Extrapolation method and device for total longitudinal stress of ships under actual sea conditions

Abstract

The present invention relates to a method and device for extrapolating the total longitudinal stress of a ship under actual sea conditions. The method comprises: constructing a high sea condition total longitudinal stress extrapolation prediction model IETLSF for extrapolating the total longitudinal stress of a longitudinal strength member under actual sea conditions, wherein the high sea condition total longitudinal stress extrapolation prediction model IETLSF is used to extrapolate a node of the longitudinal strength member and the measured total longitudinal stress value of the node under actual sea conditions to determine the extrapolated predicted value of the total longitudinal stress of the node under high sea conditions. The present invention can accurately extrapolate and determine the total longitudinal stress under high sea conditions based on the measured total longitudinal stress under medium and low sea conditions, thereby meeting the need for using the total longitudinal stress under high sea conditions to assess ship safety.

Description

Technical Field

[0001] The present invention relates to an extrapolation method and device, in particular to an extrapolation method and device for the total longitudinal stress of a ship under actual sea conditions. Background Art

[0002] The total longitudinal stress of the hull's longitudinal strength members is an important part of the hull strength verification. When conducting the hull strength verification, the ship is generally placed on the waves, the total longitudinal bending moment is calculated, and the calculated total longitudinal bending moment is loaded onto the equivalent beam or the finite element model of the ship. The total longitudinal stress of the hull's longitudinal strength members can be obtained through finite element theory calculation for comparison with the allowable stress. This is the main method for calculating the total longitudinal stress of the hull's longitudinal strength members to date.

[0003] During full-scale ship testing, ships are typically placed in a Class 5 sea state environment with a wave height of approximately 3 meters. Class 5 sea state environment refers to medium-low sea conditions, and the high sea conditions corresponding to medium-low sea conditions generally refer to Class 8-9 sea conditions. For ships in medium-low sea conditions, the total longitudinal stress of a longitudinal strength member can be measured using existing common methods. However, in high sea conditions, the total longitudinal stress of longitudinal strength members cannot generally be obtained through direct measurement. Therefore, it is generally necessary to extrapolate the total longitudinal stress obtained in medium-low sea conditions to determine the total longitudinal stress in high sea conditions. This extrapolation of the total longitudinal stress in high sea conditions can then be used to evaluate the ship's seaworthiness in high sea conditions.

[0004] At present, the extrapolation method is generally used to determine the total longitudinal stress under high sea conditions. However, the use of empirical formulas for extrapolation will lead to large deviations and need to be corrected. For example, periodic correction is calculated by numerical simulation, nonlinear slamming is corrected by model test results, and wave height is corrected by linear correction.

[0005] Taking into account the three-dimensional data of model tests, numerical calculations and actual ship tests, there will be differences and insufficient consideration of nonlinear terms. Therefore, how to accurately use the extrapolation method to determine the total longitudinal stress under high sea conditions is a technical problem that urgently needs to be solved. Summary of the Invention

[0006] The purpose of the present invention is to overcome the shortcomings of the existing technology and provide a method and device for extrapolating the total longitudinal stress of a ship under actual sea conditions. Based on the actual measured total longitudinal stress under medium and low sea conditions, the method and device can accurately extrapolate the total longitudinal stress under high sea conditions, thereby meeting the need for using the total longitudinal stress under high sea conditions to assess the safety of a ship.

[0007] According to the technical solution provided by the present invention, a method for extrapolating the total longitudinal stress of a ship under actual sea conditions is used to extrapolate the total longitudinal stress of a longitudinal strength member on the ship under actual sea conditions to determine the total longitudinal stress of the longitudinal strength member under high sea conditions. The method for extrapolating the total longitudinal stress under actual sea conditions includes:

[0008] A high sea condition total longitudinal stress extrapolation prediction model IETLSF is constructed for extrapolating the total longitudinal stress of the longitudinal strength member under actual sea conditions, wherein:

[0009] When constructing the IETLSF model for extrapolating and predicting total longitudinal stress under high sea conditions, a basic model for extrapolating and predicting total longitudinal stress under multiple sea conditions, MCSFP, is first constructed based on a neural network. When constructing the basic model for extrapolating and predicting total longitudinal stress under multiple sea conditions, a standard dataset for extrapolating stress under multiple sea conditions in a computational domain generated based on the longitudinal strength component working condition design space is used for training, so that the basic model for extrapolating and predicting total longitudinal stress under multiple sea conditions, MCSFP, is generated after the training reaches a target state of the basic model.

[0010] The standard data set for stress extrapolation under multiple sea conditions in the computational domain includes stress extrapolation data samples under multiple sea conditions in the computational domain, and any stress extrapolation data sample under multiple sea conditions in the computational domain includes a wave load under a working condition and a total longitudinal stress under the wave load under the working condition, wherein the working condition includes medium, low and high sea conditions;

[0011] Producing a standard dataset of the actual sea condition and actual ship test of the longitudinal strength member, configuring the produced standard dataset of the actual sea condition and actual ship test as a target domain, and performing knowledge transfer learning on the basic model MCSFP for extrapolation prediction of total longitudinal stress under multiple sea conditions based on the configured target domain, so as to generate the extrapolation prediction model IETLSF for total longitudinal stress under high sea conditions after the knowledge transfer learning;

[0012] For a node of the longitudinal strength member and the measured total longitudinal stress value of the node under actual sea conditions, the total longitudinal stress extrapolation prediction model IETLSF under high sea conditions is used to extrapolate and determine the extrapolated predicted value of the total longitudinal stress of the node under high sea conditions.

[0013] For the basic model MCSFP for extrapolating total longitudinal stress prediction under multiple sea conditions, we have:

[0014]

[0015] Among them, R D′ (X (i) ,θ MCSFP ) is the loss function of the MCSFP basic model for total longitudinal stress extrapolation prediction under multiple sea conditions; N1*M1 is the number of stress extrapolation data samples under multiple sea conditions in the standard data set for stress extrapolation under multiple sea conditions in the computational domain; θ MCSFPThe weights w trained between neurons in the basic model MCSFP for extrapolating and predicting total longitudinal stress under multiple sea conditions are ij and bias b j The set of f(X (i) θ MCSFP ) is the output function of the output layer of the MCSFP basic model for total longitudinal stress extrapolation prediction under multiple sea conditions, X (i) is the wave load of the total longitudinal stress of the i-th computational domain multi-sea state stress extrapolation data sample in the computational domain multi-sea state stress extrapolation standard data set, is the total longitudinal stress of the i-th computational domain multi-sea state stress extrapolation data sample in the computational domain multi-sea state stress extrapolation standard data set, The predicted value of MCSFP of the basic model is extrapolated to predict the total longitudinal stress under multiple sea conditions.

[0016] For a standard dataset of stress extrapolation under multiple sea conditions in a computational domain, a method for generating the standard dataset of stress extrapolation under multiple sea conditions in a computational domain includes:

[0017] For the longitudinal strength member, a wave load sample data set is generated based on the Latin hypercube test method and the working condition design space;

[0018] For the longitudinal strength member, generating a sample data set of total longitudinal stress of the longitudinal strength member based on a wave load sample data set and a finite element theory analysis method;

[0019] The component total longitudinal stress sample data set is mean normalized to generate a standard data set for stress extrapolation in multiple sea conditions in the computational domain suitable for transmission between neurons.

[0020] For the sample data set of total longitudinal stress of components, we have:

[0021]

[0022] Where D is the sample data set of total longitudinal stress of components, is the wave load subset, is the total longitudinal stress subset corresponding to the wave load subset, {x1, x2, x3} are the coordinate values of the grid nodes of the longitudinal strength components, {x4, x5, x6, x7} are the ship speed, wave height, period, and heading angle under a working condition, respectively, and {y1} is the total longitudinal stress value;

[0023] N1 finite element mesh nodes of the longitudinal strength member; is the minimum coordinate value of the finite element mesh node of the longitudinal strength member, is the maximum coordinate value of a finite element mesh node of the longitudinal strength member; is the Latin hypercube test sample condition with the number of samples being M1 after 5 iterations; is the lower boundary value of the Latin hypercube test condition, is the upper boundary value of the Latin hypercube test condition;

[0024] is the wave load calculation for the Latin hypercube test sample condition with sample number M1; Mesh is the mesh node coordinate generation function of the finite element of the longitudinal strength member; Patran is the total longitudinal stress calculation function of the longitudinal strength member.

[0025] based on When calculating the wave loads for the Latin hypercube test sample condition with sample number M1, the calculation conditions configured include hull lines, weight distribution, shear area and draft;

[0026] When calculating the total longitudinal stress using the total longitudinal stress calculation function Patran, the configured finite element calculation conditions include the wave loads on the longitudinal strength components and the bow and stern boundary conditions.

[0027] When performing mean normalization on the component total longitudinal stress sample data set, the coordinate values of the grid nodes, speed, wave height, period, heading angle and total longitudinal stress values are all normalized.

[0028] The mean normalization process is as follows:

[0029]

[0030] in: is the mean value of coordinate value, ship speed, period, wave height, heading angle or total longitudinal stress value, D s is the variance corresponding to the coordinate value, speed, period, wave height, heading angle or total longitudinal stress value; D′ is the normalized value of the mean corresponding to the coordinate value, speed, period, wave height, heading angle or total longitudinal stress value.

[0031] When performing knowledge transfer learning on the MCSFP basic model for extrapolating and predicting total longitudinal stress under multiple sea conditions, it includes:

[0032] Five layers of neurons are added to the output layer of the basic model MCSFP for extrapolating and predicting total longitudinal stress under multiple sea conditions. The standard dataset of real sea condition and real ship test is used as the target domain, and the weight and bias set θ between neurons in the basic model MCSFP for extrapolating and predicting total longitudinal stress under multiple sea conditions is frozen. CSFP The added five-layer neurons are trained until the target training state of the prediction model is reached to generate the total longitudinal stress extrapolation prediction model IETLSF under high sea conditions.

[0033] For the total longitudinal stress extrapolation prediction model IETLSF under high sea conditions, we have:

[0034]

[0035] in, is the loss function of the IETLSF extrapolation prediction model for total longitudinal stress under high sea conditions; θ exp The weights and bias sets between the five added layers of neurons; is the output function of the output layer of the IETLSF model for total longitudinal stress extrapolation prediction under high sea conditions, is the total longitudinal stress value of the ith real ship test in the standard data set of real sea condition real ship test, is the predicted value of the total longitudinal stress extrapolation prediction model IETLSF under high sea conditions; M2 is the number of actual ship test conditions of the longitudinal strength components, N3 is the number of boundary characteristic points of the longitudinal strength components, and N2 is the number of measured nodes of the longitudinal strength components.

[0036] For the standard data set of real sea condition and real ship test, there are:

[0037]

[0038] Among them, D exp It is the basic data set for real sea condition and real ship test; is the input set of real ship test samples, the input set of real ship test samples Including coordinates, speed, wave height, period, heading angle and the output values of the basic model MCSFP for total longitudinal stress extrapolation prediction under multiple sea conditions. The total longitudinal stress value set of the actual sea condition and actual ship, which is the calculated values of the boundary characteristic points of the N3 longitudinal strength members and the measured values of the total longitudinal stress of the N2 measured nodes corresponding to the M2 test conditions;

[0039] Basic data set D for real sea condition and real ship test exp Mean normalization is performed to form a standard data set for real sea conditions and real ship tests.

[0040] An extrapolation device for total longitudinal stress of a ship under actual sea conditions includes a total longitudinal stress extrapolation processor, wherein:

[0041] For the measured total longitudinal stress value of any longitudinal strength member, the total longitudinal stress extrapolation processor uses the above extrapolation method to generate the extrapolated predicted value of the total longitudinal stress under high sea conditions.

[0042] The advantages of the present invention are as follows: the basic model MCSFP for extrapolating and predicting total longitudinal stress under multiple sea conditions based on a neural network replaces the traditional finite element theory. From the perspective of machine learning, this method learns the finite element theoretical knowledge through the calculation domain stress field data set, and is verified to obtain the basic model MCSFP for extrapolating and predicting total longitudinal stress under multiple sea conditions that satisfies a smaller loss function. Therefore, the total longitudinal stress value at any point in the structural space can be obtained.

[0043] Taking into account the stress errors between actual ship tests and finite element calculations at the same structural node under the same sea conditions and navigation status, an extrapolation prediction model for total longitudinal stress under high sea conditions, IETLSF, is established based on transfer learning. That is, a new network is constructed and trained on the basis of the basic model for extrapolation prediction of total longitudinal stress under multiple sea conditions, MCSFP, and finally an extrapolation prediction model for total longitudinal stress under high sea conditions, IETLSF, is obtained. The present invention uses deep learning and transfer learning to integrate wave loads and finite element theory with historical experience of sea tests in large winds and waves, realizing the application of machine learning in the field of stress inference under high sea conditions. The total longitudinal stress under high sea conditions can be accurately extrapolated based on the measured total longitudinal stress under medium and low sea conditions, meeting the demand for ship safety assessment using the total longitudinal stress under high sea conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 The figure is a flow chart of an embodiment of the extrapolation of total longitudinal stress under actual sea conditions according to the present invention.

[0045] Figure 2 Schematic diagram of an embodiment of constructing the IETLSF model for extrapolating and predicting total longitudinal stress under high sea conditions according to the present invention.

[0046] Figure 3 This is a schematic diagram of the features of the standard data set for actual ship tests using the deck as an example of longitudinal strength member selection in the present invention. DETAILED DESCRIPTION

[0047] The present invention will be further described below with reference to specific drawings and embodiments.

[0048] In order to accurately extrapolate the total longitudinal stress under high sea conditions based on the total longitudinal stress actually measured under medium and low sea conditions, a method for extrapolating the total longitudinal stress of a ship under actual sea conditions is provided. In one embodiment of the present invention, the total longitudinal stress of a longitudinal strength member on the ship under actual sea conditions is extrapolated to determine the total longitudinal stress of the longitudinal strength member under high sea conditions. The method for extrapolating the total longitudinal stress under actual sea conditions includes:

[0049] A high sea condition total longitudinal stress extrapolation prediction model IETLSF is constructed for extrapolating the total longitudinal stress of the longitudinal strength member under actual sea conditions, wherein:

[0050] When constructing the IETLSF model for extrapolating and predicting total longitudinal stress under high sea conditions, a basic model for extrapolating and predicting total longitudinal stress under multiple sea conditions, MCSFP, is first constructed based on a neural network. When constructing the basic model for extrapolating and predicting total longitudinal stress under multiple sea conditions, a standard dataset for extrapolating stress under multiple sea conditions in a computational domain generated based on the longitudinal strength component working condition design space is used for training, so that the basic model for extrapolating and predicting total longitudinal stress under multiple sea conditions, MCSFP, is generated after the training reaches a target state of the basic model.

[0051] The standard data set for stress extrapolation under multiple sea conditions in the computational domain includes stress extrapolation data samples under multiple sea conditions in the computational domain, and any stress extrapolation data sample under multiple sea conditions in the computational domain includes a wave load under a working condition and a total longitudinal stress under the wave load under the working condition, wherein the working condition includes medium, low and high sea conditions;

[0052] Producing a standard dataset of the actual sea condition and actual ship test of the longitudinal strength member, configuring the produced standard dataset of the actual sea condition and actual ship test as a target domain, and performing knowledge transfer learning on the basic model MCSFP for extrapolation prediction of total longitudinal stress under multiple sea conditions based on the configured target domain, so as to generate the extrapolation prediction model IETLSF for total longitudinal stress under high sea conditions after the knowledge transfer learning;

[0053] For a node of the longitudinal strength member and the measured total longitudinal stress value of the node under actual sea conditions, the total longitudinal stress extrapolation prediction model IETLSF under high sea conditions is used to extrapolate and determine the extrapolated predicted value of the total longitudinal stress of the node under high sea conditions.

[0054] It can be seen from the above description that the longitudinal strength member can be one of the deck, outer plate or inner bottom plate mentioned above. Of course, it can also be other longitudinal strength members in the ship. The type of longitudinal strength member can be selected according to needs, so as to meet the extrapolation of the total longitudinal stress of the longitudinal strength.

[0055] In specific implementations, the actual sea conditions refer to the aforementioned medium and low sea conditions, and high sea conditions refer to sea conditions with wave heights higher than those in medium and low sea conditions. The specific conditions of medium and low sea conditions and high sea conditions may be consistent with existing conditions. Extrapolating the total longitudinal stress under actual sea conditions specifically refers to determining the total longitudinal stress of a longitudinal strength member under high sea conditions based on the predicted total longitudinal stress of the longitudinal strength member under actual sea conditions.

[0056] In one embodiment of the present invention, after the type of longitudinal strength member is selected, it is necessary to construct an IETLSF (Intelligent Extrapolation for Total Longitudinal Stress Field) prediction model for the selected longitudinal strength member under high sea conditions. Figure 1 The flowchart of an embodiment of constructing the IETLSF model for extrapolating total longitudinal stress under high sea conditions is shown in FIG. Figure 1 The process shown in the figure details the process of constructing the IETLSF extrapolation prediction model for total longitudinal stress under high sea conditions.

[0057] Figure 1 and Figure 2When constructing the IELTSF model for extrapolating total longitudinal stress under high sea conditions, it is generally necessary to first construct the MCSFP (Stress field prediction for multi-condition calculation) basic model for extrapolating total longitudinal stress under multiple sea conditions. Specifically, the neural network is trained using a standard dataset for extrapolating stress under multiple sea conditions in the computational domain. After the training reaches the target state of the basic model, the MCSFP basic model for extrapolating total longitudinal stress under multiple sea conditions is obtained. The neural network based on or adopted by the MCSFP basic model for extrapolating total longitudinal stress under multiple sea conditions can be a commonly used neural network in the art, such as a convolutional neural network (CNN) or a deep neural network (DNN). The specific type of neural network can be selected as needed to ensure that the MCSFP basic model for extrapolating total longitudinal stress under multiple sea conditions can be obtained through training.

[0058] In one embodiment of the present invention, a method for generating a standard dataset for extrapolating stress under multiple sea conditions in a computational domain includes:

[0059] For the longitudinal strength member, a wave load sample data set is generated based on the Latin hypercube test method and the working condition design space;

[0060] For the longitudinal strength member, generating a sample data set of total longitudinal stress of the longitudinal strength member based on a wave load sample data set and a finite element theory analysis method;

[0061] The component total longitudinal stress sample data set is mean normalized to generate a standard data set for stress extrapolation in multiple sea conditions in the computational domain suitable for transmission between neurons.

[0062] Specifically, for longitudinal strength members, wave load sample data are generated based on the Latin hypercube test and the working condition design space, such as Figure 1 As shown; wherein, based on the Latin hypercube test, specifically refers to the boundary value based on a sea condition, the Latin hypercube test is used to generate the speed, period, wave height and heading angle under different sea conditions; generally, the sea condition includes wave height and period, and the navigation state includes speed and heading angle. The Latin hypercube test method is a commonly used test method in this technical field. The Latin hypercube test method can be used to generate the speed, period, wave height and heading angle under different sea conditions. The speed, period, wave height and heading angle under different sea conditions are used to generate a wave load sample data set based on the working condition design space; the different sea conditions here specifically include medium and low sea conditions and high sea conditions, that is, the wave load sample data set includes wave loads under medium and low sea conditions and wave loads under high sea conditions.

[0063] By loading the generated wave load sample dataset into the finite element seaworthiness model of the longitudinal strength member, a sample dataset of the total longitudinal stress of the longitudinal strength member can be generated based on the wave load sample dataset and finite element theoretical analysis methods. The sample dataset of the total longitudinal stress of the member is mean-normalized to generate a standard dataset for stress extrapolation under multiple sea conditions in the computational domain suitable for transfer between neurons.

[0064] In one embodiment of the present invention, for a sample dataset of total longitudinal stress of a component, there is:

[0065]

[0066] Where D is the sample data set of total longitudinal stress of components, is the wave load subset, For the wave load subset The corresponding total longitudinal stress subset, {x1, x2, x3} are the coordinate values of the grid nodes of the longitudinal strength components, {x4, x5, x6, x7} are the ship speed, wave height, period, and heading angle under a working condition, respectively, and {y1} is the total longitudinal stress value;

[0067] is the coordinate set of N1 finite element mesh nodes of the longitudinal strength member; is the minimum coordinate value of the finite element mesh node of the longitudinal strength member, is the maximum coordinate value of a finite element mesh node of the longitudinal strength member; is the Latin hypercube test sample condition with the number of samples being M1 after 5 iterations; is the lower boundary value of the Latin hypercube test condition, is the upper boundary value of the Latin hypercube test condition;

[0068] is the wave load calculation for the Latin hypercube test sample condition with sample number M1; Mesh is the mesh node coordinate generation function of the finite element of the longitudinal strength member; Patran is the total longitudinal stress calculation function of the longitudinal strength member.

[0069] From the above expression for constructing the total longitudinal stress sample dataset, we can see that the component total longitudinal stress sample dataset D includes the wave load subset and the wave load subset The corresponding total longitudinal stress subset, here, is the same as the wave load subset Correspondingly, specifically refers to the total longitudinal stress subset Total longitudinal stress and wave load subset The wave loads in the subgroup are in one-to-one correspondence, that is, the total longitudinal stress is the total longitudinal stress of the node of the longitudinal strength member under the corresponding wave load. It includes N1*M1 wave loads. Similarly, the total longitudinal stress subset It includes N1*M1 total longitudinal stresses.

[0070] X1 is the grid node coordinate of the longitudinal strength member, that is, the three-dimensional coordinate of the grid node. The three-dimensional coordinate is the coordinate in the spatial coordinate system established based on the ship, such as the X-axis can be constructed by the length of the ship, the Y-axis can be constructed by the width of the ship, and the Z-axis can be constructed by the height direction of the ship. The coordinates of the grid nodes obtained by finite element software correspond one to one with the coordinates of the corresponding nodes of the longitudinal strength member, which is consistent with the existing ones. In the above {x1, x2, x3}, x1 can generally be the X-axis coordinate value, x2 can generally be the Y-axis coordinate value, and x3 can generally be the Z-axis coordinate value. {x4, x5, x6, x7} are respectively the speed, wave height, period, and heading angle under a working condition, that is, x4 is the speed under a working condition, x5 is the wave height under the same working condition, x6 is the period under the same working condition, and x7 is the heading angle under the same working condition.

[0071] When using the finite element model of the longitudinal strength component of the finite element software, the mesh node coordinate generation function Mesh can be used to generate the coordinates of N1 finite element mesh nodes respectively, and then the coordinate set is obtained based on the coordinates of N1 finite element mesh nodes. When generating coordinates, the minimum coordinate value of the finite element mesh node of the longitudinal strength member is Maximum coordinate value of a finite element mesh node of a longitudinal strength member It can be obtained through finite element software, which is consistent with the existing ones.

[0072] In order to use the Latin hypercube test method, after 5 iterations, the Latin hypercube test sample condition with a sample number of M1 can be obtained, that is, the Latin hypercube test sample condition It includes M1 Latin hypercube test sample conditions. Based on the sea conditions and navigation status, the Latin hypercube test method is used to generate Latin hypercube test sample conditions. That is, for any generated Latin hypercube test sample condition, the speed, period, wave height and heading angle under the same condition are generally included.

[0073] When using the Latin hypercube test method to generate Latin hypercube test sample conditions, it is generally necessary to configure the lower boundary value X of the test condition in the condition design space. low And the upper boundary value X of the test condition up , where, for the speed, the lower limit value of the speed is X low , upper boundary value X up Generally, it can be set according to the working scenario of the ship, and the lower boundary value of the cycle X low With the upper boundary value X upThe range formed by the time is usually 3.5-18s; for wave height, the wave height lower boundary value X low With the upper boundary value X up The range formed by the time is usually 0.5-16.5m; for the heading angle, the lower boundary value of the heading angle X low With the upper boundary value X up The range formed by the time is usually 0-180 degrees. For speed, the lower limit value of speed is X low The maximum speed is generally related to the characteristics of the ship and can be determined by the design parameters of the ship.

[0074] Wave load calculation for the Latin hypercube test sample case with sample number M1 Specifically refers to the wave load corresponding to the M1 Latin hypercube test sample conditions, wave load calculation Generally, it can be provided and implemented by the wave load calculation tool MSWLF commonly used in this technical field, based on When calculating wave loads for M1 Latin Hypercube test sample conditions, the configured calculation conditions include hull lines, weight distribution, shear area, and draft. The specific configuration of these calculation conditions is determined by the specific vessel on which the longitudinal strength member is located. Specifically, after configuring the wave load calculation conditions, the corresponding wave loads can be directly calculated for M1 Latin Hypercube test sample conditions.

[0075] Finite element theory calculations can be used to calculate the total longitudinal stress under each wave load for the finite element model of the longitudinal strength member. In one embodiment of the present invention, the total longitudinal stress is calculated and generated using a total longitudinal stress calculation function, Patran. The total longitudinal stress calculation function, Patran, can be provided by the commonly used existing total longitudinal stress calculation tool, Patran. When calculating the total longitudinal stress using the total longitudinal stress calculation function, the configured finite element calculation conditions include the wave loads applied to the longitudinal strength member and the bow and stern boundary conditions. The wave loads applied are the wave loads determined above, and the bow and stern boundary conditions can be determined based on the characteristic parameters of the ship.

[0076] In one embodiment of the present invention, when performing mean normalization processing on the component total longitudinal stress sample data set, the coordinate values, ship speed, wave height, period, heading angle and total longitudinal stress values of the grid nodes are all subjected to mean normalization processing, wherein,

[0077] The mean normalization process is as follows:

[0078]

[0079] in: is the mean value of coordinate value, ship speed, period, wave height, heading angle or total longitudinal stress value, Ds is the variance corresponding to the coordinate value, speed, period, wave height, heading angle or total longitudinal stress value; D′ is the normalized value of the mean corresponding to the coordinate value, speed, period, wave height, heading angle or total longitudinal stress value.

[0080] To obtain a standard dataset for stress extrapolation under multiple sea conditions in the computational domain, the component total longitudinal stress sample dataset must be mean-normalized. This includes normalizing the coordinate values, speed, period, wave height, heading angle, or total longitudinal stress values. The above-mentioned methods can be used for mean normalization. For details, refer to the mean normalization used in the above expression. This results in a mean-normalized subset of wave loads and a mean-normalized subset of total longitudinal stress.

[0081] In one embodiment of the present invention, the basic model MCSFP for extrapolating and predicting total longitudinal stress under multiple sea conditions is:

[0082]

[0083] Among them, R D′ (X (i) ,θ MCSFP ) is the loss function of the MCSFP basic model for total longitudinal stress extrapolation prediction under multiple sea conditions; N1*M1 is the number of stress extrapolation data samples under multiple sea conditions in the standard data set for stress extrapolation under multiple sea conditions in the computational domain; θ MCSFP The weights w trained between neurons in the basic model MCSFP for extrapolating and predicting total longitudinal stress under multiple sea conditions are ij and bias b j The set of f(X (i) θ MCSFP ) is the output function of the output layer of the MCSFP basic model for total longitudinal stress extrapolation prediction under multiple sea conditions, X (i) is the wave load of the total longitudinal stress of the i-th computational domain multi-sea state stress extrapolation data sample in the computational domain multi-sea state stress extrapolation standard data set, is the total longitudinal stress of the i-th computational domain multi-sea state stress extrapolation data sample in the computational domain multi-sea state stress extrapolation standard data set, The predicted value of MCSFP of the basic model is extrapolated to predict the total longitudinal stress under multiple sea conditions.

[0084] Specifically, the predicted values of the MCSFP basic model for total longitudinal stress extrapolation prediction under multiple sea conditions are That is, the output function f(X (i) θ MCSFP ) outputs the predicted value.

[0085] From the above description, it can be seen that the basic model MCSFP for extrapolating and predicting total longitudinal stress under multiple sea conditions is based on a deep neural network. That is, after the DNN is trained using the standard dataset for extrapolating stress under multiple sea conditions in the computational domain and reaches the target training state, the basic model MCSFP for extrapolating and predicting total longitudinal stress under multiple sea conditions can be obtained.

[0086] Figure 2 An embodiment of the basic model MCSFP for extrapolating and predicting total longitudinal stress under multiple sea conditions is shown in the figure. Specifically, the basic model MCSFP for extrapolating and predicting total longitudinal stress under multiple sea conditions uses the coordinate values of the finite element model grid nodes as input and the total longitudinal stress values of the nodes as output. Therefore, when training using the standard dataset for stress extrapolation under multiple sea conditions in the computational domain, the sample nodes after mean normalization are used as the input of the basic model MCSFP for extrapolating and predicting total longitudinal stress under multiple sea conditions, and the total longitudinal stress subset after mean normalization is used as the output of the basic model MCSFP for extrapolating and predicting total longitudinal stress under multiple sea conditions.

[0087] During specific training, the convergence of the loss function is used as the basic model target training state. When the training state is reached, θ can be obtained. MCSFP The MCSFP basic model for extrapolating and predicting total longitudinal stress under multiple sea conditions is used. Figure 2 In the network structure, there are five hidden layers, where the number of neurons in the hidden layer is Nh1; the weights and biases between neurons in the entire network structure are w ij and b j During training, the training conditions can be configured as follows: 500 iterations, the Adam algorithm is used for gradient descent of the loss function, the ReLU function is used for the activation function, the initial learning rate α is 0.0001, and Nh1 = 256.

[0088] Therefore, after reaching the target training state of the basic model, the loss function R of the basic model MCSFP for total longitudinal stress extrapolation prediction under multiple sea conditions is obtained: D′ (X (i) ,θ MCSFP ), when the standard dataset of stress extrapolation under multiple sea conditions in the computational domain is used for training, the size of the total longitudinal stress subset after mean normalization is N1*M1.

[0089] In one embodiment of the present invention, when performing knowledge transfer learning on the MCSFP basic model for extrapolating and predicting total longitudinal stress under multiple sea conditions, the method includes:

[0090] Five layers of neurons are added to the output layer of the basic model MCSFP for extrapolating and predicting total longitudinal stress under multiple sea conditions. The standard dataset of real sea condition and real ship test is used as the target domain, and the weight and bias set θ between neurons in the basic model MCSFP for extrapolating and predicting total longitudinal stress under multiple sea conditions is frozen. CSFPThe added five-layer neurons are trained until the target training state of the prediction model is reached to generate the total longitudinal stress extrapolation prediction model IETLSF under high sea conditions.

[0091] Transfer learning, as a machine learning method, is increasingly becoming a research hotspot in various fields, such as image, audio, natural language processing, and autonomous driving. It can transfer the knowledge obtained from training on the source task to the target task. It is suitable for application scenarios such as stress calculation and actual ship testing where data from different domains are corrected. In essence, it is a learning process of knowledge transfer.

[0092] Figure 2 In the high sea condition extrapolation prediction model IETLSF, the total longitudinal stress extrapolation prediction model includes the basic model MCSFP for extrapolation prediction of total longitudinal stress under multiple sea conditions and five layers of neurons. The five layers of neurons are sequentially added to the output layer of the basic model MCSFP for extrapolation prediction of total longitudinal stress under multiple sea conditions. The added five layers of neurons are directly added to the output layer of the basic model MCSFP for extrapolation prediction of total longitudinal stress under multiple sea conditions, and the neurons in the fifth layer are used as the output layer of the high sea condition extrapolation prediction model IETLSF. In specific implementation, after adding the five layers of neurons, the added five layers of neurons need to be trained. During training, the weights w between the neurons in the basic model MCSFP for extrapolation prediction of total longitudinal stress under multiple sea conditions are maintained. ij and bias b j The set θ CSFP , that is, the weights w between neurons in the basic model MCSFP for extrapolating total longitudinal stress under multiple frozen sea conditions ij and bias b j The set θ CSFP , only the weights and bias θ between the five added layers of neurons are changed exp .

[0093] When the five added layers of neurons are trained to the target training state, the IETLSF (extrapolated total longitudinal stress prediction model) for high sea conditions is obtained. Specifically, the training conditions can be configured as follows: 500 iterations, the Adam algorithm for gradient descent of the loss function, the ReLU function for the activation function, an initial learning rate of 0.00001, and K2 = 256*256*256*256*256. This means that K2 determines the number of neurons in each layer.

[0094] Figure 2In the IETLSF model for extrapolating total longitudinal stress under high sea conditions, the sea conditions, navigation status, node coordinates of longitudinal strength components, and the output of the basic model for extrapolating total longitudinal stress under multiple sea conditions MCSFP are used as the input layer. The output of the basic model for extrapolating total longitudinal stress under multiple sea conditions MCSFP is the total longitudinal stress prediction value corresponding to the standard dataset for stress extrapolation under multiple sea conditions in the computational domain. The hidden layer is a network of five layers of neurons, and the output layer is the total longitudinal stress prediction value under high sea conditions. During training, the weights and biases θ between the neurons in the basic model for extrapolating total longitudinal stress under multiple sea conditions MCSFP are maintained. MCSFP , that is, the weights and biases θ between neurons in the basic model MCSFP for extrapolating total longitudinal stress under multiple frozen sea conditions MCSFP , only the weights and bias θ between the five layers of neurons are changed exp .

[0095] In one embodiment of the present invention, for the total longitudinal stress extrapolation prediction model IETLSF under high sea conditions, the following is obtained:

[0096]

[0097] Among them, R De′xp (X exp ,θ MCSFP θ exp ) is the loss function of the total longitudinal stress extrapolation prediction model IETLSF under high sea conditions; θ exp The weights and bias sets between the five added layers of neurons; is the output function of the output layer of the IETLSF model for total longitudinal stress extrapolation prediction under high sea conditions, is the total longitudinal stress value of the ith real ship test in the standard data set of real sea condition real ship test, is the predicted value of the total longitudinal stress extrapolation prediction model IETLSF under high sea conditions; M2 is the number of actual ship test conditions of the longitudinal strength components, N3 is the number of boundary characteristic points of the longitudinal strength components, and N2 is the number of measured nodes of the longitudinal strength components.

[0098] Specifically, when the total longitudinal stress extrapolation prediction model IETLSF under high sea conditions is obtained through training, the loss function of the total longitudinal stress extrapolation prediction model IETLSF under high sea conditions can be obtained. Predicted values of the total longitudinal stress extrapolation prediction model IETLSF under high sea conditions This is the output function The corresponding output value.

[0099] In one embodiment of the present invention, for a standard data set of real sea condition and real ship test, there is:

[0100]

[0101] Among them, D exp It is the basic data set for real sea condition and real ship test; is the input set of real ship test samples, the input set of real ship test samples Including coordinates, speed, wave height, period, heading angle and the output values of the basic model MCSFP for total longitudinal stress extrapolation prediction under multiple sea conditions. The total longitudinal stress value set of the actual sea condition and actual ship, which is the calculated values of the boundary characteristic points of the N3 longitudinal strength members and the measured values of the total longitudinal stress of the N2 measured nodes corresponding to the M2 test conditions;

[0102] Basic data set D for real sea condition and real ship test exp Mean normalization is performed to form a standard data set for real sea conditions and real ship tests.

[0103] Specifically, x8 is the predicted value output by the MCSFP basic model for total longitudinal stress extrapolation prediction under multiple sea conditions. exp When performing mean normalization, please refer to the above-mentioned mean normalization method. Specifically, when performing mean normalization, it is necessary to input the actual ship test sample set as well as All elements in are mean normalized.

[0104] Set of total longitudinal stress values of real ships in real sea conditions As the actual ship test sample output set, and the actual ship test sample input set After mean normalization, they can be configured as the target domain.

[0105] The number of actual ship test conditions M2 of longitudinal strength components refers to the number of actual ship tests conducted under different sea conditions. Therefore, the basic data set D of the actual sea condition actual ship test of longitudinal strength components is exp , including a number of measured node data groups located at the longitudinal strength members and a number of boundary node data groups located at the longitudinal strength members, Figure 3 In the description, the longitudinal strength member is taken as an example of the deck. Several measured nodes are arranged on the deck, and several boundary nodes are selected. The boundary nodes are the nodes on the boundary contour of the deck. The number of measured nodes N2 and the number of boundary nodes N3 are related to the longitudinal strength member and can be selected according to needs.

[0106] After arranging the measured nodes, the existing commonly used technical means can be used to obtain the total longitudinal stress values of the measured nodes. At this time, the measured node set of the longitudinal strength components and the total longitudinal stress measured value set corresponding to the measured nodes of the longitudinal strength components can be obtained. Similarly, after selecting the boundary nodes, the above-mentioned total longitudinal stress prediction method can be used to obtain the boundary node set of the longitudinal strength components and the total longitudinal stress calculated value set of the boundary nodes of the longitudinal strength components. At this time, the total longitudinal stress value set of the actual sea condition and the actual ship can be obtained.

[0107] Basic data set D for real ship test exp The mean normalization process is performed on the above-mentioned real ship test sample input set. And the total longitudinal stress value set of the actual ship in the actual sea condition The elements in are respectively subjected to mean normalization processing. The specific process of mean normalization processing can refer to the above description.

[0108] Therefore, the total longitudinal stress value of the ith actual ship test in the actual ship test standard data set is That is the measured value of the total longitudinal stress or the calculated value of the total longitudinal stress after the above-mentioned mean normalization processing.

[0109] For standard datasets tested under real sea conditions and on real ships, it is generally necessary to partition them into training, test, and validation sets. Therefore, the specific number of test sets, n, can be selected and determined based on actual needs to meet the requirements of training and automatic adjustment. This will not be discussed here. Of course, when using the standard dataset for stress extrapolation under multiple sea conditions in the computational domain to train the basic model for total longitudinal stress extrapolation prediction under multiple sea conditions, MCSFP generally also needs to be partitioned to form training, test, and validation sets. The specific details can be consistent with existing ones.

[0110] In summary, an extrapolation device for the total longitudinal stress of a ship under actual sea conditions can be obtained. In one embodiment of the present invention, a total longitudinal stress extrapolation processor is included, wherein:

[0111] For the measured total longitudinal stress value of any longitudinal strength member, the total longitudinal stress extrapolation processor uses the above extrapolation method to generate the extrapolated predicted value of the total longitudinal stress under high sea conditions.

[0112] Specifically, the total longitudinal stress extrapolation processor can generally be a commonly used computer device, that is, a total longitudinal stress extrapolation prediction model IETLSF under high sea conditions is constructed in the computer device. The total longitudinal stress extrapolation prediction model IETLSF under high sea conditions can be used to predict the actual measured value of the total longitudinal stress of the longitudinal strength member to generate an extrapolated predicted value of the total longitudinal stress under high sea conditions.

[0113] During extrapolation, it is generally necessary to determine the node locations of the longitudinal strength members, the measured total longitudinal stress values at the node locations, and the speed, wave height, heading, and period of the vessel on which the longitudinal strength members are located. This can then be achieved by using the IETLSF model for extrapolating the total longitudinal stress under high sea conditions to output the predicted extrapolated value of the total longitudinal stress under high sea conditions. Of course, during extrapolation, the node locations of the longitudinal strength members, the measured total longitudinal stress values at the node locations, and the speed, wave height, heading, and period of the vessel on which the longitudinal strength members are located need to be mean-normalized. The specific method for performing mean-normalization can be found in the above description. The mean and variance required for the specific mean-normalization can be calculated based on the corresponding values in the above-mentioned standard data set for actual sea condition and actual ship tests.

[0114] The measured total longitudinal stress value of the longitudinal strength member is obtained by using existing common total longitudinal stress detection equipment to obtain the total longitudinal stress value at the node. The specific method for obtaining the measured total longitudinal stress value can be consistent with existing methods. Of course, in this case, the measured total longitudinal stress value must be obtained under actual sea conditions.

[0115] In summary, the neural network-based basic model MCSFP for extrapolating and predicting total longitudinal stress under multiple sea conditions in the present invention replaces the traditional finite element theory. From the perspective of machine learning, this method learns the finite element theoretical knowledge through the computational domain stress field data set, and is verified to obtain the basic model MCSFP for extrapolating and predicting total longitudinal stress under multiple sea conditions that satisfies a small loss function. Therefore, the total longitudinal stress value at any point in the structural space can be obtained.

[0116] Taking into account the stress errors between actual ship tests and finite element calculations at the same structural node under the same sea conditions and navigation status, an extrapolation prediction model for total longitudinal stress under high sea conditions, IETLSF, is established based on transfer learning. That is, a new network is constructed and trained on the basis of the basic model for extrapolation prediction of total longitudinal stress under multiple sea conditions, MCSFP, and finally an extrapolation prediction model for total longitudinal stress under high sea conditions, IETLSF, is obtained. The present invention uses deep learning and transfer learning to integrate wave loads and finite element theory with historical experience of sea tests in large winds and waves, realizing the application of machine learning in the field of stress inference under high sea conditions. The total longitudinal stress under high sea conditions can be accurately extrapolated based on the measured total longitudinal stress under medium and low sea conditions, meeting the demand for ship safety assessment using the total longitudinal stress under high sea conditions.

Claims

1. A method for extrapolating the total longitudinal stress of a ship under actual sea conditions, characterized by: The method is used to extrapolate the total longitudinal stress of a longitudinal strength member on a ship under actual sea conditions to determine the total longitudinal stress of the longitudinal strength member under high sea conditions, wherein the extrapolation method of the total longitudinal stress under actual sea conditions includes: A high sea condition total longitudinal stress extrapolation prediction model IETLSF is constructed for extrapolating the total longitudinal stress of the longitudinal strength member under actual sea conditions, wherein: When constructing the IETLSF model for extrapolating and predicting total longitudinal stress under high sea conditions, a basic model for extrapolating and predicting total longitudinal stress under multiple sea conditions, MCSFP, is first constructed based on a neural network. When constructing the basic model for extrapolating and predicting total longitudinal stress under multiple sea conditions, a standard dataset for extrapolating stress under multiple sea conditions in a computational domain generated based on the longitudinal strength component working condition design space is used for training, so that the basic model for extrapolating and predicting total longitudinal stress under multiple sea conditions, MCSFP, is generated after the training reaches a target state of the basic model. The standard data set for stress extrapolation under multiple sea conditions in the computational domain includes stress extrapolation data samples under multiple sea conditions in the computational domain, and any stress extrapolation data sample under multiple sea conditions in the computational domain includes a wave load under a working condition and a total longitudinal stress under the wave load under the working condition, wherein the working condition includes medium, low and high sea conditions; Producing a standard dataset of the actual sea condition and actual ship test of the longitudinal strength member, configuring the produced standard dataset of the actual sea condition and actual ship test as a target domain, and performing knowledge transfer learning on the basic model MCSFP for extrapolating and predicting total longitudinal stress under multiple sea conditions based on the configured target domain, so as to generate the extrapolating and predicting model IETLSF for total longitudinal stress under high sea conditions after the knowledge transfer learning; For a node of the longitudinal strength member and the measured total longitudinal stress value of the node under actual sea conditions, use the total longitudinal stress extrapolation prediction model IETLSF under high sea conditions to determine the extrapolated predicted value of the total longitudinal stress of the node under high sea conditions; For the basic model MCSFP for extrapolating total longitudinal stress prediction under multiple sea conditions, we have: Among them, R D′ (X (i) ,θ MCSFP ) is the loss function of the MCSFP basic model for total longitudinal stress extrapolation prediction under multiple sea conditions; N1*M1 is the number of stress extrapolation data samples under multiple sea conditions in the standard data set for stress extrapolation under multiple sea conditions in the computational domain; θ MCSFP The weights w trained between neurons in the basic model MCSFP for extrapolating and predicting total longitudinal stress under multiple sea conditions are ij and bias b j The set of f(X (i) θ MCSFP ) is the output function of the output layer of the MCSFP basic model for total longitudinal stress extrapolation prediction under multiple sea conditions, X (i) is the wave load of the total longitudinal stress of the i-th computational domain multi-sea state stress extrapolation data sample in the computational domain multi-sea state stress extrapolation standard data set, is the total longitudinal stress of the i-th computational domain multi-sea state stress extrapolation data sample in the computational domain multi-sea state stress extrapolation standard data set, The predicted values of MCSFP of the basic model are extrapolated for total longitudinal stress under multiple sea conditions; When performing knowledge transfer learning on the MCSFP basic model for extrapolating and predicting total longitudinal stress under multiple sea conditions, it includes: Five layers of neurons are added to the output layer of the basic model MCSFP for extrapolating and predicting total longitudinal stress under multiple sea conditions. The standard dataset of real sea condition and real ship test is used as the target domain, and the weight and bias set θ between neurons in the basic model MCSFP for extrapolating and predicting total longitudinal stress under multiple sea conditions is frozen. CSFP Under the condition of , the added five layers of neurons are trained until the target training state of the prediction model is reached to generate the total longitudinal stress extrapolation prediction model IETLSF under high sea conditions; For the total longitudinal stress extrapolation prediction model IETLSF under high sea conditions, we have: in, is the loss function of the IETLSF extrapolation prediction model for total longitudinal stress under high sea conditions; θ exp The weights and bias sets between the five added layers of neurons; is the output function of the output layer of the IETLSF model for total longitudinal stress extrapolation prediction under high sea conditions, is the total longitudinal stress value of the ith real ship test in the standard data set of real sea condition real ship test, is the predicted value of the total longitudinal stress extrapolation prediction model IETLSF under high sea conditions; M2 is the number of actual ship test conditions of the longitudinal strength components, N3 is the number of boundary characteristic points of the longitudinal strength components, and N2 is the number of measured nodes of the longitudinal strength components.

2. The method for extrapolating the total longitudinal stress of a ship under actual sea conditions according to claim 1 is characterized by: For a standard dataset of stress extrapolation under multiple sea conditions in a computational domain, a method for generating the standard dataset of stress extrapolation under multiple sea conditions in a computational domain includes: For the longitudinal strength member, a wave load sample data set is generated based on the Latin hypercube test method and the working condition design space; For the longitudinal strength member, generating a sample data set of total longitudinal stress of the longitudinal strength member based on a wave load sample data set and a finite element theory analysis method; The component total longitudinal stress sample data set is mean normalized to generate a standard data set for stress extrapolation in multiple sea conditions in the computational domain suitable for transmission between neurons.

3. The method for extrapolating the total longitudinal stress of a ship under actual sea conditions according to claim 2 is characterized by: For the sample data set of total longitudinal stress of components, we have: Where D is the sample data set of total longitudinal stress of components, is the wave load subset, is the total longitudinal stress subset corresponding to the wave load subset, {x1, x2, x3} are the coordinate values of the grid nodes of the longitudinal strength components, {x4, x5, x6, x7} are the ship speed, wave height, period, and heading angle under a working condition, respectively, and {y1} is the total longitudinal stress value; N1 finite element mesh nodes of the longitudinal strength member; is the minimum coordinate value of the finite element mesh node of the longitudinal strength member, is the maximum coordinate value of a finite element mesh node of the longitudinal strength member; is the Latin hypercube test sample condition with the number of samples being M1 after 5 iterations; is the lower boundary value of the Latin hypercube test condition, is the upper boundary value of the Latin hypercube test condition; is the wave load calculation for the Latin hypercube test sample condition with sample number M1; Mesh is the mesh node coordinate generation function of the finite element of the longitudinal strength member; Patran is the total longitudinal stress calculation function of the longitudinal strength member.

4. The method for extrapolating the total longitudinal stress of a ship under actual sea conditions according to claim 3 is characterized by: based on When calculating the wave loads for the Latin hypercube test sample condition with sample number M1, the calculation conditions configured include hull lines, weight distribution, shear area and draft; When calculating the total longitudinal stress using the total longitudinal stress calculation function Patran, the configured finite element calculation conditions include the wave loads on the longitudinal strength components and the bow and stern boundary conditions.

5. The method for extrapolating the total longitudinal stress of a ship under actual sea conditions according to claim 3 is characterized by: When performing mean normalization on the component total longitudinal stress sample data set, the coordinate values of the grid nodes, speed, wave height, period, heading angle and total longitudinal stress values are all normalized. The method of the mean normalization process is: in: is the mean value of coordinate value, ship speed, period, wave height, heading angle or total longitudinal stress value, D s is the variance corresponding to the coordinate value, speed, period, wave height, heading angle or total longitudinal stress value; D′ is the normalized value of the mean corresponding to the coordinate value, speed, period, wave height, heading angle or total longitudinal stress value.

6. The method for extrapolating the total longitudinal stress of a ship under actual sea conditions according to claim 1 is characterized by: For the standard data set of real sea condition and real ship test, there are: Among them, D exp It is the basic data set for real sea condition and real ship test; is the input set of real ship test samples, the input set of real ship test samples Including coordinates, speed, wave height, period, heading angle and the output values of the basic model MCSFP for total longitudinal stress extrapolation prediction under multiple sea conditions. The total longitudinal stress value set of the actual sea condition and actual ship, which is the calculated values of the boundary characteristic points of the N3 longitudinal strength members and the measured values of the total longitudinal stress of the N2 measured nodes corresponding to the M2 test conditions; Basic data set D for real sea condition and real ship test exp Mean normalization is performed to form a standard data set for real sea conditions and real ship tests.

7. A device for extrapolating the total longitudinal stress of a ship under actual sea conditions, characterized by: Includes a total longitudinal stress extrapolation processor, where For the measured total longitudinal stress value of any longitudinal strength member, the total longitudinal stress extrapolation processor uses the extrapolation method of any one of claims 1 to 6 to generate an extrapolated predicted value of the total longitudinal stress under high sea conditions.

Images