Water turbine blade topological optimization method combining double-line learning neural network and conformal mapping

Abstract

The invention discloses a water turbine blade topological optimization method combining a double-line learning neural network and conformal mapping. Firstly, a water turbine blade model is mapped to a two-dimensional plane through a conformal mapping algorithm, and a data sample library and corresponding conformal factors are generated. And obtaining an external force field and a displacement field of a model node under a specific flow field condition by adopting fluid-structure interaction analysis. According to the method, a global attention U-NET neural network is introduced, and a conformal factor, a structural topology, an external force field and a displacement field are used as input and output of the network to train the neural network. In the optimization process, dependence on traditional finite element analysis is reduced, the optimization efficiency is improved, efficient topological optimization of the water turbine blade is achieved through combination of off-line learning and on-line learning, and then the performance and reliability of a water turbine are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of fluid machinery structure optimization, and in particular to a turbine blade topology optimization method combining a dual-line learning neural network and conformal mapping. Background Art

[0002] As an important water energy conversion equipment, hydraulic turbines are widely used in hydropower stations around the world. Their efficiency and reliability directly affect the energy conversion efficiency and economic benefits of hydropower stations. As one of the key components of hydraulic turbines, the improvement of the mechanical properties of hydraulic turbine blades is crucial to improving the performance of hydraulic turbines. Topological optimization can significantly improve the mechanical properties of blades. However, traditional topological optimization methods for blades usually rely on empirical design and finite element analysis, which are often time-consuming, costly, and difficult to adapt to complex working conditions and changing design requirements. Therefore, it is necessary to design a hydraulic turbine blade optimization method that can reduce reliance on finite element processes and is versatile.

[0003] The information disclosed in this background technology section is only intended to enhance the understanding of the overall background of the invention and should not be regarded as an acknowledgment or any form of suggestion that the information constitutes the prior art already known to a person skilled in the art. Summary of the invention

[0004] The purpose of the present invention is to provide a turbine blade topology optimization method combining a dual-line learning neural network with conformal mapping, thereby overcoming the shortcomings of the existing turbine blade topology optimization, such as long time consumption, high cost and low efficiency.

[0005] To achieve the above object, the present invention provides a method for topological optimization of turbine blades combining a dual-line learning neural network and conformal mapping, comprising the following steps:

[0006] (1) Create a data sample library consisting of turbine blade models, use a conformal mapping algorithm for each sample, map the sample to a two-dimensional plane, and obtain a mapped two-dimensional grid and a conformal factor corresponding to the grid node;

[0007] (2) Perform fluid-structure coupling analysis on each sample under different structural topologies to obtain the external force field and displacement field of the model nodes under specific flow field conditions;

[0008] (3) Replace the smooth transition layer of the U-NET neural network with the global attention layer to obtain the global attention U-NET neural network;

[0009] (4) taking the conformal factor and structural topology of each sample described in step (1) as the input of the network, and the external force field and displacement field of the model node generated in step (2) as the output of the network, integrating the input and output of the network to form a separate training sample matrix;

[0010] (5) All training sample matrices obtained in step (4) are integrated as a whole to serve as the global attention U-NET neural network for offline learning dataset training step (3);

[0011] (6) The turbine blade model to be optimized is processed by a common mapping algorithm to obtain the mapped mesh and conformal factor, and on this basis, the level set topology optimization method is initialized and multiple optimization iterations are performed;

[0012] (7) integrating the external force field, displacement field, structural topology and conformal factor generated in step (6), and obtaining an online learning data set by referring to step (5);

[0013] (8) retraining the global attention U-NET neural network trained in step (5) using an online learning dataset;

[0014] (9) replacing the fluid-solid coupling analysis step of the level set method with the global attention U-NET neural network trained twice in step (8) to generate a level set method combined with dual-line learning;

[0015] (10) The level set method combined with dual-line learning obtained in step (9) is used for multiple subsequent optimization iterations of the model selected in step (5).

[0016] Preferably, in the above technical solution, the common mapping algorithm comprises the following steps:

[0017] 1) The RICCI flow algorithm for discrete triangular meshes is used, and the two-dimensional plane mapped to is a bounded rectangular plane;

[0018] 2) Find the largest inscribed rectangle on the surface of the turbine blade, and set the four nodes of the rectangle as fixed points of mapping;

[0019] 3) The obtained conformal factor is converted into a matrix with the same size as the two-dimensional grid through the centroid interpolation algorithm.

[0020] Preferably, in the above technical solution, the external force field and displacement field generated in step (2) are represented in the form of a matrix, and the size of the matrix is ​​consistent with the matrix in step 3).

[0021] Preferably, in the above technical solution, the global attention layer in step (3) is a network architecture that convolves the matrix for each channel separately and finally superimposes the restored network architecture.

[0022] Preferably, in the above technical solution, the naming of the external force field and displacement field in step (4) is unified with the conformal factor obtained in step (1), so as to realize the correspondence between the node factor and the external force and displacement on the grid node.

[0023] Preferably, in the above technical solution, the offline learning data set in step (5) is divided into two parts, one part is used to train the global attention U-NET neural network, and the other part is used to verify the global attention U-NET neural network.

[0024] Preferably, in the above technical solution, training the global attention U-NET neural network in step (5) includes: continuously inputting the data set into the global attention U-NET neural network according to batch conditions, and adaptively adjusting the learning rate parameter of the global attention U-NET neural network until the loss function value calculated using the cross entropy loss function converges.

[0025] Preferably, in the above technical solution, the level set topology optimization method used in step (6) is an extended level set method combined with a conformal factor, which drives the evolution of the turbine blade topology according to the results of fluid-structure coupling analysis.

[0026] Preferably, in the above technical solution, the multiple optimization iterations of the model are 10-30 optimization iterations.

[0027] Compared with the prior art, the present invention has the following beneficial effects:

[0028] (1) The present invention combines a dual-line learning neural network with a conformal mapping method for turbine blade topology optimization. First, a data sample library consisting of turbine blade models is created, and each sample is mapped to a two-dimensional plane using a conformal mapping algorithm to obtain a mapped two-dimensional grid and a corresponding conformal factor. Subsequently, a fluid-solid coupling analysis is performed on each sample to obtain the external force field and displacement field of the model node under specific flow field conditions. By replacing the smooth transition layer of the U-NET neural network with a global attention layer, a global attention U-NET neural network is obtained to improve the accuracy of the prediction.

[0029] (2) The method of the present invention not only includes the step of preprocessing the turbine blade model using a conformal mapping algorithm, but also involves using a global attention U-NET neural network to train the model offline and online, and finally realizes the topology optimization of the turbine blade. This method can effectively reduce the finite element analysis process that traditional topology optimization relies on, improve the optimization efficiency, and maintain or improve the performance of the turbine blade.

[0030] (3) Through the implementation of the present invention, the mechanical properties of turbine blades can be significantly improved, and their structural design can be optimized, thereby improving the overall performance and reliability of the turbine, providing technical support for improving the energy conversion efficiency and economic benefits of the hydropower station. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 is a step diagram of a method for topological optimization of a turbine blade combining a dual-line learning neural network and conformal mapping according to the present invention;

[0032] Figure 2 It is a diagram showing the changes in the blade meshes before and after mapping in the method according to the present invention.

[0033] Figure 3 is a schematic diagram of training an offline network in a method according to the present invention;

[0034] Figure 4 Schematic diagram of the structure of the global attention U-NET neural network in the method according to the present invention.

[0035] Figure 5 Schematic diagram of the detailed structure of the global attention layer in the method according to the present invention.

[0036] Figure 6 It is a schematic diagram of the first 20 iterations of the online learning neural network in the method according to the present invention.

[0037] Figure 7 It is a schematic diagram of the iterative process after online learning of the neural network 20 times in the method according to the present invention. DETAILED DESCRIPTION

[0038] The specific implementation modes of the present invention are described in detail below in conjunction with the accompanying drawings, but it should be understood that the protection scope of the present invention is not limited by the specific implementation modes.

[0039] Unless explicitly stated otherwise, throughout the specification and claims, the term “comprise” or variations such as “include” or “comprising”, etc., will be understood to include the stated elements or components but not to exclude other elements or components.

[0040] like Figures 1 to 7 As shown, according to a specific embodiment of the present invention, a method for topology optimization of a turbine blade combining a dual-line learning neural network and conformal mapping comprises the following steps:

[0041] S1: Create a data sample library consisting of turbine blade models. Figure 2 , a conformal mapping algorithm is used for each sample to map the sample to a two-dimensional plane, so as to obtain the mapped two-dimensional grid and the conformal factor corresponding to the grid node.

[0042] In this embodiment, 50 turbine blade models with different shapes are collected, and 60 different structural topologies are set for each model. Sample individuals are defined based on the different shapes and structural topologies. These 3,000 samples form a sample database. At the same time, the conformal mapping algorithm shown in step S1 can be specifically divided into:

[0043] 1): The conformal mapping method used in step S1 is the RICCI flow algorithm for discrete triangle meshes. At the same time, the two-dimensional plane mapped is a bounded rectangular plane.

[0044] 2): Find the largest inscribed rectangle on the surface of the turbine blade. Set the four nodes of the rectangle as fixed points for mapping.

[0045] 3): The obtained conformal factor is converted into a matrix with the same size as the two-dimensional grid through the centroid interpolation algorithm. In this embodiment, the size of the two-dimensional grid is 150x150.

[0046] Step S1 achieves the purpose of reducing the dimension of the turbine blade model to a two-dimensional plane, and at the same time collects the conformal factors that characterize the original surface features of the model for subsequent processing.

[0047] S2: Perform fluid-solid coupling analysis on each sample to obtain the external force field and displacement field of the model nodes under specific flow field conditions.

[0048] In the topology optimization of turbine blades, the specific form of finite element analysis is fluid-solid coupling analysis. In this embodiment, all samples are placed in a uniform incompressible flow field with a flow rate of 2m / S. Fluid-solid coupling analysis is performed on each data sample in step S1 to obtain the external force field and displacement field generated by each blade in a given flow field. These two field data will provide a basis for subsequent optimization links.

[0049] S3: Replace the smooth transition layer of the U-NET neural network with the global attention layer to obtain the global attention U-NET neural network.

[0050] The network architecture adopted by the present invention is improved based on the U-NET neural network. Figure 3 As shown in FIG. 1 , the present invention replaces the smooth transition layer at the bottom of the U-NET neural network with a global attention layer. The global attention layer is a network architecture that allows the matrix to be convolved along different dimensions and then superimposed and restored. Compared with the smooth transition layer, the global attention layer performs convolution processing in different dimensions, so it has better analytical ability for the local information of the matrix and is more suitable for topology optimization problems.

[0051] S4: Take the conformal factor and structural topology of each sample in step S1 as the input of the network, and the external force field and displacement field of the model node generated in step S2 as the output of the network, integrate the input and output of the network to form a separate training sample matrix.

[0052] Finite element analysis in topology optimization is mainly used to calculate the external force field and displacement field of the model under the current structural topology. Therefore, the neural network used in the present invention should be able to predict the corresponding external force field and displacement field based on the conformal factor of the model and the current structural topology. Therefore, in this embodiment, the conformal factor and structural topology of each sample in step S1 are used as the input of the network; the external force field and displacement field of the model node generated by each sample in step S2 are used as the output of the network.

[0053] S5: Integrate all training sample matrices obtained in step S4 and use them as the global attention U-NET neural network for offline learning data set training step S3.

[0054] The two-line neural network proposed in the present invention is a neural network trained by two processes, online learning and offline learning. The offline learning process is as follows: Steps S1 to S5, Figure 4 As shown. The purpose of offline learning is to let the neural network learn the relationship between the two input factors, the conformal factor and structural topology of the turbine blades in a wide range of applications compared with the external force field and the displacement field in the two output factors through a wide range of data samples. At the same time, for the convenience of data processing, the input data and output data of step S4 are both matrices of the same size.

[0055] The optimizer for training the neural network is Adam; in the offline learning data set, the ratio of training set to test set is 7:3; the loss function used in training is the cross entropy loss function.

[0056] S6: The turbine blade model to be optimized is processed by the conformal mapping algorithm of step S1 to obtain the mapped mesh and conformal factor, and on this basis, the level set topology optimization method is initialized and 20 optimization iterations are performed.

[0057] The level set method used in step S6 is an extended level set method combined with conformal mapping, which is suitable for optimization objects with complex surfaces such as turbine blades.

[0058] S7: Integrate the external force field, displacement field, structural topology and conformal factor generated in step S6, and obtain an online learning data set with reference to step S5.

[0059] In order to ensure the transferability of the global attention U-NET neural network, offline learning and online learning are consistent in learning form, the only difference is the training data.

[0060] S8: Perform secondary training on the global attention U-NET neural network trained in step S5 using an online learning dataset.

[0061] The process of online learning is as follows: Steps S6 to S8, Figure 5 As shown in the figure. The purpose of online learning is to let the neural network learn the relationship between the structural topology of the current blade to be optimized and the external force field and displacement field. The combination of offline learning and online learning can reduce the network training burden for a specific turbine blade model while ensuring the grid prediction accuracy.

[0062] S9: The global attention U-NET neural network trained by online learning in step S8 replaces the fluid-solid coupling analysis step of the level set method to generate a level set method combined with dual-line learning.

[0063] The overall flow chart of dual-line learning based on neural network and conformal mapping is as follows Figure 6 The global attention U-NET neural network trained by offline learning and online learning can accurately predict the corresponding external force field and displacement field according to the structural topology of the model to be optimized in step S6.

[0064] S10: The level set method combined with dual-line learning obtained in step S9 is used for the optimization iterations after 20 of the model selected in step S5.

[0065] The foregoing description of specific exemplary embodiments of the present invention is for the purpose of illustration and demonstration. These descriptions are not intended to limit the present invention to the precise form disclosed, and it is clear that many changes and variations can be made based on the above teachings. The purpose of selecting and describing the exemplary embodiments is to explain the specific principles of the present invention and its practical application, so that those skilled in the art can realize and utilize various different exemplary embodiments of the present invention and various different selections and changes. The scope of the present invention is intended to be limited by the claims and their equivalents.

Claims

1. A method for topological optimization of turbine blades combining bilinear learning neural network and conformal mapping, characterized in that: The following steps are involved: (1) Create a data sample library consisting of turbine blade models, use a conformal mapping algorithm for each sample, map the sample to a two-dimensional plane, and obtain the mapped two-dimensional grid and the conformal factor corresponding to the grid node; (2) Perform fluid-structure coupling analysis on each sample under different structural topologies to obtain the external force field and displacement field of the model nodes under specific flow field conditions; (3) Replace the smooth transition layer of the U-NET neural network with the global attention layer to obtain the global attention U-NET neural network; (4) taking the conformal factor and structural topology of each sample described in step (1) as the input of the network, and the external force field and displacement field of the model node generated in step (2) as the output of the network, integrating the input and output of the network to form a separate training sample matrix; (5) All training sample matrices obtained in step (4) are integrated as a whole to serve as the global attention U-NET neural network for offline learning dataset training step (3); (6) The turbine blade model to be optimized is processed by a common mapping algorithm to obtain the mapped mesh and conformal factor, and on this basis, the level set topology optimization method is initialized and multiple optimization iterations are performed; (7) integrating the external force field, displacement field, structural topology and conformal factor generated in step (6), and obtaining an online learning data set by referring to step (5); (8) retraining the global attention U-NET neural network trained in step (5) using an online learning dataset; (9) replacing the fluid-solid coupling analysis step of the level set method with the global attention U-NET neural network trained twice in step (8) to generate a level set method combined with dual-line learning; (10) The level set method combined with dual-line learning obtained in step (9) is used for multiple subsequent optimization iterations of the model selected in step (5).

2. The method for topological optimization of turbine blades combining bilinear learning neural network and conformal mapping according to claim 1, characterized in that: The common mapping algorithm comprises the following steps: 1) The RICCI flow algorithm for discrete triangular meshes is used, and the two-dimensional plane mapped to is a bounded rectangular plane; 2) Find the largest inscribed rectangle on the surface of the turbine blade, and set the four nodes of the rectangle as fixed points of mapping; 3) The obtained conformal factor is converted into a matrix with the same size as the two-dimensional grid through the centroid interpolation algorithm.

3. The method for topology optimization of turbine blades combining bilinear learning neural network and conformal mapping according to claim 1, characterized in that: The external force field and displacement field generated in step (2) are expressed in the form of a matrix, and the size of the matrix is ​​consistent with the matrix in step 3).

4. The method for topology optimization of turbine blades combining bilinear learning neural network and conformal mapping according to claim 1, characterized in that: The global attention layer in step (3) convolves the matrix for each channel separately and finally superimposes the restored network architecture.

5. The method for topology optimization of turbine blades combining bilinear learning neural network and conformal mapping according to claim 1, characterized in that: The external force field and displacement field in step (4) are named in a unified manner with the conformal factor obtained in step (1), so as to realize the correspondence between the node factor and the external force and displacement on the grid node.

6. The method for topology optimization of turbine blades combining bilinear learning neural network and conformal mapping according to claim 1, characterized in that: The offline learning dataset in step (5) is divided into two parts, one part is used to train the global attention U-NET neural network, and the other part is used to verify the global attention U-NET neural network.

7. The method for topology optimization of turbine blades combining bilinear learning neural network and conformal mapping according to claim 1, characterized in that: Training the global attention U-NET neural network in step (5) includes: continuously inputting the data set into the global attention U-NET neural network according to batch conditions, and adaptively adjusting the learning rate parameter of the global attention U-NET neural network until the loss function value calculated using the cross entropy loss function converges.

8. The method for topology optimization of turbine blades combining bilinear learning neural network and conformal mapping according to claim 1, characterized in that: The level set topology optimization method used in step (6) is an extended level set method combined with a conformal factor, which drives the evolution of the turbine blade topology based on the results of the fluid-structure coupling analysis.

9. The method for topology optimization of turbine blades combining bilinear learning neural network and conformal mapping according to claim 1, characterized in that: The multiple optimization iterations of the model were 10-30 optimization iterations.

Images