A surface vehicle ship bottom holographic water load reconstruction method and system

Abstract

The application discloses a water surface aircraft ship bottom holographic water load reconstruction method and system, and relates to the flow field reconstruction technical field of fluid mechanics. The method comprises the following steps: constructing a three-dimensional model of a holographic grid unit corresponding to a target water surface aircraft according to preset fineness expectation information; inputting ship type parameters of the target water surface aircraft, working condition parameter of the target water surface aircraft and center point coordinates of each grid in the three-dimensional model of the holographic grid unit corresponding to the target water surface aircraft into a holographic water load reconstruction model to obtain a holographic water load of a ship bottom of the target water surface aircraft. The application can reduce the time of holographic water load reconstruction.

Description

Technical Field

[0001] This invention relates to the field of flow field reconstruction technology in fluid mechanics, and in particular to a method and system for reconstructing holographic water loads on the bottom of a surface aircraft. Background Technology

[0002] The distribution of hull loads on a surface vehicle (SAV) is a crucial indicator of its hydrodynamic performance and plays a vital role in its design. Computational Fluid Dynamics (CFD) assisted by tank testing is the mainstream method for obtaining holographic water loads. First, a limited number of sensors are placed at specific locations on the SAV's hull surface to collect corresponding hydrodynamic physical quantities during tank testing; these quantities are typically considered as true values. Then, CFD-based analysis and calculations are performed under identical operating conditions to obtain holographic data. Finally, the true values ​​collected from the tank test are used to correct the accuracy of the CFD analysis and calculation results, ensuring the validity of the simulation results. However, high-quality CFD analysis and calculations require significant computational resources and place high demands on the personnel. This has significantly extended the experimental and product verification cycles, resulting in longer holographic water load reconstruction times. Therefore, a key challenge we now face is how to reduce the experimental and product verification cycles to shorten the holographic water load reconstruction time. Summary of the Invention

[0003] The purpose of this invention is to provide a method and system for reconstructing holographic water loads on the bottom of a surface aircraft, which can reduce the time required for holographic water load reconstruction.

[0004] To achieve the above objectives, the present invention provides the following solution:

[0005] A method for reconstructing holographic water loads on the hull of a surface aircraft includes:

[0006] Construct a holographic mesh 3D model of the target surface aircraft based on the preset precision expectation information;

[0007] The target surface aircraft's hull parameters, operating condition parameters, and the center point coordinates of each grid in the holographic mesh unit 3D model corresponding to the target surface aircraft are input into the holographic water load reconstruction model to obtain the target surface aircraft's hull holographic water load.

[0008] Optionally, the process of determining the holographic water load reconstruction model includes:

[0009] Obtain a sample database; the sample database includes multiple samples, each sample including raw data from the pool experiment of the sample surface aircraft; the raw data from the pool experiment of the sample surface aircraft includes the load values ​​measured by each sensor on the bottom of the sample surface aircraft, the position of each sensor, the hull parameters of the sample surface aircraft, and the operating condition parameters of the sample surface aircraft;

[0010] Using the positions of the sensors on the bottom of each sample surface aircraft in the sample database, the hull parameters of each sample surface aircraft, and the operating condition parameters of each sample surface aircraft as inputs, and the load values ​​measured by the sensors on the bottom of each sample surface aircraft in the sample database as outputs, the attention neural network process model is trained to obtain a holographic water load reconstruction model.

[0011] Optionally, the attention neural network process model includes: a deterministic path encoder, a hidden path encoder, and a conditional decoder; the conditional decoder includes a first input terminal and a second input terminal, the input terminal of the deterministic path encoder and the input terminal of the hidden path encoder are connected; the output terminal of the deterministic path encoder and the output terminal of the hidden path encoder are both connected to the first input terminal of the conditional decoder; the input terminal of the deterministic path encoder and the input terminal of the hidden path encoder serve as the first input terminal of the attention neural network process model; the second input terminal of the conditional decoder serves as the second input terminal of the attention neural network process model.

[0012] Optionally, the deterministic path encoder includes a fully connected module, a self-attention module, and a cross-attention module connected in sequence.

[0013] Optionally, the hidden path encoder includes a fully connected module, a self-attention module, and an average aggregation module connected in sequence.

[0014] Optionally, the conditional decoder includes six fully connected layers connected in sequence.

[0015] Optionally, using the positions of sensors on the hull of each sample surface aircraft in the sample database, the hull type parameters of each sample surface aircraft, and the operating condition parameters of each sample surface aircraft as inputs, and the load values ​​measured by the sensors on the hull of each sample surface aircraft in the sample database as outputs, an attention neural network process model is trained to obtain a holographic water load reconstruction model, specifically including:

[0016] A sample sub-database is obtained by selecting a set number of samples from the sample database.

[0017] In the current iteration, the attention neural network process model in the previous iteration is tuned to obtain the attention neural network process model in the current iteration.

[0018] The first global latent variable distribution is obtained by inputting the samples corresponding to the current iteration number in the sample sub-database into the hidden path encoder of the attention neural network process model at the current iteration number.

[0019] The sample corresponding to the current iteration number in the sample database is input into the hidden path encoder of the attention neural network process model at the current iteration number to obtain the second global latent variable distribution;

[0020] The intermediate representation is obtained by inputting the samples corresponding to the current iteration number in the sample sub-database, the positions of each sensor on the bottom of the sample surface aircraft corresponding to the current iteration number in the sample database, the ship type parameters of the sample surface aircraft corresponding to the current iteration number, and the operating condition parameters of the sample surface aircraft corresponding to the current iteration number into the deterministic path encoder of the attention neural network process model at the current iteration number.

[0021] The sampled values ​​of the first global latent variable distribution are obtained based on the mean and variance of the first global latent variable distribution;

[0022] The intermediate representation, the first global latent variable distribution sample value, the position of each sensor on the bottom of the sample surface aircraft corresponding to the current iteration number in the sample database, the hull type parameters of the sample surface aircraft corresponding to the current iteration number, and the operating condition parameters of the sample surface aircraft corresponding to the current iteration number are input into the conditional decoder of the attention neural network process model at the current iteration number to obtain the predicted load value distribution.

[0023] The loss function value is obtained based on the first global latent variable distribution, the second global latent variable distribution, the predicted load value distribution, and the load values ​​measured by each sensor on the bottom of the sample surface aircraft corresponding to the current iteration number in the sample database.

[0024] If the loss function value reaches the set condition, then the attention neural network process model at the current iteration number is determined to be the holographic water load reconstruction model;

[0025] If the loss function value reaches the set condition, the iteration count is updated and the next iteration begins.

[0026] Optionally, the expected precision information is the total number of grids in the holographic grid unit 3D model corresponding to the target surface aircraft.

[0027] A computer system includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method described above.

[0028] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0029] This invention constructs a holographic mesh unit 3D model of the target surface aircraft based on preset precision expectation information; inputting the ship type parameters of the target surface aircraft, the operating condition parameters of the target surface aircraft, and the center point coordinates of each mesh in the holographic mesh unit 3D model of the target surface aircraft into the holographic water load reconstruction model, the holographic water load of the target surface aircraft's bottom can be obtained, which can reduce the dependence on CFD simulation calculation, thereby reducing the time for holographic water load reconstruction. Attached Figure Description

[0030] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0031] Figure 1 A flowchart illustrating the holographic water load reconstruction method for the hull of a surface aircraft provided in this embodiment of the invention;

[0032] Figure 2 A schematic diagram of the hardware environment provided for an embodiment of the present invention;

[0033] Figure 3 This is a schematic diagram of the sensor arrangement in a water tank test according to an embodiment of the present invention;

[0034] Figure 4 A schematic diagram illustrating the generation of a 3D model of a holographic mesh unit provided in an embodiment of the present invention;

[0035] Figure 5 A schematic diagram of the structure of the attention neural process model provided in an embodiment of the present invention;

[0036] Figure 6 A framework diagram of the holographic water load reconstruction method for the bottom of a surface aircraft provided in an embodiment of the present invention;

[0037] Figure 7 A modular composition diagram of the device provided in the embodiments of the present invention;

[0038] Figure 8 This is a diagram of the internal structure of a computer system. Detailed Implementation

[0039] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0040] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0041] like Figure 1 As shown in the figure, an embodiment of the present invention provides a method for reconstructing the holographic water load on the bottom of a surface aircraft, comprising:

[0042] Step 101: Construct a 3D model of the target surface aircraft using holographic mesh units based on the preset precision expectation information.

[0043] Step 102: Input the ship type parameters of the target surface aircraft, the operating condition parameters of the target surface aircraft, and the center point coordinates of each grid in the holographic mesh unit 3D model of the target surface aircraft into the holographic water load reconstruction model to obtain the holographic water load on the bottom of the target surface aircraft.

[0044] In practical applications, the process of determining the holographic water load reconstruction model includes:

[0045] Obtain a sample database; the sample database includes multiple samples, each sample including raw data from the pool experiment of the sample surface aircraft; the raw data from the pool experiment of the sample surface aircraft includes the load values ​​measured by each sensor on the bottom of the sample surface aircraft, the position of each sensor, the hull parameters of the sample surface aircraft, and the operating condition parameters of the sample surface aircraft.

[0046] Using the positions of the sensors on the bottom of each sample surface aircraft in the sample database, the hull parameters of each sample surface aircraft, and the operating condition parameters of each sample surface aircraft as inputs, and the load values ​​measured by the sensors on the bottom of each sample surface aircraft in the sample database as outputs, the attention neural network process model is trained to obtain a holographic water load reconstruction model.

[0047] In practical applications, the attention neural network process model includes: a deterministic path encoder, a hidden path encoder, and a conditional decoder; the conditional decoder includes a first input terminal and a second input terminal, and the input terminals of the deterministic path encoder and the hidden path encoder are connected; the output terminals of the deterministic path encoder and the hidden path encoder are both connected to the first input terminal of the conditional decoder; the input terminals of the deterministic path encoder and the hidden path encoder serve as the first input terminal of the attention neural network process model; the second input terminal of the conditional decoder serves as the second input terminal of the attention neural network process model.

[0048] In practical applications, the deterministic path encoder includes a fully connected module, a self-attention module, and a cross-attention module connected in sequence.

[0049] In practical applications, the hidden path encoder includes a fully connected module, a self-attention module, and an average aggregation module connected in sequence.

[0050] In practical applications, the conditional decoder comprises six fully connected layers connected in sequence.

[0051] In practical applications, the holographic water load reconstruction model is obtained by training an attention neural network process model with the positions of sensors on the bottom of each sample surface aircraft in the sample database, the hull type parameters of each sample surface aircraft, and the operating condition parameters of each sample surface aircraft as inputs, and the load values ​​measured by each sensor on the bottom of each sample surface aircraft in the sample database as outputs. Specifically, this includes:

[0052] A sample sub-database is obtained by selecting a set number of samples from the sample database.

[0053] At the current iteration number, the attention neural network process model from the previous iteration number is tuned to obtain the attention neural network process model for the current iteration number.

[0054] The first global latent variable distribution is obtained by inputting the samples corresponding to the current iteration number in the sample sub-database into the hidden path encoder of the attention neural network process model at the current iteration number.

[0055] The second global latent variable distribution is obtained by inputting the samples corresponding to the current iteration number in the sample database into the hidden path encoder of the attention neural network process model at the current iteration number.

[0056] The intermediate representation is obtained by inputting the samples corresponding to the current iteration number in the sample sub-database, the positions of each sensor on the bottom of the sample surface aircraft corresponding to the current iteration number in the sample database, the hull type parameters of the sample surface aircraft corresponding to the current iteration number, and the operating condition parameters of the sample surface aircraft corresponding to the current iteration number into the deterministic path encoder of the attention neural network process model at the current iteration number.

[0057] The sampled values ​​of the first global latent variable distribution are obtained based on the mean and variance of the first global latent variable distribution. Specifically, the first global latent variable distribution is a normal distribution Normal(mean, std), where mean is the mean and std is the variance. First, random sampling is performed on the standard normal distribution N(0,1) to obtain sampled values, and then (mean + std * sampled value) is output to obtain the sampled values ​​of the first global latent variable distribution.

[0058] The intermediate representation, the first global latent variable distribution sample value, the positions of each sensor on the bottom of the sample surface aircraft corresponding to the current iteration number in the sample database, the hull type parameters of the sample surface aircraft corresponding to the current iteration number, and the operating condition parameters of the sample surface aircraft corresponding to the current iteration number are input into the conditional decoder of the attention neural network process model at the current iteration number to obtain the predicted load value distribution.

[0059] The loss function value is obtained based on the first global latent variable distribution, the second global latent variable distribution, the predicted load value distribution, and the load values ​​measured by each sensor on the bottom of the sample surface aircraft corresponding to the current iteration number in the sample database. Specifically, the loss function includes the KL divergence between the first and second global latent variable distributions and the negative of the probability that the sensor measured load value conforms to the predicted load value distribution.

[0060] If the loss function value reaches the set condition, then the attention neural network process model for the current iteration number is determined to be a holographic water load reconstruction model.

[0061] If the loss function value reaches the set condition, the iteration count is updated and the next iteration begins.

[0062] In practical applications, the expected precision information is the total number of grids in the holographic grid unit 3D model corresponding to the target surface aircraft.

[0063] This invention provides embodiments for using the above method in pool tests, which can be applied to, for example... Figure 2 In the hardware system shown, Figure 2The hardware system shown stores the original data from the water tank experiment, ship type and operating condition parameters (ship type parameters and operating condition parameters), expected precision information, and holographic water load on the client side. The server includes a training sample database and is also used to train the attention neural process model, generate holographic mesh cells with expected precision, and reconstruct the holographic water load on the ship's bottom. The specific working process is as follows: Figure 6 As shown, it includes:

[0064] S1. Receive the raw data and precision expectation information (precision expectation value) of the water tank test sent by the client.

[0065] S2. Construct a training sample database using the original data from the water tank test.

[0066] S3. Use the training sample database to train the attention neural process model.

[0067] S4. Construct a holographic mesh 3D model of the surface vehicle to be reconstructed (i.e., the target surface vehicle in the previous embodiment) based on the expected refinement information. Specifically, using the geometry of the surface of the bottom of the surface vehicle to be reconstructed as the boundary, generate a holographic mesh 3D model that matches the expected refinement information.

[0068] S5. Using the trained attention neural process model (holographic water load reconstruction model) and the holographic mesh unit three-dimensional model, the holographic water load on the bottom of the surface aircraft to be reconstructed is reconstructed, and the processing result is sent to the client.

[0069] In practical applications, the client collects load values ​​on the bottom surface of the water-based aircraft model using sensors placed at designated locations on the model's hull. These designated locations are typical sensor placements used in pool tests, such as... Figure 3 As shown.

[0070] In practical applications, the raw data from a water tank test specifically refers to the load values ​​collected by sensors placed at multiple locations on the bottom surface of a surface aircraft model during a water tank test of a certain ship type under specific operating conditions. For each sensor, the raw attribute data is in the form of (x, y, z, α, A / F, λ, v, C). p ), where x, y, z represent the sensor's three-dimensional coordinates in space, α represents the climb angle, A / F represents the aft body length / forward body length, λ represents the wavelength, v represents the speed, and C p This represents the time-series load value. The hull parameters include the heave angle and aft / fore hull length, while the operating condition parameters include wavelength and speed. These, along with the sensor's three-dimensional spatial coordinates, constitute the input data for the attention neural process model. Therefore, the raw data corresponding to a single pool test consists of multiple (x, y, z, α, A / F, λ, v, C) values. pVectors, such as Figure 3 As shown, the raw data under this sensor arrangement is a 1486×(7+1800) matrix, where 1486 represents the number of sensors, 7 represents the dimension of the input (x,y,z,α,A / F,λ,v), and 1800 represents the time-series load value C. p The number of time frames. This invention can use raw data under multiple ship type operating condition parameter combinations to construct a training sample database. Therefore, the raw data of the pool test provided by the client is in the form of a (N×1486)×(7+1800) matrix, where N is the number of ship type operating condition groups, 1486 is the number of sensors in each ship type operating condition group, 7 is the dimension of the input (x,y,z,α,A / F,λ,v), and 1800 is the time-series load value C. p The number of time frames.

[0071] In practical applications, the expected refinement information refers to the total number of holographic mesh cells suitable for holographic pressure coefficient reconstruction, i.e., the total number of meshes in the 3D holographic mesh cell model corresponding to the surface aircraft to be reconstructed. The holographic mesh is generated by mesh design software, and the mesh construction module generates a holographic mesh with the corresponding number of cells based on the expected refinement information. Therefore, the client inputs the expected refinement information in the format of (number of mesh cells). Based on this indicator, the server can output holographic water loads with different levels of refinement. Figure 4 As shown, the expected number of mesh elements is set to 22118, and the mesh building module generates a three-dimensional mesh that conforms to the bottom surface of the surface vehicle to be reconstructed.

[0072] In practical applications, methods for placing sensors at designated locations on the bottom surface of a surface aircraft model include:

[0073] Cross sections are selected at intervals from bow to stern, and sensors are arranged at intervals on each selected section. At least 80 cross sections are selected, and the number of sections in areas with drastic load changes near the fault steps should be greater than in other areas. That is, within each cross section, the density of sensors near the fault steps is greater than in other locations within the section. In this embodiment, the selected cross sections should cover as much of the bottom of the water-based aircraft model as possible. Figure 3 The 81 cross-sections are evenly distributed across the entire hull. For the arrangement of sensors on individual cross-sections, the mid-section and bilge areas should be denser than other locations.

[0074] In the preferred scheme, 81 cross-sections are selected from the bow to the stern to arrange sensors, with a total of 1486 sensors. Among them, the number of sensors is increased near the step break, with more cross-sections selected than in other areas. Figure 3 The image shows the arrangement of sensors during a water tank test.

[0075] The process of constructing a training sample database using raw data from the pool test includes: acquiring the spatial coordinates, temporal load values, and ship type operating condition parameters of each sensor arranged on the bottom surface of the water-based aircraft model, and using this as the raw sensor data. Treating the load data of each time frame as a dataset and the load distribution pattern of each time frame as an objective function, the temporal load data can be considered a family of datasets, and the load distribution patterns of all time frames constitute a family of objective functions. Therefore, the original data matrix of form (N×1486)×(7+1800) is processed into a matrix of form 1800×(N×1486)×8, where 1800 represents the number of time frames, N represents the number of ship type operating condition groups, and 8 represents the dimensions of the three-dimensional spatial coordinates, two-dimensional ship type parameters, two-dimensional operating condition parameters, and one-dimensional single-time-frame load values.

[0076] In practical applications, training an attention neural process model using a training sample database involves: using the sensor's three-dimensional spatial coordinates, ship type parameters, and operating condition parameters as inputs to the attention neural process model, and using the corresponding water tank test load measurements as labels to obtain a data matrix for model training. The attention neural process model structure includes a deterministic path encoder, a hidden path encoder, and a conditional decoder. The deterministic path encoder consists of a fully connected module, a self-attention module, and a cross-attention module; the hidden path encoder consists of a fully connected module, a self-attention module, and an averaging and aggregation module; and the conditional decoder consists of a fully connected module. Specifically, training the attention neural process model involves using seven dimensions of features—three-dimensional spatial coordinates, two-dimensional ship type parameters, and two-dimensional operating condition parameters—as model input (x, y, z, α, A / F, λ, v), and a one-dimensional target—the single-time-frame load value—as model output (C). p1 Therefore, the input data is a matrix of 1800×(N×1486)×7, and the output is a matrix of 1800×(N×1486)×1. The water load on the hull of the surface aircraft is a time-varying physical quantity, and different regions exhibit varying degrees of mutual influence. This mutual influence is referred to as spatial dependency between nodes. This invention models an attention neural process model, processing the time-series load data into multiple datasets and synchronously learning the load variation patterns across all time frames. Specifically, a cross-attention module captures the mutual influence strength of contextual sensor data to obtain richer contextual information, and a cross-attention module captures the spatial dependency strength between target region nodes and contextual sensor data to obtain more relevant contextual sensor data information. The attention neural process model is implemented in a computer language, and the training results are stored on a server for subsequent use.

[0077] During the training phase, m samples are selected from n samples to form a set. All samples constitute a set (x,y) i Let x represent the location of the sensor, ship type parameters, and operating condition parameters, and y represent the label; the goal is to minimize the loss function (even if the conditional probability p(y) is less than the value of the i-th sample in the set). i |z,r,x i The model is trained with the goal of increasing KL divergence and decreasing KL divergence. The loss function used for model training is: Where KL(·||·) is the KL divergence, q(z|x 1:n ,y 1:n ) represents the second global latent variable distribution, q(z|x) 1:m ,y 1:m (x) represents the first global latent variable distribution, which implicitly represents the distribution pattern of all sample data. Both distributions are normal distributions, and their variances and means are obtained from two neurons in the output layer of the fully connected module of the hidden path encoder. 1:m ,y 1:m (x) represents the m samples taken, and (x) represents the m samples taken. 1:n ,y 1:n ) represents all samples, x i Let z be the sensor coordinates, ship type parameters, and operating condition parameters of the i-th sample among all samples, z be the sampled value of the first global latent variable distribution, which has global characteristics, and r be the intermediate representation obtained by the deterministic path encoder; p(y i |z,r,x i This is the actual load (the load value y measured by the i-th sensor on the bottom of the sample surface aircraft in the sample database). i The probability of occurrence in the predicted load distribution is obtained from the predicted load value distribution and the load value measured by the i-th sensor on the bottom of the sample surface aircraft in the sample database.

[0078] In the preferred embodiment, the deterministic path encoder has a five-layer fully connected module with implicit output dimensions of 64, 256, 128, 64, and 32, respectively; the self-attention module uses an 8-head attention mechanism; and the cross-attention module uses an 8-head attention mechanism. The hidden path encoder also has a five-layer fully connected module with implicit output dimensions of 64, 256, 128, 64, and 32, respectively; the self-attention module uses an 8-head attention mechanism; and the averaging module performs a mean-taking operation. The conditional decoder has a six-layer fully connected module with implicit output dimensions of 64, 256, 512, 256, 64, and 1, respectively. Figure 5The diagram shows the structure of the attention neural process model. The data flow of this model is as follows: During the training phase, the complete data is divided into context nodes and target nodes. The context data is encoded using deterministic path encoding to obtain fine-grained intermediate representations, and then encoded using hidden path encoding to obtain global context global latent variables. The target node is encoded using hidden paths to obtain global target global latent variables. The joint vector of the target input, intermediate representations, and target global latent variables is encoded using a conditional encoder to obtain the target output. The loss function uses KL(·||·) to align the context global latent variables and the target global latent variables. During the generation phase, the context data is encoded using deterministic path encoding to obtain fine-grained intermediate representations, and the context data is encoded using hidden path encoding to obtain global latent variables. The joint vector of the target input, intermediate representations, and global latent variables is encoded using a conditional encoder to obtain the target output.

[0079] Then, the holographic water load on the hull of the surface aircraft is reconstructed using a trained attention neural process model and holographic mesh units. Specifically: Figure 6 As shown, the server loads a pre-trained attentional neural process model and corresponding sensor data as context data. It takes the coordinates of the center point of each grid cell, ship type parameters, and operating condition parameters as input, and outputs the load values ​​of all grid cells. Taking a precision expectation of 22118 as an example, this invention first automatically generates holographic grid cells based on this precision expectation information. Then, it combines the three-dimensional spatial coordinates of each grid center with the ship type parameters and operating condition parameters input by the client into a 7-dimensional vector, i.e., (x, y, z, α, A / F, λ, v). Next, the server loads the pre-trained attentional neural process model locally and predicts the holographic grid cell input to obtain the holographic water load, which is a 1800×22118×1 matrix.

[0080] The method provided in this invention is used for holographic water load reconstruction of the bottom of a surface aircraft in a water tank test. It involves sensor placement technology in the water tank test. It can reconstruct a reasonable holographic water load based on real data from the water tank test, reducing the dependence on CFD simulation calculations, reducing the time for holographic water load reconstruction, reducing the amount of calculation, and improving the utilization rate of water tank test data.

[0081] This embodiment also provides a holographic water load reconstruction device for the hull of a surface aircraft. Specifically, the reconstruction device can run on a laboratory server, such as... Figure 7 As shown, the reconstruction device includes:

[0082] The data receiving module is used to receive raw data from the pool test, ship type parameters, operating condition parameters, and precision expectation information sent by the client.

[0083] The sample processing module is used to construct a training sample database using the raw data from the pool test.

[0084] The training module is used to train an attention neural process model using the training sample database.

[0085] The mesh construction module is used to construct a holographic mesh unit 3D model of the surface aircraft to be reconstructed (i.e., the target surface aircraft in the previous embodiment) based on the expected level of detail.

[0086] The data processing module is used to reconstruct the holographic water load on the bottom of the surface aircraft using the trained attention neural process model and the holographic mesh unit 3D model.

[0087] The data sending module is used to send the processing results to the client.

[0088] This invention provides a method and apparatus for reconstructing holographic water loads on the hull of a surface vehicle. Using raw data from a water tank test provided by the client and the expected precision of the reconstructed holographic water load, the method performs modeling and analysis of the holographic water load on the server side. First, the raw data from the water tank test is acquired, a training sample database is constructed, and an attentional neural process model is trained. Based on the client's expected precision, holographic mesh units suitable for holographic water load reconstruction are constructed. The trained model is then used to reconstruct the holographic water load on the hull of the surface vehicle. Finally, the holographic water load that meets the precision requirements is returned to the client for selection. This invention eliminates the need for computationally expensive CFD simulations to acquire holographic data; it reconstructs reasonable holographic data solely based on real data from the water tank test, reducing computational load, dependence on simulation calculations, and reconstruction time, while improving the utilization rate of water tank test data. Furthermore, the attentional neural process model involved in this invention is easy to implement, and once trained, its efficiency in reconstructing holographic data far exceeds that of CFD simulation calculations. For example, a comparative experiment on acquiring holographic water loads on the hull of a certain type of surface vehicle can be conducted while maintaining consistent computing equipment. The results show that acquiring 1800 frames of temporal holographic water loads and calculating holographic data using CFD takes 36 hours, while predicting holographic data using the model trained in this invention takes no more than 5 minutes.

[0089] In one embodiment, a computer system is also provided, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps in the above method embodiments, and its internal structure diagram can be as follows. Figure 8As shown in the diagram, the computer system also includes input / output interfaces (I / O) and communication interfaces. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores pending transactions. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When the computer program is executed by the processor, it implements a data processing method.

[0090] It should be noted that the object information (including but not limited to object device information, object personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the object or fully authorized by all parties, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0091] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0092] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0093] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for reconstructing holographic water loads on the hull of a surface aircraft, characterized in that, include: Construct a holographic mesh 3D model of the target surface aircraft based on the preset precision expectation information; The target surface aircraft's hull parameters, operating condition parameters, and center point coordinates of each grid in the holographic mesh unit 3D model corresponding to the target surface aircraft are input into the holographic water load reconstruction model to obtain the target surface aircraft's hull holographic water load. The hull parameters include the lift angle and aft / fore hull length, and the operating condition parameters include wavelength and speed. The process of determining the holographic water load reconstruction model includes: Obtain a sample database; the sample database includes multiple samples, each sample including raw data from the pool experiment of the sample surface aircraft; the raw data from the pool experiment of the sample surface aircraft includes the load values ​​measured by each sensor on the bottom of the sample surface aircraft, the position of each sensor, the hull parameters of the sample surface aircraft, and the operating condition parameters of the sample surface aircraft; Using the positions of sensors on the hull of each sample surface aircraft in the sample database, the hull type parameters of each sample surface aircraft, and the operating condition parameters of each sample surface aircraft as inputs, and the load values ​​measured by each sensor on the hull of each sample surface aircraft in the sample database as outputs, an attention neural network process model is trained to obtain a holographic water load reconstruction model. The attention neural network process model includes: a deterministic path encoder, a hidden path encoder, and a conditional decoder. The conditional decoder includes a first input terminal and a second input terminal. The input terminals of the deterministic path encoder and the hidden path encoder are connected. The output terminals of the deterministic path encoder and the hidden path encoder are both connected to the first input terminal of the conditional decoder. The input terminals of the deterministic path encoder and the hidden path encoder serve as the first input terminal of the attention neural network process model. The second input terminal of the conditional decoder serves as the second input terminal of the attention neural network process model. The deterministic path encoder includes a fully connected module, a self-attention module, and a cross-attention module connected in sequence. The hidden path encoder includes a fully connected module, a self-attention module, and an average aggregation module connected in sequence. The conditional decoder includes six fully connected layers connected in sequence.

2. The method for reconstructing the holographic water load on the bottom of a surface aircraft according to claim 1, characterized in that, Using the positions of sensors on the hull of each sample surface aircraft in the sample database, the hull type parameters of each sample surface aircraft, and the operating condition parameters of each sample surface aircraft as inputs, and the load values ​​measured by the sensors on the hull of each sample surface aircraft in the sample database as outputs, an attention neural network process model is trained to obtain a holographic water load reconstruction model, specifically including: A sample sub-database is obtained by selecting a set number of samples from the sample database. In the current iteration, the attention neural network process model in the previous iteration is tuned to obtain the attention neural network process model in the current iteration. The first global latent variable distribution is obtained by inputting the samples corresponding to the current iteration number in the sample sub-database into the hidden path encoder of the attention neural network process model at the current iteration number. The sample corresponding to the current iteration number in the sample database is input into the hidden path encoder of the attention neural network process model at the current iteration number to obtain the second global latent variable distribution; The intermediate representation is obtained by inputting the samples corresponding to the current iteration number in the sample sub-database, the positions of each sensor on the bottom of the sample surface aircraft corresponding to the current iteration number in the sample database, the ship type parameters of the sample surface aircraft corresponding to the current iteration number, and the operating condition parameters of the sample surface aircraft corresponding to the current iteration number into the deterministic path encoder of the attention neural network process model at the current iteration number. The sampled values ​​of the first global latent variable distribution are obtained based on the mean and variance of the first global latent variable distribution; The intermediate representation, the first global latent variable distribution sample value, the position of each sensor on the bottom of the sample surface aircraft corresponding to the current iteration number in the sample database, the hull type parameters of the sample surface aircraft corresponding to the current iteration number, and the operating condition parameters of the sample surface aircraft corresponding to the current iteration number are input into the conditional decoder of the attention neural network process model at the current iteration number to obtain the predicted load value distribution. The loss function value is obtained based on the first global latent variable distribution, the second global latent variable distribution, the predicted load value distribution, and the load values ​​measured by each sensor on the bottom of the sample surface aircraft corresponding to the current iteration number in the sample database. If the loss function value reaches the set condition, then the attention neural network process model at the current iteration number is determined to be the holographic water load reconstruction model; If the loss function value reaches the set condition, the iteration count is updated and the next iteration begins.

3. The method for reconstructing the holographic water load on the bottom of a surface aircraft according to claim 1, characterized in that, The expected level of detail is the total number of grids in the holographic grid unit 3D model corresponding to the target surface aircraft.

4. A computer system, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the steps of the method according to any one of claims 1-3.

Images