A fast prediction method of airfoil flow field based on super-resolution reconstruction

Abstract

A method for rapid prediction of airfoil flow field based on super-resolution reconstruction is provided, which belongs to the field of fluid mechanics and artificial intelligence technology, including generating an airfoil sample data set; processing the original flow field data to obtain high- and low-resolution flow field data; using a deep neural network to establish a prediction model from working conditions and shape parameters to low-resolution flow field data; using an ultra-deep super-resolution neural network to establish a prediction model from low-resolution flow field data to high-resolution flow field data; connecting two models in series to build a flow field prediction model; and using the built model for rapid prediction of airfoil flow field. The present invention is based on super-resolution reconstruction technology and combined with a deep neural network, which can well learn the flow field characteristics, realize rapid and accurate prediction of the flow field, and reduce the time consumption of flow field calculation. In the process of flow field data processing, the present invention first maps the flow field data from the aerodynamic calculation domain to the learning domain, performs interpolation processing on the data with large gradients, improves the modeling accuracy, and reduces the flow field prediction error.

Description

Technical Field

[0001] The invention belongs to the technical field of fluid mechanics and artificial intelligence, and specifically relates to a method for quickly predicting an airfoil flow field based on super-resolution reconstruction. Background Art

[0002] In the field of airfoil aerodynamic analysis and optimization design, a large number of time-consuming and high-precision computational fluid dynamics (CFD) numerical simulations are required. When the airfoil shape and calculation conditions change, the CFD simulation needs to be re-executed, resulting in low efficiency in airfoil aerodynamic analysis or optimization design. In fact, the airfoil flow fields obtained under different geometric shapes and calculation conditions have similar physical properties and flow characteristics. Machine learning methods such as deep learning and neural networks can learn these characteristics and achieve rapid prediction of the flow field. Therefore, it is necessary and widely needed to develop a rapid prediction method for airfoil flow fields with variable shapes and conditions. Therefore, it is necessary to propose improvements. Summary of the invention

[0003] Technical problem solved by the present invention: In order to reduce the time-consuming problem of a large amount of flow field calculation caused by changes in the shape and calculation conditions in the analysis or optimization design of the aerodynamic characteristics of the airfoil, the present invention provides a method for rapid prediction of the airfoil flow field based on super-resolution reconstruction. The method can extract the characteristics of the airfoil flow field data with variable shapes and conditions, realize rapid and accurate prediction of the steady flow field of the airfoil under different shapes and calculation conditions, and improve the efficiency of flow field calculation.

[0004] The technical solution adopted by the present invention is a method for rapid prediction of airfoil flow field based on super-resolution reconstruction, comprising the following steps:

[0005] Step 1): Generate a sample data set of airfoil with variable working conditions and shapes; the specific steps are as follows:

[0006] Step 1-1): Select a certain type of airfoil, determine the airfoil shape parameter design space and calculation condition design space, and use Latin hypercube sampling to obtain a series of airfoil samples;

[0007] Step 1-2): Generate an aerodynamic calculation grid for the airfoil;

[0008] Step 1-3): Perform CFD simulation calculation on the airfoil sample obtained in step 1-1) to obtain original flow field data of the airfoil sample;

[0009] Step 2): The original flow field data in steps 1-3) are subjected to domain mapping, data interpolation, and downsampling to obtain high- and low-resolution flow field data; the specific steps are as follows:

[0010] Step 2-1): The original flow field data in step 1-3) has the same grid size as that in step 1-2), which is 210×150. The original flow field data is mapped from the aerodynamic calculation domain to the learning domain and represented as a 210×150 orthogonally uniformly distributed pixel image;

[0011] Step 2-2): interpolate the area with large gradient change of the flow field data after domain mapping in step 2-1) to obtain a 315×150 pixel image as high-resolution flow field data;

[0012] Step 2-3): The downsampling scaling factor is 5, and the high-resolution flow field data in step 2-2) is downsampled to obtain low-resolution flow field data, whose size is 63×30;

[0013] Step 3): Use a deep neural network DNN to establish a prediction model from the working conditions and shape parameters in step 1-1) to the low-resolution flow field data in step 2);

[0014] Step 4): Use an ultra-deep super-resolution neural network VDSR to establish a prediction model from the low-resolution flow field data in step 2) to the high-resolution flow field data in step 2);

[0015] Step 5): connect the DNN in step 3) and the VDSR in step 4) into a DNN-VDSR model to complete the modeling of the shape and working condition parameters to the high-resolution flow field data;

[0016] Step 6): Use the DNN-VDSR model in step 5) to quickly predict the steady flow field of the airfoil with changed shape and working conditions. The specific steps are as follows:

[0017] Step 6-1): Given the shape parameters and geometric parameters of the new airfoil, the RBF mesh deformation technology of the aerodynamic calculation mesh in step 1-2) is used to obtain the mesh of the new airfoil, with a size of 210×150;

[0018] Step 6-2): Use the shape and geometric parameters in step 6-1) as the input of the DNN-VDSR trained in step 5) to predict high-resolution flow field data with a size of 315×150;

[0019] Step 6-3): Perform reverse interpolation on the high-resolution flow field data predicted in step 6-2) to obtain a size of 210×150, and map the data to the grid in step 6-1) to obtain the predicted flow field data.

[0020] In the above step 1-1), the certain type of airfoil is an elliptical airfoil, and the total number of samples is 500, of which 90% are used for training and 10% are used for testing; the shape parameters are the camber and thickness of the elliptical airfoil, and the calculation conditions of the thickness are the angle of attack and the Mach number, the camber value range is 0 to 0.03c, the thickness value range is 0.13c to 0.19c, the angle of attack value range is 1 to 4°, and the Mach number value range is 0.15 to 0.45.

[0021] In the above steps 1-2), the aerodynamic calculation grid is an O-type grid, the grid size is 210×150 in the circumferential direction and the first layer height of the boundary layer is 8×10 -6 c. First, an O-shaped grid is divided for any airfoil as the base grid. The grids of the remaining airfoils are obtained based on the base grid using the RBF grid deformation technology. All grid sizes are 210×150.

[0022] In the above steps 1-3), the CFD calculation model adopts the SST k-omega turbulence model.

[0023] In the above step 3), a total of 4 parameters including the shape parameters and geometric parameters in step 1-1) are used as the input of the deep neural network DNN, and the low-resolution flow field data in step 2-3) is divided into 30 regions, each with a dimension of 60, as the output of the DNN, and a total of 30 groups of DNN models are established.

[0024] In the above step 4), the low-resolution flow field data in step 2-3) is used as the input of the ultra-deep super-resolution neural network VDSR, and the high-resolution flow field data in step 2-2) is used as the output of the ultra-deep super-resolution neural network VDSR, and a VDSR model is established.

[0025] In the above step 5), the model established using DNN in step 3) and the model established using VDSR in step 4) are connected in series to form a DNN-VDSR airfoil flow field prediction model and train it.

[0026] The advantages of the present invention compared with the prior art are:

[0027] 1. This solution is based on super-resolution reconstruction technology and combined with deep neural networks, which can well learn the flow field characteristics, realize fast and accurate prediction of the flow field, and reduce the time consumption of flow field calculation;

[0028] 2. The rapid flow field prediction model proposed in this scheme can be applied to the prediction of steady flow fields of airfoils whose shapes and working conditions change simultaneously, and has a high model generalization ability;

[0029] 3. In the process of flow field data processing, this scheme first maps the flow field data from the aerodynamic calculation domain to the learning domain, and performs interpolation processing on the data with large gradients, which can effectively improve the modeling accuracy and reduce the prediction error of the flow field. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 A flow chart of the modeling process of the flow field prediction model based on high-resolution reconstruction in the present invention;

[0031] Figure 2 Schematic diagram of the geometric shape of the airfoil in the present invention;

[0032] Figure 3 A schematic diagram of sampling of airfoil calculation conditions in the present invention;

[0033] Figure 4 Schematic diagram of the airfoil aerodynamic grid in the present invention, (a) is the global grid, (b) is the local grid;

[0034] Figure 5 A flow chart of domain mapping, data interpolation and down sampling operations for the original flow field data of the airfoil in the present invention;

[0035] Figure 6 The airfoil flow field prediction results of the present invention are as follows: (a) is the actual pressure field cloud map, (b) is the predicted pressure field cloud map, (c) is the prediction error cloud map of the pressure field, and (d) is a comparison diagram of the actual and predicted results of the isobars. DETAILED DESCRIPTION

[0036] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0037] See also Figure 1-6 , details of the embodiments of the present invention.

[0038] A method for rapid prediction of airfoil flow field based on super-resolution reconstruction comprises the following steps:

[0039] Step 1): Generate a sample data set of airfoil with variable working conditions and shapes; the specific steps are as follows:

[0040] Step 1-1): In this embodiment, an elliptical airfoil is selected, and the shape parameters are the camber and thickness of the elliptical airfoil, and the calculation conditions of the thickness are the angle of attack and the Mach number. The design space of the shape parameters of the airfoil and the design space of the calculation conditions are: the camber value range is 0 to 0.03c, the thickness value range is 0.13c to 0.19c, c is the chord length of the airfoil, the angle of attack value range is 1 to 4°, and the Mach number value range is 0.15 to 0.45. The geometric shape diagram of the airfoil is as follows Figure 2 The schematic diagram of airfoil calculation condition sampling is shown in Figure 3 A series of airfoil samples are obtained using Latin hypercube sampling, with a total of 500 samples, 90% of which are used for training and 10% for testing.

[0041] Step 1-2): Generate the aerodynamic calculation grid of the airfoil. Specifically, the specific operation of generating the aerodynamic grid of the airfoil is as follows: the aerodynamic calculation grid is an O-type grid, the grid size is 210×150 (circumferential×radial), the first layer height of the boundary layer is 8×10 -6 c. First, an O-shaped grid is divided into any one of the airfoils as the reference grid. The grids of the other airfoils are obtained based on the reference grid using the RBF grid deformation technology. The size of all grids is 210×150. The global schematic diagram of the aerodynamic grid is shown in Figure 4 (a), partial schematic diagram see Figure 4 (b).

[0042] Step 1-3): Perform CFD simulation calculation on the airfoil sample obtained in step 1-1), and the CFD calculation model adopts the SST k-omega turbulence model to obtain the original flow field data of the airfoil sample.

[0043] Step 2): Perform domain mapping, data interpolation, and downsampling on the original flow field data in steps 1-3) to obtain high- and low-resolution flow field data, such as Figure 5 The specific steps are as follows:

[0044] Step 2-1): The original flow field data in step 1-3) has the same grid size as that in step 1-2), which is 210×150. The data is mapped from the aerodynamic calculation domain to the learning domain and represented as a 210×150 orthogonally uniformly distributed pixel image.

[0045] Step 2-2): interpolate the area with large gradient change of the flow field data after domain mapping in step 2-1) to obtain a 315×150 pixel image as high-resolution flow field data;

[0046] Step 2-3): The downsampling scaling factor is 5, and the high-resolution flow field data in step 2-2) is downsampled to obtain low-resolution flow field data, whose size is 63×30.

[0047] Step 3): Use a deep neural network DNN to establish a prediction model from the working conditions and shape parameters in step 1-1) to the low-resolution flow field data in step 2);

[0048] Step 4): Use the ultra-deep super-resolution neural network VDSR to establish a prediction model from the low-resolution flow field data in step 2) to the high-resolution flow field data in step 2); specifically, the low-resolution flow field data in step 2-3) is used as the input of the ultra-deep super-resolution neural network VDSR, and the high-resolution flow field data in step 2-2) is used as the output of the ultra-deep super-resolution neural network VDSR, and a VDSR model is established.

[0049] Step 5): The model established using DNN in step 3) and the model established using VDSR in step 4) are connected in series to form a DNN-VDSR airfoil flow field prediction model and train it.

[0050] Step 6): Use the DNN-VDSR model in step 5) to quickly predict the steady flow field of the airfoil with changed shape and working conditions. The specific steps are as follows:

[0051] Step 6-1): Given the shape parameters and geometric parameters of the new airfoil, the RBF mesh deformation technology in step 1-2) is used to obtain the mesh of the new airfoil with a size of 210×150.

[0052] Step 6-2): The shape and geometric parameters in step 6-1) are used as the input of the DNN-VDSR trained in step 5) to predict high-resolution flow field data with a size of 315×150.

[0053] Step 6-3): Perform reverse interpolation on the high-resolution flow field data predicted in step 6-2) to obtain a size of 210×150, and map the data to the grid in step 6-1) to obtain the predicted flow field data. Figure 6 The prediction results of the airfoil flow field under a given new shape and new working conditions are given, where (a) is the actual pressure field cloud map, (b) is the predicted pressure field cloud map, (c) is the prediction error cloud map of the pressure field, and (d) is the comparison between the actual and predicted results of the isobars.

[0054] The technical solution realizes the steady flow field prediction of variable geometry airfoils under variable working conditions based on high-resolution reconstruction technology. Compared with the prior art, the present invention is based on super-resolution reconstruction technology combined with deep neural networks, which can well learn the flow field characteristics, realize fast and accurate prediction of the flow field, and reduce the time consumption of flow field calculation; the rapid flow field prediction model proposed in the present invention can be applied to the steady flow field prediction of airfoils whose appearance and working conditions change simultaneously, and has a high model generalization ability; in the process of processing flow field data, the present invention first maps the flow field data from the aerodynamic calculation domain to the learning domain, and performs interpolation processing on the places where the data gradient is large, which can effectively improve the modeling accuracy and reduce the prediction error of the flow field.

[0055] It is obvious to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential features of the present invention. Therefore, the embodiments should be regarded as exemplary and non-limiting from any point of view, and the scope of the present invention is defined by the appended claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention, and any reference numerals in the claims should not be regarded as limiting the claims involved.

[0056] In addition, it should be understood that although the present specification is described according to implementation modes, not every implementation mode contains only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment may also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.

Claims

1. A method for rapid prediction of airfoil flow field based on super-resolution reconstruction, characterized in that: The following steps are involved: Step 1): Generate a sample data set of airfoil with variable working conditions and shapes; the specific steps are as follows: Step 1-1): Select a certain type of airfoil, determine the airfoil shape parameter design space and calculation condition design space, and use Latin hypercube sampling to obtain a series of airfoil samples; Step 1-2): Generate aerodynamic calculation grid of airfoil; Step 1-3): Perform CFD simulation calculation on the airfoil sample obtained in step 1-1) to obtain original flow field data of the airfoil sample; Step 2): The original flow field data in steps 1-3) are processed by domain mapping, data interpolation, and downsampling to obtain high- and low-resolution flow field data; the specific steps are as follows: Step 2-1): The original flow field data in step 1-3) has the same grid size as that in step 1-2), which is 210×150. The original flow field data is mapped from the aerodynamic calculation domain to the learning domain and represented as a 210×150 orthogonally uniformly distributed pixel image; Step 2-2): interpolate the area with large gradient change of the flow field data after domain mapping in step 2-1) to obtain a 315×150 pixel image as high-resolution flow field data; Step 2-3): The downsampling scaling factor is 5, and the high-resolution flow field data in step 2-2) is downsampled to obtain low-resolution flow field data, whose size is 63×30; Step 3): Use a deep neural network DNN to establish a prediction model from the working conditions and shape parameters in step 1-1) to the low-resolution flow field data in step 2); Step 4): Use an ultra-deep super-resolution neural network VDSR to establish a prediction model from the low-resolution flow field data in step 2) to the high-resolution flow field data in step 2); Step 5): connect the DNN in step 3) and the VDSR in step 4) into a DNN-VDSR model to complete the modeling of the shape and working condition parameters to the high-resolution flow field data; Step 6): The DNN-VDSR model in step 5) is used to quickly predict the steady flow field of the airfoil with changed shape and working conditions; the specific steps are as follows: Step 6-1): Given the shape parameters and geometric parameters of the new airfoil, the RBF mesh deformation technology of the aerodynamic calculation mesh in step 1-2) is used to obtain the mesh of the new airfoil, with a size of 210×150; Step 6-2): Use the shape and geometric parameters in step 6-1) as the input of the DNN-VDSR trained in step 5) to predict high-resolution flow field data with a size of 315×150; Step 6-3): Perform reverse interpolation on the high-resolution flow field data predicted in step 6-2) to obtain a size of 210×150, and map the data to the grid in step 6-1) to obtain the predicted flow field data.

2. The method for rapid prediction of airfoil flow field based on super-resolution reconstruction according to claim 1 is characterized in that: In the above step 1-1), the certain type of airfoil is an elliptical airfoil, and the total number of samples is 500, of which 90% are used for training and 10% are used for testing; the shape parameters are the camber and thickness of the elliptical airfoil, and the calculation conditions of the thickness are the angle of attack and the Mach number, the camber value range is 0 to 0.03c, the thickness value range is 0.13c to 0.19c, the angle of attack value range is 1 to 4°, and the Mach number value range is 0.15 to 0.

45.

3. The method for rapid prediction of airfoil flow field based on super-resolution reconstruction according to claim 1 is characterized in that: In the above steps 1-2), the aerodynamic calculation grid is an O-type grid, the grid size is 210×150 in the circumferential direction and the first layer height of the boundary layer is 8×10 -6 c. First, an O-shaped grid is divided for any airfoil as the base grid. The grids of the remaining airfoils are obtained based on the base grid using the RBF grid deformation technology. The size of all grids is 210×150.

4. The method for rapid prediction of airfoil flow field based on super-resolution reconstruction according to claim 1 is characterized in that: In the above steps 1-3), the CFD calculation model adopts the SST k-omega turbulence model.

5. The method for rapid prediction of airfoil flow field based on super-resolution reconstruction according to claim 1 is characterized in that: In the above step 3), a total of 4 parameters including the shape parameters and geometric parameters in step 1-1) are used as the input of the deep neural network DNN, and the low-resolution flow field data in step 2-3) is divided into 30 regions, each with a dimension of 60, as the output of the DNN, and a total of 30 groups of DNN models are established.

6. The method for rapid prediction of airfoil flow field based on super-resolution reconstruction according to claim 1 is characterized in that: In the above step 4), the low-resolution flow field data in step 2-3) is used as the input of the ultra-deep super-resolution neural network VDSR, and the high-resolution flow field data in step 2-2) is used as the output of the ultra-deep super-resolution neural network VDSR, and a VDSR model is established.

7. The method for rapid prediction of airfoil flow field based on super-resolution reconstruction according to claim 1 is characterized in that: In the above step 5), the model established using DNN in step 3) and the model established using VDSR in step 4) are connected in series to form a DNN-VDSR airfoil flow field prediction model and train it.

Images