Composite material propeller design method based on regression neural network and genetic algorithm

Abstract

The invention discloses a composite propeller design method based on a regression neural network and a genetic algorithm, and belongs to the field of composite propellers. According to the method, the minimization of the distance from the pitch angle of the blade tip of the composite propeller to the optimal pitch angle of the propeller type under each working condition is taken as the optimization target of the propulsive efficiency, and the minimization of the absolute value of the difference value of the thrust coefficient KTA under the working condition of the low advance speed coefficient and the thrust coefficient KTB under the working condition of the high advance speed coefficient is taken as the optimization target of vibration and noise reduction; the constraint is that the strength of the structural performance of the composite material needs to meet a Tsai-Wu strength criterion, and constructing a ply pre-deformation optimization model; through combination of a regression neural network and a genetic algorithm, a layering angle sequence and pre-deformation geometric parameters of the composite material propeller are optimized, performance optimization of the composite material propeller in a full working condition range is realized under steady-state and non-steady-state working conditions, the balance between the minimum propulsion efficiency and vibration and noise reduction is realized, and the comprehensive performance of the composite material propeller is improved. Therefore, the working performance of the composite propeller is improved.

Description

Technical Field

[0001] The present invention relates to an integrated design method for composite propellers based on regression neural network and genetic algorithm, and particularly to an optimization design method based on GA-GRNN and NSGA-II that comprehensively considers pre-deformation geometric parameters and ply angle sequences, belonging to the field of composite propellers. Background Art

[0002] Composite propellers have many advantages compared with traditional metal propellers: the lightweight characteristic reduces the overall weight and improves the fuel efficiency, which is particularly significant for large ships and naval vessels; the high strength and stiffness endow the propeller with durability, reducing the damage during operation and the maintenance requirements; the corrosion resistance and seawater erosion resistance of composite materials extend the service life and reduce the maintenance cost; in addition, the vibration damping and noise reduction characteristics create a quieter ship environment, which is particularly important for military vessels; finally, the design flexibility of composite materials enables the shape and profile of the propeller to be optimized, improving the propulsion efficiency and overall performance.

[0003] At present, most marine fiber composite propellers adopt the form values of metal propellers, without considering the influence of the fluid-structure interaction effect of the propeller blades. When the composite propeller blades work, they will deform, resulting in a reduction in propulsion efficiency and unable to meet the propulsion requirements of the ship under all working conditions. The unique bending-torsion coupling effect of the fiber-reinforced materials in composite materials can improve the propulsion efficiency of the propeller by reasonably arranging the fiber direction and material ply sequence according to the propeller load conditions and blade structure shape. Therefore, in order to improve the hydrodynamic performance and vibration and noise performance of the propeller and increase the service life of the propeller, domestic and foreign scholars have conducted a large number of studies on the two-way fluid-structure interaction numerical calculation of composite propellers and proposed a ply angle optimization method for composite propellers, mainly by combining different ply angles of composite propellers to obtain a new ply structure of composite propellers.

[0004] However, only changing the ply angle sequence without optimizing the geometric shape of the propeller cannot optimize the performance under all working conditions. Summary of the Invention

[0005] In order to solve the problem that the optimization parameters are single and the performance optimization under the full working condition range cannot be achieved in the design process of existing composite propellers, the purpose of the present invention is to provide a design method for composite propellers based on regression neural network and genetic algorithm, comprehensively optimizing the ply angle sequence of the propeller blades and the pre-deformation geometric parameters of composite propellers, so as to realize the full working condition performance optimization of the balanced vibration and propulsion efficiency of composite propellers. According to the optimization results, the propulsion efficiency of composite propellers is improved, and vibration can be reduced and noise can be reduced.

[0006] The purpose of the present invention is achieved by the following technical solutions:

[0007] The composite propeller design method based on regression neural network and genetic algorithm disclosed by the present invention uses the ply angle sequence and pre-deformed geometry of the composite blade as the optimization parameters of the regression neural network and genetic algorithm, and adopts the Latin hypercube sampling method to perform preliminary ply angle sampling; at the same time, based on the ply angle sequence, the pre-deformed geometry of the composite propeller is optimized, the two-way fluid-structure coupling calculation is carried out through the composite propeller, and the deformations of the leading and trailing edges at each radius of the blade are extracted; considering the requirements of high propulsion efficiency under steady-state conditions and low vibration characteristics under unsteady-state conditions, the corresponding optimization objective functions f1 and f2 are constructed; a finite element model of the composite propeller is established, the two-way fluid-structure coupling calculation is carried out between the finite element model of the composite propeller and the computational fluid dynamics model of the composite propeller, and the optimization objective function is solved; a GA-GRNN network model is constructed to obtain the mapping relationship between the ply angle sequence, pre-deformed geometry parameters and the optimization objective function; based on the NSGA-II multi-objective genetic algorithm, the optimal ply angle sequence and pre-deformed geometry parameters that balance the propeller vibration and propulsion efficiency are obtained through optimization iteration. Based on the optimal ply angle sequence, a pre-deformation amount is applied in the opposite direction of the deformation of the composite propeller, and the geometry of the pre-deformed composite propeller is further calculated, so that its performance under the design conditions is equivalent to that of the original propeller and superior to the original propeller under non-design conditions, realizing the design of the composite propeller based on the regression neural network and genetic algorithm.

[0008] The composite propeller design method based on regression neural network and genetic algorithm disclosed by the present invention includes the following steps:

[0009] Step 1: Select the ply angle sequence θ = [(θ1)2, (θ2)2, (θ3)2, (θ4)2, (θ5)2, (θ6)2] of the optimization design variable, and determine the design variable range [LB, UB]: Use the Optimal Latinhypercube design (OLHD) method of Latin hypercube experimental design to sample the variable ply angle θ, and within [LB, UB], N trial designs are carried out to obtain N sets of discrete samples for subsequent calculations.

[0010] Step 2: Optimize the pre-deformed geometry of the composite propeller using the current ply combination, extract the coordinates of the original leading and trailing edge feature points and convert them into three-dimensional Cartesian coordinates, perform two-way fluid-structure coupling calculation on the composite propeller to obtain the deformation values of the leading and trailing edges, and solve the pitch angle and the trim R a ' of the composite propeller according to the deformation values. Further, the skew C ' of the pre-deformed composite propeller of each blade section is obtained by the pitch angle s ' of each blade section. The pitch angle trim Ra ', side skew C s ' constitute the pre-deformation geometric parameters Convert the pre-deformation geometric parameters into three-dimensional Cartesian coordinates and add them to the original coordinates of the composite propeller to obtain the geometric parameters of the pre-deformed composite propeller, and import them into the modeling software for geometric modeling of the pre-deformed propeller.

[0011] Step 2.1: Determine the profile coordinates corresponding to the two characteristic points, namely the leading edge point and the trailing edge point, in the profile value table of each blade of the propeller, and convert the profile points of the characteristic points into three-dimensional Cartesian coordinate points;

[0012] Step 2.2: Perform two-way fluid-structure interaction calculation on the composite propeller, and extract the deformation values of the leading edge point and the trailing edge point of the composite propeller in the three directions of X, Y, and Z at each radius in the structural finite element analysis module; then pre-deform the leading edge point and the trailing edge point on the above composite propeller, that is, reversely superimpose the extracted deformation values into the coordinates of the leading edge point and the trailing edge point to obtain the pre-deformed coordinates of the two points. Denote the coordinates of the two characteristic points after pre-deformation as (xl1, yl1, zl1), (xt1, yt1, zt1);

[0013] Step 2.3: Solve the pitch angle of the pre-deformed composite propeller at the blade section where the leading edge point and the trailing edge point are located according to the axial coordinates of the leading edge point and the trailing edge point after pre-deformation and the longitudinal tilt R a ', repeat steps 2.1 and 2.2 to obtain the pitch angle of the pre-deformed composite propeller for each blade section and the longitudinal tilt R a ';

[0014] Step 2.4: Use the pitch angle of each blade section of the pre-deformed composite propeller obtained in step 2.3 to calculate the side skew C of the pre-deformed composite propeller for each blade section s '. The pitch angle of the obtained pre-deformed composite propeller longitudinal tilt R a ' and side skew C s ' are used to determine the three-dimensional Cartesian coordinates of the entire blade of the pre-deformed composite propeller.

[0015] Step 2.5: Based on the three-dimensional Cartesian coordinates of the entire blade of the pre-deformed composite propeller obtained in step 2.4, import the three-dimensional Cartesian coordinates into the modeling software for geometric modeling of the pre-deformed propeller.

[0016] Step 3: To simultaneously meet the requirements of high propulsion efficiency of the composite propeller under steady-state conditions and low vibration characteristics under unsteady-state conditions, and to ensure the structural strength of the propeller, considering different operating conditions, i.e., different advance coefficient J = V / nD, in order to balance and improve the propulsion efficiency η = JK T / 2πK Q and the vibration reduction and noise reduction performance, taking the minimization of the distance from the blade tip pitch angle of the composite propeller to the optimal pitch angle of this propeller type under each working condition as the optimization goal of the propulsion efficiency, and minimizing the absolute value of the difference between the thrust coefficient K TA under low advance coefficient conditions and the thrust coefficient K TB under high advance coefficient conditions as the optimization goal of vibration reduction and noise reduction; the constraint is that the strength of the composite material structure performance needs to meet the Tsai-Wu strength criterion and set the range of the ply angle constraint space θ ∈ [LB, UB], where the thrust coefficient is K T = T / ρn 2 D 4 , the torque coefficient is K Q = Q / ρn 2 D 5 , and then a ply pre-deformation optimization model is constructed.

[0017] The ply pre-deformation optimization model is as follows:

[0018] Min f1(θ, X), f2(θ, X) (1.1)

[0019] θ ∈ [LB, UB] (1.2)

[0020] s.t. G(θ, X) ≤ 0 (1.3)

[0021] where f1 represents the propulsion efficiency optimization objective function in the full operating condition range, f2 represents the vibration performance optimization objective function; θ represents the ply angle design variable of the composite propeller, and X represents the pre-deformation geometric parameter of the composite propeller. LB and UB are the minimum and maximum values of the ply angle design variable respectively. G represents the Tsai-Wu strength failure criterion constraint in the composite material structure strength; the finite element model of the composite propeller and the computational fluid dynamics model of the composite propeller are used for two-way fluid-structure interaction calculation to obtain the multi-objective response f1, the objective response f2 and the constraint G corresponding to each group of sample parameters.

[0022] Step 4: Construct a GA-GRNN network model for characterizing the mapping relationship between the ply pre-deformation parameters of the composite propeller and the optimization objectives based on the sample parameter data obtained in Step 3. The GA-GRNN network model consists of two parts: the genetic algorithm GA and the generalized neural network GRNN. The prediction performance is improved by using the genetic algorithm GA to optimize the smoothing factor σ of the GRNN. The GRNN is composed of an input layer, a pattern layer, a normalization layer, and an output layer. The ply angle design variable θ, the pre-deformed pitch angle Longitudinal tilt R a ' and the skew C s ' are used as the input variables of the generalized neural network GRNN, and the target responses f1 and f2 are used as the prediction outputs. Construct a normalized data set of the input and output to train the GRNN. At the same time, set the initialization parameters of the genetic algorithm GA to optimize the smoothing factor σ of the GRNN. Use the loss functions of f1 and f2 predicted by the GRNN as the fitness function, perform genetic operations, i.e., selection, crossover, and mutation, and calculate the fitness values. When the maximum number of generations or the accuracy is less than the set threshold, obtain the optimal smoothing factor of the generalized neural network GRNN. According to the optimal smoothing factor, obtain the ply angle design variable θ, the pre-deformed pitch angle Longitudinal tilt R a ' and the skew C s ' and the GA-GRNN prediction surrogate model between the two objective functions f1 and f2. Sample and predict N groups of results within the constraint space in Step 3 to form the initial sample space for balancing and optimizing the propulsion efficiency and vibration reduction and noise reduction performance.

[0023] The regression neural network preferably selects the GA-GRNN network model.

[0024] Step 4.1: Based on the training sample data and the GRNN network model obtained in Step 3, establish the fitness function of the GA-GRNN model. Evaluate the performance of each candidate smoothing factor σ through the fitness function to optimize the prediction accuracy of the GRNN. Set the initialization parameters of the GA, including the population size, chromosome length, crossover probability, mutation probability, and maximum number of iterations.

[0025] Step 4.2: Encode the smoothing factor σ in real number coding to establish the initial population P = {p1, p2,..., p N}, set the selection rate, crossover probability P c , mutation probability P m , and the maximum number of generations t max , and given the value range U σ of the smoothing factor as the constraint condition.

[0026] Step 4.3: Randomly generate U σAmong the N individuals, namely each candidate smoothing factor combination, form the initial population S = {p1, p2,..., p N}, and set the generation number as t = 1. Calculate the regression error or prediction accuracy of each individual's corresponding GRNN model by computing the fitness function.

[0027] Step 4.4: If the fitness meets the predetermined goal or reaches the maximum genetic generation number t max , then select the individual with the best fitness as the optimal solution to obtain the best smoothing factor σ * . Otherwise, according to the selection strategy, randomly select individuals from the population S for replication to form a new population S1.

[0028] Step 4.5: Randomly select individuals from the population S1 for crossover operation according to the crossover probability P c to generate new individuals and form a new population S2. Mutate the individuals in the population S2 according to the mutation probability P m to obtain a new population S3. Randomly vary the smoothing factors through crossover and mutation operations to ensure the extensiveness of the search space.

[0029] Step 4.6: Take the population S3 as the new generation population, replace the current population S with S3, and update the generation number to t = t + 1. Continue to calculate the fitness and judge whether the termination condition is met according to the criteria of the genetic algorithm. If the termination condition is not reached, return to Step 4.4 and repeat the genetic operations until the best smoothing factor σ * is obtained.

[0030] Step 4.7: Use the best smoothing factor σ * to train the GRNN network and predict the test samples to obtain the mapping relationship between the best ply angle design variables and the target response. Sample and predict N groups of results within the constraint space in Step 3 to form the initial sample space for balancing and optimizing the propulsion efficiency and vibration and noise reduction performance.

[0031] Step 5: Optimize the ply angles and pre-deformation design variables of the composite propeller through the multi-objective genetic algorithm NSGA-II based on the initial sample space obtained in Step 4. Take the objective responses f1 and f2 as the fitness functions of NSGA-II, set the initial parameters for multi-generation evolutionary search, and calculate the Pareto front that balances the optimized propulsion efficiency and vibration reduction and noise reduction performance through non-dominated sorting. Conduct two-way fluid-structure coupling calculations on the composite propeller design schemes corresponding to some of the optimal solutions in the Pareto front to determine whether the composite propeller meets the requirements of preset high propulsion efficiency, low vibration and noise, and strength. If not, iterate the multi-objective genetic algorithm NSGA-II to optimize the ply angles and pre-deformation design variables of the composite propeller. If satisfied, complete the optimization to obtain the composite propeller with integrated optimization design based on the regression neural network and genetic algorithm.

[0032] Step 5.1: Construct the fitness function of the NSGA-II optimization algorithm by using the mapping relationship between the design variables and the two optimization objectives established by GA-GRNN. This fitness function includes the objective functions of maximizing the propulsion efficiency and minimizing the vibration reduction and noise reduction effect, and simultaneously considers the constraint condition of the Tsai-Wu strength failure criterion of the material.

[0033] Step 5.2: Set the initial parameters of the NSGA-II algorithm, including population size, crossover probability, mutation probability, selection strategy, maximum number of iterations, and the calculation method of non-dominated sorting crowding degree.

[0034] Step 5.3: Run the NSGA-II algorithm for optimization search. Through multi-generation evolutionary search, continuously iterate and optimize the ply angle sequence θ of the design variables and the pre-deformation geometric parameters X until the termination condition is met. In each generation, select high-quality individuals according to the fitness function, perform crossover and mutation operations to generate new offspring individuals, and maintain the diversity of the population through non-dominated sorting and crowding degree calculation to promote the expansion of the Pareto front.

[0035] Step 5.4: Extract the optimal solution set of the Pareto front. Conduct two-way fluid-structure coupling calculations on the composite propeller design schemes corresponding to some of the optimal solutions in the Pareto front to determine whether the hydrodynamic performance parameters of the composite propeller are equivalent to those of the metal propeller under the design conditions, and superior to the metal propeller under non-design conditions, and whether the absolute value of the difference in thrust coefficients of the composite propeller at high and low advance coefficient values after pre-deformation is less than the absolute value of the difference in thrust coefficients of the metal propeller. At the same time, determine whether the material fails. If all the above conditions are met simultaneously, complete the ply pre-deformation optimization design of the composite propeller.

[0036] Step 5.5: If none of the above judgment criteria are met, return to Step 5.2 until the hydrodynamic performance requirements and structural strength requirements of the composite marine propeller described in Step 5.4 are satisfied, then complete the final pre-deformation optimization design of the layup, that is, realize the full-condition performance optimization of the balanced vibration and propulsion efficiency of the composite propeller, and obtain the parameters of the composite propeller based on the integrated optimization design of the regression neural network and genetic algorithm.

[0037] It further includes Step Six: According to the optimized design parameters of the composite propeller obtained in Step Five, guide the performance verification and engineering application of the propeller. At the same time, provide a basis for the production process and material selection of the composite propeller through the optimization results, improve the propulsion efficiency of the composite propeller, and be able to reduce vibration and noise, thereby improving the overall working performance and reliability of the propeller.

[0038] Beneficial effects:

[0039] 1. The design method of the composite propeller based on the regression neural network and genetic algorithm disclosed in the present invention combines the regression neural network and genetic algorithm to optimize the layup angle sequence and pre-deformation geometric parameters of the composite propeller, and realizes the full-condition range performance optimization of the composite propeller under steady-state and non-steady-state conditions, significantly improving the propulsion efficiency and the effect of vibration and noise reduction.

[0040] 2. The design method of the composite propeller based on the regression neural network and genetic algorithm disclosed in the present invention, through a multi-objective optimization model, on the basis of considering the thrust coefficient and pitch angle under different advance coefficient conditions, takes the minimum distance from the pitch angle at the tip of the composite propeller blade to the optimal pitch angle of the propeller type under each condition as the optimization goal of propulsion efficiency, and minimizes the absolute value of the difference between the thrust coefficient K TA under low advance coefficient conditions and the thrust coefficient K TB under high advance coefficient conditions as the optimization goal of vibration and noise reduction; the constraint is that the strength of the composite structure performance needs to meet the Tsai-Wu strength criterion, and a pre-deformation optimization model of the layup is constructed; by constructing the pre-deformation optimization model of the layup, the trade-off between minimizing propulsion efficiency and vibration and noise reduction is realized, and the problem that the existing design method fails to balance propulsion efficiency and vibration and noise reduction during the optimization process is solved, thereby improving the working performance of the composite propeller.

[0041] 3. The design method of the composite propeller based on the regression neural network and genetic algorithm disclosed in the present invention accurately predicts the relationship between design variables and optimization goals through the GA-GRNN network model, and combines the NSGA-II multi-objective genetic algorithm for global search, which can efficiently find the optimal design parameters, reduce the computational cost and time in the optimization design process, and improve the optimization design efficiency of the composite propeller. Description of the Drawings

[0042] Figure 1 This is the flowchart of the composite propeller design method based on the regression neural network and genetic algorithm of the present invention. Specific implementation manners

[0043] The present invention will be further described below in conjunction with embodiments and the accompanying drawings:

[0044] An integrated optimization method for the ply structure - pre - deformed geometric shape of a composite propeller based on a regression neural network and genetic algorithm is a more perfect optimization design method, which combines the optimization design of the ply structure and the pre - deformed geometry optimization design. The design variables for the ply structure optimization design are the ply angle sequence of the composite propeller, and the design variables for the pre - deformed geometry optimization are the pitch angle, trim, and skew.

[0045] The example is based on the Seiun - Maru highly skewed marine propeller (HSP), and its geometric parameters are: diameter: 3.6 m; number of blades 5, hub - diameter ratio: 0.1972, disk area ratio: 0.7; pitch ratio: 0.92; skew angle 45°; trim angle - 3.03°; design advance coefficient J design = 0.851.

[0046] As Figure 1 shown, the specific implementation steps of the composite propeller design method based on the regression neural network and genetic algorithm disclosed in this embodiment are as follows:

[0047] Step 1: Select Epoxy Carbon UD (230GPa) Prepreg composite material as the ply material for this time, and the ply thickness is 0.3 mm. Since the HSP propeller has a large skew angle and the blade geometry is relatively complex, a ply method starting from the mid - surface of the propeller and laying towards the blade back and blade face respectively is adopted. Taking two consecutive composite material thin sheets as a ply, and then using this ply to lay the composite propeller; the optimization variable is taken as the ply angle θ of the fibers in each composite material thin sheet, and the upper and lower bounds of the ply are determined according to the design requirements:

[0048]

[0049] Since the ply angle is a discrete variable, the range of ply angle values is [-75°, 90°]. Considering the actual processing application of the composite propeller ply, the ply angles of 0°, 90°, +45°, -45° ± 30°, and ±60° are selected to form a discrete design space. The Optimal Latin hypercube design (OLHD) method is used for sampling. The obtained sample points are based on the continuous design space. Therefore, a mapping relationship needs to be established between the continuous samples and the corresponding discrete points in the discrete sample space. Using this method, 40 sets of discrete samples are obtained through 40 trial designs in the discrete design space for subsequent calculations.

[0050] Step 2: Use the current ply combination for the pre-deformation geometric optimization design of the composite propeller to obtain the pre-deformation geometric parameters. Where is the pitch angle, R a ' is the longitudinal tilt C s ' is the skew. The specific implementation method is as follows:

[0051] Step 2.1: First, determine the profile coordinates (XL, YL) and (XT, YT) of the two characteristic points, namely the leading edge point and the trailing edge point, in the profile value table of each blade section of the propeller from 0.2R to 0.95R. Substitute the profile points of the characteristic points into formula (1.5) to convert them into three-dimensional Cartesian coordinate points.

[0052]

[0053] Among them, x, y, and z are the three-dimensional Cartesian coordinate values of the propeller; C s is the skew value; R a is the longitudinal tilt value; Ri is the radius value corresponding to each blade section of the propeller; (X, Y) is the profile point of the blade section; is the pitch angle; the three-dimensional Cartesian coordinates of the leading edge point and the trailing edge point are obtained as (x l , y l , z l ), (x t , y t , z t );

[0054] Step 2.2: Perform two-way fluid-structure interaction calculations on the composite propeller. Extract the deformation values of the leading edge point and the trailing edge point of the composite propeller at each radius from 0.2R to 0.95R in the X, Y, and Z directions in the structural finite element analysis module. The deformation value of the leading edge point is The deformation value of the trailing edge point is Then, pre-deformation is performed on the leading edge points and trailing edge points of the above composite propeller, that is, the extracted deformation values are reversely superimposed into the coordinates of the leading edge points and trailing edge points to obtain the pre-deformation coordinates of the two points. and Let the coordinates of the two characteristic points after pre-deformation be and

[0055] Step 2.3: Substitute the axial coordinates of the pre-deformed leading edge points and trailing edge points into formula (1.6):

[0056]

[0057] where are the axial coordinates of the pre-deformed leading edge points and trailing edge points respectively; b is the chord length of the blade section corresponding to the leading edge points and trailing edge points; R a ' and are the longitudinal pitch and pitch angle of the blade section corresponding to the pre-deformed leading edge points and trailing edge points respectively. Repeating steps (i) and (ii) of step two can respectively obtain the pitch angle and longitudinal pitch R a ' of the pre-deformed composite propeller for each blade section from 0.2R to 0.95R;

[0058] Step 2.4: Substitute the pitch angles of each blade section from 0.2R to 0.95R of the pre-deformed composite propeller obtained in step 2.3 into the formula to calculate the skew C s ' of each blade section from 0.2R to 0.95R of the pre-deformed composite propeller. Substitute the obtained pitch angle longitudinal pitch R a ' and skew C s ' into formula (1.5) respectively to determine the three-dimensional Cartesian coordinates of the entire blade of the pre-deformed composite propeller.

[0059] Step 2.5: Based on the three-dimensional Cartesian coordinates of the entire blade of the pre-deformed composite propeller obtained in step 2.4, import the three-dimensional Cartesian coordinates into the modeling software to perform geometric modeling of the pre-deformed propeller.

[0060] Step three: Considering the requirements of high propulsion efficiency under steady-state conditions, low vibration characteristics under unsteady-state conditions, and the structural strength requirements of the composite propeller, determine the objective function and constraints in the design process and establish a mathematical model for ply pre-deformation optimization:

[0061] When the propeller is under different operating conditions, that is, the advance coefficient J = V / nD is different, to ensure that the propeller has as high a propulsion efficiency η = JK T / 2πK Q, generally, it is required that the pitch angle of the composite propeller after deformation The change curve of the pitch angle with the advance coefficient J is as close as possible to the optimal pitch angle curve of this propeller type That is, it is required that the composite propeller low At low advance coefficient J The distance from the pitch angle To the optimal pitch angle of the propeller high At high advance coefficient J The distance from the pitch angle To the optimal pitch angle of the propeller Is minimized, that is In the vibration performance, it is required to minimize the thrust pulsation formed by the uneven incoming flow as much as possible, which is equivalent to minimizing the absolute value of the difference between the thrust coefficient K TJlow Under low advance coefficient conditions and the thrust coefficient K TJhigh Under high advance coefficient conditions, that is, min(|K TJlow -K TJhigh |); among them, the design advance coefficient J of the HSP propeller design =0.851, and the high and low advance coefficient conditions are selected according to thrust equivalence as J high =0.95; J low =0.62. The constraint is that the strength of the composite material structure performance needs to meet the Tsai-Wu strength criterion, where the thrust coefficient is K T =T / ρn 2 D 4 , the torque coefficient is K Q =Q / ρn 2 D 5 , the mathematical model is as follows:

[0062] Min f1(θ,X),f2(θ,X) (1.7)

[0063] θ i ∈[0°,90°,+45°,-45°,+30°,-30°,+60°,-60°],i = 1,2,3... (1.8)

[0064] s.t.G(θ,X)≤0 (1.9)

[0065] Among them, f1 represents the hydrodynamic performance optimization objective function, and f2 represents the vibration performance optimization objective function.

[0066]

[0067] f2(θ)=|K TJlow -K TJhigh|(1.11)

[0068] Among them, f1 represents the propulsion efficiency optimization objective function within the full operating range, and f2 represents the vibration performance optimization objective function; θ represents the composite propeller ply angle design variable, and X represents the pre-deformed geometric parameters of the composite propeller. G represents the Tsai-Wu strength failure criterion constraint in the composite material structure strength, G(θ) = maximum failure index values - 1; The composite two-way fluid-structure interaction calculation method is used to calculate the target responses and constraints corresponding to the samples obtained in Step 1: Call the ACP module in ANSYS Workbench to perform the composite propeller ply modeling, and then based on the three-dimensional solid finite element model obtained from the ply, import the established composite propeller finite element model into the finite element analysis software to calculate the blade structure response, add boundary conditions such as propeller speed and fixed constraints, and set the composite propeller blade as the fluid-structure interaction interface. The composite propeller structure control equation is Among them, [M s is the structural mass matrix, [C s is the structural damping matrix, [K s is the structural stiffness matrix; {X} is the structural displacement, is the structural velocity, is the structural acceleration; F CFD represents the fluid force acting on the structure under fluid-structure interaction solved by CFD software. The composite propeller hydrodynamic analysis module is used to solve the Reynolds-averaged Navier-Stokes equations (RANS) for the composite propeller flow field to obtain the hydrodynamic loads of the composite propeller flow field. Through the System Coupling module in the WorkBench platform, the blade structure response and the composite propeller flow field hydrodynamic loads obtained by the solution are used to perform two-way fluid-structure interaction calculations using the step-by-step algorithm. In the numerical simulation, first, the loads acting on the structure by the fluid are obtained from the flow field calculation, and the loads are then transferred to the structure field through the fluid-structure interaction surface. The structure deforms under the action of the loads, the flow field updates the grid according to the changes in the structure field, and the flow field calculation is performed again to obtain the loads acting on the structure. The calculations of the structure field and the flow field are alternated until the calculation accuracy meets the requirements. The multi-objective responses f1 and f2 and the constraint G corresponding to each group of sample points obtained by the two-way fluid-structure interaction calculation;

[0069] Step 4: Construct a GA-GRNN network model for characterizing the mapping relationship between the pre-deformation parameters of the composite propeller ply and the optimization objectives based on the sample parameter data obtained in Step 3. The GA-GRNN network model consists of two parts: the genetic algorithm GA and the generalized neural network GRNN. The prediction performance is improved by using the genetic algorithm GA to optimize the smoothing factor σ of the GRNN. The GRNN is composed of an input layer, a pattern layer, a normalization layer, and an output layer. The ply angle design variable θ, the pitch angle of longitudinal inclination R a ', and the skew angle C s ' are used as the input variables of the generalized neural network GRNN, and the objective responses f1 and f2 are used as the prediction outputs. A normalized data set of the input and output is constructed to train the GRNN. At the same time, the initial parameters of the genetic algorithm GA are set to optimize the smoothing factor σ of the GRNN. The loss functions of f1 and f2 predicted by the GRNN are used as the fitness functions, and genetic operations, namely selection, crossover, and mutation, are performed and the fitness values are calculated. When the maximum number of generations or the accuracy is less than the set threshold, the optimal smoothing factor of the generalized neural network GRNN is obtained. Based on the optimal smoothing factor, the ply angle design variable θ, the pitch angle of longitudinal inclination R a ', and the skew angle C s ' and the GA-GRNN prediction surrogate model between the two objective functions f1 and f2 are obtained. Samples are taken and N groups of results are predicted within the constraint space in Step 3 to form the initial sample space for balancing and optimizing the propulsion efficiency and the vibration and noise reduction performance.

[0070] The regression neural network preferably uses the GA-GRNN network model.

[0071] Step 4.1: Based on the training sample data and the GRNN network model obtained in Step 3, establish the fitness function of the GA-GRNN model. The performance of each candidate smoothing factor is evaluated through the fitness function to optimize the prediction accuracy of the GRNN. Set the initial parameters of the GA, including the population size P = 50, the chromosome length L = 10, the crossover probability P c = 0.85, the mutation probability P m = 0.15, and the maximum number of iterations MaxIter = 100.

[0072] Step 4.2: Encode the smoothing factor σ in a real number coding manner to establish the initial population. Set the selection rate to 0.75, the crossover probability P c = 0.85, the mutation probability P m = 0.15, and the maximum number of generations to 100. Given the value range of the smoothing factor as the constraint condition U σ = [0.01, 1.0], optimization is carried out within this range.

[0073] Step 4.3: Randomly generate N = 50 individuals in σ , that is, each candidate smoothing factor combination, to form the initial population S = {p1, p2,..., p N}, and set the generation number to 0. Calculate the regression error or prediction accuracy of the GRNN model corresponding to each individual by calculating the fitness function. The fitness function is used to evaluate the prediction performance of each smoothing factor. The higher the fitness, the better the prediction accuracy.

[0074] Step 4.4: If the fitness meets the predetermined goal or reaches the maximum genetic generation t max , then select the individual with the best fitness as the optimal solution to obtain the best smoothing factor σ * . Otherwise, according to the selection strategy, randomly select individuals from the population S for replication to form a new population S1. The replication method is through roulette wheel selection or ranking selection methods, and individuals with higher fitness are preferentially selected to be passed on to the next generation.

[0075] Step 4.5: According to the crossover probability P c = 0.85, randomly select individuals from the population S1 for crossover operations to generate new individuals. Crossover operations can adopt methods such as single-point crossover and multi-point crossover. In this example, single-point crossover is adopted, that is, randomly select a crossover point in the gene sequence and exchange the gene information of two individuals to form a new population S2. According to the mutation probability P m = 0.15, perform mutation operations on the individuals in the population S2. Mutation operations can randomly select a gene locus for mutation to ensure the extensiveness of the search space to avoid falling into local optimal solutions, and obtain a new population S3. Through crossover and mutation operations, the smoothing factors are randomly changed to ensure the extensiveness of the search space.

[0076] Step 4.6: Take the population S3 as the new generation population, replace the current population S with S3, and update the generation number to t = t + 1. Continue with fitness calculations and judge whether the termination condition is met according to the criteria of the genetic algorithm. If the termination condition is not reached, return to Step 4.4 and repeat the genetic operations until the optimal smoothing factor σ * is obtained.

[0077] Step 4.7: Use the best smoothing factor σ * to train the GRNN network and predict the test samples to obtain the mapping relationship between the best ply angle design variables and the target response. Sample and predict N groups of results within the constraint space in Step 3 to form the initial sample space for balancing and optimizing the propulsion efficiency and vibration reduction and noise reduction performance.

[0078] ​​Step 5: Based on the initial sample space obtained in Step 4, optimize the ply angles and pre-deformation design variables of the composite propeller through the multi-objective genetic algorithm NSGA-II. Use the objective responses f1 and f2 as the fitness functions of NSGA-II, set the initial parameters for multi-generation evolutionary search, and calculate the Pareto front that balances the optimized propulsion efficiency and vibration reduction and noise reduction performance through non-dominated sorting. Perform two-way fluid-structure interaction calculations on the composite propeller design schemes corresponding to some of the optimal solutions in the Pareto front to determine whether the composite propeller meets the requirements of preset high propulsion efficiency, low vibration and noise, and strength. If not, iterate the multi-objective genetic algorithm NSGA-II to optimize the ply angles and pre-deformation design variables of the composite propeller. If satisfied, complete the optimization to obtain a composite propeller with integrated optimization design based on the regression neural network and genetic algorithm.

[0079] Step 5.1: Construct the fitness functions f1 and f2 of the NSGA-II optimization algorithm by using the mapping relationships between the design variables and the two optimization objectives established by GA-GRNN. This fitness function includes the objective functions of maximizing the propulsion efficiency and minimizing the vibration reduction and noise reduction effects, and at the same time considers the constraint condition G of the Tsai-Wu strength failure criterion for the material strength.

[0080] Step 5.2: Set the initial parameters of the NSGA-II algorithm, including the population size, crossover probability, mutation probability, selection strategy, maximum number of iterations, and non-dominated sorting crowding degree calculation method. Specifically, the population size is set to 100, indicating that each generation of the population contains 100 individuals. A larger population size helps to enhance the global search ability of the algorithm. The crossover probability is set to 0.9, indicating that 90% of the individuals will perform crossover operations to accelerate the search and increase the population diversity through gene recombination; the mutation probability is set to 0.1, indicating that 10% of the individuals will perform mutation operations to prevent getting stuck in local optimal solutions. The selection strategy adopts roulette wheel selection or tournament selection to adapt to the fitness of individuals and ensure that better individuals enter the next generation first. The maximum number of iterations is set to 200 to ensure that the algorithm has enough generations to optimize the solution and avoid excessive calculation. Non-dominated sorting is used to sort individuals to ensure that high-quality solutions are not replaced by inferior solutions; while the crowding degree calculation is used to measure the distribution density of individuals in the population, which helps to maintain the diversity of the population and ensure the balanced distribution of different solutions in the objective space.

[0081] Step 5.3: Run the NSGA-II algorithm for optimization search. Through multi-generation evolutionary search, the algorithm continuously iteratively optimizes the design variables, namely the ply angle sequence θ and the pre-deformation geometric parameter X, until the termination condition is met. In each generation, high-quality individuals are selected according to the fitness function. Individuals with higher fitness have a greater probability of entering the next generation, thus ensuring the transmission of excellent solutions. Crossover and mutation operations continuously generate new offspring individuals during this process. The crossover operation accelerates the search through gene recombination, while the mutation operation ensures the diversity of the population and avoids premature convergence during the search process. Non-dominated sorting is used to sort individuals, and dominant solutions are preferentially selected, that is, solutions that are not dominated by other solutions in the objective space. The crowding distance calculation is used to measure the distribution density of individuals in the population, ensuring the diversity of the population and avoiding the population concentrating in a specific area. Through these operations, the algorithm can promote the expansion of the Pareto front, further optimize the design scheme, and ensure a balance among multiple objectives.

[0082] Step 5.4: Extract the Pareto front optimal solution set. Perform two-way fluid-structure coupling calculations on the composite propeller design schemes corresponding to some of the optimal solutions in the Pareto front to determine whether the hydrodynamic performance parameters of the composite propeller are equivalent to those of the metal propeller under the design conditions and superior to those of the metal propeller under non-design conditions. In addition, it is also necessary to evaluate the difference in the thrust coefficient of the pre-deformed composite propeller at different advance coefficient values, especially whether the absolute value of the thrust coefficient difference is less than that of the metal propeller at high and low advance coefficient values. Finally, confirm whether the material meets the strength requirements and does not fail. Only when all these conditions are met simultaneously is the optimized design of the ply pre-deformation of the composite propeller considered completed, ensuring that the design meets the goals of efficient propulsion and vibration reduction and noise reduction, while ensuring structural strength.

[0083] Step 5.5: If any of the above judgment criteria is not met, return to Step 5.2 until the hydrodynamic performance requirements and structural strength requirements of the composite marine propeller described in Step 5.4 are met, then the final optimized design of the ply pre-deformation is completed, that is, the full-condition performance optimization of the balanced vibration and propulsion efficiency of the composite propeller is achieved, and the parameters of the composite propeller based on the integrated optimization design of the regression neural network and genetic algorithm are obtained.

[0084] Step Six: Based on the final optimization results obtained in Step Five, conduct performance verification of the composite propeller and provide a basis for the production process and material selection of the propeller to ensure that the optimized design achieves the goals of improving propulsion efficiency and vibration reduction and noise reduction.

[0085] The specific description above further elaborates on the purpose, technical solution, and beneficial effects of the invention. It should be understood that the above is only a specific embodiment of the present invention and is not used to limit the protection scope of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A design method for composite propellers based on recurrent neural networks and genetic algorithms, characterized in that: It includes the following steps: Step 1: Select the optimized design variable ply angle sequence θ = [(θ1)2, (θ2)2, (θ3)2, (θ4)2, (θ5)2, (θ6)2], and determine the design variable range [LB, UB]: Use the Orthogonal Latin Hypercube Design (OLHD) method to sample the variable ply angle θ. Within [LB, UB], conduct N trial designs to obtain N sets of discrete samples for subsequent calculations. Step 2: Perform geometric optimization of the pre-deformation of the composite propeller using the current ply combination, extract the coordinate values of the original leading edge and trailing edge feature points and convert them into three-dimensional Cartesian coordinates, perform two-way fluid-structure interaction calculations on the composite propeller to obtain the deformation values of the leading edge and trailing edge, and solve for the pitch angle of the composite propeller according to the deformation values. and trim R a '; Further, through the pitch angle of each blade section, calculate the skew C s ' of the pre-deformed composite propeller of each blade section; The pitch angle , trim R a ', and skew C s ' constitute the pre-deformation geometric parameters Convert the pre-deformation geometric parameters into three-dimensional Cartesian coordinates and add them to the original coordinates of the composite propeller to obtain the geometric parameters of the pre-deformed composite propeller, and import them into the modeling software to perform geometric modeling of the pre-deformed propeller. Step 3: To simultaneously meet the requirements of high propulsion efficiency of the composite propeller under steady-state conditions and low vibration characteristics under unsteady-state conditions, and to ensure the structural strength of the propeller, considering different operating conditions, i.e., different advance coefficients J = V / nD, in order to balance the improvement of the propulsion efficiency η = JK T / 2πK Q and the vibration reduction and noise reduction performance, taking the minimization of the distance from the blade tip pitch angle of the composite propeller to the optimal pitch angle of the propeller type under each working condition as the optimization goal of the propulsion efficiency, and minimizing the absolute value of the difference between the thrust coefficient K TA under low advance coefficient conditions and the thrust coefficient K TB under high advance coefficient conditions as the optimization goal of vibration reduction and noise reduction; the constraint is that the strength of the composite structure performance needs to meet the Tsai-Wu strength criterion and set the ply angle constraint space range θ ∈ [LB, UB], where the thrust coefficient is K T = T / ρn 2 D 4 , the torque coefficient is K Q = Q / ρn 2 D 5 , and then construct the ply pre-deformation optimization model; Step 4: Construct a GA-GRNN network model for characterizing the mapping relationship between the pre-deformation parameters of the composite propeller ply and the optimization objectives based on the sample parameter data obtained in Step 3. The GA-GRNN network model consists of two parts: the genetic algorithm GA and the generalized neural network GRNN. The prediction performance is improved by optimizing the smoothing factor σ of GRNN using the genetic algorithm GA. Among them, GRNN is composed of an input layer, a pattern layer, a normalization layer, and an output layer; the ply angle design variable θ, the pre-deformed pitch angle Longitudinal tilt R a ' and the skew C s ' are used as the input variables of the generalized neural network GRNN, and the target responses f1 and f2 are used as the prediction outputs; construct a normalized data set of the input and output to train GRNN, and at the same time set the initial parameters of the genetic algorithm GA to optimize the smoothing factor σ of GRNN. Take the loss functions of f1 and f2 predicted by GRNN as the fitness function, perform genetic operations, namely selection, crossover, and mutation, and calculate the fitness value. When the maximum number of generations or the accuracy is less than the set threshold, obtain the optimal smoothing factor of the generalized neural network GRNN; according to the optimal smoothing factor, obtain the ply angle design variable θ, the pre-deformed pitch angle Longitudinal tilt R a ' and the skew C s ' and the GA-GRNN prediction surrogate model between the two objective functions f1 and f2, sample and predict N groups of results within the constraint space in Step 3 to form the initial sample space for balancing and optimizing the propulsion efficiency and vibration reduction and noise reduction performance; Step 5: Based on the initial sample space obtained in Step 4, optimize the ply angle and pre-deformation design variables of the composite propeller through the multi-objective genetic algorithm NSGA-II. Take the objective responses f1 and f2 as the fitness functions of NSGA-II, set the initial parameters for multi-generation evolutionary search, calculate the Pareto front that balances the optimized propulsion efficiency and vibration and noise reduction performance through non-dominated sorting. Conduct two-way fluid-structure interaction calculations on the composite propeller design schemes corresponding to some of the optimal solutions in the Pareto front to determine whether the composite propeller meets the requirements of preset high propulsion efficiency, low vibration and noise, and strength. If not, iterate the multi-objective genetic algorithm NSGA-II to optimize the ply angle and pre-deformation design variables of the composite propeller. If satisfied, complete the optimization to obtain a composite propeller based on the integrated optimization design of the regression neural network and genetic algorithm.

2. The design method of the composite propeller based on the recurrent neural network and the genetic algorithm according to claim 1, wherein: The implementation method of Step 2 is as follows: Step 2.1: Determine the profile coordinates corresponding to the two characteristic points, namely the leading edge point and the trailing edge point, in the profile value table of each blade section of the propeller, and convert the profile points of the characteristic points into three-dimensional Cartesian coordinate points. Step 2.2: Conduct two-way fluid-structure interaction calculations on the composite propeller, and extract the deformation values of the leading edge point and the trailing edge point of the composite propeller in the X, Y, and Z directions at each radius in the structural finite element analysis module; then pre-deform the leading edge point and the trailing edge point on the above composite propeller, that is, reversely superimpose the extracted deformation values onto the coordinates of the leading edge point and the trailing edge point to obtain the pre-deformed coordinates of the two points; denote the coordinates of the two characteristic points after pre-deformation as (xl1, yl1, zl1) and (xt1, yt1, zt1). Step 2.3: Solve the pitch angle and trim R of the pre-deformed composite propeller at the blade section where the guide edge points and trailing edge points are located according to the axial coordinates of the guide edge points and trailing edge points after pre-deformation and trim R a ', and repeat Steps 2.1 and 2.2 to obtain the pitch angle and trim R a ' of the pre-deformed composite propeller at each blade section Step 2.4: Using the pitch angles of each blade section of the pre-deformed composite propeller obtained in Step 2.3 the skew C of the pre-deformed composite propeller for each blade section can be calculated s '; The pitch angle of the obtained pre-deformed composite propeller the longitudinal tilt R a ' and the skew C s ' are used to determine the three-dimensional Cartesian coordinates of the entire blade of the pre-deformed composite propeller Step 2.5: Based on the three-dimensional Cartesian coordinates of the overall pre-deformed composite propeller blade obtained in Step 2.4, import the three-dimensional Cartesian coordinates into the modeling software to perform geometric modeling of the pre-deformed propeller.

3. The composite propeller design method based on the regression neural network and genetic algorithm according to claim 3, characterized in that: The ply pre-deformation optimization model is as follows: Min f1(θ, X), f2(θ, X) (1.1) θ ∈ [LB, UB] (1.2) s.t. G(θ, X) ≤ 0 (1.3) Among them, f1 represents the optimization objective function of the propulsion efficiency within the full operating conditions range, and f2 represents the optimization objective function of the vibration performance; θ represents the design variable of the composite propeller ply angle, and X represents the pre-deformation geometric parameters of the composite propeller; LB and UB are respectively the minimum and maximum values of the ply angle design variable; G represents the Tsai-Wu strength failure criterion constraint in the composite material structure strength; the finite element model of the composite propeller and the computational fluid dynamics model of the composite propeller are subjected to two-way fluid-structure interaction calculations to obtain the multi-objective response f1, the objective response f2, and the constraint G corresponding to each group of sample parameters.

4. The composite propeller design method based on a recurrent neural network and a genetic algorithm according to claim 3, characterized in that: The implementation method of Step Four is as follows. Step 4.1: According to the training sample data and the GRNN network model obtained in Step Three, establish the fitness function of the GA-GRNN model; evaluate the performance of each candidate smoothing factor σ through the fitness function to optimize the prediction accuracy of GRNN; set the initialization parameters of GA, including population size, chromosome length, crossover probability, mutation probability, and maximum number of iterations. Step 4.2: Encode the smoothing factor σ in real number coding to establish the initial population P = {p1, p2,..., p N}, set the selection rate, crossover probability P c , mutation probability P m , and the maximum number of genetic generations t max , and given the value range U σ of the smoothing factor as a constraint condition; Step 4.3: Randomly generate N individuals in σ , that is, each candidate smoothing factor combination, to form the initial population S = {p1, p2,..., p N}, and set the generation number to t = 1; calculate the regression error or prediction accuracy of the GRNN model corresponding to each individual by computing the fitness function. Step 4.4: If the fitness meets the predetermined goal or reaches the maximum genetic algebra t max , then select the individual with the best fitness as the optimal solution to obtain the best smoothing factor σ * ; otherwise, according to the selection strategy, randomly select individuals from the population S for replication to form a new population S1; Step 4.5: According to the crossover probability P c Randomly select individuals from the population S1 for crossover operation to generate new individuals and form a new population S2; According to the mutation probability P m Perform mutation operation on the individuals in the population S2 to obtain a new population S3; Randomly vary the smoothing factor through crossover and mutation operations to ensure the extensiveness of the search space; Step 4.6: Take the population S3 as the new generation population, replace the current population S with S3, and update the generation number to t = t + 1; continue with fitness calculation, and determine whether the termination condition is met according to the criteria of the genetic algorithm; if the termination condition is not reached, return to Step 4.4 and repeat the genetic operations until the optimal smoothing factor σ is obtained * ; Step 4.7: Use the optimal smoothing factor σ * Train the GRNN network and predict the test samples to obtain the mapping relationship between the optimal ply angle design variables and the target response. Sample and predict N groups of results within the constraint space in Step 3 to form the initial sample space for balancing and optimizing the propulsion efficiency and vibration and noise reduction performance.

5. The design method of the composite propeller based on the recurrent neural network and the genetic algorithm according to claim 4, characterized in that: The implementation method of Step Five is as follows. Step 5.1: Construct the fitness function of the NSGA-II optimization algorithm by using the mapping relationship between the design variables established by GA-GRNN and the two optimization objectives; this fitness function includes the objective functions of maximizing the propulsion efficiency and minimizing the vibration reduction and noise reduction effects, and at the same time considers the constraint conditions of the Tsai-Wu strength failure criterion of the material strength. Step 5.2: Set the initialization parameters of the NSGA-II algorithm, including population size, crossover probability, mutation probability, selection strategy, maximum number of iterations, and non-dominated sorting crowding degree calculation method. Step 5.3: Run the NSGA-II algorithm for optimization search; through multi-generation evolutionary search, continuously iterate and optimize the ply angle sequence θ and the pre-deformation geometric parameters X of the design variables until the termination conditions are met; in each generation, select high-quality individuals according to the fitness function, perform crossover and mutation operations to generate new offspring individuals, and maintain the diversity of the population through non-dominated sorting and crowding degree calculation to promote the expansion of the Pareto front. Step 5.4: Extract the Pareto front optimal solution set, perform two-way fluid-structure interaction calculations on the composite propeller design schemes corresponding to some of the optimal solutions in the Pareto front, and judge whether the hydrodynamic performance parameters of the composite propeller are equivalent to those of the metal propeller under the design conditions, superior to the metal propeller under non-design conditions, and whether the absolute value of the difference in the thrust coefficient of the composite propeller after pre-deformation at high and low advance coefficient is less than the absolute value of the difference in the thrust coefficient of the metal propeller, and at the same time judge whether the material fails; if all the above conditions are met at the same time, the ply pre-deformation optimization design of the composite propeller is completed. Step 5.5: If none of the above judgment criteria are met, return to Step 5.2 until the hydrodynamic performance requirements and structural strength requirements of the composite marine propeller described in Step 5.4 are satisfied, then complete the final pre-laying deformation optimization design, that is, achieve the full-condition performance optimization of the balanced vibration and propulsion efficiency of the composite propeller, and obtain the parameters of the composite propeller based on the integrated optimization design of the regression neural network and genetic algorithm.

6. The design method of composite propellers based on recurrent neural network and genetic algorithm according to claim 1, 2, 3, 4 or 5, characterized in that: It also includes Step Six: According to the optimized design parameters of the composite propeller obtained in Step Five, guide the performance verification and engineering application of the propeller; at the same time, provide a basis for the production process and material selection of the composite propeller through the optimization results, improve the propulsion efficiency of the composite propeller, and be able to reduce vibration and noise, thereby improving the overall working performance and reliability of the propeller.

Images