Compressor blade profile parameter determination method and compressor blade profile parameter determination device

Abstract

The invention provides a gas compressor blade profile parameter determination method and a gas compressor blade profile parameter determination device.The method comprises the steps that a training set is obtained, the training set comprises a plurality of training samples, aerodynamic performance labels and strength characteristic labels, the aerodynamic performance labels and the strength characteristic labels correspond to the training samples, and the training samples comprise a plurality of target design parameters of a gas compressor blade profile; training the initial pneumatic agent model by using the plurality of training samples and the plurality of pneumatic performance labels to obtain a target pneumatic agent model, and training the initial strength agent model by using the plurality of training samples and the plurality of strength characteristic labels to obtain a target strength agent model; generating a plurality of design parameter combinations by using a genetic algorithm; for each design parameter combination, processing the design parameter combination by using the target aerodynamic proxy model and the target strength proxy model to generate predicted aerodynamic performance and predicted strength characteristics; and determining a target design parameter combination from a plurality of design parameter combinations of which the predicted aerodynamic performance and the predicted strength characteristic meet preset conditions.

Description

Technical Field

[0001] This application relates to the field of compressor technology, and more specifically, to a method and apparatus for determining compressor airfoil parameters. Background Technology

[0002] As the core unit for gas compression in a compressor, the design quality of the compressor blades is a fundamental factor determining the compressor's aerodynamic performance and structural reliability. During compressor operation, the blades must withstand complex aerodynamic, centrifugal, thermal, and vibrational loads while rotating at high speeds. This requires the blade design to simultaneously meet the dual demands of excellent aerodynamic performance and reliable strength characteristics. On the one hand, blades with excellent aerodynamic performance must possess low flow losses, high boosting efficiency, and a wide stability margin to reduce energy loss during compression and improve compressor efficiency and overall performance. On the other hand, blades with reliable strength characteristics must have sufficient structural stiffness and fatigue life to ensure that no fractures or deformations occur under extreme conditions, guaranteeing the safe operation of the aero-engine.

[0003] However, when designing airfoils, the relevant technologies struggle to effectively capture the complex coupling relationship between the design parameters of the compressor airfoil and its aerodynamic performance and strength characteristics, resulting in compromised design efficiency and accuracy, which in turn affects the operational safety of the compressor. Summary of the Invention

[0004] In view of this, this application provides a method and apparatus for determining compressor airfoil parameters.

[0005] One aspect of this application provides a method for determining compressor blade parameters, comprising: acquiring a training set, wherein the training set includes multiple training samples, aerodynamic performance labels and strength characteristic labels corresponding to each training sample, the training samples including multiple target design parameters of the compressor blade; training an initial aerodynamic surrogate model using the multiple training samples and multiple aerodynamic performance labels to obtain a target aerodynamic surrogate model, and training an initial strength surrogate model using the multiple training samples and multiple strength characteristic labels to obtain a target strength surrogate model; generating multiple design parameter combinations using a genetic algorithm, wherein the design parameter combinations include multiple design parameters of the compressor blade; processing each design parameter combination using the target aerodynamic surrogate model and the target strength surrogate model respectively to generate predicted aerodynamic performance and predicted strength characteristics; and determining a target design parameter combination from the multiple design parameter combinations whose predicted aerodynamic performance and predicted strength characteristics satisfy preset conditions.

[0006] Another aspect of this application provides a device for determining compressor blade parameters, comprising: an acquisition module for acquiring a training set, wherein the training set includes multiple training samples, aerodynamic performance labels and strength characteristic labels corresponding to each training sample, the training samples including multiple target design parameters of the compressor blade; a training module for training an initial aerodynamic surrogate model using the multiple training samples and multiple aerodynamic performance labels to obtain a target aerodynamic surrogate model, and training an initial strength surrogate model using the multiple training samples and multiple strength characteristic labels to obtain a target strength surrogate model; a generation module for generating multiple design parameter combinations using a genetic algorithm, wherein the design parameter combinations include multiple design parameters of the compressor blade; a prediction module for processing each design parameter combination using the target aerodynamic surrogate model and the target strength surrogate model respectively, generating predicted aerodynamic performance and predicted strength characteristics; and a determination module for determining a target design parameter combination from the multiple design parameter combinations whose predicted aerodynamic performance and predicted strength characteristics satisfy preset conditions.

[0007] Another aspect of this application provides an electronic device comprising: one or more processors; and a memory for storing one or more programs, wherein, when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to perform the method as described above.

[0008] Another aspect of this application provides a computer-readable storage medium storing computer-executable instructions that, when executed, are used to implement the method described above.

[0009] Another aspect of this application provides a computer program product comprising computer-executable instructions which, when executed, are used to implement the method described above.

[0010] According to embodiments of this application, a target aerodynamic surrogate model and a target strength surrogate model are trained using training samples. After generating multiple design parameter combinations using a genetic algorithm, predicted aerodynamic performance and predicted strength characteristics are generated using the target aerodynamic surrogate model and the target strength surrogate model. The optimal design parameter combination is determined by the genetic algorithm, the target aerodynamic surrogate model, and the target strength surrogate model. Since this application uses the genetic algorithm, the target aerodynamic surrogate model, and the target strength surrogate model to jointly find the optimal design parameter combination, it can effectively capture the complex coupling relationship between the compressor blade design parameters and aerodynamic performance and strength characteristics, thereby improving design efficiency and design accuracy, and effectively ensuring the working safety of the compressor. Attached Figure Description

[0011] The above and other objects, features and advantages of this application will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0012] Figure 1 An exemplary system architecture for determining compressor blade parameters according to embodiments of this application is shown;

[0013] Figure 2 A flowchart illustrating a method for determining compressor blade profile parameters according to an embodiment of this application is shown;

[0014] Figure 3 A schematic diagram of the structure of an initial pneumatic proxy model according to an embodiment of this application is shown;

[0015] Figure 4 A schematic diagram of the structure of the initial strength proxy model according to an embodiment of this application is shown;

[0016] Figure 5 A schematic diagram of the compressor blade profile deconstruction according to an embodiment of this application is shown;

[0017] Figure 6 A schematic diagram of the CFD mesh of the blade-shaped flow channel according to an embodiment of this application is shown;

[0018] Figure 7 A schematic diagram of a CFD computation model according to an embodiment of this application is shown;

[0019] Figure 8 A schematic diagram of the finite element mesh of a blade according to an embodiment of this application is shown;

[0020] Figure 9 A schematic diagram of finite element analysis of a blade according to an embodiment of this application is shown;

[0021] Figure 10 A block diagram of a compressor blade profile parameter determination apparatus according to an embodiment of this application is shown;

[0022] Figure 11 A block diagram of an electronic device suitable for implementing the methods described above, according to an embodiment of this application, is shown. Detailed Implementation

[0023] The embodiments of this application will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of this application. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of this application for ease of explanation. However, it will be apparent that one or more embodiments may be implemented without these specific details. Furthermore, descriptions of well-known structures and technologies are omitted in the following description to avoid unnecessarily obscuring the concepts of this application.

[0024] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0025] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0026] When using expressions such as "at least one of A, B and C", they should generally be interpreted in accordance with the meaning that is commonly understood by those skilled in the art (e.g., "a system having at least one of A, B and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B and C, etc.).

[0027] In the embodiments of this application, the collection, updating, analysis, processing, use, transmission, provision, disclosure, and storage of data (e.g., including but not limited to user personal information) comply with relevant laws and regulations, are used for legitimate purposes, and do not violate public order and good morals. In particular, necessary measures have been taken to prevent unauthorized access to user personal information data and to safeguard user personal information security and network security.

[0028] Current compressor airfoil design faces significant technical challenges, making it difficult to efficiently and accurately balance aerodynamic performance and strength characteristics. From an aerodynamic design perspective, the airflow motion inside a compressor exhibits complex three-dimensional viscous, compressible, and unsteady characteristics, with complex flow phenomena such as shock waves, boundary layer separation, and secondary flows. The geometric parameters of the airfoil and its aerodynamic performance exhibit a highly nonlinear and strongly coupled mapping relationship. Traditional design methods based on empirical formulas or simplified flow field models struggle to accurately capture this complex mapping relationship, resulting in airfoils that often fail to achieve optimal aerodynamic performance or require extensive iterative optimization to approach the design target.

[0029] From the perspective of strength verification and design iteration, the traditional design process typically adopts a serial mode of "aerodynamic design - strength verification - feedback modification," which suffers from serious efficiency bottlenecks. After completing the initial aerodynamic blade design, the blade needs to be strength verified using methods such as finite element analysis to confirm whether its stress and strain under actual operating conditions meet the strength standards. If the strength does not meet the standards, it is necessary to return to the aerodynamic design stage, adjust the blade geometry parameters, and re-perform aerodynamic performance analysis and strength verification. Since there is often a contradiction between aerodynamic performance and strength characteristics (e.g., increasing blade thickness may improve strength, but it will lead to increased aerodynamic drag and flow losses), this serial iterative process often needs to be repeated many times, consuming a large amount of computational resources and time, which seriously restricts the R&D efficiency and design quality of compressor components.

[0030] Current compressor blade designs suffer from multi-dimensional and deep-seated shortcomings in balancing aerodynamic performance and strength characteristics, as well as improving design efficiency and accuracy. These shortcomings can be summarized as follows:

[0031] (1) The design process is serial and fragmented, resulting in low iteration efficiency.

[0032] Current technologies generally employ a sequential design process of "aerodynamic-first design - subsequent strength verification - feedback modification and iteration," where aerodynamic design and strength design are independent and fragmented. In this process, designers must first complete the blade geometry design based on aerodynamic performance targets, then import the design results into strength analysis software (such as ANSYS, ABAQUS, etc.) for stress and strain verification. If the strength does not meet the design standards, the blade geometry parameters must be manually adjusted, and the process must return to the aerodynamic design stage for performance analysis, repeating this iterative process. Since each iteration requires separate aerodynamic and strength simulations, a single iteration cycle typically lasts from several hours to several days. Furthermore, aerodynamic and strength requirements often conflict; for example, while thickening the blade can improve strength, it disrupts the continuity of the aerodynamic flow field and increases flow losses, often resulting in more than ten or even dozens of iterations. This significantly consumes computational resources and time, severely restricting the development cycle of compressor blades and failing to meet the rapid iteration requirements of aero-engines.

[0033] (2) It is difficult to accurately establish the aerodynamic-strength coupling mapping relationship, which limits the design accuracy.

[0034] Existing technologies cannot effectively capture the complex coupling relationship between compressor blade geometry parameters and aerodynamic performance and strength characteristics. On the one hand, for aerodynamic performance, traditional design methods often rely on empirical formulas or linearly simplified models, which are difficult to accurately describe the highly nonlinear and strongly coupled mapping relationship between blade geometry parameters and aerodynamic performance indicators, easily leading to large deviations in the aerodynamic performance of the initial blade design. On the other hand, for strength characteristics, existing technologies often treat strength verification as a "post-verification step," failing to incorporate strength constraints in the early design stage, and thus failing to predict the impact of blade geometry parameters on strength in advance. Furthermore, the coupling mechanism between aerodynamic performance and strength characteristics is not quantified, resulting in designers lacking clear directional guidance during iterative modifications. They can only rely on experience to blindly adjust parameters, and the final designed blade often fails to achieve the optimal balance between aerodynamic performance and strength characteristics simultaneously, significantly reducing design accuracy.

[0035] (3) The parameter optimization methods are limited and lack global optimization capabilities.

[0036] Existing compressor airfoil parameter optimization methods often employ local optimization algorithms (such as gradient descent), which have significant limitations. Local optimization algorithms are prone to getting trapped in local optima and cannot find the globally optimal solution within the broad airfoil parameter space. For example, a certain parameter combination may minimize aerodynamic losses in a local area, but it may not be the optimal solution in the global parameter space that also considers strength, leading to performance shortcomings in the final design. Furthermore, this method cannot dynamically adjust the optimization direction or optimize the parameter search strategy in real time based on aerodynamic and strength feedback results, resulting in low optimization efficiency, high randomness in optimization results, and difficulty in consistently obtaining airfoil design schemes that meet engineering requirements.

[0037] In view of this, embodiments of this application provide a method and apparatus for determining compressor airfoil parameters. The method includes obtaining a training set, wherein the training set includes multiple training samples, aerodynamic performance labels and strength characteristic labels corresponding to each training sample, and the training samples include multiple target design parameters of the compressor airfoil; training an initial aerodynamic surrogate model using the multiple training samples and multiple aerodynamic performance labels to obtain a target aerodynamic surrogate model, and training an initial strength surrogate model using the multiple training samples and multiple strength characteristic labels to obtain a target strength surrogate model; generating multiple design parameter combinations using a genetic algorithm, wherein the design parameter combinations include multiple design parameters of the compressor airfoil; for each design parameter combination, processing the design parameter combination using the target aerodynamic surrogate model and the target strength surrogate model respectively to generate predicted aerodynamic performance and predicted strength characteristics; and determining a target design parameter combination from multiple design parameter combinations whose predicted aerodynamic performance and predicted strength characteristics satisfy preset conditions.

[0038] Figure 1An exemplary system architecture for determining compressor blade parameters, according to embodiments of this application, is shown. It should be noted that... Figure 1 The examples shown are merely examples of system architectures that can be applied to the embodiments of this application, in order to help those skilled in the art understand the technical content of this application, but do not mean that the embodiments of this application cannot be used in other devices, systems, environments or scenarios.

[0039] like Figure 1 As shown, the system architecture 100 according to this embodiment may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 serves as a medium for providing communication links between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired and / or wireless communication links, etc.

[0040] Users can use the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 via the network 104 to receive or send messages, etc. Various communication client applications can be installed on the first terminal device 101, the second terminal device 102, and the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, and / or social media platform software, etc. (for example only).

[0041] The first terminal device 101, the second terminal device 102, and the third terminal device 103 can be various electronic devices with displays and support web browsing, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0042] Server 105 can be a server that provides various services, such as a backend management server that supports websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103 (this is just an example). The backend management server can analyze and process data such as received user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal devices.

[0043] It should be noted that the method for determining compressor blade profile parameters provided in this application embodiment can generally be executed by server 105. Correspondingly, the device for determining compressor blade profile parameters provided in this application embodiment can generally be located in server 105. The method for determining compressor blade profile parameters provided in this application embodiment can also be executed by a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105. Correspondingly, the device for determining compressor blade profile parameters provided in this application embodiment can also be located in a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105.

[0044] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0045] Figure 2 A flowchart illustrating a method for determining compressor blade parameters according to an embodiment of this application is shown.

[0046] like Figure 2 As shown, the method for determining the compressor blade parameters includes operations S201~S205.

[0047] In operation S201, a training set is obtained, which includes multiple training samples, aerodynamic performance labels and strength characteristic labels corresponding to each training sample, and multiple target design parameters of the compressor blade profile.

[0048] In operation S202, an initial aerodynamic surrogate model is trained using multiple training samples and multiple aerodynamic performance labels to obtain a target aerodynamic surrogate model. In addition, an initial strength surrogate model is trained using multiple training samples and multiple strength characteristic labels to obtain a target strength surrogate model.

[0049] In operation S203, a genetic algorithm is used to generate multiple design parameter combinations, which include multiple design parameters of the compressor blade profile.

[0050] In operation S204, for each combination of design parameters, the target aerodynamic surrogate model and the target strength surrogate model are used to process the combination of design parameters to generate predicted aerodynamic performance and predicted strength characteristics.

[0051] In operation S205, the target design parameter combination is determined from multiple design parameter combinations that satisfy preset conditions for predicted aerodynamic performance and predicted strength characteristics.

[0052] According to embodiments of this application, the target design parameters for the compressor blade profile include the leading edge radius, trailing edge radius, maximum thickness, maximum thickness location, blade bend angle, chord length, and root fillet radius. The aerodynamic performance label reflects the performance data of the compressor blade corresponding to the training sample in terms of flow loss, boosting efficiency, and stability margin. The strength characteristic label reflects the strength characteristics of the compressor blade corresponding to the training sample in terms of stress distribution, structural stiffness, and fatigue life.

[0053] According to an embodiment of this application, an initial aerodynamic surrogate model is trained using multiple training samples and multiple aerodynamic performance labels to obtain a target aerodynamic surrogate model. At the same time, an initial strength surrogate model is trained using multiple training samples and multiple strength characteristic labels to obtain a target strength surrogate model. During the training process, a deep neural network can be selected as the initial aerodynamic surrogate model and the initial strength surrogate model. The deep neural network has powerful nonlinear fitting and complex feature extraction capabilities. It can accurately capture the high-dimensional coupling mapping relationship between blade parameters and aerodynamic performance and strength characteristics through multiple hidden layers. Using a deep neural network can further reduce prediction errors and better meet the subsequent high-precision optimization requirements.

[0054] According to an embodiment of this application, a genetic algorithm is first used to generate multiple design parameter combinations, each corresponding to a compressor blade profile. The target aerodynamic surrogate model and target strength surrogate model, obtained through the aforementioned training, predict the predicted aerodynamic performance and predicted strength characteristics of each design parameter combination. The genetic algorithm then iteratively optimizes the design parameter combinations, resulting in multiple design parameter combinations whose predicted aerodynamic performance and predicted strength characteristics satisfy preset conditions. Finally, a target design parameter combination is determined from these multiple design parameter combinations that satisfy the preset conditions.

[0055] According to embodiments of this application, a target aerodynamic surrogate model and a target strength surrogate model are trained using training samples. After generating multiple design parameter combinations using a genetic algorithm, predicted aerodynamic performance and predicted strength characteristics are generated using the target aerodynamic surrogate model and the target strength surrogate model. The optimal design parameter combination is determined by the genetic algorithm, the target aerodynamic surrogate model, and the target strength surrogate model. Since this application uses the genetic algorithm, the target aerodynamic surrogate model, and the target strength surrogate model to jointly find the optimal design parameter combination, it can effectively capture the complex coupling relationship between the compressor blade design parameters and aerodynamic performance and strength characteristics, thereby improving design efficiency and design accuracy, and effectively ensuring the working safety of the compressor.

[0056] According to an embodiment of this application, obtaining a training set includes: analyzing the compressor blade profile based on its aerodynamic and structural characteristics to obtain multiple initial design parameters; performing sensitivity analysis and sample generation on the multiple initial design parameters to obtain multiple training samples; and calculating the aerodynamic performance label and strength characteristic label of the training samples based on the training samples.

[0057] According to the embodiments of this application, based on the aerodynamic and structural characteristics of the compressor blade, the key geometric features of the blade are deconstructed, including multiple initial design parameters such as leading / trailing edge morphology, thickness distribution, bend angle, chord length, and root / tip transition structure, to ensure that the selected parameters can comprehensively affect aerodynamic performance (such as flow loss and pressurization efficiency) and strength characteristics (such as stress distribution and stiffness).

[0058] According to the embodiments of this application, NURBS (Non-Uniform Rational B-Splines) or Bezier curve methods are used for leaf shape modeling. This method can accurately describe the complex curve shape with a small number of control parameters and avoid parameter redundancy.

[0059] According to embodiments of this application, sensitivity analysis and sample generation are performed on multiple initial design parameters to obtain multiple training samples. The aerodynamic performance label and strength characteristic label for each training sample can then be calculated as follows:

[0060] For aerodynamic performance labeling, computational fluid dynamics (CFD) software such as ANSYS CFX and Fluent can be used, based on the three-dimensional Reynolds-Averaged Navier-Stokes equations, selecting the SST k-ω or k-ε turbulence model (balancing computational accuracy and efficiency). Structured mesh generation of the compressor channel is performed using NUMECA Autogrid or ANSYS ICEM. Based on the compressor design conditions (e.g., rated speed, takeoff speed, cruise speed), boundary conditions such as inlet total pressure, total temperature, outlet back pressure, and speed are set. Aerodynamic performance indicators of training samples under multiple operating conditions are calculated, and the results are batch imported into CFD post-processing software (e.g., CFD POST). A script is used to batch extract aerodynamic performance results such as pressure ratio, adiabatic efficiency, total pressure loss coefficient, stability margin, and airflow angle distribution.

[0061] For the strength characteristic labels, software such as ANSYS Mechanical and ABAQUS were used to establish a three-dimensional solid model of the blade, and high-precision meshes were generated (with finer meshes in stress concentration areas such as the blade root and leading edge). The element type was SOLID 186 or C3D8R. Since the solution of the strength characteristics needs to be based on the aerodynamic loads on the blade surface obtained from aerodynamic calculations, an interpolation mapping from the flow field mesh to the structural finite element mesh needs to be established. Data transfer from the flow field load to the structural surface is achieved through unidirectional coupling. Centrifugal loads (calculated based on rotational speed), aerodynamic loads (extracted from the flow field and mapped to the blade surface), and thermal loads (set according to the operating temperature field) were applied. The degrees of freedom of the blade root mounting surface were constrained, and the strength characteristic indices (maximum equivalent stress, first / second natural frequencies, maximum deformation, fatigue life, etc.) of the training samples under multiple operating conditions were solved.

[0062] According to the embodiments of this application, the data of "training samples - aerodynamic performance labels of multiple working conditions - strength characteristic labels of multiple working conditions" are organized into a structured database (formatted as Excel, SQL or HDF5), and samples with simulation non-convergence or abnormal data (such as unreasonable samples with stress far exceeding the allowable value of materials) are removed. Finally, the valid training samples are retained for subsequent model training.

[0063] According to an embodiment of this application, sensitivity analysis and sample generation are performed on multiple initial design parameters to obtain multiple training samples, including: performing sensitivity analysis on multiple initial design parameters to screen multiple initial design parameters and obtain multiple target design parameters; and performing parameter sampling processing based on multiple target design parameters and their respective parameter value ranges to obtain multiple training samples.

[0064] According to embodiments of this application, oversensitivity analysis (such as Morris screening) eliminates variables with weak effects on aerodynamics / strength, ensuring that design variables are both representative and independent, and reducing the sample size for subsequent database construction.

[0065] According to the embodiments of this application, each target design parameter has its own engineering value range, that is, parameter value range, which can be determined based on the design experience of similar compressor blades and material properties. For example, the maximum thickness of GH4169 alloy blades is usually 3.5-6.0 mm.

[0066] According to the embodiments of this application, Latin hypercube sampling (LHS) or orthogonal experimental design can be used to generate 500-1000 sets of non-repeating parameter combination samples within the range of design variable values, ensuring that the samples uniformly cover the entire parameter space and avoiding model bias caused by local sample density.

[0067] Figure 3 A schematic diagram of the structure of an initial pneumatic proxy model according to an embodiment of this application is shown. Figure 4 A schematic diagram of the structure of the initial strength proxy model according to an embodiment of this application is shown.

[0068] According to an embodiment of this application, either the target aerodynamic surrogate model or the target strength surrogate model is trained in the following manner: iteratively performing the following operations: for any training sample, inputting the training sample into a deep neural network and outputting the predicted performance parameters corresponding to the training sample, wherein the predicted performance parameters include predicted aerodynamic parameters or predicted strength parameters; calculating the target loss result based on the predicted performance parameters and the corresponding performance labels; and determining the deep neural network of the last iteration as the surrogate model if the target loss result meets the loss threshold or the number of iterations reaches the number threshold.

[0069] According to the embodiments of this application, Z-Score normalization is applied to multiple parameters (such as chord length and leading edge radius) in the input training samples and different parameters (such as total pressure loss coefficient and maximum stress) in the output predicted performance parameters to eliminate the interference of differences in the magnitude of different parameters (such as "mm-level chord length" and "degree-level bending angle") on model training; at the same time, outliers in the predicted performance parameters (such as stress values ​​that exceed the reasonable range of engineering) are processed by truncation to avoid extreme data affecting model convergence.

[0070] According to embodiments of this application, for the initial aerodynamic proxy model, a deep neural network structure of "input layer - 3 hidden layers - output layer" is adopted, such as... Figure 3 As shown. The number of neurons in the input layer is consistent with the number of target design parameters in the training samples (e.g., 10-15); the hidden layers are set with 64, 128, and 64 neurons respectively, and the activation function is ReLU (to alleviate the gradient vanishing problem); the number of neurons in the output layer matches the number of multiple aerodynamic indices in the predicted aerodynamic parameters (e.g., total pressure loss coefficient, pressure ratio, stability margin, etc., usually 3-5), and the activation function of the output layer is Linear (suitable for continuous value regression prediction).

[0071] According to embodiments of this application, for the initial strength proxy model, since the mapping between strength characteristics and leaf shape parameters is more complex, hidden layers need to be added to improve the fitting ability. Therefore, a deep neural network structure of "input layer - 4 hidden layers - output layer" is adopted, such as... Figure 4 As shown. The number of neurons in the input layer is consistent with the number of target design parameters in the training samples; the hidden layers are set with 128, 256, 128, and 64 neurons respectively, and the activation function is Leaky ReLU, which aims to solve the problem of neuron death in the negative region of ReLU; the number of neurons in the output layer matches the number of intensity characteristic indicators in the predicted intensity parameters (such as maximum stress, natural frequency, deformation, etc., usually 3-4), and the activation function of the output layer is Linear.

[0072] According to the embodiments of this application, the optimizer is the Adam optimizer, with an initial learning rate of 0.001. The learning rate is gradually reduced in the later stages of training to achieve stable convergence. The root mean square error (RMSE) is used as the loss function to more accurately reflect the deviation between the model's predicted values ​​and the true values. Training employs mini-batch gradient descent with a batch size of 32-64 and 200-300 training epochs. Dropout technology is also used, with a hidden layer dropout rate of 0.2-0.3, randomly discarding some neurons to improve the model's generalization ability.

[0073] According to embodiments of this application, test samples from the test set can also be used to verify and optimize the model's accuracy. An early stopping mechanism is introduced during testing, terminating training when the validation set loss shows no decrease for 20 consecutive rounds to avoid model overfitting. In addition to calculating the "mean relative error (MRE) between predicted and true values" on the test set, a coefficient of determination (R²) is introduced to evaluate the model's goodness of fit. Key indicators (such as total pressure loss coefficient and maximum stress) are required to have an MRE ≤ 3% and an R² ≥ 0.95, ensuring that the model's prediction accuracy is superior to traditional algorithms.

[0074] According to embodiments of this application, if the MRE exceeds the threshold, the model is optimized first by increasing the number of neurons in the hidden layer or adjusting the learning rate decay strategy; if it still does not meet the standard, sample data for highly sensitive design variables are supplemented, such as increasing the sampling density for key variables like maximum thickness and leaf root radius, and the model is retrained. If the model is overfitting, it is optimized by increasing the Dropout rate, increasing the data augmentation intensity, or reducing the number of hidden layers until the model's accuracy on both the training and test sets meets the requirements.

[0075] According to the embodiments of this application, the strong fitting ability of deep neural networks is used to accurately mine the complex nonlinear mapping relationship between airfoil parameters and aerodynamic performance and strength characteristics, and to construct a high-precision surrogate model. This provides a reliable performance prediction tool for subsequent genetic algorithms to quickly find the optimal solution. At the same time, compared with traditional machine learning algorithms, it further reduces the prediction error of key performance indicators and improves design accuracy.

[0076] According to an embodiment of this application, before inputting the training samples into the deep neural network, the method further includes: for any training sample, using a Gaussian noise injection method to process some of the design parameters in the multiple target design parameters of the training sample to obtain multiple new training samples, so as to train the deep neural network using the multiple training samples and the multiple new training samples.

[0077] According to the embodiments of this application, for sample data of some highly sensitive indicators (such as total pressure loss coefficient and maximum equivalent stress) in aerodynamic performance and strength characteristics, a Gaussian noise injection method (with noise intensity controlled at 1%-3% of the original training samples) is used to generate additional new training samples, thereby expanding the sample size of key indicators and improving the model's response accuracy to changes in highly sensitive parameters.

[0078] According to an embodiment of this application, a genetic algorithm is used to generate multiple design parameter combinations, including: configuring multiple initial populations based on multiple target design parameters and their respective parameter value ranges, wherein the initial populations include multiple initial individuals, and the initial individuals include multiple design parameters of the compressor blade profile; performing genetic mutation on the multiple initial populations to obtain multiple target individuals, wherein the target individuals represent the design parameter combinations.

[0079] According to embodiments of this application, multiple initial populations are configured based on multiple target design parameters and their respective parameter value ranges. The population size of the initial populations is 80-120 to ensure population diversity and avoid local optima. Then, genetic mutations are performed on the initial populations to obtain multiple target individuals.

[0080] According to an embodiment of this application, genetic mutation is performed on multiple initial populations to obtain multiple target individuals, including: performing a crossover operation on multiple initial populations to obtain multiple crossover populations; and performing a mutation operation on multiple crossover populations to obtain multiple mutated populations, wherein the mutated populations include multiple target individuals.

[0081] According to the embodiments of this application, in specific genetic mutations, the crossover probability can be 0.7-0.9 to control the gene exchange frequency of the population and balance the convergence speed and diversity; the mutation probability can be 0.01-0.05 to introduce a small number of mutations and avoid the algorithm from getting trapped in local optima.

[0082] According to embodiments of this application, the design parameter combination is processed using a target aerodynamic surrogate model and a target strength surrogate model respectively to generate predicted aerodynamic performance and predicted strength characteristics, including: iteratively performing the following iterative genetic operation: for any i-th individual among the initial individual or target individuals, the i-th individual is processed using the target aerodynamic surrogate model and the target strength surrogate model respectively to generate predicted aerodynamic performance and predicted strength characteristics; if the predicted strength characteristics satisfy the strength constraint conditions, the fitness of the i-th individual is calculated based on the predicted aerodynamic performance; and individuals for which genetic mutation operation is performed are selected from multiple i-th individuals with fitness.

[0083] According to embodiments of this application, strength constraints refer to setting strength limits under multiple working conditions as rigid constraints for optimization. Examples include maximum stress ≤ allowable material stress, natural frequency ≥ resonance safety threshold, and maximum deformation ≤ engineering allowable value. Preset selection strategies can employ roulette wheel selection or tournament selection methods.

[0084] According to embodiments of this application, for each individual, it is input into a target aerodynamic surrogate model and a target strength surrogate model to obtain the corresponding predicted aerodynamic performance and predicted strength characteristics. If the predicted strength characteristics satisfy the strength constraints, the fitness of the i-th individual is calculated based on the predicted aerodynamic performance and the optimization objective. The optimization objective can be to achieve optimal aerodynamic performance under multiple operating conditions, such as minimizing the total pressure loss coefficient, maximizing the pressure ratio, and maximizing the stability margin under rated operating conditions. A weighted summation method is used to transform multiple objectives into a single objective. The weights are determined according to the compressor design requirements; for example, the weight of the total pressure loss coefficient is set to 0.5, the pressure ratio to 0.3, and the stability margin to 0.2. Individuals that do not meet the strength constraints are directly eliminated.

[0085] According to embodiments of this application, roulette wheel selection or tournament selection is used to select individuals with high fitness from multiple individuals with fitness to enter the next genetic mutation operation, until the process is iterated 50-100 times (the specific number can be adjusted according to actual needs).

[0086] According to an embodiment of this application, determining a target design parameter combination from multiple design parameter combinations that satisfy preset conditions for predicted aerodynamic performance and predicted strength characteristics includes: determining multiple parameter combinations that satisfy preset conditions for predicted aerodynamic performance and predicted strength characteristics from multiple design parameter combinations; and determining the parameter combination corresponding to the maximum fitness value as the target design parameter combination when the number of iterations reaches a threshold number of iterations.

[0087] According to an embodiment of this application, after the iteration is completed, multiple parameter combinations that satisfy the preset conditions for predicted aerodynamic performance and predicted strength characteristics are determined from multiple design parameter combinations, and the individual with the highest fitness among the multiple individuals obtained by screening is taken as the final target design parameter combination.

[0088] According to an embodiment of this application, after determining the target design parameter combination, the optimal target design parameter combination is input into CFD and finite element analysis software for high-precision simulation, thereby verifying whether its aerodynamic performance and strength characteristics meet the design requirements.

[0089] According to the embodiments of this application, if used for engineering applications, a physical prototype is manufactured based on the combination of target design parameters, and the actual performance is verified by bench tests (pneumatic bench, strength bench). If the verification fails, the constraints or algorithm parameters are adjusted and re-optimized until the requirements are met.

[0090] According to the embodiments of this application, the method for determining compressor blade parameters in this embodiment can shorten the compressor blade design cycle by more than 60%. On the one hand, by using a target aerodynamic surrogate model and a target strength surrogate model trained by machine learning, the time for a single blade performance evaluation is compressed from the current hours to days to seconds or minutes, avoiding repeated time-consuming CFD aerodynamic simulations and finite element strength analyses. On the other hand, by integrating strength constraints into the design stage in advance, combined with the global rapid optimization capability of genetic algorithms, the serial iterative mode of "aerodynamic design - strength verification - repeated modification" is abandoned. Typically, only 1-2 rounds of optimization are needed to obtain a blade scheme that meets the requirements, rather than the more than ten or even dozens of iterations of related technologies. Taking the design of a high-pressure compressor blade for a certain type of aero-engine as an example, related technologies require 2-3 months to complete the design. After adopting this embodiment, the cycle can be shortened to 2-3 weeks, significantly reducing time costs and facilitating rapid iterative upgrades of aero-engines.

[0091] According to the embodiments of this application, the method for determining compressor blade parameters in this embodiment can accurately output a blade design that achieves "strength within limits and optimal aerodynamic performance," improving design accuracy by more than 30% compared to related technologies. Firstly, the surrogate model trained on a massive database can accurately capture the nonlinear coupling relationship between blade parameters and aerodynamic performance and strength characteristics, controlling the prediction error to within 5%, far lower than the 15%-20% error rate of related technologies. Secondly, the genetic algorithm, during the optimization process, aims to maximize aerodynamic performance while adhering to strength compliance constraints. Through multi-generational evolutionary selection, it can lock the optimal solution in the global parameter space, avoiding the performance bottleneck problem of existing local optimization algorithms. Actual verification shows that the blade designed using this embodiment, while meeting strength requirements such as allowable material stress and vibration frequency, significantly reduces the total pressure loss coefficient, significantly improves the pressure ratio and stability margin, and achieves a better level of aerodynamic performance.

[0092] According to the embodiments of this application, the compressor blade design method of this embodiment exhibits excellent performance stability across the entire operating condition range, effectively solving the problem of poor adaptability to varying operating conditions in related technologies. During the training set construction phase, load data and performance requirements under typical operating conditions such as takeoff, cruise, and landing have been covered. The machine learning model can accurately predict aerodynamic and strength performance under different operating conditions. During the genetic algorithm optimization, multi-condition constraints are simultaneously incorporated to ensure that the blade not only meets the standards under rated operating conditions but also satisfies the requirements of "no aerodynamic instability (such as surge, rotating stall) + no strength risk (such as stress over-limit, fatigue damage)" under varying operating conditions. Simulation tests show that the compressor blade designed using this patented technology maintains a maximum stress consistently below 90% of the material's allowable stress within a varying operating condition range of ±15% speed fluctuation and ±20% flow rate change, with aerodynamic efficiency fluctuation not exceeding 3%, fully meeting the robustness requirements of complex aero-engine operating scenarios.

[0093] According to the embodiments of this application, the method for determining compressor airfoil parameters has good engineering scalability and can be flexibly adapted to different types of compressors, such as axial compressors and centrifugal compressors, as well as compressor airfoil designs of different levels, such as high-pressure, low-pressure, and medium-pressure compressors. By simply adjusting the sample range of the training set according to the specific airfoil's structural characteristics (such as blade chord length and aspect ratio) and operating conditions (such as inlet airflow parameters and rotational speed), and retraining the surrogate model, it can be quickly applied to the design of new types of airfoils without requiring significant modifications to the core algorithm. For example, after completing the airfoil design of a certain type of high-pressure axial compressor, only the airfoil parameters and performance data of the centrifugal compressor need to be supplemented, and the model needs to be retrained to apply this method to the centrifugal compressor airfoil design. The adaptation cycle is only 1-2 weeks, significantly improving the engineering application value of the technology.

[0094] Figure 5 A schematic diagram of the compressor blade profile deconstruction according to an embodiment of this application is shown. Figure 6 A schematic diagram of a CFD mesh for a blade-shaped flow channel according to an embodiment of this application is shown. Figure 7 A schematic diagram of a CFD computation model according to an embodiment of this application is shown. Figure 8 A schematic diagram of a blade finite element mesh according to an embodiment of this application is shown. Figure 9 A schematic diagram of finite element analysis of a blade according to an embodiment of this application is shown.

[0095] In a complete embodiment, the implementation process is described in detail based on the method for determining compressor blade parameters of this application, combined with the step breakdown and key parameter description in the technical solution.

[0096] (1) Implementation preparation

[0097] a) Software tools: NREC Axcent 8.9 was used for airfoil parametric modeling, ANSYS CFX 2023 R2 was used for aerodynamic performance calculation, ANSYS Mechanical 2023 R2 was used for strength characteristic calculation, and Python 3.12 (with Scikit-learn, XG Boost, and DEAP libraries) was used for database construction, machine learning surrogate model training, and genetic algorithm optimization.

[0098] b) Basic parameters: The rated operating conditions are determined according to the design requirements of this type of high-pressure axial flow compressor. In addition to the rated operating conditions, the variable operating conditions cover takeoff and cruise conditions; the allowable stress of GH4169 alloy is 650 MPa, the first-order natural frequency safety threshold of the blade is 300 Hz, and the maximum allowable deformation is 0.15 mm.

[0099] (2) Definition of leaf shape parameters

[0100] a) Leaf-shaped geometric deconstruction

[0101] Geometric analysis of the high-pressure axial compressor blades was performed using NREC Axcent 8.9 to identify key geometric features affecting aerodynamic performance and strength characteristics, such as... Figure 5 As shown, the leading edge has a rounded transition structure, which affects the inlet flow loss; the trailing edge has a small rounded structure, which affects the outlet separation loss; the thickness distribution varies asymmetrically along the chord length, which affects the stress distribution and aerodynamic load; the blade bend angle determines the airflow deflection angle, which affects the pressure ratio; the blade root transition fillet affects the stress concentration sensitive area; at the same time, the chord length and aspect ratio are directly related to the overall blade stiffness and centrifugal load bearing capacity.

[0102] b) Selection of parameterization method and determination of design variables

[0103] NURBS curves were used for airfoil modeling, accurately describing the airfoil profile by controlling the vertex coordinates, thus avoiding parameter redundancy. Combined with sensitivity analysis (Morris screening method), variables with negligible impact on aerodynamics / strength (sensitivity coefficient < 0.1) were eliminated, resulting in 12 core design variables (i.e., target design parameters) and their engineering value ranges, as shown in Table 1.

[0104] Table 1

[0105]

[0106] c) Variable correlation verification

[0107] By using Python to call the Scikit-learn library to perform variable correlation analysis, the Pearson correlation coefficients among the design variables were calculated. The results showed that the absolute values ​​of the correlation coefficients among all variables were <0.3, indicating that the variables were independent and there was no significant coupling interference, which can be used for subsequent sample generation and model training.

[0108] (3) Construction of parameter-performance related database

[0109] a) Sample generation

[0110] The Latin hypercube sampling (LHS) method was used to generate 800 unique parameter combination samples (i.e., training samples) within the range of 12 design variables (target design parameters). This ensured that the samples were evenly distributed within the parameter space (sampling bias <5%), avoiding model training bias caused by local sample density. The sampling process was implemented using the Python DEAP library, and the generated samples were stored in Excel format, containing the specific values ​​of the 12 design variables in each sample group.

[0111] b) Aerodynamic performance calculation

[0112] CFD Mesh Generation: The blade profile parameters of each sample are imported into the modeling software to generate a 3D blade model. After importing into NUMECA Autogrid, a computational domain containing the blade and flow channel is constructed, and a structured mesh is used for generation, such as... Figure 6 As shown. The blade surface mesh size is 0.01 mm, the boundary layer mesh has 5 layers, y⁺ < 1, and the mesh quality check shows orthogonality > 0.8, which meets the calculation accuracy requirements.

[0113] Solution setup: Based on the three-dimensional Reynolds-averaged Navier-Stokes (RANS) equations, CFD solution is performed using ANSYS CFX software, such as... Figure 7 As shown. The SST k-ω turbulence model was selected to balance the simulation accuracy and computational efficiency of the separated flow. The boundary conditions were set as follows: the inlet was a total pressure-total temperature boundary (parameters were set separately for rated, takeoff, and cruise conditions), the outlet was a back pressure boundary, the blade surface was a non-slip adiabatic wall, and the hub and casing were rotating walls (matching the operating speed). The solver used the SIMPLEC algorithm, and the convergence criterion was that the residual was less than 1×10⁻⁻⁻⁶. 6 Simultaneously monitor changes in the total pressure ratio and flow coefficient at the inlet and outlet to ensure stable convergence.

[0114] Performance index extraction: After the calculation is completed, the aerodynamic performance indexes of each sample under three working conditions are extracted, such as total pressure loss coefficient (η_loss), pressure ratio (π_c), adiabatic efficiency (η_ad), stability margin (SM), and location of suction surface boundary layer separation point, etc., and are organized to form an aerodynamic performance dataset, i.e. aerodynamic performance label.

[0115] c) Strength property calculation

[0116] Finite element model construction: Import the generated 3D blade model into ANSYS Mechanical, and mesh it using SOLID186 elements, such as... Figure 8 As shown. The mesh is finer in the stress concentration areas at the blade root and leading edge, with an element size of 0.5 mm, while the element size in other areas is 1-2 mm. The total number of meshes is approximately 500,000, and mesh quality checks show an element distortion rate of <0.1%.

[0117] Load and constraint application: Centrifugal load is calculated using the centrifugal force formula; aerodynamic load is extracted from CFX calculation results and mapped to the blade surface through the fluid-structure interaction interface; thermal load is set according to the operating temperature field, and after calculating the temperature distribution using steady-state thermal analysis, the thermal stress is coupled to the structural analysis; the constraint condition is that the blade root mounting surface is fixed with all degrees of freedom (simulating the actual connection state with the wheel disk).

[0118] Strength index calculation: Static analysis is used to solve the stress distribution of the blade, modal analysis is used to solve the first and second natural frequencies, and fatigue analysis is used to calculate the fatigue life, such as... Figure 9 As shown, where Figure 9 (a) is a schematic diagram of stress analysis. Figure 9 (b) is a schematic diagram of deformation analysis. The strength characteristic indicators of each sample under three working conditions are extracted: maximum equivalent stress (σ_max), first natural frequency (f1), second natural frequency (f2), maximum deformation (δ_max), and fatigue life (L), forming a strength characteristic dataset, i.e., strength characteristic labels.

[0119] d) Database cleanup and cleaning

[0120] The data on "blade parameter combination - multi-condition aerodynamic performance - multi-condition strength characteristics" were integrated into a structured database (stored using SQL Server). Data cleaning was performed using the Python Pandas library, and samples with non-convergent aerodynamic simulations (residuals not meeting the convergence criterion) and abnormal strength calculations (such as σ_max > 1000MPa, significantly exceeding material properties) were removed. The remaining samples were used for subsequent training of machine learning surrogate models.

[0121] (4) Training of machine learning agent model

[0122] a) Data preprocessing

[0123] Data partitioning: Valid samples were randomly divided into a training set and a test set in an 8:2 ratio using random sampling. The training set was used for model parameter learning, and the test set was used for accuracy verification. Normalization: The 12 input airfoil design variables and the output aerodynamic performance and strength characteristics were Z-score normalized and mapped to the [0,1] interval to eliminate the impact of parameter magnitude differences on model training.

[0124] b) Model Algorithm Selection and Training

[0125] Deep neural networks (DNNs) were selected as the core algorithm for the surrogate model. Aerodynamic and strength surrogate models were constructed separately. Utilizing the feature extraction capabilities of DNNs' multiple hidden layers, the high-dimensional nonlinear mapping relationship between airfoil parameters and performance indicators was accurately fitted. The structural designs of the initial aerodynamic and strength surrogate models are described in [reference needed]. Figure 3 and Figure 4 .

[0126] c) Hyperparameter optimization and accuracy verification

[0127] Hyperparameter optimization: The hyperparameters of the two models were optimized using the Bayesian optimization algorithm (implemented through the Python Optuna library). The optimization objective was to minimize the mean relative error (MRE) between the predicted and true values ​​on the test set.

[0128] Accuracy verification: Input the test set samples into the optimized surrogate model and calculate the mean repetition rate (MRE) of the key performance indicators (KRI). If the MRE of all key indicators is ≤5%, the model accuracy requirement is met, and it can be used for subsequent genetic algorithm optimization.

[0129] (5) Optimization and verification based on genetic algorithm

[0130] a) Setting optimization goals and constraints

[0131] Optimization objective: To achieve optimal aerodynamic performance under multiple operating conditions, a weighted summation method is used to transform the multi-objective into a single-objective fitness function. Based on the design priority of this type of compressor, the weights are set as follows: rated operating condition η_loss weight 0.3, rated operating condition π_c weight 0.25, rated operating condition SM weight 0.2, takeoff operating condition η_loss weight 0.15, and cruise operating condition SM weight 0.1. The fitness function formula is shown in formula (1): F = 0.3 × (1 - η_loss rated) + 0.25 × π_c rated + 0.2 × SM rated + 0.15 × (1 - η_loss takeoff) + 0.1 × SM cruise (1);

[0132] Among them, the smaller the η_loss, the better the aerodynamic performance. Therefore, "1-η_loss" is used to convert it into a positive index. The larger the F value, the higher the fitness.

[0133] Strength constraints: Set hard strength constraints under multiple operating conditions, as follows: Rated operating condition: σ_max≤650MPa, f1≥300Hz, δ_max≤0.15mm; Takeoff operating condition: σ_max≤650MPa, δ_max≤0.15mm; Cruise operating condition: σ_max≤650MPa, δ_max≤0.15mm.

[0134] b) Genetic Algorithm Parameter Configuration

[0135] Configure the genetic algorithm parameters using the Python DEAP library to ensure that the algorithm balances global search capability and convergence speed:

[0136] Population size: 100, balancing population diversity and computational efficiency. Crossover probability: 0.8, controlling gene exchange frequency to avoid premature convergence. Mutation probability: 0.03, introducing a small number of mutations to prevent the algorithm from getting trapped in local optima. Number of iterations: 80 generations, verified through preliminary experiments that the fitness function tends to stabilize after 80 generations with no significant improvement.

[0137] Selection strategy: Tournament selection method, randomly selects 5 individuals each time, and selects the individual with the highest fitness to enter the next generation, thereby increasing the probability of high-quality individuals being inherited.

[0138] c) Global optimization execution

[0139] Initial population generation: Within the range of 12 design variables, 100 sets of airfoil parameter combinations that meet the engineering constraints are randomly generated as the initial population.

[0140] Fitness calculation: Input the parameters of each group in the initial population into the aerodynamic surrogate model and the strength surrogate model to obtain the corresponding aerodynamic performance index and strength characteristic index; determine whether the strength constraint is satisfied. If it is satisfied, calculate the F value according to the fitness function. If it is not satisfied, set the F value to 0 (directly eliminate).

[0141] Genetic operations: Perform crossover (using single-point crossover, randomly selecting a variable as the crossover point, and exchanging the variable values ​​of two parent individuals) and mutation (randomly adjusting the value of a variable in an individual within the range of its mutation probability) on individuals that meet the constraints to generate new individuals; merge the new individuals with the parent individuals, recalculate the fitness, and eliminate individuals with F=0, retaining individuals that meet the constraints to form the next generation population.

[0142] Iterative convergence: Repeat the "fitness calculation-crossover-mutation-population update" process for 80 generations. If the maximum fitness value shows no significant change for 10 consecutive generations (fluctuation < 0.01), the algorithm is considered converged. Output the individual with the highest fitness in the 80th generation, which is the optimal leaf shape parameter combination (i.e., the target design parameter combination). Specific parameters are shown in Table 2.

[0143] Table 2

[0144]

[0145] d) Optimal solution verification

[0146] High-precision simulation verification: The optimal airfoil parameters are imported into NREC Axcent to generate a three-dimensional model. The simulation parameters are consistent with those in the database construction stage (CFD uses SST k-ω model, and finite element method uses SOLID 186 element). High-precision aerodynamic and strength simulation verification under multiple working conditions is performed. All aerodynamic performance indicators and strength characteristics meet the design objectives and constraints.

[0147] Figure 10 A block diagram of a compressor blade profile parameter determination apparatus according to an embodiment of this application is shown.

[0148] like Figure 10 As shown, the compressor blade profile parameter determination device 1000 includes an acquisition module 1010, a training module 1020, a generation module 1030, a prediction module 1040, and a determination module 1050.

[0149] The acquisition module 1010 is used to acquire a training set, wherein the training set includes multiple training samples, aerodynamic performance labels and strength characteristic labels corresponding to each training sample, and the training samples include multiple target design parameters of the compressor blade profile.

[0150] The training module 1020 is used to train an initial aerodynamic surrogate model using multiple training samples and multiple aerodynamic performance labels to obtain a target aerodynamic surrogate model, and to train an initial strength surrogate model using multiple training samples and multiple strength characteristic labels to obtain a target strength surrogate model.

[0151] The generation module 1030 is used to generate multiple design parameter combinations using a genetic algorithm, wherein the design parameter combinations include multiple design parameters of the compressor blade profile.

[0152] The prediction module 1040 is used to process the design parameter combination using the target aerodynamic surrogate model and the target strength surrogate model for each design parameter combination, and generate predicted aerodynamic performance and predicted strength characteristics.

[0153] The determination module 1050 is used to determine the target design parameter combination from a plurality of design parameter combinations in which the predicted aerodynamic performance and predicted strength characteristics meet preset conditions.

[0154] According to embodiments of this application, a target aerodynamic surrogate model and a target strength surrogate model are trained using training samples. After generating multiple design parameter combinations using a genetic algorithm, predicted aerodynamic performance and predicted strength characteristics are generated using the target aerodynamic surrogate model and the target strength surrogate model. The optimal design parameter combination is determined by the genetic algorithm, the target aerodynamic surrogate model, and the target strength surrogate model. Since this application uses the genetic algorithm, the target aerodynamic surrogate model, and the target strength surrogate model to jointly find the optimal design parameter combination, it can effectively capture the complex coupling relationship between the compressor blade design parameters and aerodynamic performance and strength characteristics, thereby improving design efficiency and design accuracy, and effectively ensuring the working safety of the compressor.

[0155] Any one or more of the modules, submodules, units, and subunits according to the embodiments of this application, or at least part of the functions of any one or more of them, can be implemented in one module. Any one or more of the modules, submodules, units, and subunits according to the embodiments of this application can be implemented by dividing them into multiple modules. Any one or more of the modules, submodules, units, and subunits according to the embodiments of this application can be at least partially implemented as hardware circuits, such as field-programmable gate arrays (FPGAs), programmable logic arrays (PLAs), systems-on-a-chip, systems-on-a-substrate, systems-on-package, application-specific integrated circuits (ASICs), or implemented by hardware or firmware in any other reasonable manner by integrating or packaging circuits, or implemented in any one of software, hardware, and firmware, or in a suitable combination of any of these. Alternatively, one or more of the modules, submodules, units, and subunits according to the embodiments of this application can be at least partially implemented as computer program modules, which, when run, can perform corresponding functions.

[0156] It should be noted that the compressor airfoil parameter determination device part in the embodiments of this application corresponds to the compressor airfoil parameter determination method part in the embodiments of this application. For a detailed description of the compressor airfoil parameter determination device part, please refer to the compressor airfoil parameter determination method part, which will not be repeated here.

[0157] Figure 11 A block diagram of an electronic device suitable for implementing the methods described above, according to an embodiment of this application, is shown. Figure 11 The electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0158] like Figure 11As shown, an electronic device 1100 according to an embodiment of this application includes a processor 1101, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1102 or a program loaded from a storage portion 1108 into a random access memory (RAM) 1103. The processor 1101 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 1101 may also include onboard memory for caching purposes. The processor 1101 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of this application.

[0159] RAM 1103 stores various programs and data required for the operation of electronic device 1100. Processor 1101, ROM 1102, and RAM 1103 are interconnected via bus 1104. Processor 1101 executes various operations of the method flow according to embodiments of this application by executing programs in ROM 1102 and / or RAM 1103. It should be noted that the programs may also be stored in one or more memories other than ROM 1102 and RAM 1103. Processor 1101 may also execute various operations of the method flow according to embodiments of this application by executing programs stored in said one or more memories.

[0160] According to embodiments of this application, the electronic device 1100 may further include an input / output (I / O) interface 1105, which is also connected to a bus 1104. The electronic device 1100 may also include one or more of the following components connected to the input / output (I / O) interface 1105: an input section 1106 including a keyboard, mouse, etc.; an output section 1107 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 1108 including a hard disk, etc.; and a communication section 1109 including a network interface card such as a LAN card, modem, etc. The communication section 1109 performs communication processing via a network such as the Internet. A drive 1110 is also connected to the input / output (I / O) interface 1105 as needed. A removable medium 1111, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 1110 as needed so that computer programs read from it can be installed into the storage section 1108 as needed.

[0161] According to embodiments of this application, the method flow according to embodiments of this application can be implemented as a computer software program. For example, embodiments of this application include a computer program product comprising a computer program carried on a computer-readable storage medium, the computer program containing program code for performing the method shown in the flowchart. In such embodiments, the computer program can be downloaded and installed from a network via communication section 1109, and / or installed from removable medium 1111. When the computer program is executed by processor 1101, it performs the functions defined in the system of embodiments of this application. According to embodiments of this application, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0162] This application also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs, which, when executed, implement the method according to the embodiments of this application.

[0163] According to embodiments of this application, the computer-readable storage medium can be a non-volatile computer-readable storage medium. Examples include, but are not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0164] Embodiments of this application also include a computer program product comprising a computer program containing program code for performing the methods provided in the embodiments of this application. When the computer program product is run on an electronic device, the program code is used to enable the electronic device to implement the methods provided in the embodiments of this application.

[0165] When the computer program is executed by the processor 1101, it performs the functions defined in the system / apparatus of this application embodiment. According to the embodiments of this application, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0166] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and may be downloaded and installed via the communication section 1109, and / or installed from the removable medium 1111. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0167] The embodiments of this application have been described above. However, these embodiments are merely illustrative and not intended to limit the scope of this application. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. This application does not depart from its scope, and those skilled in the art can make various substitutions and modifications, all of which should fall within the scope of this application.

Claims

1. A method of determining parameters of a compressor blade profile, characterized in that, The method comprises the following steps: obtaining a training set, wherein the training set comprises a plurality of training samples, an aerodynamic performance label corresponding to each training sample, and a strength characteristic label, and the training sample comprises a plurality of target design parameters of a compressor blade profile; training an initial aerodynamic proxy model using the plurality of training samples and the plurality of aerodynamic performance labels to obtain a target aerodynamic proxy model, and training an initial strength proxy model using the plurality of training samples and the plurality of strength characteristic labels to obtain a target strength proxy model; generating a plurality of design parameter combinations using a genetic algorithm, wherein the design parameter combination comprises a plurality of design parameters of a compressor blade profile; for each of the design parameter combinations, respectively processing the design parameter combination using the target aerodynamic proxy model and the target strength proxy model to generate a predicted aerodynamic performance and a predicted strength characteristic; determining a target design parameter combination from the plurality of design parameter combinations whose predicted aerodynamic performance and predicted strength characteristic meet a preset condition.

2. The method of claim 1, wherein, The method comprises the following steps: obtaining a training set, comprising: analyzing the compressor blade profile based on its aerodynamic characteristics and structural characteristics to obtain a plurality of initial design parameters; performing sensitivity analysis on the plurality of initial design parameters and generating samples to obtain a plurality of training samples; 3. The method of claim 2, wherein, calculating the aerodynamic performance label and the strength characteristic label of the training sample according to the training sample. The method comprises the following steps: performing sensitivity analysis on the plurality of initial design parameters to filter the plurality of initial design parameters to obtain a plurality of target design parameters; 4. The method of claim 1, wherein, performing parameter sampling processing based on the plurality of target design parameters and their respective parameter value ranges to obtain a plurality of training samples. Any one of the target aerodynamic proxy model and the target strength proxy model is trained in the following manner: iteratively performing the following operations: for any training sample, inputting the training sample into a deep neural network to output a predicted performance parameter corresponding to the training sample, wherein the predicted performance parameter comprises a predicted aerodynamic parameter or a predicted strength parameter; calculating a target loss result according to the predicted performance parameter and the corresponding performance label; 5. The method of claim 4, wherein, in the case where the target loss result meets a loss threshold or the number of iterations reaches a number threshold, determining the deep neural network of the last iteration as the proxy model. Before inputting the training sample into the deep neural network, the method further comprises the following steps:

6. The method of claim 1, wherein, for any training sample, performing processing on part of the plurality of target design parameters in the training sample using a Gaussian noise injection method to obtain a plurality of new training samples, so as to train the deep neural network using the plurality of training samples and the plurality of new training samples. Based on the aerodynamic performance constraint condition and the strength constraint condition, the genetic algorithm is used to generate a plurality of design parameter combinations, comprising: based on the plurality of target design parameters and their respective parameter value ranges, configuring a plurality of initial populations, wherein the initial population comprises a plurality of initial individuals, and the initial individual comprises a plurality of design parameters of a compressor blade profile; The initial populations are subjected to genetic mutation to obtain a plurality of target individuals, wherein the target individuals represent the design parameter combinations.

7. The method of claim 6, wherein, The initial populations are subjected to genetic mutation to obtain a plurality of target individuals, comprising: The initial populations are subjected to cross operation to obtain a plurality of cross populations; The cross populations are subjected to mutation operation to obtain a plurality of mutation populations, wherein the mutation populations include a plurality of target individuals.

8. The method of claim 6, wherein, The design parameter combinations are processed by the target aerodynamic proxy model and the target strength proxy model respectively to generate predicted aerodynamic performance and predicted strength characteristics, comprising: The following iterative genetic operation is iteratively performed: For any ith individual in the initial individuals or target individuals, the ith individual is processed by the target aerodynamic proxy model and the target strength proxy model respectively to generate predicted aerodynamic performance and predicted strength characteristics; In the case that the predicted strength characteristics satisfy the strength constraint condition, the fitness of the ith individual is calculated according to the predicted aerodynamic performance; Based on a preset selection strategy, an individual subjected to genetic mutation operation is selected from a plurality of ith individuals with fitness.

9. The method of claim 8, wherein, A target design parameter combination is determined from a plurality of design parameter combinations whose predicted aerodynamic performance and predicted strength characteristics satisfy a preset condition, comprising: A plurality of parameter combinations whose predicted aerodynamic performance and predicted strength characteristics satisfy a preset condition are determined from a plurality of design parameter combinations; In the case that the number of iterations reaches a genetic number threshold, a parameter combination corresponding to a maximum fitness value is determined as the target design parameter combination.

10. An apparatus for determining a compressor blade profile parameter, characterized by Comprising: An acquisition module is configured to acquire a training set, wherein the training set includes a plurality of training samples, aerodynamic performance labels corresponding to each training sample, and strength characteristic labels, and each training sample includes a plurality of target design parameters of a compressor blade profile; A training module is configured to train an initial aerodynamic proxy model by using the plurality of training samples and the plurality of aerodynamic performance labels to obtain a target aerodynamic proxy model, and train an initial strength proxy model by using the plurality of training samples and the plurality of strength characteristic labels to obtain a target strength proxy model; A generation module is configured to generate a plurality of design parameter combinations by using a genetic algorithm, wherein each design parameter combination includes a plurality of design parameters of a compressor blade profile; A prediction module is configured to process each design parameter combination by using the target aerodynamic proxy model and the target strength proxy model respectively to generate predicted aerodynamic performance and predicted strength characteristics; A determination module is configured to determine a target design parameter combination from a plurality of design parameter combinations whose predicted aerodynamic performance and predicted strength characteristics satisfy a preset condition.

Images