A neural network-based rapid prediction method for load separation aerodynamic forces

Abstract

This invention specifically relates to a method for rapid prediction of load separation aerodynamic forces based on neural networks, comprising: constructing a data-driven rapid prediction model of aerodynamic forces by combining physical information and neural network technology; learning the complex mapping relationship between flow field physical information and aerodynamic forces / torques through multi-layer nonlinear transformation; which can replace high-cost experiments and high-fidelity simulations; achieving rapid evaluation of aerodynamic characteristics in the early stages of development; significantly reducing R&D costs and cycle time; and meeting the urgent need for rapid iteration of solutions.

Description

Technical Field

[0001] This invention relates to the field of machine learning technology, and specifically to a method for rapid prediction of load separation aerodynamic forces based on neural networks. Background Technology

[0002] In modern aviation missions, the safe separation of the payload from the carrier aircraft is a crucial step in the successful execution of the mission. To ensure a stable and reliable separation process, the payload must be able to detach from the aircraft quickly and smoothly, and maintain attitude stability in the initial stage of separation to avoid any contact with the carrier aircraft. Therefore, during the research and development phase, it is essential to accurately acquire aerodynamic data during the separation process, analyze and predict the separation trajectory, and thus fully assess and verify the safety of the separation process. For example, in the separation of an aircraft from an external store, after release, the external store is in the airflow field that interferes with the aircraft, and its separation trajectory and attitude are affected by the airflow field. Similarly, in the separation of a transport aircraft from airdropped cargo, after exiting the cargo bay, changes in position in the wake / door flow field affect the aerodynamic forces of the cargo separation and the transport aircraft's separation attitude.

[0003] Currently, obtaining load separation aerodynamic forces mainly relies on computational fluid dynamics (CFD) numerical simulations or wind tunnel testing. While high-precision CFD methods can accurately simulate flow field details, the excessively long simulation cycles are difficult to adapt to the overall R&D process. Although wind tunnel testing and other experimental methods provide reliable data, they require significant human and material resources, resulting in high costs and making them unsuitable for the rapid verification needs of multiple iterations in the early stages of development.

[0004] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of the present invention, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] This invention provides a method for rapid prediction of load separation aerodynamic forces based on neural networks, a computer-readable storage medium, and a computer program product, which can effectively overcome the defects existing in the prior art.

[0006] Other features and advantages of the invention will become apparent from the following detailed description, or may be learned in part by practice of the invention.

[0007] According to a first aspect of the present invention, a method for rapid prediction of load separation aerodynamic forces based on neural networks is provided, the method comprising: Obtain the load model to be predicted, and divide the load model into several parts along the axial direction to obtain several local segments; The local flow direction data and local Mach number of each local segment are determined based on computational fluid dynamics algorithms when the load model is located at the location to be predicted; Based on the type of separation aerodynamic force to be predicted, the corresponding pre-trained load separation aerodynamic force prediction model is called, and the local flow direction data and local Mach number that match the called load separation aerodynamic force prediction model are input into the called load separation aerodynamic force prediction model. The load separation aerodynamic prediction model is used to predict and calculate the input local flow direction data and local Mach number, and the prediction results of the separation aerodynamics when the load model is located at the position to be predicted are obtained. The load separation aerodynamic prediction model includes: a first network branch corresponding to each local segment and used to receive the local flow direction data and local Mach number data of the corresponding local segment; and a second network branch used to receive global average information, which is the average value of the local flow direction data and the average value of the local Mach number of all local segments.

[0008] In some exemplary embodiments, determining the local flow direction data and local Mach number of each segment when the load model is located at the location to be predicted based on computational fluid dynamics algorithms includes: Based on the aircraft's flight parameters, geometric model, load model, and position parameters of the load model relative to the aircraft, computational fluid dynamics algorithms are used to solve the aircraft's disturbance flow field. Based on the load model and the position parameters of the load model relative to the carrier aircraft, the center position of each local segment is determined, and the velocity vector of the flow field and the local sound speed at each center position are obtained. The velocity vectors of the flow field at each center location are transformed into the volume coordinate system of the load model, and the local flow direction data of the corresponding local segment are calculated based on the transformed velocity vectors. Calculate the local Mach number for the corresponding local segment based on the ratio of the magnitude of the velocity vector at each center location to the local speed of sound.

[0009] In some exemplary embodiments, the local flow data includes: local angle of attack, local sideslip angle, and local flow angle; the types of separation aerodynamic forces to be predicted include: normal force, pitching moment, lateral force, yaw moment, and axial force; Based on the type of separation aerodynamic force to be predicted, the corresponding pre-trained load separation aerodynamic force prediction model is invoked, and the local flow direction data and local Mach number matching the invoked load separation aerodynamic force prediction model are input into the invoked load separation aerodynamic force prediction model, including: When the type of separation aerodynamic force to be predicted is normal force, the load separation aerodynamic force prediction model corresponding to the normal force is called, and the local angle of attack and local Mach number are input into the called load separation aerodynamic force prediction model. When the type of separation aerodynamic force to be predicted is pitching moment, the load separation aerodynamic force prediction model corresponding to the pitching moment is called, and the local angle of attack and local Mach number are input into the called load separation aerodynamic force prediction model. When the type of separation aerodynamic force to be predicted is lateral force, the load separation aerodynamic force prediction model corresponding to the pitching moment is called, and the local sideslip angle and local Mach number are input into the called load separation aerodynamic force prediction model. When the type of separation aerodynamic force to be predicted is yaw moment, the load separation aerodynamic force prediction model corresponding to the yaw moment is called, and the local sideslip angle and local Mach number are input into the called load separation aerodynamic force prediction model. When the type of separation aerodynamic force to be predicted is axial force, the load separation aerodynamic force prediction model corresponding to the axial force is called, and the local flow direction angle and local Mach number are input into the called load separation aerodynamic force prediction model.

[0010] In some exemplary embodiments, the step of using the invoked load separation aerodynamic prediction model to predict the local flow direction data and local Mach number of each input segment, and obtaining the separation aerodynamic prediction result when the load model is located at the position to be predicted, includes: Input the local flow direction data and local Mach number of each local segment into the first network branch corresponding to each local segment, and obtain the output results of each first network branch; Input the global average information into the second network branch to obtain the output of the second network branch; Based on the output results of each first network branch and the output results of the second network branch, the predicted results of load separation aerodynamics are determined.

[0011] In some exemplary embodiments, the step of inputting the local flow direction data and local Mach number of each local segment into the first network branch corresponding to each local segment, and obtaining the output results of each first network branch, includes: Input the local Mach number and local flow direction data of the local segments into the corresponding first network branch; The first network branch is used to perform feature mapping on the local Mach number and local flow direction data to obtain the corresponding intermediate output; The intermediate output is multiplied by the local flow data to obtain the output result of the corresponding first network branch.

[0012] In some exemplary embodiments, the method further includes: Under the selected flight conditions, the training load model is divided into several training sections along the axial direction, and the local flow direction data and local Mach number of each training section are determined based on computational fluid dynamics algorithms when the training load model is located at multiple positions. The separated aerodynamic data of the training payload model are calculated based on the overlapping grid algorithm when the training payload model is located in multiple positions. The separated aerodynamic data includes: normal force, pitching moment, lateral force, yaw moment and axial force. The local flow direction data and local Mach number of each training section are input into the first network branch of the initial load separation aerodynamic prediction model, and the average value of the local flow direction data and the average value of the local Mach number of each training section are input into the second network branch of the initial load separation aerodynamic prediction model, with the separation aerodynamic data at the corresponding location as the supervision label. The weight parameters of the initial load separation aerodynamic prediction model are updated using the backpropagation algorithm based on the loss function until the iteration termination condition is met, thus obtaining the trained load separation aerodynamic prediction model.

[0013] In some exemplary embodiments, the step of inputting the local flow direction data and local Mach number of each training segment into the first network branch of the initial load separation aerodynamic prediction model, inputting the average value of the local flow direction data and the average value of the local Mach number of each training segment into the second network branch of the initial load separation aerodynamic prediction model, and using the separation aerodynamic data and flight conditions at the corresponding locations as supervision labels includes: When the separated aerodynamic data is normal force or pitching moment, the local flow direction data of the first network branch and the second network branch is the local angle of attack. When the separated aerodynamic data is lateral force or yaw moment, the local flow direction data of the first network branch and the second network branch is the local sideslip angle. When the separated aerodynamic data is axial force, the local flow direction data of the first network branch and the second network branch is the local flow direction angle.

[0014] According to a second aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium including a stored executable program, wherein, when the executable program is executed, the device where the storage medium is located executes the above-described method for rapid prediction of load separation aerodynamic forces based on a neural network.

[0015] According to a third aspect of the present invention, a computer program product is provided, comprising a computer program that, when executed by a processor, implements the above-described method for rapid prediction of load separation aerodynamic forces based on a neural network.

[0016] According to a fourth aspect of the present invention, an electronic device is provided, comprising: A processor; and a memory for storing executable instructions of the processor; The processor is configured to implement the aforementioned neural network-based rapid prediction method for load separation aerodynamic forces by executing the executable instructions.

[0017] The embodiments of this invention provide a method for rapid prediction of load separation aerodynamics based on neural networks. The method acquires a load model to be predicted and divides the load model into several local segments along the axial direction. Based on computational fluid dynamics algorithms, it determines the local flow direction data and local Mach number of each local segment. The local flow direction data and local Mach number of each local segment are input into a pre-trained load separation aerodynamic prediction model to obtain the predicted load separation aerodynamics. The load separation aerodynamic prediction model includes a first network branch corresponding to each local segment and a second network branch corresponding to global statistical information. The global statistical information includes the average local flow direction data and the average local Mach number of each local segment. By extracting the local flow field characteristics of each local segment of the load and combining the local and global branches for aerodynamic modeling, the method can more accurately characterize the load separation aerodynamic characteristics in the aircraft's interference flow field, thereby achieving rapid prediction of load separation aerodynamics.

[0018] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit the invention. Attached Figure Description

[0019] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention. It is obvious that the drawings described below are merely some embodiments of the invention, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0020] Figure 1 The flowchart illustrates an exemplary embodiment of the present invention: a method for rapid prediction of load separation aerodynamic forces based on neural networks. Figure 2A This schematic diagram illustrates the force diagram of a load in a disturbed flow field in the xy plane in an exemplary embodiment of the present invention. Figure 2B This schematic diagram illustrates the force diagram of a load in a disturbed flow field in the xz plane of an exemplary embodiment of the present invention. Figure 3 This diagram schematically illustrates the network architecture of a load separation aerodynamic prediction model according to an exemplary embodiment of the present invention. Figure 4A The diagram illustrates a comparison of experimental results for the normal force coefficient in a verification experiment of an exemplary embodiment of the present invention. Figure 4B The diagram illustrates a comparison of experimental results for the pitch moment coefficient in a verification experiment of an exemplary embodiment of the present invention. Figure 4CThe diagram schematically illustrates a comparison of experimental results for the lateral force coefficient in a verification experiment of an exemplary embodiment of the present invention. Figure 4D The diagram illustrates a comparison of experimental results for the yaw moment coefficient in a verification experiment of an exemplary embodiment of the present invention. Figure 4E The diagram illustrates a comparison of experimental results for the axial force coefficient in a verification experiment of an exemplary embodiment of the present invention. Figure 5 The diagram illustrates the composition of an electronic device according to an exemplary embodiment of the present invention. Detailed Implementation

[0021] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the invention will be more comprehensive and complete, and will fully convey the concept of the exemplary embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0022] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0023] To address the shortcomings and deficiencies of existing technologies, this example embodiment provides a method for rapid prediction of load separation aerodynamic forces based on neural networks. (Reference) Figure 1 As shown, it can specifically include: Step S10: Obtain the load model to be predicted, and divide the load model into several parts along the axial direction to obtain several local segments; Step S12: Determine the local flow direction data and local Mach number of each local segment when the load model is located at the location to be predicted based on computational fluid dynamics algorithms; Step S14: Based on the type of separation aerodynamic force to be predicted, call the corresponding pre-trained load separation aerodynamic force prediction model, and input the local flow direction data and local Mach number that match the called load separation aerodynamic force prediction model into the called load separation aerodynamic force prediction model. Step S16: Use the invoked load separation aerodynamic prediction model to predict and calculate the input local flow direction data and local Mach number to obtain the separation aerodynamic prediction result when the load model is located at the position to be predicted. The load separation aerodynamic prediction model includes: a first network branch corresponding to each local segment and used to receive the local flow direction data and local Mach number data of the corresponding local segment; and a second network branch used to receive global average information, which is the average value of the local flow direction data and the average value of the local Mach number of all local segments.

[0024] The following will describe in more detail each step of a neural network-based method for rapid prediction of load separation aerodynamic forces in this exemplary embodiment, with reference to the accompanying drawings and embodiments.

[0025] For example, in step S10, the aforementioned load model refers to a three-dimensional geometric model of the load's aerodynamic shape. A load refers to an object carried by a carrier aircraft and released or separated from the carrier aircraft under predetermined operating conditions.

[0026] Specifically, taking the load carried by the aircraft as a carrier as an example, which has a slender body shape and a drastic change in the flow field along the axial direction, we select different positions of the load under the training carrier and divide the load into N segments along the axial direction to obtain N local segments.

[0027] For example, in step S12, determining the local flow direction data and local Mach number of each segment when the load model is located at the location to be predicted based on the computational fluid dynamics algorithm includes: Step S121: Based on the flight condition parameters of the carrier aircraft and the geometric model of the carrier aircraft, the computational fluid dynamics algorithm is used to solve the interference flow field of the carrier aircraft. Specifically, the flight parameters and geometric model of the carrier aircraft are obtained. The flight parameters include, but are not limited to, flight Mach number, angle of attack, sideslip angle, flight altitude, and air density; the geometric model of the carrier aircraft is used to characterize the external configuration of the carrier aircraft. After obtaining the above parameters, computational fluid dynamics (CFD) algorithms were used to calculate the aircraft interference process. The aircraft interference flow field is the non-uniform local flow field formed around the load under the influence of the flow around the aircraft. Since the load is located near the aircraft during separation, the surrounding flow is not an ideal uniform free flow, but is affected by the flow around the aircraft's nose, fuselage, wings, pylons, and other structures. Therefore, it is necessary to solve this complex flow field using computational fluid dynamics algorithms.

[0028] Alternatively, computational fluid dynamics algorithms can employ the finite volume method, finite difference method, finite element method, or other numerical calculation methods suitable for solving compressible flows.

[0029] Step S122: Based on the load model and the position parameters of the load model relative to the carrier aircraft, determine the center position of each local segment, and obtain the velocity vector of the flow field and the local sound speed at each center position; Specifically, after solving the interference flow field of the carrier aircraft, the center position of each local segment obtained by dividing the load model along the axial direction is determined. Each local segment corresponds to a center position, which can be used as a representative sampling point for the local flow field information of that local segment.

[0030] Subsequently, local flow field parameters were read at the center locations of each segment within the aircraft's interference flow field. These local flow field parameters include, but are not limited to, the velocity vector and local sound speed at the center location. The velocity vector characterizes the flow direction and magnitude at that location, while the local sound speed characterizes the speed of sound propagation in the medium at that location. Due to the significant spatial non-uniformity of the flow field around the load, the magnitude and direction of the velocity at the center locations of different segments may vary; therefore, it is necessary to extract the local flow field parameters at the center locations of each segment separately.

[0031] Step S123: Transform the velocity vector of the flow field at each center position to the volume coordinate system of the load model, and calculate the local flow direction data of the corresponding local segment based on the transformed velocity vector; Specifically, after obtaining the velocity vectors at the center positions of each local segment, the velocity vectors are transformed to the volume coordinate system of the load model. Here, the volume coordinate system is fixed to the load body and is used to characterize the directional relationship between the incoming flow direction and the load's own attitude. Since the velocity vectors output by computational fluid dynamics algorithms are usually located in the global coordinate system, inertial coordinate system, or flow field solution coordinate system, directly using these velocity vectors would be difficult to accurately reflect the directional characteristics of the local incoming flow relative to the load body; therefore, coordinate system transformation is necessary.

[0032] After coordinate transformation, local flow direction data for the corresponding local segments are calculated based on the coordinate components of the transformed velocity vector. Local flow direction data consists of angular information characterizing the local incoming flow direction, including the local angle of attack, local sideslip angle, and local flow direction angle.

[0033] Step S124: Calculate the local Mach number of the corresponding local segment based on the ratio of the magnitude of the velocity vector at each center position to the local speed of sound.

[0034] Specifically, after obtaining the velocity vector and local sound speed at the center of each local segment, the local Mach number of the corresponding local segment is calculated based on the ratio of the magnitude of the velocity vector to the local sound speed. For the nth local segment, the local velocity can be calculated first based on the velocity vector at its center, and then the local velocity can be divided by the local sound speed to obtain the local Mach number of the nth local segment.

[0035] For details, please refer to Figures 2A-2B As shown, where, Figure 2A This is a force diagram of the load in the xy plane under the disturbance flow field. Figure 2B This is a force diagram of the load in the xz plane under the influence of the disturbed flow field. The load, carried by the aircraft, is a slender body with a rapidly changing axial flow field. The pressure gradients on the upper and lower surfaces of the load are as follows: ,in, The curvature of the flow field is given. The pressure gradient on the upper and lower surfaces of the load is related to the streamline direction, i.e., it is related to the local angle of attack in the vertical plane. In the lateral plane, and with the equivalent local sideslip angle It can be assumed that, under a given Mach number, the normal force and pitching moment of the load are related to the local angle of attack. The lateral force and yaw moment are related to the local sideslip angle. The axial force and the local flow angle are related. Related to this. The local flow angle is the angle between the fluid at a specific location in the flow field and the axis of the object at that location; the local angle of attack is also relevant. with local sideslip angle It is the local flow angle Projected onto the vertical and lateral planes of the projectile.

[0036] Typically, the load separation time is short, and the roll of axisymmetric objects is limited, which has little impact on the separation safety assessment. Furthermore, the roll torque calculated numerically has a large error and requires a large amount of experimental data. Therefore, this invention does not perform aerodynamic modeling for the roll torque during the separation process.

[0037] In a non-uniform flow field, there are many factors that affect aerodynamic forces and torque coefficients. Apart from the local flow direction data, the aerodynamic forces and torque coefficients of the load are completely different when the carrier aircraft is deployed at different Mach numbers. The Mach number is a dimensionless value that describes the relationship between the velocity of an object in a fluid and the local speed of sound. It is defined as the ratio of the velocity of the object to the local speed of sound, as shown in expression (1).

[0038] (1) in, The velocity of an object relative to a fluid (usually the velocity of an aircraft). Refers to the local speed of sound.

[0039] Therefore, to broaden the scope of the model, in addition to local flow data, local Ma number variables can be introduced into the equation.

[0040] For example, the local flow direction data includes: local angle of attack, local sideslip angle, and local flow direction angle; the separation aerodynamic force type to be predicted includes: normal force, pitching moment, lateral force, yaw moment, and axial force. In step S14, based on the type of separation aerodynamic force to be predicted, the corresponding pre-trained load separation aerodynamic force prediction model is invoked, and the local flow direction data and local Mach number matching the invoked load separation aerodynamic force prediction model are input into the invoked load separation aerodynamic force prediction model, including: Step S141: When the type of separation aerodynamic force to be predicted is normal force or pitching moment, call the corresponding load separation aerodynamic force prediction model and input the local angle of attack and local Mach number into the called load separation aerodynamic force prediction model. Step S142: When the type of separation aerodynamic force to be predicted is lateral force or yaw moment, call the corresponding load separation aerodynamic force prediction model and input the local sideslip angle and local Mach number into the called load separation aerodynamic force prediction model. Step S143: When the type of separation aerodynamic force to be predicted is axial force, call the load separation aerodynamic force prediction model corresponding to the axial force, and input the local flow direction angle and local Mach number into the called load separation aerodynamic force prediction model.

[0041] For example, in step S16, the method of using the invoked load separation aerodynamic prediction model to predict the local flow direction data and local Mach number of each input segment, and obtaining the separation aerodynamic prediction result when the load model is located at the position to be predicted, includes: Step S161: Input the local flow direction data and local Mach number of each local segment into the first network branch corresponding to each local segment, and obtain the output results of each first network branch; Step S162: Input the global average information into the second network branch to obtain the output result of the second network branch; Step S163: Based on the output results of each first network branch and the output results of the second network branch, determine the predicted results of the load separation aerodynamics.

[0042] For example, in step S161, the step of inputting the local flow direction data and local Mach number of each local segment into the first network branch corresponding to each local segment, and obtaining the output results of each first network branch, includes: Step S1611: Input the local Mach number and local flow direction data of the local segments into the corresponding first network branch respectively; Step S1612: Use the first network branch to perform feature mapping on the local Mach number and local flow direction data to obtain the corresponding intermediate output; Step S1613: Multiply the intermediate output with the local flow data to obtain the output result of the corresponding first network branch.

[0043] For details, please refer to Figure 3 As shown, Figure 3 This is a network architecture diagram of the load separation aerodynamic prediction model. The model employs a parallelized, modular deep neural network architecture, with the overall network consisting of N+1 independent sub-neural network modules (labeled as...). This is the structure of each subgrid. The input vector consists of physical state parameters, and each module is a fully connected neural network with sub-grids. Output Directly enter the summation node, submesh Output Before entering the summation node, it will first be compared with the input variables. Multiplying these together, the final output is the aerodynamic force and aerodynamic torque in each direction. The mathematical expression for this network structure is shown in expression (2).

[0044] (2) Where N is the number of local segments into which the load is manually divided. For each local segment, the corresponding local flow direction data, This refers to the local Mach number corresponding to a specific segment; hour, Let N be the average of local flow direction data. It is the average of N local Mach numbers.

[0045] Furthermore, For the second network branch, It is the first network branch.

[0046] When the type of separation aerodynamic force to be predicted is normal force or pitching moment When k=0, For the local angle of attack That is, the input data of the first network branch is the local Mach number of each local segment corresponding to the location to be predicted. and local attack angle The input data for the second network branch is the local Mach number of each segment corresponding to the location to be predicted. Average and local angle of attack The average value.

[0047] When the type of separation aerodynamic force to be predicted is lateral force or yaw moment When k=1, For the local sideslip angle That is, the input data of the first network branch is the local Mach number of each local segment corresponding to the location to be predicted. and local sideslip angle The input data for the second network branch is the local Mach number of each segment corresponding to the location to be predicted. The average value and local sideslip angle The average value.

[0048] When the type of separation aerodynamic force to be predicted is axial force When k=2, For local flow angle That is, the input data of the first network branch is the local Mach number of each local segment corresponding to the location to be predicted. and local flow angle The input data for the second network branch is the local Mach number of each segment corresponding to the location to be predicted. The average value and local flow angle The average value.

[0049] For example, the method further includes: Step S21: Under the selected flight conditions, the training load model is divided into several training sections along the axial direction, and the local flow direction data and local Mach number of each training section are determined based on the computational fluid dynamics algorithm when the training load model is located at multiple positions. Step S22: Calculate the separated aerodynamic data of the training load model when it is located at multiple positions based on the overlapping grid algorithm; wherein, the separated aerodynamic data includes: normal force, pitching moment, lateral force, yaw moment and axial force; Step S23: Input the local flow direction data and local Mach number of each training section into the first network branch of the initial load separation aerodynamic prediction model, input the average value of the local flow direction data and the average value of the local Mach number of each training section into the second network branch of the initial load separation aerodynamic prediction model, and use the separation aerodynamic data at the corresponding position as supervision labels. Step S24: Update the weight parameters of the initial load separation aerodynamic prediction model using the backpropagation algorithm based on the loss function until the iteration termination condition is met, and obtain the trained load separation aerodynamic prediction model.

[0050] For example, in step S23, the step of inputting the local flow direction data and local Mach number of each training segment into the first network branch of the initial load separation aerodynamic prediction model, inputting the average value of the local flow direction data and the average value of the local Mach number of each training segment into the second network branch of the initial load separation aerodynamic prediction model, and using the separation aerodynamic data at the corresponding position as supervision labels includes: Step S231: When the separated aerodynamic data is normal force or pitching moment, input the local flow direction data of the first network branch and the second network branch as the local angle of attack. Step S232: When the separated aerodynamic data is lateral force or yaw moment, input the local flow direction data of the first network branch and the second network branch as the local sideslip angle; Step S233: When the separated aerodynamic data is axial force, input the local flow direction data of the first network branch and the second network branch as the local flow direction angle.

[0051] Specifically, the model training process for the initial load separation aerodynamic prediction model is as follows: Step 1: Determine the training payload model to be predicted and train the carrier model.

[0052] Step Two: Divide the training load into segments along the axial direction. In this example, we take dividing the load into 5 equal segments. Select different positions of the load under the training aircraft. The selected positions must be within the aircraft's interference flow field. These positions are determined under the following flight conditions: and Each of the following sets of 60 locations is selected, for a total of 120 locations.

[0053] Step 3: Calculate the separated aerodynamic forces in each direction under load at the selected location using overlapping mesh technology. (i=0, 1, 2, 3, 4), that is ( ).in, Normal force coefficient; This is the pitch moment coefficient; This is the lateral force coefficient; This is the yaw moment coefficient; This is the axial force coefficient.

[0054] Step 4: Based on the locations selected in Step 2, calculate the flow field of the training aircraft at each location individually, and obtain local flow direction data and local... Collect data and construct a training set.

[0055] Step 5: Based on the network structure, initialize the weight coefficients and other network parameters of each sub-module neural network.

[0056] Step 6: Read in the training set data and calculate the loss function.

[0057] Step 7: Train the neural network. Update the neural network weights using the backpropagation algorithm based on the loss function. Stop training when the maximum number of iterations is reached or the error meets the preset requirements, and output the optimal neural network. Otherwise, continue to Step 6.

[0058] This embodiment is respectively in and The following CFD technology was used to calculate the flow field of the training aircraft, thereby obtaining local flow direction data for each part of the load at 120 selected locations. With local Mach number When n=0, Let N be the average of local flow direction data. It is the average of N local Mach numbers.

[0059] Based on the aerodynamic model described above, the normal force and pitching moment acting on the load are related to the local angle of attack. The normal force coefficient (…) is constructed. When predicting the model, i=0, k=0, For the local angle of attack The input data for the first network branch of the initial load separation aerodynamic prediction model is the local Mach number corresponding to each training segment at each location. and local attack angle There are a total of 120 positions. The data for each position is ( ,…,( The second network branch input data is the local Mach number corresponding to each training section at each location. Average and local angle of attack The average value. Normal force in separated aerodynamic data. The model is trained using supervisory labels. The initial normal force coefficients are updated using backpropagation based on the loss function. The prediction model's weight parameters are calculated until the iteration termination condition is met, resulting in the normal force coefficients after training is complete. Predictive models.

[0060] Constructing the pitch moment coefficient ( When using the prediction model, i=1, k=0, For the local angle of attack The input data for the first network branch of the initial load separation aerodynamic prediction model is the local Mach number corresponding to each training segment at each location. and local attack angle There are a total of 120 positions. The data for each position is ( ,…,( The second network branch input data is the local Mach number corresponding to each training section at each location. Average and local angle of attack The average value. Pitch moment coefficient in separated aerodynamic data. The model is trained using supervisory labels. The initial pitch moment coefficients are updated using backpropagation based on the loss function. The prediction model's weight parameters are calculated until the iteration termination condition is met, resulting in the normal force coefficients after training is complete. Predictive models.

[0061] According to the aerodynamic model described in the technical solution, the lateral force and yaw moment acting on the load are related to the local sideslip angle. A lateral force coefficient is constructed (…). When using the prediction model, i=2, k=1, For the local sideslip angle The input data for the first network branch of the initial load separation aerodynamic prediction model is the local Mach number corresponding to each training segment at each location. and local sideslip angle There are a total of 120 positions. The data for each position is ( ,…,( The second network branch input data is the local Mach number corresponding to each training section at each location. The average value and local sideslip angle The average value. Lateral force coefficient in the separated aerodynamic data. The model is trained using supervisory labels. The initial lateral force coefficients are updated using backpropagation based on the loss function. The weight parameters of the prediction model are calculated until the iteration termination condition is met, resulting in the lateral force coefficients after training is complete. Predictive models.

[0062] Constructing the yaw moment coefficient ( When using the prediction model, i=3, k=1. For the local sideslip angle The input data for the first network branch of the initial load separation aerodynamic prediction model is the local Mach number corresponding to each training segment at each location. and local sideslip angle There are a total of 120 positions. The data for each position is ( ,…,( The second network branch input data is the local Mach number corresponding to each training section at each location. The average value and local sideslip angle The average value of the yaw moment coefficient in the separated aerodynamic data. The model is trained using supervisory labels. The initial yaw moment coefficients are updated using backpropagation based on the loss function. The weight parameters of the prediction model are calculated until the iteration termination condition is met, resulting in the yaw moment coefficients after training. Predictive models.

[0063] According to the aerodynamic model described in the technical solution, the axial force on the load is related to the local flow angle. An axial force coefficient is constructed ( When using the prediction model, i=4, k=2. For local flow angle The input data for the first network branch of the initial load separation aerodynamic prediction model is the local Mach number corresponding to each training segment at each location. and local flow angle There are a total of 120 positions. The data for each position is ( ,…,( The second network branch input data is the local Mach number corresponding to each training section at each location. The average value and local sideslip angle The average value. Axial force coefficient in separated aerodynamic data. The model is trained using supervisory labels. The initial axial force coefficients are updated using backpropagation based on the loss function. The weight parameters of the prediction model are adjusted until the iteration termination condition is met, resulting in the axial force coefficients after training. Predictive models.

[0064] The loss function is shown in expression (3): (3) For the sample size, For the true value, These are predicted values.

[0065] The specific parameters during model training are shown in Table 1.

[0066] Table 1 Neural Network Parameter Table

[0067] Step 8: Select different locations of the load under the predicted carrier according to actual needs, use CFD to calculate the flow field of the carrier to be predicted, and obtain the local flow direction data and local Ma number information of each local segment of the load at the selected location.

[0068] Step 9: Input the local flow direction data and local Ma number information of the load into the trained neural network to obtain the corresponding load separation aerodynamic force.

[0069] This example implementation provides a verification experiment of a fast aerodynamic prediction method based on a neural network for load separation. The prediction results are compared with CFD results, and the comparison results are referenced. Figures 4A-4E As shown. Among them, Figure 4A A comparison chart of experimental results for the normal force coefficient; Figure 4B A comparison chart of experimental results for pitch moment coefficient; Figure 4C A comparison chart of experimental results for the lateral force coefficient; Figure 4D A comparison chart of experimental results for the yaw moment coefficient; Figure 4E This is a comparison chart of experimental results for the axial force coefficient; the horizontal axis in the chart represents 11 locations selected uniformly in the vertical direction after the load leaves the bracket, and the vertical axis represents the corresponding predicted results.

[0070] The aerodynamic data input for training is and Aerodynamic forces and aerodynamic moments under the operating conditions were selected to verify the model's generalization ability. After the load leaves the hanger, 11 positions are evenly selected along the vertical direction. The specific prediction errors are shown in Table 2.

[0071] Table 2 Prediction Error Table

[0072] As can be seen from the experimental results in Figure 1 and Table 2, compared with the CFD calculation results, the errors of force and torque in each direction are all less than 30%, which shows that the constructed neural network model has high accuracy in dealing with load separation aerodynamic prediction problems. It also further demonstrates the feasibility of applying this method to practical engineering problems and solving the problem of slow load separation trajectory prediction in practical engineering projects.

[0073] The beneficial effects of this invention are as follows: (1) Traditional fully connected neural networks are usually a "black box", making it difficult to analyze the specific influence mechanism of input variables on output results. This invention simulates the physical laws of aerodynamics to a certain extent by constructing a topology structure that integrates parallel subnetworks and multiplication, thereby improving the physical interpretability of the model. (2) Pure data-driven deep neural networks are prone to overfitting in sparse regions of training data. This invention introduces multiplication nodes as “soft constraints” at the architecture level. Compared with unconstrained general neural networks, this structure introduces prior knowledge to reduce the hypothesis space and improve the generalization performance of the model. (3) By effectively mining the information in the computational data, this invention can quickly assess the safety of payload separation in the early stage of aircraft development. The overall algorithm has high computational efficiency and is suitable for engineering scenarios that require rapid analysis.

[0074] It should be noted that the above figures are merely illustrative of the processes included in the method according to exemplary embodiments of the present invention, and are not intended to be limiting. It is readily understood that the processes shown in the above figures do not indicate or limit the temporal order of these processes. Furthermore, it is readily understood that these processes may, for example, be executed synchronously or asynchronously in multiple modules.

[0075] It should be noted that although several modules or units of the device for performing actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to embodiments of the present invention, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0076] Figure 5 A schematic diagram of an electronic device suitable for implementing embodiments of the present invention is shown.

[0077] It should be noted that, Figure 5 The electronic device 1000 shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0078] like Figure 5 As shown, the electronic device 1000 includes a Central Processing Unit (CPU) 1001, which can perform various appropriate actions and processes based on programs stored in Read-Only Memory (ROM) 1002 or programs loaded from storage section 1008 into Random Access Memory (RAM) 1003. The RAM 1003 also stores various programs and data required for system operation. The CPU 1001, ROM 1002, and RAM 1003 are interconnected via a bus 1004. An Input / Output (I / O) interface 1005 is also connected to the bus 1004. Furthermore, the electronic device 1000 also includes an FPGA device and a System-on-a-Chip (SoC) device.

[0079] The following components are connected to I / O interface 1005: an input section 1006 including a keyboard, mouse, etc.; an output section 1007 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 1008 including a hard disk, etc.; and a communication section 1009 including a network interface card such as a LAN (Local Area Network) card, modem, etc. The communication section 1009 performs communication processing via a network such as the Internet. A drive 1010 is also connected to I / O interface 1005 as needed. Removable media 1011, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., are installed on drive 1010 as needed so that computer programs read from them can be installed into storage section 1008 as needed.

[0080] In particular, according to embodiments of the present invention, the processes described below with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a storage medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 1009, and / or installed from removable medium 1011. When the computer program is executed by central processing unit (CPU) 1001, it performs various functions defined in the system of this application.

[0081] Specifically, the aforementioned electronic devices can be airborne intelligent electronic devices.

[0082] It should be noted that the storage medium shown in the embodiments of the present invention can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable compact disc read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In the present invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In the present invention, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, wherein computer-readable program code is carried. Such transmitted data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. The computer-readable signal medium can also be any storage medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the storage medium can be transmitted using any suitable medium, including but not limited to wireless, wired, etc., or any suitable combination thereof.

[0083] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0084] The units described in the embodiments of the present invention can be implemented in software or hardware, and the described units can also be located in a processor. The names of these units do not necessarily limit the specific unit itself.

[0085] It should be noted that, as another aspect, this application also provides a storage medium, which may be included in an electronic device or may exist independently without being assembled into the electronic device. The aforementioned storage medium carries one or more programs, which, when executed by an electronic device, cause the electronic device to perform the methods described in the following embodiments. For example, the electronic device may perform... Figure 1 The steps of the method shown.

[0086] In one embodiment, this application provides a computer program product including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0087] Furthermore, the above figures are merely illustrative of the processes included in the method according to exemplary embodiments of the present invention, and are not intended to be limiting. It is readily understood that the processes shown in the above figures do not indicate or limit the temporal order of these processes. Additionally, it is readily understood that these processes may be executed synchronously or asynchronously, for example, in multiple modules.

[0088] Other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention herein. This application is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein. The specification and embodiments are to be considered exemplary only, and the true scope and spirit of the invention are indicated by the claims.

[0089] It should be understood that the present invention is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A neural network-based fast prediction method for load separation aerodynamic forces, characterized in that, The method includes: Obtain the load model to be predicted, and divide the load model into several parts along the axial direction to obtain several local segments; The local flow direction data and local Mach number of each local segment are determined based on computational fluid dynamics algorithms when the load model is located at the location to be predicted; Based on the type of separation aerodynamic force to be predicted, the corresponding pre-trained load separation aerodynamic force prediction model is called, and the local flow direction data and local Mach number that match the called load separation aerodynamic force prediction model are input into the called load separation aerodynamic force prediction model. The load separation aerodynamic prediction model is used to predict and calculate the input local flow direction data and local Mach number, and the prediction results of the separation aerodynamics when the load model is located at the position to be predicted are obtained. The load separation aerodynamic prediction model includes: a first network branch corresponding to each local segment and used to receive the local flow direction data and local Mach number data of the corresponding local segment; and a second network branch used to receive global average information, which is the average value of the local flow direction data and the average value of the local Mach number of all local segments.

2. The method of claim 1, wherein, The determination of local flow direction data and local Mach number for each segment when the load model is located at the location to be predicted, based on computational fluid dynamics algorithms, includes: Based on the aircraft's flight parameters, geometric model, load model, and position parameters of the load model relative to the aircraft, computational fluid dynamics algorithms are used to solve the aircraft's disturbance flow field. Based on the load model and the position parameters of the load model relative to the carrier aircraft, the center position of each local segment is determined, and the velocity vector of the flow field and the local sound speed at each center position are obtained. The velocity vectors of the flow field at each center location are transformed into the volume coordinate system of the load model, and the local flow direction data of the corresponding local segment are calculated based on the transformed velocity vectors. Calculate the local Mach number for the corresponding local segment based on the ratio of the magnitude of the velocity vector at each center location to the local speed of sound.

3. The method of claim 1, wherein, The local flow data includes: local angle of attack, local sideslip angle, and local flow direction angle; the types of separation aerodynamic forces to be predicted include: normal force, pitching moment, lateral force, yaw moment, and axial force; Based on the type of separation aerodynamic force to be predicted, the corresponding pre-trained load separation aerodynamic force prediction model is invoked, and the local flow direction data and local Mach number matching the invoked load separation aerodynamic force prediction model are input into the invoked load separation aerodynamic force prediction model, including: When the type of separation aerodynamic force to be predicted is normal force or pitching moment, the corresponding load separation aerodynamic force prediction model is called, and the local angle of attack and local Mach number are input into the called load separation aerodynamic force prediction model. When the type of separation aerodynamic force to be predicted is lateral force or yaw moment, the corresponding load separation aerodynamic force prediction model is called, and the local sideslip angle and local Mach number are input into the called load separation aerodynamic force prediction model. When the type of separation aerodynamic force to be predicted is axial force, the load separation aerodynamic force prediction model corresponding to the axial force is called, and the local flow direction angle and local Mach number are input into the called load separation aerodynamic force prediction model.

4. The method of claim 1, wherein, The method utilizes the invoked load separation aerodynamic prediction model to predict and calculate the local flow direction data and local Mach number of each input segment, obtaining the separation aerodynamic prediction results when the load model is located at the position to be predicted, including: Input the local flow direction data and local Mach number of each local segment into the first network branch corresponding to each local segment, and obtain the output results of each first network branch; The global average information is input into the second network branch to obtain the output result of the second network branch; Based on the output results of each first network branch and the output results of the second network branch, the predicted results of load separation aerodynamics are determined.

5. The method of claim 4, wherein, The step of inputting the local flow direction data and local Mach number of each local segment into the first network branch corresponding to each local segment, and obtaining the output results of each first network branch, includes: Input the local Mach number and local flow direction data of the local segments into the corresponding first network branch; The first network branch is used to perform feature mapping on the local Mach number and local flow direction data to obtain the corresponding intermediate output; The intermediate output is multiplied by the local flow data to obtain the output result of the corresponding first network branch.

6. The method of any one of claims 1-5, wherein, The method further includes: Under the selected flight conditions, the training load model is divided into several training sections along the axial direction, and the local flow direction data and local Mach number of each training section are determined based on computational fluid dynamics algorithms when the training load model is located at multiple positions. The separated aerodynamic data of the training payload model are calculated based on the overlapping grid algorithm when the training payload model is located in multiple positions. The separated aerodynamic data includes: normal force, pitching moment, lateral force, yaw moment and axial force. The local flow direction data and local Mach number of each training section are input into the first network branch of the initial load separation aerodynamic prediction model, and the average value of the local flow direction data and the average value of the local Mach number of each training section are input into the second network branch of the initial load separation aerodynamic prediction model, with the separation aerodynamic data at the corresponding location as the supervision label. The weight parameters of the initial load separation aerodynamic prediction model are updated using the backpropagation algorithm based on the loss function until the iteration termination condition is met, thus obtaining the trained load separation aerodynamic prediction model.

7. The method of claim 6, wherein, The process involves inputting the local flow direction data and local Mach number of each training segment into the first network branch of the initial load separation aerodynamic prediction model, inputting the average value of the local flow direction data and the average value of the local Mach number of each training segment into the second network branch of the initial load separation aerodynamic prediction model, and using the separation aerodynamic data at the corresponding location as supervision labels, including: When the separated aerodynamic data is normal force or pitching moment, the local flow direction data of the first network branch and the second network branch is the local angle of attack. When the separated aerodynamic data is lateral force or yaw moment, the local flow direction data of the first network branch and the second network branch is the local sideslip angle. When the separated aerodynamic data is axial force, the local flow direction data of the first network branch and the second network branch is the local flow direction angle.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored executable program, wherein, when the executable program is executed, it controls the device on which the storage medium is located to perform the method according to any one of claims 1 to 7.

9. A computer program product, characterised in that, Includes a computer program that, when executed by a processor, implements the method according to any one of claims 1 to 7.

Images