LSTM-Smith hull beam ultimate strength evaluation method considering initial defects
By modifying the Smith method using an LSTM model, the influence of initial defects in the hull beam is accurately predicted, solving the problem of evaluation error in the traditional Smith method and achieving efficient and accurate ultimate strength evaluation of the hull beam.
Patent Information
- Application Number
- CN202511927893.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-19
- Publication Date
- 2026-03-20
AI Technical Summary
The traditional Smith method fails to adequately consider initial defects in the ultimate strength assessment of ship hull beams, resulting in significant errors in the assessment results.
By employing an LSTM model combined with the Huber loss function, and through preprocessing of training sample data and an initial point penalty term, the ultimate strength calculation in the Smith method is corrected, and the stress-strain curve of stiffened plate elements with initial defects is accurately predicted.
It improves the accuracy of hull beam ultimate strength assessment, reduces the amount of calculation, and is suitable for rapid assessment of ship design and structural safety, with errors controlled within 3%.
Smart Images
Figure CN121706591A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to a method for assessing the structural strength of ships, specifically to an LSTM-Smith method for assessing the ultimate strength of hull beams that takes into account initial defects. Background Technology
[0002] The ultimate strength of hull beams is a core indicator of structural safety under extreme loads, and its accurate assessment is crucial for ship structural design and navigation safety. Currently, commonly used methods for assessing the ultimate strength of hull beams include the nonlinear finite element method, the Smith method (step-by-step collapse method), and the ideal structural element method. Among these, the Smith method is simple to operate and computationally efficient, and has long been an important method for assessing the ultimate strength of ship structures. Especially in engineering practice, the Smith method is widely used in ship structural design and safety assessment, effectively handling the ultimate strength calculation of complex hull structures. However, the traditional Smith method has significant limitations: it uses idealized assumptions for stiffened plate elements and does not fully consider the impact of initial geometric defects during ship manufacturing and transportation on the ultimate strength, leading to significant errors in the assessment results for structures with defects. Summary of the Invention
[0003] Purpose of the invention: The purpose of this invention is to address the problem that the idealized assumptions of the traditional Smith method lead to insufficient accuracy in the ultimate strength assessment of hull beams with initial defects, and to provide a high-precision LSTM-Smith ultimate strength assessment method for hull beams that takes into account initial defects.
[0004] Technical solution: The present invention provides an LSTM-Smith method for assessing the ultimate strength of ship hull beams, taking into account initial defects, comprising:
[0005] S1: Obtain sample data of stiffened plate units with initial defects within a typical ship size range. Each sample data consists of stiffener type, plate flexibility coefficient, stiffener flexibility coefficient, material yield strength, and stress-strain curve.
[0006] S2: Sample data preprocessing;
[0007] S3: Construct an LSTM model using the Huber loss function with initial point penalty; train the LSTM model using preprocessed sample data;
[0008] S4: Embed the trained LSTM model into the Smith method's computational flow to correct the ultimate strength calculation of stiffened plate elements with defects.
[0009] Furthermore, in step S1, the stiffened plate element model containing initial defects covers different stiffener types, geometric parameters, and material parameters; finite element ultimate strength analysis is performed on each stiffened plate element model containing initial defects to obtain curves composed of stress and relative strain data.
[0010] Furthermore, the types of reinforcing ribs include angle steel, T-shaped steel, and flat steel; the geometric parameters include the flexibility coefficient of the plate and the flexibility coefficient of the ribs; and the material parameters include the material type and yield strength.
[0011] Furthermore, initial defects include the initial deformation of the plate. Initial cylindrical deformation of the rib Initial deformation of ribs by lateral tilt The initial defect is applied using a formula method.
[0012] Furthermore,
[0013]
[0014]
[0015]
[0016] in, , , The coordinates represent the length, width, and thickness of the stiffening plate, respectively. , The flexibility coefficient of the display board; Indicates the thickness of the stiffening strip; , , For the length of the stiffening plate, The width of the stiffening plate. The web height of the tendon. Let the half-wave number of plate buckling be defined as satisfying The smallest positive integer.
[0017] Further, in step S2, sample data preprocessing includes: performing unique thermal encoding on the stiffener type and converting the category data into binary vectors; using the Z-score method to standardize the geometric and material parameters of the stiffened plate element model to ensure dimensional balance; performing interpolation on the stress-strain curves to ensure consistent sample time step lengths; and dividing all preprocessed sample data into training and test sets according to a specific ratio.
[0018] Furthermore, the training set and the test set are divided at 80%:20%.
[0019] Furthermore, in step S3, the LSTM model adopts a four-layer stacked structure, with the number of nodes in each layer being 512, 256, 128, and 64 respectively.
[0020] Furthermore, in step S3, the Huber loss function, which includes an initial point penalty, is as follows:
[0021]
[0022] in, For the true value, For predicted values, This is the critical value of the error;
[0023] The purpose of introducing an "initial point penalty term" is to force the model to minimize the prediction error in the first time step, thus avoiding the accumulation of bias over time.
[0024]
[0025] in, This is the predicted value for the first time step. This is the true value at the first time step. For penalty weighting;
[0026] Ultimately, the total loss function consists of the Huber loss and the initial point penalty:
[0027]
[0028] in, This represents the mean of the loss.
[0029] Further, in step S4, the cross section of the hull beam is divided into hard-corner elements, plate elements, and stiffened plate elements, and each element is numbered to determine the model input information for each element; the location of the initial defect is determined, the stiffened plate element number where the defect is located is obtained, the corresponding model input information is input into the LSTM model, the corrected stress-strain curve is obtained, and the stress calculation in the traditional Smith method is corrected.
[0030] Beneficial Effects: Compared with existing technologies, this invention has the following significant advantages: This invention accurately predicts the stress-strain curves of stiffened plate elements with initial defects using an LSTM model, solving the error problem caused by the idealized assumptions of the traditional Smith method, making the evaluation results closer to actual engineering conditions. At the same time, it retains the advantages of the Smith method, such as simple operation and high computational efficiency. This invention can also achieve rapid assessment of the impact of local defects on ultimate strength without the need for a complex finite element modeling process, significantly reducing the computational load and making it suitable for rapid assessment of ship design and structural safety. Attached Figure Description
[0031] Figure 1 This is a flowchart of an LSTM-Smith method for evaluating the ultimate strength of a ship hull beam that takes into account initial defects, provided by an embodiment of the present invention.
[0032] Figure 2 This is a schematic diagram of the initial defect application in an embodiment of the present invention;
[0033] Figure 3 These are the ultimate strength calculation results of the ideal model and the defective model of the Nishihara box girder in the embodiments of the present invention;
[0034] Figure 4 These are the ultimate strength calculation results of the ideal model and the defective model of the Container ship hull in the embodiments of the present invention. Detailed Implementation
[0035] The invention will now be further described with reference to the accompanying drawings.
[0036] like Figure 1 As shown, this embodiment of the invention provides an LSTM-Smith method for evaluating the ultimate strength of a ship hull beam that takes into account initial defects, including the following steps:
[0037] S1: Obtain sample data of stiffened plate elements containing initial defects.
[0038] 923 stiffened plate element models with initial defects were constructed within a typical ship size range, covering different stiffener types (angle steel, T-beams, flat steel), geometric parameters, and material parameters. Geometric parameters included the plate's compliance coefficient and the stiffener's compliance coefficient, while material parameters included material type and yield strength. Finite element ultimate strength analysis was performed on each stiffened plate element model with initial defects to obtain curves composed of stress and relative strain data. Each sample data consisted of stiffener type, plate compliance coefficient, stiffener compliance coefficient, material yield strength, and stress-strain curve.
[0039] In this embodiment, the material used is common ship hull steel with an elastic modulus of 211,000 MPa, a Poisson's ratio of 0.3, and yield strengths of 235 MPa, 313.6 MPa, and 352.8 MPa, respectively.
[0040] Referring to the stiffening plate size range of 6 typical ships in the ISSC2012 report, the stiffening plate size meets the following requirements: plate flexibility coefficient 0.6~3.0, stiffener flexibility coefficient 0.25~0.9, covering most ship stiffening plate size specifications.
[0041] like Figure 2 As shown, the initial defects include the initial deformation of the plate. Initial cylindrical deformation of the rib Initial deformation of ribs by lateral tilt The initial defect is applied using a formula method.
[0042]
[0043]
[0044]
[0045] in, , , The coordinates represent the length, width, and thickness of the stiffening plate, respectively. , The flexibility coefficient of the display board; Indicates the thickness of the stiffening strip; , , For the length of the stiffening plate, The width of the stiffening plate. The web height of the tendon. Let the half-wave number of plate buckling be defined as satisfying The smallest positive integer.
[0046] According to the boundary constraints of the stiffened plate set in the ISSC2015 report, the loading method is displacement loading at both ends, and the stress and strain data of each stiffened plate element are obtained through nonlinear finite element analysis.
[0047] S2: Sample data preprocessing.
[0048] Each sample data is preprocessed, including: performing unique thermal encoding on the stiffener type and converting the category data into binary vectors; using the Z-score method to standardize the geometric and material parameters of the stiffened plate element model to ensure that the data has balanced dimensions during training; interpolating the stress-strain curves and standardizing them to 101 time steps to ensure that the sample time step lengths are consistent; and dividing all preprocessed sample data into training and test sets according to an 80%:20% ratio.
[0049] S3: Construct an LSTM model using the Huber loss function with initial point penalty; train and validate the LSTM model using the training and test sets respectively; the inputs to the LSTM model are the plate flexibility coefficient, stiffener flexibility coefficient, stiffener type, material yield strength, and relative strain, and the output of the LSTM model is the stress corresponding to the relative strain.
[0050] The LSTM model employs a four-layer stacked structure, with 512, 256, 128, and 64 nodes per layer, respectively. A Dropout layer (with a Dropout rate of 0.4) is added after each layer to prevent overfitting. The Adam optimizer is used for optimization, with a learning rate of 0.0001. The ReLU activation function is used, combined with an early stopping mechanism to improve the model's generalization ability.
[0051] The Huber loss function, including the initial point penalty, is as follows:
[0052]
[0053] in, For the true value, For predicted values, This is the critical value of the error.
[0054] This invention introduces an additional "initial point penalty term" to force the model to minimize the prediction error in the first time step, thus avoiding the accumulation of time-series bias.
[0055]
[0056] in, This is the predicted value for the first time step. This is the true value at the first time step. For penalty weights.
[0057] Ultimately, the total loss function consists of the Huber loss and the initial point penalty:
[0058]
[0059] in, This represents the mean of the loss.
[0060] When evaluating the accuracy of model predictions, specific evaluation metrics (such as mean squared error, mean squared error, mean squared error, mean squared error) are used. (etc.) Quantitative analysis is performed on the difference between the prediction results and the actual test data; after training, the accuracy and generalization ability of the model are verified through the test set.
[0061] In this embodiment, the coefficient of determination between the final model prediction and the true value is verified through a test set. The accuracy reached 0.9953, and the mean absolute error (MAE) was 2.8, ensuring the prediction accuracy of the LSTM model.
[0062] S4: Embed the trained LSTM model into the Smith method's computational flow to correct the ultimate strength calculation of stiffened plate elements with defects.
[0063] The cross section of the hull beam is divided into hard-corner elements, plate elements, and stiffened plate elements, and each element is numbered to determine the model input information for each element. The location of the initial defect is determined manually, the stiffened plate element number where the defect is located is obtained, and the corresponding model input information is input into the LSTM model to obtain the corrected stress-strain curve, thereby correcting the stress calculation in the traditional Smith method.
[0064] For imperfect stiffened plate elements, the corrected stress values output by the LSTM model replace the stress calculation formula in the traditional Smith method, dynamically adjusting the impact of imperfections on the ultimate strength. For elements without imperfections, the traditional Smith method is still used for calculation. After the corrected stress values output by the LSTM model replace the stress calculation in the traditional Smith method, the standard calculation process in the Smith method is still followed for iteration, ultimately yielding the ultimate strength of the hull beam.
[0065] Two specific examples are given below.
[0066] Example 1: Ultimate Strength Calculation of Nishihara Box Girder
[0067] The Nishihara box girder MST-3 model was selected. The model was divided into cross-sectional elements, specifically stiffened plate elements and hard-corner elements. For the ideal model, the ultimate strength was calculated using the traditional Smith method. For the model with defects, a modified Smith method incorporating LSTM was used. Defects were applied to the strips and reinforcement of the stiffened plate.
[0068] Under the ideal model, the relative error between the calculation results of the traditional Smith method and the finite element method is 1.72%, which is in high agreement with the numerical results, verifying the theoretical correctness and calculation accuracy of the Smith method itself.
[0069] Under the defective model, the ultimate strength under the modified method of this invention is calculated to be 563.606 kNm, with a relative error of 0.60% compared with the finite element result with defects (560.222 kNm) and a relative error of ≤2% compared with the experimental value (575 kNm).
[0070] The specific calculation data for each working condition are shown in Table 1. This table clearly compares the ultimate strength values of the ideal model and the model with initial defects under different calculation methods, as well as the differences with experimental values in the literature; Figure 3 The results of calculations using the finite element method and the Smith method for the ideal model and the defective model of the Nishihara box girder are presented intuitively in the form of moment-curvature curves, further demonstrating the accuracy of the correction method of this invention.
[0071] Table 1 Comparison of ultimate strength results for Nishihara box girders
[0072]
[0073] Example 2: Ultimate Strength Calculation of a Typical Container Ship Hull
[0074] The typical container ship hull model from the ISSC 2012 report was used. The model was divided into cross-sectional elements, specifically stiffened plate elements and hard-corner elements. For the ideal model, the ultimate strength was calculated using the traditional Smith method. For the model with defects, a modified Smith method incorporating LSTM was used. Defects were applied at multiple buckling-prone stiffener locations.
[0075] The method of this invention was used to calculate the ultimate strength of the central arch condition as 7.258 GNm, with a relative error of 0.12% compared to the finite element result of 7.249 GNm with defects; and the ultimate strength of the central sag condition as 6.562 GNm, with a relative error of 2.37% compared to the finite element result of 6.721 GNm with defects, which meets the engineering accuracy requirements.
[0076] The specific calculation data for each working condition are shown in Table 2. This table clearly compares the ultimate strength values of the ideal model and the model with initial defects under different calculation methods, as well as the differences between the results calculated using the Smith method for the ideal model in the literature; and Figure 4 The results of calculations using the finite element method and the Smith method for the ideal model and the model with defects are presented intuitively in the form of moment-curvature curves, further demonstrating the accuracy of the correction method of this invention.
[0077] Table 2 Comparison of Ultimate Strength Results for Container Ship Hull
[0078]
[0079] The above examples show that, when the model considers initial defects, the relative error between the modified Smith method and the finite element method is controlled within 3%, and the results show a high degree of agreement with experimental values. This demonstrates the ability to accurately capture the influence of initial defects on the ultimate strength of the hull beam. It overcomes the shortcomings of the traditional Smith method's idealization, accurately assessing the ultimate strength of models with defects, while avoiding the complex workload of finite element calculations. This invention is applicable to structural strength verification and navigation safety assessment during the ship design phase, and has broad application prospects.
Claims
1. A method for assessing the ultimate strength of ship hull beams using LSTM-Smith technology, taking into account initial defects, characterized in that... include: S1: Obtain sample data of stiffened plate units with initial defects within a typical ship size range. Each sample data consists of stiffener type, plate flexibility coefficient, stiffener flexibility coefficient, material yield strength, and stress-strain curve. S2: Sample data preprocessing; S3: Construct an LSTM model using the Huber loss function with initial point penalty; train the LSTM model using preprocessed sample data; S4: Embed the trained LSTM model into the Smith method's computational flow to correct the ultimate strength calculation of stiffened plate elements with defects.
2. The LSTM-Smith hull beam ultimate strength assessment method considering initial defects as described in claim 1, characterized in that, In step S1, the stiffened plate element model containing initial defects covers different stiffener types, geometric parameters, and material parameters; finite element ultimate strength analysis is performed on each stiffened plate element model containing initial defects to obtain curves composed of stress and relative strain data.
3. The LSTM-Smith hull beam ultimate strength assessment method considering initial defects as described in claim 2, characterized in that, The types of reinforcing ribs include angle steel, T-shaped steel, and flat steel. The geometric parameters include the flexibility coefficient of the plate and the flexibility coefficient of the ribs. The material parameters include the material type and yield strength.
4. The LSTM-Smith hull beam ultimate strength assessment method considering initial defects as described in claim 1, characterized in that, Initial defects include the initial deformation of the plate. Initial cylindrical deformation of the rib Initial deformation of ribs by lateral tilt The initial defect is applied using a formula method.
5. The LSTM-Smith hull beam ultimate strength assessment method considering initial defects as described in claim 4, characterized in that, in, , , The coordinates represent the length, width, and thickness of the stiffening plate, respectively. , The flexibility coefficient of the display board; Indicates the thickness of the stiffening strip; , , For the length of the stiffening plate, The width of the stiffening plate. The web height of the tendon. Let the half-wave number of plate buckling be defined as satisfying The smallest positive integer.
6. The LSTM-Smith hull beam ultimate strength assessment method considering initial defects as described in claim 1, characterized in that, In step S2, sample data preprocessing includes: performing unique thermal encoding on stiffener types and converting category data into binary vectors; using the Z-score method to standardize the geometric and material parameters of the stiffened plate element model to ensure dimensional balance; interpolating the stress-strain curves to ensure consistent sample time step lengths; and dividing all preprocessed sample data into training and test sets according to a specific ratio.
7. The LSTM-Smith hull beam ultimate strength assessment method considering initial defects as described in claim 6, characterized in that, The training set and the test set are divided into 80% and 20% respectively.
8. The LSTM-Smith hull beam ultimate strength assessment method considering initial defects according to claim 1, characterized in that, In step S3, the LSTM model adopts a four-layer stacked structure, with the number of nodes in each layer being 512, 256, 128, and 64 respectively.
9. The LSTM-Smith method for assessing the ultimate strength of hull beams taking into account initial defects as described in claim 1, characterized in that, In step S3, the Huber loss function, which includes an initial point penalty, is as follows: in, For the true value, For predicted values, This is the critical value of the error; The purpose of introducing an "initial point penalty term" is to force the model to minimize the prediction error in the first time step and avoid the accumulation of bias over time. in, This is the predicted value for the first time step. This is the true value at the first time step. For penalty weighting; Ultimately, the total loss function consists of the Huber loss and the initial point penalty: in, This represents the mean of the loss.
10. The LSTM-Smith method for assessing the ultimate strength of hull beams taking into account initial defects as described in claim 1, characterized in that, In step S4, the cross section of the hull beam is divided into hard-corner elements, plate elements, and stiffened plate elements, and each element is numbered to determine the model input information for each element; the location of the initial defect is determined, the stiffened plate element number where the defect is located is obtained, the corresponding model input information is input into the LSTM model, the corrected stress-strain curve is obtained, and the stress calculation in the traditional Smith method is corrected.