Rapid prediction method for three-dimensional temperature field of turbine guide vane cooling structure

Abstract

The invention discloses a rapid prediction method for a three-dimensional temperature field of a turbine guide blade cooling structure. The method comprises the following steps: sequentially constructing an original temperature field matrix, a temperature field snapshot matrix and a temperature field covariance matrix according to a training set; solving the covariance matrix of the temperature field to obtain a feature vector and a feature value, calculating an energy value by using the feature value, and determining the number of primary functions and spectral coefficients and corresponding values; determining the number of ANN models according to the number of the spectral coefficients and the number of input boundary conditions; constructing a loss function of a spectral coefficient ANN model matched with the POD method characteristics according to the POD characteristics; and predicting a spectral coefficient of an unknown working condition by using the trained ANN model, and combining the spectral coefficient with the primary function to quickly reconstruct a three-dimensional temperature field of the turbine guide vane cooling structure of the unknown working condition. According to the method, a plurality of ANN models with mutually overlapped prediction results are constructed, and the defect that the prediction precision is low due to the fact that the characteristics of the POD method are not considered when an existing POD method is combined with a neural network method is overcome.

Description

Technical Field

[0001] The invention relates to the technical field of turbine guide blade research, and in particular to a method for quickly predicting a three-dimensional temperature field of a turbine guide blade cooling structure. Background Art

[0002] Turbine guide vanes are key components in high-performance mechanical systems such as aircraft engines and gas turbines. Their main function is to convert the kinetic energy and thermal energy of high-temperature gases into mechanical energy. When operating for a long time in an extremely high temperature environment, turbine guide vanes are prone to thermal fatigue and metal material aging, which seriously affects the safety of the engine. Blade thermal analysis is the key to the design and optimization of turbine guide vane cooling structure. At present, the thermal analysis calculation of blades generally uses a full three-dimensional gas-heat coupling numerical simulation method to conduct a detailed analysis of the internal flow characteristics and internal and external heat transfer characteristics of the blades. However, the design and optimization of the blade cooling structure usually requires continuous adjustment of the structure. The repeated three-dimensional temperature field calculations caused by a large number of adjustments consume a lot of computing resources and time costs, resulting in a significant reduction in the design efficiency of the turbine guide vane cooling structure.

[0003] The proper orthogonal decomposition method (POD) can extract the main features from complex high-dimensional data, and realize the low-dimensional representation of high-dimensional physical fields through the linear combination of a small number of high-energy modes (basis functions), thereby greatly improving the efficiency of solving physical fields while ensuring that the results have physical meaning. Obtaining the spectral coefficients corresponding to the basis functions under different solution conditions is the key to the POD method, and it is also the main calculation object and time-consuming source for the POD method to reconstruct the physical field. However, for complex problems, the commonly used snapshot POD method is difficult to obtain the calculation expression of the spectral coefficients, and the solution takes a certain amount of time. To overcome this problem, the existing technology establishes a neural network model of the spectral coefficients to achieve rapid and accurate reconstruction of the physical field of complex problems, making the POD method more widely used. However, for the fluid-solid coupled heat transfer problem of the cooling structure of the complex turbine guide blade with turbulent flow, the combination of conventional neural networks and POD methods has the problem of low prediction accuracy. The root cause lies in two aspects:

[0004] First, the POD reconstruction of the physical field of complex problems requires more high-energy basis functions and corresponding spectral coefficients; the spectral coefficients are the output objects of the neural network, and the boundary conditions are the inputs of the neural network; for the problem of predicting the temperature field of the turbine guide blade cooling structure, the number of boundary conditions is small, and fewer inputs are used to predict more outputs, resulting in low prediction accuracy of the neural network; second, the basis functions of the POD method are obtained by sorting from large to small according to the energy value. The more the basis functions are ranked, the greater the weight of their spectral coefficients on the prediction results. It is difficult to consider this characteristic of the POD method when training the neural network using conventional loss functions. Therefore, it is urgent to construct a neural network method that matches the characteristics of the POD method to achieve rapid reconstruction of the temperature field of the complex turbine guide blade cooling structure while ensuring accuracy. Summary of the invention

[0005] The purpose of the present invention is to provide a method for rapid prediction of the three-dimensional temperature field of a turbine guide blade cooling structure. The method constructs an ANN model in which multiple prediction results (spectral coefficients) overlap with each other, and the loss function takes into account the influence weight of the spectral coefficients on the prediction results, thereby overcoming the defect of low prediction accuracy caused by combining the existing POD method with the neural network method without considering the characteristics of the POD method.

[0006] The objective of the present invention is achieved through the following technical solutions:

[0007] A method for quickly predicting a three-dimensional temperature field of a turbine guide blade cooling structure, the method comprising:

[0008] Step 1: Set the research conditions, randomly select the training set, validation set and test set, and construct the original temperature field matrix, temperature field snapshot matrix and temperature field covariance matrix in sequence according to the training data set;

[0009] Step 2: Obtain eigenvectors and eigenvalues ​​by solving the temperature field covariance matrix, calculate the corresponding energy values ​​using the eigenvalues, and determine the number and corresponding values ​​of basis functions and spectral coefficients according to the energy value sorting;

[0010] Step 3: Determine the number of artificial neural network ANN models to be constructed according to the number of spectral coefficients and the number of input boundary conditions, and allocate the spectral coefficient objects predicted by each ANN model;

[0011] Step 4: According to the characteristics of the intrinsic orthogonal decomposition method POD, a loss function of the spectral coefficient ANN model matching the characteristics of the POD method is constructed;

[0012] Step 5: Use the trained ANN model to predict the spectral coefficients of the unknown working condition, and combine the predicted spectral coefficients with the basis function to quickly reconstruct the three-dimensional temperature field of the turbine guide blade cooling structure of the unknown working condition.

[0013] It can be seen from the technical solution provided by the present invention that the above method constructs an ANN model with multiple overlapping prediction results (spectral coefficients), and the loss function takes into account the influence weight of the spectral coefficient on the prediction result, overcoming the defect of low prediction accuracy caused by combining the existing POD method with the neural network method without considering the characteristics of the POD method, and can achieve high-precision and rapid reconstruction of the temperature field of the turbine guide blade cooling structure, providing technical support for the research on the design and optimization of new cooling structures. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0015] Figure 1 A schematic flow chart of a method for rapidly predicting a three-dimensional temperature field of a turbine guide blade cooling structure provided by an embodiment of the present invention;

[0016] Figure 2 This is a schematic diagram of the ANN model structure according to an embodiment of the present invention;

[0017] Figure 3 is a schematic diagram of a reconstructed temperature field obtained according to the method of the present invention;

[0018] Figure 4 is a schematic diagram of an actual temperature field according to the method of the present invention;

[0019] Figure 5 is a schematic diagram of a relative error field according to the method of the present invention;

[0020] Figure 6 This is a schematic diagram of the reconstructed temperature field obtained without considering the characteristics of the POD method;

[0021] Figure 7 It is a schematic diagram of the actual temperature field without considering the characteristics of the POD method;

[0022] Figure 8 Schematic diagram of the relative error field without considering the characteristics of the POD method. DETAILED DESCRIPTION

[0023] The following is a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings 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, which does not constitute a limitation of the present invention. 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.

[0024] like Figure 1 The figure is a schematic flow chart of a method for rapid prediction of a three-dimensional temperature field of a turbine guide blade cooling structure provided by an embodiment of the present invention, the method comprising:

[0025] Step 1: Set the research working conditions, randomly select the training set (M working conditions), the validation set (L working conditions) and the test set (P working conditions), and construct the original temperature field matrix, the temperature field snapshot matrix and the temperature field covariance matrix in sequence according to the training data set;

[0026] In this step, the computational fluid dynamics method CFD is used to obtain the three-dimensional temperature field (N grid nodes) of the turbine guide blade cooling structure under the training data set. The training data set contains N discrete spaces and M temperature field snapshots. These data are used to construct the original temperature field matrix F:

[0027]

[0028] f represents the one-dimensional vector transformed from the three-dimensional temperature field of the turbine guide vane cooling structure;

[0029] Solve the mean of M temperature field snapshots in N discrete spaces and obtain a mean vector of length N It is expressed as:

[0030]

[0031] Subtract this mean vector from each snapshot in the original temperature field matrix F The temperature field snapshot matrix F′ is obtained, which is expressed as:

[0032]

[0033] Then the temperature field snapshot matrix F′ is used to construct the temperature field covariance matrix R, which is expressed as:

[0034]

[0035] The superscript T is the mathematical transpose symbol.

[0036] Step 2: Obtain eigenvectors and eigenvalues ​​by solving the temperature field covariance matrix, calculate the corresponding energy values ​​using the eigenvalues, and determine the number and corresponding values ​​of basis functions and spectral coefficients according to the energy value sorting;

[0037] In this step, the eigenvalue λ and eigenvector of the temperature field covariance matrix R are solved Right now:

[0038]

[0039] Use the calculated characteristic value to get the corresponding energy value k, assuming there are n 0 eigenvalues, then:

[0040]

[0041] Sort the eigenvalues ​​from large to small, determine the eigenvalues ​​and corresponding eigenvectors whose cumulative energy contribution C exceeds 99.99%, and assume that the first n 1 The cumulative energy contribution C of the eigenvalues ​​exceeds 99.99%, then:

[0042]

[0043] Using the temperature field snapshot matrix F′ and high energy eigenvalues ​​λ and eigenvectors Calculate the basis function Φ i , expressed as:

[0044]

[0045] For each working condition, calculate the spectral coefficient a corresponding to the basis function i , expressed as:

[0046]

[0047] Step 3: Determine the number of artificial neural network ANN models to be constructed according to the number of spectral coefficients and the number of input boundary conditions, and allocate the spectral coefficient objects predicted by each ANN model;

[0048] In this step, the boundary conditions B of the three-dimensional temperature field simulation of the turbine guide blade cooling structure under each known working condition are i (including blowing ratio, inlet Reynolds number and other S boundary conditions) are quantified as input and the corresponding spectral coefficient a i Divided into n s Segments are used as outputs to construct multiple ANN models. Each spectral coefficient corresponds to a neural network, thus solving the problem of the number of spectral coefficients n due to the POD method. 1 This is usually caused by the number of boundary conditions being several times S, which leads to low prediction accuracy of the ANN model.

[0049] Among them, the spectral coefficient a i Number of segments n s The calculation method is:

[0050]

[0051] S is the number of boundary conditions; n 1 is the number of spectral coefficients;

[0052] The last predicted value of each segment is required to be the starting value of the next segment prediction. The specific output of each ANN model is:

[0053] ANN 1 :(a 1 ,a 2 ,......,a S )

[0054] ANN 2 :(a S ,a S+1 ,......,a 2S-1 ) .......

[0056]

[0057] In the formula, due to It is not necessarily divisible, so the number of spectral coefficients output by the last ANN model is less than or equal to S. There is a connection between the spectral coefficients. Using the overlap of some spectral coefficient prediction values ​​to construct the output of the neural network model can enhance the connection between different neural network models and improve the prediction accuracy of the spectral coefficients.

[0058] The above operation can avoid the reduction in accuracy caused by the number of spectral coefficients output by a single ANN model being significantly greater than the number of input boundary conditions, and at the same time, the repetition of some spectral coefficients in different ANN models can be used to enhance the connection between different neural network models.

[0059] Step 4: According to the characteristics of the intrinsic orthogonal decomposition method POD, a loss function of the spectral coefficient ANN model matching the characteristics of the POD method is constructed;

[0060] In this step, the characteristic of the proper orthogonal decomposition method POD is that the larger the eigenvalue, the greater the weight of influence on the result. The loss function calculation method of the spectral coefficient ANN model is:

[0061]

[0062] Where: a i and are the true value and predicted value of the i-th spectral coefficient respectively.

[0063] In the specific implementation, Figure 2 The figure shows the structure diagram of the ANN model according to the embodiment of the present invention. Figure 2 The artificial neural network structure shown in the figure is used to carry out artificial neural network training, in which Gaussian error linear unit (GELU) is used as the activation function:

[0064]

[0065] The Optuna optimization algorithm with automatic optimization capability was used for hyperparameter optimization. The optimized batch size, hidden layer size, number of layers, and learning rate were 11, 150, 5, and 0.008, respectively.

[0066] Step 5: Use the trained ANN model to predict the spectral coefficients of the unknown working condition, and combine the predicted spectral coefficients with the basis function to quickly reconstruct the three-dimensional temperature field of the turbine guide blade cooling structure of the unknown working condition.

[0067] In this step, for an unknown working condition, multiple trained ANN models are used to predict the basis function Φ i The corresponding spectral coefficient a i , and take the average of the spectral coefficients of repeated predictions, expressed as:

[0068]

[0069] Where: the superscript represents the number of the ANN model;

[0070] Using the basis function Φ i and the predicted spectral coefficient a i Calculate the temperature field and mean vector under the unknown working condition The difference between them is calculated and the three-dimensional temperature field of the turbine guide blade cooling structure under the unknown working condition is reconstructed, which is expressed as:

[0071]

[0072] t indicates unknown conditions.

[0073] It is worth noting that the contents not described in detail in the embodiments of the present invention belong to the prior art known to professional and technical personnel in the field.

[0074] In order to verify the effectiveness of a method for rapid prediction of the three-dimensional temperature field of a turbine guide blade cooling structure by coupling intrinsic orthogonal decomposition (POD) and artificial neural network (ANN) described in an embodiment of the present invention, a comparative test was conducted between the method described in the present invention and the POD-ANN method without considering the characteristics of the POD method. By selecting different inlet velocities and blowing ratios, a data set containing 121 working conditions was obtained, and 99 of them were randomly selected as training sets, 11 as validation sets, and 11 as test sets. The simulation results of the commercial software FLUENT were used as a reference object, and the worst prediction result in the test set was used as an example for illustration.

[0075] like Figure 3 FIG. 1 is a schematic diagram of the reconstructed temperature field obtained according to the method of the present invention, as shown in FIG. Figure 4 The actual temperature field is shown in Figure 2. Figure 5 The relative error field is shown as follows. Figure 3-Figure 5 It can be seen that the three-dimensional temperature field distribution of the turbine guide blade cooling structure predicted by the POD-ANN method disclosed in the present invention is almost consistent with the temperature field distribution calculated by the FLUENT software, and the maximum relative deviation between the two at any grid node is 2.61%, and the average deviation of all grids is only 0.048%.

[0076] like Figure 6 The figure shows the schematic diagram of the reconstructed temperature field without considering the characteristics of the POD method. Figure 7 The actual temperature field is shown in Figure 2. Figure 8 The relative error field is shown as follows. Figure 6-Figure 8 It can be seen that: under the same circumstances, the three-dimensional temperature field distribution of the turbine guide blade cooling structure of the POD-ANN method without considering the characteristics of the POD method is not much different from the temperature field distribution calculated by the FLUENT software. The maximum relative deviation between the two at any grid node is 13.2%, and the average deviation of all grids is 1.28%.

[0077] Compared with the maximum relative deviation, the calculation accuracy of the POD-ANN method described in the embodiment of the present invention is 5.06 times that of the POD-ANN method that does not consider the characteristics of the POD method; compared with the average deviation, the calculation accuracy of the POD-ANN method described in the embodiment of the present invention is 26.67 times that of the POD-ANN method that does not consider the characteristics of the POD method.

[0078] For the 11 working conditions in the test set, the average deviation of the temperature field predicted by the POD-ANN method described in the embodiment of the present invention is 0.032%, and the average deviation of the temperature field predicted by the POD-ANN method without considering the characteristics of the POD method is 1.07%, and the accuracy difference between the two is 33.44 times.

[0079] It can be seen from the above comparison results that the three-dimensional temperature field of the turbine guide blade cooling structure predicted by the POD-ANN method described in the embodiment of the present invention is more accurate than the POD-ANN method that does not consider the characteristics of the POD method, and can provide technical support for the design and optimization of new turbine guide blade cooling structures.

[0080] In summary, the method for rapid prediction of the three-dimensional temperature field of the turbine guide blade cooling structure by coupling the proper orthogonal decomposition (POD) and the artificial neural network (ANN) described in the embodiment of the present invention, namely the POD-ANN method, achieves rapid and accurate reconstruction of the temperature field of the turbine guide blade cooling structure by adopting an ANN model prediction spectral coefficient that matches the characteristics of the POD method, which is conducive to promoting the design and optimization research of new cooling structures.

[0081] In addition, a person skilled in the art can understand that all or part of the steps in the above-mentioned embodiment method can be implemented by instructing related hardware through a program, and the corresponding program can be stored in a computer-readable storage medium. The above-mentioned storage medium can be a read-only memory, a disk or an optical disk, etc.

[0082] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by any technician familiar with the technical field within the technical scope disclosed in the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims. The information disclosed in the background technology section of this article is only intended to deepen the understanding of the overall background technology of the present invention, and should not be regarded as an admission or in any form that the information constitutes prior art known to those skilled in the art.

Claims

1. A method for rapid prediction of three-dimensional temperature field of turbine guide blade cooling structure, characterized in that: The method comprises: Step 1: Set the research conditions, randomly select the training set, validation set and test set, and construct the original temperature field matrix, temperature field snapshot matrix and temperature field covariance matrix in sequence according to the training data set; Step 2: Obtain eigenvectors and eigenvalues ​​by solving the temperature field covariance matrix, calculate the corresponding energy values ​​using the eigenvalues, and determine the number and corresponding values ​​of basis functions and spectral coefficients according to the energy value sorting; Step 3: Determine the number of artificial neural network ANN models to be constructed according to the number of spectral coefficients and the number of input boundary conditions, and allocate the spectral coefficient objects predicted by each ANN model; Step 4: According to the characteristics of the intrinsic orthogonal decomposition method POD, a loss function of the spectral coefficient ANN model matching the characteristics of the POD method is constructed; Step 5: Use the trained ANN model to predict the spectral coefficients of the unknown working condition, and combine the predicted spectral coefficients with the basis function to quickly reconstruct the three-dimensional temperature field of the turbine guide blade cooling structure of the unknown working condition.

2. The method for rapid prediction of three-dimensional temperature field of turbine guide blade cooling structure according to claim 1 is characterized in that: In step 1, the computational fluid dynamics method CFD is used to obtain the three-dimensional temperature field of the turbine guide blade cooling structure under the training data set. The training data set contains N discrete spaces and M temperature field snapshots. These data are used to construct the original temperature field matrix F: f represents the one-dimensional vector transformed from the three-dimensional temperature field of the turbine guide vane cooling structure; Solve the mean of M temperature field snapshots in N discrete spaces to obtain a mean vector of length N It is expressed as: Subtract this mean vector from each snapshot in the original temperature field matrix F The temperature field snapshot matrix F′ is obtained, which is expressed as: Then the temperature field snapshot matrix F′ is used to construct the temperature field covariance matrix R, which is expressed as: The superscript T is the mathematical transpose symbol.

3. The method for rapid prediction of three-dimensional temperature field of turbine guide blade cooling structure according to claim 2 is characterized in that: In step 2, solve the eigenvalue λ and eigenvector of the temperature field covariance matrix R Right now: Use the calculated eigenvalues ​​to get the corresponding energy value k. Assuming there are n0 eigenvalues, we have: Sort the eigenvalues ​​from large to small, determine the eigenvalues ​​whose cumulative energy contribution C exceeds 99.99% and the corresponding eigenvectors. Assuming that the cumulative energy contribution C of the first n1 eigenvalues ​​exceeds 99.99%, then: Using the temperature field snapshot matrix F′ and high energy eigenvalues ​​λ and eigenvectors Calculate the basis function Φ i , expressed as: For each working condition, calculate the spectral coefficient a corresponding to the basis function i , expressed as:

4. The method for rapid prediction of three-dimensional temperature field of turbine guide blade cooling structure according to claim 3 is characterized in that: In step 3, the boundary condition B of the three-dimensional temperature field simulation of the turbine guide blade cooling structure under each known working condition is i After quantization, the corresponding spectral coefficient a is used as input i Divided into n s The segment is used as the output to build multiple ANN models, and each spectral coefficient corresponds to a neural network; Among them, the spectral coefficient a i Number of segments n s The calculation method is: S is the number of boundary conditions; n1 is the number of spectral coefficients; The last predicted value of each segment is required to be the starting value of the next segment prediction. The specific output of each ANN model is: In the formula, due to It is not necessarily divisible, so the number of spectral coefficients output by the last ANN model is less than or equal to S.

5. The method for rapid prediction of three-dimensional temperature field of turbine guide blade cooling structure according to claim 4 is characterized in that: In step 4, the loss function of the spectral coefficient ANN model is calculated as: Where: a i and are the true value and predicted value of the i-th spectral coefficient respectively.

6. The method for rapid prediction of three-dimensional temperature field of turbine guide blade cooling structure according to claim 5 is characterized in that: In step 5, for an unknown working condition, multiple trained ANN models are used to predict the basis function Φ i The corresponding spectral coefficient a i , and take the average of the spectral coefficients of repeated predictions, expressed as: Where: the superscript represents the number of the ANN model; Using the basis function Φ i and the predicted spectral coefficient a i Calculate the temperature field and mean vector under the unknown working condition The difference between them is calculated and the three-dimensional temperature field of the turbine guide blade cooling structure under the unknown working condition is reconstructed, which is expressed as: t indicates unknown conditions.

Images