Water erosion damage prediction method, system and device for turbine blade

Abstract

The invention discloses a steam turbine blade water erosion damage prediction method, system and device, and relates to the technical field of water erosion damage prediction.The method comprises the steps that a three-dimensional blade and outer watershed geometric model is established, grid division is conducted on the geometric model, and a graph structure capable of expressing grid units is obtained; performing numerical simulation by adopting a graph structure to obtain a water erosion data training set; constructing a water erosion damage evaluation model of the blade, wherein the water erosion damage evaluation model comprises a flow field prediction network and a movable blade erosion rate prediction network; inputting a water erosion data training set into a water erosion damage evaluation model for training, and predicting a moving blade erosion rate caused by secondary water drop impact at different radial heights; according to the method, sufficient blade water erosion analysis results serve as sample data, the flow field liquid phase parameter distribution-based prediction network and the moving blade erosion rate prediction network are trained, rapid prediction from design parameters to the last-stage moving blade erosion rate is achieved, and meanwhile, the accuracy of moving blade erosion rate prediction is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of prediction of water erosion damage, and particularly to a method, a system and a device for predicting water erosion damage of steam turbine blades. Background Art

[0002] The last-stage blades of the low-pressure cylinders of conventional condensing steam turbines in power plants and the last-stage blades of nuclear power steam turbines operate in wet steam regions. As the steam in the cascade continuously expands and does work, after passing through the Wilson saturation line, steam condensation nuclei will occur and droplet growth will be accompanied. These small droplets will seriously disrupt the steam flow and reduce the efficiency of the steam turbine stage. Part of the droplets will deposit on the blade surface to form a water film, and the larger-sized secondary droplets formed by the tearing of the water film at the trailing edge of the blade by the steam flow will impact the surface of the moving blade at high speed, thereby causing water erosion damage. Therefore, researching a method for evaluating the degree of water erosion damage is of great significance for evaluating the water erosion risk of the blades in the low-pressure cylinder under different working conditions, improving the economic benefits of steam turbine generator sets, and ensuring the safe and stable operation of the units.

[0003] At present, most scholars judge the degree of water erosion damage of steam turbine blades by means of pure numerical simulation methods. For example, in the Journal of Power Engineering, a method for controlling the impact behavior of secondary water droplets by curved blades is proposed. By defining a material erosion model to simulate the erosion effect of water droplets impacting the blade, the influence of stator blade reverse bending on blade water erosion is further explored. In the Research on Erosion of the Last-stage Long Blades of the Low-pressure Cylinder of a Steam Turbine, numerical simulation methods are used in combination with a particle transport model to analyze the wet steam two-phase flow field of the last three stages of the low-pressure cylinder, and the position where the last-stage moving blade is most severely impacted by secondary water droplets is predicted.

[0004] However, in the above-mentioned prior art, the calculation process using pure numerical simulation methods is quite cumbersome, and when studying the water erosion characteristics under variable boundary conditions such as small volume flow, each working condition needs to be calculated separately, which greatly increases the time and energy costs of the research work. At the same time, some boundary conditions used in the numerical simulation process, such as parameters such as the size and distribution of inlet water droplets, are taken as range values by theory or experience, resulting in a low accuracy of the predicted erosion rate of the moving blade. Summary of the Invention

[0005] Aiming at the deficiencies of the prior art that the prediction efficiency using numerical simulation methods is slow and the accuracy of the predicted erosion rate of the moving blade is low, the present invention proposes a method, a system and a device for predicting water erosion damage of steam turbine blades. By training a flow field prediction network and a moving blade erosion rate prediction network based on depth graph convolution, rapid prediction from flow field condition control parameters to flow field parameters and then to the erosion rate of the last-stage moving blade is realized, thereby solving the problems existing in the prior art.

[0006] A method for predicting water erosion damage of steam turbine blades includes the following steps: Obtain the geometric parameters and operating condition control parameters of the steam turbine blade samples. Based on the geometric parameters, establish a geometric model including the three-dimensional blade surface and the outer flow field of the blade, and perform mesh division on the geometric model to obtain a graph structure that can represent the mesh elements. Use the graph structure for numerical simulation to obtain the liquid phase parameters at the moving and static blade cross-sections at different radial heights; the liquid phase parameters at the moving and static blade cross-sections include the number of primary droplets, droplet diameter, and droplet impact velocity; based on the number of primary droplets, droplet diameter, and droplet impact velocity, obtain the spatial distribution characteristics and motion distribution characteristics of secondary droplets; based on the spatial distribution characteristics and motion distribution characteristics of secondary droplets, obtain the erosion rate of the moving blade caused by secondary water droplet impacts at different radial heights. Use the operating condition control parameters of the blade samples as the input and the liquid phase parameters at the moving and static blade cross-sections as the output to train the neural network, and construct a flow field prediction network; use the output of the flow field prediction network as the input, use the spatial distribution characteristics and motion distribution characteristics of secondary droplets as the influencing parameters of the moving blade erosion rate, and use the moving blade erosion rate as the output to construct a moving blade erosion rate prediction network; establish a water erosion damage assessment model based on the flow field prediction network and the moving blade erosion rate prediction network. Input the actual operating condition control parameters of the steam turbine blade to be tested into the water erosion damage assessment model to predict the erosion rate of the moving blade caused by secondary water droplet impacts at different radial heights.

[0007] Further, obtaining the erosion rate of the moving blade caused by secondary water droplet impacts at different radial heights according to the spatial distribution characteristics and motion distribution characteristics of secondary droplets specifically includes the following steps: Use wet steam as the working medium to perform non-equilibrium condensation flow calculations on blade samples with different structures to obtain the liquid phase mass flow rate in the last-stage stator blade row. G ; According to the density of wet steam , the absolute velocity vector of wet steam , the absolute velocity vector of primary droplets generated by wet steam condensation , and the surface tension of primary droplets , use the Weber calculation formula to calculate the diameter of secondary droplets in different blade height regions d ; According to the liquid phase mass flow rate in the last-stage stator blade row, the humidity at the relative blade height, the volume and density of primary droplets, obtain the number of secondary droplets at different blade heights. M ; According to the impact velocity of secondary droplets, the diameter of secondary droplets, and the erosion resistance number of the moving blade material, calculate the volume damage of the material caused by a single droplet on the moving blade surface, and further obtain the moving blade erosion rate; among them, the circumferential velocity of the secondary droplets is the same as the circumferential velocity of the corresponding blade height region, and its radial velocity is the same as the mainstream radial component velocity.

[0008] Further, using the operating condition control parameters of the blade sample as the input and the liquid phase parameters of the stator and rotor blade cross-sections as the output to train the neural network, a flow field prediction network is constructed, which specifically includes the following steps: Set the initial boundary conditions according to the actual operating conditions of the blade; Analyze the liquid phase parameters of the stator-rotor blade cross-section. At the same blade height of this cross-section, select m points along the tangential direction, extract the liquid phase parameters at each data point, and then obtain the liquid phase parameters in the channel at the same radial height; Extract the liquid phase parameter values of the stator-rotor blade cross-sections at different radial heights at a total of k locations of this cross-section, and obtain the liquid phase parameter vector space of the stator-rotor blade cross-section; Based on the liquid phase parameter vector space of the stator-rotor blade cross-section, the operating condition control parameters of the blade sample, and the neural network parameters, establish a flow field prediction network.

[0009] Further, in the flow field prediction network and the moving blade erosion rate prediction network, three-layer graph convolutional layers are respectively used to extract the intermediate feature information of the flow field, and global pooling is used to aggregate the extracted feature information and then map it to the output space, and the prediction result is output through a fully connected layer.

[0010] Further, the mean square error between the predicted flow field of the flow field prediction network and the real flow field, and the mean square error between the moving blade erosion rate predicted by the moving blade erosion rate prediction network and the real moving blade erosion rate are respectively used as loss functions to optimize the flow field prediction network and the moving blade erosion rate prediction network.

[0011] The present invention further includes a steam turbine blade water erosion damage prediction system, including: An acquisition module, configured to acquire the geometric parameters and operating condition control parameters of the steam turbine blade sample, establish a geometric model including the three-dimensional blade surface and the outer flow domain of the blade according to the geometric parameters, and perform mesh division on the geometric model to obtain a graph structure that can represent the mesh cells; A training set construction module, configured to perform numerical simulation using the graph structure to obtain the liquid phase parameters of the stator-rotor blade cross-sections at different radial heights; the liquid phase parameters of the stator-rotor blade cross-section include the number of primary droplets, droplet diameter, and droplet impact velocity; according to the number of primary droplets, droplet diameter, and droplet impact velocity, obtain the spatial distribution characteristics and motion distribution characteristics of secondary droplets; according to the spatial distribution characteristics and motion distribution characteristics of secondary droplets, obtain the moving blade erosion rate caused by the impact of secondary water droplets at different radial heights; A model training module is used to train a neural network with the operating condition control parameters of blade samples as inputs and the liquid phase parameters of the stator and rotor blade sections as outputs, thereby constructing a flow field prediction network. Using the output of the flow field prediction network as an input, with the spatial distribution characteristics and motion distribution characteristics of secondary droplets as influencing parameters of the rotor blade erosion rate and the rotor blade erosion rate as an output, a rotor blade erosion rate prediction network is constructed. A water erosion damage assessment model is established based on the flow field prediction network and the rotor blade erosion rate prediction network. A prediction module is used to input the actual operating condition control parameters of the steam turbine blade to be measured into the water erosion damage assessment model to predict the rotor blade erosion rate caused by the impact of secondary water droplets at different radial heights.

[0012] The present invention further includes a computer device for predicting water erosion damage of steam turbine blades, comprising: a memory, a processor, and a computer program stored in the memory. When the processor executes the computer program, the steps of the method for predicting water erosion damage of the last-stage steam turbine blades are implemented.

[0013] The present invention further includes a readable storage medium storing a computer program, where the computer program includes program instructions. When the program instructions are executed by a processor, the steps of the method for predicting water erosion damage of the last-stage steam turbine blades are executed.

[0014] The present invention provides a method for predicting water erosion damage of steam turbine blades, having the following beneficial effects: The present invention uses a graph structure to represent the data in the grid, retaining the information in the original flow field data to the greatest extent and greatly reducing the influence of the grid structure on the flow field prediction. By training a flow field prediction network and a rotor blade erosion rate prediction network based on deep graph convolution, the flow field prediction network can predict the liquid phase parameters inside the entire flow field by inputting the operating condition parameters, and has a wide applicable range and high generality for the last-stage blade cascades of steam turbines under different operating conditions. Selecting the secondary droplet size and spatial distribution characteristics directly related to water erosion damage as intermediate quantities for prediction by the rotor blade erosion rate prediction network improves the correlation between the two, making the prediction of the rotor blade erosion rate more reliable. This method has a faster prediction speed compared to traditional numerical simulation methods that calculate through droplet deposition and droplet motion trajectories, and the accuracy of the rotor blade erosion rate is more precise. Description of the Drawings

[0015] Figure 1 It is a schematic diagram of the numerical calculation domain and boundary conditions of the last-stage blade in the embodiment of the present invention; Figure 2 It is a schematic diagram of the channel grid division of the calculation domain in the embodiment of the present invention; Figure 3 It is a schematic diagram of the method for representing the grid diagram in the embodiment of the present invention; Figure 4Schematic diagram of the parameter distribution of the cross-section between the stator and rotor blades in the embodiments of the present invention; Figure 5 Schematic diagram of the flow field prediction network in the embodiments of the present invention; Figure 6 Schematic diagram of the moving blade erosion rate prediction network in the embodiments of the present invention; Figure 7 Flow chart of the method for predicting the water erosion damage of steam turbine blades in the embodiments of the present invention. Detailed implementation manners

[0016] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments.

[0017] The present invention proposes a method for predicting the water erosion damage of steam turbine blades. This method uses the analysis results of the wet steam condensation flow in the sufficient last-stage blade cascade of the steam turbine as samples to train the flow field prediction network and the moving blade erosion rate prediction network based on deep graph convolution, so as to achieve rapid prediction from the flow field condition control parameters to the flow field parameters and then to the erosion rate of the last-stage moving blade. As Figure 7 shown, the method specifically includes the following steps: S1. Establish a geometric model including the three-dimensional blade surface and the outer flow field of the blade: Use n design parameters, such as blade installation angle, chord length, maximum thickness, etc., to constrain the blade geometric parameters, and adopt a parametric design method to generate a geometric model including the surface of the last-stage blade of the steam turbine and the outer flow field of the blade. The vector space of the blade geometric structure control parameters is: ; The aerodynamic outer flow field of the divided last-stage blade cascade is as Figure 1 shown. The left side is the inlet boundary, the right side is the outlet boundary, and the upper and lower sides are periodic symmetry boundaries, which are set to be rotationally periodic symmetric about the rotation axis of the turbomachine.

[0018] S2. Adaptive grid division and graph structure data representation: Perform structured hexahedral grid division on the geometric model, encrypt the grid near the wall surface and fillets, generate boundary layer grids and refine them to ensure the accuracy of local parameters in the aerodynamic analysis. Structured hexahedral grids or unstructured tetrahedral grids can be used inside the channel according to the structural needs, and at the same time, grid independence verification is required to ensure that the calculation is not affected by the number of grids. Adaptive grid division can use structured grids or unstructured grids, and use graph structure to represent the data in the grid, which greatly reduces the influence degree of the grid structure on the flow field prediction. The three-dimensional blade surface and outer flow field grids obtained by division are as Figure 2 shown, and structured grids are used for division.

[0019] The flow field information obtained using adaptive grid calculation is stored in grid nodes. To perform aerodynamic flow field prediction using deep learning, these flow field data need to be converted into matrix form. Using a graph structure can flexibly represent the topological structure of grid cells, and boundary conditions and initial conditions are processed as node feature vectors, maximizing the retention of information in the original flow field data. Taking X as a graph node, each graph node X i includes cell center coordinates, Mach number, temperature, etc. n feature dimensions, E represents an edge, and each edge connects two graph nodes . The mathematical expression of the grid graph structure is: ; .

[0020] S3. Establishment and integration of water erosion data training samples; The training samples should include all data from blade geometric parameters and aerodynamic design parameters to flow field parameter distribution and then to the water erosion characteristic variable matrix. The establishment process includes the following specific steps: a) First, use the Latin hypercube sampling technique to randomly sample from multiple blade geometric parameter distribution spaces to obtain N sets of last-stage turbine blade profiles that meet the design requirements. Then, establish the outer flow field for aerodynamic calculation using the methods of the first and second steps and complete the grid division.

[0021] b) Secondly, using wet steam as the working medium, perform non-equilibrium condensation flow calculations on samples with different structures to obtain the liquid-phase mass flow rate G inside the last-stage stator blade row. The liquid droplets generated by the condensation of wet steam in the channel are primary droplets with relatively small diameters, usually less than 1 μm in diameter, and have little impact on water erosion. The secondary droplets that cause greater water erosion damage to the last-stage rotor blades are generally generated by the tearing of the water film on the surface of the stator blades. Research shows that the share of secondary water droplets in the last-stage stator blade row is about 10%. Considering that the radial velocity of the secondary droplets is greatly affected by centrifugal force after entering the rotor blade channel due to their larger diameter, it is necessary to correct the radial distribution of the secondary droplets.

[0022] The average humidity ratio at different relative blade heights is defined as: ; In the formula, represents the humidity at the relative blade height, represents the average humidity.

[0023] The empirical formula for the radial distribution of secondary droplets at the inlet of the last-stage stator blade considering the action of centrifugal force is: , ; , ; In the formula, L represents the relative blade height; For the secondary water droplets just detached from the trailing edge, the axial initial velocity is very small, taking 3 m / s. The circumferential velocity is the same as the circumferential velocity of the corresponding blade height region, and the radial velocity is taken as the same as the radial component velocity of the mainstream. The diameters of the secondary water droplets in different blade height regions d can be obtained from the Weber calculation formula: ; In the formula, represents the Weber number, represents the steam density, represents the absolute velocity vector of the steam, represents the absolute velocity vector of the water droplet, represents the surface tension of the water droplet.

[0024] The number of secondary droplets at different blade heights is defined as: ; In the formula, represents the volume of a single droplet, represents the droplet density.

[0025] The volume damage of the material caused by a single droplet on the moving blade surface is defined as: ; In the formula, is the droplet impact velocity, d is the droplet diameter, is the erosion resistance number of the material, which is related to the Vickers hardness of the material.

[0026] From the above analysis, it can be seen that the erosion rate of the last-stage moving blade is only related to 3 variables, namely the number of secondary droplets M , the droplet diameter d and the droplet impact velocity V .

[0027] c) Finally, integrate the results of N sample data as the training sets for the flow field prediction network and the moving blade erosion rate prediction network; The cross-section between the moving and stationary blades is a specific flow analysis cross-section in the steam turbine, located in the transition region between the rotating moving blades (rotor blades) and the stationary stationary blades (guide vanes / nozzles).

[0028] S4. Establish a flow field prediction network. The output parameters of the flow field prediction network are selected according to actual conditions. For the factors affecting the secondary droplet distribution in the last stage passage of the steam turbine, they should include humidity, droplet diameter, and droplet absolute velocity. To enhance the representation ability and learning ability of the network, different activation functions can be introduced to add non-linear elements to the input of the neural network and enhance the robustness of the training process.

[0029] Set the initial boundary conditions according to the actual operating conditions of the blade. The operating condition parameters are denoted as θ, including inlet pressure, temperature, outlet pressure, etc., totaling j variables. The vector space representation of the operating condition parameters is: ; Post-process the calculation results of the non-equilibrium condensation flow in S3. Analyze the liquid phase parameters of a certain cross-section between the stator and rotor blades. Along the tangential direction, m points are selected at the same blade height of this cross-section to extract the liquid phase parameters at each data point. Then, the liquid phase parameters in the passage at the same radial height are represented as: ; Extract the liquid phase parameters at a total of k different radial heights on this cross-section. Therefore, the vector space representation of the liquid phase parameters of the cross-section between the stator and rotor blades is: ; In the formula, represents the liquid phase parameter value of the cross-section between the stator and rotor blades. The superscript represents different radial height positions, and the subscript represents different types of liquid phase parameters; The input variable of the flow field prediction network is the operating condition control parameter of the to-be-solved last stage flow field, and the output is the liquid phase parameter distribution of the cross-section between the stator and rotor blades. Therefore, the flow field prediction network is represented as: ; In the formula, represents the flow field predicted by the flow field prediction network, represents the input parameters of the flow field prediction network, i.e., the operating condition parameters, represents the parameters of the flow field prediction network, represents the flow field prediction network.

[0030] The flow field prediction network predicts the distribution of flow field parameters through the input operating condition control parameters. Through multiple convolutions, it can effectively capture feature information of different scales in the flow field. The stochastic gradient descent algorithm is used to optimize the weight parameters to improve the prediction accuracy. In the flow field prediction network, three graph convolutional layers are set up to train the graph structure. Through multiple convolution operations, the intermediate feature information of the flow field can be mapped to the physical field feature information on the nodes. The parameters of the convolutional layers from the first layer to the third layer are set to 56*56, 56*112, and 112*224 respectively. The operations within the specific network layers also include a 3*3 convolutional kernel, a pooling layer, an activation layer, etc. The ReLU function is used in the activation layer. The propagation method between layers of the graph convolutional neural network is as follows: ; where is the feature representation of the l -th layer graph structure, is the graph node adjacency matrix, is 's degree matrix, is the node feature of the network, is the weight matrix, is the non-linear activation function ReLU.

[0031] is to integrate the features extracted by the graph convolutional layer and finally map them to the output space. A fully connected layer needs to be added after the max pooling layer to map the features obtained by convolution to a higher output dimension. Therefore, three fully connected layers are added after the max pooling layer, and the output parameters of the final fully connected layer are 10*1. Considering that too many parameters in the fully connected layer are likely to cause overfitting problems, regularization can be introduced, such as adding Dropout between the fully connected layers to alleviate it. Dropout is set to 0.2 in this model.

[0032] S5. Establish a moving blade erosion rate prediction network. Select the liquid phase parameters of the flow field predicted by the flow field prediction network as the input of the moving blade erosion rate prediction network. The size and motion characteristics of the secondary droplets are the direct influencing parameters of the moving blade erosion rate and have the strongest correlation with the moving blade erosion rate, which improves the credibility of the erosion rate prediction. The moving blade erosion rate prediction network realizes the prediction from the liquid phase parameters at the cross-section between the stationary and moving blades to the erosion rate of the secondary droplets impinging on the moving blade.

[0033] The input parameters of the moving blade erosion rate prediction network are the prediction results of the flow field prediction network, that is , which is expressed by the following formula: ; The output parameters of the moving blade erosion rate prediction network are the erosion rates of the moving blade caused by the impact of secondary water droplets at different radial heights: ; In the formula, Indicates the moving blade erosion rate.

[0034] The moving blade erosion rate prediction network is expressed by the following formula: ; In the formula, Indicates the value of the moving blade erosion rate obtained through the moving blade erosion rate prediction network, Indicates the parameters of the moving blade erosion rate prediction network.

[0035] The moving blade erosion rate prediction network receives the output parameters of the flow field prediction network and further performs feature extraction. A global pooling aggregation output node feature can be added at the end of the model. The output of the final fully connected layer is the predicted moving blade erosion rate. The method of predicting the moving blade erosion rate at the last stage by predicting the distribution characteristics of the secondary droplet liquid phase parameters in the flow field only needs to predict the size and motion parameters of the secondary droplets, and simply calculates according to the droplet impact material damage formula to obtain the moving blade erosion rate. This method greatly reduces the problem of many parameters to be predicted when directly predicting the moving blade erosion rate, reduces the output variables of the moving blade erosion rate prediction network, reduces the complexity of the network, and improves the training speed of the network.

[0036] S6. Co-training and application of the two networks.

[0037] The training process of the neural network requires a large number of data samples. In this process, the deep learning model continuously adjusts the weights and biases of the neural network through the optimization algorithm, so that the value of the loss function continuously decreases, and finally reaches the goal that the network output result is close to the real result. A larger learning rate can be set at the beginning of the training to improve the solution efficiency, and then the learning rate is gradually reduced according to the residual convergence situation to refine the parameter adjustment. In the initial stage of training, in order to ensure the residual convergence, the two networks are trained separately to improve the prediction accuracy. The input and output parameters of the depth map convolutional neural network are represented by a regular spatial structure, so that the last-stage cascades of steam turbines with different structures and operating conditions can be represented by a regular data structure to directly participate in the training and prediction of the neural network.

[0038] The loss function of the flow field prediction network is defined as the mean square error between the predicted flow field and the real flow field, and is defined by the following formula: ; The loss function of the moving blade erosion rate prediction network is defined as the mean square error between the predicted moving blade erosion rate and the real moving blade erosion rate, and is defined by the following formula: ; After the convergence of the two network residuals, the output result of the flow field prediction network is used as the input of the moving blade erosion rate prediction network for collaborative training. After the training is completed, for any last-stage steam turbine blade, it is possible to achieve rapid prediction from the blade operating conditions parameters to the liquid-phase parameters of the flow field and then to the moving blade erosion rate.

[0039] The present invention predicts the water erosion rate on the surface of the moving blade by predicting the distribution parameters of the flow field of the last-stage cascade through a deep graph convolutional neural network. This method has a faster prediction speed compared with the traditional numerical simulation method that simultaneously calculates droplet deposition and droplet movement trajectories, more accurate prediction of droplet size and distribution characteristics, and high prediction efficiency. The present invention includes two networks. The flow field prediction network can predict the liquid-phase parameters inside the entire flow field by inputting the operating conditions parameters, and has a wide applicable range and high generality for the last-stage cascades of steam turbines under different operating conditions. Selecting the secondary droplet size and spatial distribution characteristics directly related to water erosion damage as the intermediate quantity for prediction improves the correlation between them, making the prediction of the moving blade erosion rate more reliable.

[0040] Based on the same inventive concept, the present invention also proposes a steam turbine blade water erosion damage prediction system, including: An acquisition module, configured to acquire the geometric parameters and operating condition control parameters of a steam turbine blade sample, establish a geometric model including the three-dimensional blade surface and the outer flow domain of the blade according to the geometric parameters, and perform mesh division on the geometric model to obtain a graph structure that can represent grid cells.

[0041] A training set construction module, configured to perform numerical simulation using the graph structure to obtain the liquid-phase parameters at the moving and stationary blade interface at different radial heights; the liquid-phase parameters at the moving and stationary blade interface include the number of primary droplets, droplet diameter, and droplet impact velocity; according to the number of primary droplets, droplet diameter, and droplet impact velocity, obtain the spatial distribution characteristics and motion distribution characteristics of secondary droplets; obtain the moving blade erosion rate caused by the impact of secondary water droplets at different radial heights according to the spatial distribution characteristics and motion distribution characteristics of secondary droplets.

[0042] A model training module, configured to use the operating condition control parameters of the blade sample as the input, use the liquid-phase parameters at the moving and stationary blade cross-sections as the output to train the neural network, and construct a flow field prediction network; use the output of the flow field prediction network as the input, use the spatial distribution characteristics and motion distribution characteristics of secondary droplets as the influencing parameters of the moving blade erosion rate, and use the moving blade erosion rate as the output to construct a moving blade erosion rate prediction network; establish a water erosion damage assessment model according to the flow field prediction network and the moving blade erosion rate prediction network.

[0043] A prediction module, configured to input the actual operating condition control parameters of the steam turbine blade to be measured into the water erosion damage assessment model to predict the moving blade erosion rate caused by the impact of secondary water droplets at different radial heights.

[0044] The present invention also provides a computer device for predicting water erosion damage of steam turbine blades, comprising: a memory, a processor, and a computer program stored in the memory. When the processor executes the computer program, the steps of the method for predicting water erosion damage of the last-stage blades of a steam turbine are implemented.

[0045] The present invention also provides a readable storage medium storing a computer program, the computer program including program instructions which, when executed by a processor, are used to execute the steps of the method for predicting water erosion damage of the last-stage blades of a steam turbine.

[0046] As mentioned above, the above are only the preferred specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, making equivalent substitutions or changes should be covered by the protection scope of the present invention.

Claims

1. A method for predicting water erosion damage of steam turbine blades, characterized in that: The following steps are involved: The geometric parameters and operating control parameters of the steam turbine blade sample are obtained, a geometric model including a three-dimensional blade surface and a flow field outside the blade is established according to the geometric parameters, and the geometric model is meshed to obtain a graph structure capable of representing mesh units; Numerical simulation is performed using a graph structure to obtain liquid phase parameters of the cross section between the moving and stationary blades at different radial heights; the liquid phase parameters of the cross section between the moving and stationary blades include the number of primary droplets, the droplet diameter, and the droplet impact velocity; the spatial distribution characteristics and motion distribution characteristics of the secondary droplets are obtained based on the number of primary droplets, the droplet diameter, and the droplet impact velocity; the moving blade erosion rate caused by the secondary water droplet impact at different radial heights is obtained based on the spatial distribution characteristics and motion distribution characteristics of the secondary droplets; The working condition control parameters of the blade samples are used as input, and the liquid phase parameters of the moving and stationary blade sections are used as output to train the neural network and construct a flow field prediction network. The output of the flow field prediction network is used as input, the spatial distribution characteristics and motion distribution characteristics of the secondary droplets are used as the influencing parameters of the moving blade erosion rate, and the moving blade erosion rate is used as output to construct a moving blade erosion rate prediction network; a water erosion damage assessment model is established based on the flow field prediction network and the moving blade erosion rate prediction network; The actual operating control parameters of the turbine blade to be tested are input into the water erosion damage assessment model to predict the blade erosion rate caused by secondary water droplet impact at different radial heights.

2. A method for predicting water erosion damage of steam turbine blades according to claim 1, characterized in that: The method of obtaining the moving blade erosion rate caused by the impact of secondary water droplets at different radial heights according to the spatial distribution characteristics and motion distribution characteristics of the secondary droplets specifically includes the following steps: Using wet steam as the working fluid, non-equilibrium condensation flow calculations were performed on blade samples with different structures to obtain the liquid mass flow rate in the last-stage stationary blade cascade. G ; According to the density of wet steam , the absolute velocity vector of wet steam , the absolute velocity vector of the primary droplet generated by the condensation of wet steam and the surface tension of the primary droplet , the diameter of the secondary droplets in different blade height areas is calculated using the Weber calculation formula d ; According to the liquid mass flow rate in the last stage static blade cascade, the relative humidity at the blade height, the volume of the primary droplets and the primary droplet density, the number of secondary droplets at different blade heights is obtained. M ; According to the secondary droplet impact velocity, the secondary droplet diameter and the erosion resistance number of the moving blade material, the material volume damage caused by a single droplet on the moving blade surface is calculated, and then the moving blade erosion rate is obtained; wherein, the circumferential velocity of the secondary droplet is the same as the circumferential velocity of the corresponding blade height area, and its radial velocity is the same as the radial component velocity of the mainstream.

3. The method for predicting water erosion damage of steam turbine blades according to claim 1, characterized in that: The method uses the working condition control parameters of the blade sample as input and the liquid phase parameters of the moving and stationary blade sections as output to train the neural network and construct a flow field prediction network, which specifically includes the following steps: Set initial boundary conditions according to the actual operating conditions of the blades; The liquid phase parameters of the section between the moving and stationary blades are analyzed. At the same blade height in the section, the liquid phase parameters of the moving and stationary blades are analyzed. m points, extract the liquid phase parameters at each data point, and then obtain the liquid phase parameters in the channel at the same radial height; The total amount on this section k Liquid phase parameter values ​​of the section between the moving and stationary blades at different radial heights Extraction is performed to obtain the liquid phase parameter vector space of the cross section between the moving and stationary blades; According to the liquid phase parameter vector space of the cross section between the moving and stationary blades, the operating control parameters of the blade samples and the neural network parameters, the outflow field prediction network is established.

4. A method for predicting water erosion damage of steam turbine blades according to claim 1, characterized in that: The flow field prediction network and the moving blade erosion rate prediction network respectively use three-layer graph convolution layers to extract intermediate feature information of the flow field, and use global pooling to aggregate the extracted feature information and then map it to the output space, and output the prediction results through the fully connected layer.

5. The method for predicting water erosion damage of steam turbine blades according to claim 1, characterized in that: It also includes using the mean square error between the predicted flow field of the flow field prediction network and the actual flow field, and the mean square error between the predicted blade erosion rate of the blade erosion rate prediction network and the actual blade erosion rate as loss functions, respectively, to optimize the flow field prediction network and the blade erosion rate prediction network.

6. A turbine blade erosion damage prediction system, characterized in that: include: An acquisition module is used to acquire geometric parameters and operating control parameters of a steam turbine blade sample, establish a geometric model including a three-dimensional blade surface and a flow field outside the blade according to the geometric parameters, and mesh the geometric model to obtain a graph structure capable of representing mesh units; A training set construction module is used to perform numerical simulation using a graph structure to obtain liquid phase parameters of the cross section between the moving and stationary blades at different radial heights; the liquid phase parameters of the cross section between the moving and stationary blades include the number of primary droplets, the droplet diameter, and the droplet impact velocity; the spatial distribution characteristics and motion distribution characteristics of the secondary droplets are obtained based on the number of primary droplets, the droplet diameter, and the droplet impact velocity; the moving blade erosion rate caused by the secondary water droplet impact at different radial heights is obtained based on the spatial distribution characteristics and motion distribution characteristics of the secondary droplets; A model training module is used to train the neural network using the operating control parameters of the blade sample as input and the liquid phase parameters of the moving and stationary blade sections as output to construct a flow field prediction network; The output of the flow field prediction network is used as input, the spatial distribution characteristics and motion distribution characteristics of the secondary droplets are used as the influencing parameters of the moving blade erosion rate, and the moving blade erosion rate is used as output to construct a moving blade erosion rate prediction network; a water erosion damage assessment model is established based on the flow field prediction network and the moving blade erosion rate prediction network; The prediction module is used to input the actual operating control parameters of the turbine blade to be tested into the erosion damage assessment model to predict the erosion rate of the moving blade caused by the secondary water drop impact at different radial heights.

7. A computer device for predicting water erosion damage of the last stage blade of a steam turbine, characterized in that: include: A memory, a processor and a computer program stored in the memory, wherein when the processor executes the computer program, the steps of the method for predicting erosion damage of a last-stage blade of a steam turbine according to any one of claims 1 to 5 are implemented.

8. A readable storage medium, characterized in that: The readable storage medium stores a computer program, which includes program instructions. When the program instructions are executed by a processor, they are used to execute the steps of the method for predicting erosion damage of a last-stage blade of a steam turbine according to any one of claims 1 to 5.

Images