Method and system for predicting two-stage topological configuration of constrained damping plate shell based on neural network

Abstract

The invention discloses a constraint damping plate shell two-stage topological configuration prediction method and system based on a neural network. The method comprises the following steps: constructing a Res-CB-Unet neural network model; acquiring substrate unit strain energy of the constrained damping structure, acquiring an optimal topological configuration of the constrained damping structure by using an SIMP model, and constructing a training data set based on the substrate unit strain energy and the optimal topological configuration; performing two-stage training on the Res-CB-Unet neural network model based on the training data set to obtain a configuration prediction model; and based on the configuration prediction model, performing topological configuration prediction on the to-be-predicted constrained damping structure. According to the method, efficient prediction of the topological configuration of the constrained damping plate shell can be realized under the condition that comprehensive optimization calculation is not carried out, so that the optimization process is accelerated, the prediction precision is improved, compared with a traditional topological optimization method, the calculation efficiency can be remarkably improved while the prediction precision is improved, and the method has an obvious time advantage.

Description

Technical Field

[0001] The invention belongs to the technical field of constrained damping structure design, and in particular relates to a two-stage topological configuration prediction method and system for constrained damping plates and shells based on a neural network. Background Art

[0002] The constrained layer damping structure mainly relies on the shear deformation of viscoelastic damping materials to consume vibration energy and realize the vibration and noise control of the whole plate and shell. It has the advantages of simple structure, convenient engineering technology application, significant vibration and noise reduction effect, and high reliability. It has been widely used in the vibration and noise reduction design of load-bearing plates and shells in large ships, aerospace, and automobiles. The constrained damping structure optimization design methods developed so far have met the engineering requirements for lightweight design of structures, but the optimization time is too long and the topology optimization model needs to be rebuilt, which is obviously not conducive to long-term engineering applications. Therefore, the development of a fast prediction method for the optimal topological configuration of a constrained damping structure has important engineering application significance.

[0003] The optimization design research of constrained damping plate and shell structures has made significant progress. Most of the objective functions are set as peak response and loss factor, and the geometric dimensions and material properties of the constrained damping materials are optimized to achieve lightweight and high efficiency of the structure. Ansari et al. optimized the distribution and geometric shape of damping materials in constrained damping thin plates based on the level set method, and the experimental results further verified the effectiveness of their numerical simulation; Wang Mingxu et al. used the solid isotropic microstructure penalty (SIMP) model to perform topological optimization on constrained damping plate and shell structures to maximize the modal loss factor; Dong Li et al. proposed a dual-material hybrid topology optimization method, which introduced the material mixing ratio as a design variable to achieve multi-material and structural integrated topology optimization based on the SIMP method, and the optimization results simultaneously considered the material distribution and mixing ratio; Wu Yonghui et al. realized the topology optimization of the constrained damping (CLD) structure by maximizing the modal loss factor based on the finite element model and the parameterized level set method (PLSM), and explored the influence of material thickness changes on the modal loss factor and the optimal configuration. The above literature shows that the topological optimization of the current constrained damping structure can effectively achieve structural lightweight and significantly suppress vibration response. However, when the size of the base plate and the size parameters of the constrained damping material of the constrained damping plate and shell structure change, it is necessary to re-establish the numerical model for optimization design, resulting in low computational efficiency and difficulty in applying to the generalized fast topological configuration design of the constrained damping plate and shell structure. Therefore, it is necessary to propose a fast prediction method and system for the topological configuration of similar structures of constrained damping plates and shells. Summary of the invention

[0004] The present invention aims to solve the deficiencies of the prior art and provides the following solutions:

[0005] A two-stage topological configuration prediction method for constrained damping plates and shells based on neural networks includes the following steps:

[0006] Construct the Res-CB-Unet neural network model;

[0007] Obtaining the strain energy of a substrate unit of a constrained damping plate-shell structure, obtaining the optimal topological configuration of the constrained damping structure using a SIMP model, and constructing a training data set based on the substrate unit strain energy and the optimal topological configuration;

[0008] Performing two-stage training on the Res-CB-Unet neural network model based on the training data set to obtain a configuration prediction model;

[0009] Based on the configuration prediction model, the topological configuration prediction is performed on the constrained damping structure to be predicted.

[0010] Preferably, the Res-CB-Unet neural network model includes: an encoding layer, a convolutional attention layer, a decoding layer, a concatenation layer and a residual skip layer;

[0011] The encoding layer is used to convert the mechanical information of the constrained damping structure into a tensor and perform feature extraction to obtain a feature tensor;

[0012] The convolutional attention layer includes: channel attention and spatial attention, which are used to improve the ability to distinguish the feature tensor;

[0013] The decoding layer is used to reorganize the feature tensor to obtain an output graph;

[0014] The splicing layer is used to connect the encoding layer and the decoding layer;

[0015] The residual jump is used for the internal connection between the encoding layer and the decoding layer to retain the shallow low-level features in the deep layer.

[0016] Preferably, the substrate unit strain energy is:

[0017] MSE=U T ×K p ×U

[0018] Where MSE represents the unit strain energy of the substrate in the constrained damping plate shell with constrained damping material, U represents the displacement vector of the substrate, and K p represents the stiffness matrix of the base unit, and T represents the transpose of the matrix.

[0019] Preferably, the method for training the Res-CB-Unet neural network model comprises:

[0020] Mapping the plane diagram of the substrate unit strain energy onto a blank background diagram in equal proportion to obtain a mapping image;

[0021] Taking the mapping image as input and the optimal topological configuration as output, training the Res-CB-Unet neural network model to obtain an intermediate configuration prediction model;

[0022] Using the intermediate configuration prediction model to predict the constrained damping plate-shell structure corresponding to the mapping image, to obtain a suboptimal topological configuration;

[0023] The intermediate configuration prediction model is trained with the suboptimal topological configuration as input and the optimal topological configuration as output to obtain the configuration prediction model.

[0024] The present invention also provides a two-stage topological configuration prediction system for constrained damping plates and shells based on a neural network, wherein the prediction system applies any of the prediction methods described above, and comprises: a model construction module, a data set construction module, a model training module and a configuration prediction module;

[0025] The model building module is used to build a Res-CB-Unet neural network model;

[0026] The data set construction module is used to obtain the substrate unit strain energy of the constrained damping structure, obtain the optimal topological configuration of the constrained damping structure using the SIMP model, and construct a training data set based on the substrate unit strain energy and the optimal topological configuration;

[0027] The model training module performs two-stage training on the Res-CB-Unet neural network model based on the training data set to obtain a configuration prediction model;

[0028] The configuration prediction module performs topological configuration prediction on the constrained damping structure to be predicted based on the configuration prediction model.

[0029] Preferably, the Res-CB-Unet neural network model includes: an encoding layer, a convolutional attention layer, a decoding layer, a concatenation layer and a residual skip layer;

[0030] The encoding layer is used to convert the mechanical information of the constrained damping structure into a tensor and perform feature extraction to obtain a feature tensor;

[0031] The convolutional attention layer includes: channel attention and spatial attention, which are used to improve the ability to distinguish the feature tensor;

[0032] The decoding layer is used to reorganize the feature tensor to obtain an output graph;

[0033] The splicing layer is used to connect the encoding layer and the decoding layer;

[0034] The residual jump is used for the internal connection between the encoding layer and the decoding layer to retain the shallow low-level features in the deep layer.

[0035] Preferably, the substrate unit strain energy is:

[0036] MSE=U T ×K p ×U

[0037] Where MSE represents the unit strain energy of the substrate in the constrained damping plate shell with constrained damping material, U represents the displacement vector of the substrate, and K p represents the element stiffness matrix of the base plate, and T represents the transpose of the matrix.

[0038] Preferably, the model training module includes: a mapping unit, a first-stage training unit, a suboptimal configuration prediction unit, and a second-stage training unit;

[0039] The mapping unit maps the plane diagram of the strain energy of the substrate unit of the existing constrained damping structure onto a blank background diagram in equal proportion to obtain a mapping image;

[0040] The first-stage training unit uses the mapping image as input and the optimal topological configuration as output to train the Res-CB-Unet neural network model to obtain an intermediate configuration prediction model;

[0041] The suboptimal configuration prediction unit predicts the existing constrained damping structure corresponding to the mapping image using the intermediate configuration prediction model to obtain a suboptimal topological configuration;

[0042] The second-stage training unit uses the suboptimal topological configuration as input and the optimal topological configuration as output to train the intermediate configuration prediction model to obtain the configuration prediction model.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] The present invention can achieve efficient prediction of the topological configuration of constrained damping plates and shells without performing comprehensive optimization calculations, thereby accelerating the optimization process and improving prediction accuracy. Compared with traditional topological optimization methods (such as the SIMP method), it can significantly improve calculation efficiency while improving prediction accuracy, and has obvious time advantages. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0046] Figure 1 A schematic diagram of a method flow of an embodiment of the present invention;

[0047] Figure 2 A schematic diagram of the Res-CB-Unet neural network structure of an embodiment of the present invention;

[0048] Figure 3 Schematic diagram of the CBAM layer structure of an embodiment of the present invention;

[0049] Figure 4 It is a schematic structural diagram of a finite element dynamics model of a constrained damping plate-shell structure according to an embodiment of the present invention. DETAILED DESCRIPTION

[0050] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0051] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0052] Embodiment 1

[0053] In this embodiment, if Figure 1 As shown, a two-stage topological configuration prediction method for constrained damping plates and shells based on neural networks includes the following steps:

[0054] S1. Construct the Res-CB-Unet neural network model.

[0055] The Res-CB-Unet neural network model includes: encoding layer, convolutional attention layer, decoding layer, splicing layer and residual jump layer; the encoding layer is used to convert the mechanical information of the constrained damping structure into a tensor and perform feature extraction to obtain a feature tensor; the convolutional attention layer includes: channel attention and spatial attention, which are used to improve the ability to distinguish feature tensors; the decoding layer is used to reorganize the feature tensor to obtain an output graph; the splicing layer is used to connect the encoding layer and the decoding layer; the residual jump is used for the internal connection of the encoding layer and the decoding layer to retain shallow low-level features in deep layers.

[0056] In this embodiment, the components of the Res-CB-UNet neural network are Encoder (encoding layer), CBAM layer (CBAM: Convolutional Block Attention Module, convolutional attention module), Decoder (decoding layer), Crop (splicing layer), and a residual skip layer is added. The residual skip layer allows the low-level features of the shallow layer (such as edges, textures, etc.) to be preserved in the deep layer, thereby helping the model to better understand the details of the image. The constructed neural network structure is as follows Figure 2 shown.

[0057] The encoding layer converts the appropriate structural mechanics information of the constrained damping plate into a tensor, and extracts features through the input layer, convolution layer and downsampling. The tensor undergoes several padding, convolution, normalization (Batch Normalization), activation function and downsampling (the size of the tensor is further reduced after pooling) in the encoding layer to extract features at different levels. The size of the convolution kernel in the encoding layer is set to 3×3, and the step size of the downsampling is set to 2, so that the size of the output tensor is reduced to 1 / 64 of the input tensor.

[0058] The decoding layer is responsible for reorganizing the feature tensor output by the encoding layer, which includes convolutional layers, upsampling, splicing, and output layers. In the decoding layer, multiple padding and interpolation deconvolution operations are performed, and the image is restored to a larger size after splicing and upsampling, where the size of the convolution kernel is the same as that of the encoding layer. The splicing mechanism fuses part of the down-sampled information with the up-sampled features, retaining some features of the original image and further enhancing the learning ability of the network. Finally, the output image generated by the decoding layer maintains the same size as the original topological configuration.

[0059] In order to enhance the efficiency of information transfer in the topology optimization process, the input channels are increased and the information is dispersed; in order to improve the information correlation between channels, a CBAM layer is added between the encoding layer and the decoding layer, such as Figure 3 As shown in the figure. This module includes the SAM module (channel attention) and the CAM module (spatial attention), with the aim of improving the ability of the multi-channel input model to distinguish features; the SAM module includes operations such as adaptive pooling (adaptive maximum pooling, adaptive average pooling), convolution, and activation function (Relu) to capture the weight differences between channels; the CAM module uses the Sigmoid activation function to smoothly complete the activation process and nonlinear mapping of the neural network; in addition, the channel superposition method is used instead of element-by-element addition to better mine the weight differences between pixels.

[0060] S2. Obtain the strain energy of the substrate unit of the constrained damping plate-shell structure, obtain the optimal topological configuration of the constrained damping structure using the SIMP model, and construct a training data set based on the substrate unit strain energy and the optimal topological configuration.

[0061] In this embodiment, a finite element dynamics model of a constrained damping structure is first constructed, such as Figure 4 As shown, it is divided into a constraint layer, a damping layer, and a substrate layer. Each unit has four nodes, and each node has 7 degrees of freedom, which are the displacement u along the x and y directions in the neutral plane of the constraint layer c ; Displacement u along the x and y directions in the neutral plane of the substrate layer b ; The lateral displacement w of the composite unit and the deflections of the x-axis and y-axis. Based on vibration mechanics and plate-shell theory, the shear modulus of the damping layer material is set as a complex constant model, then the finite element dynamics model of the constrained damping structure is:

[0062]

[0063] Where X represents the generalized displacement vector of all nodes, M represents the global mass matrix, and K R represents the real part of the global stiffness matrix, K I represents the imaginary part of the global stiffness matrix, represents the generalized acceleration vector of all nodes, and j represents the imaginary unit. According to the above formula, the characteristic equation of free vibration of the constrained damping plate and shell structure can be obtained:

[0064]

[0065] Among them, in the formula: represents the r-th order complex eigenvalue, Represents the r-th order complex eigenvector. Based on the modal loss factor, the vibration suppression capability of the constrained damping structure is characterized, and the r-th order modal loss factor is defined as:

[0066]

[0067] Among them, ψ r represents the r-th order real mode shape vector corresponding to the complex eigenvector, and T represents the transpose of the matrix. The topological optimization configuration of the constrained damping material with the modal loss factor as the objective function has a certain similarity with the MSE distribution of the substrate; and the strain energy distribution of the base structure is easier to obtain, and the construction of input information is convenient and simple. Therefore, the unit strain energy of the substrate is:

[0068] MSE=U T ×K p ×U

[0069] Where MSE represents the strain energy of the substrate unit in the constrained damping plate shell with constrained damping material, U represents the displacement vector of the substrate, and K p represents the stiffness matrix of the base unit, and T represents the transpose of the matrix.

[0070] Afterwards, the SIMP model is used to obtain the optimal topological configuration of the constrained damping structure: Based on the SIMP interpolation model, the mass matrix and stiffness matrix of the constrained damping structure can be written as:

[0071]

[0072] Where M represents the mass matrix of the constrained damping structure, K represents the stiffness matrix of the constrained damping structure, and M b , K b , They represent the global mass matrix of the substrate layer, the global stiffness matrix of the substrate layer, the unit mass matrix of the substrate layer, and the unit stiffness matrix of the substrate layer respectively; M v , K R , K I , They represent the global mass matrix of the damping layer, the real part of the global stiffness matrix of the constrained damping structure, the imaginary part of the global stiffness matrix of the constrained damping structure, the real part of the shear stiffness matrix of the constrained damping structural unit, and the imaginary part of the shear stiffness matrix of the constrained damping structural unit; M c , K c , They represent the global mass matrix of the constraint layer, the global stiffness matrix of the constraint layer, the element mass matrix of the constraint layer, and the element stiffness matrix of the constraint layer respectively; represents the unit mass matrix of the damping layer, K v-R represents the real part of the global stiffness matrix of the damping layer; n represents the number of elements; x i Indicates the existence state of the i-th constrained damping structure unit, which refers to the unit density here; p represents the penalty factor, and p=3 is selected in this embodiment. Taking the unit density of the constrained damping structure as the design variable, the maximization of the modal loss factor as the design goal, and the material consumption as the constraint condition, a mathematical model for topological optimization of the constrained damping structure is constructed as follows:

[0073] find:x i ,(i=1,2,…,n)

[0074] Min:

[0075]

[0076] s¨t

[0077]

[0078] 0 <x min ≤x i ≤1

[0079] Among them, ξ(x) is the objective function; η r is the modal loss factor, is the modal order; x i is the existence state of the i-th constrained damping material, which refers to the design variable. x = 1 means that the damping composite material exists completely, and x = 0 means that only the substrate unit is included; V 0 represents the volume of the substrate layer surface fully covered with constrained damping material; x min represents the minimum value of unit density; f represents the volume fraction of the constrained damping material. Finally, the mathematical model of topological optimization of constrained damping structure is used to perform topological optimization on the constrained damping structure and obtain the optimal topological configuration.

[0080] In this embodiment, the input data in the training data set is composed of a two-dimensional image of the strain energy of the substrate unit; when constructing the input data set image, the parameters of the constrained damping structure are set: the constraint mode is four-side clamping, the substrate size of the constrained damping plate: the width is 0.4m-0.3m, the length is 0.5m-0.3m, the damping layer thickness is 0.3mm-0.9mm, and the constraint layer thickness is 0.1mm-0.7mm; the substrate width and length, the damping layer thickness, and the constraint layer thickness are uniformly sampled by a uniform sampling method, and then 300 constrained damping plates of different sizes and specifications are randomly combined; finite element analysis is performed on these 300 constrained damping plates to obtain the modal strain energy of the substrate and convert it into a two-dimensional image. When constructing the output data set image, the SIMP model is used to perform topological optimization on these 300 constrained damping plates to obtain the optimal topological configuration, and it is converted into a two-dimensional image as the output data of the data set.

[0081] S3. Based on the training data set, the Res-CB-Unet neural network model is trained in two stages to obtain the configuration prediction model.

[0082] The method for training the Res-CB-Unet neural network model includes: mapping the plane diagram of the substrate unit strain energy onto a blank background image in proportion to obtain a mapping image; training the Res-CB-Unet neural network model with the mapping image as input and the optimal topological configuration as output to obtain an intermediate configuration prediction model; using the intermediate configuration prediction model to predict the constrained damping structure corresponding to the mapping image to obtain a suboptimal topological configuration; training the intermediate configuration prediction model with the suboptimal topological configuration as input and the optimal topological configuration as output to obtain a configuration prediction model.

[0083] In order to solve the problem that the pixels of the images of the substrate unit strain energy and the optimal topological configuration formed by constrained damping plates of different sizes and specifications are different, the proportional mapping method is then used to map the obtained two-dimensional image to a blank background of 384×512. In this way, the pixel ratios of the images of the input information and the output information can be kept consistent in any damping plate size, and both the input information and the output information are converted into encoding tensors.

[0084] Based on a computer equipped with an Intel(R) Core(TM) i5-12500H CPU and an Nvidia 3060 GPU, the Pytorch framework was used to train the Res-CB-UNet neural network using the Adam algorithm with the substrate unit strain energy as input information and the optimal topological configuration as output information. The initial learning rate was set to 0.001. If the loss value did not show a downward trend after 20 consecutive training rounds, the learning rate was reduced to 0.1 times the original value. In order to evaluate the prediction effect of the neural network model, the cross entropy loss function value was used to characterize the prediction accuracy. With the increase of training rounds, the loss value of the training data set gradually decreased; after the training round was greater than 207, the loss value of the test set tended to stabilize and converged to around 0.1036, while the loss value of the training set was still slowly decreasing, reaching 0.0894 at the end of training. The neural network model at the training round of 205 was selected as the intermediate configuration prediction model. In the 205th training round, the strain energy cloud map of the substrate unit of the validation set was imported into the intermediate configuration prediction model as input information to obtain the suboptimal topological configuration. For the constrained damping structure, in order to improve the obvious limitations of the Res-CB-UNet neural network with single input information in processing edge positions and details, the second stage of neural network training was carried out: in this stage of training, the intermediate configuration prediction model was used to continue to learn the detailed information of the optimized configuration of the constrained damping structure, and the suboptimal topological configuration obtained in the first stage of training was used as input information and the optimal topological configuration was used as output information to establish a data set for the second training. Then, the Adam algorithm was used to train the intermediate configuration prediction model in the second stage: the initial learning rate was set to 0.001. If the loss value did not show a downward trend after 20 consecutive training rounds, the learning rate was decayed to 0.1 times the original value to ensure that the model maintained a stable optimization direction during the training process. In order to evaluate the prediction effect of the neural network model, the cross entropy loss function value was still used to characterize the prediction accuracy. After the training was completed, the configuration prediction model was obtained.

[0085] S4. Based on the configuration prediction model, the topological configuration prediction of the constrained damping structure to be predicted is performed.

[0086] Embodiment 2

[0087] In this embodiment, the accuracy of the constructed model is evaluated.

[0088] Before evaluating the prediction accuracy of the model, its stability needs to be examined first. The input information of 300 samples is loaded into the trained neural network to obtain the corresponding prediction image, the cross entropy loss value of the prediction result and the output information is calculated, and the loss value distribution graph is established based on this. According to the loss value distribution graph, the loss values ​​of the vast majority of samples are concentrated in areas with smaller values, and the loss values ​​of a small number of samples are relatively large, indicating that the prediction effect is poor, but it does not affect the overall prediction quality of the model. The neural network model after two-stage training has a more concentrated distribution of loss values ​​when predicting the optimal configuration, showing good stability, which is significantly better than the neural network model that has only undergone a single-stage training.

[0089] The topology optimization evaluation indicators of constrained damping structure include pixel value error, modal loss factor error and computational efficiency. The specific evaluation results are shown in Table 1. Compared with the SIMP method, the prediction results of the single-stage neural network have an average reduction of 11.6%, -3.79% and 96.78% in pixel value error, quality score error and computational time of any 100 samples, respectively; while the prediction results of the two-stage neural network have an average reduction of 5.71%, -2.84% and 94.53% compared with the SIMP method, respectively.

[0090] Table 1

[0091]

[0092] In summary, based on the SIMP operation results, the optimal topological configuration formed by the two-stage Res-CB-UNet neural network performs better in terms of pixel value error, especially at the edges of the structure. The prediction results of the two-stage neural network show clearer details, which are more consistent with the SIMP operation results. However, there is a decrease in the quality score error, mainly due to the reduction of discontinuities and the increase of details in the prediction results. The quality score errors of the single-stage and two-stage prediction results are lower than the benchmark values ​​of the SIMP method. The prediction model error of the two-stage prediction method is smaller, so the impact on the modal loss factor is also smaller. The computational efficiency is slightly reduced due to the secondary import of the weight file. Considering the significant improvement in accuracy, the overall performance of the two-stage prediction method is still better than that of the single-stage prediction method.

[0093] Embodiment 3

[0094] In this embodiment, a two-stage topological configuration prediction system for constrained damping plates and shells based on a neural network includes: a model construction module, a data set construction module, a model training module and a configuration prediction module.

[0095] The model building module is used to build the Res-CB-Unet neural network model.

[0096] The Res-CB-Unet neural network model includes: encoding layer, convolutional attention layer, decoding layer, splicing layer and residual jump layer; the encoding layer is used to convert the mechanical information of the constrained damping structure into a tensor and perform feature extraction to obtain a feature tensor; the convolutional attention layer includes: channel attention and spatial attention, which are used to improve the ability to distinguish feature tensors; the decoding layer is used to reorganize the feature tensor to obtain an output graph; the splicing layer is used to connect the encoding layer and the decoding layer; the residual jump is used for the internal connection of the encoding layer and the decoding layer to retain shallow low-level features in deep layers.

[0097] The data set construction module is used to obtain the substrate unit strain energy of the constrained damping structure, obtain the optimal topological configuration of the constrained damping structure using the SIMP model, and construct a training data set based on the substrate unit strain energy and the optimal topological configuration.

[0098] The strain energy of the substrate unit is:

[0099] MSE=U T ×K p ×U

[0100] Where MSE represents the substrate unit strain energy, U represents the displacement vector, and K p represents the element stiffness matrix and T represents the transpose of the matrix.

[0101] The model training module performs two-stage training on the Res-CB-Unet neural network model based on the training data set to obtain the configuration prediction model.

[0102] The model training module includes: a mapping unit, a first-stage training unit, a suboptimal configuration prediction unit and a second-stage training unit; the mapping unit maps the plane diagram of the strain energy of the substrate unit of the existing constrained damping structure to a blank background image in proportion to obtain a mapping image; the first-stage training unit uses the mapping image as input and the optimal topological configuration as output to train the Res-CB-Unet neural network model to obtain an intermediate configuration prediction model; the suboptimal configuration prediction unit uses the intermediate configuration prediction model to predict the existing constrained damping structure corresponding to the mapping image to obtain a suboptimal topological configuration; the second-stage training unit uses the suboptimal topological configuration as input and the optimal topological configuration as output to train the intermediate configuration prediction model to obtain a configuration prediction model.

[0103] The configuration prediction module performs topological configuration prediction on the constrained damping structure to be predicted based on the configuration prediction model.

[0104] The above embodiments are only descriptions of the preferred embodiments of the present invention, and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should fall within the protection scope determined by the claims of the present invention.

Claims

1. A two-stage topological configuration prediction method for constrained damping plates and shells based on neural networks, characterized in that: The following steps are involved: Construct the Res-CB-Unet neural network model; Obtaining the strain energy of the substrate unit of the constrained damping structure, obtaining the optimal topological configuration of the constrained damping plate-shell structure using the SIMP model, and constructing a training data set based on the substrate unit strain energy and the optimal topological configuration; Performing two-stage training on the Res-CB-Unet neural network model based on the training data set to obtain a configuration prediction model; Based on the configuration prediction model, the topological configuration prediction is performed on the constrained damping plate-shell structure to be predicted.

2. According to the neural network-based two-stage topological configuration prediction method for constrained damping plates and shells according to claim 1, it is characterized in that: The Res-CB-Unet neural network model includes: an encoding layer, a convolutional attention layer, a decoding layer, a concatenation layer and a residual skip layer; The encoding layer is used to convert the mechanical information of the constrained damping plate-shell structure into a tensor and perform feature extraction to obtain a feature tensor; The convolutional attention layer includes: channel attention and spatial attention, which are used to improve the ability to distinguish the feature tensor; The decoding layer is used to reorganize the feature tensor to obtain an output graph; The splicing layer is used to connect the encoding layer and the decoding layer; The residual jump is used for the internal connection between the encoding layer and the decoding layer to retain the shallow low-level features in the deep layer.

3. According to the neural network-based two-stage topological configuration prediction method for constrained damping plates and shells in claim 1, it is characterized in that: The substrate unit strain energy is: MSE=U T ×K p ×U Where MSE represents the unit strain energy of the substrate in the constrained damping plate shell with constrained damping material, U represents the displacement vector of the substrate, and K p represents the stiffness matrix of the base unit, and T represents the transpose of the matrix.

4. According to the neural network-based two-stage topological configuration prediction method for constrained damping plates and shells in claim 1, it is characterized in that: The method for training the Res-CB-Unet neural network model includes: Mapping the plane diagram of the substrate unit strain energy onto a blank background diagram in equal proportion to obtain a mapping image; Taking the mapping image as input and the optimal topological configuration as output, training the Res-CB-Unet neural network model to obtain an intermediate configuration prediction model; Using the intermediate configuration prediction model to predict the constrained damping plate-shell structure corresponding to the mapping image, to obtain a suboptimal topological configuration; The intermediate configuration prediction model is trained with the suboptimal topological configuration as input and the optimal topological configuration as output to obtain the configuration prediction model.

5. A two-stage topological configuration prediction system for constrained damping plates and shells based on neural networks, wherein the prediction system applies the prediction method according to any one of claims 1 to 4, characterized in that: include: Model building module, data set building module, model training module and configuration prediction module; The model building module is used to build a Res-CB-Unet neural network model; The data set construction module is used to obtain the substrate unit strain energy of the constrained damping structure, obtain the optimal topological configuration of the constrained damping structure using the SIMP model, and construct a training data set based on the substrate unit strain energy and the optimal topological configuration; The model training module performs two-stage training on the Res-CB-Unet neural network model based on the training data set to obtain a configuration prediction model; The configuration prediction module performs topological configuration prediction on the constrained damping structure to be predicted based on the configuration prediction model.

6. The neural network-based two-stage topological configuration prediction system for constrained damping plates and shells according to claim 5, characterized in that: The Res-CB-Unet neural network model includes: an encoding layer, a convolutional attention layer, a decoding layer, a concatenation layer and a residual skip layer; The encoding layer is used to convert the mechanical information of the constrained damping structure into a tensor and perform feature extraction to obtain a feature tensor; The convolutional attention layer includes: channel attention and spatial attention, which are used to improve the ability to distinguish the feature tensor; The decoding layer is used to reorganize the feature tensor to obtain an output graph; The splicing layer is used to connect the encoding layer and the decoding layer; The residual jump is used for the internal connection between the encoding layer and the decoding layer to retain the shallow low-level features in the deep layer.

7. The neural network-based two-stage topological configuration prediction system for constrained damping plates and shells according to claim 5 is characterized in that: The substrate unit strain energy is: MSE=U T ×K p ×U Where MSE represents the unit strain energy of the substrate in the constrained damping plate shell with constrained damping material, U represents the displacement vector of the substrate, and K p represents the stiffness matrix of the base unit, and T represents the transpose of the matrix.

8. The neural network-based two-stage topological configuration prediction system for constrained damping plates and shells according to claim 5, characterized in that: The model training module includes: a mapping unit, a first-stage training unit, a suboptimal configuration prediction unit, and a second-stage training unit; The mapping unit maps the plane diagram of the strain energy of the substrate unit of the existing constrained damping structure onto a blank background diagram in equal proportion to obtain a mapping image; The first-stage training unit uses the mapping image as input and the optimal topological configuration as output to train the Res-CB-Unet neural network model to obtain an intermediate configuration prediction model; The suboptimal configuration prediction unit predicts the existing constrained damping structure corresponding to the mapping image using the intermediate configuration prediction model to obtain a suboptimal topological configuration; The second-stage training unit uses the suboptimal topological configuration as input and the optimal topological configuration as output to train the intermediate configuration prediction model to obtain the configuration prediction model.

Images