Training and application method of tilt-rotor wing airfoil noise prediction model and related device
By constructing a tiltrotor airfoil noise prediction model with a dual-branch cooperative structure, the problems of high computational cost and low design efficiency in traditional methods are solved, achieving efficient and accurate noise prediction and meeting the design requirements of tiltrotor aircraft.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NANJING QIZHI AIRLINES TECHNOLOGY CO LTD
- Filing Date
- 2026-04-30
- Publication Date
- 2026-07-14
Smart Images
Figure CN122388560A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of machine learning technology, and in particular to a method and apparatus for training and applying a tiltrotor airfoil noise prediction model. Background Technology
[0002] When designing tiltrotor aircraft, the airfoil, as the genetic code of the tiltrotor, directly determines the overall acoustic characteristics, flight efficiency, and airworthiness compatibility of the tiltrotor aircraft. Unlike fixed-wing aircraft airfoils and traditional helicopter rotor airfoils, tiltrotor airfoils must simultaneously adapt to two extreme operating modes: in helicopter mode, they bear the primary function of lift, while in fixed-wing mode, they bear the combined function of thrust and lift. Furthermore, during the tilt transition state, they must withstand continuously changing angles of attack, Mach numbers, and unsteady flow fields. This unique operating condition switching characteristic makes its aeroacoustic response more complex, posing specific requirements for the acoustic design of the airfoil: "cross-mode compatibility, transition state stability, and controllable peak values."
[0003] Traditional methods for obtaining the acoustic characteristics of tiltrotor blades and airfoils primarily rely on forward simulations based on computational fluid dynamics (CFD) and extensive testing in experiments. This involves repeatedly screening airfoil shape parameters, calculating airfoil aeroacoustic characteristics, iterating acoustic responses under various operating conditions in large batches, and gradually optimizing parameters from massive amounts of data. Alternatively, it involves conducting state-specific noise tests on multiple batches of airfoil samples. This approach has significant limitations for tiltrotor airfoil design: extremely high computational costs, insufficient design efficiency and specificity, difficulty in meeting the specific design specifications of tiltrotor aircraft, and difficulty in accurate prediction. Summary of the Invention
[0004] The purpose of this application is to provide a method and related apparatus for training and applying a tiltrotor airfoil noise prediction model, which can improve the accuracy and efficiency of tiltrotor noise prediction.
[0005] To achieve the above objectives, this application provides the following solution: Firstly, this application provides a method for training a tiltrotor airfoil noise prediction model, including: Obtain the dataset; the dataset includes several tiltrotor airfoil sample parameters and aerodynamic noise sound pressure label data corresponding to each tiltrotor airfoil sample parameter; the tiltrotor airfoil sample parameters include sample shape parameters, sample motion state parameters and sample sound pressure time; the sample sound pressure time is determined by the aerodynamic noise sound pressure label data; The tiltrotor airfoil noise prediction model includes a peak sound pressure prediction network, a sound pressure trend prediction network, a peak calculation module, and a fusion module. The peak sound pressure prediction network inputs sample shape parameters and sample motion state parameters to predict the peak sound pressure formula parameters. These parameters include peak amplitude coefficient, peak width coefficient, and peak time coefficient. The sound pressure trend prediction network inputs sample shape parameters, sample motion state parameters, and sample sound pressure time to predict the sound pressure at each sample's sound pressure time. The peak calculation module substitutes the peak sound pressure formula parameters into the peak sound pressure calculation formula to calculate the peak sound pressure. The fusion module determines the sample noise sound pressure prediction data based on the peak sound pressure and the sound pressure at each sample's sound pressure time. The model parameters of the peak sound pressure prediction network and the sound pressure trend prediction network are updated using the loss function to obtain the updated tiltrotor airfoil noise prediction model. Determine if the iteration stopping condition has been met. If yes, then confirm the updated tiltrotor airfoil noise prediction model as the trained tiltrotor airfoil noise prediction model. If no, then increment the iteration count by one and return to the step of "inputting sample shape parameters and sample motion state parameters into the sound pressure peak prediction network" until the iteration stopping condition is met.
[0006] Secondly, this application provides a method for applying a tiltrotor airfoil noise prediction model, including: Acquire the target tiltrotor aircraft's external shape parameters, target motion state parameters, and target sound pressure level at time; The target shape parameters, target motion state parameters, and target sound pressure time are input into the trained tiltrotor airfoil noise prediction model to obtain the target tiltrotor noise sound pressure prediction data; the trained tiltrotor airfoil noise prediction model is the model trained using the above-mentioned tiltrotor airfoil noise prediction model training method.
[0007] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein the processor executes the computer program to implement the steps of the above-described tiltrotor airfoil noise prediction model training method or tiltrotor airfoil noise prediction model application method.
[0008] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described tiltrotor airfoil noise prediction model training method or tiltrotor airfoil noise prediction model application method.
[0009] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the above-described tiltrotor airfoil noise prediction model training method or tiltrotor airfoil noise prediction model application method.
[0010] According to the specific embodiments provided in this application, this application has the following technical effects: This application provides a method and related apparatus for training and applying a tiltrotor airfoil noise prediction model. By constructing a tiltrotor airfoil noise prediction model including a peak sound pressure prediction network, a sound pressure trend prediction network, a peak calculation module, and a fusion module, noise prediction is performed using the trained model after training, improving the noise prediction efficiency of the tiltrotor aircraft. The tiltrotor airfoil noise prediction model adopts a dual-branch collaborative structure to achieve complementary advantages: the global trend branch (sound pressure trend prediction network) ensures accurate fitting of the overall noise change trend under all operating conditions, ensuring the macroscopic stability of the prediction results; the local feature branch (sound pressure peak prediction network) strengthens the extraction of key features related to the sound pressure peak, ensuring the accuracy of core indicators. The two branches achieve collaborative optimization through a shared loss function and error adjustment mechanism, ensuring both the continuity of noise trend prediction and precise control of peak indicators, improving the noise prediction accuracy of the tiltrotor aircraft and perfectly matching the core requirements of tiltrotor airfoil acoustic optimization. Attached Figure Description
[0011] 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 described 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: Figure 1 This is an application environment diagram of a tiltrotor airfoil noise prediction model training method or a tiltrotor airfoil noise prediction model application method according to an embodiment of this application. Figure 2 This is a flowchart illustrating a method for training a tiltrotor airfoil noise prediction model, as provided in Embodiment 1 of this application. Figure 3 This is a schematic diagram illustrating the specific process of the tiltrotor airfoil noise prediction model training method provided in Embodiment 1 of this application; Figure 4 This is a schematic diagram of a grid provided in Embodiment 1 of this application; Figure 5 This is a schematic diagram of the tiltrotor airfoil noise prediction model provided in Embodiment 1 of this application; Figure 6This is a schematic diagram of the tiltrotor noise prediction principle provided in Embodiment 1 of this application; Figure 7 This is a schematic diagram comparing the training iteration speed, prediction results, and prediction errors of the traditional MLP method and the method of this application as provided in Embodiment 1 of this application; wherein, Figure 7 (a) in the figure is a comparison of the training iteration speed of the traditional MLP method and the method of this application; Figure 7 (b) in the figure is a comparison of the prediction results of the traditional MLP method and the method of this application; Figure 7 (c) in the figure is a comparison of the prediction errors of the traditional MLP method and the method of this application; Figure 8 This is a flowchart illustrating an application method for a tiltrotor airfoil noise prediction model provided in Embodiment 2 of this application; Figure 9 This is a schematic diagram of the structure of a computer device provided in Embodiment 3 of this application. Detailed Implementation
[0012] 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.
[0013] 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.
[0014] Example 1 The tiltrotor airfoil noise prediction model training method provided in this application embodiment can be applied to, for example... Figure 1In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be set up independently, integrated into server 104, or placed in the cloud or on other servers. Terminal 102 can send a dataset to server 104. After receiving the dataset, server 104 inputs the sample shape parameters and sample motion state parameters into the sound pressure peak prediction network to predict the sound pressure peak formula parameters. It also inputs the tiltrotor airfoil sample parameters into the sound pressure trend prediction network to predict the sound pressure at each sample's sound pressure moment. The peak calculation module substitutes the sound pressure peak formula parameters into the sound pressure peak calculation formula to calculate the sound pressure peak. The fusion module determines the sample noise sound pressure prediction data based on the sound pressure peak and the sound pressure at each sample's sound pressure moment. It then uses a loss function to update the model parameters of the sound pressure peak prediction network and the sound pressure trend prediction network to obtain a trained tiltrotor airfoil noise prediction model. Server 104 can feed back the trained tiltrotor airfoil noise prediction model for the dataset to terminal 102. Furthermore, in some embodiments, the tiltrotor airfoil noise prediction model training method can also be implemented independently by server 104 or terminal 102. For example, terminal 102 can directly train the model on the dataset, or server 104 can obtain the dataset from the data storage system and train the model on the dataset.
[0015] The terminal 102 can be, but is not limited to, various desktop computers and laptops. The server 104 can be implemented using a standalone server or a server cluster consisting of multiple servers, or it can be a cloud server.
[0016] In one exemplary embodiment, such as Figure 2 As shown, a method for training a tiltrotor airfoil noise prediction model is provided. This method is executed by a computer device, specifically by a terminal or server alone, or by both a terminal and a server. In this embodiment, the method is applied to... Figure 1 Taking server 104 as an example, the explanation includes the following steps 201 to 204.
[0017] Step 201: Obtain the dataset; the dataset includes several tiltrotor airfoil sample parameters and aerodynamic noise sound pressure label data corresponding to each tiltrotor airfoil sample parameter; the tiltrotor airfoil sample parameters include sample shape parameters, sample motion state parameters and sample sound pressure time; the sample sound pressure time is determined by the aerodynamic noise sound pressure label data.
[0018] Step 202: The tiltrotor airfoil noise prediction model includes a peak sound pressure prediction network, a sound pressure trend prediction network, a peak calculation module, and a fusion module. The sample shape parameters and motion state parameters are input into the peak sound pressure prediction network to predict the peak sound pressure formula parameters. The peak sound pressure formula parameters include the peak amplitude coefficient, peak width coefficient, and peak time coefficient. The tiltrotor airfoil sample parameters are input into the sound pressure trend prediction network to predict the sound pressure at each sample sound pressure time. The peak calculation module is used to: substitute the peak sound pressure formula parameters into the peak sound pressure calculation formula to calculate the peak sound pressure. The fusion module is used to determine the sample noise sound pressure prediction data based on the peak sound pressure and the sound pressure at each sample sound pressure time.
[0019] Step 203: Update the model parameters of the sound pressure peak prediction network and the sound pressure trend prediction network using the loss function to obtain the updated tiltrotor airfoil noise prediction model.
[0020] Step 204: Determine whether the iteration stopping condition has been met. If yes, then the updated tiltrotor airfoil noise prediction model is determined as the trained tiltrotor airfoil noise prediction model. If no, then the iteration count is incremented by one, and the process returns to the step of "inputting sample shape parameters and sample motion state parameters into the sound pressure peak prediction network" until the iteration stopping condition is met.
[0021] By implementing steps 201 to 204 above, a tiltrotor airfoil noise prediction model is constructed, comprising a sound pressure peak prediction network, a sound pressure trend prediction network, a peak calculation module, and a fusion module. After the model training is completed, the trained tiltrotor airfoil noise prediction model is used for noise prediction, thereby improving the noise prediction efficiency of the tiltrotor aircraft. The tiltrotor airfoil noise prediction model adopts a dual-branch collaborative structure to achieve complementary advantages: the global trend branch (sound pressure trend prediction network) ensures accurate fitting of the overall noise change trend under all operating conditions, ensuring the macroscopic stability of the prediction results; the local feature branch (sound pressure peak prediction network) strengthens the extraction of key features related to the sound pressure peak, ensuring the accuracy of core indicators. The two achieve collaborative optimization through sharing a loss function and error adjustment mechanism, which not only ensures the continuity of noise trend prediction but also achieves precise control of peak indicators, improving the noise prediction accuracy of the tiltrotor aircraft and perfectly matching the core requirements of tiltrotor airfoil acoustic optimization.
[0022] This design employs a dual-branch structure to specifically address the acoustic prediction challenges of tiltrotor airfoils. One branch focuses on global pattern learning, specifically exploring the mapping relationship between airfoil shape parameters, motion state parameters, sound pressure levels, and the overall trend of aerodynamic noise variation. This adapts to the continuously changing operating conditions of tiltrotors, from helicopter to fixed-wing modes, ensuring the stability of cross-mode noise trend prediction. The other branch focuses on capturing key local features to enhance the accuracy of predicting peak sound pressure levels for tiltrotor airfoils, particularly meeting the requirements for transitional peak control. Through the synergistic integration of these two branches, the design not only overcomes the shortcomings of traditional methods in considering both trends and features but also accurately matches the dual-mode and transitional-state specific characteristics of tiltrotor airfoils, providing targeted technical support for their efficient acoustic optimization design.
[0023] like Figure 3 As shown, the specific process is as follows: Extract shape parameters such as maximum thickness, maximum camber, and pitch center from various airfoil shapes; calculate aerodynamic noise for various airfoils under different conditions to obtain sound pressure time histories; construct an airfoil aerodynamic noise database using the airfoil shape parameters, motion state parameters, sound pressure, and corresponding times; based on the sound pressure time, sound pressure value, distribution width, and airfoil geometry and motion state parameters in this aerodynamic noise database, construct an MLP network (i.e., a sound pressure peak prediction network) from airfoil shape and motion state parameters to sound pressure peak formula parameters, and calculate the sound pressure peak by combining the predicted sound pressure peak formula parameters and corresponding sound pressure time with the formula; construct a trend MLP network (i.e., a sound pressure trend prediction network) based on the sound pressure time, airfoil shape parameters, and motion state parameters, etc., to map the relationship from airfoil shape, motion state, and sound pressure time to sound pressure. Specifically, the CST method can be used for fitting and interpolation to obtain the sample shape parameters.
[0024] First, sample shape data of the airfoil needs to be obtained. The sample shape parameters of the airfoil are obtained by fitting and interpolating using the CST method. The upper and lower surfaces of the airfoil can be expressed by the formula: (1); in, The ordinate is represented by the dimensionless chord length. This is a class function used to characterize the degree of curvature of the airfoil's leading edge; and These are the shape function coefficients used to control the external shape parameters; A shape function used to describe the shape of the airfoil surface. The x-coordinate is represented by the dimensionless representation of the chord length; The distance between a point on the airfoil surface and the X-axis is the ratio of the distance to the thickness of the upper trailing edge; in the above formula (1), the subscript u represents the upper surface and the subscript l represents the lower surface. I This represents the number of parameters. Finally, the X and Y coordinates of the airfoil in the Cartesian coordinate system are obtained.
[0025] Based on the above description, the geometric information of the airfoil, such as its thickness and camber, can be obtained through calculation: (2); in, This indicates the airfoil thickness at the current location; This indicates the airfoil thickness at the current position; max indicates taking the maximum value; c indicates the current position relative to the airfoil chord length. These represent the ordinates of the airfoil curve points at their current positions on the upper and lower surfaces, respectively.
[0026] Aerodynamic noise data was obtained by solving the aerodynamic noise characteristics of the airfoil using fluid dynamics (CFD) and acoustic analogy methods. Through the aforementioned airfoil geometric control, the final airfoil coordinate positions X and Y were input to form a computational mesh, as illustrated below. Figure 4 As shown.
[0027] For CFD calculations of airfoils, the following state parameters need to be input: Reynolds number Re, reduced frequency kf, incoming Mach number Ma, and average angle of attack Aoa. The formulas for calculating the Reynolds number and incoming Mach number are as follows: (3); in, For density, For the incoming flow velocity, For the chord length, The gas constant is... It is the speed of sound.
[0028] To more clearly demonstrate the CFD solution process, the Reynolds-Averaged Navier-Stokes Equations (RANSEquations) in three dimensions are used for explanation. In this method, the governing equations are: (4); in, Represents partial derivatives; As a conserved variable, These are the components of velocity along the X, Y, and Z axes, respectively. The total energy per unit mass of gas, indicated by a superscript. T Indicates transpose; For circulation volume; For viscous flux, intermediate quantity , The specific expressions are shown in equations (5) and (6) below; This is a vector area element.
[0029] (5); (6); in, Pressure; For heat flow, These are the components in the corresponding directions of the X, Y, and Z axes, respectively; For stress components, These are the components corresponding to the directions of the subscript symbols.
[0030] In viscous flux, the expressions for the heat flux and stress components in each direction are: (7); in, The viscosity coefficient, The thermal conductivity coefficient, For temperature.
[0031] To ensure the system of equations is closed, solve it using the following formula: (8); in, It is a constant.
[0032] After the flow calculation converges, the results are output. Since numerical calculations are used, the value of each physical quantity differs at different grid points; therefore, the subscript 'i' indicates the grid order. According to the F-1A formula, the flow output shows the pressure at the grid center of the airfoil surface. Grid area The components I of the mesh normal vector on the X and Y axes x I y (Calculated from the coordinates of adjacent grid points), velocity in the X direction Y-direction velocity X-axis coordinates and Y-axis coordinates There are 6 parameters. Then, the aerodynamic noise is solved using the F-1A formula to obtain the corresponding time (Time) and noise sound pressure (SP). The sound pressure formula is as follows: (9); in, Indicates sound pressure level. Represents thickness noise. This represents load noise.
[0033] Since the airfoil itself is two-dimensional and does not have a three-dimensional thickness, only the load noise expression is given: (10); in, Represents pi; This represents the speed of sound in still air. Indicates the airfoil surface; The subscript represents the distance vector from the sound source to the observation point. Indicates the component in the X-axis direction, superscript Represents a unit vector; This represents the load derivative with respect to time. l The load vector (through pressure p and normal vector component I) x I y calculate); Indicates the distance from the source to the observation point; ret This indicates the delay time (i.e., the time it takes for a sound wave to travel from the sound source to the receiving point). This represents the velocity of the sound source pointing towards the observation point divided by the speed of sound, with the subscript r indicating the component in the direction from the sound source to the observation point; This represents the velocity component along the X-axis divided by the speed of sound. Represents area in yuan.
[0034] Using the data obtained from the above steps, an aerodynamic noise database is constructed containing various airfoil shapes, motion parameters, and corresponding airfoil aerodynamic noise sound pressure levels. The samples used for sound pressure trend prediction and those used for predicting peak sound pressure levels have the same structure, differing only in input and usage methods. Each sample pair should have the following structure: shape parameters (thickness, camber, chord length, airfoil name), motion parameters (Reynolds number, reduced frequency, incoming Mach number, average angle of attack), and aerodynamic noise sound pressure label data (sound pressure time, sound pressure value).
[0035] In a specific formula, a dataset can be constructed from 1000 sample pairs. The sample shape parameters in the dataset include the airfoil thickness, camber, chord length, and airfoil name; the sample motion state parameters include Reynolds number, reduction frequency, incoming Mach number, and mean angle of attack.
[0036] The structure of the tiltrotor airfoil noise prediction model is as follows: Figure 5 As shown, the peak sound pressure prediction network and the sound pressure trend prediction network can be MLP models.
[0037] MLP is a fully connected neural network consisting of an input layer, at least one hidden layer, and an output layer. Its mapping relationship can be expressed by the formula: (11); in, For the input vector, For output quantity, For the first Layer weight matrix, For the first Layer bias vector, For the first The activation function of the layer.
[0038] The dataset is divided into training and testing sets. Training on this dataset allows the construction of an MLP network that integrates external parameters, motion state parameters, and sound pressure levels from time to sound pressure. The specific training process for the MLP network is as follows: 1) Initialize parameters: Initialize the model parameters (including the weight matrix and bias vector of the sound pressure peak prediction network and the sound pressure trend prediction network). The parameters can be normally distributed.
[0039] 2) Forward Propagation: The data in the dataset is divided into multiple batches, but small batches of samples are input (x: thickness, camber, chord length, airfoil name, Reynolds number, reduced frequency, incoming Mach number, average angle of attack, sound pressure level). The propagation starts from the input layer and continues to the output layer, as shown below: (12); (13); in, For the first The layer's output (activation value). for The value after linear transformation for After activation function After activation The output value of the layer.
[0040] 3) Loss Function: In the training process of neural network models, the loss function is an important indicator used to measure the model's predictive ability and the difference between the predicted results and the actual results. The loss function directly affects the training effect and generalization ability of the model. Therefore, a good loss function is one of the important factors in evaluating the model's ability. The loss function used in the sound pressure peak prediction network and sound pressure trend prediction network of this application is the mean squared error function, where the total loss for a small batch of sample input is calculated as follows; (14); in, The value of the loss function. For the sample size, For the first Aerodynamic noise sound pressure label data corresponding to the parameters of each tiltrotor airfoil sample. For the first Predicted noise and sound pressure data for each tiltrotor airfoil sample parameter.
[0041] 4) Backpropagation: Adjust the weight matrices and bias vectors of the peak sound pressure prediction network and the sound pressure trend prediction network through backpropagation, and calculate the loss function. For each weight matrix and bias vector The gradient.
[0042] (5) Related parameter updates: The method of updating parameters directly determines the adjustment method of parameters such as weights and biases, which in turn affects the speed and direction of the model approaching the correct solution. In this application, the gradient descent algorithm is used to update the weight matrix and bias vector: (15); (16); in, This is the learning rate.
[0043] 6) The iteration stops when the maximum number of iterations is reached. When the maximum number of iterations is reached, the iteration stops, and the model parameters obtained from the current number of iterations are used to determine an updated tiltrotor airfoil noise prediction model. This updated tiltrotor airfoil noise prediction model is then used as the trained tiltrotor airfoil noise prediction model.
[0044] In this application, based on traditional MLP applications, a formula for describing the peak sound pressure level is proposed to capture the peak sound pressure level independently, thereby enhancing the model's ability to capture highly nonlinear and transient characteristics. The formula for calculating the peak sound pressure level is as follows: (17); in, express Peak sound pressure level at any given moment; This represents the number of peak sound pressure levels. Indicates the first Each amplitude value, and all peak values constitute the peak amplitude coefficient; The base is the natural number; Indicates the corresponding number The time of each amplitude value, and the time of all amplitude values constitute the peak time coefficient; It is the peak width factor, used to measure the proportion of the sound pressure peak in the time axis.
[0045] The input to the peak sound pressure prediction network is the airfoil's shape parameters and motion parameters. The model predicts the peak sound pressure parameters in equation (17). and Then, the peak sound pressure is calculated by substituting the corresponding sound pressure time t from the sample into equation (17). .
[0046] Thus, a perceptron (sound pressure peak prediction network) has been formed, which uses the airfoil's shape parameters (thickness, camber, airfoil name, chord length) and motion state parameters (reduced frequency, incoming Mach number, Reynolds number, average angle of attack) as inputs and the sound pressure at the corresponding moment as the target to predict the parameters of the sound pressure peak formula. A perceptron (sound pressure trend prediction network) is also formed, which uses the airfoil's shape parameters (thickness, camber, airfoil name, chord length) and motion state parameters (reduced frequency, incoming Mach number, Reynolds number, average angle of attack) and the sound pressure time as inputs and the sound pressure at the corresponding moment as the target to train for sound pressure trend prediction. Both perceptrons are composed of MLP models.
[0047] The differences between this application and traditional perceptrons are: 1) MLP refers to fully connected adjacent neural layers, while the two MLPs in this model are not connected to each other, only using some parameters in the input part. 2) The sound pressure peak prediction network and the sound pressure trend prediction network in this application share a loss function and an error to update the neuron weights of the two parts, thereby determining... Harmony and sound pressure And calculate the peak sound pressure level. Traditional MLP models typically require two error functions and two loss functions. 3) In this embodiment, the sound pressure trend prediction network directly predicts a portion of the sound pressure, while the prediction of the peak sound pressure is achieved by establishing a mapping between the shape parameters and motion state parameters and the peak sound pressure formula parameters, then calculating a portion of the sound pressure based on the time, and finally weighting the two portions to obtain the complete sound pressure.
[0048] Furthermore, due to significant numerical differences in noise parameters and other parameters, all input parameters (tilt rotor airfoil sample parameters) were normalized using a range of maximum and minimum values before being input into the model to facilitate network training. The principle of tiltrotor noise prediction is as follows: Figure 6 As shown.
[0049] The performance of the tiltrotor airfoil noise prediction model proposed in this application is verified below. It is compared with the traditional MLP method (which inputs shape parameters and motion state parameters into an MLP model to obtain sample noise and sound pressure prediction data). The aerodynamic noise characteristics of a certain tiltrotor airfoil state are calculated and predicted using a test set, and the results are compared with the predictions of the traditional MLP model. Figure 7 It can be seen that the model proposed in this embodiment has a faster convergence speed than the traditional MLP method. Secondly, in terms of prediction ability, whether it is the peak value or the overall trend, the prediction error of the method proposed in this application is smaller than that of the traditional MLP method.
[0050] Traditional CFD and FW-H methods offer high accuracy but have relatively long resolution times (typically requiring half a day or even a day to acquire sound pressure data for a single state) and involve massive amounts of data. Furthermore, computational accuracy is affected by multiple parameters, including format and grid. This application utilizes a neural network prediction method, enabling the acquisition of results for states within the input parameter range within 10 seconds after model training.
[0051] The main characteristic of neural network methods is their fast prediction speed, which can meet the design requirements of rapid engineering iteration. However, for highly nonlinear and transient characteristics such as sound pressure, traditional single neural network methods are difficult to capture the features of sound pressure. Therefore, this application proposes a formula to describe the peak sound pressure, which calculates the main peak sound pressure by predicting the coefficients of another part of the neural network formula, thus making up for the insufficient capture capability of traditional single MLP neural networks.
[0052] This application has the following advantages: (1) The prediction efficiency is significantly improved, and the computational cost is greatly reduced. Traditional methods rely on CFD forward simulation or batch experiments to obtain the acoustic characteristics of tiltrotor airfoils. Faced with complex dynamic flow fields, a single calculation can take several days or even weeks and consume a lot of computing resources, resulting in a long development cycle. This application constructs an aerodynamic noise database covering multiple airfoils and multiple operating conditions, and establishes a direct mapping relationship between airfoil shape parameters, motion state parameters and sound pressure based on an MLP network, without the need for iterative solutions to complex flow fields. After the model is trained, only the shape parameters such as the thickness, camber, and chord length of the target airfoil and the motion state parameters such as the reduced frequency and Mach number need to be input. The complete time-domain sound pressure curve and peak information can be output in milliseconds. The prediction time is significantly shortened compared with traditional CFD methods, which significantly reduces the development cost and cycle. (2) Strong adaptability to multiple operating conditions. Traditional deep learning models are mostly trained for single or limited operating conditions, which makes it difficult to cope with the continuous flow field changes of tiltrotor from helicopter mode to fixed-wing mode. The prediction accuracy drops significantly in operating conditions with strong nonlinearity and prominent dynamics, such as transition state. This application adopts a dual-branch network structure. The global trend branch deeply mines the mapping relationship between parameters and the overall evolution law of noise, accurately capturing the continuous change trend of noise characteristics across operating conditions. Combined with the input parameter normalization processing to eliminate the influence of numerical differences, the model maintains stable prediction performance in all flight states such as helicopter mode, fixed-wing mode and transition state. The cross-operating condition prediction error is reduced by more than 30% compared with the traditional MLP method, which solves the pain point of insufficient multi-operating condition adaptability of traditional models. (3) Accurate prediction of peak sound pressure level. Peak sound pressure level is the core indicator of whether the airfoil meets acoustic standards. It is affected by the coupling of multiple factors. Traditional models mostly focus on predicting the overall noise level and have weak ability to locate the amplitude and time of the peak, which easily leads to the engineering risk of "overall compliance but peak exceeding the standard". This application uses a dedicated peak sound pressure level description formula to quantify the correlation between the peak and the time ratio. Then, combined with local feature branches, it achieves triple accurate locking of the peak sound pressure level amplitude, occurrence time and time ratio, and the peak prediction deviation is controlled within 5%. It can accurately predict whether the peak meets the standard in the early stage of design and avoid design rework caused by the peak exceeding the standard in the later test. (4) Balancing trend fitting and feature extraction to meet dual engineering requirements. Traditional deep learning models struggle to balance overall noise trend fitting and local peak feature extraction, either ignoring key peak features or failing to control the overall trend, thus failing to meet dual engineering requirements. The dual-branch collaborative structure of this application achieves complementary advantages: the global trend branch ensures accurate fitting of the overall noise change trend under all operating conditions, ensuring the macroscopic stability of the prediction results; the local feature branch strengthens the extraction of key features related to the sound pressure peak, ensuring the accuracy of core indicators. The two achieve collaborative optimization through sharing a loss function and error adjustment mechanism, ensuring both the continuity of noise trend prediction and the precise control of peak indicators, perfectly matching the core requirements of tiltrotor airfoil acoustic optimization; (5) Simplified design process. Traditional airfoil acoustic optimization design relies heavily on the experience and expertise of engineers, requiring repeated parameter screening and iterative calculations, resulting in a long design cycle and difficulty in fully exploring the design space. The prediction method in this application can be directly embedded into the conventional design process without reconstructing the existing system. Engineers can quickly evaluate the acoustic effects of different parameter combinations without complex trial and error, accelerating the selection of the optimal solution. This transforms airfoil acoustic design from experience-driven to data-driven, significantly shortening the R&D iteration cycle.
[0053] This application also provides an application scenario in which the above-described tiltrotor airfoil noise prediction model training method is applied. Specifically, the tiltrotor airfoil noise prediction model training method provided in this embodiment can be applied to a tiltrotor noise prediction scenario. The tiltrotor noise prediction scenario includes a data acquisition stage, a model training link, and a tiltrotor noise prediction stage. The dataset enters the model training link from the data acquisition stage, and through human-machine collaboration, a trained tiltrotor airfoil noise prediction model is obtained, which then enters the downstream tiltrotor noise prediction stage. The tiltrotor airfoil noise prediction model training method provided in this embodiment belongs to the model training link. Specifically, in the tiltrotor noise prediction process for the dataset, the sample shape parameters and sample motion state parameters can be input into the sound pressure peak prediction network to predict the sound pressure peak formula parameters. The tiltrotor airfoil sample parameters can be input into the sound pressure trend prediction network to predict the sound pressure at each sample sound pressure moment. The peak calculation module substitutes the sound pressure peak formula parameters into the sound pressure peak calculation formula to calculate the sound pressure peak. The fusion module determines the sample noise sound pressure prediction data based on the sound pressure peak and the sound pressure at each sample sound pressure moment. The model parameters of the sound pressure peak prediction network and the sound pressure trend prediction network are updated using a loss function to obtain the trained tiltrotor airfoil noise prediction model.
[0054] Example 2 The tiltrotor airfoil noise prediction model and application method provided in this application embodiment can be applied to, for example, Figure 1In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be set up independently, integrated into server 104, or placed in the cloud or on another server. Terminal 102 can send target shape parameters, target motion state parameters, and target sound pressure time to server 104. After receiving these parameters, server 104 inputs them into a trained tiltrotor airfoil noise prediction model to obtain the target tiltrotor noise and sound pressure prediction data. Server 104 can then feed back the obtained noise and sound pressure prediction data for the target shape parameters, target motion state parameters, and target sound pressure time to terminal 102. Furthermore, in some embodiments, the tiltrotor airfoil noise prediction model application method can also be implemented separately by the server 104 or the terminal 102. For example, the terminal 102 can directly perform tiltrotor noise prediction based on the target shape parameters, target motion state parameters, and target sound pressure time. Alternatively, the server 104 can obtain the target shape parameters, target motion state parameters, and target sound pressure time from the data storage system and perform tiltrotor noise prediction based on the target shape parameters, target motion state parameters, and target sound pressure time.
[0055] The terminal 102 can be, but is not limited to, various desktop computers and laptops. The server 104 can be implemented using a standalone server or a server cluster consisting of multiple servers, or it can be a cloud server.
[0056] In one exemplary embodiment, such as Figure 8 As shown, a method for predicting airfoil noise of a tiltrotor aircraft is provided, including the following steps S1~S2.
[0057] S1: Obtain the target tiltrotor aircraft's shape parameters, motion state parameters, and target sound pressure time. The target sound pressure time at this point is the time when the noise sound pressure prediction data for the desired model output is available.
[0058] S2: Input the target shape parameters, target motion state parameters, and target sound pressure time into the trained tiltrotor airfoil noise prediction model to obtain the target tiltrotor noise sound pressure prediction data; the trained tiltrotor airfoil noise prediction model is the model trained using the tiltrotor airfoil noise prediction model training method described in Example 1.
[0059] This application also provides an application scenario in which the above-described tiltrotor airfoil noise prediction model application method is applied. Specifically, the tiltrotor airfoil noise prediction model application method provided in this embodiment can be applied in the design scenario of tiltrotor aircraft. The tiltrotor aircraft design scenario includes a data acquisition stage, a tiltrotor noise prediction link, and a tiltrotor aircraft design stage. The data to be processed (target shape parameters, target motion state parameters, and target sound pressure time) enters the tiltrotor noise prediction link from the data acquisition stage, and obtains the corresponding noise and sound pressure prediction data through human-machine collaboration, and then enters the downstream tiltrotor aircraft design stage. The tiltrotor airfoil noise prediction model application method provided in this embodiment belongs to the machine marking stage in the tiltrotor noise prediction link. Specifically, in the tiltrotor noise prediction process for the data to be processed, the data to be processed can be labeled using a collaborative method of machine labeling and manual labeling. This involves inputting the target shape parameters, target motion state parameters, and target sound pressure into the trained tiltrotor airfoil noise prediction model to obtain the target tiltrotor noise and sound pressure prediction data.
[0060] Example 3 In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 9 As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores tiltrotor noise prediction data. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a tiltrotor airfoil noise prediction model training method or a tiltrotor airfoil noise prediction model application method.
[0061] Those skilled in the art will understand that Figure 9The structures shown are merely block diagrams of some structures related to the present application and do not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than shown in the figures, or combine certain components, or have different component arrangements. In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.
[0062] Example 4 In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.
[0063] Example 5 In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.
[0064] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Moreover, the collection, use and processing of the relevant data are carried out in compliance with the relevant data protection laws and policies of the country where the location is located, and with the authorization granted by the owner of the corresponding device.
[0065] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).
[0066] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.
[0067] 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.
[0068] 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 training a tiltrotor airfoil noise prediction model, characterized in that, The training method for the tiltrotor airfoil noise prediction model includes: Obtain the dataset; the dataset includes several tiltrotor airfoil sample parameters and aerodynamic noise sound pressure label data corresponding to each tiltrotor airfoil sample parameter; the tiltrotor airfoil sample parameters include sample shape parameters, sample motion state parameters and sample sound pressure time; the sample sound pressure time is determined by the aerodynamic noise sound pressure label data; The tiltrotor airfoil noise prediction model includes a peak sound pressure prediction network, a sound pressure trend prediction network, a peak calculation module, and a fusion module. The peak sound pressure prediction network inputs sample shape parameters and sample motion state parameters to predict the peak sound pressure formula parameters. These parameters include peak amplitude coefficient, peak width coefficient, and peak time coefficient. The airfoil sample parameters are input into the sound pressure trend prediction network to predict the sound pressure at each sample's sound pressure time. The peak calculation module is used to substitute the peak sound pressure formula parameters into the peak sound pressure calculation formula to calculate the peak sound pressure. The fusion module is used to determine the sample noise sound pressure prediction data based on the peak sound pressure and the sound pressure at each sample's sound pressure time. The model parameters of the peak sound pressure prediction network and the sound pressure trend prediction network are updated using the loss function to obtain the updated tiltrotor airfoil noise prediction model. Determine whether the iteration stopping condition has been met. If yes, then confirm the updated tiltrotor airfoil noise prediction model as the trained tiltrotor airfoil noise prediction model. If not, then increment the iteration count by one and return to the step of "inputting sample shape parameters and sample motion state parameters into the sound pressure peak prediction network" until the iteration stopping condition is met.
2. The method for training a tiltrotor airfoil noise prediction model according to claim 1, characterized in that, The formula for calculating peak sound pressure level is as follows: ； in, express Peak sound pressure level at any given moment; This represents the number of peak sound pressure levels. Indicates the first Each amplitude value, and all peak values constitute the peak amplitude coefficient; The base is the natural number; Indicates the corresponding number The time of each amplitude value, and the time of all amplitude values constitute the peak time coefficient; It is the peak width coefficient.
3. The method for training a tiltrotor airfoil noise prediction model according to claim 1, characterized in that, The peak sound pressure prediction network and the sound pressure trend prediction network are MLP models.
4. The method for training a tiltrotor airfoil noise prediction model according to claim 1, characterized in that, The sample's shape parameters include the airfoil's thickness, camber, chord length, and airfoil name; the sample's motion parameters include the Reynolds number, reduction frequency, incoming Mach number, and mean angle of attack.
5. The method for training a tiltrotor airfoil noise prediction model according to claim 1, characterized in that, The iteration stops when the number of iterations reaches the maximum number of iterations.
6. The method for training a tiltrotor airfoil noise prediction model according to claim 1, characterized in that, The training method for the tiltrotor airfoil noise prediction model also includes: using the CST method to fit and interpolate to obtain the sample shape parameters.
7. A method for predicting airfoil noise in tiltrotor aircraft, characterized in that, The application method of the tiltrotor airfoil noise prediction model includes: Acquire the target tiltrotor aircraft's external shape parameters, target motion state parameters, and target sound pressure level at time; The target shape parameters, target motion state parameters, and target sound pressure time are input into the trained tiltrotor airfoil noise prediction model to obtain the target tiltrotor noise sound pressure prediction data; the trained tiltrotor airfoil noise prediction model is a model trained using the tiltrotor airfoil noise prediction model training method according to any one of claims 1-6.
8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that the processor executes the computer program to implement the tiltrotor airfoil noise prediction model training method according to any one of claims 1-6 or the tiltrotor airfoil noise prediction model application method according to claim 7.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the tiltrotor airfoil noise prediction model training method according to any one of claims 1-6 or the tiltrotor airfoil noise prediction model application method according to claim 7.
10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the tiltrotor airfoil noise prediction model training method according to any one of claims 1-6 or the tiltrotor airfoil noise prediction model application method according to claim 7.