A multi-fidelity fusion method and system for aircraft distributed loads combined with spatial coordinates

Abstract

This invention discloses a multi-fidelity fusion method and system for aircraft distributed loads combined with spatial coordinates, comprising: (1) calculating the distributed load dataset of the aircraft under different operating conditions through numerical simulation, defining it as low-fidelity aerodynamic data; obtaining the distributed load dataset of the aircraft under different operating conditions through actual experiments, defining it as high-fidelity aerodynamic data; (2) in the pre-trained original data fusion layer, using high-fidelity data to initially fuse the low-fidelity aerodynamic data, improving the quality of the low-fidelity aerodynamic data; (3) in the pre-trained custom spatial coordinate data layer, optimizing the spatial coordinate features of the high-fidelity data and the initially fused low-fidelity data, generating a custom high-fidelity data and a custom low-fidelity data combination; (4) in the pre-trained combined spatial coordinate data fusion layer, further fusing the custom high-fidelity data and the custom low-fidelity data. Using this invention, high-precision distributed load data of aircraft can be obtained at low cost.

Description

Technical Field

[0001] This invention belongs to the field of aircraft aerodynamics, and in particular relates to a multi-fidelity fusion method and system for aircraft distributed loads that incorporates spatial coordinates. Background Technology

[0002] In the field of aerodynamics, wind tunnel experiments, numerical calculations, and simulated flight tests are commonly used to obtain aerodynamic data for aircraft. However, each method has its advantages and disadvantages. Currently, improving the accuracy of any of these aerodynamic assessment methods in a short period of time is extremely difficult and costly.

[0003] For example, Chinese patent document CN112182771A discloses a data processing method based on numerical simulation, including: determining the flow field numerical calculation type in fluid dynamics numerical simulation; correcting the flow field numerical values ​​solved by the NS equation solution type, the flow field numerical values ​​solved by the boundary condition solution type, and the flow field numerical values ​​solved by the turbulence equation solution type according to the flow field numerical calculation type; and performing numerical simulation of the aerodynamic or thermal characteristics of the aircraft based on the correction results.

[0004] Chinese patent document CN114168796A discloses a method for establishing an upper aerodynamic database for an aircraft, relating to the field of aircraft research. The method includes: obtaining the aircraft's external shape parameters; setting state parameters and generating several virtual flight states based on the state parameters; performing numerical simulation calculations on each virtual flight state based on the aircraft's external shape parameters, obtaining corresponding aerodynamic data for each virtual flight state; and obtaining an upper aerodynamic database for the aircraft based on the aerodynamic data corresponding to all the virtual flight states.

[0005] To shorten development cycles and reduce design costs, establishing aerodynamic surrogate models has attracted significant attention as an efficient means of acquiring aerodynamic data. The establishment of such models typically involves understanding the physical characteristics and mathematical statistical laws of the aircraft, or using fitting and interpolation of existing data to predict responses under arbitrary flight conditions. While establishing aerodynamic surrogate models is an efficient method for acquiring aerodynamic data, a sufficient and high-precision aerodynamic data is a prerequisite for building a good model. This is because it is difficult to significantly improve the accuracy of aerodynamic data in a surrogate model using only a single data point.

[0006] To address the aforementioned issues, there is an urgent need to design a new data fusion technology to integrate aerodynamic data from different sources, thereby improving the quality and applicability of aerodynamic research. Furthermore, data accuracy can be enhanced through data management, data fusion, and data reuse technologies. Summary of the Invention

[0007] This invention provides a method and system for multi-fidelity fusion of distributed loads of aircraft using spatial coordinates. It can obtain high-precision distributed load data of aircraft at low cost by utilizing limited experimental data and numerical simulation data.

[0008] A multi-fidelity fusion method for distributed payloads of an aircraft that incorporates spatial coordinates includes the following steps:

[0009] (1) The distributed load dataset of the aircraft under different operating conditions is obtained by numerical simulation calculation and is defined as low-fidelity aerodynamic data; the distributed load dataset of the aircraft under different operating conditions is obtained by actual test and is defined as high-fidelity aerodynamic data.

[0010] (2) In the pre-trained raw data fusion layer, high-fidelity data is used to perform preliminary fusion of low-fidelity aerodynamic data to obtain high-quality low-fidelity aerodynamic data, while retaining the original high-fidelity aerodynamic data.

[0011] (3) In the pre-trained custom spatial coordinate data layer, optimize the spatial coordinate features of high-fidelity aerodynamic data and low-fidelity aerodynamic data after preliminary fusion, and generate custom high-fidelity data and custom low-fidelity data sample sets.

[0012] (4) In the pre-trained spatial coordinate data fusion layer, the spatial coordinate feature learning of sample points is introduced to enrich the relationship information between data sample points. The custom high-fidelity and custom low-fidelity data sets in the previous layer are further fused to obtain the final fused aircraft target distribution payload data.

[0013] In step (2), the network structure of the original data fusion layer is a fully connected layer. During the pre-training process, the losses of the two high-fidelity original datasets and the low-fidelity original datasets are calculated separately, using the following formula:

[0014]

[0015] in, The predicted values ​​are calculated in the forward pass of the high-fidelity raw data at the raw data fusion layer. For low-fidelity aerodynamic data, the forward calculation prediction values ​​are generated at the raw data fusion layer. For high-fidelity raw data, The data consists of low-fidelity raw data, l represents the number of high-fidelity raw data points, and q represents the number of low-fidelity raw data points; the total loss is calculated using a weighted average method. Calculate the total loss using a weighted method. The formula is as follows:

[0016]

[0017] Here, ρ is an adjustable parameter, and the optimal solution is found through grid optimization.

[0018] In step (3), the custom spatial coordinate data layer includes a generator and a discriminator. The generator is used to generate samples with specified spatial coordinate features. During pre-training, the loss function of the generator is...

[0019]

[0020] The discriminator selects a custom dataset that matches the real data. The discriminator's loss function is:

[0021]

[0022] Where z is a random noise vector from a custom space, x is a real sample, G(z) is a generated data sample with random noise as input and generated as output, D(x) represents the probability that sample x is identified as a real sample, and D(G(z)) represents the probability that the generated sample is identified as a real sample. p represents the average of all possible inputs in the latent space. data (x) represents the distribution of the real data, p z (z) represents the distribution of the potential space.

[0023] In step (4), the spatial coordinate data fusion layer is combined with the encoder and decoder to generate spatial position sequence relationships between samples, and the self-attention mechanism is used to focus on the spatial coordinate sequence between samples, capture the relationship between different samples, and extract the detailed features of the data.

[0024] During the pre-training process of the spatial coordinate data fusion layer, the total loss is calculated using a weighted method. The formula is

[0025]

[0026] In the formula, and To combine spatial coordinate data with the fusion layer for forward calculation of predicted values, For high-fidelity raw data, The data consists of low-fidelity raw data, m represents the number of high-fidelity raw data sets, n represents the number of low-fidelity raw data sets, and λ is an adjustable parameter used to control the proportion of the loss of the two datasets in the total loss. This layer combines the sample space coordinate sequence to further fuse the custom high-fidelity and low-fidelity data.

[0027] Within the HyperBand multi-armed gambling machine framework, Bayesian optimization is used to select the configuration of the adjustable parameter λ. The specific process is as follows:

[0028] First, initialize the maximum resource R, reduce the factor η, and calculate the maximum number of steps s. max =logη (R); then execute the following loop: randomly select hyperparameter configurations based on the current step s, configure training cycles for each hyperparameter and evaluate its performance, update the Bayesian optimization model using the evaluation results; check if the termination condition is met, if not, continue the optimization process; if the termination condition is met, output the hyperparameter λ that performs best in all evaluations.

[0029] Furthermore, the termination condition is the total loss. Once the maximum number of evaluations is reached or the performance of the fusion layer combining spatial coordinate data during the pre-training phase is no longer significantly improved.

[0030] A multi-fidelity fusion system for distributed payloads of an aircraft incorporating spatial coordinates includes a memory and one or more processors. The memory stores executable code, and when the one or more processors execute the executable code, they implement the aforementioned multi-fidelity fusion method for distributed payloads of an aircraft incorporating spatial coordinates.

[0031] Compared with the prior art, the present invention has the following beneficial effects:

[0032] This invention improves the quality of low-fidelity aerodynamic data by initially fusing high-fidelity data in the original data fusion layer. In the spatial coordinate data customization layer, it optimizes the spatial coordinate features of both the high-fidelity data and the initially fused low-fidelity data, generating custom high-fidelity and low-fidelity data sample sets. In the combined spatial coordinate data fusion layer, it introduces spatial coordinate feature learning of sample points to enrich the relationship information between data sample points, further fusing the custom high-fidelity and low-fidelity data sets from the previous layer. Within the HyperBand multi-arm fusion framework, Bayesian optimization is used to select hyperparameter configurations, achieving the fusion of aircraft distributed payload data obtained through different methods. By combining spatial coordinates and fusing data twice, the model efficiently achieves data fusion of two aerodynamic datasets. Attached Figure Description

[0033] Figure 1 This is a flowchart of a multi-fidelity fusion method for distributed payloads of an aircraft that combines spatial coordinates, according to an embodiment of the present invention.

[0034] Figure 2 This is a numerical simulation diagram of the ONERA M6 standard wing model according to an embodiment of the present invention;

[0035] Figure 3 This is a wind tunnel test image of the ONERA M6 standard wing, as described in an embodiment of the present invention.

[0036] Figure 4 The figure shows a comparison between the results of the embodiments of the present invention and the numerical calculation results at the x-axis X = 0.2.

[0037] Figure 5 The figure shows a comparison between the results of the embodiments of the present invention and the numerical calculation results at the x-axis X = 0.44.

[0038] Figure 6 The figure shows a comparison between the results of the embodiments of the present invention and the numerical calculation results at the x-axis X = 0.65.

[0039] Figure 7 The figure shows a comparison between the results of the embodiments of the present invention and the numerical calculation results at the x-axis X = 0.8.

[0040] Figure 8 The figure shows a comparison between the results of the embodiments of the present invention and the numerical calculation results at the x-axis X = 0.9;

[0041] Figure 9 The figure shows a comparison between the results of the embodiments of the present invention and the numerical calculation results at the x-axis X = 0.96. Detailed Implementation

[0042] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be noted that the embodiments described below are intended to facilitate the understanding of the present invention and do not constitute any limitation thereof.

[0043] This invention requires the fusion of two types of data: the first is experimental data, which is high-fidelity aerodynamic data; the second is CFD numerical calculation data, which is low-fidelity aerodynamic data. The fusion of these two aerodynamic datasets is achieved through three data processing layers in the model.

[0044] like Figure 1 As shown, a multi-fidelity fusion method for spacecraft distributed payloads combining spatial coordinates includes the following steps:

[0045] S01, the distributed load dataset of the aircraft under different operating conditions is obtained through numerical simulation calculation and is defined as low-fidelity aerodynamic data; the distributed load dataset of the aircraft under different operating conditions is obtained through actual experiments and is defined as high-fidelity aerodynamic data.

[0046] S02, in the raw data fusion layer after pre-training, high-fidelity data is used to perform preliminary fusion of low-fidelity aerodynamic data to improve the quality of low-fidelity aerodynamic data;

[0047] S03, in the pre-trained custom spatial coordinate data layer, optimize the spatial coordinate features of high-fidelity data and low-fidelity data after initial fusion, and generate custom high-fidelity data and custom low-fidelity data sample sets.

[0048] S04. In the pre-trained spatial coordinate data fusion layer, spatial coordinate feature learning of sample points is introduced to enrich the relationship information between data sample points. The custom high-fidelity and custom low-fidelity data sets in the previous layer are further fused to obtain the final fused aircraft target distribution payload data.

[0049] The following section provides a detailed description of the pre-training process of the original data fusion layer, the custom spatial coordinate data layer, and the combined spatial coordinate data fusion layer in the model.

[0050] First, a distributed pressure dataset 1 of the aircraft under different operating conditions was obtained through numerical calculation. The operating condition parameters (Mach number, angle of attack, and coordinates of the stress point) in dataset 1 were then used as input features x. l Distributed pressure at different locations (C) p ) as output label y com Data set 2, representing the distributed pressure of the aircraft under different operating conditions, was obtained through experiments. The operating parameters (Mach number, angle of attack, and coordinates of the stress point) in dataset 2 were used as input features x. h Distributed pressure at different locations (C) p ) as output label y exp Two working conditions from dataset 1 and dataset 2 were selected as test datasets, and 80% of the remaining data were combined as training data and 20% of the remaining data were combined as validation data.

[0051] Step S02, during the pre-training of the original data fusion layer, the model uses x l and x h Obtain the predicted value Calculated using a custom loss function, the formula is as follows:

[0052]

[0053] in, For high-fidelity aerodynamic data, the forward calculation prediction value is performed at the raw data fusion layer. For low-fidelity aerodynamic data, the forward calculation prediction values ​​are generated at the raw data fusion layer. For high-fidelity raw data, The data consists of low-fidelity raw data, q represents the number of low-fidelity raw data points, l represents the number of high-fidelity raw data points, and ρ is an adjustable parameter. The optimal solution is found through grid optimization.

[0054] The predicted values ​​of the labels were obtained by forward computation using high-fidelity data 2 and low-fidelity data 1, respectively. and Through weighted loss function Calculate the total loss, and in each training epoch, update the model parameters through backpropagation and the optimizer to minimize the total loss. This layer is used to learn high-fidelity and low-fidelity data. It initially fuses the high-fidelity and low-fidelity data to generate high-quality low-fidelity aerodynamic data. The loss value for forward calculation is obtained, where ρ is set to 0.01. After calculation by this layer, the data quality of dataset 1 is improved, resulting in high-quality low-fidelity aerodynamic dataset 3, while the original high-fidelity aerodynamic dataset 2 is retained.

[0055] In step S03, during the pre-training of the custom spatial coordinate data layer, based on the spatial coordinate features of the ONERA M6 standard wing model, samples with x-coordinates of 0.2, 0.44, 0.65, 0.8, 0.9, and 0.96 are extracted as groups, and the dataset is divided based on this sequence. This layer consists of a generator and a discriminator. The loss function of the generator is...

[0056]

[0057] The discriminator selects a custom dataset that matches the real data. The discriminator's loss function is:

[0058]

[0059] Where z is a random noise vector from a custom space, x is a real sample, G(z) is a generated data sample with random noise as input and generated as output, D(x) represents the probability that sample x is identified as a real sample, and D(G(z)) represents the probability that the generated sample is identified as a real sample. p represents the average of all possible inputs in the latent space. data (x) represents the distribution of the real data, p z (z) represents the distribution of the potential space.

[0060] In each training epoch, the model parameters are updated through backpropagation and the optimizer. The generator minimizes the loss L under the condition that the sample x-coordinates are 0.2, 0.44, 0.65, 0.8, 0.9, and 0.96. G Generate samples that conform to the custom spatial coordinate features, and then use a discriminator to calculate and minimize L. D The loss function is used to filter out target sample data. This layer processes the high-fidelity aerodynamic dataset 1 and the high-quality low-fidelity aerodynamic dataset 3 from step 1 to generate sample data with specific spatial coordinate features and corresponding sample labels, thereby obtaining custom high-fidelity aerodynamic dataset 4 and custom low-fidelity aerodynamic dataset 5.

[0061] In step S03, during the pre-training process of the spatial coordinate data fusion layer, the encoder and decoder process the sample groups with x-coordinates of 0.2, 0.44, 0.65, 0.8, 0.9, and 0.96, generating spatial position sequence relationships between samples. A self-attention mechanism is used to focus on the spatial coordinate sequences between samples, capturing the relationships between different samples and extracting detailed features of the data. The total loss is calculated using a weighted method between custom high-fidelity aerodynamic data and custom low-fidelity aerodynamic dataset. The formula is

[0062]

[0063] In the formula, and These are the forward computation predictions of high-fidelity aerodynamic data and low-fidelity aerodynamic data, respectively, calculated using a fusion layer combining spatial coordinate data. For high-fidelity raw data, λ represents the low-fidelity raw data, m represents the number of high-fidelity raw data sets, and n represents the number of low-fidelity raw data sets. λ is an adjustable parameter used to control the proportion of the loss from the two datasets in the total loss.

[0064] In each training cycle, custom high-fidelity aerodynamic data 4 and custom low-fidelity aerodynamic data 5 are generated by the encoder and decoder, and combined with the sample space coordinate order during the computation process, respectively, to obtain the predicted values ​​of the labels through forward computation. and Through weighted loss function Calculate the total loss, and in each training epoch, update the model parameters through backpropagation and the optimizer to minimize the total loss. This layer learns to combine high-fidelity and low-fidelity aerodynamic data from spatial coordinate sequences. By fusing these high-fidelity and low-fidelity aerodynamic data from spatial coordinate sequences, target aerodynamic data is generated.

[0065] During the pre-training process of the spatial coordinate data fusion layer, a surrogate model (Gaussian process) is constructed using different initial values ​​λ for the hyperparameters. The total loss is used as the basis for this process. The surrogate model is trained using the previously evaluated λ and corresponding performance values ​​for the objective function. Resources are allocated according to the HyperBand multi-armed gambling mechanism, with the resource amount for each hyperparameter configuration gradually increased. Specifically, the maximum resource R is initialized first, the factor η is reduced, and the maximum number of steps s is calculated. max =log η (R). Then, execute the following loop: based on the current step s, randomly select a hyperparameter configuration, perform training cycles for each hyperparameter configuration and evaluate its performance, and update the Bayesian optimization model using the evaluation results. Check if the termination condition (total loss) is met. Once the maximum number of evaluations is reached or the performance of the fusion layer combining spatial coordinate data during the pre-training phase no longer significantly improves, the output hyperparameter λ, which performs best in all evaluations, is determined.

[0066] In a specific embodiment of this invention, the object is the ONERAM6 standard wing. Fluent is a general-purpose fluid flow simulation software capable of rapidly evaluating aerodynamic performance, used to provide simulation calculation datasets for the M6 ​​standard wing. The ONERAM6 wing test data is described in the AGARD report; see details below. Figure 2 and Figure 3 The Mach number, angle of attack, and coordinates of the stress point are taken as input features, and the distributed pressure at different locations is taken as the output label. The Mach number ranges from 0.6998 to 0.9262, corresponding to an angle of attack range of 0.04 to 4.08. The experimental dataset contains 679 samples, and the computational simulation data contains 7906 samples. The test dataset results of this embodiment show that the mean absolute error of the multi-fidelity fusion method model for aircraft distributed loads combining spatial coordinates is better than that of the numerical calculation results. The results of this embodiment and the numerical calculation results are compared at the x-coordinates X = 0.2, 0.44, 0.65, 0.8, 0.9, and 0.96 respectively. Figures 4-9 As shown in the figure, the horizontal axis represents the spatial horizontal coordinate of the sample point, the vertical axis represents the pressure value at the sample point, the CFD curve represents the numerically calculated pressure result, the EXP curve represents the actual pressure value of the aircraft, and the DNN curve represents the pressure calculation result of the example of this invention.

[0067] The embodiments described above provide a detailed explanation of the technical solutions and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, additions, and equivalent substitutions made within the scope of the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A multi-fidelity fusion method for distributed payloads of an aircraft combining spatial coordinates, characterized in that, Includes the following steps: (1) The distributed load dataset of the aircraft under different operating conditions is obtained by numerical simulation calculation and is defined as low-fidelity aerodynamic data; the distributed load dataset of the aircraft under different operating conditions is obtained by actual test and is defined as high-fidelity aerodynamic data. (2) In the original data fusion layer after pre-training, high-fidelity aerodynamic data is used to perform preliminary fusion of low-fidelity aerodynamic data to obtain high-quality low-fidelity aerodynamic data, while retaining the original high-fidelity aerodynamic data. (3) In the pre-trained custom spatial coordinate data layer, the spatial coordinate features of the high-fidelity aerodynamic data and the initially fused low-fidelity aerodynamic data are optimized to generate custom high-fidelity data and custom low-fidelity data sample sets. The custom spatial coordinate data layer contains a generator and a discriminator. The generator is used to generate samples with specified spatial coordinate features. During pre-training, the loss function of the generator is... The discriminator selects a custom dataset that matches the real data. The discriminator's loss function is: Where z is a random noise vector from a custom space, x is a real sample, G(z) is a generated data sample with random noise as input and generated as output, D(x) represents the probability that sample x is identified as a real sample, and D(G(z)) represents the probability that the generated sample is identified as a real sample. p represents the average of all possible inputs in the latent space. data (x) represents the distribution of the real data, p z (z) represents the distribution of the potential space; (4) In the pre-trained spatial coordinate data fusion layer, the spatial coordinate feature learning of sample points is introduced to enrich the relationship information between data sample points. The custom high-fidelity and custom low-fidelity data sets in the previous layer are further fused to obtain the final fused aircraft target distribution payload data. By combining spatial coordinate data fusion layer, encoder and decoder are used to generate spatial position sequence relationships between samples. Self-attention mechanism is used to focus on spatial coordinate sequences between samples, capture the relationship between different samples, and extract detailed features of data.

2. The multi-fidelity fusion method for spacecraft distributed payloads combining spatial coordinates according to claim 1, characterized in that, In step (2), the network structure of the original data fusion layer is a fully connected layer. During the pre-training process, the losses of the high-fidelity original dataset and the low-fidelity original dataset are calculated separately, using the following formula: in, For high-fidelity aerodynamic data, the forward calculation prediction value is performed at the raw data fusion layer. For low-fidelity aerodynamic data, the forward calculation prediction values ​​are generated at the raw data fusion layer. For high-fidelity raw data, The data consists of low-fidelity raw data, l represents the number of high-fidelity raw data points, and q represents the number of low-fidelity raw data points; the total loss is calculated using a weighted average method. By minimizing the total loss Let's learn about high-fidelity and low-fidelity data information.

3. The multi-fidelity fusion method for spacecraft distributed payloads combining spatial coordinates according to claim 2, characterized in that, Calculate the total loss using a weighted method. The formula is as follows: Here, ρ is an adjustable parameter, and the optimal solution is found through grid optimization.

4. The multi-fidelity fusion method for spacecraft distributed payloads combining spatial coordinates according to claim 1, characterized in that, In step (4), during the pre-training process of the spatial coordinate data fusion layer, the total loss is calculated using a weighted method. The formula is In the formula, and These are the forward computation predictions of high-fidelity aerodynamic data and low-fidelity aerodynamic data, respectively, calculated using a fusion layer combining spatial coordinate data. For high-fidelity raw data, The data consists of low-fidelity raw data, m represents the number of high-fidelity raw data sets, n represents the number of low-fidelity raw data sets, and λ is an adjustable parameter used to control the proportion of the loss of the two datasets in the total loss. This layer combines the sample space coordinate sequence to further fuse the custom high-fidelity and low-fidelity data.

5. The multi-fidelity fusion method for spacecraft distributed payloads combining spatial coordinates according to claim 4, characterized in that, Within the HyperBand multi-armed gambling machine framework, Bayesian optimization is used to select the configuration of the adjustable parameter λ. The specific process is as follows: First, initialize the maximum resource R, reduce the factor η, and calculate the maximum number of steps s. max =log η (R); then execute the following loop: randomly select hyperparameter configurations based on the current step s, configure training cycles for each hyperparameter and evaluate its performance, update the Bayesian optimization model using the evaluation results; check if the termination condition is met, and if the termination condition is not met, continue the optimization process; If the termination condition is met, output the hyperparameter λ that performs best among all evaluations.

6. The multi-fidelity fusion method for spacecraft distributed payloads combining spatial coordinates according to claim 5, characterized in that, The termination condition is the total loss J(θ1). MSE *The performance of the fusion layer combining spatial coordinate data no longer improves significantly after reaching the maximum number of evaluations or during the pre-training phase.

7. A multi-fidelity fusion system for distributed payloads of an aircraft combining spatial coordinates, characterized in that, The system includes a memory and one or more processors, wherein the memory stores executable code, and the one or more processors execute the executable code to implement the multi-fidelity fusion method for spacecraft distributed payloads incorporating spatial coordinates as described in any one of claims 1-6.

Images