Hull structure trepanning strength forecasting method based on improved deep neural network model

Abstract

The invention provides a hull structure trepanning strength forecasting method based on an improved deep neural network model, and the method comprises the steps: obtaining coarse grid stress distribution data before trepanning of a hull structure and fine grid stress distribution data after trepanning of the hull structure, constructing a deep neural network model, and during model training, carrying out the prediction of the trepanning strength of the hull structure. A label of coarse grid stress distribution data is introduced into a plurality of neurons of the last layer of the hidden layer for performance evaluation, and coarse grid stress supervision is carried out by limiting the difference between the output value of the neurons introduced with the coarse grid label and a coarse grid reference stress value. And training the model by adopting a composite loss function formed by weighting and combining a fine grid stress prediction error and a coarse grid stress supervision error, and obtaining an optimal training model as a trepanning strength prediction model. According to the method, the machine learning model is introduced to forecast the opening strength, improvement is made in model application, the stress distribution overall trend provided by coarse grid data before opening is considered, and therefore efficient prediction of the opening strength of the hull structure is achieved.

Description

Technical Field

[0001] This invention belongs to the fields of ship hull structure opening strength assessment technology and machine learning technology, specifically involving a method for predicting the strength of ship hull structure openings based on an improved deep neural network model. Background Technology

[0002] In shipbuilding engineering design and analysis, predicting the structural strength of perforated structures is a crucial step in ensuring overall structural safety. However, traditional methods often struggle to accurately predict the stress distribution in perforated areas. For example, progressive damage analysis and characteristic length methods, while capable of predicting the strength of porous materials to some extent, rely heavily on extensive experimental data and cannot simulate complex failure processes.

[0003] In recent years, machine learning and neural network methods have shown great application potential in the field of structural stress analysis. However, traditional neural network models often adopt a "black box" end-to-end learning approach, lacking guidance from physical knowledge, and thus have limited generalization ability in structural stress analysis. Summary of the Invention

[0004] The purpose of this invention is to provide a method for predicting the strength of openings in ship hull structures based on an improved deep neural network model. The improved deep neural network model improves the accuracy and efficiency of the prediction.

[0005] To achieve the above objectives, the technical solution provided by the present invention is as follows:

[0006] A method for predicting the strength of openings in a ship's hull structure based on an improved deep neural network model, the method comprising:

[0007] Historical data on openings in the hull structure are obtained, including parameter data affecting stress after the opening and stress distribution data around the opening; the stress distribution data around the opening includes fine-grid stress distribution data after the opening and coarse-grid stress distribution data at the corresponding position before the opening.

[0008] A training set is constructed based on the historical data, and a deep neural network model containing an input layer, multiple hidden layers, and an output layer is trained, wherein:

[0009] The training set is constructed using the parameter data as input features and the fine-grid stress distribution data as label data. Several neurons in the last layer of the hidden layer are introduced with labels from the coarse-grid stress distribution data for performance evaluation. Coarse-grid stress supervision is performed by limiting the difference between the output value of the neurons with coarse-grid labels and the coarse-grid reference stress value. The model is trained using a composite loss function composed of a weighted combination of fine-grid stress prediction error and coarse-grid stress supervision error.

[0010] The optimal training model is obtained as the opening strength prediction model, and the stress distribution data of the opening area to be predicted is obtained using the prediction model.

[0011] In some embodiments of the present invention, the parameters affecting the stress after the hole is opened are the hole diameter, the length of the long side, the presence or absence of a support column, the location of the hole, and the external load.

[0012] In some embodiments of the present invention, the length of the long side is determined based on the shape of the opening. For an opening with a waist-shaped shape, the length of its long side is taken as the input. For an opening with a perfect circle shape, the length of the long side is 0.

[0013] The value for whether a support exists is determined as follows: if a support exists, the value is 1; otherwise, the value is 0.

[0014] In some embodiments of the present invention, the fine mesh stress distribution data, after preprocessing, is used to construct a training set, including:

[0015] Using a preset grid size as the analysis unit size, the area around the opening is divided into several grids along the circumferential direction. The average stress value in each grid is taken as the stress value on the corresponding grid, which is used to construct a tag set.

[0016] In some embodiments of the present invention, when the number of grids is less than N, the number of grids is supplemented by interpolation.

[0017] During interpolation, the mesh with the highest stress remains unchanged in size, while the remaining meshes are interpolated.

[0018] In some embodiments of the present invention, the historical data is obtained through finite element simulation, wherein the stress distribution data around the opening is obtained in the following way:

[0019] After modeling the ship, a coarse mesh is divided according to the spacing of the aggregate. After applying the load, the stress distribution data of the coarse mesh at the location where the hole is to be opened is obtained.

[0020] A hole model is constructed at the desired hole location, and a fine mesh is generated at the hole location. After applying the same load, the stress distribution data of the fine mesh around the hole is obtained.

[0021] In some embodiments of the present invention, the last layer of the hidden layer uses labels from the first two neurons to introduce coarse grid stress distribution data.

[0022] In some embodiments of the present invention, the fine mesh stress prediction error and the coarse mesh stress monitoring error are calculated using a mean square error function;

[0023] The weight of the stress prediction error in the fine mesh is greater than the weight of the stress monitoring error in the coarse mesh.

[0024] In some embodiments of the present invention, the deep neural network model is a multilayer perceptron (MLP).

[0025] In some embodiments of the present invention, the method for obtaining the optimal training model is to adopt an early stopping strategy during the model training process. When the validation set loss does not decrease within a set number of consecutive rounds, the training is automatically stopped and the current optimal model is output.

[0026] In some embodiments of the present invention, the method further includes performing inverse normalization processing on several grid stress value data output by the hole strength prediction model, and displaying the stress value data through cloud map visualization technology to show the stress distribution in the hole area.

[0027] The present invention has the following beneficial effects:

[0028] This invention utilizes a deep neural network model to predict the strength of openings in ship hull structures. The deep neural network learning model is improved by selecting neurons in the last hidden layer to introduce physical labels of the pre-opening coarse-grid stress, and studying the stress mapping from coarse to fine grids. This invention considers the overall trend of stress distribution reflected in the pre-opening coarse-grid data, explicitly distinguishing between coarse and fine-grid data input channels in the network architecture. This allows the model to learn stress transformation characteristics across scales, thereby improving the interpretability and reliability of the neural network model for assessing beam opening stress. The improved model further enhances the accuracy of the prediction. Attached Figure Description

[0029] The invention, its features, and advantages will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings. Like reference numerals denote like parts throughout the drawings. The drawings are not intentionally drawn to scale; their purpose is to illustrate the spirit of the invention.

[0030] Figure 1 This is a schematic diagram of a deep neural network architecture based on a multilayer perceptron (MLP) in one embodiment.

[0031] Figure 2 This is a schematic diagram of a coarse grid model of a certain deck of a roll-on / roll-off ship in one embodiment.

[0032] Figure 3 This is a schematic diagram of a uniformly distributed load condition in one embodiment.

[0033] Figure 4 This is a schematic diagram of a fine mesh model with partial openings in one embodiment.

[0034] Figure 5 This is a schematic diagram of stress cloud diagram in one embodiment.

[0035] Figure 6This is an improved MLP and MLP prediction error angle distribution in one embodiment. Detailed Implementation

[0036] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. It should also be noted that the terminology used in this invention is for describing specific implementations only and is not intended to limit the exemplary implementations according to this application.

[0037] Example 1

[0038] This embodiment provides an efficient method for predicting the strength of openings in ship hull structures based on an improved MLP, comprising the following steps:

[0039] S101: Obtain historical data of the opening, including external load (uniformly distributed load), opening diameter, long side length, number of supports (whether supports exist), opening location, and stress distribution around the opening. The stress distribution around the opening includes the stress distribution of the coarse mesh before the opening and the stress distribution of the fine mesh after the opening.

[0040] S102: Preprocessing based on historical data, including:

[0041] S201: Based on the shape of the opening, it is divided into a perfect circle and an oval shape. The oval shape is input with the length of its long side, and the perfect circle is input with 0.

[0042] The value for the number of supports is determined by whether there are supports at the beam where the opening is located: 1 if supports are present, 0 otherwise.

[0043] The opening position is determined based on the coordinates (x, y, z). The coordinate value of each axis is used as an input. The selection of the origin has no impact on the modeling. In this embodiment, the center of the beam is selected as the origin.

[0044] S202: Divide the stress distribution of the fine mesh around the opening into 28 meshes, and take the average stress value in each mesh as the stress value on the corresponding mesh.

[0045] During mesh generation, this embodiment extracts the stress data of a 50mm × 50mm fine mesh around the opening using finite element analysis as the key analysis area. Based on this analysis area size, when the opening diameter is at its maximum, the mesh around the opening consists of exactly 28 meshes. For smaller diameter holes, the surrounding mesh may be less than 28. To ensure the consistency of the machine learning model's output dimensions, interpolation is used to supplement the mesh data to 28. One optional operation is as follows: if there are less than 28 meshes, select the mesh with the highest stress in the fine mesh as the starting point, and perform interpolation and generation between the two meshes counterclockwise until the number of meshes reaches 28. During interpolation, the size of the mesh with the highest stress remains unchanged; interpolation and generation are only performed between other meshes.

[0046] S103: Construct a dataset based on the processed historical data, including:

[0047] S203: Use the external load, opening diameter, long side length, number of supports, and opening location contained in the historical data as input features to construct the input dataset X; use the stress values ​​on the corresponding 28 grids as output features to construct the output dataset Y (label set).

[0048] In some embodiments of this invention, the finite element method is used to simulate the stress distribution at the openings in the ship's beams. First, a finite element model of the ship's beams is established using Patran, a widely used finite element software in the shipbuilding industry, and simulation results are obtained using Nastran. In the coarse mesh model, all major structures of the ship are modeled. Deck, outer plating, longitudinal and transverse bulkheads, as well as major supporting components (trusses), web plates, etc., are all modeled using 2D plate elements, while ribs, stiffeners, and other components are modeled using 1D beam elements. Components that do not significantly affect the calculation results are not modeled. The coarse mesh is generated according to the rib spacing, such as... Figure 2 The diagram is divided into upper and lower sections. The load conditions are set to typical loads encountered in actual operation of a roll-on / roll-off ship, primarily uniformly distributed loads. Figure 3 This is a load view for a certain working condition.

[0049] After completing the loading calculations for each working condition, the crossbeams under high stress were selected as typical working conditions for analysis based on the coarse mesh results. The mesh of the selected strong crossbeam openings was refined for analysis, and the size and location of the openings were parametrically designed. The simulated opening sizes covered circular holes of commonly used diameters in ships, while the opening locations were selected from typical arrangements at the center and edge of the crossbeam to comprehensively analyze the stress distribution variation. According to the Norwegian Classification of Shipping (DNV) standard, the stress data of the 50mm × 50mm fine mesh around the openings was extracted through finite element analysis as the key analysis area, focusing primarily on the central values ​​of the Mises stress in the upper and lower element layers. The fine mesh circular hole model is as follows... Figure 4 As shown.

[0050] Input features that significantly affect the stress after the opening are selected, including beam geometric parameters (opening diameter, opening location), load parameters (uniformly distributed load, etc.) and the number of supports, and the stress values ​​of the corresponding coarse and fine meshes are recorded as data labels for the model. These data points cover the stress values ​​and their corresponding geometric coordinates at key locations around the opening, forming a high-dimensional feature dataset for training the neural network.

[0051] S104: Train the artificial intelligence model using the constructed training set to obtain the optimal model;

[0052] Specifically, the dataset is divided into a training set and a test set according to a proportional method, with the training set accounting for 80% of the total data and the test set accounting for 20%. The model is trained using the divided dataset, and the accuracy is validated using the test set to obtain the optimal model.

[0053] To improve overall prediction performance and enhance forecast reliability, this embodiment employs an improved deep neural network model to train the forecast model, such as... Figure 1 As shown. This embodiment uses an MLP model as the deep neural network model. The improved MLP model's input layer accepts seven dimensions of input features, including external load, opening diameter, long side length, number of supports, and opening location (X, Y, and Z axis coordinates each occupy one input dimension). It contains five hidden layers. The outputs of the first and second neurons in the last hidden layer serve as the coarse-grid stress prediction value. The hidden layers use the Sigmoid activation function, and the output layer uses the Softplus activation function. The loss function is a weighted average of the fine-grid stress error and the coarse-grid stress error, as detailed below:

[0054] The model incorporates a coarse-grid stress prediction loss into its loss function by calling the output values ​​of the first two neurons in the last hidden layer. These output values ​​correspond to the physical labels of the coarse-grid stress data. Its form is:

[0055]

[0056]

[0057] in, and These are the prediction losses for fine-grid and coarse-grid predictions, respectively. and The first in the dataset The first sample One predicted value and one actual value. and These are hyperparameters used to balance the weights of the two parts of the loss. In this paper... , .

[0058] In the last hidden layer of the model, the network reflects the coarse mesh stress distribution of the structure through the calculation of the output values ​​of the first two neurons. We define the last hidden layer of the network as... ,in Among them, the output values ​​of the first two neurons Compare the label values ​​with the coarse mesh stress data.

[0059]

[0060] in, and These are the first two neurons in the last hidden layer of the dataset. The output value of each sample and These are the corresponding coarse mesh stress data labels.

[0061] During the training of the artificial intelligence model, the Adam optimizer is used with mean squared error (MSE) as the loss function. The model parameters are optimized based on the training set at each iteration, and the training ends after 10,000 iterations.

[0062] S105: Use the trained model to predict the opening strength under actual working conditions and output stress distribution data;

[0063] S106: Visualize the output data to obtain a data cloud map;

[0064] Specifically, to facilitate the processing and understanding of the output data and allow users to more intuitively comprehend complex data analysis results, based on the S205 processing, the stress values ​​of the 28 meshes output by the model are inversely normalized. These stress values ​​are then displayed using cloud map visualization technology to show the stress distribution in the opening area. A schematic diagram of the stress cloud map is shown below. Figure 5 As shown.

[0065] Through the above steps, the artificial intelligence model is trained using the acquired historical data to obtain the optimal model that can predict the stress distribution in the opening area of ​​the structure; the stress distribution data output by the optimal model is then visualized to obtain a stress distribution cloud map.

[0066] Example 2

[0067] This embodiment compares the performance of the improved deep neural network (improved MLP) of this invention with that of the traditional deep neural network (MLP) in stress prediction in the opening area of ​​a ship's beam.

[0068] As shown in Table 1, the improved MLP model exhibits superior performance on the test set: the average relative error for fine mesh stress is 13.09%, lower than the 23.38% of the MLP model; the average relative error for maximum stress is 8.08%, slightly lower than the 14.5% of the MLP model. Notably, the improved MLP model achieves an average relative error (MAPE) of 18.28% in coarse mesh stress prediction, which the MLP model cannot handle, demonstrating its comprehensive advantages in multi-scale prediction of opening stress.

[0069] Table 1. Comparison of prediction accuracy of improved MLP and MLP for ship opening strength

[0070]

[0071] Meanwhile, through systematic analysis of stress prediction errors at different angles in the opening region, the study found that the average relative error distributions of the improved MLP and the MLP model exhibit the same characteristics. The mesh angles were calculated starting from the top of the circular hole and recorded in a counter-clockwise increment. Figure 6 The average relative error distribution characteristics of the improved MLP and MLP models at different angles in the opening region are presented. The results show that the angular distribution of the average relative error of both the improved MLP and MLP models exhibits two obvious peaks, appearing in the first 30% and last 30% of the angle region, respectively, with a peak of approximately 20%, corresponding to the complex fluctuation region of stress concentration in the opening. This is mainly because in the complex fluctuation region of stress concentration, due to the drastic changes in stress values, the model has more difficulty capturing nonlinear features, thus leading to a decrease in prediction accuracy. In other angular regions, the stress changes are relatively gradual, and the average relative error remains at a level of approximately 11%. Even in another region of stress concentration—the region near the lower end of the opening—the relative error predicted by the model remains at a low level. The improved MLP and MLP models exhibit similar prediction error characteristics in the symmetrical region of stress distribution, indicating that in the symmetrical region of stress distribution, the models encounter similar difficulties in the stress concentration region of the opening, but the improved MLP performs better overall.

Claims

1. A method for predicting the strength of openings in a ship's hull structure based on an improved deep neural network model, characterized in that, The method includes: Historical data on openings in the hull structure are obtained, including parameter data affecting stress after the opening and stress distribution data around the opening; the stress distribution data around the opening includes fine-grid stress distribution data after the opening and coarse-grid stress distribution data at the corresponding position before the opening. A training set is constructed based on the historical data, and a deep neural network model containing an input layer, multiple hidden layers, and an output layer is trained, wherein: The training set is constructed using the parameter data as input features and the fine-grid stress distribution data as label data. Several neurons in the last layer of the hidden layer are introduced with labels from the coarse-grid stress distribution data for performance evaluation. Coarse-grid stress supervision is performed by limiting the difference between the output value of the neurons with coarse-grid labels and the coarse-grid reference stress value. The model is trained using a composite loss function composed of a weighted combination of fine-grid stress prediction error and coarse-grid stress supervision error. The optimal training model is obtained as the opening strength prediction model, and the stress distribution data of the opening area to be predicted is obtained using the prediction model.

2. The method for predicting the opening strength according to claim 1, characterized in that, The parameters that affect the stress after the hole is opened are the hole diameter, the length of the long side, the presence of a support column, the location of the hole, and the external load.

3. The method for predicting the opening strength according to claim 2, characterized in that, The length of the long side is determined based on the shape of the opening. For an opening with a waist-shaped shape, its long side length is taken as the input. For an opening with a perfect circle shape, the long side length is 0. The value for whether a support exists is determined as follows: if a support exists, the value is 1; otherwise, the value is 0.

4. The method for predicting the opening strength according to claim 1, characterized in that, The fine-mesh stress distribution data, after preprocessing, is used to construct a training set, including: Using a preset grid size as the analysis unit size, the area around the opening is divided into several grids along the circumferential direction. The average stress value in each grid is taken as the stress value on the corresponding grid, which is used to construct a tag set.

5. The method for predicting the opening strength according to claim 4, characterized in that, When the number of grids is less than N, the number of grids is supplemented by interpolation. During interpolation, the mesh with the highest stress remains unchanged in size, while the remaining meshes are interpolated.

6. The method for predicting the opening strength according to claim 1 or 4, characterized in that, The historical data was obtained through finite element simulation, wherein the stress distribution data around the opening was obtained in the following way: After modeling the ship, a coarse mesh is divided according to the spacing of the aggregate. After applying the load, the stress distribution data of the coarse mesh at the location where the hole is to be opened is obtained. A hole model is constructed for the hole location, and a fine mesh is generated at the hole location. After applying the same load, the stress distribution data of the fine mesh around the hole is obtained.

7. The method for predicting the opening strength according to claim 1, characterized in that, The last layer of the hidden layer uses labels from the first two neurons to introduce coarse grid stress distribution data.

8. The method for predicting the opening strength according to claim 1, characterized in that, The fine-grid stress prediction error and the coarse-grid stress monitoring error are calculated using the mean square error function. The weight of the stress prediction error in the fine mesh is greater than the weight of the stress monitoring error in the coarse mesh.

9. The method for predicting the opening strength according to claim 1, characterized in that, The deep neural network model is a multilayer perceptron (MLP).

10. The method for predicting the opening strength according to claim 1, characterized in that, The method further includes performing inverse normalization on several grid stress value data output by the hole strength prediction model, and displaying the stress value data through cloud map visualization technology to show the stress distribution in the hole area.

Images