Ultralow-lift impeller vortex grid bionic design method for blade resistance reduction
By constructing a groove structure on the surface of the axial flow pump blades and utilizing biomimetic design and neural network optimization methods, the problem of increased resistance in axial flow pump blades under forward and reverse flow conditions was solved, improving operating efficiency and stability, and simplifying the design and manufacturing process.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-10
- Publication Date
- 2026-04-03
AI Technical Summary
Existing axial flow pump blades experience increased resistance and enhanced flow field fluctuations under both forward and reverse flow conditions, resulting in poor operating efficiency and stability. Existing drag reduction technologies have long design cycles, unoptimized parameters, complex structures, and are difficult to manufacture, making it difficult to maintain high-efficiency operation under bidirectional conditions.
A biomimetic design approach was adopted, and a groove structure was constructed on the blade surface using a neural network prediction model and a vortex lattice method. The groove parameters were optimized by combining the gray wolf optimization algorithm and the non-dominated sorting genetic algorithm. The optimal groove parameters were quickly obtained by using CFTurbo, CREO and ANSYS software for three-dimensional modeling and numerical simulation.
It effectively reduces flow separation and wake development, improves flow adhesion, reduces flow resistance, enhances the operating performance of axial flow pumps, has good engineering adaptability and manufacturing feasibility, and saves calculation and modification costs.
Smart Images
Figure CN121786995A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of hydraulic machinery and pumping station engineering technology, specifically to a biomimetic design method for ultra-low head impeller vortex grids for blade drag reduction. Background Technology
[0002] Axial flow pumps, as key water conveying equipment commonly used in large and medium-sized pumping stations operating under low-head, high-flow conditions, work by rapidly transferring liquid along the axial direction at high speed through impeller rotation. Under complex operating conditions such as flood drainage and water diversion, the axial flow pump impeller must possess stable hydraulic performance in both forward and reverse directions. However, in actual operation, due to issues such as fluid viscosity, boundary layer separation, and wake interference, turbulent boundary layers and flow shedding easily form on the blade surface, especially the upstream side. This leads to increased energy loss within the pump, enhanced flow field fluctuations, and a significant increase in surface resistance, severely impacting the pump's operating efficiency and system stability. Therefore, blade drag reduction has become an important research direction for improving the operating performance of ultra-low-head bidirectional axial flow pumps.
[0003] Existing blade designs often employ smooth blade surfaces or optimized camber distribution to mitigate drag. However, in bidirectional operation, the flow paths differ significantly in both directions, making it difficult for the original design to maintain high efficiency in both directions. This problem is closely related to insufficient blade geometric versatility, weak boundary layer control, and poor adjustability of the operating flow field, especially prominent in scenarios with fluctuating head or rapid start-stop. While existing drag reduction technologies include methods that improve flow conditions by designing local microstructures on the blade surface, these methods typically suffer from long design and calculation cycles, lack of optimized microstructure dimensional parameters, non-universal optimization processes and results, complex structures that are difficult to manufacture, and difficulty in guaranteeing impeller efficiency after optimization.
[0004] Biological structures in nature have developed efficient and stable flow adaptability over a long period of evolution, providing inspiration for the optimized design of hydraulic machinery. Therefore, how to apply biomimetic design concepts to the research on reducing blade surface resistance and improving near-wall flow conditions, thereby enhancing the operating performance of ultra-low head bidirectional axial flow pumps, is an urgent problem to be solved. Summary of the Invention
[0005] To address the aforementioned technical problems, this invention provides a biomimetic design method for ultra-low head impeller vortex grids for blade drag reduction.
[0006] To achieve the above objectives, the present invention adopts the following technical solution: A biomimetic design method for ultra-low head impeller vortex grids for blade drag reduction, targeting an axial flow pump, involves constructing a neural network prediction model for predicting blade performance data according to steps S1 to S4, and then using the neural network prediction model to obtain the optimal groove structure parameters in step A. Step S1: Generate candidate axial flow pump blades based on various target design parameters, and arrange vortex lines on the surface of the candidate axial flow pump blades using the vortex grid method to form a vortex grid on the surface of the candidate axial flow pump blades. Calculate the performance of the candidate axial flow pump blades, and further determine whether the performance of the candidate axial flow pump blades meets the target ultra-low head operating conditions. If yes, set a groove structure at the center of each vortex grid and execute step S2; otherwise, execute step S1. Step S2: Determine the sampling range of each target groove structure parameter, and use orthogonal experimental design to construct a groove structure parameter dataset containing each groove structure parameter group; Step S3: Based on the groove structure parameter dataset, construct a three-dimensional impeller model of the target axial flow pump containing the groove structure. Further integrate the three-dimensional impeller model into the pre-constructed overall fluid domain model of the ultra-low head bidirectional vertical pump device to generate a water body integrated model. Based on the low Reynolds number turbulence model and combined with the preset blade performance index, perform network segmentation and numerical simulation on the water body integrated model. Based on the numerical simulation results, obtain the blade performance simulation data corresponding to each groove structure parameter group, and thus form a sample set with a single groove structure parameter group and its corresponding blade performance simulation data as samples. Step S4: Divide the sample set into a training set, a validation set, and a test set according to a preset ratio; Based on the validation set, use the Grey Wolf optimization algorithm to optimize the hyperparameters of the pre-built neural network model; Further, based on the training set, use the parameter group of a single groove structure in the training sample as input and the blade performance simulation data corresponding to the parameter group of a single groove structure in the training sample as output to train the optimized neural network model and obtain the neural network prediction model. Step A: Construct a dataset of groove structure parameters for the blade to be analyzed using Step S2, and input each set of groove structure parameters into the trained neural network prediction model to obtain blade performance simulation data corresponding to each set of groove structure parameters. Further, in conjunction with the preset optimization objective, use a non-dominated sorting genetic algorithm to determine the optimal groove structure parameters from each set of groove structure parameters.
[0007] Furthermore, in step S1, the flow rate, head, speed, and net positive suction head (NPSH) are input into the CFTurbo software as target design parameters to generate candidate axial flow pump blades.
[0008] Furthermore, the target groove structure parameters in step S2 include the structure length, structure width, groove depth, and the radius of curvature at the top of the groove.
[0009] Furthermore, in step S3, a three-dimensional model of the impeller is constructed using three-dimensional modeling software.
[0010] Furthermore, the blade performance indicators in step S3 include the surface friction drag coefficient under forward and reverse operating conditions, the degree of flow separation under boundary vortex diagnosis, local losses, pressure distribution, and overall efficiency.
[0011] Furthermore, the neural network model in step S4 includes convolutional layers, activation layers, pooling layers, and fully connected layers. The convolutional layers are used to receive the set of groove structure parameters and perform feature extraction. The activation layers are used to perform element-wise nonlinear function mapping on the features output by the convolutional layers. The pooling layers are used to maintain feature invariance and reduce data volume and computational complexity. The fully connected layers are used to classify features, integrate all learned distributed features, and map them to the final sample space.
[0012] Furthermore, after training the optimized neural network model in step S4, the process further includes: based on the test set, using the neural network model to obtain the blade performance prediction data corresponding to each test sample, and determining whether the root mean square error between the blade performance prediction data contained in the test set and the blade performance simulation data contained in the test set is less than a preset threshold. If yes, the model training is completed; otherwise, step S3 is executed to optimize the hyperparameters of the pre-built neural network model using the Grey Wolf optimization algorithm, and the neural network model is retrained.
[0013] Furthermore, the preset optimization objectives mentioned in step A include minimizing blade surface resistance, minimizing local losses, maximizing efficiency, and minimizing the degree of flow separation on the blade surface under forward and reverse operating conditions.
[0014] The beneficial effects of adopting the above technical solution are as follows: (1) By introducing a groove structure with certain geometric regularity on the surface of the blade, the present invention effectively enhances the near-wall flow adhesion, weakens the separation and wake development caused by the sudden change of velocity gradient, and thus improves the overall performance of the ultra-low head axial flow pump device, and solves the problem of vortex concentration and streamline shedding in the blade root and blade tip areas of the existing blades. (2) By combining biomimetic structure, vortex method, simulation analysis and neural network prediction, this invention can quickly obtain the optimal groove structure parameters that can significantly reduce flow resistance and optimize the flow characteristics of impeller surface, effectively saving computational load; (3) The present invention has good engineering adaptability and manufacturing feasibility. It can be directly embedded into the existing blade manufacturing process for surface finishing or laser etching without making major adjustments to the overall structure, thereby significantly saving structural modification costs and construction time. Attached Figure Description
[0015] Figure 1 This is a flowchart of the present invention; Figure 2 This is a schematic diagram of the blade vortex arrangement of the present invention; Figure 3 A schematic diagram of the vortex line arrangement on the blade surface; Figure 4 This is a schematic diagram of the biomimetic structural arrangement of the present invention. Detailed Implementation
[0016] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings.
[0017] refer to Figure 1 A biomimetic design method for ultra-low head impeller vortex grids for blade drag reduction is proposed. For a target axial flow pump, a neural network prediction model is constructed according to steps S1 to S4 to predict blade performance data. Using this neural network prediction model, the optimal groove structure parameters are obtained in step A. Step S1: Using flow rate, head, speed, and net positive suction head (NPSH) as target design parameters, candidate axial flow pump blades are generated using CFTurbo software, and then referenced... Figure 2 and Figure 3 A vortex lattice method program was written using MATLAB to arrange vortex lines on the surface of candidate axial flow pump blades. After several vortex lattices on the surface of the candidate axial flow pump blades, the lift effect generated by simulating fluid flow around the blades was used to calculate the blade performance. The performance of the candidate axial flow pump blades was then determined to meet the requirements of the target ultra-low head condition. If yes, a groove was arranged at the center of each vortex lattice and step S2 was executed; otherwise, step S1 was executed. In this embodiment, the normal velocity induced by a single horseshoe vortex at a control point is calculated. According to the principle of velocity superposition, the total normal velocity at a control point is a linear superposition of the velocities induced by all horseshoe vortices at that point. Based on this principle, a system of linear equations is established. Solving this system of linear equations using Gaussian elimination yields the circulation distribution on each vortex cell. This circulation distribution directly reflects the pressure differential distribution on the blade surface. Based on the circulation distribution, the lift of each vortex cell can be calculated using the Kuta-Zhukovsky theorem. By summing the lift of all vortex cells on the entire blade, the total lift of the blade can be obtained, and the induced drag of the blade can also be calculated. By integrating the tangential components of lift and drag, torque and power are obtained. Combining the calculated lift and torque with the velocity triangle theory, the head and flow rate that the blade can provide under the current geometry, i.e., the blade performance, can be calculated.
[0018] Step S2: Select key biomimetic parameters as the target groove structure parameters, wherein the target groove structure parameters include the structure length. l Structural width b groove depth sand the radius of curvature at the top of the groove r Furthermore, the range of groove parameters is limited by the blade thickness and vortex grid, and an orthogonal experimental design is used to construct a groove structure parameter dataset containing 200 groove structure parameter groups; Step S3: Based on the groove structure parameter dataset, a 3D impeller model of the target axial flow pump containing the groove structure is generated using CREO software. This 3D impeller model is then integrated into a pre-constructed overall fluid domain model of the ultra-low head bidirectional vertical pump device, generating a water body integrated model. Structured meshing is performed using ANSYS Mesh, with localized refinement applied to key areas such as vortex grid locations to ensure simulation accuracy. Furthermore, steady-state CFD numerical simulations are conducted using ANSYS CFX for preset blade performance indicators. Based on the numerical simulation results, blade performance simulation data corresponding to each groove structure parameter group are obtained, forming a sample set with individual groove structure parameter groups and their corresponding blade performance simulation data. A low Reynolds number turbulence model is introduced during the numerical simulation to enhance the ability to capture near-wall flow behavior.
[0019] Step S4: Divide the sample set into a training set, a validation set, and a test set, with 140 groups in the training set, 30 groups in the validation set, and 30 groups in the test set. Based on the validation set, use the Grey Wolf optimization algorithm to optimize the hyperparameters of the pre-built neural network model. Further, based on the training set, use the parameter set of a single groove structure in the training samples as input and the simulated blade performance data corresponding to the parameter set of a single groove structure in the training samples as output to train the optimized neural network model, thus obtaining the neural network prediction model. Step A involves constructing a dataset of groove structure parameters for the blade to be analyzed using step S2. Each set of groove structure parameters is then input into the trained neural network prediction model to obtain simulated blade performance data corresponding to each set of parameters. Further, combined with a preset optimization objective, a non-dominated sorting genetic algorithm is used to determine the optimal groove structure parameters from each set of parameters. And referencing... Figure 4 After obtaining the optimal groove structure parameters, a solid impeller model can be manufactured by CNC machining or laser micro-engraving.
[0020] Furthermore, the blade performance indicators in step S3 include the surface friction drag coefficient under forward and reverse operating conditions, the degree of flow separation under boundary vortex diagnosis, local losses, pressure distribution, and overall efficiency.
[0021] Furthermore, the neural network model in step S4 includes convolutional layers, activation layers, pooling layers, and fully connected layers. The convolutional layers are used to receive the set of groove structure parameters and perform feature extraction. The activation layers are used to perform element-wise nonlinear function mapping on the features output by the convolutional layers. The pooling layers are used to maintain feature invariance and reduce data volume and computational complexity. The fully connected layers are used to classify features, integrate all learned distributed features, and map them to the final sample space.
[0022] Furthermore, after training the optimized neural network model in step S4, the process further includes: based on the test set, using the neural network model to obtain the blade performance prediction data corresponding to each test sample, and determining whether the root mean square error between the blade performance prediction data contained in the test set and the blade performance simulation data contained in the test set is less than a preset threshold. If yes, the model training is completed; otherwise, step S3 is executed to optimize the hyperparameters of the pre-built neural network model using the Grey Wolf optimization algorithm, and the neural network model is retrained.
[0023] Furthermore, the preset optimization objectives mentioned in step A include minimizing blade surface resistance, minimizing local losses, maximizing efficiency, and minimizing the degree of flow separation on the blade surface under forward and reverse operating conditions.
[0024] The above description is merely a preferred embodiment of the present invention and does not constitute any limitation on the present invention. Any simple modifications, alterations, or equivalent structural changes made to the above embodiments based on the technical essence of the present invention shall still fall within the protection scope of the present invention.
Claims
1. A biomimetic design method for ultra-low head impeller vortex lattice for blade drag reduction, characterized in that, For the target axial flow pump, a neural network prediction model for predicting blade performance data is constructed according to steps S1 to S4. Using the neural network prediction model, the optimal groove structure parameters are obtained according to step A. Step S1: Generate candidate axial flow pump blades based on various target design parameters, and arrange vortex lines on the surface of the candidate axial flow pump blades using the vortex grid method to form a vortex grid on the surface of the candidate axial flow pump blades. Calculate the performance of the candidate axial flow pump blades, and further determine whether the performance of the candidate axial flow pump blades meets the target ultra-low head operating conditions. If yes, set a groove structure at the center of each vortex grid and execute step S2; otherwise, execute step S1. Step S2: Determine the sampling range of each target groove structure parameter, and use orthogonal experimental design to construct a groove structure parameter dataset containing each groove structure parameter group; Step S3: Based on the groove structure parameter dataset, construct a three-dimensional impeller model of the target axial flow pump containing the groove structure. Further integrate the three-dimensional impeller model into the pre-constructed overall fluid domain model of the ultra-low head bidirectional vertical pump device to generate a water body integrated model. Based on the low Reynolds number turbulence model and combined with the preset blade performance index, perform network segmentation and numerical simulation on the water body integrated model. Based on the numerical simulation results, obtain the blade performance simulation data corresponding to each groove structure parameter group, and thus form a sample set with a single groove structure parameter group and its corresponding blade performance simulation data as samples. Step S4: Divide the sample set into a training set, a validation set, and a test set according to a preset ratio; Based on the validation set, use the Grey Wolf optimization algorithm to optimize the hyperparameters of the pre-built neural network model; Further, based on the training set, use the parameter group of a single groove structure in the training sample as input and the blade performance simulation data corresponding to the parameter group of a single groove structure in the training sample as output to train the optimized neural network model and obtain the neural network prediction model. Step A: Construct a dataset of groove structure parameters for the blade to be analyzed using Step S2, and input each set of groove structure parameters into the trained neural network prediction model to obtain blade performance simulation data corresponding to each set of groove structure parameters. Further, in conjunction with the preset optimization objective, use a non-dominated sorting genetic algorithm to determine the optimal groove structure parameters from each set of groove structure parameters.
2. The ultra-low head impeller vortex lattice biomimetic design method for blade drag reduction according to claim 1, characterized in that, In step S1, the flow rate, head, speed, and net positive suction head are input into the CFTurbo software as target design parameters to generate candidate axial flow pump blades.
3. The ultra-low head impeller vortex lattice biomimetic design method for blade drag reduction according to claim 1, characterized in that, The target groove structure parameters in step S2 include the structure length, structure width, groove depth, and the radius of curvature at the top of the groove.
4. The ultra-low head impeller vortex grid biomimetic design method for blade drag reduction according to claim 1, characterized in that, In step S3, a 3D model of the impeller is constructed using 3D modeling software.
5. The ultra-low head impeller vortex grid biomimetic design method for blade drag reduction according to claim 1, characterized in that, The blade performance indicators in step S3 include the surface friction drag coefficient under forward and reverse operating conditions, the degree of flow separation under boundary vortex diagnosis, local losses, pressure distribution, and overall efficiency.
6. The ultra-low head impeller vortex grid biomimetic design method for blade drag reduction according to claim 1, characterized in that, The neural network model in step S4 includes convolutional layers, activation layers, pooling layers, and fully connected layers. The convolutional layers are used to receive the set of parameters of the groove structure and extract features. The activation layers are used to perform element-wise nonlinear function mapping on the features output by the convolutional layers. The pooling layers are used to maintain feature invariance and reduce data volume and computational complexity. The fully connected layers are used to classify features, integrate all the learned distributed features, and map them to the final sample space.
7. The ultra-low head impeller vortex grid biomimetic design method for blade drag reduction according to claim 1, characterized in that, Step S4, after training the optimized neural network model, also includes: based on the test set, using the neural network model to obtain the blade performance prediction data corresponding to each test sample, and determining whether the root mean square error between the blade performance prediction data contained in the test set and the blade performance simulation data contained in the test set is less than a preset threshold. If yes, the model training is completed; otherwise, step S3 is executed, using the Grey Wolf optimization algorithm to optimize the hyperparameters of the pre-built neural network model, and the neural network model is retrained.
8. The ultra-low head impeller vortex grid biomimetic design method for blade drag reduction according to claim 1, characterized in that, The preset optimization objectives mentioned in step A include minimizing blade surface resistance, minimizing local losses, maximizing efficiency, and minimizing the degree of flow separation on the blade surface under forward and reverse operating conditions.