Turbine blade cooling structure parameter recommendation method based on generative model

Abstract

The invention discloses a turbine blade cooling structure parameter recommendation method based on a generative model, and belongs to the technical field of aero-engine turbine blade cooling design, and the method comprises the steps: carrying out the separation training of a decoder and an encoder in the generative model, and enabling the feature parameters generated by the encoder to belong to a cooling structure parameter space, the physical significance of the generated parameters is ensured; the collaborative recommendation of multiple design targets can be realized through an innovative model architecture that one encoder is linked with a plurality of decoders and a training method that a loss function is associated with a plurality of design targets. According to the method, direct output of the cooling structure parameters for the given design indexes under different working conditions is achieved through the generative model, and the purpose of improving the turbine blade cooling design efficiency is achieved; and data in each design activity is incorporated into a training data set, so that the purpose of continuously improving the model recommendation capability is achieved.

Description

Technical Field

[0001] The invention belongs to the technical field of aero-engine turbine blade cooling design, and in particular relates to a turbine blade cooling structure parameter recommendation method based on a generative model. Background Art

[0002] In the current cooling design process of aircraft engine turbine blades, the initial cooling structure parameters are directly given by manual experience, and then the cooling structure parameters are optimized using an iterative optimization algorithm to achieve the design indicators.

[0003] In the face of more advanced aircraft engines in the future, turbine blades will have more severe working conditions, more demanding design indicators, and more sophisticated structural layouts. The current "empirical" initial design method is likely to seriously deviate from the feasible design area, increasing the time and difficulty of subsequent "iterative" optimization, and each design activity is independent of each other, resulting in the inability to reuse historical design results and data. Therefore, in order to improve the efficiency and design capabilities of my country's aircraft engine turbine blade cooling design, it is urgent to develop an efficient and continuously evolving turbine blade cooling structure parameter recommendation method. Summary of the invention

[0004] In order to solve the above technical problems, the present invention proposes a method for recommending turbine blade cooling structure parameters based on a generative model to solve the problems existing in the above-mentioned prior art.

[0005] To achieve the above object, the present invention provides a method for recommending turbine blade cooling structure parameters based on a generative model, comprising:

[0006] Construct a training data set of turbine blade cooling design indicators corresponding to different operating conditions and structural parameters based on numerical simulation methods;

[0007] Building a plurality of decoder models based on a neural network architecture, training the plurality of decoder models using the training data set, and generating a plurality of trained decoder models;

[0008] Building an encoder model based on a neural network architecture, training the encoder model using the training data set, and generating a trained encoder model;

[0009] Connecting the output of the trained encoder model with the input of the trained multiple decoder models to generate a turbine blade cooling structure parameter recommendation model;

[0010] Based on the turbine blade cooling structure parameter recommendation model, cooling structure parameters of the design target under current operating conditions are generated.

[0011] Preferably, the operating conditions include but are not limited to: mainstream Reynolds number, turbulence, density ratio, blowing ratio and pressure difference between inside and outside of the blade;

[0012] The structural parameters include but are not limited to: impact hole diameter, film hole diameter, film hole incident angle, aspect ratio, hole spacing and row spacing.

[0013] Preferably, the process of constructing multiple decoder models based on the neural network architecture includes: constructing multiple decoder models from operating conditions and structural parameters to cold air flow and blade comprehensive cooling efficiency design indicators based on the neural network architecture.

[0014] Preferably, the expression of the decoder model is:

[0015] Y=G(X)=G(Re,Tu,DR,M,ΔP,d i ,d f ,α,l / d f ,P / d f ,S / d f ,...);

[0016] Among them, Re is the mainstream Reynolds number, Re is the mainstream Reynolds number, Tu is the turbulence, DR is the density ratio, M is the blowing ratio, ΔP is the pressure difference between the inside and outside of the blade; di is the impact hole diameter, df is the film hole diameter, α is the film hole incident angle, l / df is the aspect ratio, P / df is the hole spacing, S / df is the row spacing, G(X) is the constructed decoder model, and Y is the design index.

[0017] Preferably, the operating conditions and structural parameters of the encoder, including cooling air flow rate and blade comprehensive cooling efficiency design indicators, are the same as those of the decoder.

[0018] Preferably, the loss function of the encoder is expressed as:

[0019]

[0020] Among them, YR is the cooling air flow rate, blade comprehensive cooling efficiency and other performance output by the decoder, Y is the design index, and K represents the dimension of the input data.

[0021] Preferably, the process of connecting the output of the trained encoder model with the input of the trained multiple decoder models comprises:

[0022] The trained encoder output is used as a hub to connect multiple decoder models in parallel, and a training method for multiple design objectives is associated through a loss function to form a unified paradigm for collaborative recommendation of multiple design objectives.

[0023] Preferably, the unified paradigm of collaborative recommendation of multiple design objectives is:

[0024]

[0025] Where k is the total number of decoders, ni is the dimension of the design indicator in the i-th decoder, is the tensor format of the predicted values ​​of the design indicators for all dimensions in the i-th decoder, Y i The tensor format of the design indicator target values ​​for all dimensions in the i-th decoder. is the predicted value of the design indicator of the jth dimension in the i-th decoder, y ij is the design indicator target value of the j-th dimension in the ith decoder.

[0026] Compared with the prior art, the present invention has the following advantages and technical effects:

[0027] The present invention innovatively proposes a generative artificial intelligence model with physical meaning and encoder-decoder separation training for the problem of turbine blade cooling structure parameter recommendation;

[0028] The present invention innovatively proposes a model architecture of an encoder linking multiple decoders and a training method of associating multiple design objectives through a loss function for the problem of collaborative recommendation of multiple design objectives for turbine blades. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] 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:

[0030] Figure 1 This is a flow chart of a method for recommending turbine blade cooling structure parameters based on a generative model according to an embodiment of the present invention. DETAILED DESCRIPTION

[0031] 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.

[0032] 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.

[0033] Embodiment 1

[0034] like Figure 1 As shown, this embodiment provides a method for recommending turbine blade cooling structure parameters based on a generative model, including:

[0035] Construct a training data set of turbine blade cooling design indicators corresponding to different operating conditions and structural parameters based on numerical simulation methods;

[0036] Building a plurality of decoder models based on a neural network architecture, training the plurality of decoder models using the training data set, and generating a plurality of trained decoder models;

[0037] Building an encoder model based on a neural network architecture, training the encoder model using the training data set, and generating a trained encoder model;

[0038] Connecting the output of the trained encoder model with the input of the trained multiple decoder models to generate a turbine blade cooling structure parameter recommendation model;

[0039] Based on the turbine blade cooling structure parameter recommendation model, the cooling structure parameters of the design target under the current working conditions are generated, specifically:

[0040] Step 1: Construction of training dataset

[0041] Based on the existing numerical simulation methods, a training data set of turbine blade cooling design indicators such as cooling air flow and blade comprehensive cooling efficiency corresponding to different operating conditions and structural parameters is constructed;

[0042] Step 2: Decoder model construction and training

[0043] Based on the neural network architecture, a decoder model is constructed from operating conditions and structural parameters to design indicators such as cooling air flow and blade comprehensive cooling effect, and the decoder model is trained based on the data set constructed in step ①. The operating parameters include parameters commonly used in engineering practice, such as mainstream Reynolds number (Re), turbulence (Tu), density ratio (DR), blowing ratio (M), blade internal and external pressure difference (ΔP), etc.; geometric parameters include impact hole diameter (d i ), air film hole diameter (d f ), film hole incident angle (α), aspect ratio (l / d f ), hole spacing (P / d f ), row spacing (S / d f ) etc. The expansion of the data set can add more parameters, and this application does not limit the type and number of parameters. The constructed decoder model G(X) is shown in formula (1).

[0044] Y=G(X)=G(Re,Tu,DR,M,ΔP,d i ,d f ,α,l / df,P / df,S / df,...)(1)

[0045] Step 3: Encoder model construction and training

[0046] Based on the neural network architecture, an encoder model is constructed from operating conditions and design indicators such as cold air flow, comprehensive cooling efficiency of blades to structural parameters, and the output of the encoder model is linked to the input of the decoder model trained in step ②, and the dataset constructed in step ① is called to train the encoder model. The operating parameters, structural parameters, design indicators and other parameters of the encoder model are completely consistent with the decoder model trained in step 2. The constructed encoder model F(Y) is shown in formula (2), and its function is to compress the high-dimensional parameter space where the design indicators Y such as cold air flow and comprehensive cooling efficiency of blades are located into the low-dimensional structural parameter space X;

[0047] X=F(Y)(2)

[0048] In order to achieve the design goal as much as possible, the training goal of the encoder is to minimize the cooling air flow output by the decoder, the comprehensive cooling effect of the blade, and other performance Y R The difference between the target performance Y of the encoder input is generally calculated by the mean square error between the two, as shown in formula (3).

[0049]

[0050] Where K represents the dimension of the input data. After the training is completed, the output of the encoder is the cooling structure parameters of the design target under the current working conditions.

[0051] For the design of turbine blade cooling structure, it is generally required to meet the design indicators of multiple disciplines such as cooling air flow, wall temperature, strength and life at the same time. However, traditional autoencoders usually have one encoder corresponding to one decoder, lacking a model architecture and training method for multiple decoders at the same time. In response to this, this application proposes a model architecture that uses the encoder output as a hub to connect multiple decoders in parallel and a training method that associates multiple design goals through a loss function, forming a unified paradigm for collaborative recommendation of multiple design goals, as shown in formula (4).

[0052]

[0053] Where k is the total number of decoders, ni is the dimension of the design indicator in the i-th decoder, is the tensor format of the predicted values ​​of the design indicators for all dimensions in the i-th decoder, Y i The tensor format of the design indicator target values ​​for all dimensions in the i-th decoder. is the predicted value of the design indicator of the jth dimension in the i-th decoder, y ij is the target value of the design indicator for the jth dimension in the i-th decoder. The values ​​of each item in the loss function are reduced during the training process of the entire autoencoder model, thereby achieving the association and satisfaction of multiple design goals.

[0054] Step 4: Cooling structure parameter recommendation

[0055] During the application phase, based on the input operating conditions and design indicators such as target cooling air flow rate and blade comprehensive cooling efficiency, the encoder model trained in step ③ is called to directly output the corresponding cooling structure parameters to complete the recommendation.

[0056] Step 5: Collection and reuse of historical design activity data

[0057] Incorporate the data from each design activity into the training dataset built in step one, thereby continuously improving the capabilities of the decoder model and encoder model built in steps two and three.

[0058] 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 recommending turbine blade cooling structure parameters based on a generative model, characterized in that: The following steps are involved: Construct a training data set of turbine blade cooling design indicators corresponding to different operating conditions and structural parameters based on numerical simulation methods; Building a plurality of decoder models based on a neural network architecture, training the plurality of decoder models using the training data set, and generating a plurality of trained decoder models; Building an encoder model based on a neural network architecture, training the encoder model using the training data set, and generating a trained encoder model; Connecting the output of the trained encoder model with the input of the trained multiple decoder models to generate a turbine blade cooling structure parameter recommendation model; Based on the turbine blade cooling structure parameter recommendation model, cooling structure parameters of the design target under current operating conditions are generated.

2. The method for recommending turbine blade cooling structure parameters based on a generative model according to claim 1, characterized in that: The operating conditions include, but are not limited to: mainstream Reynolds number, turbulence, density ratio, blowing ratio and pressure difference between inside and outside of the blade; The structural parameters include but are not limited to: impact hole diameter, film hole diameter, film hole incident angle, aspect ratio, hole spacing and row spacing.

3. The method for recommending turbine blade cooling structure parameters based on a generative model according to claim 1, characterized in that: The process of constructing multiple decoder models based on the neural network architecture includes: constructing multiple decoder models from operating conditions and structural parameters to cold air flow and blade comprehensive cooling efficiency design indicators based on the neural network architecture.

4. The method for recommending turbine blade cooling structure parameters based on a generative model according to claim 1, characterized in that: The expression of the decoder model is: Y=G(X)=G(Re,Tu,DR,M,ΔP,d i ,d f ,α,l / d f ,P / d f ,S / d f ,...); Where Re is the mainstream Reynolds number, Re is the mainstream Reynolds number, Tu is the turbulence, DR is the density ratio, M is the blowing ratio, and ΔP is the pressure difference between the inside and outside of the blade; d i is the impact hole diameter, d f is the diameter of the air film hole, α is the incident angle of the air film hole, l / d f is the aspect ratio, P / d f is the hole spacing, S / d f is the row spacing, G(X) is the constructed decoder model, and Y is the design index.

5. The method for recommending turbine blade cooling structure parameters based on a generative model according to claim 1, characterized in that: The working conditions and structural parameters of the encoder, as well as the cooling air flow rate and the blade comprehensive cooling efficiency design indicators are the same as those of the decoder.

6. The method for recommending turbine blade cooling structure parameters based on a generative model according to claim 1, characterized in that: The expression of the loss function of the encoder is: Among them, Y R is the cooling air flow rate, blade comprehensive cooling efficiency and other performance output by the decoder, Y is the design index, and K represents the dimension of the input data.

7. The method for recommending turbine blade cooling structure parameters based on a generative model according to claim 1, characterized in that: The process of connecting the output of the trained encoder model with the input of the trained multiple decoder models includes: The trained encoder output is used as a hub to connect multiple decoder models in parallel, and a training method for multiple design objectives is associated through a loss function to form a unified paradigm for collaborative recommendation of multiple design objectives.

8. The method for recommending turbine blade cooling structure parameters based on a generative model according to claim 7, characterized in that: The unified paradigm of multi-design objective collaborative recommendation is: Among them, k is the total number of decoders, n i is the dimension of the design indicator in the i-th decoder, is the tensor format of the predicted values ​​of the design indicators for all dimensions in the i-th decoder, Y i is the tensor format of the design indicator target value for all dimensions in the i-th decoder, is the predicted value of the design indicator of the jth dimension in the i-th decoder, y ij is the design indicator target value of the j-th dimension in the ith decoder.

Images