Data driving-based runner blade surface static stress neural network prediction method

Abstract

The invention discloses a runner blade surface static stress neural network prediction method based on data driving, and the method comprises the steps: building a mixed-flow hydraulic turbine set full-flow-channel model, carrying out the grid division of the model, importing a grid corresponding to a working condition point into numerical simulation software, and obtaining runner blade surface pressure data; the method comprises the following steps: establishing a hydroelectric generating set shafting model, performing unstructured grid division on the model, inputting runner blade surface pressure data into an interface of a fluid domain and a solid domain to obtain runner blade surface static stress data and corresponding three-dimensional coordinates, and dividing the static stress data into a training set and a test set; and training the prediction model by using the three-dimensional coordinates, the water head and the opening corresponding to the training set, and inputting the three-dimensional coordinates, the water head and the opening of the working condition point into the trained prediction model to obtain the surface static stress of each runner blade under the working condition point. The method can correctly reflect the distribution trend of the static stress borne by the blade under all working conditions, and is high in precision.

Description

Technical Field

[0001] The present invention relates to the technical field of runner blade state monitoring, and in particular to a data-driven neural network prediction method for static stress on the surface of a runner blade. Background Art

[0002] The runner of a water turbine is usually in a high-speed rotating state. The pressure load generated by the flow of water causes the stress on the runner to change continuously. Once this alternating stress exceeds the bearing limit of the material, fatigue damage will occur. Fatigue damage is usually manifested as the formation and expansion of cracks. Under long-term stress, tiny cracks will gradually form on the surface of the runner. These cracks will eventually lead to fatigue damage of the runner in the process of continuous expansion and merging. Fatigue damage to the runner of a water turbine will affect its performance and safety, and may cause the runner to break or even cause serious accidents. Therefore, real-time monitoring and evaluation of the stress on the runner of a water turbine and taking corresponding measures to avoid fatigue damage are crucial to ensure the safe operation of the turbine.

[0003] Since the efficiency and accuracy of runner stress prediction directly determine the efficiency and accuracy of runner stress optimization, it is particularly important to quickly and accurately calculate the stress on the surface of the turbine runner. Traditional fluid-solid coupling calculations take too long and are gradually unable to meet the actual needs of engineering. For complex hydraulic machinery runners, the acquisition of surface stress requires a lot of computing resources. A complete transition process usually takes tens of seconds, and it is very difficult to accurately reflect its stress characteristics through numerical calculations. It is imperative to explore prediction methods that take less time and occupy less computing resources.

[0004] Referring to the research progress at home and abroad, it is not difficult to see that researchers often choose to obtain key parameters related to the life of mechanical equipment through experimental measurement or finite element analysis, so as to extract and organize data sets for building machine learning prediction models. By training and predicting these samples, the prediction results of key factors affecting the life of mechanical equipment are obtained, thereby realizing the life management of mechanical equipment. In recent years, many scholars have taken minimizing the life loss of mechanical equipment as the optimization goal, and proposed a series of models based on machine learning and deep learning technology for predicting factors affecting the life of mechanical equipment, which provides new research ideas for the life management of mechanical equipment. Considering that neural network prediction takes less time and has good generalization performance, it is possible to apply neural network methods to the static stress prediction of complex hydraulic machinery blades. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a data-driven neural network prediction method for static stress on the surface of turbine blades, and to establish a static stress prediction model for the surface of turbine blades based on a fully connected neural network, which can provide guidance and modification suggestions for the operating status of the overall hydropower unit.

[0006] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0007] A data-driven neural network prediction method for static stress on the surface of a runner blade comprises the following steps:

[0008] S1. Based on the drawings of the Francis turbine unit, a full flow channel model of the Francis turbine unit is established, the model is meshed, a set number of operating points are selected according to the water head and opening, and the meshes corresponding to the operating points are imported into the numerical simulation software for flow field calculation to obtain the surface pressure data of the runner blades.

[0009] S2. Based on the hydropower unit shaft system drawing, a hydropower unit shaft system model is established, and the model is divided into unstructured grids. The surface pressure data of the runner blades in step S1 is input into the interface between the fluid domain and the solid domain. The finite element method is used to perform static analysis to obtain the static stress data of the runner blade surface and its corresponding three-dimensional coordinates, and the static stress data is divided into a training set and a test set.

[0010] S3. Select a prediction model using the three-dimensional coordinates, water head and opening of any operating point, train the prediction model using the three-dimensional coordinates, water head and opening corresponding to the training set, input the three-dimensional coordinates, water head and opening of the operating point into the trained prediction model, and obtain the static stress on the surface of each impeller blade under the operating point.

[0011] Furthermore, in step S1, obtaining the runner blade surface pressure data includes the following sub-steps:

[0012] S101. Based on the drawings of the Francis turbine unit, a full flow channel model of the Francis turbine unit is established using SolidWorks software. The model includes a volute, fixed guide vanes, movable guide vanes, a runner, a pressure reducing pipe and a tailwater pipe.

[0013] S102. Use n grid number schemes to mesh the whole flow channel model of the Francis turbine unit. Under the same boundary conditions, perform steady-state fluid numerical simulation on the whole flow channel model of the Francis turbine unit based on the rated head and rated power to obtain the corresponding runner efficiency. Select the scheme with the highest runner efficiency and the least grid number as the final mesh division scheme, divide the runner into unstructured grids, and divide the rest into structured grids.

[0014] S103, based on the pressure head and the guide vane opening corresponding to different operating conditions, set the intersection of the equal head line and the equal efficiency line on the comprehensive characteristic curve of the whole flow channel of the Francis turbine unit as the head opening, and select a set number of operating points according to the head opening.

[0015] S104. Use ICEM software to re-grid the movable guide vanes with different openings under different operating conditions to obtain a hexahedral structure grid. Other components are not meshed, and all grids are assembled so that each operating point corresponds to a set of independent grids.

[0016] S105, using the fluent solver to perform computational fluid dynamics analysis on the grids in step S104, and obtain the surface pressure data of the runner blades, the specific content is:

[0017] Fluid motion satisfies the conservation of mass and energy, and the corresponding equation is shown in formula (7):

[0018]

[0019] Among them, ρ represents density, t represents time, and u represents streamline. represents the Hamiltonian operator, p represents the static pressure, v represents the kinematic viscosity, represents the Laplace operator, and f represents the mass force.

[0020] The Reynolds average method is used to time-average the Navier-Stokes equation to obtain the time-averaged Reynolds equation, which is expressed as follows:

[0021]

[0022] Among them, Δt represents the unit time, represents the time average value, and φ represents the instantaneous physical quantity.

[0023] Decomposing the instantaneous physical quantity, we get the instantaneous equation, the specific expression is:

[0024]

[0025] Wherein, φ′ represents the pulsation value.

[0026] By simplifying formula (9) using the Cartesian tensor, we can obtain the Reynolds time-averaged NS equations, which are expressed as follows:

[0027]

[0028] Among them, u i represents the velocity of the ith streamline, u j represents the jth streamline velocity, f i represents the i-th mass force, u i ′ represents the pulsation value of the i-th streamline velocity, u j ′ represents the pulsation value of the jth streamline velocity, Indicates u i ′ and u j′The time average of the product, (x i ,x j ) represents the spatial coordinates.

[0029] The Reynolds-averaged Navier-Stokes equation is closed using the Realizable k-ε transport equation. The specific expression is as shown in formula (11):

[0030]

[0031] in, k represents turbulent kinetic energy, ε represents the turbulent kinetic energy dissipation rate, μ t represents the turbulent viscosity, G k Indicates the turbulent kinetic energy generated by the average velocity gradient, G b represents the turbulent kinetic energy generated by buoyancy, Y M represents the part of the total dissipation rate of compressive turbulence caused by fluctuation expansion, C 1ε , C2, C 3ε Both represent constants, σ k , σ ε denote the turbulent Prandtl number of k and ε, respectively, and S k and S ε Both represent user-defined source terms; μ represents dynamic viscosity, and S represents a user-defined parameter.

[0032] The runner area is processed using the multiple reference system method, and then a steady calculation is performed. The Reynolds-averaged Navier-Stokes equations are solved using the SIMPLEC method to obtain the velocity and pressure distributions inside the flow field, which include the pressure data on the runner blade surface.

[0033] Furthermore, in step S2, obtaining the static stress data of the runner blade surface includes the following sub-steps:

[0034] S201. Based on the hydropower unit shaft system drawing, a hydropower unit shaft system model is established using SolidWorks software. The model includes a mixed flow runner, a thrust bearing, a generator rotor, an upper guide bearing, a water guide bearing, a lower guide bearing and a main shaft. The surface where the blade contacts the water is set as the interface between the fluid domain and the solid domain.

[0035] S202. Use static structure finite element software to divide the solid domain into unstructured grids, and encrypt the grids of the runner blades in combination with calculation accuracy.

[0036] S203. Constraints are set for the solid domain using the physical properties of the shaft system model of the hydropower unit. The water guide bearing, the upper guide bearing and the lower guide bearing are fixedly constrained to restrain radial vibrations, the thrust bearing is fixedly constrained to restrain axial vibrations, and the rotation speed and gravity acceleration are set.

[0037] S204, input the runner blade surface pressure data as load into the interface between the fluid domain and the solid domain, perform fluid-solid coupling calculation, and obtain the runner blade surface static stress data, the specific content is:

[0038] The control equation is used to make the stress and displacement between the fluid domain and the solid domain meet the conservation conditions. The specific formula is:

[0039]

[0040] Among them, τ f n f represents the fluid normal stress, τ s n s represents the solid normal stress, d f represents the fluid displacement, d s Represents the displacement of the solid.

[0041] The finite element analysis of the interface between the fluid domain and the solid domain was carried out using the one-way steady-state fluid-solid coupling method to obtain the static stress data of the runner blade surface and its corresponding three-dimensional coordinates.

[0042] Furthermore, in step S3, predicting the static stress on the surface of the runner blade includes the following sub-steps:

[0043] S301. Input the three-dimensional coordinates, water head and opening corresponding to the surface grid nodes of any blade of the runner at any operating point into each neural network model to obtain the predicted static stress data, compare the error between the predicted static stress data and the test set, and select the neural network model corresponding to the error below the set error as the prediction model. The model includes an input layer, a fully connected layer and an output layer.

[0044] S302, input the three-dimensional coordinates, water head and opening corresponding to the training set into the prediction model, and obtain weight and bias parameters through partial differential operation and nonlinear activation function processing.

[0045] S303, based on the weight and bias parameters in step S302, the prediction model is trained and parameterized using the back propagation algorithm and the gradient descent algorithm until the error between the predicted static stress data and the test set reaches below the set error threshold, and the training is stopped to obtain a trained prediction model.

[0046] S304, inputting the three-dimensional coordinates, water head and opening of the operating point into the trained prediction model to obtain the static stress on the surface of each runner blade at the operating point.

[0047] Furthermore, the present invention also proposes an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the data-driven neural network prediction method for static stress on the surface of a runner blade when executing the computer program.

[0048] Furthermore, the present invention also proposes a computer-readable storage medium, which stores a computer program. When the computer program is run by a processor, it executes the data-driven neural network prediction method for static stress on the surface of a runner blade as described above.

[0049] Compared with the prior art, the present invention adopts the above technical solution and has the following technical effects:

[0050] 1. The neural network used in the present invention has a rapid response in the prediction of static stress of turbine blades. The results can be output in just a few seconds, and can correctly reflect the static stress distribution trend of blades under all working conditions. The absolute error of the static stress on the runner blades at the test working point is basically controlled within 1MPa, and the relative error is basically controlled below 10%, and the accuracy meets engineering requirements.

[0051] 2. The method proposed in the present invention has good generalization performance and can make effective predictions for different operating points within the operating area, and can provide guidance and modification suggestions for the operating status of the overall hydropower unit. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 It is an overall implementation flow chart of the present invention.

[0053] Figure 2 It is a schematic diagram of the full flow channel model of the mixed flow turbine unit of the present invention.

[0054] Figure 3 It is a full flow passage grid schematic diagram of the full flow passage model of the Francis turbine unit of the present invention.

[0055] Figure 4 It is a schematic diagram of verifying the grid independence of the present invention.

[0056] Figure 5 It is a schematic diagram of the water head output distribution at the working point of the present invention.

[0057] Figure 6 It is a schematic diagram of the shaft system model of the hydropower unit of the present invention, constraint settings and solid domain mesh.

[0058] Figure 7 It is a schematic diagram of the fully connected neural network model architecture of the present invention.

[0059] Figure 8 It is a schematic diagram of the water head and opening distribution of the training working condition point and the selection of the test working condition point of the present invention.

[0060] Fig. 9 It is a schematic diagram comparing the static stress neural network prediction results and finite element calculation results of the pressure surface and suction surface of a single blade of the present invention.

[0061] Fig.10 It is a schematic diagram comparing the neural network prediction results and the finite element calculation results of the static stress of all blades of the present invention. DETAILED DESCRIPTION

[0062] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.

[0063] To achieve the above object, the present invention proposes a data-driven neural network prediction method for static stress on the surface of runner blades, such as Figure 1 As shown, the specific steps are as follows:

[0064] S1. Based on the drawings of the Francis turbine unit, the full flow channel model of the Francis turbine unit was established by SolidWorks software, and the model was meshed. After the meshing, 70 operating points were selected according to the water head opening, and the meshes corresponding to the 70 operating points were imported into the numerical simulation software for flow field calculation to obtain the surface pressure data of the runner blades; the specific contents are as follows:

[0065] S101, such as Figure 2 As shown, the full flow model of the Francis turbine unit includes a volute, fixed guide vanes, movable guide vanes, a runner, a pressure reducing pipe and a tailwater pipe. Figure 2 The various components of the full flow channel model of a Francis turbine unit are displayed, distinguished by different colors, showing the structure of the fluid calculation domain.

[0066] S102, such as Figure 3 As shown in the figure, the whole flow channel model of the Francis turbine unit is meshed, the runner is divided into unstructured grids, and the rest are divided into structured grids. The model is divided into many small units through meshing to ensure the normal numerical calculation and obtain accurate results. The completion of meshing is ready for selecting the optimal grid for mesh independence verification.

[0067] In fluid numerical simulation, the quality of the grid may directly affect the results of the flow field calculation. The number of grids has an equally important impact on the calculation. If the number of grids is too small, it will be difficult to capture the flow details or lose some features of the flow components. If the number of grids is too large, a large amount of computing resources will be required, which will slow down the calculation speed. The appropriate number of grids can ensure the simulation accuracy and shorter time consumption. At the same time, it is necessary to ensure that the grid has no effect on the correctness and accuracy of the numerical simulation, so grid independence verification is required.

[0068] like Figure 4 As shown in the figure, six schemes are used for meshing. The total number of meshes under the six schemes is 3.62 million, 4.98 million, 6.41 million, 7.84 million, 10.32 million and 13.5 million respectively. The six schemes use the same boundary conditions. Based on the rated head and rated power, the steady-state fluid numerical simulation of the whole flow channel model of the Francis turbine unit is carried out to obtain the corresponding runner efficiency. The meshing scheme with the highest efficiency and the least number of meshes is selected as the final meshing scheme. Figure 4 It can be seen that the fifth solution with 10.32 million grids has a moderate number of grids and high operating efficiency, so it is selected as the optimal grid for subsequent calculations.

[0069] S103, such as Figure 5 As shown in the figure, based on the pressure head and movable guide vane opening corresponding to different operating conditions, the intersection of the equal head line and the equal efficiency line on the comprehensive characteristic curve of the whole flow channel of the Francis turbine unit is set as the head opening, and 70 operating points are selected according to the head opening. Figure 5 The power and opening data corresponding to each operating condition are displayed, providing a basis for modifying the movable guide vane grid under different subsequent operating conditions. The operating points in the figure are typical operating points selected for the hydropower station.

[0070] S104. Since the operating condition of the whole flow channel of the mixed flow turbine unit needs to change the opening of the movable guide vane, and once the opening of the movable guide vane changes, the shape of the movable guide vane will change, and the mesh of the movable guide vane part needs to be re-divided. Therefore, the ICEM software is used to re-structure the mesh of the movable guide vanes with different openings under different operating conditions to obtain a hexahedral structure mesh. The other parts are not meshed, and all the meshes are assembled so that each different operating point corresponds to a set of independent meshes, totaling 70 sets of meshes.

[0071] S105. Use the fluent solver to perform CFD (Computational Fluid Dynamics) analysis on the 70 sets of grids in step S104 to obtain the runner blade surface pressure data. The specific content is:

[0072] Fluid motion must comply with the laws of conservation of mass, momentum, and energy. These conservation laws are applied to fluid motion, and after corresponding transformations, the NS (Navier-Stokes) equations of fluid mechanics are formed. In fluid mechanics, it is generally believed that when solving incompressible fluids, the energy conservation equation can be ignored due to the relatively small amount of heat exchange. Therefore, the motion of the fluid only needs to satisfy the conservation of mass and energy. The corresponding equations are shown in formula (13):

[0073] The expressions of the mass conservation equation and momentum conservation equation are:

[0074]

[0075] Among them, ρ represents density, t represents time, and u represents streamline. represents the Hamiltonian operator, p represents the static pressure, v represents the kinematic viscosity, represents the Laplace operator, and f represents the mass force.

[0076] When solving turbulent problems, since it is difficult to obtain an analytical solution to the Navier-Stokes equations describing turbulent flow, the Reynolds average method is used to regard turbulence as the superposition of two flows, and the Navier-Stokes equations are time-averaged to obtain the time-averaged Reynolds equations. The specific expression is:

[0077]

[0078] Among them, Δt represents the unit time, represents the time average value, and φ represents the instantaneous physical quantity.

[0079] Decomposing the instantaneous physical quantity, we get the instantaneous equation, the specific expression is:

[0080]

[0081] Wherein, φ′ represents the pulsation value.

[0082] By simplifying formula (15) using Cartesian tensor, we get the RANS (Reynolds Average Navier-Stokes, Reynolds time-averaged NS) equations, which are expressed as follows:

[0083]

[0084] Among them, u i represents the velocity of the ith streamline, u j represents the jth streamline velocity, f i represents the i-th mass force, u i ′ represents the pulsation value of the i-th streamline velocity, u j ′ represents the pulsation value of the jth streamline velocity, Indicates u i ′ and u j ′The time average of the product, (x i ,x j ) represents the spatial coordinates.

[0085] Since the equations are no longer closed, it is necessary to re-establish the connection between the time-averaged value and the pulsating value. Therefore, according to the actual situation, the appropriate turbulent transport equation is introduced as a new expression to close the RANS equations and then solve them. The Realizable k-ε transport equation is used to close the RANS equations. The specific expression is as follows:

[0086]

[0087] in, k represents turbulent kinetic energy, ε represents the turbulent kinetic energy dissipation rate, μ t represents the turbulent viscosity, G k Indicates the turbulent kinetic energy generated by the average velocity gradient, G b represents the turbulent kinetic energy generated by buoyancy, Y M represents the part of the total dissipation rate of compressible turbulence caused by fluctuation expansion, C 1ε , C2, C 3ε Both represent constants, σ k , σ ε denote the turbulent Prandtl number of k and ε, respectively, and S k and S ε Both represent user-defined source terms, μ represents dynamic viscosity, and S represents user-defined parameters.

[0088] The runner area is processed using the MRF (Multiple Reference Frame) method, and steady-state calculations are performed at the same time. The RANS equations are solved using the SIMPLEC method to obtain the velocity and pressure distribution inside the flow field, which includes the pressure data on the blade surface. These data are saved in the software in the form of a dat result file, and the operating software exports the blade surface pressure data in a txt file format for subsequent finite element analysis.

[0089] S2. Based on the shaft system drawing of the hydropower unit, a shaft system model of the hydropower unit is established using SolidWorks software, and the model is imported into the static structure software for unstructured meshing. The runner blade surface pressure data in step S1 is input into the interface between the fluid domain and the solid domain, and a static analysis is performed using the finite element method to obtain the static stress data on the runner blade surface and its corresponding three-dimensional coordinates, and the static stress data is divided into a training set and a test set; the specific contents are as follows:

[0090] S201. The shaft system model of the hydropower unit includes the Francis runner, thrust bearing, generator rotor, upper guide bearing, water guide bearing, lower guide bearing and main shaft. The surface where the blade contacts the water is set as the interface between the fluid domain and the solid domain.

[0091] S202. Use static structure finite element software to divide the solid domain into unstructured grids, and encrypt the grids of the runner blades in combination with calculation accuracy.

[0092] S203. For the shaft system model of the hydropower unit, the mixed flow impeller is the most important flow-passing component. The physical properties of the shaft system model of the hydropower unit are set (including material density, Young's modulus of the material, Poisson's ratio of the material, bulk elastic modulus of the material, shear elastic modulus of the material, tensile and compressive yield strength of the material, and tensile ultimate strength of the material), and constraints are set for the solid domain. The water guide bearing, upper guide bearing and lower guide bearing are fixed to constrain radial vibration, and the thrust bearing is fixed to constrain axial vibration. At the same time, the rotation speed and gravity acceleration are set.

[0093] Figure 6 This is a schematic diagram of the shaft system model of the hydropower unit, constraint settings, and solid domain mesh. Figure 6 (a) is the shaft system model and constraint setting of the hydropower unit. Figure 6 (b) is a schematic diagram of the solid domain mesh. It can be seen that the constraint setting can make the shaft system model of the hydropower unit closer to the actual operation situation, and the solid domain mesh divides the shaft system of the hydropower unit into sufficiently small units to facilitate subsequent fluid-solid coupling calculations.

[0094] S204, input the runner blade surface pressure data as load into the interface between the fluid domain and the solid domain, perform fluid-solid coupling calculation, and obtain the runner blade surface static stress data, the specific content is:

[0095] When performing fluid-solid coupling calculations, in order to make the calculation results accurate and reliable, it is necessary to ensure that the transfer of physical quantities between the fluid domain and the solid domain strictly follows the conservation law. In particular, the temperature, heat flow, stress, and displacement parameters of the interface between the fluid domain and the solid domain must remain consistent. Since the calculation of the present invention does not take into account the influence of heat exchange, it is only necessary to ensure that the stress and displacement meet the conservation conditions. The control equation is used to make the stress and displacement between the fluid domain and the solid domain meet the conservation conditions. The specific formula is:

[0096]

[0097] Among them, τ f n f represents the fluid normal stress, τ s n s represents the solid normal stress, d f represents the fluid displacement, d s Represents the displacement of the solid.

[0098] The finite element analysis of the interface between the fluid domain and the solid domain was carried out using the one-way steady-state fluid-solid coupling method to obtain the static stress data of the runner blade surface and its corresponding three-dimensional coordinates.

[0099] The static stress distribution is reflected in the visualization results of the static structure finite element software, and the static stress distribution on the blade surface and the static stress data on the runner blade surface at each grid point and their corresponding three-dimensional coordinates are derived.

[0100] S3. Select a prediction model using the three-dimensional coordinates, water head and opening of any operating point, train the prediction model using the three-dimensional coordinates, water head and opening corresponding to the training set, input the three-dimensional coordinates, water head and opening of the operating point into the trained prediction model, and obtain the static stress on the surface of each runner blade under the operating point; the specific content is:

[0101] S301, such as Figure 7 As shown, the three-dimensional coordinates (i.e., x1 to x5 are the three-dimensional coordinates x, y, z of each grid node on the specified blade), water head and opening corresponding to the surface grid nodes of any blade of the runner at any operating point are input into each neural network model respectively to obtain the predicted static stress data SS, and the error between the predicted static stress data and the test set is compared, and the neural network model corresponding to the error below 10% is selected as the prediction model. In this embodiment, a fully connected neural network model is selected as the prediction model, which includes an input layer, a fully connected layer and an output layer, wherein the lth layer is a fully connected layer (or a linear layer). There are many fully connected layers inside a fully connected neural network, which are briefly described here. The eth layer is the output layer. a eΩ represents the result of the nonlinear activation function operation corresponding to the Ωth neuron in the eth layer, ζ eΩ It represents the loss function operation result corresponding to the Ωth neuron in the eth layer, where a represents the nonlinear activation function and ζ represents the loss function.

[0102] S302, input the three-dimensional coordinates, water head and opening corresponding to the training set into the prediction model, and obtain weight and bias parameters through partial differential operation and nonlinear activation function processing.

[0103] S303, based on the weight and bias parameters in step S302, the prediction model is trained and parameterized using the back propagation algorithm and the gradient descent algorithm, and the hyperparameters such as the number of training rounds, the learning rate, and the batch size are adjusted through training until the error between the predicted static stress data and the test set reaches less than 0.0001, and the training is stopped to obtain a trained prediction model. The weight and bias parameters of the network are adjusted by minimizing the loss function, so that the model can fit the training data more accurately and has good generalization ability and nonlinear solving ability.

[0104] The back-propagation algorithm is based on the chain rule. It starts from the output layer and calculates the gradient of each layer's parameters layer by layer. First, the error of the output layer is calculated, that is, the partial derivative of the loss function with respect to the output of the output layer neurons. Then, by back-propagating these errors, the error of each layer of neurons is calculated, that is, the partial derivative of the loss function with respect to the output of the neurons in that layer; these errors are further used to calculate the gradient of the loss function with respect to the model parameters.

[0105] (1) Update various parameters:

[0106] For the output layer nodes, the error E is y e function, and y e Represents the output of the output layer, which is also the nonlinear activation function value net e Function, net e is the weight w le and bias b e function.

[0107]

[0108] Among them, γ represents the learning rate, Δw le represents the weight gradient from the fully connected layer to the output layer, Δb e represents the bias gradient of the output layer, w le represents the weight from the fully connected layer to the output layer, b e Represents the bias of the output layer.

[0109] (2) For fully connected layer nodes:

[0110]

[0111] Among them, Δw ol represents the weight gradient from the input layer to the fully connected layer, represents the derivative of the nonlinear activation function in the fully connected layer, M represents the number of nodes in each layer of the neural network, and w ql represents the weight from the qth node in the input layer to the fully connected layer, m q represents the qth training sample value, y q represents the qth output, Δθ l represents the bias gradient of the fully connected layer, G′(net q ) represents the derivative of the nonlinear activation function in the output layer, and xo represents the input of the input layer.

[0112] S304, inputting the three-dimensional coordinates, water head and opening of the operating point into the trained prediction model to obtain the static stress on the surface of each runner blade at the operating point.

[0113] In this embodiment, two operating points are selected at the center and two ends of the operating range, namely the first group of test operating points, to analyze the prediction performance of the neural network at both sides and the middle of the operating range; and four operating points are selected in the relatively open distribution of points, namely the second group of test operating points, to analyze the performance of the neural network in the location where the operating points are sparsely distributed. Figure 8 After selecting the test operating point, the test calculation can be performed to verify the prediction performance of the fully connected neural network.

[0114] The visualization is post-processed and rendered by tecplot to observe the differences between the learning results and the simulation results. Fig. 9 This is a schematic diagram comparing the static stress neural network prediction results and finite element calculation results of a single blade pressure surface and suction surface. Fig. 9 (a) is the prediction result of the neural network for the static stress on the pressure surface of a single blade. Fig. 9 (b) is the finite element calculation result of static stress on the pressure surface of a single blade. Fig. 9 (c) is the prediction result of the neural network for the static stress of the suction surface of a single blade. Fig. 9 (d) is the finite element calculation result of the static stress of the suction surface of a single blade. From the comparison, it can be seen that the fully connected neural network can make predictions that meet the accuracy requirements for the static stress of the suction surface and pressure surface of a single blade. Fig.10 This is a schematic diagram comparing the neural network prediction results and the finite element calculation results of the static stress of all blades. Fig.10 (a) is the prediction result of the neural network for the static stress of all blades. Fig.10 (b) is the finite element calculation result of the static stress of all blades. The static stress concentration area of ​​the runner blade is highlighted and enlarged to observe the prediction accuracy of the neural network. Fig.10 It can be seen that the fully connected neural network can predict the surface static stress of all runner blades with accuracy that meets the requirements.

[0115] The embodiment of the present invention further provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. It should be noted that when the processor executes the computer program, the specific steps of the method provided in the embodiment of the present invention are corresponding to the specific steps of the method provided in the embodiment of the present invention, and the processor has the functional modules and beneficial effects corresponding to the execution method. For technical details not described in detail in this embodiment, please refer to the method provided in the embodiment of the present invention.

[0116] The embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program. It should be noted that when the computer program is executed by the processor, it corresponds to the specific steps of the method provided in the embodiment of the present invention, and has the functional modules and beneficial effects corresponding to the execution method. For technical details not described in detail in this embodiment, please refer to the method provided in the embodiment of the present invention.

[0117] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A data-driven neural network prediction method for static stress on the surface of runner blades, characterized in that: include: S1. Based on the drawings of the Francis turbine unit, a full flow channel model of the Francis turbine unit is established, the model is meshed, a set number of operating points are selected according to the water head opening, and the meshes corresponding to the operating points are imported into the numerical simulation software for flow field calculation to obtain the surface pressure data of the runner blades; S2. Based on the shaft system drawing of the hydropower unit, a shaft system model of the hydropower unit is established, and the model is divided into unstructured grids. The surface pressure data of the runner blades in step S1 is input into the interface between the fluid domain and the solid domain, and a static analysis is performed using the finite element method to obtain the static stress data of the runner blade surface and its corresponding three-dimensional coordinates, and the static stress data is divided into a training set and a test set; S3. Select a prediction model using the three-dimensional coordinates and water head opening of any operating point, train the prediction model using the three-dimensional coordinates and water head opening corresponding to the training set, input the three-dimensional coordinates and water head opening of the operating point into the trained prediction model, and obtain the static stress on the surface of each impeller blade under the operating point.

2. The data-driven neural network prediction method for static stress on the surface of runner blades according to claim 1 is characterized in that: In step S1, obtaining the runner blade surface pressure data includes the following sub-steps: S101. Based on the drawings of the Francis turbine unit, a full flow channel model of the Francis turbine unit is established using SolidWorks software. The model includes a volute, fixed guide vanes, movable guide vanes, a runner, a pressure reducing pipe and a draft tube. S102, meshing the whole flow channel model of the Francis turbine unit using n grid number schemes, performing steady-state fluid numerical simulation on the whole flow channel model of the Francis turbine unit based on the rated head and rated power under the same boundary conditions, obtaining the corresponding runner efficiency, selecting the scheme with the highest runner efficiency and the least grid number as the final meshing scheme, dividing the runner into unstructured grids, and dividing the rest into structured grids; S103, based on the pressure head and the opening of the movable guide vane corresponding to different working conditions, the intersection of the equal head line and the equal efficiency line on the comprehensive characteristic curve of the whole flow passage of the Francis turbine unit is set as the head opening, and a set number of working condition points are selected according to the head and the opening; S104, using ICEM software to re-grid the movable guide vanes with different openings under different working conditions to obtain a hexahedral structure grid, and other components are not meshed, and all grids are assembled so that each working point corresponds to a set of independent grids; S105, using the fluent solver to perform computational fluid dynamics analysis on the grids in step S104, and obtain the surface pressure data of the runner blades, the specific content is: Fluid motion satisfies the conservation of mass and energy. The corresponding equations are shown in formula (1): Among them, ρ represents density, t represents time, and u represents streamline. represents the Hamiltonian operator, p represents the static pressure, v represents the kinematic viscosity, represents the Laplace operator, and f represents the mass force; The Reynolds average method is used to time-average the Navier-Stokes equation to obtain the time-averaged Reynolds equation, which is expressed as follows: Among them, Δt represents the unit time, represents the time average value, φ represents the instantaneous physical quantity; Decomposing the instantaneous physical quantity, we get the instantaneous equation, the specific expression is: Among them, φ′ represents the pulsation value; By simplifying formula (3) using the Cartesian tensor, we can obtain the Reynolds time-averaged NS equations, which are expressed as follows: Among them, u i represents the velocity of the ith streamline, u j represents the jth streamline velocity, f i represents the i-th mass force, u i ′ represents the pulsation value of the i-th streamline velocity, u j ′ represents the pulsation value of the jth streamline velocity, Indicates u i ′ and u j ′The time average of the product, (x i ,x j ) represents the spatial coordinate; The Reynolds-averaged Navier-Stokes equation is closed using the Realizable k-ε transport equation. The specific expression is as follows: in, k represents turbulent kinetic energy, ε represents the turbulent kinetic energy dissipation rate, μ t represents the turbulent viscosity, G k Indicates the turbulent kinetic energy generated by the average velocity gradient, G b represents the turbulent kinetic energy generated by buoyancy, Y M represents the part of the total dissipation rate of compressive turbulence caused by fluctuation expansion, C 1ε , C2, C 3ε Both represent constants, σ k , σ ε denote the turbulent Prandtl number of k and ε, respectively, and S k and S ε All represent user-defined source terms; μ represents dynamic viscosity, and S represents user-defined parameters; The runner area is processed using the multiple reference system method, and then a steady calculation is performed. The Reynolds-averaged Navier-Stokes equations are solved using the SIMPLEC method to obtain the velocity and pressure distributions inside the flow field, which include the pressure data on the runner blade surface.

3. The data-driven neural network prediction method for static stress on the surface of runner blades according to claim 1 is characterized in that: In step S2, obtaining the static stress data of the runner blade surface includes the following sub-steps: S201. Based on the hydropower unit shaft system drawing, a hydropower unit shaft system model is established using SolidWorks software. The model includes a Francis runner, a thrust bearing, a generator rotor, an upper guide bearing, a water guide bearing, a lower guide bearing and a main shaft. The surface where the blade contacts the water is set as the interface between the fluid domain and the solid domain. S202, using static structure finite element software to divide the solid domain into unstructured grids, and encrypting the grids of the runner blades in combination with calculation accuracy; S203, using the physical properties of the shaft system model of the hydropower unit to set constraints on the solid domain, the water guide bearing, the upper guide bearing and the lower guide bearing are fixedly constrained to restrain radial vibration, the thrust bearing is fixedly constrained to restrain axial vibration, and the rotation speed and gravity acceleration are set; S204, input the runner blade surface pressure data as load into the interface between the fluid domain and the solid domain, perform fluid-solid coupling calculation, and obtain the runner blade surface static stress data, the specific content is: The control equation is used to make the stress and displacement between the fluid domain and the solid domain meet the conservation conditions. The specific formula is: Among them, τ f n f represents the fluid normal stress, τ s n s represents the solid normal stress, d f represents the fluid displacement, d s represents the displacement of the solid; The finite element analysis of the interface between the fluid domain and the solid domain was carried out using the one-way steady-state fluid-solid coupling method to obtain the static stress data of the runner blade surface and its corresponding three-dimensional coordinates.

4. The data-driven neural network prediction method for static stress on the surface of runner blades according to claim 1, characterized in that: In step S3, predicting the static stress on the runner blade surface includes the following sub-steps: S301, inputting the three-dimensional coordinates, water head and opening corresponding to the surface grid nodes of any blade of the runner at any operating point into each neural network model, obtaining predicted static stress data, comparing the error between the predicted static stress data and the test set, and selecting the neural network model corresponding to the error below the set error as the prediction model, the model includes an input layer, a fully connected layer and an output layer; S302, inputting the three-dimensional coordinates, water head and opening corresponding to the training set into the prediction model, and obtaining weight and bias parameters through partial differential operation and nonlinear activation function processing; S303, based on the weight and bias parameters in step S302, the prediction model is trained and parameterized using the back propagation algorithm and the gradient descent algorithm, until the error between the predicted static stress data and the test set reaches below a set error threshold, the training is stopped, and a trained prediction model is obtained; S304, inputting the three-dimensional coordinates, water head and opening of the operating point into the trained prediction model to obtain the static stress on the surface of each runner blade at the operating point.

5. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the data-driven neural network prediction method for static stress on the surface of a runner blade are implemented as described in any one of claims 1 to 4.

6. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the data-driven neural network prediction method for static stress on the surface of a runner blade is executed as described in any one of claims 1 to 4.

Images