataset of the model. The convolutional neural network uses the Adam optimization algorithm, and the initial learning rate is set to 0.001. The graph convolutional neural network uses the Adamax optimization algorithm, and the initial learning rate is set to 0.005. Both use the MSE loss function; the batch sample size is 64, and the adaptive learning rate descent algorithm based on the loss function is used. The loss function change curve of the model during training is shown as follows. Figure 6 As shown in the figure, during the training process, the loss function value of GCCNN shows a downward trend as a whole. From the enlarged figure, we can see that when the training is about to end, when the iteration round exceeds 550, the loss function decline rate slows down further, and the loss function value of the test set gradually becomes significantly higher than that of the training set. At this time, the loss function value has dropped to a small enough level, so the model training is terminated after 600 iterations. The average loss function value of the model in the last 100 iteration rounds of the training set and the test set is 2.59×10 -5 and 2.87×10 -5 .

[0213] Table 3 further compares the computational overhead of GCCNN and the finite element method for calculating the turbine blade root groove stress field. Compared to finite element analysis, the proposed model enables rapid reconstruction of the turbine blade root groove stress field after deployment. The table shows that the average computation time with GPU acceleration is reduced from 12 seconds to 0.16 seconds, a 75-fold increase in average computation speed.

[0214] Table 3 Comparison of average calculation time of reconstructed model and finite element analysis

[0215]

[0216] Example of a root wheel groove profile.

[0217] Table 4 shows six representative samples from the turbine blade root groove test sample set. Figure 7 The geometric shapes of the blade roots and wheel groove lines of these 6 samples are displayed, which are typical structures of the 4-tooth fir tree-shaped blade root wheel groove line.

[0218] Table 4 Six typical design schemes

[0219]

[0220] Example of reconstruction of displacement, strain, and stress components.

[0221] Take sample T1 as an example to illustrate. Figure 8 The comparison between the finite element analysis results and the reconstruction results of the displacement field in the x and y directions is shown. Under the action of centrifugal force and equivalent tension, the u x The displacement field is numerically symmetrical along the center line. The blade root teeth shrink along the x direction. Most of the area on the left side of the center line is u x ≥0, most of the area on the right u x ≤0, the opposite is true for the wheel groove, the wheel groove opening tends to expand along the x direction, and the left side of the center line u x ≤0, right u x ≥0. u of blade root groove y The displacement field is also symmetrical along the center line, and the u in the blade root area is y It is obviously larger than the wheel groove area. In terms of reconstruction accuracy, u x and u y The reconstruction deviation of the displacement field is within 5%, and u y The overall relative prediction error is much smaller than u x .from Figure 8 (c) and Figure 8 As can be seen from (d), the finite element analysis results of the displacement on the blade root groove profile are in good agreement with the reconstruction results; since the maximum value is often concerned in strength analysis, its relative deviation is given, and the u on the blade root profile is x 、u y The relative deviation of the maximum value is 0.40%, 0.07%, and the wheel groove line u x 、u y The relative deviations of the maximum values ​​were 1.00% and 0.05%.

[0222] Figure 9 The finite element analysis results and reconstruction results of the normal strain stress field and shear stress field in the x and y directions are further compared. The results show that GCCNN can reconstruct the three stress and strain component fields with high precision, and the distribution of stress and strain fields is consistent. Figure 9 (a) and (b) show the reconstruction results of the normal strain and normal stress in the x direction. The distribution of the field still maintains a highly symmetrical form. The maximum ε at the blade root x , σ x The value appears at the concave fillet below the non-load bearing surface of the first pair of teeth, where the maximum ε of the wheel groove x , σ x The value appears in the concave corner at the bottom. Figure 9The finite element analysis results of the displacement on the profile of the blade root groove in (c) and (d) are in good agreement with the reconstruction results. x , σ x The relative prediction deviation of the maximum value is 2.13% and 1.60%, and the ε x , σ x The relative prediction deviations of the maximum values ​​are 0.08% and 0.42%.

[0223] Figure 10 The reconstruction results of the normal strain and normal stress in the y direction are shown, and the distribution of the field still maintains a highly symmetrical form. As shown in the figure, the maximum ε at the blade root y , σ y The value appears at the concave fillet above the first pair of tooth bearing surfaces, and ε y , σ y There are multiple peaks of similar size, located at the concave fillet below the bearing surface of each pair of teeth. Figure 10 The finite element analysis results of the displacement on the profile of the blade root groove in (c) and (d) are in good agreement with the reconstruction results. y , σ y The relative prediction deviation of the maximum value is 1.22% and 0.99%, and the ε y , σ y The relative prediction deviations of the maximum values ​​are 0.43% and 0.40%.

[0224] Figure 11 The reconstruction results of shear strain and shear stress in the xy plane are shown. The distribution of the field still maintains a highly symmetrical form. The maximum γ xy , τ xy The value is located at the concave fillet above the second pair of tooth bearing surfaces, where the maximum γ xy , τ xy The value is located at the concave fillet at the bottom of the profile. Figure 11 The finite element analysis results of the displacement of the blade root groove in (c) and (d) are in good agreement with the reconstruction results. xy , τ xy The relative prediction deviation of the maximum value is 0.16% and 0.26%, and the γ xy , τ xy The relative prediction deviations of the maximum values ​​are 0.62% and 0.12%.

[0225] In general, the three strain and stress components of the turbine blade root groove have multiple peaks along the profile, all located in the concave fillet area of ​​the blade root groove, so the parameters of this area are considered in the parametric modeling. y , σ yOverall, it is greater than the other two strain and stress components. The reconstruction results show that the maximum absolute deviation of the stress component reconstruction is less than 36MPa, and the maximum relative deviation on the blade root groove line is 2.13%, which further proves that GCCNN has very high accuracy in reconstructing the stress field of the blade root groove of the gas turbine turbine blade.

[0226] Analysis of equivalent stress reconstruction results.

[0227] This section will give a comparison of the equivalent stress field results of six typical samples of turbine blade root grooves under finite element analysis and reconstruction model, and then randomly select three batches of samples to evaluate the prediction accuracy of the maximum equivalent stress. Figure 12 The finite element analysis and reconstruction results of the equivalent stress field for specimen T1 are presented. As shown in the figure, T1's maximum equivalent stress is the smallest among the six specimens. The tooth width control parameters, contact surface tooth width, and contact surface inclination angle for this specimen are significantly higher than those of the other five specimens. The maximum absolute deviation of the reconstructed field is 36 MPa, and the maximum equivalent stress in the blade root groove is 742 MPa, occurring at the concave fillet above the bearing surface of the first pair of teeth at the blade root. The reconstruction deviation at this location is 0.56%. The finite element analysis results and the reconstruction results for the equivalent stress along the blade root groove profile agree well. Figures 13 to 17 The equivalent stress reconstruction results of samples T2 to T6 are further shown.

[0228] Compared to sample T1, T2 has smaller tooth width control parameters, contact surface width, and contact surface inclination, and the radius of the concave fillet of the blade root groove is larger. The maximum absolute deviation of the reconstructed field is 35 MPa, and the maximum equivalent stress of the blade root groove is 800 MPa, occurring at the concave fillet above the bearing surface of the first pair of blade root teeth. The reconstruction deviation at this location is 0.89%. The equivalent stress finite element analysis results along the blade root groove profile are in good agreement with the reconstructed results.

[0229] The tooth width control parameters, contact surface inclination angle, and concave arc radius of the blade root groove of sample T3 are smaller than those of T1 and T2. The maximum absolute deviation of the reconstructed field is 48 MPa, and the maximum equivalent stress of the blade root groove is 895 MPa, which occurs at the concave fillet at the bottom of the groove. The reconstruction deviation at this location is 0.08%. The finite element analysis results of the equivalent stress along the blade root groove profile are in good agreement with the reconstruction results.

[0230] The stress level of specimen T4 was significantly higher than that of the previous three specimens. Its contact surface width and the concave and convex fillet radii of the root wheel groove were further reduced. The maximum absolute deviation of the reconstructed field was 53 MPa, and the maximum equivalent stress of the root wheel groove was 990 MPa, occurring at the concave fillet above the bearing surface of the first pair of teeth at the root. The reconstruction deviation at this location was 1.08%. The equivalent stress finite element analysis results along the root wheel groove profile line were in good agreement with the reconstructed results.

[0231] The throat width of sample T5 is comparable to that of T2 and T4. The concave and convex arc radii of the blade root groove are both larger than those of T4. The contact surface inclination angle is the smallest among the six samples. The maximum absolute deviation of the reconstructed field is 50 MPa. The maximum equivalent stress of the blade root groove is 831 MPa, which occurs at the concave fillet below the bearing surface of the first pair of teeth of the groove. The reconstruction deviation at this location is 1.11%. The finite element analysis results of the equivalent stress along the blade root groove profile are in good agreement with the reconstruction results.

[0232] Considering the stress level as the design target, the design scheme of T6 is only better than T4 among the six samples. The concave and convex fillet radii of its blade root and wheel groove are both smaller, and the tooth width control parameter is also the smallest among the six samples.

[0233] like Figure 17 As shown in the figure, the maximum absolute deviation of the reconstructed field is 49 MPa, and the maximum equivalent stress of the blade root groove is 907 MPa, which occurs at the concave fillet above the bearing surface of the first pair of teeth at the blade root. The reconstruction deviation at this location is 0.63%. The finite element analysis results of the equivalent stress along the blade root groove profile are in good agreement with the reconstruction results.

[0234] In the structural design of blade root groove profiles, the maximum equivalent stress is often the most concerned indicator, so it is necessary to evaluate the prediction accuracy of the model for the maximum equivalent stress of the blade root groove. Figure 18 The evaluation results of the model's maximum equivalent stress prediction accuracy for three random batches of samples (batch sample size is 64) in the turbine blade root groove test dataset are given. Figure 18 (a) is a maximum equivalent stress scatter plot, where the horizontal axis of the stress scatter plot is the result of finite element analysis, and the vertical axis is the result of the reconstructed model. The closer the stress scatter plot is to the straight line y=x, the higher the prediction accuracy of the reconstructed model. It can be seen from the figure that the maximum equivalent stress prediction deviation of the reconstructed model of the gas turbine blade root groove stress field is within 5%; Figure 18 (b) is the relative deviation distribution of the maximum equivalent stress prediction results. It can also be found that the maximum equivalent stress prediction deviations of the blade root groove in the random batch samples are all within 5%, and the median deviation is 0.46%.

[0235] The deviation analysis results of the above six samples and the maximum equivalent stress show that the stress field reconstruction model of the turbine blade root groove based on GCCNN can accurately reconstruct the displacement, strain and stress field of the profiles with different design parameters in the test data set, and accurately predict the maximum equivalent stress of the blade root groove.

[0236] Based on the same concept, the present invention also provides a gas turbine blade root groove stress field reconstruction system, which includes an acquisition module, a calculation module, a training module and a reconstruction module.

[0237] The acquisition module is used to construct multiple finite element models of the blade root groove of the gas turbine turbine blade under different profile design schemes, and to collect multiple sets of design variables, multiple sets of measurement point data and graphic data corresponding to the multiple finite element models; in the finite element model, the blade root groove profile area uses a structured grid, and the internal area of ​​the blade root uses an unstructured grid.

[0238] The calculation module is used to perform finite element calculation on multiple finite element models to obtain multiple first stress fields corresponding to structured grids and multiple second stress fields corresponding to unstructured grids.

[0239] The training module is used to construct a stress field reconstruction model, which includes a parallel convolutional neural network and a GMM graph convolutional neural network; the convolutional neural network is trained with multiple sets of design variables and multiple sets of measurement point data as input and multiple first stress fields as output; the graph convolutional neural network is trained with multiple sets of design variables, multiple sets of measurement point data and graph data as input and multiple second stress fields as output.

[0240] The reconstruction module is used to obtain the design variables, measurement point data and map data of the blade root groove of the gas turbine turbine blade in real time, and input them into the trained stress field reconstruction model to obtain the reconstructed blade root groove stress field.

[0241] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0242] Obviously, those skilled in the art may make various modifications and variations to the present invention without departing from the spirit and scope of the present invention. Thus, if such modifications and variations fall within the scope of the claims and their equivalents, the present invention is intended to include such modifications and variations.

Claims

1. A method for reconstructing the stress field of a blade root groove of a gas turbine turbine blade, characterized in that: The following steps are involved: Construct multiple finite element models of gas turbine blade root grooves under different profile design schemes, and collect multiple sets of design variables, multiple sets of measurement point data, and graphic data corresponding to the multiple finite element models; In the finite element model, the root wheel groove line area uses a structured grid, and the inner area of ​​the blade root uses an unstructured grid; Performing finite element calculations on the multiple finite element models to obtain multiple first stress fields corresponding to the structured grids and multiple second stress fields corresponding to the unstructured grids; Constructing a stress field reconstruction model, wherein the stress field reconstruction model includes a parallel convolutional neural network and a GMM graph convolutional neural network; The convolutional neural network is trained with multiple sets of design variables and multiple sets of measurement point data as input and multiple first stress fields as output; The graph convolutional neural network is trained with multiple sets of design variables, multiple sets of measurement point data and graph data as input and multiple second stress fields as output; The design variables, measurement point data and map data of the blade root groove of the gas turbine turbine blade are obtained in real time, and are input into the trained stress field reconstruction model to obtain the reconstructed blade root groove stress field.

2. The method for reconstructing the stress field of a blade root groove of a gas turbine according to claim 1, characterized in that: The convolutional neural network includes 1 input layer, 3 upsampling layers and 1 output layer. The input layer includes 2 fully connected layers. Each upsampling layer includes 1 interpolation layer, 1 batch normalization layer and 2 first convolution layers. The output layer is 1 second convolution layer. The training of the convolutional neural network specifically includes the following steps: Combine the design variables and the measurement point data to obtain the input set; The input layer linearly transforms and reshapes the input set to obtain the feature map tensor; The feature map tensor is interpolated, batch normalized, and convolved once through the upsampling layer to obtain the output feature map; The output feature map is convolved twice through the output layer to obtain the reconstructed first stress field.

3. The method for reconstructing the stress field of a blade root groove of a gas turbine according to claim 1, characterized in that: The graph data is specifically as follows: D = {A', B', X}; Where D is the graph data, A' and B' are the adjacency matrix and degree matrix of the unstructured grid, and X is the information matrix of the unstructured grid nodes.

4. A method for reconstructing the stress field of a blade root groove of a gas turbine according to claim 3, characterized in that: The graph convolutional neural network adopts the GMM graph convolutional neural network, which includes 1 input layer, 5 intermediate layers and 1 output layer. The input layer and the intermediate layers all use Gaussian deviation linear unit as the activation function, and the output layer does not use an activation function; The training of the graph convolutional neural network specifically includes the following steps: Weighting of design variables, measurement point data and graph data through the input layer; The nonlinearity is added to the weighted data through the intermediate layer to obtain the reconstructed second stress field; The reconstructed second stress field is output through the output layer.

5. The method for reconstructing the stress field of a blade root groove of a gas turbine according to claim 1, characterized in that: When training convolutional neural networks and graph convolutional neural networks, network parameters are optimized by minimizing sub-loss functions, which are specifically as follows: L T =L F +L S +L gx +L gy ; Where, L T is the sub-loss function, L F is the loss function of the global stress field, L S is the loss function of the field maximum, L gx and L gy is the loss function of the global stress field gradient in the x and y directions.

6. The method for reconstructing the stress field of a blade root groove of a gas turbine according to claim 1, characterized in that: The stress field reconstruction model is trained as a whole through the total loss function, which is specifically as follows: L C =L T,1 +L T,2 +L′; Where, L C is the total loss function, L T,1 With L T,2 is the sub-loss function of the convolutional neural network and the graph convolutional neural network, and L′ is the loss function of the convolutional neural network and the graph convolutional neural network for the predicted values ​​of the common nodes of the structured grid and the unstructured grid.

7. The method for reconstructing the stress field of a blade root groove of a gas turbine according to claim 1, characterized in that: The measuring point data includes temperature data and displacement data.

8. A gas turbine blade root groove stress field reconstruction system, characterized in that: include: The acquisition module is used to construct multiple finite element models of the blade root groove of the gas turbine turbine blade under different profile design schemes, and to collect multiple sets of design variables, multiple sets of measurement point data and graphic data corresponding to the multiple finite element models; In the finite element model, the root wheel groove line area uses a structured grid, and the inner area of ​​the blade root uses an unstructured grid; A calculation module, configured to perform finite element calculations on the multiple finite element models to obtain multiple first stress fields corresponding to the structured grids and multiple second stress fields corresponding to the unstructured grids; A training module is used to construct a stress field reconstruction model, wherein the stress field reconstruction model includes a parallel convolutional neural network and a GMM graph convolutional neural network; The convolutional neural network is trained with multiple sets of design variables and multiple sets of measurement point data as input and multiple first stress fields as output; The graph convolutional neural network is trained with multiple sets of design variables, multiple sets of measurement point data and graph data as input and multiple second stress fields as output; The reconstruction module is used to obtain the design variables, measurement point data and map data of the blade root groove of the gas turbine turbine blade in real time, and input them into the trained stress field reconstruction model to obtain the reconstructed blade root groove stress field.

Images