Neural network model-based reticulated shell structure node parameter automatic optimization method

Abstract

The invention relates to the technical field of space structure design and parameter optimization, in particular to a neural network model-based reticulated shell structure node parameter automatic optimization method, which comprises the following steps of: acquiring an input-output data pair of a single-layer cylindrical reticulated shell constructed by aluminum alloy plate type nodes, the input parameters comprise design parameters and initial node parameters of the single-layer cylindrical reticulated shell. According to the method, node optimization data under different parameters are obtained through interaction of a genetic algorithm, finite element software and a programming tool, an example is supplemented to construct a training data set covering a common design parameter range of a project, and then a three-layer full-connection neural network model containing two hidden layers is trained; and the model precision is ensured by matching an early stop strategy and a learning rate attenuation strategy. During application, target single-layer cylindrical reticulated shell design parameters are input, the model can automatically output optimized node parameters, repeated iteration and manual intervention of a traditional method are not needed, optimization time consumption is greatly shortened, and the multi-working-condition batch optimization requirement is efficiently met.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of spatial structure design and parameter optimization, and particularly relates to a net shell structure node parameter automatic optimization method based on a neural network model. BACKGROUND

[0002] As a kind of efficient space bearing structure, the net shell structure is widely used in large-span buildings such as stadiums, transportation hubs and exhibition centers due to good mechanical properties and flexible architectural forms. Among them, the aluminum alloy single-layer cylindrical net shell becomes an important choice for modern green buildings due to the advantages of light weight, high strength, corrosion resistance and low life cycle maintenance cost. The core stress unit of this kind of net shell is the node and the rod, and especially the plate-type node as a commonly used semi-rigid node form, the node parameters (including node vertical coordinates, node bending stiffness, etc.) directly affect the overall stress state of the net shell, the node coordinates determine the surface form of the net shell, and then affect the load transfer path; the node stiffness determines the semi-rigid characteristics of the net shell, and if the parameters are not reasonably set, the ultimate bearing capacity of the net shell is easily reduced or local stress concentration is caused. In actual design, the node parameters need to be optimized so that the net shell can maintain the macro-aesthetic form while realizing the thin film stress dominance (improving the rod axial force and reducing the bending moment), thereby enhancing the overall stability performance.

[0003] At present, the optimization of the node parameters of the aluminum alloy single-layer cylindrical net shell is mostly realized by the interactive mode of the genetic algorithm combined with the finite element software (such as ANSYS) and the programming tool (such as MATLAB). Specifically, the node parameter adjustment scheme is generated by the genetic algorithm iteration, and then the finite element model of the net shell is established by using ANSYS to calculate the bearing capacity of the net shell, and MATLAB and ANSYS are repeatedly interacted to select the optimal scheme. However, this method has significant technical limitations: in order to ensure the optimization accuracy, a large population size and iteration generation (such as a population size of 50 and an iteration of 50 generations) are usually required, and the time consumption of a single optimization calculation is extremely long - even if the time is reduced by avoiding repeated gene calculation, it still takes about 74 hours to complete the optimization of a typical model; if the number of net shell nodes increases and the number of parameter variables (such as the rise-span ratio, the length-span ratio and the load ratio combination) increases, the time consumption will increase sharply, which makes it difficult to meet the demand of rapid optimization in engineering design. More importantly, the whole process needs manual intervention in parameter setting and software interaction, and cannot realize the automatic output of the node parameters, and when facing the batch optimization demand under multiple working conditions (such as different load distribution and support conditions), the problem of low efficiency is more prominent, which restricts the engineering application efficiency of the aluminum alloy net shell structure. SUMMARY

[0004] To address the shortcomings of existing technologies, this invention provides an automatic optimization method for node parameters of a net-shell structure based on a neural network model. This method solves the problems of long optimization time, the need for manual intervention, the inability to automatically output parameters, and low efficiency in batch optimization under multiple operating conditions compared to existing technologies.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an automatic optimization method for node parameters of a reticulated shell structure based on a neural network model, comprising: S1. Collect the "input-output" data pairs of a single-layer cylindrical reticulated shell constructed with aluminum alloy plate nodes. The input parameters include the design parameters and initial node parameters of the single-layer cylindrical reticulated shell. The output parameters include the optimized node vertical coordinate adjustment value and node bending stiffness adjustment value. S2. Normalize the input parameters to eliminate dimensional differences and form a training dataset for the neural network model. S3. Construct a three-layer fully connected neural network model, which includes an input layer, a hidden layer, and an output layer. The hidden layer uses the ReLU activation function to enhance the model's non-linear fitting ability. S4. The neural network model is trained using the training dataset. The training includes dividing the training dataset into a training set, a validation set, and a test set, and adopting an early stopping strategy to avoid model overfitting. S5. Verify the neural network model. Verification includes comparing the optimized node parameters output by the model with the traditional optimization results. If the ultimate bearing capacity deviation of the single-layer cylindrical reticulated shell corresponding to the optimized node parameters is less than or equal to 5% and the maximum displacement deviation is less than or equal to 8%, then the model verification is successful. S6. Apply the neural network model to optimize node parameters. The application includes inputting the design parameters of the single-layer cylindrical reticulated shell. The neural network model automatically outputs the optimized node vertical coordinate adjustment value and node bending stiffness adjustment coefficient. S7. Based on the cubic spline interpolation method, the cross-sectional curve of the single-layer cylindrical shell is corrected according to the vertical coordinate adjustment value of the node, so that the corrected curve is smooth and continuous. S8. Import the optimized node parameters into the finite element software for bearing capacity verification. If the bearing capacity verification does not meet the engineering requirements, re-enter the parameters for fine-tuning and then optimize again.

[0006] Furthermore, the input parameters include the rise-to-span ratio, length-to-span ratio, half-span load ratio, support conditions, initial node x-coordinate, initial node y-coordinate, initial node bending stiffness, and grid shell span of the single-layer cylindrical reticulated shell. The rise-to-span ratio is set in the range of 1 / 6, 1 / 5, 1 / 4, and 1 / 3; the length-to-span ratio is set in the range of 1.4, 1.8, 2.2, and 2.6; the half-span load ratio is set in the range of 0, 1 / 4, 1 / 2, and 1; and the support conditions include two-sided support and four-sided support.

[0007] Furthermore, the "input-output" data pairs are derived from node optimization data under different parameters obtained through interactive optimization using a genetic algorithm combined with the finite element software ANSYS and the programming tool MATLAB. The training dataset is supplemented with 100 sets of reticulated shell optimization examples with different parameter combinations to ensure that it covers the range of common design parameters in engineering. The normalization process includes converting the node coordinate parameters and node bending stiffness parameters into relative values ​​to eliminate the influence of dimensional differences on the training results of the neural network model.

[0008] Furthermore, the number of nodes in the input layer is 8, and the 8 input parameters correspond in sequence to the span-to-span ratio, length-to-span ratio, half-span load ratio, support condition code value, initial node x-coordinate, initial node y-coordinate, initial node bending stiffness, and grid shell span of the reticulated shell.

[0009] Furthermore, in the three-layer fully connected neural network model, the hidden layer is set to two layers, wherein the number of nodes in the first hidden layer is 64 and the number of nodes in the second hidden layer is 32.

[0010] Furthermore, the number of nodes in the output layer is 2, corresponding to the optimized node y-coordinate adjustment value and node bending stiffness adjustment coefficient. The neural network model uses the mean square error function as the loss function to measure the deviation between the optimized parameters output by the model and the actual parameters, and adopts the Adam optimizer, with the learning rate of the Adam optimizer initially set to 0.001.

[0011] Furthermore, the training process divides the complete dataset into a training set, a validation set, and a test set in a 7:2:1 ratio, and the training rounds of the neural network model are set to 500 rounds.

[0012] Furthermore, the early stopping strategy is to stop training if the validation set loss does not decrease for 20 consecutive rounds, in order to avoid overfitting of the neural network model; the Adam optimizer adopts a learning rate decay strategy, specifically, the learning rate is decayed to 0.9 every 100 iterations to improve the stability of the model training.

[0013] Furthermore, the verification includes selecting typical cases from the test set. The input parameters of the typical cases include a combination of parameters for a reticulated shell model with a span-to-span ratio of 1 / 6, a length-to-span ratio of 2.6, a load ratio of 0, and four-sided support. The input parameters are then imported into the trained neural network model, and the optimized node parameters are output and compared with the optimization results of a traditional genetic algorithm. If the deviation of the ultimate bearing capacity of the reticulated shell or the maximum displacement deviation exceeds a preset range, the number of hidden layer nodes or the learning rate is adjusted, and the model is retrained until the accuracy requirements are met.

[0014] Furthermore, when the neural network model is used to optimize the node parameters, the design parameters of the target single-layer cylindrical reticulated shell include the rise-to-span ratio, length-to-span ratio, load ratio, support conditions, initial node parameters, and reticulated shell span information. The neural network model automatically outputs the optimized node y-coordinate adjustment value and node bending stiffness adjustment coefficient.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention first uses a genetic algorithm combined with finite element software and programming tools to interactively obtain node optimization data under different parameters. It then supplements this data with additional examples to construct a training dataset covering common engineering design parameter ranges. Next, it trains a three-layer fully connected neural network model with two hidden layers, employing early stopping and learning rate decay strategies to ensure model accuracy. When applied, the model automatically outputs optimized node parameters upon inputting the target single-layer cylindrical reticulated shell design parameters. This eliminates the need for repeated iterations and manual intervention required by traditional methods, significantly reducing optimization time and efficiently handling batch optimization needs across multiple working conditions. Simultaneously, it corrects the reticulated shell cross-sectional curve using cubic spline interpolation to ensure smoothness and continuity. The optimized parameters are then imported into finite element software for verification, guaranteeing that the ultimate bearing capacity and maximum displacement deviation of the reticulated shell meet engineering requirements. This avoids structural performance problems caused by unreasonable parameters, effectively improving the overall stability and engineering design efficiency of the aluminum alloy single-layer cylindrical reticulated shell. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the overall method of the present invention; Figure 2 This is a diagram of the neural network model architecture of the present invention; Figure 3 This is a flowchart of the training strategy of the present invention; Figure 4 This is a schematic diagram of the node curve optimization of the present invention; Figure 5 This is a flowchart illustrating the engineering application of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Please see Figures 1-5 This invention provides an automatic optimization method for node parameters of a reticulated shell structure based on a neural network model, comprising: S1. Collect the "input-output" data pairs of a single-layer cylindrical reticulated shell constructed with aluminum alloy plate nodes. The input parameters include the design parameters and initial node parameters of the single-layer cylindrical reticulated shell. The output parameters include the optimized node vertical coordinate adjustment value and node bending stiffness adjustment value. S2. Normalize the input parameters to eliminate dimensional differences and form a training dataset for the neural network model. S3. Construct a three-layer fully connected neural network model, which includes an input layer, a hidden layer, and an output layer. The hidden layer uses the ReLU activation function to enhance the model's non-linear fitting ability. S4. The neural network model is trained using the training dataset. The training includes dividing the training dataset into a training set, a validation set, and a test set, and adopting an early stopping strategy to avoid model overfitting. S5. Verify the neural network model. Verification includes comparing the optimized node parameters output by the model with the traditional optimization results. If the ultimate bearing capacity deviation of the single-layer cylindrical reticulated shell corresponding to the optimized node parameters is less than or equal to 5% and the maximum displacement deviation is less than or equal to 8%, then the model verification is successful. S6. Apply the neural network model to optimize node parameters. The application includes inputting the design parameters of the single-layer cylindrical reticulated shell. The neural network model automatically outputs the optimized node vertical coordinate adjustment value and node bending stiffness adjustment coefficient. S7. Based on the cubic spline interpolation method, the cross-sectional curve of the single-layer cylindrical shell is corrected according to the vertical coordinate adjustment value of the node, so that the corrected curve is smooth and continuous. S8. Import the optimized node parameters into the finite element software for bearing capacity verification. If the bearing capacity verification does not meet the engineering requirements, re-enter the parameters for fine-tuning and then optimize again.

[0019] Specifically, addressing the issues of traditional optimization methods relying on the interaction between genetic algorithms and the finite element software MATLAB, the time-consuming nature of each optimization cycle, the need for manual intervention, and the problems of uneven optimized reticulated shell curves and low verification accuracy due to simplified material models, this implementation method proceeds as follows: First, input-output data pairs are collected. Taking an aluminum alloy plate-node single-layer cylindrical reticulated shell as the object, the data collection logic is determined based on the optimization objective function and constraints. The objective function is:

[0020] (in Adjustment values ​​for the vertical coordinates of independent control points. For the span of the reticulated shell, (The scaling factor is recommended to be ≤1 / 100 to avoid changes in the macroscopic shape). An optimization program is written using a genetic algorithm, and the finite element software ANSYS is called to build the model. MATLAB is used to achieve automatic interactive iteration among the three to obtain multiple sets of node optimization data. At the same time, 100 sets of calculation examples covering different sag-to-span ratios, length-to-span ratios, load ratios, and support conditions are added to ensure that the data covers the range of common engineering design parameters.

[0021] Next, the input parameters are normalized, converting node coordinate parameters into relative values ​​relative to the shell span and node bending stiffness parameters into relative values ​​relative to the initial node bending stiffness, eliminating dimensional differences and forming a training dataset. Subsequently, a three-layer fully connected neural network model is constructed. The input layer receives 8 input parameters, and the hidden layers are set to two layers (64 nodes in the first layer and 32 nodes in the second layer). The ReLU activation function is used to enhance nonlinear fitting ability. The output layer outputs the adjusted values ​​of the node vertical coordinates and the node bending stiffness adjustment coefficients. During the training phase, the dataset is divided into training, validation, and test sets in a 7:2:1 ratio. An early stopping strategy is adopted (stopping if the validation set loss does not decrease for 20 consecutive rounds) to avoid overfitting. The Adam optimizer is used (initial learning rate 0.001, decaying to 0.9 every 100 rounds) to improve training stability.

[0022] During model validation, typical cases with a rise-to-span ratio of 1 / 6, a length-to-span ratio of 2.6, a load ratio of 0, and four-sided bracing were selected from the test set. The optimized parameters output by the model were compared with the results of the traditional genetic algorithm. In addition to checking that the deviation of the ultimate bearing capacity is ≤5% and the deviation of the maximum displacement is ≤8%, the formula for the internal force ratio of the members was also used.

[0023] ( axial force ratio, The bending moment ratio, , For actual axial force and bending moment, Non-proportional elongation strength, For cross-sectional area, For the moment of inertia, The rationality of the force is judged by the cross-sectional height. If the axial force ratio is concentrated and the bending moment ratio is reduced, the effectiveness of the model is further confirmed.

[0024] When applying the model, input the target reticulated shell design parameters, and the model automatically outputs the optimized adjusted values ​​of the nodal vertical coordinates and the nodal bending stiffness adjustment coefficients; the cross-sectional curve is corrected based on the cubic spline interpolation method, using the formula: ( , , ,,,, Piecewise curves were calculated (for curve coefficients) to ensure a smooth and continuous corrected curve. Finally, the optimized parameters were imported into finite element software for verification, using the Ramberg-Osgood material constitutive formula. ( In response, For stress, For elastic modulus, Set material properties for strain hardening parameters to ensure calculation accuracy; if they do not meet engineering requirements, fine-tune the input parameters and re-optimize.

[0025] Through this implementation method, the neural network model can automatically output optimized parameters without repeated iterations and manual intervention, greatly reducing the time required. At the same time, with the help of cubic spline interpolation and accurate material models, the aesthetics of the reticulated shell curve and the reliability of the calculation are ensured.

[0026] Furthermore, the input parameters include the rise-to-span ratio, length-to-span ratio, half-span load ratio, support conditions, initial node x-coordinate, initial node y-coordinate, initial node bending stiffness, and grid shell span of the single-layer cylindrical reticulated shell. The rise-to-span ratio is set in the range of 1 / 6, 1 / 5, 1 / 4, and 1 / 3; the length-to-span ratio is set in the range of 1.4, 1.8, 2.2, and 2.6; the half-span load ratio is set in the range of 0, 1 / 4, 1 / 2, and 1; and the support conditions include two-sided support and four-sided support.

[0027] Specifically, the selection of input parameters must strictly adhere to the preset range and be compatible with the input layer nodes of the neural network model and the actual engineering requirements. The rise-to-span ratio is selected from 1 / 6, 1 / 5, 1 / 4, and 1 / 3. For example, a 1 / 6 ratio can be used for a large-span convention center reticulated shell to balance space and stability. The length-to-span ratio is selected from 1.4, 1.8, 2.2, and 2.6. A 2.2 or 2.6 ratio can be used for a narrow and long transportation hub reticulated shell to adapt to the layout. The half-span load ratio is determined according to the actual load distribution. It is taken as 0 when only bearing the full-span dead load, and 1 / 4, 1 / 2, or 1 when bearing a combination of half-span live load and full-span dead load. The support conditions are divided into two longitudinal side supports (scenarios with reliable constraints on both sides) and four-sided supports (scenarios with constraints on all four sides). The initial node x-coordinate and initial node y-coordinate are determined according to the reticulated shell mesh division scheme. They need to be matched with the span and rise, and are converted to relative values ​​relative to the span during subsequent normalization. The initial node bending stiffness is calculated based on structural parameters such as the node plate thickness and the number of bolts. The reticulated shell span is determined according to the actual covered space size.

[0028] Clearly defining the range of values ​​for each parameter ensures that the input parameters conform to common engineering scenarios and provides a regular input for subsequent data normalization and model training, avoiding model fitting deviations caused by parameter confusion. At the same time, engineers can directly select parameters according to the range, reducing the difficulty of operation and indirectly improving optimization efficiency.

[0029] Furthermore, the "input-output" data pairs are derived from node optimization data under different parameters obtained through interactive optimization using a genetic algorithm combined with the finite element software ANSYS and the programming tool MATLAB. The training dataset is supplemented with 100 sets of reticulated shell optimization examples with different parameter combinations to ensure that it covers the range of common design parameters in engineering. The normalization process includes converting the node coordinate parameters and node bending stiffness parameters into relative values ​​to eliminate the influence of dimensional differences on the training results of the neural network model.

[0030] Specifically, the acquisition of "input-output" data pairs is based on traditional optimization methods, combined with the optimization objective function and constraints. A genetic algorithm program is written to construct the population initialization, selection, crossover, and mutation logic in MATLAB, and the ANSYS parametric language is used to build a reticulated shell model based on the objective function. The optimal node parameters are iteratively selected, and automatic interaction between MATLAB and ANSYS is achieved to obtain multiple sets of optimized data under different parameters. At the same time, 100 sets of calculation examples are added, covering different sag-to-span ratios, length-to-span ratios, load ratios and support conditions to ensure that the data covers common engineering scenarios.

[0031] During normalization, the node coordinate parameters are converted to relative values ​​relative to the span of the reticulated shell, and the node bending stiffness parameters are converted to relative values ​​relative to the initial node bending stiffness, eliminating dimensional differences. For example, when the node x-coordinate is 5m and the span is 30m, the normalized value is 5 / 30≈0.167. When the initial node bending stiffness is 4000kN·m / rad and the bending stiffness in a certain example is 4400kN·m / rad, the normalized value is 4400 / 4000=1.1. This avoids the impact of different parameter magnitudes on the model training accuracy and provides a reliable data foundation for the subsequent model to learn the mapping relationship between input and output.

[0032] Furthermore, the number of nodes in the input layer is 8, and the 8 input parameters correspond in sequence to the span-to-span ratio, length-to-span ratio, half-span load ratio, support condition code value, initial node x-coordinate, initial node y-coordinate, initial node bending stiffness, and grid shell span of the reticulated shell.

[0033] Specifically, the eight nodes in the input layer correspond to the key parameters of the reticulated shell in a fixed order to ensure that the model accurately resolves the input information. The first node corresponds to the sag-to-span ratio (e.g., 1 / 5), the second to the length-to-span ratio (e.g., 2.0), the third to the half-span load ratio (e.g., 1 / 4), the fourth to the support condition coding value (two longitudinal side supports are coded as 1, four side supports are coded as 2), the fifth to the initial node x-coordinate (normalized relative value), the sixth to the initial node y-coordinate (normalized relative value), the seventh to the initial node bending stiffness (normalized relative value), and the eighth to the reticulated shell span (normalized relative value).

[0034] This sequence is consistent with the parameter organization logic in the data acquisition and normalization stages. Engineers can directly input the parameters in sequence without adjusting the arrangement, thus avoiding deviations in model output due to incorrect order. At the same time, the clear node-parameter correspondence enables the model to accurately extract the features of each parameter, providing a correct foundation for subsequent hidden layer feature processing and output layer parameter calculation, and ensuring reliable optimization results.

[0035] Furthermore, in the three-layer fully connected neural network model, the hidden layer is set to two layers, wherein the number of nodes in the first hidden layer is 64 and the number of nodes in the second hidden layer is 32.

[0036] Specifically, the three-layer fully connected neural network model has two hidden layers, determined based on the complexity of the shell node parameter optimization problem. The first hidden layer has 64 nodes, which can fully extract the complex correlation features between the input parameters such as the span-to-span ratio, length-to-span ratio, support conditions, and node parameters, avoiding insufficient feature extraction due to too few nodes. The second hidden layer has 32 nodes, which can further filter and integrate the features extracted by the first layer, eliminate redundant information, and retain the features that are key to the optimization result, avoiding an increase in model complexity and training time due to too many nodes.

[0037] The combination of two hidden layers and the ReLU activation function can enhance the model's ability to fit nonlinear optimization problems, control the overall complexity of the model, enable rapid convergence during training, and quickly process input parameters and output results during the application stage. This adapts to the engineering requirements for optimization efficiency and avoids time-consuming or accuracy problems caused by unreasonable model structure.

[0038] Furthermore, the number of nodes in the output layer is 2, corresponding to the optimized node y-coordinate adjustment value and node bending stiffness adjustment coefficient. The neural network model uses the mean square error function as the loss function to measure the deviation between the optimized parameters output by the model and the actual parameters, and adopts the Adam optimizer, with the learning rate of the Adam optimizer initially set to 0.001.

[0039] Specifically, the number of output layer nodes is set to 2, corresponding to the optimized node y-coordinate adjustment value and the node bending stiffness adjustment coefficient, respectively. The node y-coordinate adjustment value is directly used to correct the vertical position of the node, and the node bending stiffness adjustment coefficient needs to be multiplied by the initial node bending stiffness to obtain the final optimized stiffness. No additional calculation is required, which improves the convenience of engineering applications.

[0040] The mean squared error function is chosen as the loss function to quantify the deviation between the model output and the actual optimal parameters, providing a clear direction for parameter adjustment and ensuring that the model converges in the direction of reducing error. The Adam optimizer is selected, with an initial learning rate of 0.001. This learning rate can ensure the convergence speed in the initial training phase, avoiding slow training due to being too small, and also avoid model oscillation due to being too large. Combined with the subsequent learning rate decay strategy, the model can finely adjust the parameters in the later stages, further improving accuracy and ensuring that the output parameters can accurately reflect the optimal state of the reticulated shell.

[0041] Furthermore, the training process divides the complete dataset into a training set, a validation set, and a test set in a 7:2:1 ratio, and the training rounds of the neural network model are set to 500 rounds.

[0042] Specifically, the training dataset is divided into training, validation, and test sets in a 7:2:1 ratio—70% of the training set is used for the model to learn the mapping relationship between input and output, ensuring a full grasp of optimization rules; 20% of the validation set is used to monitor the training status in real time to avoid model overfitting; and 10% of the test set is used to finally test the generalization ability, ensuring the model is reliable in new scenarios.

[0043] The training rounds are set to 500 rounds, which is determined based on pre-training experiments. This ensures that the model converges sufficiently to achieve stable accuracy while avoiding excessive training time due to too many rounds. It balances training effectiveness and efficiency, avoids the time-consuming problem of excessive iterations in traditional optimization methods, and ensures that the model can be quickly deployed in engineering applications.

[0044] Furthermore, the early stopping strategy is to stop training if the validation set loss does not decrease for 20 consecutive rounds, in order to avoid overfitting of the neural network model; the Adam optimizer adopts a learning rate decay strategy, specifically, the learning rate is decayed to 0.9 every 100 iterations to improve the stability of the model training.

[0045] Specifically, the early stopping strategy is set to stop training if the validation set loss does not decrease for 20 consecutive rounds. When the validation set loss does not decrease for 20 consecutive rounds, it means that the model has mastered the optimization rules, and continuing training will lead to overfitting. Stopping at this time can ensure accuracy and avoid unnecessary time consumption.

[0046] The Adam optimizer employs a strategy of reducing the learning rate to 0.9 every 100 iterations. In the early stages of training, a larger learning rate (0.001) is used to accelerate convergence, while the learning rate is gradually reduced in the later stages to allow for more precise parameter adjustments, avoid oscillations, improve training stability, ensure that the model converges stably to a better state, and reduce accuracy fluctuations.

[0047] Furthermore, the verification includes selecting typical cases from the test set. The input parameters of the typical cases include a combination of parameters for a reticulated shell model with a span-to-span ratio of 1 / 6, a length-to-span ratio of 2.6, a load ratio of 0, and four-sided support. The input parameters are then imported into the trained neural network model, and the optimized node parameters are output and compared with the optimization results of a traditional genetic algorithm. If the deviation of the ultimate bearing capacity of the reticulated shell or the maximum displacement deviation exceeds a preset range, the number of hidden layer nodes or the learning rate is adjusted, and the model is retrained until the accuracy requirements are met.

[0048] Specifically, model validation selected a typical case from the test set with a span-to-span ratio of 1 / 6, a length-to-span ratio of 2.6, a load ratio of 0, and four-sided support. This case represents a basic engineering scenario; if the model meets accuracy requirements in this scenario, reliability in other scenarios is also assured. The input parameters of the case were normalized and imported into the model, outputting optimized node parameters. Simultaneously, a traditional genetic algorithm was combined with ANSYS and MATLAB, based on the objective function... The traditional optimization results were obtained, and the ultimate bearing capacity and maximum displacement deviation of the two sets of results were compared.

[0049] If the deviation exceeds the preset range (e.g., a 7% deviation in ultimate bearing capacity), adjust the number of hidden layer nodes (e.g., change the first layer from 64 to 70, and the second layer from 32 to 35) or the learning rate (e.g., change the initial value to 0.0008), and retrain the model. During the verification process, the internal force ratio formula can also be used. The optimization process can be used to further validate the model's effectiveness if the axial force ratio increases and the bending moment ratio decreases, ensuring comprehensive and reliable verification.

[0050] Furthermore, when the neural network model is used to optimize the node parameters, the design parameters of the target single-layer cylindrical reticulated shell include the rise-to-span ratio, length-to-span ratio, load ratio, support conditions, initial node parameters, and reticulated shell span information. The neural network model automatically outputs the optimized node y-coordinate adjustment value and node bending stiffness adjustment coefficient.

[0051] Specifically, when applying the neural network model, the input design parameters of the target single-layer cylindrical reticulated shell must be complete and meet the requirements, including the rise-to-span ratio (e.g., 1 / 4), length-to-span ratio (e.g., 1.8), load ratio (e.g., 0), support conditions (e.g., two longitudinal supports), initial node x-coordinate (normalized relative value), initial node y-coordinate (normalized relative value), initial node bending stiffness (normalized relative value), and reticulated shell span (normalized relative value).

[0052] After the parameters are organized and normalized according to the input layer order, they are input into the model. The model automatically outputs the optimized nodal y-coordinate adjustment values ​​and nodal bending stiffness adjustment coefficients; after output, they can be adjusted according to the constraints. The initial check is to verify whether the adjustment values ​​are reasonable, so as to avoid excessive fluctuations that could lead to abnormal shapes. This process does not require manual intervention in the software and can quickly complete the optimization of a single reticulated shell. When dealing with batch optimization under multiple working conditions, it is only necessary to input the parameters of each reticulated shell in sequence, which greatly improves the efficiency of engineering design.

[0053] In summary, this invention first obtains node optimization data under different parameters through the interaction of genetic algorithms, finite element software, and programming tools. It then supplements this data with additional examples to construct a training dataset covering the range of common engineering design parameters. Next, it trains a three-layer fully connected neural network model with two hidden layers, employing early stopping and learning rate decay strategies to ensure model accuracy. When applied, the target single-layer cylindrical reticulated shell design parameters are input, and the model automatically outputs the optimized node parameters, eliminating the need for repeated iterations and manual intervention required by traditional methods. This significantly reduces optimization time and efficiently addresses the need for batch optimization under multiple working conditions. Simultaneously, the cubic spline interpolation method is used to correct the reticulated shell cross-sectional curve to ensure its smoothness and continuity. The optimized parameters are then imported into the finite element software for verification, ensuring that the ultimate bearing capacity and maximum displacement deviation of the reticulated shell meet engineering requirements. This avoids structural performance problems caused by unreasonable parameters and effectively improves the overall stability and engineering design efficiency of the aluminum alloy single-layer cylindrical reticulated shell.

[0054] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0055] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An automatic optimization method for node parameters of a reticulated shell structure based on a neural network model, characterized in that, include: S1. Collect "input-output" data pairs of a single-layer cylindrical reticulated shell constructed with aluminum alloy plate nodes. The input parameters include the design parameters and initial node parameters of the single-layer cylindrical reticulated shell. The output parameters include the optimized node vertical coordinate adjustment value and node bending stiffness adjustment value. S2. Normalize the input parameters to eliminate dimensional differences and form a training dataset for the neural network model. S3. Construct a three-layer fully connected neural network model, which includes an input layer, a hidden layer, and an output layer. The hidden layer uses the ReLU activation function to enhance the model's nonlinear fitting ability. S4. The neural network model is trained using the training dataset. The training includes dividing the training dataset into a training set, a validation set, and a test set, and adopting an early stopping strategy to avoid model overfitting. S5. Verify the neural network model. Verification includes comparing the optimized node parameters output by the model with the traditional optimization results. If the ultimate bearing capacity deviation of the single-layer cylindrical reticulated shell corresponding to the optimized node parameters is less than or equal to 5% and the maximum displacement deviation is less than or equal to 8%, then the model verification is successful. S6. Apply the neural network model to optimize node parameters. The application includes inputting the design parameters of the single-layer cylindrical reticulated shell. The neural network model automatically outputs the optimized node vertical coordinate adjustment value and node bending stiffness adjustment coefficient. S7. Based on the cubic spline interpolation method, the cross-sectional curve of the single-layer cylindrical shell is corrected according to the vertical coordinate adjustment value of the node, so that the corrected curve is smooth and continuous. S8. Import the optimized node parameters into the finite element software for bearing capacity verification. If the bearing capacity verification does not meet the engineering requirements, re-enter the parameters for fine-tuning and then optimize again.

2. The automatic optimization method for node parameters of a mesh shell structure based on a neural network model according to claim 1, characterized in that, The input parameters include the rise-to-span ratio, length-to-span ratio, half-span load ratio, support conditions, initial node x-coordinate, initial node y-coordinate, initial node bending stiffness, and grid shell span of the single-layer cylindrical reticulated shell. The rise-to-span ratio is set in the range of 1 / 6, 1 / 5, 1 / 4, and 1 / 3; the length-to-span ratio is set in the range of 1.4, 1.8, 2.2, and 2.6; the half-span load ratio is set in the range of 0, 1 / 4, 1 / 2, and 1; and the support conditions include two-sided support and four-sided support.

3. The automatic optimization method for node parameters of a mesh shell structure based on a neural network model according to claim 1, characterized in that, The "input-output" data pairs are derived from node optimization data under different parameters obtained through interactive optimization using a genetic algorithm combined with the finite element software ANSYS and the programming tool MATLAB. The training dataset is supplemented with 100 sets of reticulated shell optimization examples with different parameter combinations to ensure that it covers the range of common design parameters in engineering. The normalization process includes converting the node coordinate parameters and node bending stiffness parameters into relative values ​​to eliminate the influence of dimensional differences on the training results of the neural network model.

4. The automatic optimization method for node parameters of a mesh shell structure based on a neural network model according to claim 1, characterized in that, The number of nodes in the input layer is 8, and the 8 input parameters correspond to the span-to-span ratio, length-to-span ratio, half-span load ratio, support condition code value, initial node x-coordinate, initial node y-coordinate, initial node bending stiffness, and grid shell span, respectively.

5. The automatic optimization method for node parameters of a reticulated shell structure based on a neural network model according to claim 1, characterized in that, In the three-layer fully connected neural network model, the hidden layer is set to 2 layers, with the first hidden layer having 64 nodes and the second hidden layer having 32 nodes.

6. The automatic optimization method for node parameters of a mesh shell structure based on a neural network model according to claim 5, characterized in that, The output layer has 2 nodes, corresponding to the optimized node y-coordinate adjustment value and node bending stiffness adjustment coefficient. The neural network model uses the mean square error function as the loss function to measure the deviation between the optimized parameters output by the model and the actual parameters, and uses the Adam optimizer with the learning rate of the Adam optimizer initially set to 0.

001.

7. The automatic optimization method for node parameters of a mesh shell structure based on a neural network model according to claim 1, characterized in that, The training process divides the complete dataset into a training set, a validation set, and a test set in a ratio of 7:2:1, and the training rounds of the neural network model are set to 500 rounds.

8. The automatic optimization method for node parameters of a reticulated shell structure based on a neural network model according to claim 7, characterized in that, The early stopping strategy is to stop training if the validation set loss does not decrease for 20 consecutive rounds, in order to avoid overfitting of the neural network model. The Adam optimizer adopts a learning rate decay strategy, specifically, the learning rate is decayed to 0.9 every 100 iterations to improve the stability of the model training.

9. The automatic optimization method for node parameters of a mesh shell structure based on a neural network model according to claim 1, characterized in that, The verification process includes selecting typical cases from the test set. The input parameters of the typical cases include a combination of parameters for a reticulated shell model with a span-to-span ratio of 1 / 6, a length-to-span ratio of 2.6, a load ratio of 0, and four-sided support. The input parameters are then imported into the trained neural network model, and the optimized node parameters are output and compared with the optimization results of a traditional genetic algorithm. If the deviation of the ultimate bearing capacity of the reticulated shell or the maximum displacement deviation exceeds a preset range, the number of hidden layer nodes or the learning rate is adjusted, and the model is retrained until the accuracy requirements are met.

10. The automatic optimization method for node parameters of a reticulated shell structure based on a neural network model according to claim 1, characterized in that, When the neural network model is used to optimize the node parameters, the design parameters of the target single-layer cylindrical reticulated shell include the rise-to-span ratio, length-to-span ratio, load ratio, support conditions, initial node parameters, and reticulated shell span information. The neural network model automatically outputs the optimized node y-coordinate adjustment value and node bending stiffness adjustment coefficient.

Images