Rotor craft aeroelastic stability prediction method

Abstract

The invention discloses a rotor craft aeroelastic stability prediction method, and relates to the field of rotor dynamics, the method comprises the following steps: constructing a nonlinear aeroelastic dynamics control equation, and determining dynamics sensitive parameters; constructing an input sample data set; obtaining an output sample data set based on the input sample data set and the nonlinear aeroelastic dynamics control equation; constructing a sample data set based on the input sample data set and the output sample data set; constructing an initial aeroelastic stability prediction model based on the de-noising diffusion probability model; training the initial aeroelastic stability prediction model by adopting the sample data set to obtain an aeroelastic stability prediction model; acquiring the working condition of the rotorcraft in real time to determine the value of the dynamic sensitive parameter of the rotorcraft; inputting the values of the dynamic sensitive parameters into the aeroelastic stability prediction model to obtain a corresponding modal damping ratio; and determining the aeroelastic stability of the rotorcraft based on the modal damping ratio. According to the method, the aeroelastic stability of the rotorcraft can be accurately predicted.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of rotor dynamics, in particular to a rotor aircraft aeroelastic stability prediction method. BACKGROUND

[0002] Rotor aircraft (such as helicopters, distributed rotorcraft, tilt rotor aircraft, etc.) as a key equipment with vertical take-off and landing and low-altitude low-speed flight capability has irreplaceable application value in military and civilian fields. With the development of aviation technology, improving flight speed has become the core demand to expand the task radius and operation efficiency, but in the high-speed forward flight state, the coupling effect between the aircraft rotor system and the airflow field will be significantly enhanced, which is easy to cause aeroelastic instability phenomena such as whirl flutter, blade flapping / pitch coupling (classic flutter), etc. Such phenomena not only cause the vibration of the aircraft to intensify and the control performance to decline, but in severe cases, it may cause structural damage or disintegration of the aircraft in a short time, directly causing flight accidents. At present, the traditional aeroelastic stability prediction method (such as linearized theoretical analysis) is difficult to fully capture the nonlinear and unsteady characteristics of the airflow field at high speed, as well as the strong coupling effect of the rotor system aerodynamics and structure, resulting in insufficient prediction accuracy of whirl flutter and other instability phenomena, which cannot meet the stringent requirements of safety and reliability for the high-speed development of rotor aircraft.

[0003] In recent years, data-driven technology has gradually attracted attention in the field of aircraft aeroelastic characteristic analysis, among which neural network (Neural Network, NN) and generative adversarial network (Generative Adversarial Networks, GAN) are relatively common methods. However, neural network (NN) is prone to model overfitting and poor generalization when dealing with high-dimensional and strongly nonlinear aeroelastic coupling data, and has extremely high requirements for the number and quality of training data, which makes it difficult to play an effective role in the scene of aeroelastic instability phenomena with scarce samples. Although generative adversarial network (GAN) has certain advantages in data generation, it has defects such as unstable training process and mode collapse (insufficient diversity of generated samples), which cannot accurately reproduce the complex dynamic characteristics of the aeroelastic system at high speed. SUMMARY

[0004] The purpose of the present application is to provide a rotor aircraft aeroelastic stability prediction method that can accurately predict the aeroelastic stability of a rotor aircraft.

[0005] To achieve the above purpose, the present application provides the following solutions: The present application provides a rotor aircraft aeroelastic stability prediction method, comprising: obtaining parameter data of a rotor aircraft; Based on the parameter data, the nonlinear aeroelastic dynamics control equations of the rotorcraft are constructed, and the dynamic sensitive parameters are determined; Define the value range of the dynamic sensitive parameter to construct an input sample dataset; the sample dataset includes multiple combinations of dynamic sensitive parameter value samples. Based on the input sample dataset and the nonlinear aeroelastic dynamics control equations, an output sample dataset corresponding to the input sample dataset is obtained; the output sample dataset includes sample modal damping ratios corresponding to all sample combinations of the values ​​of the dynamic sensitive parameters in the input sample dataset; Construct a sample dataset based on the input sample dataset and the output sample dataset; An initial aeroelastic stability prediction model is constructed based on a denoised diffusion probability model; The initial aeroelastic stability prediction model is trained using the sample dataset to obtain the aeroelastic stability prediction model; The operating conditions of the rotorcraft are acquired in real time to determine the values ​​of the rotorcraft's dynamic sensitive parameters; the values ​​of the dynamic sensitive parameters are input into the aeroelastic stability prediction model to obtain the corresponding modal damping ratio; and the aeroelastic stability of the rotorcraft is determined based on the modal damping ratio.

[0006] In one embodiment, the nonlinear aeroelastic dynamics control equations of the rotorcraft are constructed based on the parameter data, and dynamically sensitive parameters are determined, including: Dynamic units are divided according to the structural topology of the rotorcraft; Based on the parameter data, the elastic potential energy and kinetic energy of each dynamic unit are determined, and the external force work of the aerodynamic load is determined. Based on the elastic potential energy, kinetic energy, and external force work of each dynamic unit, a nonlinear aeroelastic dynamics control equation is constructed. Test factors are set based on the parameters of the rotorcraft, and the values ​​of the test factors are set to construct multiple test groups; each test group includes test factors and their values. L27 (3) 13 The orthogonal experimental design scheme and multiple test groups are used to simulate the nonlinear aeroelastic dynamics control equations to obtain the modal damping ratios corresponding to each test group; based on the modal damping ratios corresponding to each test group, the experimental factors are screened to obtain the dynamic sensitive parameters.

[0007] In one embodiment, a nonlinear aeroelastic dynamics control equation is constructed based on the elastic potential energy, kinetic energy, and external force work of each dynamic unit, including: The aeroelastic dynamics equations are constructed based on the elastic potential energy, kinetic energy, and external force work of each dynamic unit using Hamilton's variational principle. The Leishman-Beddoes model was used to simulate the dynamic stall process of a rotorcraft airfoil in an unsteady airflow, and the aerodynamic coefficients under different rates of change of angle of attack were determined. The aeroelastic dynamics equations are modified based on the aerodynamic coefficients to obtain the nonlinear aeroelastic dynamics control equations.

[0008] In one embodiment, the aeroelastic dynamics equations are modified based on the aerodynamic coefficients to obtain the nonlinear aeroelastic dynamics control equations, including: The aerodynamic coefficients are used to replace the static coefficients of the aerodynamic load terms in the aeroelastic dynamics equations to construct time-varying aerodynamic load terms; By embedding the time-varying aerodynamic load term into the aeroelastic dynamics equation, the nonlinear aeroelastic dynamics control equation is obtained.

[0009] In one embodiment, a sample range of values ​​for the dynamic sensitive parameter is defined to construct an input sample dataset, including: Based on the typical operating conditions of high-speed forward flight of a rotorcraft, the sample range of values ​​for the aforementioned dynamic sensitive parameters is determined; The Latin hypercube sampling method is used to sample the value sample interval to obtain multiple combinations of value samples of the dynamic sensitive parameters, so as to construct the input sample dataset.

[0010] In one embodiment, based on the input sample dataset and the nonlinear aeroelastic dynamics control equations, an output sample dataset corresponding to the input sample dataset is obtained, including: After inputting the sample values ​​of the dynamic sensitive parameters into the nonlinear aeroelastic dynamics control equation, the nonlinear aeroelastic dynamics control equation is converted into a state-space equation. The QR decomposition method is used to solve for the eigenvalues ​​of the state matrix in the state-space equations. Based on the eigenvalues, determine the modal damping ratio corresponding to the sample combination of values ​​of the dynamic sensitive parameter.

[0011] In one embodiment, constructing a sample dataset based on the input sample dataset and the output sample dataset includes: The Min-Max normalization method is used to standardize the sample combinations of dynamic sensitive parameters in the input sample dataset and the modal damping ratio in the output sample dataset, respectively, to obtain the standardized input sample dataset and the standardized output sample dataset. Based on the standardized input sample dataset, parameter labels are added to the modal damping ratios in the standardized output sample dataset to establish the correspondence between the sample combinations of dynamic sensitive parameters and the modal damping ratios, and a sample dataset is constructed based on the output sample dataset with added parameter labels.

[0012] In one embodiment, the initial aeroelastic stability prediction model is trained using the sample dataset to obtain the aeroelastic stability prediction model, including: The sample dataset is divided into a training set and a validation set according to a set ratio; The training set is augmented to obtain an augmented training set; The hyperparameters of the initial aeroelastic stability prediction model are defined; the hyperparameters include the initial learning rate, batch size, maximum number of training rounds, number of diffusion steps, and weight decay coefficient. The initial gaseous stability prediction model is trained using the enhanced training set, and the loss value of the trained initial gaseous stability prediction model is calculated using the validation set and loss function. The process continues until the loss value of the trained initial gaseous stability prediction model reaches a set condition, thus obtaining the gaseous stability prediction model.

[0013] In one embodiment, the Adam optimizer is used to update the hyperparameters during the training of the initial aeroelastic stability prediction model.

[0014] In one embodiment, the loss function is a mean squared error function, expressed as: ; In the formula, Indicates the loss value. This indicates the number of samples in the validation set. Indicates the first in the verification set Modal damping ratio of each sample, The first output of the model represents the... Modal damping ratio of each sample.

[0015] According to the specific embodiments provided in this application, this application has the following technical effects: This application provides a method for predicting the aeroelastic stability of a rotorcraft. An initial aeroelastic stability prediction model is constructed based on a denoised diffusion probability model, and this model is trained using a sample dataset to obtain the aeroelastic stability prediction model. Leveraging the technical advantages of the denoised diffusion probability model, it addresses the problems of high requirements for the quantity and quality of training data in neural networks, poor adaptability to small sample scenarios, and the instability and insufficient sample diversity in generative adversarial networks. This ensures strong generalization ability and prediction stability even in small sample scenarios. Furthermore, a nonlinear aeroelastic dynamics control equation for the rotorcraft is constructed, which can accurately capture the complex dynamic characteristics of the rotorcraft's aeroelastic system during high-speed forward flight, such as nonlinearity, unsteadiness, and strong coupling, thereby improving the accuracy and reliability of the aeroelastic stability prediction model. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart of a method for predicting the aeroelastic stability of a rotorcraft according to an embodiment of this application; Figure 2 This is a schematic diagram of the structure of a noise addition module provided in an embodiment of this application; Figure 3 This is a schematic diagram of a noise reduction module structure provided in an embodiment of this application; Figure 4 This is a schematic diagram of the Transformer submodule structure in a noise reduction module provided in an embodiment of this application. Detailed Implementation

[0018] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] Actual scenarios of aeroelastic instability in rotorcraft are scarce. Wind tunnel and flight tests are too costly and pose significant safety risks, resulting in a limited number of training samples that are difficult to obtain. Neural networks (NNs), when processing high-dimensional, strongly nonlinear aeroelastic coupled data, not only rely on a large amount of high-quality labeled data but also have weak model generalization ability. The combination of these two factors makes NNs prone to overfitting in aeroelastic stability prediction, hindering their effectiveness. Generative adversarial networks (GANs) rely on adversarial training between the generator and discriminator. This architecture is prone to imbalance in the training process, and its ability to control the diversity of generated samples is weak, making it unable to reproduce the complex dynamic characteristics of aeroelastic systems. Ultimately, this leads to unstable GAN training, pattern collapse problems, and difficulty in accurately predicting aeroelastic characteristics during high-speed forward flight.

[0020] To address the shortcomings of data-driven technologies (such as neural networks and generative artificial intelligences) in aeroelastic stability prediction, this application aims to leverage the technical advantages of conditional diffusion models to solve the problems of high dependence on training data and poor adaptability to small sample scenarios in neural networks, as well as the instability and insufficient sample diversity in GAN training. Conditional diffusion models, through progressively adding noise and reverse learning of the denoising process, can efficiently model the probability distribution of high-dimensional nonlinear systems under given conditions (such as flight state parameters and structural design parameters). This not only possesses stronger generalization ability and stability but also achieves accurate prediction of complex phenomena in small sample data scenarios, providing a new technical path to solve the problem of insufficient accuracy in aeroelastic stability prediction for rotorcraft during high-speed forward flight. Simultaneously, it accurately captures the nonlinear, unsteady, and strongly coupled characteristics of aeroelastic systems during high-speed forward flight, improving the prediction accuracy of unstable phenomena such as gyroscopic flutter and blade flapping / variable pitch coupling. Ultimately, it provides a precise, efficient, and highly adaptable new method for predicting the aeroelastic stability of rotorcraft during high-speed forward flight, meeting the stringent safety and reliability requirements of the high-speed development of aircraft.

[0021] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0022] In one exemplary embodiment, such as Figure 1 As shown, a method for predicting the aeroelastic stability of a rotorcraft is provided, including: Step 100: Obtain the parameter data of the rotorcraft. Based on the parameter data, construct the nonlinear aeroelastic dynamics control equations of the rotorcraft and determine the dynamically sensitive parameters.

[0023] Step 200: Define the sample range of values ​​for the dynamic sensitive parameters to construct the input sample dataset. The sample dataset includes multiple combinations of sample values ​​for the dynamic sensitive parameters.

[0024] Step 300: Based on the input sample dataset and the nonlinear aeroelastic dynamics control equations, obtain the output sample dataset corresponding to the input sample dataset. The output sample dataset includes the sample modal damping ratios corresponding to the sample combinations of all dynamically sensitive parameters in the input sample dataset.

[0025] Step 400: Construct a sample dataset based on the input sample dataset and the output sample dataset.

[0026] Step 500: Construct an initial aeroelastic stability prediction model based on the denoised diffusion probability model. The initial aeroelastic stability prediction model is trained using a sample dataset to obtain the aeroelastic stability prediction model.

[0027] Step 600: Real-time acquisition of the rotorcraft's operating conditions to determine the values ​​of its dynamic sensitive parameters. Inputting these dynamic sensitive parameter values ​​into the aeroelastic stability prediction model yields the corresponding modal damping ratios. The aeroelastic stability of the rotorcraft is then determined based on these modal damping ratios.

[0028] As an optional implementation, in order to accurately reproduce the aerodynamic-structural-inertial coupling characteristics of the aircraft under high-speed forward flight conditions, step 100 involves constructing the nonlinear aeroelastic dynamics control equations of the rotorcraft based on parameter data and determining the dynamically sensitive parameters, including: Step 110: Divide the dynamic units according to the topology of the rotorcraft structure.

[0029] Step 120: Determine the elastic potential energy and kinetic energy of each dynamic unit based on the parameter data, and determine the external force work of the aerodynamic load.

[0030] Step 130: Construct nonlinear aeroelastic dynamics control equations based on the elastic potential energy, kinetic energy, and external work of each dynamic unit.

[0031] Step 140: Based on the rotorcraft parameters, set the test factors and their values ​​to construct multiple test groups. Each test group includes the test factors and their values. Using L27 (3 13 An orthogonal experimental design scheme and multiple test groups were used to simulate the nonlinear aeroelastic dynamics control equations, obtaining the modal damping ratios corresponding to each test group. Based on the modal damping ratios corresponding to each test group, experimental factors were screened to obtain dynamic sensitive parameters.

[0032] The implementation process of step 130 includes: Step 131: Using Hamilton's variational principle, construct the aeroelastic dynamics equations based on the elastic potential energy, kinetic energy, and external work of each dynamic unit.

[0033] Step 132: The Leishman-Beddoes model is used to simulate the dynamic stall process of the rotorcraft airfoil in unsteady airflow, and the aerodynamic coefficients under different rates of change of angle of attack are determined.

[0034] Step 133: Based on aerodynamic coefficients, modify the aeroelastic dynamics equations to obtain the nonlinear aeroelastic dynamics control equations. This step involves replacing the static coefficients of the aerodynamic load terms in the aeroelastic dynamics equations with aerodynamic coefficients to construct time-varying aerodynamic load terms. Embedding these time-varying aerodynamic load terms into the aeroelastic dynamics equations yields the nonlinear aeroelastic dynamics control equations.

[0035] For example, in order to construct nonlinear aeroelastic dynamics control equations that can accurately reproduce the aerodynamic-structural-inertial coupling characteristics of rotorcraft under high-speed forward flight conditions, the sensitivity of key parameters to aircraft stability is quantitatively analyzed, the law of parameter action is clarified, and support that combines theoretical rigor and engineering practicality is provided for the generation of subsequent sample datasets.

[0036] The dynamic units are divided according to the structural topology of the aircraft (such as the wing-nacelle-rotor connection relationship of a tiltrotor aircraft), and the generalized coordinates of each dynamic unit are defined. The wing unit includes geometric attitude parameters such as sweep angle, dihedral angle, and angle of attack; the connections between components need to consider the offsets along the x, y, and z axes. The rotor unit includes degrees of freedom such as flapping angle, flare angle, and torsion angle. The nacelle unit includes pitch and yaw degrees of freedom. Based on the acquired parameter data of the tiltrotor aircraft, the elastic potential energy (generated by the elastic deformation of the material) and kinetic energy (generated by the translation and rotation of the components) of each dynamic unit are calculated, and the work done by aerodynamic loads (lift, drag, and torque of the high-speed airflow on the wing and rotor blades) is determined (i.e., external force work). Hamilton's variational principle is used. Based on the elastic potential energy, kinetic energy, and external work of each dynamic unit, an aeroelastic dynamics equation incorporating aerodynamic-structural-inertial coupling characteristics is constructed. This ensures that the equation can effectively characterize the unsteady effects of the airflow field (such as dynamic stall) and the time-varying characteristics of the coupling terms during high-speed forward flight of a rotorcraft. For variational notation, and These represent the start time and end time of the action, respectively. As kinetic energy, It is elastic potential energy. Work done by external force; This represents the virtual kinetic energy variation. This represents the variational component of the elastic imaginary energy. Let represent the variation of virtual work done by external forces. Taking the partial derivative of the virtual kinetic energy variation yields the mass matrix, damping matrix, stiffness matrix, and nonlinear force vector of the kinetic energy contribution; taking the partial derivative of the virtual work variation of external forces yields the aerodynamic damping matrix and aerodynamic stiffness matrix of the aerodynamic force (i.e., aerodynamic load) contribution; taking the partial derivative of the virtual potential energy variation yields the stiffness matrix of the elastic potential energy contribution. The constructed aeroelastic dynamics equations are expressed as: In the formula, The coupling stiffness matrix includes the stiffness matrix contributed by kinetic energy, the aerodynamic stiffness matrix contributed by aerodynamic load, and the stiffness matrix contributed by elastic potential energy. This is the coupled damping matrix (including the damping matrix contributed by kinetic energy and the aerodynamic damping matrix contributed by aerodynamic load). These are the coupling mass matrices (including the mass matrix of kinetic energy contribution), which can reflect the aerodynamic-structural-inertial coupling characteristics; It is a nonlinear force vector. Indicates the degree of freedom of motion.

[0037] After constructing the aeroelastic dynamics equations, the Leishman-Beddoes model is introduced to correct the aerodynamic loads, and then to correct the external work done by the aerodynamic loads. The Leishman-Beddoes model is used to simulate the dynamic stall process of a rotorcraft airfoil in unsteady airflow (including the attached flow, separated flow, and wake flow stages), and the aerodynamic coefficients (including lift and drag coefficients and moment coefficients) under different angle of attack rates are calculated. First, the relationship between the aerodynamic load terms (i.e., the aerodynamic stiffness matrix in the coupled stiffness matrix and the aerodynamic damping matrix in the coupled damping matrix) and the aerodynamic coefficients in the aeroelastic dynamics equations is clarified. Then, aerodynamic coefficients are introduced to replace static coefficients to construct time-varying aerodynamic load terms. The time-varying aerodynamic load terms are embedded into the aeroelastic dynamics equations and iteratively corrected and coupled to ensure that the aerodynamic load terms always match the dynamic motion state of the rotorcraft. Finally, the aeroelastic dynamics equations are accurately corrected. Then, the specific data of the rotorcraft's structural material properties (such as density and component stiffness coefficients) and geometric parameters (including rotor radius, blade chord length, wing aspect ratio, nacelle length, etc.) are substituted to obtain the nonlinear aeroelastic dynamics control equations.

[0038] Rotor speed, flapping-pitch coupling coefficient, pre-cone angle, hub suspension height, and structural stiffness were set as experimental factors in the rotorcraft parameters. Each factor was configured with three levels (covering commonly used engineering ranges) to construct multiple test groups. The L27 (3 13 An orthogonal experimental design scheme and multiple experimental groups were used. The minimum modal damping ratio of each experimental group was calculated by simulation, and the Pareto effect diagram was plotted to quantify the influence weight of each parameter on the minimum modal damping ratio. Finally, parameters with significant influence (weight ratio greater than 10%) were selected from the experimental factors as dynamic sensitive parameters.

[0039] The dynamic modes of the rotorcraft encompass key modes directly related to aeroelastic stability during high-speed forward flight, specifically including: wing basic bending and torsional modes, nacelle pitching and yaw modes, combined, forward, and backward modes of rotor flapping and shimmy, and rotor torsional modes. The range of values ​​for the test factors (i.e., the three levels set) references relevant standards to ensure consistency with actual engineering application scenarios.

[0040] As an optional implementation, step 200 includes: determining the value sample range of dynamic sensitive parameters based on typical operating conditions of high-speed forward flight of a rotorcraft; sampling the value sample range using the Latin hypercube sampling method to obtain multiple value sample combinations of dynamic sensitive parameters to construct the input sample dataset.

[0041] The implementation process of step 300 includes: inputting the sample combination of values ​​of the dynamic sensitive parameters into the nonlinear aeroelastic dynamics control equations, and then converting the nonlinear aeroelastic dynamics control equations into state-space equations. The eigenvalues ​​of the state matrix in the state-space equations are solved using the QR decomposition method. Based on the eigenvalues, the modal damping ratio corresponding to the sample combination of values ​​of the dynamic sensitive parameters is determined.

[0042] For example, the dynamic sensitive parameters ultimately determined in step 100 of the above embodiments include the flapping-pitch coupling coefficient, rotor speed, and structural stiffness. Based on typical operating conditions of high-speed forward flight of the aircraft, the sample range of values ​​for the dynamic sensitive parameters is determined. Subsequently, Latin hypercube sampling (LHS) is used to sample within the sample range of values ​​for the dynamic sensitive parameters, generating 1000 sets of sample combinations of values ​​for the dynamic sensitive parameters to construct the input sample dataset. Latin hypercube sampling ensures that the sample values ​​of each parameter are uniformly distributed within their sample range, and that the sample values ​​of different parameters are not duplicated, effectively improving the representativeness and coverage of the samples and avoiding a decrease in the generalization ability of the aeroelastic stability prediction model due to sample bias.

[0043] The sample combinations of values ​​for each dynamic sensitive parameter in the input sample dataset are input into the nonlinear aeroelastic dynamics control equations, and the nonlinear aeroelastic dynamics control equations are linearized and transformed into state-space form. ,in, For state vectors, (The state matrix) is obtained by solving the state-space equations using the QR decomposition method. The eigenvalues, and the negative of the real part of each eigenvalue, represent the modal damping ratios corresponding to the sample combinations of values ​​for the dynamic sensitive parameters. In solving for the modal damping ratios, it is crucial to focus on those directly related to rotational flutter and classical flutter, such as the damping ratios of the wing's fundamental bending and torsional modes, the damping ratio of the rotor flapping collective mode, and the damping ratio of the rotor tumbling backward mode. Simultaneously, wind tunnel test data of similar rotorcraft from publicly available literature can be collected. The dynamic sensitive parameters and corresponding measured modal damping ratios for key operating conditions can be compared with the above-mentioned sample combinations and calculated modal damping ratios. The least squares method is used to correct the aerodynamic coefficients (delay time constants in the time-varying aerodynamic load terms) of the nonlinear aeroelastic dynamics control equations, ensuring that the relative error between the calculated and measured values ​​is less than 5%. Finally, an output sample dataset containing sample modal damping ratios corresponding one-to-one with the sample combinations of values ​​for the dynamic sensitive parameters in the input sample dataset is formed. Modal damping ratio is a core indicator of rotorcraft stability: when the modal damping ratio is less than 0, the rotorcraft is in an aeroelastic unstable state and is at risk of flutter; when the modal damping ratio is greater than or equal to 0, the rotorcraft is in an aeroelastic stable state.

[0044] As an optional implementation, step 400 includes: standardizing the sample combinations of dynamic sensitive parameter values ​​in the input sample dataset and the modal damping ratios in the output sample dataset using the Min-Max normalization method, respectively, to obtain standardized input sample datasets and standardized output sample datasets. Based on the standardized input sample dataset, parameter labels are added to the modal damping ratios in the standardized output sample dataset to establish the correspondence between the sample combinations of dynamic sensitive parameter values ​​and the modal damping ratios, and a sample dataset is constructed based on the output sample dataset with added parameter labels.

[0045] For example, in the data processing stage, normalization is used to eliminate the interference of parameter dimension differences on the training of the prediction model, and labels are added to establish a mapping relationship between the sample combination of dynamic sensitive parameter values ​​and the modal damping ratio, ensuring that the data input to the prediction model has a uniform format and clear logic.

[0046] The Min-Max normalization method (mapping the data to the [0,1] interval) is used to standardize the data (i.e., the sample combinations of dynamic sensitive parameters and modal damping ratios) in both the input and output sample datasets. This avoids over-summing of large parameters during the prediction model training process due to differences in parameter magnitudes, while also improving the convergence speed of the model optimizer and reducing the number of training iterations. The Min-Max normalization formula is expressed as: In the formula, For data in the input or output sample dataset, The maximum value of this data. The minimum value of this data. The data is standardized (ultimately resulting in a standardized input sample dataset and a standardized output sample dataset).

[0047] Based on the standardized input sample dataset, dynamic sensitive parameter labels are added to the modal damping ratios in the standardized output sample dataset. The label format is a key-value pair of "dynamic sensitive parameter name-parameter value sample". The sample dataset is constructed based on the labeled output sample dataset to ensure that the correspondence between dynamic sensitive parameters and damping ratios can be established during the training of the prediction model.

[0048] Step 500, which involves training an initial gas bullet stability prediction model using a sample dataset, includes the following steps: dividing the sample dataset into a training set and a validation set according to a set ratio; performing data augmentation on the training set to obtain an augmented training set; setting the hyperparameters of the initial gas bullet stability prediction model, including the initial learning rate, batch size, maximum training epochs, diffusion steps, and weight decay coefficient; training the initial gas bullet stability prediction model using the augmented training set; and calculating the loss value of the trained initial gas bullet stability prediction model using the validation set and a loss function, continuing until the loss value of the trained initial gas bullet stability prediction model reaches a set condition, thus obtaining the gas bullet stability prediction model.

[0049] In the initial training of the aeroelastic stability prediction model, the Adam optimizer is used to update the hyperparameters. The loss function is the mean squared error function, expressed as: .

[0050] In the formula, Indicates the loss value. This indicates the number of samples in the validation set. Indicates the first in the verification set Modal damping ratio of each sample, The first output of the model represents the... Modal damping ratio of each sample.

[0051] For example, the sample dataset is divided into a training set (700 sets of modal damping ratio data and labels, used for learning prediction model parameters), a validation set (200 sets of modal damping ratio data and labels, used for tuning prediction model hyperparameters), and a test set (100 sets of modal damping ratio data and labels, used for evaluating the generalization ability of the prediction model) in a 7:2:1 ratio. Data augmentation is performed on the training set by adding small Gaussian noise to each set of modal damping ratio data. The Gaussian noise has a mean of 0 and a standard deviation (intensity baseline value) of 0.05. This process simulates the measurement errors of sensors in real-world engineering (e.g., the measurement errors of laser vibrometers), improving the robustness of the prediction model to noise and thus increasing the prediction accuracy.

[0052] A conditional diffusion network framework based on the Denoising Diffusion Probabilistic Model (DDPM) is constructed as the initial aeroelastic stability prediction model. This conditional diffusion network framework includes a noise addition module and a noise reduction module.

[0053] (1) Noise addition module, such as Figure 2 As shown, the noise addition module consists of a time encoding submodule and a Transformer submodule. Its core function is to gradually add Gaussian noise to the modal damping ratio data in the training set (or validation set) according to a preset number of diffusion steps until the data is completely transformed into pure noise. The specific process is as follows: First, the data-augmented modal damping ratio data and dynamic sensitive parameter label data in the training set (or the modal damping ratio data and dynamic sensitive parameter label data in the validation set) are mapped to the Transformer submodule through a 2-layer fully connected feedforward network (with ReLU activation function) to generate sequence data 1; Sequence data 1 is copied into two copies (denoted as data I and data II). Data I is processed by layer normalization to obtain sequence data A; Data II is processed by a 1-layer fully connected feedforward network and then added and fused with the time embedding vector (with the same dimension as data II) generated by the time encoding submodule to obtain sequence data B.

[0054] When training the initial aeroelastic stability prediction model, sequence data A and sequence data B are concatenated and used as input to the denoising module. This design can enhance the module's ability to perceive noise.

[0055] When using an aeroelastic stability prediction model for prediction, for example, the generalization ability of the model can be tested using data from a test set. The dynamic sensitive parameter values ​​sampled in the test set are used as the original sequence data C, which is then concatenated with sequence data A and sequence data B and used as the input to the denoising module. The final output is the predicted modal damping ratio corresponding to the dynamic sensitive parameter values ​​sampled in the test set.

[0056] (2) Noise reduction module, such as Figure 3 As shown. The core of the denoising module is the U-Net network, which includes a first residual convolutional submodule, a first downsampling submodule, a second residual convolutional submodule, a second downsampling submodule, a third residual convolutional submodule, a first Transformer submodule, a fourth residual convolutional submodule, a second Transformer submodule, a fifth residual convolutional submodule, a first upsampling submodule, a sixth residual convolutional submodule, a second upsampling submodule, a seventh residual convolutional submodule, a first feature fusion submodule, and a second feature fusion submodule (implementing the classic U-Net skip connections). Its function is to learn the denoising process from the noisy data and output the predicted value of the denoised modal damping ratio. The specific process is as follows: 1) Downsampling feature extraction stage: The output data of the noise addition module is input into the denoising module. First, it is input into the first residual convolution submodule (containing two 3×3 convolutional layers, a Batch Normalization layer, and a ReLU activation function) to extract shallow local features. Then, it is dimensionality reduced by the first downsampling submodule (using 2×2 max pooling with a stride of 2) to obtain the first downsampling feature map. Repeat the "residual convolution-downsampling" operation once: input the first downsampling feature map into the second residual convolution submodule (with the same structure as the first residual convolution submodule), and then reduce the dimensionality by the second downsampling submodule (with the same structure as the first downsampling submodule) to obtain the second downsampling feature map with a dimension of 1 / 4 of the initial data. This second downsampling feature map is then input into the third residual convolution submodule to further extract deep features.

[0057] 2) Global Feature Enhancement Stage: The output of the third residual convolution submodule is input to the first Transformer submodule (including a 2-head self-attention mechanism) to capture the global dependency relationship between parameters and damping ratio in the aeroelastic data; after the features are optimized by the fourth residual convolution submodule, they are input to the second Transformer submodule to enhance the global features a second time, resulting in a deep global feature map.

[0058] 3) Upsampling and Feature Fusion Stage: The deep global feature map is input into the fifth residual convolution submodule (with the same structure as the first residual convolution submodule), and then upsampled by the first upsampling submodule (using 2×2 transposed convolution with a stride of 2) to obtain the first upsampled feature map. At this time, the first upsampled feature map and the second downsampled feature map (corresponding to scale features) are channel-wise concatenated and fused by the first feature fusion submodule to supplement shallow local details. The fused data is input into the sixth residual convolution submodule to optimize features, and then upsampled by the second upsampling submodule (with the same structure as the first upsampling submodule) to obtain the second upsampled feature map. The second upsampled feature map and the first downsampled feature map are concatenated and fused by the second feature fusion submodule to further restore detailed features. 4) Output stage: The fused data is input into the seventh residual convolutional submodule to finally optimize the features. The feature channel number is mapped to the modal damping ratio data dimension through a 1×1 convolutional layer, and the output modal damping ratio prediction value is consistent with the input data dimension.

[0059] Among them, such as Figure 4 As shown, the Transformer submodules in the denoising module (including the first and second Transformer submodules) consist of an encoder and a decoder. Both include a multi-head self-attention mechanism (with 2 heads), residual connections and layer normalization, and a feedforward network module. The encoder is used to extract features from the input sequence data, and the decoder is used to generate the denoised features. This design can effectively capture the coupling relationship between parameters and damping ratio in aeroelastic data. The U-Net network can select a one-dimensional, two-dimensional, or three-dimensional structure according to the dimension of the input data (e.g., one-dimensional data of modal damping ratio as a single-channel sequence line graph, or two-dimensional or three-dimensional data of modal damping ratio as a multi-channel graph). The number of downsampling and upsampling operations can be adjusted according to the actual data volume (e.g., 3 times when the number of samples is >1000, and 2 times when the number of samples is ≤1000) and task requirements to ensure that the network complexity matches the data scale.

[0060] During the initial training of the gaseous stability prediction model, based on the characteristics of the gaseous data, the training hyperparameters were specifically set as follows: initial learning rate 1e-4, batch size 32, maximum training epochs 2000, linear scheduling as the noise scheduling strategy, diffusion steps T=800, and weight decay coefficient 1e-5. Mean squared error (MSE) was used as the loss function, which accurately measures the deviation between the predicted and actual values. The Adam optimizer was used (…). =0.9, =0.999) Update network parameters to improve training stability.

[0061] For the predictive model training and early stopping strategy, during training, the model loss is calculated using the validation set every 10 iterations. When the validation set loss no longer decreases for 50 consecutive iterations, the learning rate is reduced by 50% (i.e., multiplied by 0.5). When the learning rate is reduced to the minimum threshold 1e-7, and the validation set loss still shows no decreasing trend for 50 consecutive iterations, training is stopped. This early stopping strategy can effectively avoid model overfitting while ensuring that the model converges to the optimal state.

[0062] After training to obtain the aeroelastic stability prediction model, a test set is used to evaluate the model's generalization ability. The labels of the dynamic sensitive parameters in the test set are input into the aeroelastic stability prediction model (as the original sequence data C), and the model outputs the corresponding modal damping ratio prediction values. The stability of the rotorcraft is determined based on the positive or negative sign of the modal damping ratio prediction values.

[0063] Based on the above embodiments, this application generates a reliable dataset through nonlinear aeroelastic dynamics control equations and, combined with the strong modeling capability of the denoised diffusion probability model, constructs an aeroelastic stability prediction model adapted to the high-dimensional and strongly coupled characteristics of aeroelastic data. This effectively solves the problem of insufficient prediction accuracy of traditional methods and related data-driven methods, achieves accurate prediction of modal damping ratio, and thus accurately determines the aeroelastic stability state of rotorcraft. This provides an efficient and accurate technical solution for predicting the aeroelastic stability of aircraft during high-speed forward flight.

[0064] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0065] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for predicting the aeroelastic stability of a rotorcraft, characterized in that, include: Acquire parameter data of a rotorcraft; Based on the parameter data, the nonlinear aeroelastic dynamics control equations of the rotorcraft are constructed, and the dynamic sensitive parameters are determined; Define the value range of the dynamic sensitive parameter to construct an input sample dataset; the sample dataset includes multiple combinations of dynamic sensitive parameter value samples. Based on the input sample dataset and the nonlinear aeroelastic dynamics control equations, an output sample dataset corresponding to the input sample dataset is obtained. The output sample dataset includes sample modal damping ratios corresponding to sample combinations of all the values ​​of the dynamic sensitive parameters in the input sample dataset; Construct a sample dataset based on the input sample dataset and the output sample dataset; An initial aeroelastic stability prediction model is constructed based on a denoised diffusion probability model; The initial aeroelastic stability prediction model is trained using the sample dataset to obtain the aeroelastic stability prediction model; The operating conditions of the rotorcraft are acquired in real time to determine the values ​​of the rotorcraft's dynamic sensitive parameters; the values ​​of the dynamic sensitive parameters are input into the aeroelastic stability prediction model to obtain the corresponding modal damping ratio; and the aeroelastic stability of the rotorcraft is determined based on the modal damping ratio.

2. The method for predicting the aeroelastic stability of a rotorcraft according to claim 1, characterized in that, Based on the parameter data, the nonlinear aeroelastic dynamics control equations of the rotorcraft are constructed, and the dynamically sensitive parameters are determined, including: Dynamic units are divided according to the structural topology of the rotorcraft; Based on the parameter data, the elastic potential energy and kinetic energy of each dynamic unit are determined, and the external force work of the aerodynamic load is determined. Based on the elastic potential energy, kinetic energy, and external force work of each dynamic unit, a nonlinear aeroelastic dynamics control equation is constructed. Test factors are set based on the parameters of the rotorcraft, and the values ​​of the test factors are set to construct multiple test groups; each test group includes test factors and their values. L27 (3) 13 The orthogonal experimental design scheme and multiple test groups are used to simulate the nonlinear aeroelastic dynamics control equations to obtain the modal damping ratios corresponding to each test group; based on the modal damping ratios corresponding to each test group, the experimental factors are screened to obtain the dynamic sensitive parameters.

3. The method for predicting the aeroelastic stability of a rotorcraft according to claim 2, characterized in that, Based on the elastic potential energy, kinetic energy, and external work of each dynamic unit, a nonlinear aeroelastic dynamics governing equation is constructed, including: The aeroelastic dynamics equations are constructed based on the elastic potential energy, kinetic energy, and external force work of each dynamic unit using Hamilton's variational principle. The Leishman-Beddoes model was used to simulate the dynamic stall process of a rotorcraft airfoil in an unsteady airflow, and the aerodynamic coefficients under different rates of change of angle of attack were determined. The aeroelastic dynamics equations are modified based on the aerodynamic coefficients to obtain the nonlinear aeroelastic dynamics control equations.

4. The method for predicting the aeroelastic stability of a rotorcraft according to claim 3, characterized in that, Based on the aerodynamic coefficients, the aeroelastic dynamics equations are modified to obtain the nonlinear aeroelastic dynamics control equations, including: The aerodynamic coefficients are used to replace the static coefficients of the aerodynamic load terms in the aeroelastic dynamics equations to construct time-varying aerodynamic load terms; By embedding the time-varying aerodynamic load term into the aeroelastic dynamics equation, the nonlinear aeroelastic dynamics control equation is obtained.

5. The method for predicting the aeroelastic stability of a rotorcraft according to claim 1, characterized in that, Define the sample range of values ​​for the dynamic sensitive parameter to construct the input sample dataset, including: Based on the typical operating conditions of high-speed forward flight of a rotorcraft, the sample range of values ​​for the aforementioned dynamic sensitive parameters is determined; The Latin hypercube sampling method is used to sample the value sample interval to obtain multiple combinations of value samples of the dynamic sensitive parameters, so as to construct the input sample dataset.

6. The method for predicting the aeroelastic stability of a rotorcraft according to claim 1, characterized in that, Based on the input sample dataset and the nonlinear aeroelastic dynamics control equations, an output sample dataset corresponding to the input sample dataset is obtained, including: After inputting the sample values ​​of the dynamic sensitive parameters into the nonlinear aeroelastic dynamics control equation, the nonlinear aeroelastic dynamics control equation is converted into a state-space equation. The QR decomposition method is used to solve for the eigenvalues ​​of the state matrix in the state-space equations. Based on the eigenvalues, determine the modal damping ratio corresponding to the sample combination of values ​​of the dynamic sensitive parameter.

7. The method for predicting the aeroelastic stability of a rotorcraft according to claim 1, characterized in that, A sample dataset is constructed based on the input sample dataset and the output sample dataset, including: The Min-Max normalization method is used to standardize the sample combinations of dynamic sensitive parameters in the input sample dataset and the modal damping ratio in the output sample dataset, respectively, to obtain the standardized input sample dataset and the standardized output sample dataset. Based on the standardized input sample dataset, parameter labels are added to the modal damping ratios in the standardized output sample dataset to establish the correspondence between the sample combinations of dynamic sensitive parameters and the modal damping ratios, and a sample dataset is constructed based on the output sample dataset with added parameter labels.

8. The method for predicting the aeroelastic stability of a rotorcraft according to claim 1, characterized in that, The initial aeroelastic stability prediction model is trained using the sample dataset to obtain the aeroelastic stability prediction model, which includes: The sample dataset is divided into a training set and a validation set according to a set ratio; The training set is augmented to obtain an augmented training set; The hyperparameters of the initial aeroelastic stability prediction model are defined; the hyperparameters include the initial learning rate, batch size, maximum number of training rounds, number of diffusion steps, and weight decay coefficient. The initial gaseous stability prediction model is trained using the enhanced training set, and the loss value of the trained initial gaseous stability prediction model is calculated using the validation set and loss function. The process continues until the loss value of the trained initial gaseous stability prediction model reaches a set condition, thus obtaining the gaseous stability prediction model.

9. The method for predicting the aeroelastic stability of a rotorcraft according to claim 8, characterized in that, During the training of the initial aeroelastic stability prediction model, the Adam optimizer is used to update the hyperparameters.

10. The method for predicting the aeroelastic stability of a rotorcraft according to claim 8, characterized in that, The loss function is the mean squared error function, expressed as: ； In the formula, Indicates the loss value. This indicates the number of samples in the validation set. Indicates the first in the verification set Modal damping ratio of each sample, The first output of the model represents the... Modal damping ratio of each sample.

Images