Feature learning and data fusion air film cooling prediction method based on neural operator

Abstract

The invention discloses a neural operator-based feature learning and data fusion air film cooling prediction method, which comprises the following steps of: obtaining a basic data set for feature learning, and establishing different geometric models by parameterizing hole distribution and blowing ratio conditions to obtain surface temperature distribution data; extracting surface adiabatic cooling efficiency data as output data, and drawing surface geometric data as input data; setting a total sample number, reading input and output data division tags, and dividing a training set and a test set; building an FNO-based iterative neural operator framework, simulating an operator solving process, and learning gas film cooling characteristics; inputting the input set data of the test set into the training model, and verifying the model by using the output set data of the test set to obtain a prediction model. According to the method, the air film cooling distribution condition is efficiently, quickly and accurately predicted through sparse dot matrix data of one sample.

Description

Technical Field

[0001] The present invention belongs to the technical field of engineering thermophysics, and in particular relates to a method for predicting film cooling based on feature learning and data fusion of a neural operator. Background Art

[0002] Film cooling technology is an important means to improve the high temperature resistance of key components such as turbine blades in aircraft engines. With the continuous increase in the operating temperature of modern aircraft engines, traditional cooling technology has gradually failed to meet the demand. Film cooling forms a layer of air film on the surface of the hot zone of the engine blade, effectively preventing the high-temperature airflow from directly contacting the blade surface, thereby cooling the blade. Film cooling plays a vital role in improving engine efficiency, extending engine service life and ensuring flight safety. Especially in the high-pressure turbine and combustion chamber areas, the design and optimization of film cooling has become the key to the development of aircraft engine technology. Accurately grasping the distribution of film cooling and optimizing the cooling scheme is one of the core research topics of current engineering thermophysics technology.

[0003] The methods for obtaining film cooling distribution mainly rely on experimental measurement and numerical simulation. Experimental measurement methods, such as thermocouples and infrared thermal imagers, can provide real-time surface temperature data, but are limited by factors such as measurement accuracy, sensor layout, and test environment, making it difficult to fully obtain the distribution of film cooling. In addition, experiments usually cannot provide dynamic changes under real operating conditions. Numerical simulation methods, especially CFD-based simulation technology, can provide more comprehensive cooling field data, but their computational costs are very high, especially under complex flow field conditions, and the simulation time and computing resources are consumed. In addition, low-precision numerical simulations often cannot obtain sufficient prediction accuracy, which makes the prediction of film cooling distribution based on these methods have great limitations. Therefore, how to achieve fast, accurate, and low-cost prediction of film cooling distribution has become a technical challenge that needs to be solved urgently.

[0004] With the rapid development of deep learning technology, more and more studies have tried to apply deep learning methods to the prediction of film adiabatic cooling efficiency. For example, convolutional neural networks (CNN) and recurrent neural networks (RNN) are used to predict film cooling distribution based on historical data and experimental measurement data. Although these methods have achieved good results when the data volume is large and the quality is high, they still have obvious shortcomings. First, deep learning models usually require a large amount of high-quality training data, which is very difficult in practical engineering applications, especially when complex physical processes are involved. High-quality training data is often scarce and difficult to obtain. Secondly, many deep learning models lack sufficient physical knowledge fusion, resulting in deficiencies in model generalization and extrapolation capabilities. This makes it difficult for these methods to provide accurate predictions when faced with new cooling conditions or special working conditions. Finally, the training process of deep learning models is complex, the consumption of computing resources is large, and it is difficult to apply them efficiently in practical engineering. Therefore, how to reduce the dependence on high-quality data while improving the generalization ability and accuracy of the model is still an important challenge facing deep learning in film cooling prediction.

[0005] As an emerging deep learning technology, the neural operator (NO) method has achieved remarkable results in solving many physical problems, especially in the fields of fluid dynamics and heat conduction. The Fourier neural operator (FNO) is a type of neural operator method that combines convolutional neural networks with Fourier transforms to achieve efficient modeling and solving of complex physical systems. The advantage of the FNO method is that it can handle data with periodic boundary conditions and high-dimensional input and output, and can effectively use the prior knowledge of the physical model to improve the accuracy and generalization ability of the model. Compared with traditional deep learning models, FNO processes in the frequency domain through Fourier transform, which significantly reduces the amount of calculation and improves the calculation efficiency. Especially when facing complex nonlinear problems, FNO shows better performance and efficiency. However, despite the good performance of FNO, its application in the prediction of film cooling distribution still faces the problem of large sample requirements. In order to better apply it to practical engineering, how to predict the film cooling distribution from a small number of samples or even sparse data of one sample is of great research significance. Summary of the invention

[0006] In order to solve the problem of efficiently, quickly and accurately predicting the film cooling distribution through the sparse lattice data of a sample, the purpose of the present invention is to provide a feature learning and data fusion film cooling prediction method based on neural operators, and learn the adiabatic film cooling distribution characteristics through feature learning of classic examples. At the same time, the data fusion method is used to fuse the learned features with the sparse lattice data of a single sample to predict the film cooling distribution under different conditions.

[0007] In order to achieve the above object, the present invention is implemented by the following technical solutions:

[0008] A method for predicting film cooling based on feature learning and data fusion of neural operators, comprising the following steps:

[0009] Step 1: Obtain the basic data set for feature learning, establish different geometric models by parameterizing the hole distribution and blowing ratio conditions, and perform calculations to obtain the surface temperature distribution data.

[0010] Step 2: Extract the surface adiabatic cooling efficiency data as output data, draw the surface geometry data as input data, and calculate the adiabatic cooling efficiency based on the wall temperature as follows:

[0011]

[0012] Where T m represents the mainstream inlet temperature, T c Indicates the cold air inlet temperature, T w represents the cooling wall temperature.

[0013] Step 3: Set the total number of samples, read the input and output data partition labels of step 2, and divide them into training set and test set.

[0014] Step 4: Build an iterative neural operator framework based on FNO, simulate the operator solution process, predict the evolution of film cooling, obtain the training model, and learn the characteristics of film cooling.

[0015] Step 5: Input the input set data of the test set into the training model, use the output set data of the test set to verify the model, obtain the prediction model, and compare the prediction results.

[0016] Step 6: Input the blowing ratio, geometric conditions and film cooling data scatter points of the new sample into the prediction model, and use the data fusion method to predict the complete film cooling distribution with the scattered data as constraints.

[0017] Furthermore, the specific steps of obtaining the basic data set in step 1 include:

[0018] Step 1-1: The total number of samples is 70. The Latin hypercube sampling method is used to randomly generate 70 groups of different flat film hole distribution point cloud data, and 70 groups of blowing ratio conditions are randomly generated.

[0019] Step 1-2: Use UG software to automatically model according to point cloud data to generate different geometric models

[0020] Step 1-3: Use fluent meshing software to automatically draw the mesh according to the generated geometric model, and use fluent software to automatically calculate with different blowing ratio conditions.

[0021] Step 1-4: Extract the surface temperature distribution as a txt file based on the calculation results.

[0022] Furthermore, the specific steps of training in step 4 include:

[0023] Step 4-1: Build an iterative neural operator framework based on FNO. The model expression of the FNO operator is as follows:

[0024] v t (x) = σ(W*v t-1 (x)+∫ D k(x,y,a(x),a(y))*v x (dy))

[0025] where v t represents the inlet temperature distribution data of different diffusion processes, σ represents the activation function, W represents the linear transformation part of the predicted value between two adjacent diffusion processes, k represents the Fourier transform and inverse transform process, and a represents the boundary condition. Therefore, the overall FNO operator model can represent the spatial diffusion evolution process.

[0026] Step 4-2: Input the initial value and boundary conditions (geometric conditions and blowing ratio conditions) of the blank. The data passes through multiple groups of FNO modules until the prediction results tend to be stable and the loop stops. In the prediction of steady-state results, this cyclic iteration process represents the process of convergence of the operator space solution results. The initial value of the blank gradually changes to an accurate prediction result. Combining the FNO operator and the cyclic iteration process, the operator solution process of spatial diffusion and solution convergence is simulated, thereby obtaining accurate prediction results.

[0027] Step 4-3: Import input and output data, standardize them, and randomly shuffle them into the input model framework for calculation.

[0028] Step 4-4: Use the mini-batch stochastic gradient descent algorithm and set the batch size to 8. Calculate the model error through forward propagation; update the model parameters through back propagation, and stop training when the loss is stable.

[0029] Furthermore, the specific steps of fusion in step 6 include:

[0030] Step 6-1: Save the feature learning model, freeze most of the parameters in the feature learning model, keep other parameters unchanged, and unfreeze the last two layers of parameters of the model layer by layer.

[0031] Step 6-2: Take the geometric conditions and blowing ratio conditions of the new sample as input, the scattered data as supervision values, constrain the prediction results, and retrain the parameters of the last two layers.

[0032] Step 6-3: After the loss is stabilized, the prediction result is output, that is, the complete film cooling distribution result in the new sample, thereby realizing the function of predicting the complete film cooling distribution from the scattered data of a single sample.

[0033] Beneficial effects of the present invention:

[0034] 1. The present invention can learn the characteristics of film cooling based on classic examples, and transfer this characteristic to a new film cooling model with different boundary conditions to predict the adiabatic cooling effect. For the new film cooling model, only a small amount of scattered data of a sample is needed to complete a fast, efficient and accurate prediction, which reduces the number of samples required for the prediction results of new research objects; at the same time, it reduces the requirement for the resolution of data points in the sample, and only a small amount (5%) of scattered data is needed to complete an accurate prediction.

[0035] 2. The present invention uses pixel difference as an indicator: in the comparison of single pixel results, the error of the training result adiabatic cooling efficiency prediction result is less than 0.01; the error of the test result adiabatic cooling efficiency prediction result is less than 0.04; in the average result comparison of all pixels in a sample, the error of the training result adiabatic cooling efficiency prediction result is less than 0.01; the error of the test result adiabatic cooling efficiency prediction result is less than 0.02. In a single sample with completely different inlet conditions and geometric conditions, when there are only 5% of the detection data points, the error is less than 0.05, and a single sample can complete efficient and accurate prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 Latin hypercube sampling hole distribution results.

[0037] Figure 2 Schematic diagram of the film cooling calculation model.

[0038] Figure 3 The full surface temperature distribution results.

[0039] Figure 4 Results of full-surface adiabatic cooling efficiency distribution.

[0040] Figure 5 Operator framework diagram based on FNO.

[0041] Figure 6 Diagram of the iterative neural operator framework based on FNO.

[0042] Figure 7 Schematic diagram of the iterative process of the sample.

[0043] Figure 8 Comparison chart of prediction results for the first test sample.

[0044] Fig. 9 Comparison chart of prediction results for the second test sample.

[0045] Fig.10 Comparison chart of prediction results for the third test sample.

[0046] Fig.11 Comparison chart of the prediction results of the 4th test sample.

[0047] Fig.12 Diagram of the new sample geometry.

[0048] Fig.13 Scattered data distribution results graph.

[0049] Fig.14 Single sample prediction result graph. DETAILED DESCRIPTION

[0050] The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, but should not be construed as limiting the present invention.

[0051] The embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0052] A method for predicting film cooling based on feature learning and data fusion of neural operators, comprising the following steps:

[0053] Step 1: First, the mathematical representation of the variables is determined by parameterizing the hole distribution and blowing ratio conditions, and combined with the Latin hypercube sampling method, 70 groups of different flat film hole distribution point cloud data are randomly generated (some samples are as follows Figure 1 As shown in the figure, 70 sets of blowing ratio conditions were randomly generated. UG software was used to automatically model the point cloud data and generate different geometric models, in which the hole diameter was 1mm, the inclination angle was 30°, and the number of holes was randomly generated in the range of 10-15. Fluent meshing software was used to automatically draw the mesh based on the generated geometric model, and fluent software was used to automatically calculate the different blowing ratio conditions. Figure 2 The figure shows the hot and cold air inlets and outlets and the geometric shapes of the model. Figure 3 The full surface temperature distribution diagram of 70 samples.

[0054] Step 2: If Figure 4 The surface adiabatic cooling efficiency data is extracted as output data, and the surface geometry data is drawn as input data. The formula for calculating the adiabatic cooling efficiency based on temperature is as follows:

[0055]

[0056] Where T m represents the mainstream inlet temperature, T c Indicates the cold air inlet temperature, T w represents the cooling wall temperature.

[0057] Step 3: Set the total number of samples to 70, read the input and output data partition labels from step 2, and divide them into training sets and test sets. The number of training sets is 60, and the number of test sets is 10.

[0058] Step 4: Build an iterative neural operator framework based on FNO. Figure 5 Based on the operator framework diagram of FNO, the model expression of the FNO operator is as follows:

[0059] v t (x) = σ(W*v t-1 (x)+∫ D k(x,y,a(x),a(y))*v x (dy))

[0060] where v t represents the inlet temperature distribution data of the spatial diffusion process, σ represents the activation function, W represents the linear transformation part of the predicted value between two adjacent diffusion steps, k represents the Fourier transform and inverse transform process, and a represents the boundary condition. Therefore, the overall FNO operator model can represent the spatial diffusion evolution process.

[0061] like Figure 6 As shown, an iterative neural operator framework based on FNO is built to simulate the operator solution process, predict the evolution process of film cooling, obtain a training model, and learn the characteristics of film cooling. The initial value and boundary conditions (geometric conditions and blowing ratio conditions) of the blank are input, and the data passes through multiple groups of FNO modules until the prediction result tends to be stable and the loop is stopped. In the prediction of steady-state results, this cyclic iterative process represents the process of convergence of the operator space solution result. The initial value of the blank gradually changes to an accurate prediction result. The hyperparameter of the number of iterations in the test process of the present invention was tested with 5, 10, 20 and other situations. It was finally found that five iterations had basically reached stability (such as Figure 7 By combining the FNO operator and the cyclic iteration process, the operator solution process of spatial diffusion and solution convergence is simulated to obtain accurate prediction results.

[0062] Import input and output data, standardize them, and randomly shuffle them into the model framework for calculation. Use the small batch stochastic gradient descent algorithm and set the batch to 8. Calculate the model error through forward propagation and update the model parameters through back propagation. Stop training when the loss is stable. A single training session takes about 1.6 seconds, and the loss function stops decreasing when the training step is 100.

[0063] Step 5: Input the input data of the test set into the training model, use the output data of the test set to verify the model, obtain the prediction model, and compare the prediction results (such as Figure 8 to Figure 11 ).

[0064] Step 6: Input the blowing ratio, geometric conditions and film cooling data of the new sample into the prediction model, and use the data fusion method to predict the complete film cooling distribution with the scattered data as constraints. The comparison between the basic 70 sample conditions and the new sample conditions is shown below:

[0065]

[0066] The geometric model of the new sample is as follows Fig.12 As shown in the figure, save the feature learning model, freeze most of the parameters in the feature learning model, keep other parameters unchanged, and unfreeze the last two layers of the model layer by layer. Take the geometric conditions and blowing ratio conditions of the new sample as input, and the scattered data as the supervision value (such as Fig.13 As shown, the scattered points only account for 5% of the total points, which are generated by random functions. The prediction results are constrained and the parameters of the last two layers are retrained. When the loss function calculation results are stable, the prediction results are output, that is, the complete film cooling distribution results in the new samples, such as Fig.14 The function shown implements the prediction of the complete film cooling distribution from the scattered data of a single sample.

[0067] The present invention learns the adiabatic film cooling distribution characteristics through feature learning of classic examples. At the same time, the data fusion method is used to fuse the learned features with the sparse lattice data of a single sample to predict the film cooling distribution under different conditions, solving the problem of efficiently, quickly and accurately predicting the film cooling distribution through the sparse lattice data of a sample.

Claims

1. A feature learning and data fusion film cooling prediction method based on neural operators, characterized in that: The following steps are involved: Step 1: Obtain the basic data set for feature learning, establish different geometric models by parameterizing the hole distribution and blowing ratio conditions, and perform calculations to obtain the surface temperature distribution data; Step 2: Extract the surface adiabatic cooling efficiency data as output data, draw the surface geometry data as input data, and calculate the adiabatic cooling efficiency based on the wall temperature as follows: Where T m represents the mainstream inlet temperature, T c Indicates the cold air inlet temperature, T w represents the cooling wall temperature; Step 3: Set the total number of samples, read the input and output data partition labels of step 2, and divide them into training sets and test sets; Step 4: Build an iterative neural operator framework based on FNO, simulate the operator solution process, predict the evolution of film cooling, obtain the training model, and learn the characteristics of film cooling; Step 5: Input the input set data of the test set into the training model, use the output set data of the test set to verify the model, obtain the prediction model, and compare the prediction results; Step 6: Input the blowing ratio, geometric conditions and film cooling data scatter points of the new sample into the prediction model, and use the data fusion method to predict the complete film cooling distribution with the scattered data as constraints.

2. The feature learning and data fusion film cooling prediction method based on neural operators according to claim 1 is characterized in that: The specific steps to obtain the basic data set in step 1 include: Step 1-1: The total number of samples is 70. The Latin hypercube sampling method is used to randomly generate 70 groups of different flat film hole distribution point cloud data, and 70 groups of blowing ratio conditions are randomly generated; Step 1-2: Use UG software to perform automatic modeling based on point cloud data to generate different geometric models; Step 1-3: Use fluent meshing software to automatically draw meshes based on the generated geometric model, and use fluent software to automatically calculate with different blowing ratio conditions; Step 1-4: Extract the surface temperature distribution as a txt file based on the calculation results.

3. The feature learning and data fusion film cooling prediction method based on neural operators according to claim 1 is characterized in that: The specific steps of training in step 4 include: Step 4-1: Build an iterative neural operator framework based on FNO. The model expression of the FNO operator is as follows: where v t represents the inlet temperature distribution data of different diffusion processes, σ represents the activation function, W represents the linear transformation part of the predicted value between two adjacent diffusion processes, k represents the Fourier transform and inverse transform process, and a represents the boundary condition. Therefore, the overall FNO operator model represents the spatial diffusion evolution process; Step 4-2: Input the initial value and boundary conditions of the blank, and the data passes through multiple groups of FNO modules until the prediction results tend to be stable, and then stop the loop; in the prediction of steady-state results, this cyclic iteration process represents the process of convergence of the operator space solution results; the initial value of the blank gradually changes to an accurate prediction result, and the FNO operator and cyclic iteration process are combined to simulate the spatial diffusion and convergence of the operator solution process, thereby obtaining an accurate prediction result; Step 4-3: Import input and output data, perform standardization, and randomly shuffle them into the model framework for calculation; Step 4-4: Use the mini-batch stochastic gradient descent algorithm, set the batch size to 8, calculate the model error through forward propagation, update the model parameters through back propagation, and stop training when the loss is stable.

4. The feature learning and data fusion film cooling prediction method based on neural operators according to claim 1 is characterized in that: The specific steps of fusion in step 6 include: Step 6-1: Save the feature learning model, freeze most of the parameters in the feature learning model, keep other parameters unchanged, and unfreeze the last two layers of parameters of the model layer by layer; Step 6-2: Take the geometric conditions and blowing ratio conditions of the new sample as input, the scattered data as supervision values, constrain the prediction results, and retrain the parameters of the last two layers; Step 6-3: After the loss is stabilized, the prediction result is output, that is, the complete film cooling distribution result in the new sample, thereby realizing the function of predicting the complete film cooling distribution from the scattered data of a single sample.

Images