Method, device and equipment for predicting additional mass coefficient and viscous damping coefficient of multi-degree-of-freedom motion of floating body, storage medium and computer program product

Abstract

The invention relates to the technical field of hydrodynamic analysis of ship and ocean engineering, in particular to a method, a device and equipment for predicting an additional mass coefficient and a viscous damping coefficient of multi-degree-of-freedom movement of a floating body, a storage medium and a computer program product. The method comprises the following steps: performing orthogonal two-dimensional slicing on each plane to generate flow field sampling points based on a floating body three-dimensional geometric model, draft and motion parameters; sampling point coordinates, initial flow field data, geometric contour discrete points, boundary conditions and motion parameters are coded into high-dimensional feature vectors through a learnable matrix, and input matrixes of all planes are constructed; predicting two-dimensional flow field data by adopting a Transform module, fusing multi-plane data through a graph attention network to generate a redundant three-dimensional flow field, and interpolating and complementing object plane missing points; the water pressure and the viscous shear force are divided along the surface area of the floating body, the inertia term and the damping term are decomposed in combination with the single-mode motion equation, the target additional mass coefficient and the target viscous damping coefficient are determined, and the accuracy of floating body hydrodynamic calculation is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of ship and ocean engineering hydrodynamic analysis, and particularly relates to a method, device, equipment, storage medium and computer program product for predicting added mass coefficients and viscous damping coefficients of multi-degree-of-freedom motion of a floating body. BACKGROUND

[0002] Rolling is more likely to occur and has a larger amplitude when a ship sails in the sea, which seriously affects the seaworthiness and living comfort of the ship. For an offshore platform, when it has six-degree-of-freedom motion (pitching, rolling, yawing, surging, heaving and swaying) in the sea, the motion amplitude is large when the sea conditions are bad, which seriously affects the safe operation of equipment on the platform and the safety of workers. The motion of the ship and the offshore platform in the wave is more obviously affected by viscosity and the nonlinearity of the motion itself. To accurately predict the motion response of the floating body, the key is to accurately calculate the hydrodynamic coefficients of the ship and the offshore platform, such as the added mass and the damping coefficient. However, the traditional method generally obtains the viscous damping coefficient by using CFD for single-mode oscillation numerical simulation, which requires a large amount of computing resources and high time cost; or the six-degree-of-freedom critical damping of the floating body is calculated by using the potential flow software, and 3% to 8% of the critical damping is used as the viscous damping matrix of the ship or the offshore platform. Without experimental data support, this method cannot guarantee the accuracy of the hydrodynamic calculation of the floating body. Therefore, how to improve the accuracy of the hydrodynamic calculation of the floating body has become a technical problem to be solved. SUMMARY

[0003] The main purpose of the present application is to provide a method, device, equipment, storage medium and computer program product for predicting added mass coefficients and viscous damping coefficients of multi-degree-of-freedom motion of a floating body, which aims to solve the technical problem of how to improve the accuracy of the hydrodynamic calculation of the floating body.

[0004] To achieve the above-mentioned purpose, the present application provides a method for predicting added mass coefficients and viscous damping coefficients of multi-degree-of-freedom motion of a floating body, which comprises the following steps:

[0005] Based on the three-dimensional geometric model, the draft depth and the frequency and amplitude of single-mode forced oscillation motion of the floating body, the outer shell of the floating body is subjected to two-dimensional slicing processing in a three-dimensional rectangular coordinate system, and flow field sampling points are generated on the X-Z plane, the Y-Z plane and the X-Y plane, respectively;

[0006] Based on the coordinates of the flow field sampling points, the initial flow field data, the discrete points of the geometric outer contour of the floating body, the boundary conditions and the motion parameters of the floating body, the learnable matrix is encoded into a high-dimensional feature vector, and an input information matrix of each plane slice is constructed;

[0007] Based on the preset floating body flow field prediction model, two-dimensional flow field prediction is performed on the input information matrix respectively, and two-dimensional flow field data of each slice is output.

[0008] Based on the graph attention network, three-dimensional fusion and correction are performed on the two-dimensional flow field data to generate a redundant three-dimensional flow field matrix, and missing data on the surface is completed by interpolation.

[0009] Based on the corrected three-dimensional flow field data, the hydrodynamic pressure and viscous shear force along the floating body surface are integrated, and the target added mass coefficient and target viscous damping coefficient are determined by combining the single modal forced oscillation motion equation.

[0010] In an embodiment, before the steps of generating flow field sampling points on the X-Z plane, Y-Z plane and X-Y plane respectively based on the three-dimensional geometric model of the floating body, the draft and the frequency and amplitude of the single modal forced oscillation motion, the method further comprises:

[0011] The first prediction module and the second prediction module are constructed based on the Transformer model, and the third prediction module is constructed based on the graph attention network, the first prediction module is used for two-dimensional flow field prediction on the X-Z plane and the Y-Z plane, the second prediction module is used for two-dimensional flow field prediction on the X-Y plane, and the third prediction module is used for integrating multi-plane slice data and correcting redundancy.

[0012] The first prediction module and the second prediction module are pre-trained based on a preset public two-dimensional computational fluid dynamics data set;

[0013] The pre-trained first prediction module and the second prediction module are optimized based on a three-dimensional self-owned data set generated by CFD, and the continuity equation, the momentum conservation equation and the surface condition are embedded as physical constraints to obtain a target first prediction module and a target second prediction module;

[0014] The parameters of the target first prediction module and the target second prediction module are fixed, and the third prediction module is trained based on a preset three-dimensional self-owned data set to obtain a target third prediction module;

[0015] The target first prediction module, the target second prediction module and the target third prediction module are integrated to obtain the preset floating body flow field prediction model.

[0016] In an embodiment, before the steps of generating flow field sampling points on the X-Z plane, Y-Z plane and X-Y plane respectively based on the three-dimensional geometric model of the floating body, the draft and the frequency and amplitude of the single modal forced oscillation motion, the method further comprises:

[0017] The three-dimensional geometric model of the floating body is geometrically verified to identify non-streamlined regions and potential numerical calculation error points in the model.

[0018] Based on the verification results, the discretization parameters of the outer shell of the float are adjusted. The discretization parameters include slice density, dividing line spacing, and surface discrete point distribution rules.

[0019] In one embodiment, after the step of performing three-dimensional fusion and correction on the two-dimensional flow field data based on a graph attention network to generate a redundant three-dimensional flow field matrix, and then filling in the missing data of the object surface through interpolation, the method further includes:

[0020] Based on the corrected three-dimensional flow field data, the confidence level of the flow field prediction is dynamically evaluated;

[0021] If the confidence level is lower than the preset threshold, the floating body motion parameters or boundary conditions are readjusted, a new input information matrix is ​​generated, and the two-dimensional flow field prediction and three-dimensional fusion steps are iteratively executed.

[0022] If the confidence level reaches the threshold, the current three-dimensional flow field data will be used as the target three-dimensional flow field data.

[0023] In one embodiment, the step of performing two-dimensional slicing of the outer shell of the floating body in a three-dimensional Cartesian coordinate system based on the three-dimensional geometric model of the floating body, its draft, and the frequency and amplitude of the single-mode forced oscillation motion, and generating flow field sampling points on the XZ, YZ, and XY planes respectively, includes:

[0024] Based on the three-dimensional geometric model of the floating body, its draft, and the frequency and amplitude of the single-mode forced oscillation motion, the outer shell of the floating body is divided into regions, resulting in regular regions and asymmetric regions.

[0025] Based on the asymmetric partitioning strategy, the asymmetric region is sliced ​​in two dimensions in a three-dimensional Cartesian coordinate system to generate asymmetric sampling points. The asymmetric partitioning strategy includes increasing the slice density in regions where the geometric curvature change is greater than a preset curvature change threshold and using non-uniform dividing line spacing in the XY plane.

[0026] Based on the rule-based partitioning strategy, the rule region is processed into two-dimensional slices in a three-dimensional Cartesian coordinate system to generate rule sampling points. The rule-based partitioning strategy includes using equidistant plane data and prioritizing the retention of planar data orthogonal to the direction of the motion degree of freedom of the floating body.

[0027] The asymmetric sampling points and the regular sampling points are used as the flow field sampling points.

[0028] In an embodiment, the step of determining the target added mass coefficient and the target viscous damping coefficient based on the corrected three-dimensional flow field data, integrating the hydrodynamic pressure and the viscous shear force along the surface of the floating body, and combining the single-mode forced oscillation equation comprises:

[0029] extracting the hydrodynamic pressure, the viscous shear force, the surface normal direction, and the position coordinates of the discrete units on the surface of the floating body in the corrected three-dimensional flow field data;

[0030] numerically integrating the hydrodynamic pressure and the viscous shear force of the discrete units respectively to obtain the total hydrodynamic force and the total hydrodynamic moment acting on the floating body by integrating the hydrodynamic pressure and the viscous shear force along the surface of the floating body;

[0031] decomposing the total hydrodynamic force and the total hydrodynamic moment into inertia terms related to acceleration and damping terms related to velocity according to the single-mode forced oscillation equation;

[0032] obtaining the target added mass coefficient and the target viscous damping coefficient by using the phase matching and amplitude scaling algorithm based on the inertia terms and the damping terms.

[0033] In addition, to achieve the above-mentioned purpose, the application further provides a floating body multi-degree-of-freedom motion added mass coefficient and viscous damping coefficient prediction device, which comprises:

[0034] a sampling module for performing two-dimensional slicing processing on the outer shell of the floating body in a two-dimensional Cartesian coordinate system based on the three-dimensional geometric model of the floating body, the draft, and the frequency and amplitude of the single-mode forced oscillation motion, and generating flow field sampling points on the X-Z plane, the Y-Z plane, and the X-Y plane, respectively;

[0035] an input matrix module for encoding the coordinates of the flow field sampling points, the initial flow field data, the discrete points of the geometric outer contour of the floating body, the boundary conditions, and the motion parameters of the floating body into high-dimensional feature vectors through a learnable matrix, and constructing an input information matrix of each plane slice;

[0036] a flow field prediction module for performing two-dimensional flow field prediction on the input information matrix based on a preset floating body flow field prediction model, and outputting two-dimensional flow field data of each slice;

[0037] a fusion correction module for performing three-dimensional fusion and correction on the two-dimensional flow field data based on a graph attention network, generating a redundant three-dimensional flow field matrix, and completing the missing data on the surface by interpolation;

[0038] The target module is configured to determine the target added mass coefficient and the target viscous damping coefficient based on the modified three-dimensional flow field data, the surface integration of hydrodynamic pressure and viscous shear force along the floating body, and the single mode forced oscillation motion equation.

[0039] In addition, to achieve the above object, the present application further provides a device for predicting added mass coefficient and viscous damping coefficient of multi-degree-of-freedom motion of a floating body, which comprises a memory, a processor, and a program for predicting added mass coefficient and viscous damping coefficient of multi-degree-of-freedom motion of a floating body stored in the memory and executable on the processor, wherein the program is configured to implement the steps of the method for predicting added mass coefficient and viscous damping coefficient of multi-degree-of-freedom motion of a floating body as described above.

[0040] In addition, to achieve the above object, the present application further provides a storage medium, wherein the storage medium stores a program for predicting added mass coefficient and viscous damping coefficient of multi-degree-of-freedom motion of a floating body, and the program is executable on a processor to implement the steps of the method for predicting added mass coefficient and viscous damping coefficient of multi-degree-of-freedom motion of a floating body as described above.

[0041] In addition, to achieve the above object, the present application further provides a computer program product, which comprises a computer program, and the computer program is executable on a processor to implement the steps of the method for predicting added mass coefficient and viscous damping coefficient of multi-degree-of-freedom motion of a floating body as described above.

[0042] The application is based on the three-dimensional geometric model of the floating body, the draft, and the frequency and amplitude of the single-mode forced oscillation motion of the floating body. The outer shell of the floating body is processed in two dimensions in the three-dimensional rectangular coordinate system, and flow field sampling points are generated on the X-Z plane, Y-Z plane and X-Y plane. Based on the coordinates of the flow field sampling points, initial flow field data, discrete points of the geometric outer contour of the floating body, boundary conditions and motion parameters of the floating body, the learnable matrix is encoded into a high-dimensional feature vector, and an input information matrix of each plane slice is constructed. Based on the preset floating body flow field prediction model, two-dimensional flow field prediction is performed on the input information matrix, and two-dimensional flow field data of each slice is output. Based on the graph attention network, the two-dimensional flow field data is fused and corrected in three dimensions to generate a redundant three-dimensional flow field matrix, and the missing data on the surface is completed by interpolation. Based on the corrected three-dimensional flow field data, the hydrodynamic pressure and viscous shear force along the surface of the floating body are integrated, and the target added mass coefficient and target viscous damping coefficient are determined by combining the single-mode forced oscillation motion equation. The two-dimensional slice processing of the three-dimensional geometric model decomposes the complex three-dimensional flow field into multiple plane high-resolution samples, reduces the calculation complexity, and improves the accuracy of the key area. The learnable matrix dynamically extracts the high-dimensional physical correlation of the flow field data, geometric contour and motion parameters, avoiding the limitations of traditional empirical formulas. The preset floating body flow field prediction model is based on multi-plane two-dimensional prediction and three-dimensional fusion and correction of the graph attention network, solving the problems of data redundancy and missing, and ensuring the spatial consistency of the flow field. Finally, through the surface integration of the hydrodynamic pressure and viscous shear force, the flow field data and inertia-damping effect are directly related by combining the single-mode motion equation, the potential flow theory assumption error is eliminated, the high-precision calculation of the six-degree-of-freedom added mass and viscous damping coefficient is realized, and the accuracy of the floating body hydrodynamic calculation is improved. BRIEF DESCRIPTION OF DRAWINGS

[0043] Figure 1 The flowchart of the first embodiment of the added mass coefficient and viscous damping coefficient prediction method of the floating body multi-degree-of-freedom motion of the application is shown.

[0044] Figure 2 The sub-flowchart of the second embodiment of the added mass coefficient and viscous damping coefficient prediction method of the floating body multi-degree-of-freedom motion of the application is shown.

[0045] Figure 3 The sub-flowchart of the third embodiment of the added mass coefficient and viscous damping coefficient prediction method of the floating body multi-degree-of-freedom motion of the application is shown.

[0046] Figure 4 The three-dimensional fluid domain diagram in one embodiment of the added mass coefficient and viscous damping coefficient prediction method of the floating body multi-degree-of-freedom motion of the application is shown.

[0047] Figure 5A slice diagram in an embodiment of the additional mass coefficient and viscous damping coefficient prediction method of the multi-degree-of-freedom motion of the floating body of the application;

[0048] Figure 6 A flow field sampling point diagram in an embodiment of the additional mass coefficient and viscous damping coefficient prediction method of the multi-degree-of-freedom motion of the floating body of the application;

[0049] Figure 7 A module structure diagram of the additional mass coefficient and viscous damping coefficient prediction device of the multi-degree-of-freedom motion of the floating body of the embodiment of the application;

[0050] Figure 8 A device structure diagram of the hardware running environment involved in the additional mass coefficient and viscous damping coefficient prediction method of the multi-degree-of-freedom motion of the floating body in the embodiment of the application.

[0051] The implementation, functional features and advantages of the application will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION

[0052] It should be understood that the specific embodiments described herein are only used to explain the application and not to limit the application.

[0053] In order to better understand the technical solutions of the application, the specific embodiments will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0054] It should be noted that the roll of the ship is more likely to occur and has a large amplitude when the ship sails in the sea, which seriously affects the seaworthiness and living comfort of the ship. For the offshore platform, when it has six-degree-of-freedom motion (pitch, roll, yaw, surge, sway, heave) in the sea, the motion amplitude is large when the sea condition is bad, which seriously affects the safe operation of the equipment on the platform and the life safety of the workers. The motion of the floating body such as the ship and the offshore platform in the wave is more obviously affected by viscosity and the nonlinear characteristics of the motion itself. To accurately predict the motion response of the floating body, the key is to accurately calculate the hydrodynamic coefficients of the ship and the offshore platform, such as the added mass and the damping coefficient. However, the traditional method generally obtains the viscous damping coefficient by using CFD for single modal oscillation numerical simulation, which requires a large amount of calculation resources and high time cost; or the six-degree-of-freedom critical damping of the floating body is calculated according to the potential flow software, and 3% to 8% of the critical damping is used as the viscous damping matrix of the ship or the offshore platform. Without experimental data support, this method often cannot guarantee the accuracy of the hydrodynamic calculation of the floating body. Therefore, how to improve the accuracy of the hydrodynamic calculation of the floating body has become a technical problem to be solved.

[0055] The main solution of the present application is: based on the three-dimensional geometric model of the floating body, the water depth, and the frequency and amplitude of the single mode forced oscillation motion, the outer shell of the floating body is processed by two-dimensional slicing in the three-dimensional rectangular coordinate system, and the flow field sampling points are generated on the X-Z plane, Y-Z plane and X-Y plane respectively; based on the coordinates of the flow field sampling points, the initial flow field data, the discrete points of the geometric outer contour of the floating body, the boundary conditions and the motion parameters of the floating body, the learnable matrix is encoded into a high-dimensional feature vector, and the input information matrix of each plane slice is constructed; based on the preset floating body flow field prediction model, the input information matrix is predicted for two-dimensional flow field respectively, and the two-dimensional flow field data of each slice is output; based on the graph attention network, the two-dimensional flow field data is fused and corrected in three dimensions to generate a redundant three-dimensional flow field matrix, and the missing data on the surface is completed by interpolation; based on the corrected three-dimensional flow field data, the hydrodynamic pressure and viscous shear force along the surface of the floating body are integrated, and the target added mass coefficient and target viscous damping coefficient are determined by combining the single mode forced oscillation motion equation.

[0056] The present application decomposes the complex three-dimensional flow field into high-resolution sampling of multiple planes through two-dimensional slicing of the three-dimensional geometric model, reduces the calculation complexity while improving the accuracy of the key area; the learnable matrix dynamically extracts the high-dimensional physical correlation of the flow field data, geometric contour and motion parameters, avoiding the limitations of traditional empirical formulas; the preset floating body flow field prediction model is based on multi-plane two-dimensional prediction and three-dimensional fusion and correction of the graph attention network, solves the problems of data redundancy and missing, and ensures the spatial consistency of the flow field; finally, through the surface integration of hydrodynamic pressure and viscous shear force, the flow field data and inertial-damping effect are directly related by combining the single mode motion equation, the hypothesis error of potential flow theory is eliminated, the high-precision calculation of six-degree-of-freedom added mass and viscous damping coefficient is realized, and the accuracy of the floating body hydrodynamic calculation is improved.

[0057] It should be noted that the execution subject of the method of the present embodiment can be a computing service device with data processing, network communication and program running functions, or a floating body multi-degree-of-freedom motion added mass coefficient and viscous damping coefficient prediction device with the same or similar functions. The present embodiment and the following embodiments will be described taking the floating body multi-degree-of-freedom motion added mass coefficient and viscous damping coefficient prediction device as an example.

[0058] Based on this, the first embodiment of the floating body multi-degree-of-freedom motion added mass coefficient and viscous damping coefficient prediction method of the present application is proposed, please refer to Figure 1 , Figure 1 The flowchart of the first embodiment of the floating body multi-degree-of-freedom motion added mass coefficient and viscous damping coefficient prediction method of the present application is shown in the figure.

[0059] In the present embodiment, the floating body multi-degree-of-freedom motion added mass coefficient and viscous damping coefficient prediction method includes the following steps:

[0060] S1: Based on the three-dimensional geometric model of the floating body, the draft, and the frequency and amplitude of the forced oscillation motion of a single mode, the outer shell of the floating body is processed in two dimensions in a three-dimensional rectangular coordinate system, and flow field sampling points are generated on the X-Z plane, Y-Z plane, and X-Y plane, respectively;

[0061] It should be noted that the three-dimensional geometric model of the floating body is a digital three-dimensional shape of the outer shell of the ship or offshore platform established by computer-aided design (CAD), which contains geometric information such as curved surfaces and edges. The draft refers to the vertical depth of the floating body immersed in still water, which affects the contact area of the floating body with the fluid and the hydrodynamic characteristics. The forced oscillation motion of a single mode refers to the periodic vibration of the floating body in a certain degree of freedom (such as roll or heave) under external excitation, while other degrees of freedom remain stationary. The three-dimensional rectangular coordinate system is a right-handed coordinate system with the center of the waterline surface of the floating body as the origin, X-axis (longitudinal), Y-axis (lateral), and Z-axis (vertical). Two-dimensional slicing refers to cutting the three-dimensional geometric model of the floating body along a certain plane (such as the X-Z plane) to generate a two-dimensional cross section for local flow field analysis. The X-Z / Y-Z / X-Y planes represent the longitudinal-vertical plane, lateral-vertical plane, and longitudinal-lateral plane, respectively, for multi-angle coverage of the flow field around the floating body. The flow field sampling points refer to the coordinate points discretely distributed on the two-dimensional slice, which are used to record flow velocity, pressure, and other flow field physical quantities.

[0062] Specifically, input the three-dimensional geometric model of the floating body (such as the CAD model of the ship hull), the draft (which determines the submerged range of the floating body), and the forced oscillation parameters of the floating body in a certain degree of freedom (such as roll frequency and amplitude). A three-dimensional rectangular coordinate system is established with the center of the waterline surface of the floating body as the origin, with the X-axis along the ship length, the Y-axis along the ship width, and the Z-axis vertically upward.

[0063] Further, the outer shell of the floating body is processed by multi-plane two-dimensional slicing, cutting along the X-Z plane (longitudinal-vertical), Y-Z plane (lateral-vertical), and X-Y plane (longitudinal-lateral) in three orthogonal directions. Increase the number of slices near the surface of the floating body and areas with significant geometric curvature changes (such as the bow and stern), and reduce the slice density in areas away from the floating body to reduce the computational load. On each slice, the grid intersection points are generated by the orthogonal intersection of the transverse and vertical dividing lines as flow field sampling points, the intersection points of the dividing lines and the outer contour of the floating body are retained, and the invalid points inside the floating body and above the free surface are deleted. According to the geometric symmetry of the floating body, non-uniform slice spacing is used for asymmetric areas (such as irregular offshore platforms), and equal interval slicing is used for regular areas (such as the middle part of the ship hull).

[0064] This step decomposes the complex three-dimensional flow field prediction problem into multiple two-dimensional local modeling through two-dimensional orthogonal slicing of the three-dimensional model, greatly reducing the computational complexity (the two-dimensional grid is of order n2 , three-dimensional n 3 ), while through dynamic sampling density adjustment, densely sampling in the area with large floating body surface curvature and severe flow field changes (such as the ship bow vortex area) to improve local accuracy, and sparsely sampling in the area away from the floating body to reduce redundant calculation. In addition, the combination of multiple plane slices (X-Z / Y-Z / X-Y) covers the full circumferential flow field of the floating body, avoids missing data from a single perspective, and ensures the comprehensiveness of the input data for subsequent neural network prediction, laying a foundation for high-precision hydrodynamic coefficient calculation.

[0065] S2: based on the coordinates of the flow field sampling points, the initial flow field data, the discrete points of the floating body geometric outer contour, the boundary conditions and the floating body motion parameters, encoding into high-dimensional feature vectors through a learnable matrix, and constructing an input information matrix of each plane slice;

[0066] It should be noted that the initial flow field data refers to the initial state parameters of the fluid under the static water working condition, including the flow velocity components (x / y / z directions), the pressure components (x / y / z directions) and the liquid volume fraction (the volume ratio of gas and liquid). The boundary condition refers to the definition of the boundary behavior of the fluid domain, such as pressure outlet, velocity inlet, wall surface (no slip condition), symmetry surface, etc. The floating body motion parameter refers to the degree of freedom direction (such as surge, roll), frequency (oscillation speed) and amplitude (oscillation amplitude) of the single mode forced oscillation motion of the floating body. The learnable matrix refers to the parameter matrix automatically optimized in the neural network training process, which is used to map the original data to a high-dimensional feature space. The high-dimensional feature vector is an abstract feature representation generated by encoding, which contains implicit correlation information of physical parameters. The input information matrix refers to the matrix composed of multiple feature vectors according to the plane slice, which is used as the input data of the neural network.

[0067] Specifically, the three-dimensional position of the sampling point, the initial flow field data, the discrete point coordinates and the sorting information (counterclockwise direction) of the material surface contour, the boundary condition and the floating body motion parameter are extracted from the flow field sampling points. Multiply the above data by different learnable matrices to generate feature vectors of uniform dimension (such as 768 dimensions). Multiply the sampling point coordinates by the learnable matrix to generate the position feature vector; multiply the flow velocity, pressure and liquid volume fraction by the learnable matrix to generate the flow field feature vector; concatenate the degree of freedom direction, frequency and amplitude to generate the motion feature vector; the boundary type and value are combined with the one-hot encoding and the learnable matrix to generate the boundary feature vector. Finally, the feature vectors are concatenated into an input information matrix according to the plane slice for subsequent neural network processing.

[0068] This step maps multiple-source heterogeneous data (coordinates, flow field, geometry, motion parameters) into high-dimensional feature vectors through learnable matrix encoding, automatically mines complex physical relationships such as flow rate-pressure coupling and motion-geometry correlation, and avoids the limitations of traditional methods that rely on artificial empirical formulas. At the same time, the construction of the input information matrix integrates spatial position, flow field state, and motion boundary conditions, providing global and local feature fusion input data for the neural network to ensure that the model can simultaneously perceive the geometric characteristics of the floating body, the dynamic response of the fluid, and external constraints, thereby significantly improving the physical consistency and accuracy of flow field prediction and laying a data foundation for the accurate calculation of additional mass and damping coefficients.

[0069] S3: Based on the preset floating body flow field prediction model, two-dimensional flow field prediction is performed on the input information matrix respectively, and two-dimensional flow field data of each slice is output;

[0070] S4: Based on the graph attention network, the two-dimensional flow field data is fused and corrected in three dimensions to generate a redundant three-dimensional flow field matrix, and missing data on the surface is completed by interpolation.

[0071] It should be noted that the preset floating body flow field prediction model is a system composed of multiple neural network modules, which is used to predict the flow field data around the floating body. Two-dimensional flow field prediction is used to predict the flow field physical quantities (flow rate, pressure, etc.) of a single plane slice (such as the X-Z plane). The graph attention network (GAT) is a graph neural network that dynamically models spatial correlation through node attention weight, and is used to fuse multi-plane data. Three-dimensional fusion and correction refers to integrating multiple two-dimensional plane prediction data into a three-dimensional flow field and correcting data redundancy and contradictions. The redundant three-dimensional flow field matrix is a three-dimensional data matrix generated by splicing multiple two-dimensional flow field data, which contains the predicted values of the same spatial point on multiple planes. Interpolation completion refers to filling in missing flow field data points through mathematical methods such as cubic spline interpolation. The missing data on the surface refers to the missing flow field data points due to incomplete coverage of the floating body surface by two-dimensional slices (such as the non-existence of surface intersection points on some planes).

[0072] Specifically, the vertical plane slice data is processed using a module based on Transformer to capture longitudinal / lateral-vertical flow field characteristics (such as boundary layer separation and vortex structure); the horizontal plane slice data is processed using a module based on Transformer to analyze longitudinal-lateral flow field distribution (such as free surface fluctuation and lateral velocity gradient); each module outputs two-dimensional flow field data for the corresponding slice, including flow velocity components, pressure components, and liquid volume fractions.

[0073] Further, the two-dimensional flow field data of each plane slice is spliced according to three-dimensional coordinates to generate a redundant three-dimensional flow field matrix (the same space point can be covered by multiple planes, and there are multiple sets of predicted values). The redundant data is modeled by GAT, and the spatial correlation of different plane data (such as the symmetry of X-Z and Y-Z planes in heave motion) is identified through the attention weight between nodes (flow field sampling points), and the conflicting or abnormal predicted values are corrected. For the missing data of the intersection point of the body surface, based on the flow field data of the adjacent slices (such as the flow velocity distribution of the same vertical position of the Y-Z plane), the physical quantities of the missing points are calculated by using cubic spline interpolation, and the body surface no-slip condition (the flow velocity is consistent with the body surface motion velocity) is forced to be satisfied.

[0074] This step decomposes the complex three-dimensional problem into local plane modeling through two-dimensional flow field prediction, uses the self-attention mechanism of the Transformer to capture the local details (such as vortices) and global correlation (such as pressure propagation path) of the flow field, significantly reduces the demand for computing resources; through three-dimensional fusion of the graph attention network (GAT), dynamically weights the redundant data of multiple planes, solves the spatial fragmentation problem caused by two-dimensional slices, and improves the consistency of the three-dimensional flow field; interpolation completion combined with physical constraints (such as the no-slip condition) ensures the integrity and reasonableness of the data near the body surface, and avoids the non-physical oscillation introduced by traditional interpolation methods. Finally, the corrected three-dimensional flow field data has high resolution and physical authenticity, providing reliable input for subsequent hydrodynamic coefficient calculation, and overall improving the accuracy of ship and offshore platform six-degree-of-freedom motion prediction.

[0075] S5: Based on the corrected three-dimensional flow field data, the hydrodynamic pressure and viscous shear force along the body surface of the floating body are integrated, and the target added mass coefficient and target viscous damping coefficient are determined by combining the single mode forced oscillation motion equation.

[0076] It should be noted that the viscous shear force refers to the tangential force caused by the viscous effect of the fluid, which is related to the flow velocity gradient and the dynamic viscosity of the fluid. The floating body surface integration refers to the process of accumulating and calculating the forces of all discrete elements on the floating body surface. The single mode forced oscillation motion equation is a mathematical equation describing the forced vibration of the floating body in a certain degree of freedom, which includes inertia term (added mass effect) and damping term (viscous damping effect). The added mass coefficient represents the equivalent mass effect caused by the surrounding fluid when the floating body moves. The viscous damping coefficient represents the influence degree of the fluid viscosity on the energy dissipation of the floating body motion.

[0077] Specifically, based on the corrected three-dimensional flow field data, the hydrodynamic pressure, viscous shear force, body surface normal direction, and body surface position coordinates of each discrete element on the floating body surface are extracted. The hydrodynamic pressure is multiplied by the element area, projected along the normal direction, and accumulated to obtain the total normal force; the shear stress is multiplied by the element area, projected along the tangential direction, and accumulated to obtain the total tangential force; the position coordinates of the element are used to calculate the moment of the force on the floating body centroid and accumulate.

[0078] Further, assuming the body is in simple harmonic motion in a certain degree of freedom (such as roll), the total hydrodynamic force (or moment) is substituted into the motion equation. Identify the force or moment component proportional to the body acceleration, determine the added mass coefficient by phase matching method (waveform phase consistent with acceleration); identify the force or moment component proportional to the body velocity, determine the viscous damping coefficient by amplitude ratio method (amplitude ratio to velocity amplitude); repeat the above steps for surge, sway, heave, roll, pitch, yaw, generate added mass matrix and damping matrix, and handle the coupling effect between degrees of freedom (such as heave-induced roll response).

[0079] This step obtains the total hydrodynamic force and moment by directly integrating the flow field data, avoiding the error of traditional potential flow theory ignoring viscous effect or relying on empirical formula; combined with single modal motion equation to decompose inertia term and damping term, accurately correlate flow field physical quantity and body motion response by phase matching and amplitude ratio method, eliminate the subjectivity of traditional critical damping ratio estimation (such as 3%-8%); multi-degree of freedom matrix output covers six degrees of freedom motion and coupling effect, provides high-precision hydrodynamic parameters for complex motion scenarios of ships and offshore platforms, and overall improves the engineering practicability of body motion response prediction.

[0080] This embodiment, based on the three-dimensional geometric model of the floating body, its draft, and the frequency and amplitude of the single-mode forced oscillation motion, performs two-dimensional slicing of the floating body's outer shell in a three-dimensional Cartesian coordinate system, generating flow field sampling points on the XZ, YZ, and XY planes respectively. Based on the coordinates of the flow field sampling points, initial flow field data, discrete points of the floating body's geometric outer contour, boundary conditions, and floating body motion parameters, a high-dimensional feature vector is encoded using a learnable matrix, and an input information matrix for each planar slice is constructed. Based on a preset floating body flow field prediction model, two-dimensional flow field prediction is performed on the input information matrix, outputting two-dimensional flow field data for each slice. The two-dimensional flow field data is fused and corrected in three dimensions using a graph attention network to generate a redundant three-dimensional flow field matrix, and missing data on the surface is filled in by interpolation. Based on the corrected three-dimensional flow field data, the surface integral of the hydrodynamic pressure and viscous shear force along the floating body, combined with the single-mode forced oscillation motion equation, determines the target added mass coefficient and the target viscous damping coefficient. This embodiment decomposes the complex three-dimensional flow field into multi-plane high-resolution sampling through two-dimensional slicing of the three-dimensional geometric model, reducing computational complexity while improving the accuracy of key areas. It uses a learnable matrix encoding to dynamically extract high-dimensional physical correlations between flow field data, geometric contours, and motion parameters, avoiding the limitations of traditional empirical formulas. A pre-defined floating body flow field prediction model, based on multi-plane two-dimensional prediction and three-dimensional fusion correction using a graph attention network, solves data redundancy and missing data issues, ensuring spatial consistency of the flow field. Finally, by combining the surface integrals of hydrodynamic pressure and viscous shear force with the single-mode motion equation to directly correlate flow field data with the inertial-damping effect, it eliminates errors in potential flow theory assumptions, achieving high-precision calculation of the six-degree-of-freedom added mass and viscous damping coefficient, thus improving the accuracy of floating body hydrodynamic calculations.

[0081] Based on the first embodiment described above, a second embodiment of the method for predicting the additional mass coefficient and viscous damping coefficient of the multi-degree-of-freedom motion of a floating body, as proposed in this application, is presented. Please refer to... Figure 2 , Figure 2 This is a schematic diagram of a sub-process in the second embodiment of the method for predicting the additional mass coefficient and viscous damping coefficient of multi-degree-of-freedom motion of a floating body in this application.

[0082] like Figure 2 As shown, in this embodiment, before step S1, the following steps are also included:

[0083] S1a: Construct a first prediction module and a second prediction module based on the Transformer model, and a third prediction module based on the graph attention network. The first prediction module is used to predict the two-dimensional flow field in the XZ plane and the YZ plane, the second prediction module is used to predict the two-dimensional flow field in the XY plane, and the third prediction module is used to integrate multi-plane slice data and correct redundancy.

[0084] S1b: pre-training the first prediction module and the second prediction module based on a preset public two-dimensional computational fluid dynamics dataset;

[0085] S1c: optimizing the pre-trained first prediction module and the second prediction module based on a three-dimensional self-owned dataset generated by CFD, and embedding continuity equation, momentum conservation equation and material surface condition as physical constraints to obtain a target first prediction module and a target second prediction module;

[0086] S1d: fixing parameters of the target first prediction module and the target second prediction module, and training the third prediction module based on a preset three-dimensional self-owned dataset to obtain a target third prediction module;

[0087] S1e: integrating the target first prediction module, the target second prediction module and the target third prediction module to obtain the preset floating body flow field prediction model.

[0088] It should be noted that the Transformer model is a deep learning model based on self-attention mechanism, which is good at capturing long-range dependencies in sequence data and is used for two-dimensional flow field prediction tasks. The graph attention network (GAT) is a graph neural network that models spatial correlation by dynamically calculating attention weights between nodes and is used for three-dimensional data fusion and correction. The three-dimensional self-owned dataset generated by CFD refers to the three-dimensional flow field data around the floating body obtained by computational fluid dynamics (CFD) simulation, including flow velocity, pressure and other physical quantities. The continuity equation is a basic equation of fluid mechanics, which describes the law of conservation of mass and requires the mass change rate of any microelement in the flow field to be zero. The momentum conservation equation is a basic equation of fluid mechanics, which describes the balance relationship between the change of fluid momentum and external force (pressure, viscous force, etc.). The physical constraint refers to the constraint condition that forces the model to satisfy the physical law (such as conservation of mass and momentum) in neural network training.

[0089] Specifically, the first prediction module is based on the Transformer architecture, inputs X-Z and Y-Z plane slice flow field data, and predicts the two-dimensional flow field of the vertical plane (longitudinal / lateral-vertical). The second prediction module is also based on the Transformer architecture, inputs X-Y plane slice flow field data, and predicts the two-dimensional flow field of the horizontal plane (longitudinal-lateral); the third prediction module is based on the graph attention network (GAT), inputs multi-plane redundant three-dimensional flow field data, corrects data conflicts and fills in missing points. The first and second modules are pre-trained using public two-dimensional CFD datasets (such as cylinder flow, square cylinder flow, etc.), to learn basic flow field features (such as boundary layer separation, vortex generation).

[0090] Further, the three-dimensional self-owned data set generated by CFD is processed into two-dimensional data by planar slicing, and is input into the pre-trained first and second modules. The residual terms of the continuity equation and the momentum equation are embedded in the training loss function, forcing the model prediction result to satisfy the mass and momentum conservation. The material surface condition constraint (such as the material surface no-slip boundary) is added to ensure that the flow field on the surface of the floating body meets the physical law. Randomly shaped floating body geometries (such as sphere, ellipsoid combination) are automatically generated and input into the first and second modules for unlabeled training, relying only on physical constraints to optimize parameters. The parameters of the first and second modules are fixed, and the third module (GAT) is trained using three-dimensional self-owned data sets to learn the redundant data fusion rules and missing data interpolation logic. The optimized first and second modules and the trained third module are connected in series to form a complete floating body flow field prediction system.

[0091] This step realizes the step-by-step improvement of model capability through a phased training strategy. In the pre-training phase, the basic flow field rules are learned using public data. In the physical constraint optimization phase, the continuity equation and the momentum equation are embedded to ensure that the prediction result meets the principles of fluid mechanics. In the unsupervised training phase, the model generalization ability is enhanced by generating random geometries, avoiding over-reliance on CFD data. Finally, through modular integration (Transformer+GAT), two-dimensional prediction and three-dimensional correction are decoupled, reducing the training complexity. At the same time, the gradual embedding of physical constraints (from no constraint pre-training to strong constraint optimization) balances the data-driven and physical consistency requirements, enabling the model to capture complex flow field characteristics (such as turbulence and free surface fluctuations) while avoiding the overfitting risk of pure data-driven models, significantly improving the accuracy and robustness of floating body hydrodynamic coefficient prediction.

[0092] Based on the above first embodiment, in this embodiment, before step S1, it further includes:

[0093] S1A: Geometric verification is performed on the three-dimensional geometric model of the floating body to identify non-streamline regions and potential numerical calculation error points in the model;

[0094] S1B: Based on the verification result, adjust the discretization parameters of the outer shell of the floating body, including slice density, partition line spacing, and material surface discrete point distribution rules.

[0095] It should be noted that geometric verification refers to checking whether the model has geometric defects (such as self-intersection, non-closed surface) or areas that do not conform to physical laws (such as sharp corners, sudden curvature). Non-streamlined areas refer to areas where the surface curvature of the floating body is suddenly changed or the geometry is discontinuous (such as the bow bulb, platform leg connection), which can easily cause flow field separation or vortex generation. Numerical calculation error points refer to local areas that may cause calculation errors in subsequent flow field simulation due to insufficient geometric discretization (such as grid distortion, low resolution). Discretization parameters refer to the rules for converting continuous geometric models into discrete data (such as grids, sampling points), including slice density, segmentation line spacing, and surface discrete point distribution method.

[0096] Specifically, use geometric analysis tools (such as CAD software or special plug-ins) to detect whether the model is closed (no gap), the surface is smooth (no self-intersection), and the topological structure is reasonable (such as no cracks at the connection between the leg and the main body). Mark the areas with sudden curvature changes (such as a curvature radius less than 10% of the ship length) through curvature analysis algorithms (such as Gaussian curvature calculation), which may cause flow field separation or turbulence intensification. Combine historical CFD simulation experience data to identify areas (such as the stern transition zone) that are prone to numerical errors due to grid distortion (such as narrow triangular grids) or insufficient resolution (such as stepped discrete profiles).

[0097] Further, increase the number of slices (such as 50% density increase) in non-streamlined areas (such as the bow bulb) to ensure that the flow field sampling points cover local details; reduce the slice density in flat areas (such as the middle of the ship body) to save computing resources. Reduce the segmentation line spacing (such as 30% of the default spacing) for areas with significant curvature changes (such as the root of the platform leg) to generate denser transverse / vertical segmentation lines. Distribute discrete points along the floating body surface profile line according to curvature weight, reduce the point distance in areas with large curvature (such as 5% of the curvature radius), and enlarge the point distance in areas with small curvature to avoid the "sawtooth effect" caused by surface discretization.

[0098] This step excludes model defects (such as self-intersection, non-closed surface) in advance through geometric verification, avoiding flow field prediction failure caused by geometric errors; identifies non-streamlined areas and dynamically adjusts discretization parameters to finely discretize (high slice density, dense segmentation lines) in key areas where the flow field changes dramatically (such as the bow, leg connection), capture complex flow characteristics (such as vortex, boundary layer separation), and reduce redundant calculations in flat areas; distribute surface discrete points according to curvature weight to ensure the smoothness of the discretized surface, avoiding numerical oscillation (such as pressure sudden change) caused by insufficient discretization. Finally, the optimized discretization parameters provide high-fidelity input data for subsequent flow field sampling point generation and neural network prediction, improving the accuracy and stability of the added mass and viscous damping coefficient calculation from the source.

[0099] The embodiment is based on the three-dimensional geometric model of the floating body, the draft, and the frequency and amplitude of the single-mode forced oscillation motion of the floating body. The outer shell of the floating body is subjected to two-dimensional slicing processing in the three-dimensional rectangular coordinate system, and flow field sampling points are generated on the X-Z plane, the Y-Z plane, and the X-Y plane. Based on the coordinates of the flow field sampling points, the initial flow field data, the discrete points of the geometric outer contour of the floating body, the boundary conditions, and the motion parameters of the floating body, the learnable matrix is encoded into a high-dimensional feature vector, and an input information matrix of each plane slice is constructed. Based on a preset floating body flow field prediction model, two-dimensional flow field prediction is performed on the input information matrix, and two-dimensional flow field data of each slice is output. Based on the graph attention network, the two-dimensional flow field data is subjected to three-dimensional fusion and correction, a redundant three-dimensional flow field matrix is generated, and the surface missing data is completed by interpolation. Based on the corrected three-dimensional flow field data, the hydrodynamic pressure and viscous shear force along the surface of the floating body are integrated, and the target added mass coefficient and the target viscous damping coefficient are determined by combining the single-mode forced oscillation motion equation. The embodiment decomposes the complex three-dimensional flow field into multiple plane high-resolution samples through two-dimensional slicing processing of the three-dimensional geometric model, reduces the calculation complexity, and improves the accuracy of the key area. The learnable matrix dynamically extracts the high-dimensional physical correlation of the flow field data, the geometric contour, and the motion parameters, avoiding the limitations of traditional empirical formulas. The preset floating body flow field prediction model is based on multi-plane two-dimensional prediction and three-dimensional fusion and correction of the graph attention network, solves the problems of data redundancy and missing, ensures the spatial consistency of the flow field, and finally integrates the hydrodynamic pressure and viscous shear force along the surface of the floating body by combining the single-mode motion equation to directly associate the flow field data and the inertia-damping effect, eliminate the assumption error of the potential flow theory, realize the high-precision calculation of the six-degree-of-freedom added mass and viscous damping coefficient, and improve the accuracy of the hydrodynamic calculation of the floating body.

[0100] Based on the second embodiment, a third embodiment of the method for predicting the added mass coefficient and the viscous damping coefficient of the floating body in multiple degrees of freedom motion is provided. Please refer to Figure 3 , Figure 3 The third embodiment of the method for predicting the added mass coefficient and the viscous damping coefficient of the floating body in multiple degrees of freedom motion is a sub-flowchart.

[0101] In the embodiment, after step S4, the following steps are further included:

[0102] S4a: based on the corrected three-dimensional flow field data, dynamically evaluating the confidence of the flow field prediction;

[0103] S4b: if the confidence is lower than the preset threshold, readjusting the motion parameters or the boundary conditions of the floating body, generating a new input information matrix, and iteratively performing the two-dimensional flow field prediction and the three-dimensional fusion steps;

[0104] S4c: if the confidence reaches the threshold, taking the current three-dimensional flow field data as the target three-dimensional flow field data.

[0105] It should be noted that the confidence is a quantitative evaluation index of the reliability of the predicted flow field data, which is usually calculated based on the error of the predicted value and the benchmark data (such as experimental or high-precision CFD results). The preset threshold is a pre-set confidence qualification standard (such as error rate <5%), which is used to determine whether iteration optimization is required. The floating body motion parameters refer to the degree of freedom direction, frequency and amplitude of the single modal forced oscillation motion of the floating body.

[0106] Specifically, the modified three-dimensional flow field data is compared with the high-precision CFD simulation results or experimental data, and the relative error of the key physical quantities (such as the peak pressure of the bow and the intensity of the stern vortex) is calculated. By integrating the flow velocity error, pressure distribution consistency, vortex position deviation and other indicators, and assigning weights according to the engineering importance (such as pressure weight 50%, flow velocity 30%, and vortex 20%), a comprehensive confidence score is generated. If the confidence score is lower than the preset threshold (such as <90%), it is determined that the prediction result is unreliable, triggering the parameter adjustment process.

[0107] Further, the error distribution is used to locate the source of the problem. If the stern vortex position deviation is large, the oscillation frequency or amplitude is adjusted; if the inlet flow velocity distribution is abnormal, the boundary type is modified (such as changing the symmetry plane to a pressure outlet). Based on the adjusted parameters, the flow field sampling points, geometric profiles, motion parameters and other data are re-encoded to construct a new input information matrix. The new input matrix is input into the floating body flow field prediction model, and the two-dimensional flow field prediction, three-dimensional fusion and modification steps are re-executed until the confidence meets the standard.

[0108] This step monitors the reliability of the prediction result in real time through dynamic confidence evaluation, avoids systematic errors caused by improper initial parameter setting (such as high frequency leading to flow field separation not being captured) or boundary condition errors (such as wall condition being mistakenly set as symmetry plane); through parameter sensitivity analysis and iteration optimization, the motion parameters or boundary conditions are adjusted adaptively to solve the local flow field prediction failure problem (such as bow pressure peak deviation and stern vortex position deviation) and reduce the cost of manual trial and error; finally, the closed-loop feedback mechanism ensures that the prediction result always meets the engineering precision requirement (such as error <5%), significantly improving the robustness and practicality of ship and offshore platform hydrodynamic coefficient prediction.

[0109] Based on the above-mentioned second embodiment, in the present embodiment, step S1 comprises:

[0110] S11: based on the three-dimensional geometric model of the floating body, the draft and the frequency and amplitude of the single modal forced oscillation motion, the outer shell of the floating body is regionally divided to obtain regular regions and asymmetric regions;

[0111] S12: performing two-dimensional slicing processing on the asymmetric region in a three-dimensional rectangular coordinate system based on an asymmetric division strategy to generate asymmetric sampling points, the asymmetric division strategy including increasing slicing density at a region where geometric curvature changes greater than a preset curvature change threshold and adopting a non-uniform division line spacing in an X-Y plane;

[0112] S13: performing the two-dimensional slicing processing on the regular region in the three-dimensional rectangular coordinate system based on a regular division strategy to generate regular sampling points, the regular division strategy including adopting an equal spacing and preferentially retaining plane data orthogonal to a direction of a degree of freedom of motion of the floating body;

[0113] S14: taking the asymmetric sampling points and the regular sampling points as the flow field sampling points.

[0114] It should be noted that the regular region refers to a region (such as a middle part of a ship body or a platform deck) where a geometric shape of a surface of a floating body is smooth, a curvature change is small, and symmetry is strong. The asymmetric region refers to a region (such as a bulbous bow or a platform leg connection) where a curvature of a surface of a floating body is abruptly changed, a geometric shape is complex, or symmetry is not strong. The asymmetric division strategy refers to a dynamic adjustment rule adopted for the asymmetric region, including locally increasing slicing and non-uniform distribution of division lines. The preset curvature change threshold is a standard for determining that a curvature change is significant (such as a curvature radius being less than 10% of a ship length), and is used to trigger slicing density adjustment. The plane data orthogonal to the direction of the degree of freedom of motion of the floating body refers to a plane (such as an X-Y plane) perpendicular to a main direction of motion (such as Z axis for heave) of the floating body, and such plane data is preferentially retained to capture key flow field characteristics.

[0115] Specifically, based on curvature analysis and symmetry detection of a three-dimensional geometric model of a floating body, a surface is divided into a regular region (such as a flat section of a ship body) and an asymmetric region (such as a bow or a leg root). The regular region determination standard is that a curvature change is continuous and has a small amplitude (such as a curvature radius being greater than 20% of a ship length), and geometric symmetry is high. The asymmetric region determination standard is that a curvature is abruptly changed or geometric symmetry is not strong (such as a curvature radius being less than 10% of a ship length). The asymmetric division strategy is adopted for the asymmetric region: slicing density is increased at a region where a curvature change is significant (such as 2 times of a default density), and a non-uniform division line spacing is adopted in an X-Y plane (such as a spacing being reduced by 50% in a region where a curvature is large). The regular division strategy is adopted for the regular region: slicing is equally spaced, division lines are uniformly arranged, and plane data orthogonal to a main direction of a degree of freedom of motion of the floating body (such as X-Y plane data when heave motion is performed) is preferentially retained.

[0116] Further, in the curvature mutation area, X-Z / Y-Z plane slices are densely cut to generate high-density sampling points. In the X-Y plane, non-uniform division lines are used (for example, 0.5 meters apart near the bow, 2 meters apart in the middle of the ship), to capture the transverse flow field gradient change. Equal-interval slices are used (for example, every 2 meters of the whole ship body), and the division lines are uniformly distributed; preferentially retaining the planes orthogonal to the direction of the motion degree of freedom (for example, retaining Y-Z plane data when the surge motion is performed), to ensure the flow field coverage in the key motion direction. The dense sampling points in the asymmetric area are integrated with the uniform sampling points in the regular area to form a complete flow field sampling point set.

[0117] This step uses dynamic area division and strategy adaptation. In the asymmetric area (for example, the bow and the leg connection), high-density slices and non-uniform division are used to accurately capture complex flow field characteristics (for example, vortex generation and boundary layer separation), to avoid the insufficient accuracy of traditional uniform sampling in the key area; in the regular area (for example, the middle of the ship body), equal-interval slices are used and orthogonal plane data are preferentially retained, to reduce redundant calculation while ensuring the integrity of the flow field information in the main motion direction. Through the differentiated sampling strategy, a balance is achieved between improving the global calculation efficiency and the local accuracy, to provide high-resolution input data for subsequent neural network prediction, and finally to significantly improve the prediction accuracy of the added mass and the viscous damping coefficient.

[0118] Based on the second embodiment described above, in this embodiment, step S5 includes:

[0119] S51: Extracting the hydrodynamic pressure, viscous shear force, surface normal direction, and position coordinates of the discrete elements on the surface of the floating body in the modified three-dimensional flow field data;

[0120] S52: Integrating the hydrodynamic pressure and viscous shear force along the surface of the floating body, and numerically integrating the hydrodynamic pressure and the viscous shear force of the discrete elements, to obtain the total hydrodynamic force and the total hydrodynamic moment acting on the floating body;

[0121] S53: According to the single-mode forced oscillation motion equation, decomposing the total hydrodynamic force and the total hydrodynamic moment into an inertia term related to acceleration and a damping term related to velocity;

[0122] S54: Based on the inertia term and the damping term, using the phase matching and amplitude scaling algorithm to obtain the target added mass coefficient and the target viscous damping coefficient.

[0123] It should be noted that the normal direction of the object surface refers to the normal vector of the discrete element on the surface of the floating body, indicating the direction of the force (perpendicular or tangential). Position coordinates refer to the three-dimensional position of the discrete element relative to the center of mass of the floating body, used for torque calculation. Numerical integration refers to the process of summing the forces or torques of the discrete elements. The inertial term refers to the force or torque component proportional to the acceleration of the floating body, characterizing the added mass effect. The damping term refers to the force or torque component proportional to the velocity of the floating body, characterizing the viscous damping effect. Phase matching is a method of separating the inertial and damping terms by analyzing the phase difference between the force / torque waveform and the motion parameters. The amplitude scaling algorithm is a method of calculating coefficients by the ratio of the force / torque amplitude to the amplitude of the motion parameters.

[0124] Specifically, the total hydrodynamic pressure is obtained by multiplying the hydrodynamic pressure of each unit by its area, projecting it along the normal direction, and summing the normal force components of all units. The total viscous shear force is obtained by multiplying the shear stress of each unit by its area, projecting it along the tangential direction, and summing the tangential force components of all units. The total hydrodynamic pressure and the total viscous shear force are then vectorively added together to obtain the total hydrodynamic force on the buoy. Using the position coordinates of each unit, the torque of the force on the buoy about its center of mass is calculated and summed to obtain the total hydrodynamic torque.

[0125] Furthermore, the waveform of the total hydrodynamic force (or torque) changes over time is analyzed to identify the component in phase with the buoy's acceleration (inertial term) and the component in phase with the velocity (damping term). The added mass coefficient is calculated by the ratio of the amplitude of the inertial term to the amplitude of the acceleration, and the viscous damping coefficient is calculated by the ratio of the amplitude of the damping term to the amplitude of the velocity. The above steps are performed on the six degrees of freedom of sway, roll, heave, pitch, yaw, and bow respectively to generate the complete added mass matrix and viscous damping matrix.

[0126] This step obtains the total hydrodynamic force and torque by directly integrating the flow field data, eliminating the errors of traditional potential flow theory that ignores viscous effects or relies on empirical formulas (such as critical damping ratio estimation). It accurately separates the inertial and damping terms through phase matching and amplitude scaling algorithms, avoiding coefficient deviations caused by artificial assumptions. The six-degree-of-freedom extension covers complex floating body motion scenarios (such as heave-roll coupling), outputting multi-degree-of-freedom coupling effects in matrix form, providing high-precision hydrodynamic parameters for predicting the motion response of ships and offshore platforms. Ultimately, the physics-driven direct calculation significantly improves the accuracy of added mass and viscous damping coefficients, reducing errors by 30%-50% compared to traditional methods, meeting the high-precision requirements of engineering.

[0127] Please see Figures 4-6 , Figure 4 This is a schematic diagram of a three-dimensional fluid domain in one embodiment of the method for predicting the additional mass coefficient and viscous damping coefficient of multi-degree-of-freedom motion of a floating body according to this application. Figure 5This is a schematic diagram of a slice in one embodiment of the method for predicting the additional mass coefficient and viscous damping coefficient of multi-degree-of-freedom motion of a floating body according to this application; Figure 6 This is a schematic diagram of the flow field sampling points in one embodiment of the method for predicting the additional mass coefficient and viscous damping coefficient of multi-degree-of-freedom motion of a floating body according to this application. In one embodiment, the specific workflow of the preset floating body flow field prediction model is as follows:

[0128] Step 1: First, determine the floating body whose viscous damping coefficient needs to be calculated and its draft h. Establish a three-dimensional rectangular coordinate system with the center of the waterline of the floating body as the origin. Next, determine the water depth H of the fluid domain for which the flow field needs to be predicted, as well as the length a along the x-direction, the width b along the y-direction, and the height c along the z-direction of the fluid domain. Subsequently, determine the flow field sampling point P. Perform two-dimensional slicing on the three-dimensional geometric model of the floating body's outer shell, including the XZ plane, YZ plane, and XY plane. Each slice is divided by n horizontal dividing lines and n vertical dividing lines, with the horizontal dividing lines orthogonal to the vertical dividing lines. From this, n 2 The intersection points are the flow field sampling points P. Each planar direction is divided by n slices, and all the dividing lines of the slices in the three planar directions are either coincident, parallel, or orthogonal. The areas where the dividing lines intersect with the floating body are densely and uniformly distributed, and the distribution becomes sparser as the dividing lines are farther away from the floating body.

[0129] Furthermore, the points where the dividing line intersects the outer contour of the float are points on the surface and need to be added to the flow field sampling points P; the points where the dividing line intersects the boundary of the fluid domain are boundary points and also need to be added to the flow field sampling points P; fluid sampling points located inside the float body should be deleted; points located above the free liquid surface at a height greater than a certain level also need to be deleted, which can be taken as 1 / 5 to 1 / 10 of the height of the remaining part of the float above the water surface. Thus, each slice in each planar direction has a set of flow field sampling points P with defined coordinates.

[0130] The flow field sampling point P includes sampling points on three planes, namely sampling point P1[x] located on the XZ plane. im ,z im The sampling point P2[y] is located on the YZ plane. jn ,z jn ] and sampling point P3[x] located on the XY plane kl ,y kl Where x, y, and z represent the x, y, and z coordinates in the coordinate system, respectively; m represents the m-th slice in the XZ plane direction, and i represents the i-th sampling point on that slice; n represents the n-th slice in the YZ plane direction, and j represents the j-th sampling point on that slice; l represents the l-th slice in the XY plane direction, and k represents the k-th sampling point on that slice.

[0131] In a three-dimensional flow field, the flow field information corresponding to a specific position coordinate [x,y,z] is [u,v,w,p]. x ,p y ,p z [,α], where u is the flow velocity in the x-direction, v is the flow velocity in the y-direction, w is the flow velocity in the z-direction, and p x The pressure in the x-direction, p y For pressure in the y direction, p z Let be the pressure in the z-direction, and α be the liquid volume fraction, representing the proportion of the liquid to the total volume of the (gas and liquid phases) space. Therefore, on the XZ plane, sampling point P1[x im ,z im The corresponding flow field information is: On the YZ plane, sampling point P2[y jn ,z jn The corresponding flow field information is: On the XY plane, sampling point P3[x kl ,y kl The corresponding flow field information is:

[0132] Step 2: Based on the draft h from Step 1, generate the corresponding initial flow field data under still water conditions for each flow field sampling point P, i.e., u = v = w = 0. Below the free surface, α = 1; above the free surface, α = 0, where ρ is the fluid density, g is the gravitational acceleration, z is the z-axis coordinate, and α is the liquid volume fraction. (When training the neural network, there are three training stages. The training datasets for stages one and two provide initial flow field data that can be directly assigned; while the initial flow field information for stage three is generated using the method described above.) Multiplying the two-dimensional coordinates of the flow field sampling points on the slice by a learnable matrix yields a 768-dimensional feature vector. (This learnable matrix is ​​continuously optimized during the training of the neural network to obtain a final, determined matrix.) For example, if the coordinates of a sampling point on a slice in the XZ plane are [2, 1], multiplying it by a 2×768-dimensional learnable matrix yields the corresponding feature vector representing the coordinates of the sampling point. Then, multiplying the flow field data of this sampling point by a learnable matrix yields a 768-dimensional feature vector. For example, the flow field data of a sampling point on a slice in the XZ plane [u, v, p]... x ,p zIf α is [4, 8, 50, 70, 1], then multiplying it with a learnable matrix of dimension 5×768 yields the corresponding feature vector representing the initial flow field data at that sampling point. Subsequently, the time (in seconds) corresponding to the flow field data to be predicted is multiplied by a learnable matrix to obtain a 768-dimensional feature vector. For example, if the time of the flow field data to be predicted is 10 seconds, multiplying time 10 by a learnable matrix of dimension 1×768 yields the corresponding feature vector representing the initial flow field data at that sampling point. (In practical applications, the time after a certain period, i.e., when the floating body returns to its original position during oscillation, i.e., when the floating body experiences the maximum force or torque, is usually taken as the time to be predicted.)

[0133] Adding the three feature vectors representing the coordinates of the sampling point, the initial flow field data, and the time corresponding to the flow field to be predicted, respectively, yields the initial flow field information feature vector containing coordinate and prediction time information. Combining the initial flow field information feature vectors of all sampling points on a slice into an m×768 matrix results in the initial flow field information matrix U. Here, each row represents the initial flow field information feature vector of a sampling point, and m represents the m sampling points on the slice.

[0134] Step 3: Based on the slicing method in Step 1, the two-dimensional geometric outline of the floating body on each slice can be obtained, and it can be discretized into N points to represent the surface geometric information L. Similar to the flow field sampling point P, the surface geometric information L also includes three types, namely discrete points L1[x] located on the XZ plane. im ,z im ], a discrete point L2[y] located on the YZ plane jn ,z jn ] and discrete point L3[x] located on the XY plane kl ,y klIn this diagram, x, y, and z represent the x, y, and z coordinates in the coordinate system, respectively; m represents the m-th slice in the XZ plane direction, and i represents the i-th discrete point of the outer contour of the floating body geometry on that slice; n represents the n-th slice in the YZ plane direction, and j represents the j-th discrete point of the outer contour of the floating body geometry on that slice; l represents the l-th slice in the XY plane direction, and k represents the k-th discrete point of the outer contour of the floating body geometry on that slice. Multiplying the two-dimensional coordinates of the discrete points of the outer contour of the floating body geometry by a learnable matrix yields a 768-dimensional feature vector. For example, if the coordinates of a discrete point of the outer contour on a slice are [6, 7], multiplying it by a learnable matrix of dimension 2×768 yields the corresponding feature vector representing the coordinate position of that discrete point. Taking the point with the smallest sum of absolute coordinate values ​​among a group of consecutive discrete points of the outer contour of the floating body geometry on a slice as the starting point, these discrete points are sorted counterclockwise from 0, starting at 0, 1, 2, ... Multiplying the indices of discrete points on the outer contour of the floating body's geometry by a learnable matrix yields a 768-dimensional feature vector. For example, if the indices of a discrete point on the outer contour of a floating body are 3, multiplying it by a 1×768 learnable matrix gives the corresponding feature vector representing the indices of that discrete point. Adding these two feature vectors, representing the coordinates and indices of the discrete points respectively, yields the surface geometry information feature vector containing both the coordinates and indices of the discrete points on the outer contour of the floating body. Combining the surface geometry information feature vectors of all discrete points on a slice into an s×768 matrix results in the surface geometry information matrix L. Each row represents the surface geometry information feature vector of a discrete point, and s represents the number of discrete points on the outer contour of the floating body on that slice.

[0135] Step 4: Encode the boundary conditions and embed them into the feature vector. In computational fluid dynamics, the boundary conditions of an incompressible fluid domain include ① pressure inlet, ② pressure outlet, ③ velocity inlet, ④ free outlet, ⑤ wall, and ⑥ symmetry plane. One-hot encoding is used to encode these six types: "100000" represents the first type, "010000" represents the second type, and so on. Four-bit binary one-hot encoding is used to encode the location of the boundary conditions in the two-dimensional fluid domain, representing the left, right, top, and bottom boundaries. For example, "1000" represents the left boundary, "0100" represents the right boundary, and so on. The boundary condition locations and their corresponding types are encoded according to the system input and then concatenated into a ten-digit code. For example, the left boundary, being a velocity inlet, is encoded as "1000100000". This code is multiplied by a 10×786 learnable matrix to obtain a 768-dimensional feature vector. Finally, the boundary condition values ​​are multiplied by a learnable matrix to obtain a 768-dimensional feature vector. For example, if "the velocity at the inlet on the left boundary is 1 m / s", then multiplying this value by a learnable matrix of dimension 1×768 will yield the corresponding feature vector representing the inlet velocity condition.

[0136] Adding the two eigenvectors above yields an eigenvector that incorporates all information about the boundary conditions. Combining the eigenvectors of the four boundaries results in a 4×768 boundary condition information matrix B. Each two-dimensional fluid domain on a slice has a corresponding boundary condition information matrix B.

[0137] Step 5: Embed the floating body motion data into the feature vector, assuming the floating body performs a single-mode forced oscillation motion. The floating body motion data includes the direction, frequency, and amplitude of the single-degree-of-freedom vibration motion, where the frequency and amplitude can be represented as [ω, ξ1, ξ2]. Here, ω is the angular frequency (rad / s), ξ1 is the angular amplitude of the rotational motion (°), and ξ2 is the distance amplitude of the translational motion (m). The floating body motion has six degrees of freedom: sway (translation along the x-axis), sway (translation along the y-axis), heave (translation along the z-axis), roll (rotation about the x-axis), pitch (rotation about the y-axis), and yaw (rotation about the z-axis). However, combined with the frequency and amplitude information mentioned above, only the axial direction of the floating body motion needs to be described to express the motion directions of the six degrees of freedom. The axial direction can be represented using a unit vector, i.e., "100" represents the x-axis direction, "010" represents the y-axis direction, and "001" represents the z-axis direction. For example, the axis direction is encoded as [1, 0, 0], and the frequency and amplitude are encoded as [0.5, 0, 10]. Combined, these represent a buoyant body undergoing a swaying motion along the x-axis with a frequency of 0.5 rad / s and an amplitude of 10 m. Concatenating the axis direction encoding with the frequency and amplitude information vectors yields a six-digit code. Multiplying this six-digit code by a learnable matrix results in a 768-dimensional feature vector, which is the buoyant body motion information matrix M. (In practical applications, for example, a set of motion frequency information data can be created at 0.1 rad / s intervals, from 0 to 1.5 rad / s, along with fixed motion amplitude data of 10 degrees or 10 meters. These are then cyclically input into the system to obtain a set of values ​​for the additional mass and viscous damping that vary with the motion frequency, facilitating table lookup. When training a neural network, if flow field data of a stationary object is used, the buoyant body motion information is [0, 0, 0, 0, 0, 0].)

[0138] Step 6: Embed the consideration of gravity's influence into the feature vector. Since gravity acts vertically downwards along the z-axis, for two-dimensional flow fields in the XZ and YZ planes, gravity causes water to accumulate at the bottom of the two-dimensional fluid domain. However, for the two-dimensional flow field in the XY plane, gravity does not affect the distribution of water within the fluid domain. To distinguish between these two cases, a two-digit "one-heat encoding" is used, where "10" indicates that gravity's influence is considered, and "01" indicates that gravity's influence is not considered. Multiplying the two-digit encoding by a learnable matrix yields a 768-dimensional feature vector, which is the gravity information matrix G.

[0139] Step 7: Sequentially concatenate and combine the initial flow field information matrix U, the object surface geometry information matrix L, the boundary condition information matrix B, the floating body motion information matrix M, and the gravity information matrix G obtained in the above steps into an input information matrix X, with a dimension of (m+s+4+1+1)×768, where m is the number of rows in U, s is the number of rows in L, 4 is the number of rows in B, 1 is the number of rows in M, and 1 is the number of rows in G.

[0140] Step 8: Based on the description in Step 7, construct the input information matrices X1, X2, and X3 for the XZ, YZ, and XY planes, respectively. The main difference between the three is that when slicing on different planes, the coordinates of the flow field sampling points, the geometric information of the object surface, the position of the boundary conditions, etc., involved in the obtained input information matrix X are all different and need to be modified accordingly.

[0141] First, input the input information matrix X1 into module 1 (the two-dimensional flow field prediction neural network module in the XZ and YZ planes). Module 1 will then output its predicted flow field data matrix Y1[u,w,p]. x ,p z Then, input the information matrix X2 into module 1, and module 1 will output its predicted flow field data matrix Y2[v,w,p]. y ,p z Finally, the input information matrix X3 is input into module 2 (XY plane two-dimensional flow field prediction neural network module), and module 2 will output its predicted flow field data matrix Y3[u,v,p]. x ,p y ,α].

[0142] It is important to note that the input information matrix X to the neural network module can only be the input information matrix for one slice at a time, not the information matrix for all slices, because the neural network module predicts the two-dimensional flow field. Therefore, by sequentially inputting the output information matrices of each slice into the neural network, the two-dimensional predicted flow field data for each slice, i.e., the predicted flow field data matrix Y, can be obtained sequentially.

[0143] Step 9: Combine and stitch together the two-dimensional predicted flow field data matrix Y from all slices according to the three-dimensional coordinate positions of the flow field sampling points to form a three-dimensional predicted flow field data matrix Y. 3D For example, for a flow field sampling point P with coordinates [1, 4, 7], it includes flow field sampling points P1[1, 7] on a slice of the XZ plane, flow field sampling points P2[4, 7] on a slice of the YZ plane, and sampling point P3[1, 4] on a slice of the XY plane. The two-dimensional predicted flow field data corresponding to these three flow field sampling points are [u1, w1, p x1 ,p z1 ,α1],[v2,w2,py2 ,p z2 ,α2],[u3,v3,p x3 ,p y3 By stitching together the two-dimensional predicted flow field data corresponding to the sampling points on the corresponding slices of these three planes, a complete three-dimensional predicted flow field data Y with data redundancy can be obtained. 3D [u1,u3,v2,v3,w1,w2,p x1 ,p x3 ,p y2 ,p y3 ,p z1 ,p z2 [α1, α2, α3], where u is the flow velocity in the x-direction, v is the flow velocity in the y-direction, w is the flow velocity in the z-direction, and p x The pressure in the x-direction, p y For pressure in the y direction, p z Let α be the pressure in the z-direction and α be the liquid volume fraction. (As can be seen, the three-dimensional predicted flow field data Y...) 3D Each physical quantity in the data contains two flow field values ​​from slices on different planes. These are two-dimensional flow field predictions generated by a planar two-dimensional flow field prediction neural network module based on input information from a two-dimensional fluid domain, without considering the third dimension. Therefore, it is necessary to fuse and correct the two-dimensional flow field prediction data from the XZ, YZ, and XY planes to obtain the truly corrected three-dimensional flow field data. This will be performed in the next step.

[0144] In addition, there are special cases that need to be handled. As described in step 1, the points where the dividing lines intersect with the outer contour of the floating body, i.e., the flow field sampling points on the object surface, are not obtained by the orthogonality of the horizontal and vertical dividing lines. Therefore, these points usually only exist on a slice in one planar direction, and there are no slices with this flow field sampling point on the object surface in the other two planar directions. Therefore, for all flow field sampling points located on the object surface, the three-dimensional predicted flow field data Y obtained after stitching them together according to the above method is... 3D There are missing data points, requiring cubic spline interpolation using flow field sampling points on surrounding surfaces to fill in the gaps. For example, a flow field sampling point p1[x1,z1] on a surface of a slice in the XZ plane has three-dimensional coordinates [x1,y1,z1], but no corresponding three-dimensional flow field sampling point exists. That is, for coordinates [y1,z1], there is no flow field sampling point on the surface of a slice in the YZ plane with matching coordinates; similarly, for coordinates [x1,y1], there is no flow field sampling point on the surface of a slice in the XY plane with matching coordinates. Therefore, the three-dimensional predicted flow field data Y corresponding to coordinates [x1,y1,z1] obtained after stitching using the above method is...3D [u1,0,0,0,w1,0,p] x1 ,0,0,0,p z1 [,0,α1,0,0]. And based on the flow field data of all slices of the YZ plane [v,w,p]... y ,p z From [α], we can obtain the velocity field v, velocity field w, and velocity field p on the object surface. y Pressure field, p z The pressure field and the α-liquid volume fraction field are used as the basis for calculating the corresponding physical quantities at coordinates [x1, y1, z1] using cubic spline interpolation, thus supplementing the three-dimensional predicted flow field data Y. 3D Similarly, the flow field data [u,v,p] for all slices of the XY plane... x ,p y By performing the same operation on α, the three-dimensional predicted flow field data Y can be completely completed. 3D [u1,u3,v2,v3,w1,w2,p x1 ,p x3 ,p y2 ,p y3 ,p z1 ,p z2 ,α1,α2,α3].

[0145] Step 10: Transfer the three-dimensional predicted flow field data Y 3D Concatenated with the corresponding three-dimensional coordinates, they form a flow field feature vector [x,y,z,u1,u3,v2,v3,w1,w2,p] with a dimension of 18. x1 ,p x3 ,p y2 ,p y3 ,p z1 ,p z2 [α1, α2, α3]. The flow field feature vectors of all flow field sampling points are combined to form a three-dimensional input flow field matrix X with dimension t×18. 3D , where t is the number of flow field sampling points P. The three-dimensional input flow field matrix X... 3D Input module 3 (a neural network module for fusing two-dimensional flow fields into three-dimensional flow fields). Module 3 will output the final three-dimensional flow field data U after fusing its predicted redundant data. 3D [x,y,z,u,v,w,p x ,p y ,p z [α]. This data represents the three-dimensional flow field around the floating body at the moment when it experiences the maximum net external force and torque during a single-mode forced oscillation motion.

[0146] The specific steps for calculating the maximum net external force and torque, as well as the additional mass coefficient and viscous damping coefficient, of a floating body during a single-mode forced oscillation motion are as follows:

[0147] Based on the flow field data around the floating body predicted by the system, more refined flow field data around the floating body can be obtained through cubic spline interpolation. The pressure on the predicted discrete surface element is p, where the hydrostatic pressure is... The hydrodynamic pressure is p d =pp s .

[0148] The predicted viscous shear force on the discrete surface element is Where u represents the three-dimensional fluid velocity vector, including x, y, and z components; μ is the fluid dynamic viscosity coefficient; and n is the normal direction of the object surface, including x, y, and z components.

[0149] Integrating the hydrodynamic pressure and viscous shear force along the surface S of the floating body, the components of the total hydrodynamic force acting on the floating body along the x, y, and z directions are F. x =∫(p d +τ)n x ds, F y =∫(p d +τ)n y ds, F z =∫(p d +τ)n z The components of the total hydrodynamic torque acting on the floating body along the x, y, and z directions are respectively M x =∫(p d +τ)·(r y n y -r z n z )ds,M y =∫(p d +τ)·(r x n x -r z n z )ds,M z =∫(p d +τ)·(r x n x -r y n y )ds. Where, n x n y and n z These are the direction vectors of the unit normal to the object's surface along the x, y, and z directions, pointing towards the inside of the buoyancy body; r x r y and r zThese represent the x, y, and z directions of the hydrodynamic center and rotation center of the surface element, respectively.

[0150] The equation of motion for a floating body undergoing single-mode forced oscillation is ξ = ξ0sinωt, where ξ, ξ0, and ω are the amplitude, maximum value, and frequency of the oscillation, respectively. For roll (or pitch or yaw) motion, the hydrodynamic torque on the floating body calculated from the flow field data is fitted using the following formula: M d = M0sin(ωt+ε), where M0 is the amplitude of the total hydrodynamic torque on the floating body under forced excitation, ε is the phase angle, and ω is the frequency, which is the same as the frequency of the floating body's motion. Hydrodynamic torque M d It can be decomposed into inertial terms and damping terms, that is Where A is the additional mass coefficient for roll (or pitch or yaw), and B is the damping coefficient for roll (or pitch or yaw). The acceleration due to the buoy's roll (or pitch or bow) motion. Let be the velocity of the floating body during roll (or pitch or yaw). Then the additional mass coefficient for roll (or pitch or yaw) is A = M0cosε / ξ0ω. 2 The roll (or pitch or yaw) damping coefficient B = -M0sinε / ξ0ω, which can be calculated using these two formulas. (Roll, pitch, and yaw correspond to rotational motion along the x-axis, y-axis, and z-axis, respectively, and their hydrodynamic torques correspond to M...) x M y M z 。 )

[0151] Similarly, for swaying (or transverse swaying or heaving) motion, the hydrodynamic force on the floating body calculated from the flow field data is fitted using the following formula: F d =F0sin(ωt+ε), where F0 is the amplitude of the hydrodynamic force on the floating body under forced excitation. The hydrodynamic force F... d It can be decomposed into inertial terms and damping terms, i.e. Where A is the sway (or transverse or heave) added mass coefficient, and B is the sway (or transverse or heave) damping coefficient. Then the sway (or transverse or heave) added mass coefficient A = F0cosε / ξ0ω 2 The sway (or transverse or vertical) damping coefficient B = -F0sinε / ξ0ω. Thus, the additional mass coefficients and damping coefficients for all six degrees of freedom can be obtained.

[0152] This embodiment, based on the three-dimensional geometric model of the floating body, its draft, and the frequency and amplitude of the single-mode forced oscillation motion, performs two-dimensional slicing of the floating body's outer shell in a three-dimensional Cartesian coordinate system, generating flow field sampling points on the XZ, YZ, and XY planes respectively. Based on the coordinates of the flow field sampling points, initial flow field data, discrete points of the floating body's geometric outer contour, boundary conditions, and floating body motion parameters, a high-dimensional feature vector is encoded using a learnable matrix, and an input information matrix for each planar slice is constructed. Based on a preset floating body flow field prediction model, two-dimensional flow field prediction is performed on the input information matrix, outputting two-dimensional flow field data for each slice. The two-dimensional flow field data is fused and corrected in three dimensions using a graph attention network to generate a redundant three-dimensional flow field matrix, and missing data on the surface is filled in by interpolation. Based on the corrected three-dimensional flow field data, the surface integral of the hydrodynamic pressure and viscous shear force along the floating body, combined with the single-mode forced oscillation motion equation, determines the target added mass coefficient and the target viscous damping coefficient. This embodiment decomposes the complex three-dimensional flow field into multi-plane high-resolution sampling through two-dimensional slicing of the three-dimensional geometric model, reducing computational complexity while improving the accuracy of key areas. It uses a learnable matrix encoding to dynamically extract high-dimensional physical correlations between flow field data, geometric contours, and motion parameters, avoiding the limitations of traditional empirical formulas. A pre-defined floating body flow field prediction model, based on multi-plane two-dimensional prediction and three-dimensional fusion correction using a graph attention network, solves data redundancy and missing data issues, ensuring spatial consistency of the flow field. Finally, by combining the surface integrals of hydrodynamic pressure and viscous shear force with the single-mode motion equation to directly correlate flow field data with the inertial-damping effect, it eliminates errors in potential flow theory assumptions, achieving high-precision calculation of the six-degree-of-freedom added mass and viscous damping coefficient, thus improving the accuracy of floating body hydrodynamic calculations.

[0153] This application also provides a device for predicting the additional mass coefficient and viscous damping coefficient of a floating body in multi-degree-of-freedom motion. Please refer to... Figure 7 , Figure 7 This is a schematic diagram of the module structure of the device for predicting the additional mass coefficient and viscous damping coefficient of a floating body in multi-degree-of-freedom motion according to an embodiment of this application. The device includes:

[0154] The sampling module 401 is used to perform two-dimensional slicing processing on the outer shell of the floating body in a three-dimensional rectangular coordinate system based on the three-dimensional geometric model of the floating body, the draft, and the frequency and amplitude of the single-mode forced oscillation motion, and to generate flow field sampling points on the XZ plane, YZ plane and XY plane respectively.

[0155] The input matrix module 402 is used to encode the coordinates of the flow field sampling points, the initial flow field data, the discrete points of the outer contour of the floating body geometry, the boundary conditions and the motion parameters of the floating body into high-dimensional feature vectors through a learnable matrix, and to construct the input information matrix of each planar slice.

[0156] The flow field prediction module 403 is used to perform two-dimensional flow field prediction on the input information matrix based on a preset floating body flow field prediction model, and output two-dimensional flow field data for each slice.

[0157] The fusion correction module 404 is used to perform three-dimensional fusion and correction on the two-dimensional flow field data based on the graph attention network, generate a redundant three-dimensional flow field matrix, and fill in the missing data of the object surface by interpolation.

[0158] The target module 405 is used to determine the target additional mass coefficient and the target viscous damping coefficient based on the corrected three-dimensional flow field data, the surface integral dynamic water pressure and viscous shear force along the floating body, and the single-mode forced oscillation motion equation.

[0159] The device for predicting the added mass coefficient and viscous damping coefficient of multi-degree-of-freedom motion of a floating body provided in this application adopts the prediction method for the added mass coefficient and viscous damping coefficient of multi-degree-of-freedom motion of a floating body in the above embodiments, and can solve the technical problem of how to improve the accuracy of hydrodynamic calculations of a floating body. Compared with the prior art, the beneficial effects of the device for predicting the added mass coefficient and viscous damping coefficient of multi-degree-of-freedom motion of a floating body provided in this application are the same as the beneficial effects of the prediction method for the added mass coefficient and viscous damping coefficient of multi-degree-of-freedom motion of a floating body provided in the above embodiments, and other technical features in the device for predicting the added mass coefficient and viscous damping coefficient of multi-degree-of-freedom motion of a floating body are the same as the features disclosed in the methods of the above embodiments, and will not be repeated here.

[0160] This application provides a device for predicting the additional mass coefficient and viscous damping coefficient of a floating body in multi-degree-of-freedom motion. The device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method for predicting the additional mass coefficient and viscous damping coefficient of a floating body in multi-degree-of-freedom motion as described in the above embodiment.

[0161] The following is for reference. Figure 8 , Figure 8 This is a schematic diagram of the hardware operating environment for the prediction method of the additional mass coefficient and viscous damping coefficient of the multi-degree-of-freedom motion of a floating body in the embodiments of this application. It shows a schematic diagram of the structure of the device suitable for implementing the prediction method of the additional mass coefficient and viscous damping coefficient of the multi-degree-of-freedom motion of a floating body in the embodiments of this application. Figure 8As shown, the device for predicting the additional mass coefficient and viscous damping coefficient of a floating body's multi-degree-of-freedom motion may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage device 1003 into a random access memory (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the device. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the device for predicting the additional mass coefficient and viscous damping coefficient of multi-degree-of-freedom motion of a floating body to exchange data wirelessly or via wired communication with other devices. Although the figure shows devices for predicting the additional mass coefficient and viscous damping coefficient of multi-degree-of-freedom motion of a floating body with various systems, it should be understood that it is not required to implement or possess all the systems shown. More or fewer systems can be implemented or possessed alternatively.

[0162] In particular, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, the embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. When the computer program is executed by the processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0163] The device for predicting the added mass coefficient and viscous damping coefficient of multi-degree-of-freedom motion of a floating body provided in this application, employing the prediction method for the added mass coefficient and viscous damping coefficient of multi-degree-of-freedom motion of a floating body in the above embodiments, can solve the technical problem of how to improve the accuracy of hydrodynamic calculations for floating bodies. Compared with the prior art, the beneficial effects of the device for predicting the added mass coefficient and viscous damping coefficient of multi-degree-of-freedom motion of a floating body provided in this application are the same as the beneficial effects of the prediction method for the added mass coefficient and viscous damping coefficient of multi-degree-of-freedom motion of a floating body provided in the above embodiments, and other technical features in this device for predicting the added mass coefficient and viscous damping coefficient of multi-degree-of-freedom motion of a floating body are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0164] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0165] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0166] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, the computer-readable program instructions being used to execute the method for predicting the additional mass coefficient and viscous damping coefficient of the multi-degree-of-freedom motion of a floating body in the above embodiments.

[0167] The aforementioned computer-readable storage medium carries one or more programs. When these programs are executed by the device for predicting the additional mass coefficient and viscous damping coefficient of the multi-degree-of-freedom motion of the floating body, the device performs the following: based on the three-dimensional geometric model of the floating body, its draft, and the frequency and amplitude of the single-mode forced oscillation motion, it performs two-dimensional slicing processing on the outer shell of the floating body in a three-dimensional rectangular coordinate system, generating flow field sampling points on the XZ, YZ, and XY planes respectively; and based on the coordinates of the flow field sampling points, initial flow field data, and discretization of the outer geometric contour of the floating body... Points, boundary conditions, and floating body motion parameters are encoded into high-dimensional feature vectors using learnable matrices, and input information matrices are constructed for each planar slice. Based on a pre-defined floating body flow field prediction model, two-dimensional flow field predictions are performed on the input information matrices, outputting two-dimensional flow field data for each slice. Three-dimensional fusion and correction of the two-dimensional flow field data are performed using a graph attention network to generate a redundant three-dimensional flow field matrix, and missing surface data is supplemented through interpolation. Based on the corrected three-dimensional flow field data, the hydrodynamic pressure and viscous shear force are integrated along the floating body surface, and combined with the single-mode forced oscillation motion equation, the target added mass coefficient and target viscous damping coefficient are determined. Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof. These programming languages ​​include object-oriented programming languages—such as Java, Smalltalk, and C++—and conventional procedural programming languages—such as C or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0168] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0169] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0170] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the above-described method for predicting the additional mass coefficient and viscous damping coefficient of multi-degree-of-freedom motion of a floating body. This method can solve the technical problem of how to improve the accuracy of hydrodynamic calculations for floating bodies. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the method for predicting the additional mass coefficient and viscous damping coefficient of multi-degree-of-freedom motion of a floating body provided in the above embodiments, and will not be elaborated upon here.

[0171] This application provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the above-described method for predicting the additional mass coefficient and viscous damping coefficient of multi-degree-of-freedom motion of a floating body.

[0172] The computer program product provided in this application can solve the technical problem of how to improve the accuracy of hydrodynamic calculations for floating bodies. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as the beneficial effects of the prediction methods for the additional mass coefficient and viscous damping coefficient of multi-degree-of-freedom motion of floating bodies provided in the above embodiments, and will not be repeated here.

[0173] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent scope of this application.

Claims

1. A method for predicting the added mass coefficient and viscous damping coefficient of a floating body in multi-degree-of-freedom motion, characterized in that, The method includes: Based on the three-dimensional geometric model of the floating body, its draft, and the frequency and amplitude of the single-mode forced oscillation motion, the outer shell of the floating body is subjected to two-dimensional slicing in a three-dimensional rectangular coordinate system, and flow field sampling points are generated on the XZ plane, YZ plane, and XY plane, respectively. Based on the coordinates of the flow field sampling points, the initial flow field data, the discrete points of the floating body's geometric outer contour, the boundary conditions, and the floating body's motion parameters, a high-dimensional feature vector is encoded using a learnable matrix, and an input information matrix for each planar slice is constructed. Based on the preset floating body flow field prediction model, two-dimensional flow field prediction is performed on the input information matrix respectively, and two-dimensional flow field data of each slice are output. The two-dimensional flow field data is fused and corrected in three dimensions based on graph attention network to generate a redundant three-dimensional flow field matrix, and the missing data of the object surface is filled by interpolation. Based on the corrected three-dimensional flow field data, the surface integral of the hydrodynamic pressure and viscous shear force along the floating body, combined with the single-mode forced oscillation motion equation, determines the target additional mass coefficient and the target viscous damping coefficient.

2. The method as described in claim 1, characterized in that, Before the step of performing two-dimensional slicing of the outer shell of the floating body in a three-dimensional Cartesian coordinate system based on the three-dimensional geometric model of the floating body, its draft, and the frequency and amplitude of the single-mode forced oscillation motion, and generating flow field sampling points on the XZ, YZ, and XY planes respectively, the method further includes: The first prediction module and the second prediction module are constructed based on the Transformer model, and the third prediction module is constructed based on the graph attention network. The first prediction module is used to predict the two-dimensional flow field in the XZ plane and the YZ plane, the second prediction module is used to predict the two-dimensional flow field in the XY plane, and the third prediction module is used to integrate multi-plane slice data and correct redundancy. Based on a pre-set publicly available two-dimensional computational fluid dynamics dataset, the first prediction module and the second prediction module are pre-trained. Based on the 3D proprietary dataset generated by CFD, the pre-trained first and second prediction modules are optimized, and the continuity equation, momentum conservation equation and surface conditions are embedded as physical constraints to obtain the target first prediction module and the target second prediction module. The parameters of the first prediction module and the second prediction module of the target are fixed, and the third prediction module is trained based on a preset three-dimensional proprietary dataset to obtain the third prediction module of the target; The first prediction module, the second prediction module, and the third prediction module are integrated to obtain the preset floating body flow field prediction model.

3. The method as described in claim 1, characterized in that, Before the step of performing two-dimensional slicing of the outer shell of the floating body in a three-dimensional Cartesian coordinate system based on the three-dimensional geometric model of the floating body, its draft, and the frequency and amplitude of the single-mode forced oscillation motion, and generating flow field sampling points on the XZ, YZ, and XY planes respectively, the method further includes: The three-dimensional geometric model of the floating body is geometrically verified to identify non-streamlined regions and potential numerical calculation error points in the model. Based on the verification results, the discretization parameters of the outer shell of the float are adjusted. The discretization parameters include slice density, dividing line spacing, and surface discrete point distribution rules.

4. The method as described in claim 1, characterized in that, After the steps of performing three-dimensional fusion and correction on the two-dimensional flow field data based on the graph attention network to generate a redundant three-dimensional flow field matrix, and then filling in the missing data of the object surface through interpolation, the method further includes: Based on the corrected three-dimensional flow field data, the confidence level of the flow field prediction is dynamically evaluated; If the confidence level is lower than the preset threshold, the floating body motion parameters or boundary conditions are readjusted, a new input information matrix is ​​generated, and the two-dimensional flow field prediction and three-dimensional fusion steps are iteratively executed. If the confidence level reaches the threshold, the current three-dimensional flow field data will be used as the target three-dimensional flow field data.

5. The method as described in claim 1, characterized in that, The step of performing two-dimensional slicing of the outer shell of the floating body in a three-dimensional Cartesian coordinate system based on the three-dimensional geometric model of the floating body, its draft, and the frequency and amplitude of the single-mode forced oscillation motion, and generating flow field sampling points on the XZ, YZ, and XY planes respectively, includes: Based on the three-dimensional geometric model of the floating body, its draft, and the frequency and amplitude of the single-mode forced oscillation motion, the outer shell of the floating body is divided into regions, resulting in regular regions and asymmetric regions. Based on the asymmetric partitioning strategy, the asymmetric region is sliced ​​in two dimensions in a three-dimensional Cartesian coordinate system to generate asymmetric sampling points. The asymmetric partitioning strategy includes increasing the slice density in regions where the geometric curvature change is greater than a preset curvature change threshold and using non-uniform dividing line spacing in the XY plane. Based on the rule-based partitioning strategy, the rule region is processed into two-dimensional slices in a three-dimensional Cartesian coordinate system to generate rule sampling points. The rule-based partitioning strategy includes using equidistant plane data and prioritizing the retention of planar data orthogonal to the direction of the motion degree of freedom of the floating body. The asymmetric sampling points and the regular sampling points are used as the flow field sampling points.

6. The method as described in claim 1, characterized in that, The steps for determining the target added mass coefficient and the target viscous damping coefficient based on the corrected three-dimensional flow field data, the surface integral of the hydrodynamic pressure and viscous shear force along the floating body, and the single-mode forced oscillation motion equation include: Extract the hydrodynamic pressure, viscous shear force, surface normal direction, and position coordinates of the discrete elements on the surface of the floating body from the corrected three-dimensional flow field data; The dynamic water pressure and viscous shear force are integrated along the surface of the floating body. The dynamic water pressure and viscous shear force of the discrete unit are numerically integrated to obtain the total hydrodynamic force and total hydrodynamic torque on the floating body. According to the single-mode forced oscillation motion equation, the total hydrodynamic force and the total hydrodynamic torque are decomposed into an inertial term related to acceleration and a damping term related to velocity. Based on the inertial term and the damping term, the target additional mass coefficient and the target viscous damping coefficient are obtained by using phase matching and amplitude scaling algorithms.

7. A device for predicting the additional mass coefficient and viscous damping coefficient of a floating body in multi-degree-of-freedom motion, characterized in that, The device includes: The sampling module is used to perform two-dimensional slicing of the outer shell of the floating body in a three-dimensional Cartesian coordinate system based on the three-dimensional geometric model of the floating body, its draft, and the frequency and amplitude of the single-mode forced oscillation motion, and to generate flow field sampling points on the XZ plane, YZ plane and XY plane respectively. The input matrix module is used to encode the coordinates of the flow field sampling points, initial flow field data, discrete points of the outer contour of the floating body geometry, boundary conditions and floating body motion parameters into high-dimensional feature vectors through learnable matrix encoding, and to construct the input information matrix of each planar slice. The flow field prediction module is used to perform two-dimensional flow field prediction on the input information matrix based on a preset floating body flow field prediction model, and output two-dimensional flow field data for each slice. The fusion correction module is used to perform three-dimensional fusion and correction on the two-dimensional flow field data based on the graph attention network, generate a redundant three-dimensional flow field matrix, and fill in the missing data of the object surface by interpolation. The target module is used to determine the target additional mass coefficient and the target viscous damping coefficient based on the corrected three-dimensional flow field data, the surface integral of the hydrodynamic pressure and viscous shear force along the floating body, and the single-mode forced oscillation motion equation.

8. A computer device, characterized in that, The device includes: a memory, a processor, and a program for predicting the additional mass coefficient and viscous damping coefficient of a floating body in multi-degree-of-freedom motion, stored in the memory and executable on the processor, the program being configured to implement the steps of the method for predicting the additional mass coefficient and viscous damping coefficient of a floating body in multi-degree-of-freedom motion as described in any one of claims 1 to 6.

9. A storage medium, characterized in that, The storage medium stores a program for predicting the additional mass coefficient and viscous damping coefficient of a floating body in multi-degree-of-freedom motion. When the processor executes the program for predicting the additional mass coefficient and viscous damping coefficient of a floating body in multi-degree-of-freedom motion, it implements the steps of the method for predicting the additional mass coefficient and viscous damping coefficient of a floating body in multi-degree-of-freedom motion as described in any one of claims 1 to 6.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the steps of the method for predicting the additional mass coefficient and viscous damping coefficient of multi-degree-of-freedom motion of a floating body as described in any one of claims 1 to 6.

Images