Gas film cooling efficiency distribution prediction method and system and computer readable medium
By constructing a film cooling efficiency prediction model based on PointNet and CGAN, the problem of long prediction time for film cooling performance is solved, and a fast and accurate prediction of film cooling efficiency distribution is achieved, which is applicable to complex structures.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-27
- Publication Date
- 2026-04-07
AI Technical Summary
Existing technologies are computationally expensive and time-consuming in predicting the performance of film cooling, making it difficult to meet the needs of rapid iterative design. Traditional models also have limitations when dealing with complex two-dimensional cooling efficiency distributions.
A film cooling efficiency prediction model based on the deep learning model PointNet and the conditional generative adversarial network CGAN is constructed. By constructing a film cooling dataset and training the model, the efficient prediction of the film cooling efficiency distribution is achieved.
It significantly reduces computing costs, shortens prediction time from hours to seconds, maintains high accuracy with an absolute error of no more than 0.05 and a relative error of no more than 6%, and is suitable for complex film cooling structures.
Smart Images
Figure CN121808685A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of gas film cooling, in particular to a gas film cooling efficiency distribution prediction method and system and a computer readable medium. BACKGROUND
[0002] Gas film cooling technology is a key technology for protecting high-temperature components (such as turbine blades) in aero-engines and gas turbines from thermal damage. The principle is to introduce cooling air flow into the surface of high-temperature components through a specific gas film hole structure to form a continuous gas film heat insulation layer, thereby effectively blocking the heat conduction and convective heat transfer of high-temperature gas to the base material, reducing the surface temperature to achieve cooling effect. With the continuous increase of turbine inlet temperature, the gas film cooling structure is becoming more and more complex, and typical representatives include fan-shaped holes, cylindrical holes, and new gas film cooling structures such as sawtooth grooves. These complex geometries not only improve the gas film cooling efficiency, but also increase the complexity of flow and heat transfer.
[0003] Currently, the research on gas film cooling performance mainly relies on computational fluid dynamics (CFD) simulation. Although the traditional CFD method can provide high-resolution flow field and cooling efficiency distribution, it is necessary to re-divide the grid and set the solution every time the geometric parameters or operating conditions (such as blowing ratio, density ratio) change, and the single case calculation often takes a lot of time. In actual engineering design, it is difficult to meet the needs of rapid iterative design.
[0004] In order to reduce the calculation cost and improve the optimization efficiency, researchers try to introduce surrogate models (such as random forest, long short-term memory network and neural network) to predict the film cooling efficiency. Although these methods reduce the calculation cost to some extent, the traditional model still has limitations in dealing with complex two-dimensional cooling efficiency distribution. Therefore, there is an urgent need for an efficient method that can quickly and accurately predict the gas film cooling efficiency distribution to support the design of cooling systems for aero-engines and gas turbines. SUMMARY
[0005] Therefore, the present application provides a gas film cooling efficiency distribution prediction method, system and computer readable medium to solve the above technical problems.
[0006] To solve the above technical problems, the technical solution adopted by the present application is as follows: a gas film cooling efficiency distribution prediction method, comprising the following steps: S1, constructing a gas film cooling data set, including gas film cooling efficiency distribution data under different parameter conditions; S2, constructing a gas film cooling efficiency prediction model based on a deep learning model PointNet and a conditional generative adversarial network CGAN, including a feature extraction module, a generator and a discriminator; S3, training the gas film cooling efficiency prediction model using the gas film cooling data set; S4, using the trained gas film cooling efficiency prediction model to predict the gas film cooling efficiency, and realizing the generation of cooling efficiency distribution under different parameter conditions.
[0007] Further, the method for constructing the gas film cooling data set comprises: S11, constructing a gas film cooling calculation model; S12, according to the set parameter condition, extracting the grid nodes located on the wall surface boundary from the calculation grid of the gas film cooling calculation model, each node containing spatial position coordinates and corresponding temperature; S13, for each parameter condition combination, obtaining the two-dimensional distribution of the bottom cooling efficiency of the calculation model, the cooling efficiency The calculation formula is: wherein, is the adiabatic wall surface temperature, is the mainstream temperature, is the coolant temperature; S14, according to the spatial position coordinates and corresponding temperature of each node under different parameter conditions, constructing the gas film cooling data set.
[0008] Further, the parameter conditions include blowing ratio, slot height and sawtooth angle.
[0009] Further, the feature extraction module is constructed by a deep learning model PointNet, the feature spaces of the generator and the discriminator are constructed by a conditional generative adversarial network CGAN; the feature extraction module extracts global features and local features from the input gas film cooling data set, and splices them into a tensor containing global features and local features, which is input into the generator, the generator maps the tensor into a two-dimensional gas film cooling efficiency distribution image, the authenticity of the generated two-dimensional gas film cooling efficiency distribution image is evaluated by the discriminator, and the generator and the discriminator are optimized by adversarial learning in the training process, realizing the mapping between high-dimensional parameters and cooling efficiency distribution.
[0010] Further, the generator inputs the spliced tensor, obtains a feature map with a resolution of 64x64 through a full connection layer, and then goes through 4 layers of convolutional network and up-sampling, the first 3 layers go through a Relu activation function, gradually improving the feature map resolution to 1024x1024, each layer uses a 3x3 convolution kernel, the stride is 1, and the padding is 1, and finally connects a Sigmoid activation function to output a two-dimensional gas film cooling efficiency distribution image with a pixel value range of [0, 1].
[0011] Further, the discriminator inputs the two-dimensional gas film cooling efficiency distribution image and the spliced tensor, the two-dimensional gas film cooling efficiency distribution image is down-sampled to 64*64 resolution through a 4-layer convolutional network, and is spliced with the reshaped and bilinear up-sampled tensor; the network adopts a LeakyReLU activation function and a Batch Normalization layer to enhance the feature extraction capability, and finally outputs a discrimination probability through a Sigmoid function.
[0012] Further, in the gas film cooling efficiency prediction model training process, a BCE loss function and an Adam optimizer are used for optimization.
[0013] Another object of the present application provides a gas film cooling efficiency distribution prediction system, comprising: A data set construction module is configured to construct a gas film cooling data set comprising gas film cooling efficiency distribution data under different parameter conditions; A model construction module is configured to construct a gas film cooling efficiency prediction model based on a deep learning model PointNet and a conditional generative adversarial network CGAN, including a feature extraction module, a generator and a discriminator; A model training module is configured to train the gas film cooling efficiency prediction model based on the deep learning model PointNet and the conditional generative adversarial network CGAN by using the gas film cooling data set; A prediction module is configured to perform gas film cooling efficiency prediction according to the trained gas film cooling efficiency prediction model.
[0014] Another object of the present application provides a computer readable medium storing a computer program, the computer program being executed to realize the above-mentioned gas film cooling efficiency distribution prediction method.
[0015] Another object of the present application provides a computer device comprising a memory and a processor, the memory storing a computer program, the computer program being configured to execute the above-mentioned gas film cooling efficiency distribution prediction method by the processor.
[0016] Compared with the prior art, the present application has the following advantages: The gas film cooling efficiency distribution prediction method of the present application trains a gas film cooling efficiency prediction model based on a deep learning model PointNet and a conditional generative adversarial network CGAN using a gas film cooling data set, and uses the trained gas film cooling efficiency prediction model to predict the gas film cooling efficiency. Compared with the traditional CFD method, the present application significantly reduces the calculation cost, shortens the prediction time from several hours to several seconds, while maintaining high accuracy (absolute error range not more than 0.05, relative error range not more than 6%); the network model captures the complex nonlinear relationship between three-dimensional space coordinates and temperature and cooling efficiency, and the generated two-dimensional cloud map is rich in details, which can be applied to new gas film cooling structures such as fan-shaped holes, cylindrical holes and zigzag grooves. BRIEF DESCRIPTION OF DRAWINGS
[0017] Figure 1 is a method flowchart of an embodiment of the present application; Figure 2 is a gas film cooling calculation model schematic diagram in an embodiment of the present application; Figure 3 is a deep learning model PointNet network structure schematic diagram in an embodiment of the present application; Figure 4 is a conditional generative adversarial network CGAN network structure schematic diagram in an embodiment of the present application; Figure 5 is a centerline gas film cooling efficiency comparison diagram in an embodiment of the present application; Figure 6 is a comparison diagram of prediction results and CFD simulation results in an embodiment of the present application; Figure 7 is an error distribution diagram of a training set in an embodiment of the present application; Figure 8 is an error distribution diagram of a test set in an embodiment of the present application; Figure 9 is a comparison diagram of CFD simulation results and prediction results when the blowing ratio is less than 0.5 in an embodiment of the present application; Figure 10 is a comparison diagram of CFD simulation results and prediction results when the blowing ratio is greater than 2.0 in an embodiment of the present application. DETAILED DESCRIPTION
[0018] To make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme of the embodiments of the present application will be described clearly and completely below with reference to the drawings of the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all the embodiments. Based on the described embodiments of the present application, all other embodiments obtained by those skilled in the art belong to the scope of protection of the present application.
[0019] Embodiment 1 Referring to Figure 1 The embodiment of the present application provides a kind of gas film cooling efficiency distribution prediction method, with sawtooth channel gas film cooling structure as example, steps are as follows: S1, construct gas film cooling dataset, including the gas film cooling efficiency distribution data under different parameter conditions;Specific construction method is as follows: (1) construct gas film cooling calculation model; The data set required for training and testing is generated by numerical simulation, specifically, CFD simulation is carried out by using commercial software ANSYS FLUENT 19.2, and sawtooth channel gas film cooling calculation model is constructed;The boundary conditions of calculation domain are as shown in Figure 2 : the main flow inlet is set as velocity inlet, the temperature is 321K, the velocity is 13.8m / s, and the turbulence intensity is 5%;The coolant inlet temperature is 296 K, the turbulence intensity is 5%, and the velocity is determined by blowing ratio and main flow velocity;The outlet is pressure outlet, and the static pressure is 1atm;Periodic boundary condition is used for side wall, and the rest of the wall is adiabatic non-slip wall.
[0020] (2) according to the set parameter condition, the grid nodes located on the wall boundary are extracted from the calculation grid of gas film cooling calculation model, each node includes spatial position coordinates and corresponding temperature; For sawtooth channel gas film cooling structure, parameter conditions include blowing ratio, channel height and sawtooth angle; To ensure the diversity and representativeness of data, Latin hypercube sampling (LHS) method is used to uniformly sample from parameter space, and 505 groups of samples are generated;Parameter conditions are set as follows: blowing ratio (M) 0.5~2.0, channel height (H) 0.0127mm~9.525mm and sawtooth angle (α) 22.5°~60.0°;Among them, 404 groups are used for training, and 101 groups are used for testing;To verify grid independence, 2.37 million to 13.1 million grid elements are tested, and it is found that 8 million grid elements can ensure the balance between calculation accuracy and cost, so this grid size is used.
[0021] (3) for each parameter condition combination, run ANSYS CFX simulation to obtain the two-dimensional distribution of cooling efficiency at the bottom surface of the calculation model, as CFD simulation result, the calculation formula of cooling efficiency is as follows: Wherein, is the adiabatic wall temperature, is the main flow temperature, is the coolant temperature; (4) according to the spatial position coordinates of each node under different parameter conditions, corresponding temperature, constitute gas film cooling dataset.
[0022] S2, a gas film cooling efficiency prediction model based on a deep learning model PointNet and a conditional generative adversarial network CGAN is constructed, including a feature extraction module, a generator and a discriminator; The feature extraction module is constructed by the deep learning model PointNet, and the feature spaces of the generator and the discriminator are constructed by the conditional generative adversarial network CGAN; Referring to Figure 3 The feature extraction module extracts global features and local features from the input gas film cooling data set, and splices them into a tensor containing global features and local features, which is input into the generator. The generator maps the tensor to a two-dimensional gas film cooling efficiency distribution image. The authenticity of the generated two-dimensional gas film cooling efficiency distribution image is evaluated by the discriminator. In the training process, the generator and the discriminator are optimized through adversarial learning to realize the mapping between high-dimensional parameters and cooling efficiency distribution. Specifically as follows: Referring to Figure 4 The generator inputs the spliced tensor, obtains a feature map with a resolution of 64x64 through a full connection layer (Linear), and then passes through 4 layers of convolutional network (Conv2d) and upsampling. The first 3 layers pass through the Relu activation function, and the feature map resolution is gradually increased to 1024x1024. Each layer uses a 3x3 convolution kernel, a stride of 1, and a padding of 1. Finally, a Sigmoid activation function is connected to output a two-dimensional gas film cooling efficiency distribution image with a pixel value range of [0, 1]; The discriminator inputs the two-dimensional gas film cooling efficiency distribution image and the 512-dimensional tensor after splicing. The two-dimensional gas film cooling efficiency distribution image is downsampled to 64x64 resolution through 4 layers of convolutional network, and is spliced with the tensor after reshaping and bilinear upsampling (Bilinear Upsampling). The network uses LeakyReLU activation function and BatchNormalization layer to enhance feature extraction capability. Finally, the Sigmoid function is used to output the discrimination probability.
[0023] S3, training the gas film cooling efficiency prediction model using the gas film cooling data set; The training process is realized using the PyTorch framework on an NVIDIA GeForce RTX 3090 GPU. In the training process, the BCE loss function and the Adam optimizer are used for optimization: The loss function of the discriminator is defined as: In the formula: is a noise vector, is a CFD real image, is conditional information, is a binary cross-entropy loss, is a generated image, To judge the discriminator in given conditions Next, the real image The judgment result; The optimizer selects Adam, and the parameter settings are: the initial learning rates of the generator and the discriminator are set to 1x10 ⁻4 and 2x10⁻ 4 , the momentum parameters =0.5, =0.999, the batch size is 8, and the training round number is 500. After training is completed, the model realizes stable prediction performance on the test set.
[0024] S4, using the trained film cooling efficiency prediction model to predict the film cooling efficiency, realizing the generation of cooling efficiency distribution under different parameter conditions, the specific process is as follows: Referring to Figure 5 , the coordinates and temperature of the case corresponding to the point to be predicted are input, the PointNet extracts the feature and inputs the generator, and the generator generates a cooling efficiency cloud map with a resolution of 1024x1024. The prediction result is compared with the CFD simulation result (see Figure 6 ), and the centerline cooling efficiency curve is highly consistent with the CFD result at M=0.5, 1.0, 1.5 and 2.0. And the time consumption of single prediction is several seconds, compared with the calculation time of several hours of CFD simulation, the efficiency is improved by hundreds of times.
[0025] To further verify the model performance, the error distribution of the training set and the test set is analyzed: as shown in Figure 7 , Figure 8 The absolute error and relative error analysis of 404 training samples and 101 test samples shows that the prediction ability of the model on the training set and the test set is good, wherein the absolute error is not more than 0.05, and the relative error is not more than 6%.
[0026] It is known that the prediction result of the model in the training set parameter is good, and the performance of the model under the condition of exceeding the parameter range of the training set is further analyzed. As shown in Figure 9 , Figure 10 When the blowing ratio is less than 0.5 or greater than 2.0, the prediction result of the model will have a large deviation, indicating that the model needs to run within the range of the training distribution.
[0027] In the embodiment of the application, the sawtooth-shaped channel is taken as an example, the parameter conditions of the premise are the blowing ratio, the channel height and the sawtooth angle, the input is the three-dimensional coordinates and the temperature, and the output cooling efficiency distribution is in the form of a two-dimensional cloud map. The method of the application has no limitation on the geometry of the film hole, and is also applicable to simple or composite hole types such as cylindrical holes and fan-shaped holes.
[0028] Embodiment 2 The embodiment of the present application provides a gas film cooling efficiency distribution prediction system, comprising: A data set construction module is configured to construct a gas film cooling data set comprising gas film cooling efficiency distribution data under different parameter conditions. A model construction module is configured to construct a gas film cooling efficiency prediction model based on a deep learning model PointNet and a conditional generative adversarial network CGAN, comprising a feature extraction module, a generator and a discriminator. A model training module is configured to train the gas film cooling efficiency prediction model based on the deep learning model PointNet and the conditional generative adversarial network CGAN by using the gas film cooling data set. A prediction module is configured to perform gas film cooling efficiency prediction according to the trained gas film cooling efficiency prediction model.
[0029] The specific configuration method of each module corresponds to the embodiment 1, and thus is not described in detail.
[0030] Embodiment 3 The embodiment of the present application provides a computer readable medium, which stores a computer program, and the computer program is executed to realize the gas film cooling efficiency distribution prediction method in the embodiment 1.
[0031] Embodiment 4 The embodiment of the present application provides a computer device, which comprises a memory and a processor, the memory stores a computer program, and the computer program is configured to execute the gas film cooling efficiency distribution prediction method in the embodiment 1 by the processor.
[0032] The above embodiments are only used to illustrate the technical solutions of the present application rather than limit the same, and although the present application is described in detail with reference to the above embodiments, the ordinary skilled in the art can still modify or equivalently replace the specific embodiments of the present application without departing from the spirit and scope of the present application, and any modification or equivalent replacement without departing from the spirit and scope of the present application is within the protection scope of the claims of the present application.
Claims
1. A method for predicting the distribution of film cooling efficiency, characterized in that, Includes the following steps: S1. Construct a film cooling dataset, including film cooling efficiency distribution data under different parameter conditions; S2. Construct a prediction model for air film cooling efficiency based on the deep learning model PointNet and the conditional generative adversarial network CGAN, including a feature extraction module, a generator, and a discriminator. S3. Train a film cooling efficiency prediction model using the film cooling dataset; S4. Use the trained film cooling efficiency prediction model to predict film cooling efficiency and generate cooling efficiency distribution under different parameter conditions.
2. The method for predicting the distribution of film cooling efficiency according to claim 1, characterized in that, Methods for constructing air-film cooling datasets include: S11. Construct a calculation model for film cooling; S12. Based on the set parameter conditions, extract the grid nodes located on the wall boundary from the calculation grid of the air film cooling calculation model. Each node contains spatial location coordinates and the corresponding temperature. S13. For each combination of parameter conditions, obtain the two-dimensional distribution of the cooling efficiency at the bottom of the calculation model, and the cooling efficiency. The calculation formula is: in, It is the temperature of the insulating wall. It is the mainstream temperature. It is the coolant temperature; S14. Based on the spatial coordinates of each node and the corresponding temperature under different parameter conditions, construct the air film cooling dataset.
3. The method for predicting the distribution of film cooling efficiency according to claim 2, characterized in that: The parameters include the air blowing ratio, channel height, and sawtooth angle.
4. The method for predicting the distribution of film cooling efficiency according to claim 2, characterized in that: The feature extraction module is constructed using the deep learning model PointNet. The feature spaces of the generator and discriminator are both constructed using the conditional generative adversarial network CGAN. The feature extraction module extracts global and local features from the input film cooling dataset and concatenates them into a tensor containing global and local features. This tensor is then input into the generator, which maps the global and local features into a two-dimensional film cooling efficiency distribution image. The discriminator evaluates the authenticity of the generated two-dimensional film cooling efficiency distribution image. During training, the generator and discriminator are optimized through adversarial learning to achieve the mapping between high-dimensional parameters and cooling efficiency distribution.
5. The method for predicting the distribution of film cooling efficiency according to claim 4, characterized in that: The generator takes the concatenated tensor as input, passes through a fully connected layer to obtain a feature map with a resolution of 64×64, then passes through a 4-layer convolutional network and upsampling. The first 3 layers pass through a ReLU activation function to gradually increase the feature map resolution to 1024×1024. Each layer uses a 3×3 convolutional kernel with a stride of 1 and padding of 1. Finally, it is connected to a Sigmoid activation function to output a two-dimensional air film cooling efficiency distribution image with pixel values ranging from [0, 1].
6. The method for predicting the distribution of film cooling efficiency according to claim 5, characterized in that: The discriminator takes a two-dimensional air film cooling efficiency distribution image and a stitched tensor as input. The two-dimensional air film cooling efficiency distribution image is downsampled to 64×64 resolution through a 4-layer convolutional network and stitched with the reshaped tensor that has been upsampled bilinearly. The network uses the LeakyReLU activation function and a Batch Normalization layer to enhance feature extraction capabilities, and finally outputs the discrimination probability through the Sigmoid function.
7. The method for predicting the distribution of film cooling efficiency according to claim 6, characterized in that: During the training of the air film cooling efficiency prediction model, the BCE loss function and Adam optimizer are used for optimization.
8. A system for predicting the distribution of film cooling efficiency, characterized in that, include: The dataset construction module is used to construct a film cooling dataset that includes film cooling efficiency distribution data under different parameter conditions. The model building module is used to build a prediction model for air film cooling efficiency based on the deep learning model PointNet and the conditional generative adversarial network CGAN, including a feature extraction module, a generator, and a discriminator. The model training module is used to train a film cooling efficiency prediction model based on the deep learning model PointNet and the conditional generative adversarial network CGAN using the film cooling dataset. The prediction module is used to predict the air film cooling efficiency based on the trained air film cooling efficiency prediction model.
9. A computer-readable medium storing a computer program, characterized in that: When the computer program is executed, it implements the gas film cooling efficiency distribution prediction method as described in any one of claims 1 to 7.
10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: The computer program is configured to be executed by the processor using the film cooling efficiency distribution prediction method as described in any one of claims 1 to 7.