Prediction method for multi-exhaust film cooling effect distribution of turbine blade

Abstract

The invention discloses a prediction method for turbine blade multi-exhaust membrane cooling efficiency distribution, which comprises the following steps: constructing a training data set based on a numerical simulation method, constructing a single-exhaust membrane cooling efficiency neural network and a multi-exhaust membrane cooling efficiency superposition principle correction coefficient neural network, and training through the training data set; respectively inputting the target parameters into the trained single-row membrane cooling effect neural network and the multi-row membrane cooling effect superposition principle correction coefficient neural network to obtain single-row membrane cooling effect distribution and standard superposition principle offset; and obtaining multi-row gas film cooling effect distribution based on the single-row gas film cooling effect distribution and the standard superposition principle offset. According to the method, the normal distribution is introduced on the basis of the conventional full-connection neural network to restrain the transverse distribution of the single-row gas film cooling efficiency, so that the prediction precision of the single-row gas film cooling efficiency distribution is improved; and based on the single-exhaust membrane cooling effect neural network, rapid and accurate prediction of multi-exhaust membrane cooling effect distribution is finally realized through correction standard superposition principle calculation.

Description

Technical Field

[0001] The invention belongs to the technical field of cooling of turbine blades of aircraft engines, and in particular relates to a method for predicting the distribution of multi-exhaust film cooling effects of turbine blades. Background Art

[0002] Increasing the temperature of the turbine inlet is the most important means to improve the thrust and other performance of aircraft engines, but it also poses a severe challenge to the thermal protection of turbine blades. Film cooling is an extremely effective cooling technology that discharges low-temperature airflow through discrete holes or slots to form an air film on the outer surface of the blade to isolate the direct heat exchange between the high-temperature combustion gas and the blade, thereby reducing the thermal load of the blade. It has been widely used in many types of turbine blades at home and abroad. At present, the evaluation of the film cooling effect distribution of turbine blades mainly relies on three-dimensional numerical simulation methods, which has the problem of long calculation cycle, resulting in low efficiency of turbine blade design. Therefore, it is urgent to develop a fast and accurate method for predicting the distribution of film cooling effect. Summary of the invention

[0003] In order to solve the above technical problems, the present invention proposes a prediction method for the distribution of multi-exhaust film cooling effect of turbine blades to solve the problems existing in the above prior art.

[0004] To achieve the above object, the present invention provides a method for predicting the distribution of multi-exhaust film cooling effect of turbine blades, comprising:

[0005] Constructing a training data set based on a numerical simulation method, constructing a single exhaust film cooling effect neural network and a multi-exhaust film cooling effect superposition principle correction coefficient neural network, and training with the training data set;

[0006] The target parameters are respectively input into the trained single exhaust film cooling effect neural network and multi-exhaust film cooling effect superposition principle correction coefficient neural network to obtain the single exhaust film cooling effect distribution and standard superposition principle offset;

[0007] The cooling effect distribution of multiple exhaust films is obtained based on the single exhaust film cooling effect distribution and the standard superposition principle offset.

[0008] Optionally, the training data set includes a single exhaust film cooling effect training data set and a multi-exhaust film cooling effect training data set.

[0009] Optionally, the expression of the single exhaust film cooling effect neural network is as follows:

[0010]

[0011] In the formula, is the single exhaust film cooling effect, Tu is the turbulence, DR is the density ratio, M is the blowing ratio; the air film hole α is the incident angle, l / df is the aspect ratio; x / df is the dimensionless flow direction coordinate point, and y / df is the transverse dimensionless coordinate vector at different flow direction coordinates.

[0012] Optionally, the process of training the single exhaust film cooling effect neural network using the training data set includes:

[0013] A single exhaust film cooling effect training data set is selected from the training data set, a normal distribution constraint is introduced into the loss function, and a single exhaust film cooling effect neural network is trained based on the single exhaust film cooling effect training data set and the loss function.

[0014] Optionally, the loss function includes two terms, one is the mean square error of all lateral distributed adiabatic cooling effect prediction values ​​and label values ​​at the current flow direction distance; the other is the mean square error between the lateral cooling effect distribution prediction value at the current flow direction distance and the normal distribution label value obtained by fitting the preprocessed training data.

[0015] Optionally, the correction coefficient neural network of the multi-exhaust film cooling effect superposition principle is a fully connected neural network, which takes the air film cooling effect when a single exhaust film exists alone as input and the offset of the standard superposition principle as output.

[0016] Optionally, the multi-exhaust film cooling effect training data set includes the difference between the multi-exhaust film adiabatic cooling effect distribution obtained by simulation and the double-exhaust film adiabatic cooling effect distribution calculated by the standard superposition principle.

[0017] Compared with the prior art, the present invention has the following advantages and technical effects:

[0018] The present invention realizes rapid prediction from operating parameters and geometric parameters to air film cooling effect distribution by adopting a fully connected neural network; and, on the basis of a conventional fully connected neural network, introduces a normal distribution to constrain the lateral distribution of a single exhaust film cooling effect, thereby improving the prediction accuracy of the single exhaust film cooling effect distribution; based on the single exhaust film cooling effect neural network, the rapid and accurate prediction of the multi-exhaust film cooling effect distribution is finally realized by modifying the standard superposition principle calculation. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0020] Figure 1 This is an overall framework diagram of a method for predicting the distribution of multi-exhaust film cooling effects on turbine blades according to an embodiment of the present invention. DETAILED DESCRIPTION

[0021] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0022] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0023] Embodiment 1

[0024] like Figure 1 As shown, this embodiment provides a method for predicting the distribution of multi-exhaust film cooling effect of turbine blades, including:

[0025] Step 1: Construction of training dataset

[0026] Based on the existing numerical simulation methods, a single exhaust film cooling effect distribution training dataset and a multi-exhaust film cooling effect distribution training dataset for turbine blades covering the range of commonly used operating parameters and geometric parameters in engineering are constructed. The operating parameters of all examples in the dataset include turbulence (Tu), density ratio (DR) and blowing ratio (M); the geometric parameters include film hole incident angle (α), aspect ratio (l / d f ), hole spacing (P / d f ) and row spacing (S / d f ). The parameter ranges of the single-row and multi-row exhaust film cooling effect data sets are shown in Tables 1 and 2, respectively.

[0027] Table 1

[0028]

[0029] Table 2

[0030]

[0031] The definition of air film cooling effect is shown in formula (1):

[0032]

[0033] Where, T ∞ and T c are the inlet temperatures of the mainstream and cold air, T aw is the adiabatic temperature of the wall when film cooling exists.

[0034] Step 2: Single exhaust film cooling effect neural network construction;

[0035] A fully connected neural network is constructed to predict the cooling efficiency distribution of a single exhaust film. The input of the neural network is the operating parameters, geometric parameters and the full-field coordinates (x / df , y / d f ), the output is the air film cooling effect value of each coordinate point in the whole field, as shown in formula (2):

[0036]

[0037] In the formula, x / d f is the dimensionless flow direction coordinate point, y / d f is the dimensionless transverse coordinate vector at different streamwise coordinates.

[0038] Step 3: Single exhaust film cooling effect neural network training under normal distribution constraints;

[0039] A normal distribution constraint is added to the loss function of the fully connected neural network of the single-exhaust film cooling effect, and the neural network is trained based on the single-exhaust film cooling effect distribution training data set constructed in step 1. For a single-exhaust film, it is physically known that the adiabatic cooling effect of the air film at different flow positions mostly presents a Gaussian distribution along the lateral direction. Therefore, the normal distribution can be used to fit the distribution of the adiabatic cooling effect of the air film at different flow positions in the training sample along the lateral direction, as shown in formula (3):

[0040]

[0041] Where a, b and c are normal distribution parameters obtained by fitting at different flow distances, y / d f is the dimensionless lateral distance of the streamwise position, y / d f = 0 means it is located on the center line of the air film hole. The vector y / d formed by the dimensionless distance input by the neural network each time f Substitute the current flow distance x / d obtained by fitting the preprocessed training samples f In the normal distribution of the lateral cooling effect, the lateral cooling effect distribution at different flow distances is calculated, as shown in formula (4):

[0042]

[0043] The loss function used for model training consists of two terms. One is the mean square error between all the predicted values ​​of the adiabatic cooling effect in the current flow direction and the label value; the other is the mean square error between the predicted value of the lateral cooling effect distribution in the current flow direction and the label value of the normal distribution obtained by fitting the preprocessed training data, as shown in equations (5) to (7):

[0044] Loss=Loss d +Loss G (5)

[0045]

[0046] Where n is the number of lateral position points at each flow distance.

[0047] Step 4: Construction of the offset neural network based on the superposition principle of the cooling efficiency standard of multiple exhaust films;

[0048] A fully connected neural network is constructed to predict the offset of the standard superposition principle of multi-exhaust film cooling effect. The input of the neural network is the film cooling effect when a single exhaust film exists alone, and the output is the offset of the standard superposition principle. The standard superposition principle of film cooling effect assumes that there is no flow interaction between the upstream and downstream air films, so that the adiabatic wall temperature formed by the film cooling of the previous row (i-1 row) can be regarded as the incoming flow temperature of the film holes in this row (i-th row). Such an assumption is relatively reasonable for the superposition of the adiabatic cooling effect of the slot-type air film with good lateral spreading and relatively gentle flow. However, for the circular holes commonly used in aircraft engine turbine blades, the lateral spreading of the cold air is limited, and generally has strong three-dimensional flow characteristics, so it is difficult to meet the assumptions required by the standard superposition principle. Therefore, directly applying the standard superposition principle will lead to higher errors. In this regard, the present application uses a neural network to predict the offset of the standard superposition principle under different conditions. The core of the application is to correct the assumption of the standard superposition principle that "the adiabatic wall temperature formed by the film cooling of the previous row (i-1 row) is regarded as the incoming flow temperature of the film holes in this row (i-th row). Taking the superposition of two rows of holes as an example, the adiabatic cooling effect when the first exhaust film hole exists alone is still:

[0049]

[0050] Assume that the second exhaust film will affect the cold air coverage of the first exhaust film f(X), so that the incoming flow temperature is no longer the adiabatic wall temperature under the first exhaust film. Rather At this time, the adiabatic cooling effect of the second exhaust film hole is:

[0051]

[0052] Therefore, the adiabatic wall temperature when two exhaust films exist can be obtained as:

[0053]

[0054] Substituting into formula (8), we can get:

[0055]

[0056] Substituting formula (8) into the equation, we can get:

[0057]

[0058] It can be seen from formula (12) that the modified standard superposition principle is to add a correction term C = f(X)(1-η2) / (T∞ -T c ), this correction term is related to the impact of the second exhaust membrane on the coverage of the first exhaust membrane, the cooling effect of the second exhaust membrane itself, and the mainstream and cold air temperatures. Physically, the impact of the second row on the first row is determined by the interaction between the two rows of cold air and the mainstream, and this interaction can be characterized by the cooling effect of a single exhaust membrane when it exists alone. Therefore, the correction term can be characterized by the cooling effect of the two exhaust membranes themselves and the mainstream and cold air temperatures. Therefore, the input layer of the neural network is the adiabatic cooling effect of the two exhaust membranes when they exist alone under the current working conditions and structural conditions, and the output layer is the correction term superimposed on the adiabatic cooling effect of the two exhaust membranes at this time.

[0059] Step 5: Multi-exhaust film cooling efficiency standard superposition principle offset neural network training;

[0060] The standard superposition principle offset neural network is trained based on the single-row and multi-row exhaust film cooling efficiency distribution training data sets constructed in step 1. The training data is the multi-row exhaust film adiabatic cooling efficiency distribution η obtained by simulation and the double exhaust film adiabatic cooling efficiency distribution η calculated by the standard superposition principle. s The difference between them is C.

[0061] C=η-η S (13)

[0062] The loss function of the model is the mean square error between the correction term prediction value and the label value:

[0063]

[0064] Where N is the number of points in the field.

[0065] Step 6: Prediction of single exhaust film cooling efficiency distribution;

[0066] Given the target operating parameters, geometric parameters and full-field coordinates, the fully connected neural network trained in step 3 is called to predict the single exhaust film cooling efficiency distribution η i .

[0067] Step 7: Standard superposition principle offset prediction;

[0068] The fully connected neural network trained in step 5 is called to predict the standard superposition principle offset C under the working parameters and geometric parameters.

[0069] Step 8: Calculation of cooling efficiency distribution of multiple exhaust films;

[0070] Based on the single exhaust film cooling efficiency distribution predicted in step six and the standard superposition principle offset predicted in step seven, the multi-exhaust film cooling efficiency distribution η under the operating parameters and geometric parameters is calculated according to formula (12).

[0071] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for predicting the distribution of multi-exhaust film cooling effect of turbine blades, characterized in that: The following steps are involved: Constructing a training data set based on a numerical simulation method, constructing a single exhaust film cooling effect neural network and a multi-exhaust film cooling effect superposition principle correction coefficient neural network, and training with the training data set; The target parameters are respectively input into the trained single exhaust film cooling effect neural network and multi-exhaust film cooling effect superposition principle correction coefficient neural network to obtain the single exhaust film cooling effect distribution and standard superposition principle offset; The cooling effect distribution of multiple exhaust films is obtained based on the single exhaust film cooling effect distribution and the standard superposition principle offset.

2. The method for predicting the distribution of multi-exhaust film cooling effect of turbine blades according to claim 1, characterized in that: The training data set includes a single exhaust film cooling effect training data set and a multi-exhaust film cooling effect training data set.

3. The method for predicting the distribution of multi-exhaust film cooling effect of turbine blades according to claim 2, characterized in that: The expression of the single exhaust film cooling effect neural network is as follows: In the formula, is the single row film cooling effect, single row Tu is the turbulence, DR is the density ratio, M is the blowing ratio; the film hole α is the incident angle, l / d f is the aspect ratio; x / d f is the dimensionless flow direction coordinate point, y / d f is the dimensionless transverse coordinate vector at different streamwise coordinates.

4. The method for predicting the distribution of multi-exhaust film cooling effect of turbine blades according to claim 2, characterized in that: The process of training the single exhaust film cooling effect neural network by using the training data set includes: A single exhaust film cooling effect training data set is selected from the training data set, a normal distribution constraint is introduced into the loss function, and a single exhaust film cooling effect neural network is trained based on the single exhaust film cooling effect training data set and the loss function.

5. The method for predicting the distribution of multi-exhaust film cooling effect of turbine blades according to claim 4, characterized in that: The loss function includes two items, one is the mean square error between all lateral distribution adiabatic cooling effect prediction values ​​and label values ​​at the current flow direction distance; the other is the mean square error between the lateral cooling effect distribution prediction value at the current flow direction distance and the normal distribution label value obtained by fitting the preprocessed training data.

6. The method for predicting the distribution of multi-exhaust film cooling effect of turbine blades according to claim 1, characterized in that: The correction coefficient neural network of the multi-exhaust film cooling effect superposition principle is a fully connected neural network, which takes the air film cooling effect when a single exhaust film exists alone as input and the offset of the standard superposition principle as output.

7. The method for predicting the distribution of multi-exhaust film cooling effect of turbine blades according to claim 2, characterized in that: The multi-exhaust film cooling efficiency training data set includes the difference between the multi-exhaust film adiabatic cooling efficiency distribution obtained by simulation and the double-exhaust film adiabatic cooling efficiency distribution calculated by the standard superposition principle.

Images