Method and device for optimizing head shape and water entry strategy of water-air cross-medium robot

Abstract

The invention provides a water-air cross-medium robot head shape and water entry strategy optimization method and device. The method comprises the steps that a design domain is determined; performing Latin hypercube sampling in a design domain to obtain a design parameter vector; an NURBS curve is drawn, and a water-air cross-medium robot model is formed; resetting the parameters; a simulation calculation domain is generated and gridding is carried out; performing finite element solution on the current grid, and extracting solution data; constructing a low-quality data set and a high-quality data set; obtaining an optimal hyper-parameter of the proxy model through Bayesian optimization; training a proxy model based on the optimal hyper-parameter; constructing an optimization target, and adopting a multi-target optimization algorithm to obtain a solution meeting bionic form constraint and component constraint under the optimization target; and judging whether the agent model converges or not, and outputting optimal design parameters. According to the method, the full-process automatic design normal form of parametric modeling-numerical simulation-parameter optimization is constructed, so that manual real-time intervention in the middle process is avoided, and the process efficiency is greatly improved.

Description

Technical Field

[0001] The present invention relates to the field of water-air cross-medium robot design, and in particular to a method and device for optimizing the head shape and water entry strategy of a water-air cross-medium robot. Background Art

[0002] The water-air cross-media robot is a new type of intelligent device capable of freely switching between air and water. The enormous impact loads it faces during water entry can easily damage the robot, thus presenting a key bottleneck hindering the practical application of this technology. Existing research indicates that head shape and entry strategy are key factors determining the impact loads. Therefore, optimization of the head shape and entry strategy of the water-air cross-media robot is necessary to minimize these impact loads.

[0003] Existing technologies mostly design head shape based on empirical formulas or single working condition simulations. Because empirical formulas and single working condition simulations rely on simplified assumptions under specific conditions, and actual water entry involves the coupling of multiple variables, resulting in insufficient model generalization ability, the optimization results are not effective in actual applications and are difficult to adapt to complex and changeable water entry conditions.

[0004] Existing water entry strategies typically utilize CFD simulation technology. This involves building a dataset using design parameters and load data predicted by CFD simulation. Models such as radial basis function neural networks (RBFNs) and dense neural networks (DNNs) are then trained to predict the water entry impact loads of water-air-crossing media robots with different head shapes. However, existing technologies require real-time manual intervention during data sampling and optimization, resulting in inefficiencies. Furthermore, the model's accuracy is limited, making efficient searches in high-dimensional design spaces impossible. Summary of the Invention

[0005] In response to the defects of the existing technology, the present invention provides a method and device for optimizing the head shape and water entry strategy of a water-air cross-medium robot.

[0006] To achieve the above object, the technical solution adopted by the present invention is as follows: In one aspect, the present invention provides a method for optimizing the head shape and water entry strategy of a water-air cross-medium robot, comprising the following steps: Determine the design domain based on bionic morphological constraints, component constraints, and motion constraints; Perform Latin hypercube sampling in the design domain to obtain the design parameter vector, and determine whether the design parameter vector meets the bionic morphological constraints and component constraints. If so, add the design parameter vector to the sample set; if not, resample. If the number of iterative sampling reaches a threshold, redefine the design domain. Draw NURBS curves based on the design parameter vectors in the current sample set to describe the head shape; generate a three-dimensional head model based on the NURBS curves, and complete component assembly to form a water-air trans-medium robot model; Reset the model, position, attitude, and motion parameters of the current water-air trans-medium robot; Automatically check and repair the surface mesh problems of the imported model, generate a simulation calculation domain and perform meshing. Among them, the pool and air adopt the preset mesh accuracy; the water surface, path, water-air trans-medium robot and its surroundings are meshed with refinement based on the preset mesh accuracy; Perform finite element solution on the current mesh, extract the solution data, and the solution data includes low-precision solution data and high-precision solution data; Based on the current low-precision solution data and the design parameter vectors in the current sample set, construct a low-quality data set; used to preliminarily screen the design parameter vectors in the sample set; Perform finite element solution on the screened design parameter vectors using a high-precision mesh to obtain high-precision solution data. Based on the screened design parameter vectors and high-precision solution data, construct a high-quality data set for training the surrogate model; Based on the high-quality data set, obtain the optimal hyperparameters of the surrogate model through Bayesian optimization; Train the surrogate model based on the optimal hyperparameters; Construct an optimization objective, use a multi-objective optimization algorithm to obtain the solutions that meet the bionic morphology constraints and component constraints under the optimization objective, and screen the solutions with large variances as labeled samples; Judge whether the surrogate model converges. If it converges, output the optimal design parameters. If it does not converge, continue to label samples and expand the high-quality data set, and retrain the surrogate model until the surrogate model converges.

[0007] Furthermore, the design domain is an eight-dimensional space, including the water entry strategy parameters and the nose shape parameters. The water entry strategy parameters include the water entry speed and the water entry angle. The nose shape parameters include the beak length, beak thickness, horizontal shape of the front end of the head, vertical shape of the front end of the head, horizontal shape of the rear end of the head, and vertical shape of the rear end of the head.

[0008] Furthermore, the design domain that satisfies the bionic morphology constraints and component constraints is calculated according to the following formula: Component constraints: ; Among them, is the feasible design domain; is the design domain; is the component constraint function; is the nose shape parameter z Under the condition ofX The function value at Bionic form constraint: ; ; ; ; wherein, is the bionic form constraint function; l is the head length.

[0009] Furthermore, it also includes determining the number of high-precision grids: Starting from the number of grids corresponding to the preset grid accuracy, gradually increase the number of grids. According to the impact load change curve in the solution data, judge the grid accuracy. As the number of grids increases, if the curve changes, continue to increase the number of grids; if the change amplitude of the curve is less than the threshold, take the number of grids before the increase as the number of grids of the high-precision grid.

[0010] Furthermore, the overlapping grid technology is adopted during the meshing process.

[0011] Furthermore, the surrogate model is a deep kernel learning model.

[0012] Furthermore, the optimal hyperparameters of the surrogate model are obtained according to the following steps: Construct the Bayesian hyperparameter optimization objective function: ; ; wherein, is the surrogate model; is the search space; n is the number of hyperparameters; is the hyperparameter combination; is the optimal hyperparameter; is the loss of the surrogate model to be trained under the hyperparameter combination in the validation set; Randomly extract from the search space k groups of initial hyperparameter combinations , calculate , and form the initial data set ; Use the acquisition function to iteratively calculate the hyperparameter combinations that satisfy the Bayesian hyperparameter optimization objective function to obtain the optimal hyperparameters of the surrogate model: ; ; ; Among them, is the mean value; K Cov(·) is the covariance kernel function; is the currently known optimal objective function value; is the data set; , T is the number of iterations.

[0013] Furthermore, in the multi-objective optimization algorithm, the optimization objectives are: ; Among them, is the mean value of the axial impact load output by the surrogate model; is the mean value of the normal impact load output by the surrogate model; is the entry velocity; The solutions that satisfy the bionic morphology constraint and the component constraint are obtained according to the following formula: ; Among them, is the solution that satisfies the bionic morphology constraint and the component constraint; is the solution obtained by the multi-objective optimization algorithm.

[0014] Furthermore, judge whether the surrogate model converges according to the following steps: Calculate the percentage absolute error between the solution data and the output result of the surrogate model. If the percentage absolute error is less than 1%, it is considered that the surrogate model converges; The percentage absolute error is calculated according to the following formula: ; ; Among them, n is the number of labeled samples; is the percentage absolute error of the axial impact load; is the percentage absolute error of the normal impact load; is the peak value of the axial impact load of the i th labeled sample in the solution data; is the peak value of the axial impact load of the i th labeled sample output by the surrogate model; is the peak value of the normal impact load of the i th labeled sample in the solution data; is the peak value of the normal impact load of the i th labeled sample output by the surrogate model.

[0015] On the other hand, the present invention provides an optimization device for the head shape and water entry strategy of an underwater-air cross-medium robot, including: The first module is used to determine the design domain based on bionic form constraints, component constraints, and motion constraints; The second module is used to perform Latin hypercube sampling within the design domain to obtain a design parameter vector, and determine whether the design parameter vector satisfies the bionic form constraints and component constraints. If it satisfies, the design parameter vector is added to the sample set; if not, resampling is performed. If the number of iterative sampling times reaches the threshold, the design domain is determined again; The third module is used to draw NURBS curves based on the design parameter vectors in the current sample set to describe the head shape; generate a three-dimensional head model based on the NURBS curves, and complete component assembly to form a water-air cross-medium robot model; The fourth module is used to reset the model, position, attitude, and motion parameters of the current water-air cross-medium robot; The fifth module is used to automatically check and repair the surface mesh problems of the imported model, generate a simulation calculation domain and perform meshing. Among them, the pool and air adopt a preset mesh accuracy; the water surface, path, water-air cross-medium robot and its surrounding areas are meshed in detail on the basis of the preset mesh accuracy; The sixth module is used to perform finite element solution on the current mesh and extract the solution data. The solution data includes low-precision solution data and high-precision solution data; The seventh module is used to construct a low-quality data set based on the current low-precision solution data and the design parameter vectors in the current sample set; and is used to preliminarily screen the design parameter vectors in the sample set; The eighth module is used to perform finite element solution on the screened design parameter vectors using a high-precision mesh to obtain high-precision solution data, and construct a high-quality data set based on the screened design parameter vectors and the high-precision solution data for training the surrogate model; The ninth module is used to obtain the optimal hyperparameters of the surrogate model through Bayesian optimization based on the high-quality data set; The tenth module is used to train the surrogate model based on the optimal hyperparameters; The eleventh module is used to construct an optimization objective, and use a multi-objective optimization algorithm to obtain solutions that satisfy the bionic form constraints and component constraints under the optimization objective, and screen out solutions with large variances as labeled samples; The twelfth module is used to determine whether the surrogate model converges. If it converges, the optimal design parameters are output; if not, continue to label samples and expand the high-quality data set, and retrain the surrogate model until the surrogate model converges.

[0016] Compared with the prior art, the beneficial technical effects of the present invention are as follows: The method and device for optimizing the head shape and water entry strategy of an air-water cross-medium robot provided by the present invention determine the design domain by integrating bionic shape constraints, component constraints, and motion constraints, and use NURBS curves to describe the head shape, enabling the head shape designed by the present invention to have diversity and be optimized based on the bionic shape, thereby achieving a good load reduction effect.

[0017] The present invention preliminarily screens the design parameter vectors in the sample set by constructing a low-quality data set, judges the approximate influence of different design parameters on the impact load, and excludes most non-critical regions through screening, reducing the number of high-precision simulations and the simulation time. At the same time, by judging whether the surrogate model converges, when it does not converge, the high-quality data set is expanded, thereby introducing an active learning strategy to specifically explore the critical regions, realizing an efficient search for the high-dimensional design space and improving the model prediction accuracy in the critical regions. On the other hand, the present invention constructs a full-process automated design paradigm of parametric modeling - numerical simulation - parameter optimization, thereby avoiding manual real-time intervention in the intermediate process and greatly improving the process efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on the structures shown in these drawings.

[0019] Figure 1 Schematic flow chart of the method for optimizing the head shape and water entry strategy of an air-water cross-medium robot provided in an embodiment; Figure 2 Relationship diagram between the control point positions and the head shape provided in an embodiment; Figure 3 Schematic diagram of the simulation calculation domain division provided in an embodiment; Figure 4 Schematic diagram of the meshed simulation calculation domain provided in an embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0020] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some of the embodiments of the present invention, rather than all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0021] Refer to Figure 1, an embodiment provides a method for optimizing the head shape and water entry strategy of a water-air cross-medium robot, including the following steps: Determine the design domain based on bionic form constraints, component constraints, and motion constraints; Perform Latin hypercube sampling within the design domain to obtain design parameter vectors, and determine whether the design parameter vectors meet the bionic form constraints and component constraints. If they meet, add the design parameter vectors to the sample set; if not, resample. If the number of iterative sampling times reaches the threshold, re-determine the design domain; Draw NURBS curves based on the design parameter vectors in the current sample set to describe the head shape; generate a three-dimensional head model based on the NURBS curves, and complete component assembly to form a water-air cross-medium robot model; Reset the model, position, attitude, and motion parameters of the current water-air cross-medium robot; Automatically check and repair the surface mesh problems of the imported model, generate a simulation calculation domain and perform meshing. Among them, the pool and air adopt a preset mesh accuracy; the water surface, path, water-air cross-medium robot and its surroundings are meshed in detail on the basis of the preset mesh accuracy; Perform finite element solution on the current mesh, extract the solution data, and the solution data includes low-precision solution data and high-precision solution data; Based on the current low-precision solution data and the design parameter vectors in the current sample set, construct a low-quality data set; used for preliminary screening of the design parameter vectors in the sample set; Perform finite element solution on the screened design parameter vectors using a high-precision mesh to obtain high-precision solution data. Based on the screened design parameter vectors and the high-precision solution data, construct a high-quality data set for training the surrogate model; Based on the high-quality data set, obtain the optimal hyperparameters of the surrogate model through Bayesian optimization; Train the surrogate model based on the optimal hyperparameters; Construct an optimization objective, use a multi-objective optimization algorithm to obtain the solutions that meet the bionic form constraints and component constraints under the optimization objective, and screen the solutions with large variances as labeled samples; Determine whether the surrogate model converges. If it converges, output the optimal design parameters; if not, continue to label samples and expand the high-quality data set, and retrain the surrogate model until the surrogate model converges.

[0022] In one embodiment, the design domain is an eight-dimensional space, including water entry strategy parameters and nose shape parameters. The water entry strategy parameters include water entry speed and water entry angle, and the nose shape parameters include beak length, beak thickness, horizontal shape at the front of the head, vertical shape at the front of the head, horizontal shape at the back of the head, and vertical shape at the back of the head. Specifically, referring to Figure 2, the non-uniform rational B-spline curve (NURBS curve) is used to segmentally represent the head shape of the kingfisher. Four control points are adopted to control the nose shape. Two of the control points respectively control the beak length and the beak thickness. Both of these two control points have one degree of freedom of movement. The first control point can move in the horizontal direction to change the beak length, and the second control point can move in the vertical direction to change the beak thickness. The other two control points control the head shape. Both of these two control points have two degrees of freedom of movement and can move in the horizontal and vertical directions to respectively control the shapes of the front and rear ends of the head, that is, the nose shape parameters are formed. Since the water entry impact load is related to the water entry strategy, the two water entry strategy parameters of the water entry speed and the water entry angle are also considered at the same time.

[0023] The motion constraint is the water entry strategy constraint determined by the actual motion ability and the lower limit of the motion index of the water-air cross-medium robot. The motion constraint is directly reflected in the upper and lower limits of the water entry strategy parameters.

[0024] The design domain that satisfies the bionic shape constraint and the component constraint is calculated according to the following formula: Component constraint: ; Among them, is the feasible design domain, and the feasible design domain is the design domain that satisfies the component constraint function; is the design domain; is the component constraint function; is the nose shape parameter z The function value of the NURBS curve at X under the condition; Bionic shape constraint: ; ; ; ; Among them, is the bionic shape constraint function; l is the head length.

[0025] In one embodiment, the design domain is shown in Table 1: Table 1 Design domain parameter range

[0026] Draw a NURBS curve based on the design parameter vector in the current sample set, discretize and sample the NURBS curve to obtain a set of coordinate points of the spline curve, generate a spline curve according to the set of coordinate points of the spline curve, and further generate a three-dimensional head model. Then assemble the generated three-dimensional head model with the prepared fuselage model to complete the component assembly and form a water-air trans-medium robot model.

[0027] Reset the model, position, attitude, and motion parameters of the current water-air trans-medium robot to adapt to the entry angle and entry speed.

[0028] Refer to Figure 3 , in one embodiment, the size of the simulation design domain is 9m×8m×9m, including a pool, water surface, water-air trans-medium robot, water-air trans-medium robot cube, and path. The water-air trans-medium robot cube completely wraps the water-air trans-medium robot; the path is set along the motion direction of the water-air trans-medium robot.

[0029] Mesh the simulation calculation domain. Among them, the pool and air adopt a preset grid accuracy; the water surface, path, water-air trans-medium robot and its surroundings are meshed with refinement on the basis of the preset grid accuracy. The meshed simulation calculation domain is as Figure 4 shown. It can be seen from the figure that the density of the grids on the water surface, path, water-air trans-medium robot and its surroundings increases significantly after refinement, which can improve the solution accuracy.

[0030] Before generating the design domain and meshing, automatically check and repair the surface grid problems of the imported model to ensure the feasibility of mesh generation.

[0031] In a preferred embodiment, the overlapping grid technology is used for meshing. The overlapping grid technology can effectively handle complex structures with relative motion without grid deformation or regeneration, so that it can adaptively refine the grid to refine or coarsen the grids on both sides of the overlapping grid interface, improving the calculation efficiency while ensuring the calculation accuracy.

[0032] In a preferred embodiment, the determination of the number of high-precision grids is also included in the meshing process: Gradually increase the number of grids starting from the number of grids corresponding to the preset grid accuracy, and judge the grid accuracy according to the impact load change curve in the solution data. As the number of grids increases, if the curve changes, continue to increase the number of grids; if the change amplitude of the curve is less than the threshold, take the number of grids before the increase as the number of grids of the high-precision grid.

[0033] By determining the number of high-precision grids as described above, the feasibility of mesh generation can be ensured.

[0034] Perform finite element solution on the current mesh, extract the solution data, which includes low-precision solution data and high-precision solution data; the peak impact load obtained from low-precision mesh simulation is used as the true value in the low-quality dataset, and the peak impact load obtained from high-precision mesh simulation is used as the true value in the high-quality dataset. Before solving, it is necessary to clear the previous solution data and then perform the solution.

[0035] Construct a low-quality dataset based on the current low-precision solution data and the design parameter vectors in the current sample set; it is used to preliminarily screen the design parameter vectors in the sample set. In one embodiment, 500 groups of low-precision solution data obtained from the simulation of a mesh with a preset mesh accuracy (i.e., low-precision mesh) are used to construct a low-precision dataset. Samples in the sample set are screened through the low-precision dataset, and 200 candidate samples with small impact loads are selected. Then, the 200 candidate samples with small impact loads are simulated through a high-precision mesh to obtain high-precision solution data, construct a high-quality dataset, and then divide the high-quality dataset into a training set, a validation set, and a test set at a ratio of 6:2:2, so as to be used for training the surrogate model.

[0036] Screen the design parameter vectors through the low-quality dataset, and utilize the influence of different design parameter vectors reflected by it on the impact load, so as to exclude most non-critical regions, reduce the number of high-precision simulations, improve the processing efficiency, and save energy consumption.

[0037] In one embodiment, the surrogate model is a deep kernel learning model (DKL model). The DKL model uses Gaussian process regression, constructs a kernel function based on DNN (deep neural network) features, and outputs the prediction expectation and variance; the output expectation (i.e., the mean) is used as the estimated value of the impact load, and the variance is used as a measure of the output uncertainty.

[0038] Obtain the optimal hyperparameters of the surrogate model through Bayesian optimization. The optimal hyperparameters of the surrogate model are obtained according to the following steps: Construct a Bayesian hyperparameter optimization objective function: ; ; Among them, is the surrogate model; is the search space; n is the number of hyperparameters; is the hyperparameter combination; is the optimal hyperparameter; is the loss of the surrogate model to be trained under the hyperparameter combination on the validation set; Randomly extract from the search space k groups of initial hyperparameter combinations , calculate , and form an initial data set ; Use the acquisition function to iteratively calculate the hyperparameter combination that satisfies the Bayesian hyperparameter optimization objective function, and obtain the optimal hyperparameters of the surrogate model: ; ; ; Among them, is the mean value; K (·) is the covariance kernel function; is the currently known optimal objective function value; is the data set; , T is the number of loops.

[0039] In one embodiment, the NSGA-III multi-objective optimization algorithm is used to obtain the solution that satisfies the bionic morphology constraint and the component constraint under the optimization objective.

[0040] In the multi-objective optimization algorithm, the optimization objective is: ; Among them, is the mean value of the axial impact load output by the surrogate model; is the mean value of the normal impact load output by the surrogate model; is the water entry speed; by adding the water entry speed to the optimization objective, the water entry rapidity requirement of the water-air cross-medium robot is satisfied.

[0041] The solution that satisfies the bionic morphology constraint and the component constraint is obtained according to the following formula: ; Among them, is the solution that satisfies the bionic morphology constraint and the component constraint; is the solution obtained by the multi-objective optimization algorithm. Sort in descending order according to the variance , and select the samples corresponding to the first n solutions with the largest sum of variances as the labeled samples.

[0042] Judge whether the surrogate model converges, so as to judge whether the accuracy of the model in the key area is high enough. If it converges, output the optimal design parameters. If it does not converge, expand the high-quality data set, and retrain the surrogate model with the expanded high-quality data set until the surrogate model converges.

[0043] In one embodiment, judge whether the surrogate model converges according to the following steps: Calculate the percentage absolute error between the calculated solution data and the output result of the surrogate model. If the percentage absolute error is less than 1%, it is considered that the surrogate model has converged; if the surrogate model has not converged, then n annotated samples will be added to the high-quality dataset to expand the high-quality dataset.

[0044] The percentage absolute error is calculated according to the following formula: ; ; where n is the number of annotated samples; is the percentage absolute error of the axial impact load; is the percentage absolute error of the normal impact load; is the peak value of the axial impact load of the i th annotated sample in the solution data; is the peak value of the axial impact load of the i th annotated sample output by the surrogate model; is the peak value of the normal impact load of the i th annotated sample in the solution data; is the peak value of the normal impact load of the i th annotated sample output by the surrogate model.

[0045] In one embodiment, an optimization device for the head shape and water-entry strategy of a water-air cross-medium robot is provided, including: The first module is used to determine the design domain based on bionic shape constraints, component constraints, and motion constraints; The second module is used to perform Latin hypercube sampling within the design domain to obtain the design parameter vector, and determine whether the design parameter vector satisfies the bionic shape constraints and component constraints. If it is satisfied, the design parameter vector is added to the sample set. If it is not satisfied, resampling is performed. If the number of iterative sampling times reaches the threshold, the design domain is re-determined; The third module is used to draw NURBS curves based on the design parameter vectors in the current sample set to describe the head shape; generate a three-dimensional head model based on the NURBS curves, and complete component assembly to form a water-air cross-medium robot model; The fourth module is used to reset the model, position, attitude, and motion parameters of the current water-air cross-medium robot; The fifth module is used to automatically check and repair the surface mesh problems of the imported model, generate a simulation calculation domain and perform meshing, where the pool and air adopt a preset mesh accuracy; the water surface, path, water-air cross-medium robot and its surroundings are meshed with refinement based on the preset mesh accuracy; The sixth module is used to perform finite element solution on the current grid and extract the solution data, where the solution data includes low-precision solution data and high-precision solution data; The seventh module is used to construct a low-quality data set based on the current low-precision solution data and the design parameter vectors in the current sample set; and is used to preliminarily screen the design parameter vectors in the sample set; The eighth module is used to perform finite element solution on the screened design parameter vectors using a high-precision grid to obtain high-precision solution data, and construct a high-quality data set based on the screened design parameter vectors and the high-precision solution data for training the surrogate model; The ninth module is used to obtain the optimal hyperparameters of the surrogate model through Bayesian optimization based on the high-quality data set; The tenth module is used to train the surrogate model based on the optimal hyperparameters; The eleventh module is used to construct an optimization objective, adopt a multi-objective optimization algorithm to obtain solutions that satisfy the bionic form constraint and component constraint under the optimization objective, and screen the solutions with large variances as labeled samples; The twelfth module is used to determine whether the surrogate model converges. If it converges, the optimal design parameters are output. If it does not converge, continue to label samples and expand the high-quality data set, and retrain the surrogate model until the surrogate model converges.

[0046] Matters not described in this invention are well-known technologies.

[0047] The technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, 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, it should be considered as the scope described in this specification.

[0048] The above-described embodiments only represent several implementation manners of this application. Their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the invention. It should be noted that for those of ordinary skill in the art, without departing from the concept of this application, several deformations and improvements can still be made, and these all belong to the protection scope of this application.

[0049] The above is only the preferred embodiment of the present invention and is not used to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. Optimization method for the head shape and water entry strategy of an air-water cross-medium robot, characterized in that It includes the following steps: Determine the design domain based on bionic form constraints, component constraints, and motion constraints; Perform Latin hypercube sampling within the design domain to obtain design parameter vectors, and judge whether the design parameter vectors meet the bionic form constraints and component constraints. If they meet, add the design parameter vectors to the sample set. If they do not meet, resample. If the iterative sampling times reach the threshold, re-determine the design domain; Draw NURBS curves based on the design parameter vectors in the current sample set to describe the head shape; generate a three-dimensional head model based on the NURBS curves, and complete component assembly to form an underwater and aerial cross-medium robot model; Reset the model, position, attitude, and motion parameters of the current underwater and aerial cross-medium robot; Automatically check and repair the surface mesh problems of the imported model, generate a simulation calculation domain and perform meshing. Among them, the pool and air adopt the preset mesh accuracy; the water surface, path, underwater and aerial cross-medium robot and its surroundings are meshed with refinement based on the preset mesh accuracy; Perform finite element solution on the current mesh, extract the solution data, and the solution data includes low-precision solution data and high-precision solution data; Based on the current low-precision solution data and the design parameter vectors in the current sample set, construct a low-quality data set; it is used to preliminarily screen the design parameter vectors in the sample set; Perform finite element solution on the screened design parameter vectors using a high-precision mesh to obtain high-precision solution data. Based on the screened design parameter vectors and high-precision solution data, construct a high-quality data set for training the surrogate model; Based on the high-quality data set, obtain the optimal hyperparameters of the surrogate model through Bayesian optimization; Train the surrogate model based on the optimal hyperparameters; Construct an optimization objective, adopt a multi-objective optimization algorithm to obtain the solutions that meet the bionic form constraints and component constraints under the optimization objective, and screen the solutions with large variances as labeled samples; Judge whether the surrogate model converges. If it converges, output the optimal design parameters. If it does not converge, continue to label samples and expand the high-quality data set, and re-train the surrogate model until the surrogate model converges.

2. The optimization method for the head shape and water entry strategy of the water-air cross-medium robot according to claim 1, wherein The design domain is an eight-dimensional space, including water entry strategy parameters and nose shape parameters. The water entry strategy parameters include water entry speed and water entry angle. The nose shape parameters include beak length, beak thickness, horizontal shape at the front of the head, vertical shape at the front of the head, horizontal shape at the rear of the head, and vertical shape at the rear of the head.

3. The method for optimizing the head shape and water entry strategy of the water-air cross-medium robot according to claim 1, characterized in that, The design domain that meets the bionic form constraints and component constraints is calculated according to the following formula: Component constraints: Among them, is the feasible design domain; is the design domain; is the component constraint function; is the nose shape parameter z The function value of the NURBS curve at X under the condition; Bionic form constraints: Among them, is the bionic form constraint function; l is the head length.

4. The method for optimizing the head shape and water entry strategy of the water-air cross-medium robot according to claim 1, characterized in that, It also includes determining the number of high-precision meshes: Gradually increase the number of meshes starting from the number of meshes corresponding to the preset mesh accuracy, and judge the mesh accuracy according to the impact load change curve in the solution data. As the number of meshes increases, if the curve changes, continue to increase the number of meshes; if the change amplitude of the curve is less than the threshold, take the number of meshes before the increase as the number of meshes of the high-precision mesh.

5. The method for optimizing the head shape and water entry strategy of the water-air cross-medium robot according to claim 1, characterized in that, The overlapping grid technology is adopted during the meshing process.

6. The optimization method for the head shape and water entry strategy of the water-air cross-medium robot according to claim 1, wherein The surrogate model is a deep kernel learning model.

7. The method for optimizing the head shape and water entry strategy of the water-air cross-medium robot according to claim 1, wherein The optimal hyperparameters of the surrogate model are obtained according to the following steps: Construct a Bayesian hyperparameter optimization objective function: Among them, is the surrogate model; is the search space; n is the number of hyperparameters; is the hyperparameter combination; is the optimal hyperparameter; is the loss of the surrogate model to be trained under the hyperparameter combination on the validation set; Randomly extract in the search space from k a set of initial hyperparameter combinations , calculate , and form an initial dataset ; The hyperparameter combination that satisfies the Bayesian hyperparameter optimization objective function is iteratively calculated using the acquisition function to obtain the optimal hyperparameters of the surrogate model: Among them, is the mean value; K Cov(·) is the covariance kernel function; is the currently known optimal objective function value; is the data set; , T is the number of iterations.

8. The method for optimizing the head shape and water entry strategy of the water-air cross-medium robot according to claim 1, characterized in that, In the multi-objective optimization algorithm, the optimization objectives are: Among them, is the mean value of the axial impact load output by the surrogate model; is the mean value of the normal impact load output by the surrogate model; is the water entry velocity; The solutions that satisfy the bionic form constraint and the component constraint are obtained according to the following formula: Among them, is the solution that satisfies the bionic form constraint and component constraint; is the solution obtained by the multi-objective optimization algorithm.

9. The method for optimizing the head shape and water entry strategy of the water-air cross-medium robot according to claim 1, wherein The following steps are used to determine whether the surrogate model converges: Calculate the percentage absolute error between the solution data and the output result of the surrogate model. If the percentage absolute error is less than 1%, the surrogate model is considered to converge; The percentage absolute error is calculated according to the following formula: Among them, n is the number of labeled samples; is the percentage absolute error of the axial impact load; is the percentage absolute error of the normal impact load; is the peak value of the axial impact load of the i th labeled sample in the solution data; is the peak value of the axial impact load of the i th labeled sample output by the surrogate model; is the peak value of the normal impact load of the i th labeled sample in the solution data; is the peak value of the normal impact load of the i th labeled sample output by the surrogate model.

10. The device for optimizing the head shape and water entry strategy of the water-air cross-medium robot is characterized in that, Including: The first module is used to determine the design domain based on the bionic form constraint, the component constraint, and the motion constraint; The second module is used to perform Latin hypercube sampling within the design domain to obtain the design parameter vector, and determine whether the design parameter vector satisfies the bionic form constraint and the component constraint. If it satisfies, the design parameter vector is added to the sample set. If it does not satisfy, resampling is performed. If the iterative sampling times reach the threshold, the design domain is re-determined; The third module is used to draw NURBS curves based on the design parameter vectors in the current sample set to describe the head shape; generate a three-dimensional head model based on the NURBS curves, and complete the component assembly to form a water-air cross-medium robot model; The fourth module is used to reset the model, the position, the attitude, and the motion parameters of the current water-air cross-medium robot; The fifth module is used to automatically check and repair the surface mesh problems of the imported model, generate the simulation calculation domain and perform meshing, where the pool and the air adopt the preset mesh accuracy; the water surface, the path, the water-air cross-medium robot and its surroundings are meshed with refinement based on the preset mesh accuracy; The sixth module is used to perform finite element solution on the current mesh and extract the solution data. The solution data includes low-precision solution data and high-precision solution data; The seventh module is used to construct a low-quality data set based on the current low-precision solution data and the design parameter vectors in the current sample set; and to preliminarily screen the design parameter vectors in the sample set; The eighth module is used to perform finite element solution on the screened design parameter vectors using a high-precision mesh to obtain high-precision solution data, and construct a high-quality data set based on the screened design parameter vectors and the high-precision solution data for training the surrogate model; The ninth module is used to obtain the optimal hyperparameters of the surrogate model through Bayesian optimization based on the high-quality data set; The tenth module is used to train the surrogate model based on the optimal hyperparameters; The eleventh module is used to construct the optimization objectives, and use the multi-objective optimization algorithm to obtain the solutions that satisfy the bionic form constraint and the component constraint under the optimization objectives, and screen the solutions with large variances as the labeled samples; The twelfth module is used to determine whether the surrogate model converges. If it converges, the optimal design parameters are output. If it does not converge, continue to label the samples and expand the high-quality data set, and retrain the surrogate model until the surrogate model converges.

Images