Transform-based aero-engine turbine flow field prediction method

Through the Transformer-based data-driven method, the multi-head self-attention mechanism and CFD solver are used to solve the complexity and time-consuming problems of turbine flow field calculation of aero engines, and efficient and accurate flow field data prediction is achieved.

CN118981832APending Publication Date: 2024-11-19HARBIN INST OF TECH
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202410945403.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-15
Publication Date
2024-11-19

AI Technical Summary

Technical Problem

In the prior art, the calculation of the turbine flow field of the aircraft engine is complex and time-consuming, making it difficult to achieve accurate and efficient prediction.

Method used

Using a data-driven method based on Transformer, a feature encoding is generated by the characteristic sampling and data processing of turbine blades is generated, and a Transformer model is constructed, and a multi-head self-attention mechanism is used to predict flow field data. Combined with the CFD solver to calculate the flow field database to generate flow field Dataset and train the model.

Benefits of technology

It improves the efficiency and accuracy of turbine flow field data prediction, reduces computing resources and time costs, and achieves fast and efficient flow field data acquisition.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118981832A_ABST
    Figure CN118981832A_ABST
Patent Text Reader

Abstract

The invention provides an aero-engine turbine flow field prediction method based on Transform, and the method comprises the steps: carrying out the feature sampling of a guide blade and a moving blade of an aero-engine turbine, constructing a feature code according to the collected feature parameters, and processing the feature code to generate a blade model feature code database; calculating a flow field database of the turbine according to the turbine blade model library, generating a flow field data set according to the flow field database, and constructing a model training data set based on the two generated databases; an encoder and a decoder of a Transform model are constructed, and the Transform model is trained based on the training data set; and obtaining a feature code of a to-be-predicted aero-engine turbine blade, inputting the feature code into the trained Transform model, and obtaining a flow field prediction result output by the model. According to the method, the flow field data of the aero-engine turbine is obtained based on the data-driven Transform model, and the efficiency and accuracy of turbine flow field prediction are improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the technical field of fluid prediction, and in particular to a Transformer-based prediction method for an aircraft engine turbine flow field. Background Art

[0002] At present, in the field of mechanical design of aircraft engine turbines, accurate and rapid calculation of turbine flow field is an important part of realizing aircraft engine design. Efficient acquisition of turbine flow field data can effectively reduce the design cycle of aircraft engines.

[0003] In the related art, when predicting the flow field data of the aircraft engine turbine, the traditional computational fluid dynamics (CFD) calculation method is generally used. However, in practical applications, only the CFD calculation simulation method is used for prediction, which is time-consuming and the calculation process is complicated.

[0004] Therefore, how to predict aircraft engine turbine flow field data more accurately and efficiently has become a problem that needs to be solved urgently. Summary of the invention

[0005] The present application aims to solve one of the technical problems in the related art at least to some extent.

[0006] To this end, the first purpose of this application is to propose a Transformer-based prediction method for aircraft engine turbine flow field. This method obtains the flow field data of the aircraft engine turbine based on a data-driven Transformer model, thereby improving the efficiency and accuracy of turbine flow field prediction.

[0007] The second object of this application is to propose a prediction system for aircraft engine turbine flow field based on Transformer;

[0008] The third object of the present application is to provide an electronic device;

[0009] A fourth objective of the present application is to provide a computer-readable storage medium.

[0010] To achieve the above-mentioned purpose, the first aspect of the present application is to propose a prediction method of an aircraft engine turbine flow field based on Transformer, the method comprising the following steps:

[0011] Performing feature sampling on guide blades and moving blades of an aircraft engine turbine, constructing feature codes according to the collected feature parameters, and processing the feature codes to generate a blade model feature code database;

[0012] Calculate the turbine flow field database according to the turbine blade model library obtained by feature sampling, generate a flow field data set according to the flow field database, and construct a model training data set based on the two generated databases, wherein the flow field data set is the target of Transformer model training;

[0013] Constructing an encoder and a decoder of a Transformer model, wherein the encoder is used to perform multi-head self-attention calculations and train the Transformer model based on the training data set;

[0014] The feature code of the current aircraft engine turbine blade to be predicted is obtained, and the current feature code is input into the trained Transformer model to obtain the flow field prediction result output by the model.

[0015] Optionally, according to one embodiment of the present application, the characteristic parameters include multiple geometric characteristic parameters of each turbine blade and multiple control point coordinates on the transverse section of the blade, and the multiple geometric characteristic parameters include: air inlet geometry angle, effective air outlet angle, upper wedge angle, outlet deflection angle, center connection angle, shaft diameter and pitch.

[0016] Optionally, according to one embodiment of the present application, the processing of the feature code to generate a blade model feature code database includes: performing data fitting on the feature code by the least squares method; sampling and screening the fitted data based on preset blade model feature parameters to construct two-dimensional aircraft engine turbine blade model data; sampling and processing the two-dimensional aircraft engine turbine blade model data by a Latin hypercube sampling algorithm to obtain the blade model feature code database.

[0017] Optionally, according to one embodiment of the present application, the turbine blade model library obtained according to feature sampling calculates the flow field database of the turbine, including: gridding the turbine blade model based on the two-dimensional aircraft engine turbine blade model data; and calculating the flow field database of the turbine through a CFD solver based on preset turbine initial flow field calculation conditions and the divided grid data.

[0018] Optionally, according to one embodiment of the present application, generating a flow field data set based on the flow field database includes: obtaining a flow field data map in the flow field database, and performing linear interpolation on the flow field data map; averaging the amplified flow field data, including: cutting each flow field to obtain a plurality of square areas with flow field characteristics; and numbering each of the square areas to generate the flow field data set.

[0019] Optionally, according to one embodiment of the present application, the training of the Transformer model based on the training data set includes: using the data in the blade model feature encoding database as the input data of the Transformer model, and mapping the input data into a high-dimensional vector through a linear embedding layer; setting a position encoding vector with the same dimension as the high-dimensional vector; inputting the high-dimensional vector and the position encoding vector into the encoder, normalizing the input data through the encoder and calculating the multi-head self-attention value; performing feature extraction on the flow field map after averaging processing to obtain the flow field feature vector, and training the Transformer model in combination with the flow field feature vector and the multi-head self-attention value.

[0020] Optionally, according to one embodiment of the present application, the encoder calculates the multi-head self-attention value by the following formula:

[0021] MSA=Concat(head1,…,head i ,…,head n );

[0022] head i =Self-attention(QW i Q ,KW i K ,VW i V );

[0023] in,

[0024] Among them, Q, K and V are different vectors obtained through training, T represents the transpose of the matrix, and d k are the dimensions of Q, K, and V, E represents the position encoding vector, n represents the number of heads for calculating multi-head self-attention, head i Represents the block after segmentation of the i-th turbine flow field diagram.

[0025] To achieve the above objectives, the second aspect of the present application further proposes a Transformer-based prediction system for aircraft engine turbine flow fields, comprising the following modules:

[0026] A sampling module is used to perform feature sampling on the guide blades and moving blades of the aircraft engine turbine, construct feature codes according to the collected feature parameters, and process the feature codes to generate a blade model feature code database;

[0027] A generation module, used to calculate the flow field database of the turbine according to the turbine blade model library obtained by feature sampling, and generate a flow field data set according to the flow field database, and construct a model training data set based on the two generated databases, wherein the flow field data set is the target of Transformer model training;

[0028] A training module, used to construct an encoder and a decoder of a Transformer model, wherein the encoder is used to perform multi-head self-attention calculations and train the Transformer model based on the training data set;

[0029] The prediction module is used to obtain the feature code of the current aircraft engine turbine blade to be predicted, input the current feature code into the trained Transformer model, and obtain the flow field prediction result output by the model.

[0030] To achieve the above-mentioned purpose, the third aspect of the present application further proposes an electronic device, comprising:

[0031] processor;

[0032] a memory for storing instructions executable by the processor;

[0033] The processor is configured to execute the instructions to implement the Transformer-based aircraft engine turbine flow field prediction method as described in any one of the first aspects above.

[0034] To achieve the above-mentioned purpose, the fourth aspect of the present application further proposes a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the Transformer-based aircraft engine turbine flow field prediction method described in any one of the above-mentioned first aspects is implemented.

[0035] The technical solution provided by the embodiments of the present application brings at least the following beneficial effects: The present application predicts turbine flow field data based on a data-driven Transformer model, and can quickly and efficiently obtain the flow field data of the aircraft engine turbine, solving the computational complexity and time-consuming problems of current traditional prediction methods, and reducing the computational resources and time costs required to obtain turbine flow field data. In addition, the present application combines a variety of algorithms and data processing methods to ensure the accuracy of the turbine flow field data prediction results. As a result, the present application improves the efficiency, convenience and accuracy of aircraft engine turbine flow field data prediction.

[0036] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0038] Figure 1 A flowchart of a Transformer-based prediction method for an aircraft engine turbine flow field proposed in an embodiment of the present application;

[0039] Figure 2 A schematic diagram of the principle of a Transformer-based prediction of an aircraft engine turbine flow field proposed in an embodiment of the present application;

[0040] Figure 3 A schematic diagram of an aircraft engine turbine blade profile database proposed in an embodiment of the present application;

[0041] Figure 4 A schematic diagram of the working principle of a Transformer model proposed in an embodiment of the present application;

[0042] Figure 5 A flowchart of a specific Transformer model training method proposed in an embodiment of the present application;

[0043] Figure 6 A schematic diagram of the structure of a Transformer-based prediction system for an aircraft engine turbine flow field proposed in an embodiment of the present application;

[0044] Figure 7 A schematic diagram of the structure of an electronic device proposed in an embodiment of the present application. DETAILED DESCRIPTION

[0045] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and should not be construed as limiting the present invention.

[0046] It should be noted that the Transformer-based aircraft engine turbine flow field prediction method implemented in the present application, as a possible implementation method, can be executed by the Transformer-based aircraft engine turbine flow field prediction system proposed in the present application. The Transformer-based aircraft engine turbine flow field prediction system can be applied to any electronic device so that the electronic device can perform the prediction function of aircraft engine turbine flow field data.

[0047] The following describes a Transformer-based prediction method, system and electronic device for an aircraft engine turbine flow field according to an embodiment of the present invention with reference to the accompanying drawings.

[0048] Figure 1 A flowchart of a method for predicting a turbine flow field of an aircraft engine based on a Transformer is provided in an embodiment of the present application, such as Figure 1 As shown, the method comprises the following steps:

[0049] Step S101: feature sampling is performed on the guide blades and moving blades of the aircraft engine turbine, feature codes are constructed according to the collected feature parameters, and the feature codes are processed to generate a blade model feature code database.

[0050] Specifically, when predicting the turbine flow field data of an aircraft engine, the present application first obtains a data set for training the Transformer model, specifically by sampling features of the turbine and then preprocessing the sampled data to obtain a training data set. Among them, feature sampling is performed on the guide vanes and moving blades of the aircraft engine turbine respectively, and feature parameters such as the blade geometry features and relative position data of the guide vanes and moving blades are selected.

[0051] In one embodiment of the present application, the collected characteristic parameters include multiple geometric characteristic parameters of each turbine blade and multiple control point coordinates on the transverse section of the blade, and the multiple geometric characteristic parameters include: inlet geometry angle, effective outlet angle, upper wedge angle, outlet deflection angle, center connection angle, shaft diameter and pitch. In addition, 8 control points on the transverse section of each blade are collected as characteristic parameters of the blade.

[0052] Furthermore, a feature code is constructed based on the above data obtained by sampling. For example, the constructed feature code X = [X1, X2, X3, X4, X5, X6, X7…, Y1… Y8, L1… L8]. Among them, Xn (n = 1, 2… 6, 7…) respectively represents the parameters such as the inlet geometry angle, the effective outlet angle, the upper wedge angle, the outlet deflection angle, the center connection angle, the shaft diameter and the pitch, and Yn and Ln (n = 1, 2… 8) respectively represent the coordinates of the 8 control points of the transverse section of the guide vane and the moving blade.

[0053] Furthermore, in order to obtain a vector set of multi-sample data, the constructed feature coding, ie, the above-mentioned control point coordinates and various parameters are adjusted, and a blade model feature coding database is generated by performing corresponding data processing on the feature coding.

[0054] In one embodiment of the present application, feature codes are processed to generate a blade model feature code database, including the following steps: first, the feature codes are fitted with data by the least squares method; then, based on preset blade model feature parameters, the fitted data is sampled and screened to construct two-dimensional aircraft engine turbine blade model data; then, the two-dimensional aircraft engine turbine blade model data is sampled and processed by the Latin hypercube sampling algorithm to obtain a blade model feature code database.

[0055] Specifically, in this embodiment, if Figure 2 As shown, first, the entire two-dimensional model obtained by feature sampling of the aircraft engine turbine is fitted by the least squares method. Then, based on the smoothness of the suction surface and pressure surface of the model blade, the fitting data is sampled and screened, and finally the two-dimensional aircraft engine turbine blade model data, that is, the turbine blade model library, is constructed. In addition, the present application also samples the obtained two-dimensional aircraft engine turbine blade model data by the Latin hypercube sampling method to obtain the blade model feature coding database of the aircraft engine turbine. For example, the following is obtained by the Latin hypercube sampling method: Figure 3 The 2000 aircraft engine turbine blade model feature coding database shown in Figure 3 Only part of the database data is shown).

[0056] Step S102: Calculate the flow field database of the turbine based on the turbine blade model library obtained by feature sampling, generate a flow field data set based on the flow field database, and construct a model training data set based on the two generated databases, wherein the flow field data set is the target of Transformer model training.

[0057] Specifically, Figure 2 As shown, the turbine flow field database is constructed based on the turbine blade model library obtained in the previous step. For example, the calculation method of computational fluid dynamics (CFD) in the prior art can be adopted.

[0058] In one embodiment of the present application, the flow field database of the turbine is calculated based on the turbine blade model library obtained by feature sampling, including: firstly, meshing the turbine blade model based on the two-dimensional aircraft engine turbine blade model data; and then, based on the preset turbine initial flow field calculation conditions and the divided grid data, calculating the flow field database of the turbine through a CFD solver.

[0059] Specifically, in this embodiment, based on the existing turbine blade model library (i.e., the two-dimensional aircraft engine turbine blade model data obtained by data fitting and screening as mentioned above), the turbine blade model in the turbine blade model library is obtained, and the turbine blade model is meshed. By giving the required turbine initial flow field calculation conditions and the divided grid data, the CFD solver is used for calculation, thereby obtaining the flow field database of the aircraft engine turbine.

[0060] Furthermore, based on the obtained turbine flow field database, a preprocessing operation is performed on the flow field data graph therein to generate a flow field data set.

[0061] In one embodiment of the present application, a flow field data set is generated based on a flow field database, including: first obtaining a flow field data map in the flow field database, and performing linear interpolation on the flow field data map; then averaging the amplified flow field data, and the specific averaging includes cutting each flow field to obtain multiple square areas with flow field characteristics; finally, numbering each square area to generate a flow field data set.

[0062] Specifically, in this embodiment, if Figure 4 As shown in the left half of the figure, the preprocessing operation of the flow field data graph specifically includes the following: first, the flow field graph in the database ( Figure 4 In the example of synthetic aperture radar (SAR) image, linear interpolation is performed to achieve the amplification of flow field data. The flow field is then evenly divided into blocks, and each flow field is cut to obtain a square area with flow field characteristics. Each square area is then numbered to obtain a flow field data set. In this application, the obtained flow field data set is used as the target of Transformer model training.

[0063] Furthermore, based on the two generated databases, namely the blade model feature encoding database and the turbine flow field database, a training data set for the Transformer model is constructed.

[0064] As a possible implementation method, the flow field data set and the feature data set are numbered according to the obtained database, and the data therein are matched one by one. The feature code X of the turbine blade is used as the input of the Transformer model, the flow field data set is used as the training set of the model, and the average data in the flow field data set is used as the output target of the Transformer model, thereby completing the preparation of the data set.

[0065] Step S103: construct an encoder and a decoder of the Transformer model, wherein the encoder is used to perform multi-head self-attention calculations and train the Transformer model based on a training data set.

[0066] Specifically, build a Transformer model, including establishing the model's encoder and decoder networks, input the constructed training data set into the Transformer model, and complete the model training.

[0067] In order to more clearly illustrate the specific implementation process of model training in this application, the following is an exemplary description of a training method of a Transformer model proposed in an embodiment of this application. Figure 5 As shown, the method includes:

[0068] Step S501, using the data in the leaf model feature encoding database as input data of the Transformer model, and mapping the input data into a high-dimensional vector through a linear embedding layer.

[0069] Specifically, the feature code X of the turbine blade of an aircraft engine is used as the input of the model and is mapped to a high-dimensional vector X0 through a linear embedding layer. As an example, X0 = FFN(X), where the FFN network is part of the encoder, namely the Feed-Forward Network in the Transformer model.

[0070] Step S502, setting a position encoding vector with the same dimension as the high-dimensional vector.

[0071] Specifically, a position vector (embedding) having the same dimension as X0 is set, and the position vector can be obtained by random initialization or the like.

[0072] Step S503: input the high-dimensional vector and the position encoding vector into the encoder, and the encoder normalizes the input data and calculates the multi-head self-attention value.

[0073] Specifically, the position encoding vector and the vector output by the linear embedding layer (i.e., the high-dimensional vector X0) are combined as the input of the encoder. Figure 4 As shown, the input data is normalized through the normalization layer LN, and the multi-head self-attention MSA is calculated.

[0074] Among them, attention is calculated by the following formula:

[0075] Among them, Q, K and V are vectors obtained through training, T represents the transpose of the matrix, and d k The dimensions are Q, K, and V. In this application, the number of encoders used in this process is 12.

[0076] It should be noted that, as described above, the present application introduces the position encoding vector E obtained in step S502 in calculating the multi-head self-attention value. Therefore, in this embodiment, the encoder calculates the multi-head self-attention value by the following formula:

[0077] MSA=Concat(head1,…,head i ,…,head n );

[0078] head i =Self-attention(QW i Q ,KW i K ,VW i V );

[0079] in,

[0080] Among them, Q, K and V are different vectors obtained through training, T represents the transpose of the matrix, and d k are the dimensions of Q, K, and V, E represents the position encoding vector, n represents the number of heads for calculating multi-head self-attention, which specifically refers to the number of blocks after the turbine flow field graph is segmented in this application, head i It represents the block after the i-th turbine flow field diagram is divided, and i is any positive number from 1 to n.

[0081] Step S504, extract features from the flow field graph after the averaging process to obtain a flow field feature vector, and train the Transformer model in combination with the flow field feature vector and the multi-head self-attention value.

[0082] It should be noted that, in order to facilitate model training, the present application also performs feature sampling on the flow field map obtained after the averaging process to obtain the flow field feature vector.

[0083] As a possible implementation method, the feature sampling of the flow field map specifically includes: extracting features from the segmented uniform flow field map, outputting a feature vector X of each square uniform flow field map through a linear embedding layer i For example, using a convolution kernel of size 16*16 and a step size of 16, the flow field image input size is H×W×C=224*224*3, where H×W is the image resolution, C is the number of channels, and the number of convolution kernels is 768. The convolution kernel can then reshape the initial input flow field image into a feature map of H×W×C=14*14*768.

[0084] Furthermore, the Transformer model is trained by combining the obtained flow field feature vector and multi-head self-attention value, including optimizing the parameters of the Transformer model so that the prediction results output by the Transformer model based on the input feature encoding are consistent with the flow field feature vector (i.e., the model is trained with the calculated flow field data as the target). Finally, the output results of 12 encoders are fused by 2D convolution channel fusion to obtain the prediction data TransCFD of the aircraft engine turbine flow field output by the Transformer model.

[0085] Furthermore, Figure 2 As shown in the figure, during the model training stage, the error between the predicted data output by the Transformer model and the original data set is also judged by the loss function. For example, the second norm is used as the loss function. If the loss function calculation result is less than the expected value, the model result is saved. If the loss function calculation result exceeds the expected value, the model parameters are changed and retrained. Among them, the original data set is the flow field data set before feature sampling.

[0086] It should be noted that the specific implementation of each step in the training of the Transformer model in this application can refer to the model training method in the relevant embodiment, and this application will not go into details. Through the above-mentioned model training steps, the training of the Transformer model is finally completed.

[0087] Step S104: Obtain the feature code of the current aircraft engine turbine blade to be predicted, input the current feature code into the trained Transformer model, and obtain the flow field prediction result output by the model.

[0088] Specifically, the aircraft engine turbine flow field data is predicted based on the trained Transformer model. The specific implementation process is that for the aircraft engine turbine whose turbine flow field is to be predicted, the feature code of the given aircraft engine turbine blade is first collected. Then the trained Transformer model is called, and the currently collected feature code is input into the Transformer model, and finally the model output is obtained for the prediction result of the flow field data of the aircraft engine turbine to be predicted.

[0089] In summary, the Transformer-based prediction method for aircraft engine turbine flow field in the embodiment of the present application predicts turbine flow field data based on a data-driven Transformer model, which can quickly and efficiently obtain the flow field data of the aircraft engine turbine, solve the problems of computational complexity and time consumption of current traditional prediction methods, and reduce the computing resources and time cost required to obtain turbine flow field data. In addition, the method combines a variety of algorithms and data processing methods to ensure the accuracy of the turbine flow field data prediction results. Therefore, the method improves the efficiency, convenience and accuracy of aircraft engine turbine flow field data prediction.

[0090] In order to implement the above embodiment, the present application also proposes a prediction system for the turbine flow field of an aircraft engine based on Transformer. Figure 6 A schematic diagram of the structure of a Transformer-based prediction system for an aircraft engine turbine flow field proposed in an embodiment of the present application is shown in FIG. Figure 6 As shown, the system includes a sampling module 100 , a generating module 200 , a training module 300 and a predicting module 400 .

[0091] The sampling module 100 is used to perform feature sampling on the guide blades and moving blades of the aircraft engine turbine, construct feature codes according to the collected feature parameters, and process the feature codes to generate a blade model feature code database.

[0092] The generation module 200 is used to calculate the flow field database of the turbine based on the turbine blade model library obtained by feature sampling, and to generate a flow field data set based on the flow field database, and to construct a model training data set based on the two generated databases, wherein the flow field data set is the target of Transformer model training.

[0093] The training module 300 is used to construct the encoder and decoder of the Transformer model, wherein the encoder is used to perform multi-head self-attention calculations and train the Transformer model based on the training data set.

[0094] The prediction module 400 is used to obtain the feature code of the aircraft engine turbine blade to be predicted, input the current feature code into the trained Transformer model, and obtain the flow field prediction result output by the model.

[0095] In one embodiment of the present application, the sampling module 100 is specifically used to: perform data fitting on the feature code by the least squares method; sample and screen the fitted data based on the preset blade model characteristic parameters to construct two-dimensional aircraft engine turbine blade model data; sample and process the two-dimensional aircraft engine turbine blade model data by the Latin hypercube sampling algorithm to obtain a blade model feature code database.

[0096] In one embodiment of the present application, the generation module 200 is specifically used to: grid the turbine blade model based on the two-dimensional aircraft engine turbine blade model data; calculate the turbine flow field database through the CFD solver based on the preset turbine initial flow field calculation conditions and the divided grid data.

[0097] In one embodiment of the present application, the generation module 200 is also used to obtain a flow field data map in a flow field database and perform linear interpolation on the flow field data map; perform equal division processing on the amplified flow field data, including: cutting each flow field to obtain multiple square areas with flow field characteristics; numbering each square area to generate a flow field data set.

[0098] In one embodiment of the present application, the training module 300 is specifically used to use the data in the blade model feature encoding database as the input data of the Transformer model, map the input data into a high-dimensional vector through a linear embedding layer; set a position encoding vector with the same dimension as the high-dimensional vector; input the high-dimensional vector and the position encoding vector into the encoder, normalize the input data through the encoder and calculate the multi-head self-attention value; perform feature extraction on the flow field map after averaging processing, obtain the flow field feature vector, and train the Transformer model in combination with the flow field feature vector and the multi-head self-attention value.

[0099] In summary, the Transformer-based aircraft engine turbine flow field prediction system of the embodiment of the present application predicts turbine flow field data based on a data-driven Transformer model, and can quickly and efficiently obtain the flow field data of the aircraft engine turbine, solving the computational complexity and time-consuming problems of current traditional prediction methods, and reducing the computational resources and time costs required to obtain turbine flow field data. In addition, the system integrates a variety of algorithms and data processing methods to ensure the accuracy of turbine flow field data prediction results. As a result, the system improves the efficiency, convenience and accuracy of aircraft engine turbine flow field data prediction.

[0100] In order to implement the above embodiments, the present application also proposes an electronic device, such as Figure 7As shown, the electronic device 700 includes: a processor 710; a memory 720 for storing executable instructions of the processor 710; wherein the processor 710 is configured to execute instructions to implement a Transformer-based aircraft engine turbine flow field prediction method as described in any of the first aspect embodiments described above.

[0101] In order to implement the above-mentioned embodiments, the present application also proposes a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, it implements the Transformer-based aircraft engine turbine flow field prediction method as described in any of the above-mentioned first aspect embodiments.

[0102] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0103] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of this application, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0104] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, fragment or portion of code comprising one or more executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by technicians in the technical field to which the embodiments of the present application belong.

[0105] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute the instructions), or in combination with these instruction execution systems, devices or apparatuses. For the purpose of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in combination with these instruction execution systems, devices or apparatuses. More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk box (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting or processing in other suitable ways if necessary, and then stored in a computer memory.

[0106] It should be understood that the various parts of the present application can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0107] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.

[0108] In addition, each functional unit in each embodiment of the present application may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0109] The storage medium mentioned above may be a read-only memory, a magnetic disk or an optical disk, etc. Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limiting the present application. A person of ordinary skill in the art may change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A Transformer-based prediction method for aircraft engine turbine flow field. It is characterized in that The following steps are involved: Performing feature sampling on guide blades and moving blades of an aircraft engine turbine, constructing feature codes according to the collected feature parameters, and processing the feature codes to generate a blade model feature code database; Calculate the turbine flow field database according to the turbine blade model library obtained by feature sampling, generate a flow field data set according to the flow field database, and construct a model training data set based on the two generated databases, wherein the flow field data set is the target of Transformer model training; Constructing an encoder and a decoder of a Transformer model, wherein the encoder is used to perform multi-head self-attention calculations and train the Transformer model based on the training data set; The feature code of the current aircraft engine turbine blade to be predicted is obtained, and the current feature code is input into the trained Transformer model to obtain the flow field prediction result output by the model.

2. The prediction method according to claim 1, It is characterized in that The characteristic parameters include multiple geometric characteristic parameters of each turbine blade and multiple control point coordinates on the transverse section of the blade. The multiple geometric characteristic parameters include: air inlet geometry angle, effective air outlet angle, upper wedge angle, outlet deflection angle, center connection angle, shaft diameter and pitch.

3. The prediction method according to claim 1, It is characterized in that The step of processing the feature codes to generate a blade model feature code database includes: Performing data fitting on the feature coding by least square method; Based on the preset blade model characteristic parameters, the fitted data is sampled and screened to construct the two-dimensional aircraft engine turbine blade model data; The two-dimensional aircraft engine turbine blade model data is sampled and processed by a Latin hypercube sampling algorithm to obtain the blade model feature coding database.

4. The prediction method according to claim 3, It is characterized in that The turbine blade model library obtained by feature sampling is used to calculate the turbine flow field database, including: Based on the two-dimensional aircraft engine turbine blade model data, meshing the turbine blade model; Based on the preset turbine initial flow field calculation conditions and the divided grid data, the flow field database of the turbine is calculated by a CFD solver.

5. The prediction method according to claim 1, It is characterized in that The generating of the flow field data set according to the flow field database comprises: Acquire a flow field data graph in the flow field database, and perform linear interpolation on the flow field data graph; The amplified flow field data are evenly divided, including: cutting each flow field to obtain a plurality of square areas with flow field characteristics; Each of the square areas is numbered to generate the flow field data set.

6. The prediction method according to claim 5, It is characterized in that The training of the Transformer model based on the training data set includes: Using the data in the leaf model feature encoding database as input data of the Transformer model, and mapping the input data into a high-dimensional vector through a linear embedding layer; Setting a position encoding vector having the same dimension as the high-dimensional vector; Inputting the high-dimensional vector and the position encoding vector into the encoder, performing a normalization operation on the input data and calculating a multi-head self-attention value through the encoder; Feature extraction is performed on the flow field graph after the averaging process to obtain a flow field feature vector, and the Transformer model is trained in combination with the flow field feature vector and the multi-head self-attention value.

7. The prediction method according to claim 6, It is characterized in that The encoder calculates the multi-head self-attention value by the following formula: MSA=Concat(head 1 ,…,head i ,…,head n ); in, Among them, Q, K and V are different vectors obtained through training, T represents the transpose of the matrix, and d k are the dimensions of Q, K, and V, E represents the position encoding vector, n represents the number of heads for calculating multi-head self-attention, head i Represents the block after segmentation of the i-th turbine flow field diagram.

8. A Transformer-based prediction system for aircraft engine turbine flow field. It is characterized in that Includes the following modules: A sampling module is used to perform feature sampling on the guide blades and moving blades of the aircraft engine turbine, construct feature codes according to the collected feature parameters, and process the feature codes to generate a blade model feature code database; A generation module, used to calculate the flow field database of the turbine according to the turbine blade model library obtained by feature sampling, and generate a flow field data set according to the flow field database, and construct a model training data set based on the two generated databases, wherein the flow field data set is the target of Transformer model training; A training module, used to construct an encoder and a decoder of a Transformer model, wherein the encoder is used to perform multi-head self-attention calculations and train the Transformer model based on the training data set; The prediction module is used to obtain the feature code of the current aircraft engine turbine blade to be predicted, input the current feature code into the trained Transformer model, and obtain the flow field prediction result output by the model.

9. An electronic device, include: processor; a memory for storing instructions executable by the processor; The processor is configured to execute the instructions to implement the Transformer-based aircraft engine turbine flow field prediction method as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, It is characterized in that When the computer program is executed by a processor, the method for predicting the turbine flow field of an aircraft engine based on Transformer as described in any one of claims 1 to 7 is implemented.

Citation Information

Cited By

  • Blade shape generation method, device and equipment of gas compressor and storage medium

    CN119989578A

  • Vision Transform-based flow field prediction model training method and device, computer equipment and storage medium

    CN120430343A