Method and system for predicting unsteady wave motion response of water surface aircraft
The unsteady hydrodynamic performance of surface aircraft is predicted by a deep learning neural network based on conditional diffusion, which solves the problems of large computational complexity and high cost of traditional methods, realizes fast and low-cost acquisition of high-precision hydrodynamic performance data, and supports the rapid design of surface aircraft.
Patent Information
- Application Number
- CN202510697976.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-28
- Publication Date
- 2025-09-23
AI Technical Summary
Traditional ship unsteady wave motion response methods are not applicable to surface aircraft. They are computationally intensive and costly, and cannot effectively predict the hydrodynamic performance of seaplanes under different ship types.
A deep learning neural network based on conditional diffusion is used to obtain the unsteady hydrodynamic performance data of surface aircraft through CFD simulation calculation and frame extraction processing. A prediction model for the unsteady hydrodynamic performance of surface aircraft is constructed. The diffusion module is used to learn the noise law of the noisy hydrodynamic performance curve, and supervised training and testing are performed.
It achieves low-cost and high-precision acquisition of unsteady hydrodynamic performance data of seaplanes, which is suitable for the rapid design of surface aircraft, reduces the amount of calculation and acquisition costs, and improves the prediction accuracy.
Smart Images
Figure CN120688226A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of seaplane design, and relates to a method and system for predicting unsteady wave motion, and in particular to a method and system for predicting the unsteady wave motion response of a surface aircraft. Background Art
[0002] Unsteady hydrodynamic performance prediction involves using hydrodynamic performance curves under a limited number of ship operating conditions to predict the hydrodynamic performance curves under unknown ship operating conditions, thereby obtaining high-precision hydrodynamic performance data at a low cost. During landing and hydroplaning, a seaplane's hydrodynamic performance varies differently under different ship operating conditions. Therefore, obtaining hydrodynamic performance curves under unknown ship operating conditions is of great significance to seaplane design. Furthermore, during landing and hydroplaning, a seaplane's hydrodynamic performance can change dramatically over time due to the impact of the water surface. This means that the hydrodynamic performance curves are highly nonlinear and time-series, making their prediction challenging.
[0003] The hydrodynamic performance of surface aircraft requires a balance between aerodynamics and hydrodynamics, so traditional technical solutions that focus on long-term hydrodynamic efficiency and stability in ships cannot be applied to surface aircraft. The differences include:
[0004] 1. Dynamics: Ships are mainly steady-state, while surface aircraft need to deal with strong unsteady cross-medium flow, that is, air and water medium.
[0005] 2. Multi-physics: Surface vehicles need to couple hydrodynamics and aerodynamics, while ships only need to deal with hydrodynamics.
[0006] 3. Data scarcity: The cross-media test data of surface aircraft is several orders of magnitude less than the sample data of ships.
[0007] 4. Different real-time requirements: The control decision response time of surface aircraft is in milliseconds, which is several orders of magnitude lower than that of ships. Summary of the Invention
[0008] In order to solve the problem that traditional methods for responding to unsteady wave motions of ships cannot be applied to surface aircraft, the present invention provides a method and system for predicting the unsteady wave motion responses of surface aircraft. The hydrodynamic performance curves under existing working conditions are used to predict the hydrodynamic performance curves under unknown working conditions, and the shortcomings of traditional methods such as large computational complexity and high acquisition cost are optimized.
[0009] The technical solutions of the present invention are as follows:
[0010] A method for predicting the unsteady wave motion response of a water surface vehicle comprises the following steps:
[0011] S1, the input-output data sample pairs of the unsteady hydrodynamic performance data of the surface aircraft obtained by computer simulation calculation;
[0012] S2, construct a deep learning neural network based on conditional diffusion for the prediction of unsteady hydrodynamic performance of surface vehicles;
[0013] S3 uses the training sample database to train the deep learning neural network of S2, and learns the noise pattern of the noisy hydrodynamic performance curve through the diffusion module;
[0014] S4, tests the deep learning neural network trained in S3.
[0015] Furthermore, in S1, the unsteady hydrodynamic performance data of the surface aircraft is calculated through CFD simulation, and the calculated unsteady hydrodynamic performance data is subjected to frame extraction processing.
[0016] Furthermore, the frame extraction process specifically includes: extracting frames from the time axis, that is, sampling a frame every other frame.
[0017] Furthermore, in S1, the unsteady hydrodynamic performance data of the surface vehicle include:
[0018] Buoyancy and hydrostatic parameters, including displacement, draft, and center of buoyancy;
[0019] Hydrodynamic performance parameters, including hydrodynamic lift, hydrodynamic drag, and glide angle;
[0020] Dynamic response parameters, pitch, roll, heave;
[0021] Splash and flow field characteristics, including splash height and splash range, and step flow separation.
[0022] Furthermore, in S2, the deep learning neural network based on conditional diffusion includes a diffusion module and a data processing module; the diffusion module contains six layers of TCN and two layers of full connection, the output dimensions of the two layers of full connection are 32 and 1 respectively, the six layers of TCN are composed of six layers of Block, including one BasicConvBlock and five ConvBlock, BasicConvBlock contains two layers of causal dilation convolution, and outputs different dimensions for water skiing data and water landing data; ConvBlock contains three layers of causal dilation convolution, and outputs different dimensions for water skiing data and water landing data; the convolution kernel size of each Block in the output dimension is 3, the step size is 1, and the expansion coefficient increases exponentially by 3 as the number of Block layers increases.
[0023] Furthermore, the data processing module performs feature alignment on the unsteady hydrodynamic performance data, position encodes the number of noise addition steps t of each hydrodynamic performance curve, and adds random Gaussian noise to the hydrodynamic performance curve according to the number of noise addition steps.
[0024] Furthermore, in S3, the diffusion module inputs the unsteady hydrodynamic performance data, the number of noise addition steps t, and the hydrodynamic performance curve corresponding to the number of noise addition steps, and outputs the predicted noise ε′; the loss function of supervised training is the MSE loss, which is the square of the difference between the predicted noise ε′ and the actual added noise ε.
[0025] A surface aircraft unsteady wave motion response prediction system includes a computer storing surface aircraft unsteady wave motion response prediction software, which runs the above-mentioned surface aircraft unsteady wave motion response prediction method.
[0026] Furthermore, the surface vehicle unsteady wave motion response prediction software includes a sample generation module, a neural network construction and training module, and a neural network testing module. The sample generation module runs S1, the neural network construction and training module runs S2 and S3, and the neural network testing module runs S4.
[0027] The beneficial effects of this application are:
[0028] The deep learning method based on the conditional diffusion model proposed in this paper for predicting the unsteady hydrodynamic performance of seaplanes primarily addresses the shortcomings of traditional methods, such as high computational complexity and acquisition costs. Compared to traditional methods, this method achieves faster and lower acquisition costs for acquiring hydrodynamic performance data, while also achieving excellent prediction accuracy. Therefore, it is suitable for acquiring high-precision, low-cost unsteady hydrodynamic performance data for seaplanes. This results in a data-driven analysis method for the unsteady hydrodynamic performance of surface aircraft, distinct from experimental research and numerical simulation, enabling rapid design of surface aircraft. BRIEF DESCRIPTION OF THE DRAWINGS
[0029] Figure 1 It is an implementation diagram of the unsteady wave motion response prediction method of the present invention. DETAILED DESCRIPTION
[0030] This section is an embodiment of the present invention, which is used to explain and illustrate the technical solution of the present invention. In the absence of conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other.
[0031] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate directions or positional relationships based on the orientations or positional relationships in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or case referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second" and the like are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance or implying the number of technical features indicated. Thus, features defined as "first", "second" and the like may explicitly or implicitly include more of the features. In the description of the present invention, unless otherwise specified, "multiple" means two or more.
[0032] In the description of the present invention, it should be noted that, unless otherwise expressly specified or limited, the terms "mounted," "connected," and "connected" should be interpreted broadly. For example, they may refer to fixed, detachable, or integrated connections; mechanical or point connections; direct or indirect connections through an intermediary; and internal communication between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.
[0033] Example 1:
[0034] A method for predicting the unsteady wave motion response of a water surface vehicle comprises the following steps:
[0035] S1, the input-output data sample pairs of the unsteady hydrodynamic performance data of the surface aircraft obtained by computer simulation calculation;
[0036] S2, construct a deep learning neural network based on conditional diffusion for the prediction of unsteady hydrodynamic performance of surface vehicles;
[0037] S3 uses the training sample database to train the deep learning neural network of S2, and learns the noise pattern of the noisy hydrodynamic performance curve through the diffusion module;
[0038] S4, tests the deep learning neural network trained in S3.
[0039] In S1, the unsteady hydrodynamic performance data of the surface aircraft is calculated through CFD simulation, and the calculated unsteady hydrodynamic performance data is subjected to frame extraction processing.
[0040] The frame extraction process is specifically as follows: extracting frames from the time axis, that is, sampling a frame every few frames.
[0041] In S1, the unsteady hydrodynamic performance data of the surface vehicle include:
[0042] Buoyancy and hydrostatic parameters, including displacement, draft, and center of buoyancy;
[0043] Hydrodynamic performance parameters, including hydrodynamic lift, hydrodynamic drag, and glide angle;
[0044] Dynamic response parameters, pitch, roll, heave;
[0045] Splash and flow field characteristics, including splash height and splash range, and step flow separation.
[0046] In S2, the deep learning neural network based on conditional diffusion includes a diffusion module and a data processing module; the diffusion module contains six layers of TCN and two layers of full connection, and the output dimensions of the two layers of full connection are 32 and 1 respectively. The six layers of TCN are composed of six layers of blocks, including one BasicConvBlock and five ConvBlocks. BasicConvBlock contains two layers of causal dilation convolution, which output different dimensions for water skiing data and water landing data; ConvBlock contains three layers of causal dilation convolution, which output different dimensions for water skiing data and water landing data; the convolution kernel size of each block in the output dimension is 3, the step size is 1, and the expansion coefficient increases exponentially by 3 as the number of block layers increases.
[0047] The data processing module performs feature alignment on the unsteady hydrodynamic performance data, position encodes the number of noise addition steps t of each hydrodynamic performance curve, and adds random Gaussian noise to the hydrodynamic performance curve according to the number of noise addition steps.
[0048] In S3, the diffusion module inputs the unsteady hydrodynamic performance data, the number of noise addition steps t, and the hydrodynamic performance curve corresponding to the number of noise addition steps, and outputs the predicted noise ε′; the loss function of supervised training is the MSE loss, which is the square of the difference between the predicted noise ε′ and the actual added noise ε.
[0049] A surface aircraft unsteady wave motion response prediction system includes a computer storing surface aircraft unsteady wave motion response prediction software, which runs the above-mentioned surface aircraft unsteady wave motion response prediction method.
[0050] The surface vehicle unsteady wave motion response prediction software includes a sample generation module, a neural network construction and training module, and a neural network testing module. The sample generation module runs S1, the neural network construction and training module runs S2 and S3, and the neural network testing module runs S4.
[0051] Example 2:
[0052] In this invention, the client provides training ship model samples and operating conditions. The server calculates the flow field based on this information and generates a hydrodynamic performance database. A deep learning model based on conditional diffusion is then trained. Next, the client provides the ship model operating condition parameters to be predicted. Finally, the server predicts the corresponding unsteady hydrodynamic performance curve based on the predicted ship model operating condition and returns it to the client.
[0053] Ship type operating parameters, i.e., the unsteady hydrodynamic performance data of surface aircraft, are key physical quantities that describe the interaction between surface aircraft and water during operations on the water surface, such as takeoff, landing, and taxiing. These parameters directly affect the stability, resistance, lift, and cross-medium transition performance of surface aircraft. The main parameters include:
[0054] 1. Buoyancy and hydrostatic parameters
[0055] Displacement: The weight of water displaced by an aircraft when it is stationary and floating on the water surface, which determines the initial buoyancy.
[0056] Draft: The vertical depth to which the aircraft's hull or pontoons are immersed in water, affecting stability and drag.
[0057] Center of buoyancy: The point of action of the resultant buoyancy force, whose relative position to the center of gravity determines static stability.
[0058] 2. Hydrodynamic performance parameters
[0059] Hydrodynamic lift: The vertical support force of water on the aircraft hull / float during gliding, which is proportional to the square of the speed.
[0060] Hydrodynamic drag is also known as frictional drag: it is caused by the wetted surface area of the aircraft hull and the viscosity of the water flow.
[0061] Pressure differential drag: caused by the pressure difference between the front and rear of the aircraft hull, including wave-making drag and splash drag.
[0062] Glide angle: The angle between the longitudinal axis of the aircraft hull and the horizontal plane. Optimizing the angle can reduce drag and increase lift.
[0063] 3. Dynamic response parameters:
[0064] Pitch / roll stability.
[0065] Pitch: The pitching oscillation during taxiing.
[0066] Roll: The tilting caused by sideways waves that needs to be suppressed to prevent the wingtip from touching the water.
[0067] Heave: Vertical movement caused by waves, which affects the safety of takeoff and landing.
[0068] 4. Splashing and flow field characteristics
[0069] Splash height / range: The distribution of water splashing after it hits the aircraft hull, which needs to be minimized to avoid damaging the engines or wings.
[0070] Step flow separation: The control effect of the stepped design of the aircraft hull on water flow separation, such as the seagull wing.
[0071] The server generates the raw data for the first step. The server generates a grid based on the training ship model and operating conditions provided by the client, and then calculates the flow field. Ship model parameters include the rise angle at the step and the rear body length / forebody length, and operating conditions include wavelength and speed.
[0072] The second step involves building a training data sample library suitable for deep learning. Specifically, the time axis of the unsteady hydrodynamic performance data obtained through CFD simulation is too dense, so a frame extraction operation is performed on the time axis. This sampling operation is performed every few frames to achieve a sparse data effect, which in turn helps accelerate model convergence. The hydrodynamic performance data for all ship operating conditions is then processed into input-output data sample pairs that are convenient for the deep learning model.
[0073] The deep learning model for unsteady hydrodynamic performance prediction constructed in the third step specifically involves: after constructing input-output data sample pairs suitable for deep learning, a deep learning model for unsteady hydrodynamic performance prediction is constructed based on the dimensional characteristics of the input-output pairs using a conditional diffusion model. The specific structure of the deep learning model based on conditional diffusion includes a diffusion module and a data processing module. The diffusion module contains a six-layer Transistor-Convolutional Network (TCN) and two fully connected layers. The output dimensions of the two fully connected layers are 32 and 1, respectively. The TCN consists of six layers of blocks, including one BasicConvBlock and five ConvBlocks. The BasicConvBlock contains two layers of causal dilated convolution, with output dimensions of 256, 256 for water skiing data and 128, 128 for landing data. The ConvBlock contains three layers of causal dilated convolution, with output dimensions of 128, 128, and 256 for water skiing data and 64, 64, and 128 for landing data. The convolution kernel size for each block in the output dimension is 3, with a stride of 1. The expansion coefficient increases exponentially with the number of blocks, reaching 3, 9, 27, 81, 243, and 729, respectively. The data processing module primarily performs feature alignment on ship operating parameters, positionally encodes the number of noise addition steps t for each hydrodynamic performance curve, and adds random Gaussian noise to the hydrodynamic performance curves based on the number of noise addition steps. The deep learning model based on conditional diffusion specifically learns the noise patterns of the noisy hydrodynamic performance curves through the diffusion module.
[0074] Because surface aircraft data is highly nonlinear and time-series, it's important to avoid disrupting the temporal relationship during frame extraction. A rolling window should be used to normalize the time series data to prevent global normalization from introducing future information. Time-series-sensitive interpolation or anomaly detection can be employed. The window size should be appropriately set, and when external variables are present, multi-input branches or spatiotemporal networks should be used for assistance.
[0075] The fourth step, training the deep learning model based on conditional diffusion, specifically involves using the training data sample library to train the deep learning model. The data flow for training the deep learning model based on conditional diffusion is as follows: the diffusion module inputs the ship's operating condition, the number of noise addition steps t, and the hydrodynamic performance curve corresponding to the number of noise addition steps, and outputs the predicted noise ε′. The loss function for supervised training is the MSE loss, which is the square of the difference between the predicted noise ε′ and the actual added noise ε, as shown in Equation 1.
[0076]
[0077] The fifth step, predicting unsteady hydrodynamic performance, involves performing unsteady hydrodynamic performance prediction based on the client's ship operating conditions and a trained deep learning model based on conditional diffusion. The diffusion module predicts the Gaussian noise corresponding to the noise curve. It then gradually removes the noise from the pure Gaussian noise curve according to the ship operating conditions, generating the corresponding hydrodynamic performance curve. This achieves the effect of regression prediction of the unsteady hydrodynamic performance curve. Finally, the predicted unsteady hydrodynamic performance data is returned to the client.
Claims
1. A method for predicting the unsteady wave motion response of a water surface vehicle, characterized in that: The following steps are involved: S1, the input-output data sample pairs of the unsteady hydrodynamic performance data of the surface aircraft obtained by computer simulation calculation; S2, construct a deep learning neural network based on conditional diffusion for the prediction of unsteady hydrodynamic performance of surface vehicles; S3 uses the training sample database to train the deep learning neural network of S2, and learns the noise pattern of the noisy hydrodynamic performance curve through the diffusion module; S4, tests the deep learning neural network trained in S3.
2. The method for predicting the unsteady wave motion response of a water surface vehicle according to claim 1, characterized in that: In S1, the unsteady hydrodynamic performance data of the surface aircraft is calculated through CFD simulation, and the calculated unsteady hydrodynamic performance data is subjected to frame extraction processing.
3. The method for predicting the unsteady wave motion response of a water surface vehicle according to claim 2, characterized in that: The frame extraction process is specifically as follows: extracting frames from the time axis, that is, sampling a frame every few frames.
4. The method for predicting the unsteady wave motion response of a water surface vehicle according to claim 1, characterized in that: In S1, the unsteady hydrodynamic performance data of the surface vehicle include: Buoyancy and hydrostatic parameters, including displacement, draft, and center of buoyancy; Hydrodynamic performance parameters, including hydrodynamic lift, hydrodynamic drag, and glide angle; Dynamic response parameters, pitch, roll, heave; Splash and flow field characteristics, including splash height and splash range, and step flow separation.
5. The method for predicting the unsteady wave motion response of a water surface vehicle according to claim 1, characterized in that: In S2, the deep learning neural network based on conditional diffusion includes a diffusion module and a data processing module; the diffusion module contains six layers of TCN and two layers of full connection, and the output dimensions of the two layers of full connection are 32 and 1 respectively. The six layers of TCN are composed of six layers of blocks, including one BasicConvBlock and five ConvBlocks. BasicConvBlock contains two layers of causal dilation convolution, which output different dimensions for water skiing data and water landing data; ConvBlock contains three layers of causal dilation convolution, which output different dimensions for water skiing data and water landing data; the convolution kernel size of each block in the output dimension is 3, the step size is 1, and the expansion coefficient increases exponentially by 3 as the number of block layers increases.
6. The method for predicting the unsteady wave motion response of a water surface vehicle according to claim 5, characterized in that: The data processing module performs feature alignment on the unsteady hydrodynamic performance data, position encodes the number of noise addition steps t of each hydrodynamic performance curve, and adds random Gaussian noise to the hydrodynamic performance curve according to the number of noise addition steps.
7. The method for predicting the unsteady wave motion response of a water surface vehicle according to claim 1, characterized in that: In S3, the diffusion module inputs the unsteady hydrodynamic performance data, the number of noise addition steps t, and the hydrodynamic performance curve corresponding to the number of noise addition steps, and outputs the predicted noise ε′; the loss function of supervised training is the MSE loss, which is the square of the difference between the predicted noise ε′ and the actual added noise ε.
8. A system for predicting the response of a surface vehicle to unsteady wave motion, comprising a computer, characterized in that: The computer stores water surface vehicle unsteady wave motion response prediction software, which runs a water surface vehicle unsteady wave motion response prediction method according to any one of claims 1 to 7.
9. The system for predicting the unsteady wave motion response of a water surface vehicle according to claim 8, characterized in that: The surface vehicle unsteady wave motion response prediction software includes a sample generation module, a neural network construction and training module, and a neural network testing module. The sample generation module runs S1, the neural network construction and training module runs S2 and S3, and the neural network testing module runs S4.