A multi-point based stress field reconstruction and test verification method for blade root

Abstract

The application discloses a kind of based on the stress field reconstruction and test verification method of blade root of multiple measuring points, comprising the following steps: blade root parameter sampling, modeling and carrying out blade root rim strength analysis;Field reconstruction model is constructed, and the input and output of model are divided, and the reconstruction model training is completed using data;Develop blade root rim tensile test, compare the numerical value measured by test with reconstruction stress value, and verify the reliability of reconstruction model.The application can obtain the stress field of blade root rim position according to input after the establishment of reconstruction model, complete strength analysis, effectively avoid the problems such as long analysis time and large calculation resource consumption in traditional finite element method, with the advantages of fast calculation speed and high precision, and can realize digital twin in industrial scene.The application verifies reconstruction stress field using blade root rim tensile test result, so that reconstruction model result is more convincing.

Description

Technical Field

[0001] This invention relates to the field of turbine blades, and in particular to a method for reconstructing and experimentally verifying the blade root stress field based on multiple measuring points. Background Technology

[0002] As a crucial component for energy conversion in turbine machinery, the strength and reliability of blades have always been a focus of attention. Blades typically operate under extremely harsh environments characterized by high temperature, high pressure, and high speed. Furthermore, deviations during design, manufacturing, installation, and operation are difficult to avoid, all of which contribute to the highly complex stress conditions experienced by blades during operation. The blade root-rim, as the primary load-bearing structure of the blade, usually demands even higher design standards for strength and reliability.

[0003] The finite element method (FEM) is currently a commonly used method for strength analysis of blade root rims. However, despite its long development, it still has some limitations. The results of the FEM analysis are greatly affected by the mesh generation. Existing automatic mesh generation algorithms can achieve high-quality meshes for relatively simple models, but for complex models, obtaining high-quality meshes often requires manual mesh generation, leading to a lengthy overall analysis process. Furthermore, increasing the number of meshes not only exponentially increases the duration of the FEM analysis but also places higher demands on computational resources. Summary of the Invention

[0004] To address the problems of the aforementioned finite element strength analysis method, this invention provides a method for reconstructing and experimentally verifying the blade root stress field based on multiple measurement points. This method, based on a neural network algorithm, can reconstruct the entire stress field at the blade root rim using stress values ​​from several points on the rim, enabling strength analysis of key components. Furthermore, a tensile test method for the blade root rim is proposed to verify the accuracy of the reconstructed stress field.

[0005] The present invention is achieved using the following technical solution:

[0006] A method for reconstructing and experimentally verifying the leaf root stress field based on multiple measurement points includes:

[0007] First, a sample set of blade root parameters is obtained, and a blade root rim model is established based on this. The stress field is then simulated and calculated using the finite element method. Second, a field reconstruction model is established, with the stress values ​​at key points of the blade root rim as input and the overall stress field as output. The model is trained using the input and output data. Finally, a tensile test is conducted on the blade root rim to obtain the stress and stress field values ​​at the measuring points. The reconstructed stress field is then compared with the calibrated test results to verify the reliability of the reconstruction method.

[0008] A further improvement of the present invention is that the method specifically includes the following steps:

[0009] Step 1: Use the Latin hypercube sampling method to sample the leaf root parameter sample set;

[0010] Step 2: Model the blade root and rim sections based on the blade root parameter samples obtained in Step 1;

[0011] Step 3: Calculate the above blade root rim model using the finite element method to obtain the stress distribution;

[0012] Step 4: Based on the stress distribution in Step 3, divide the input and output of the reconstruction model, determine the basic structure and parameters of the field reconstruction model, and complete the training of the reconstruction model according to the input and output.

[0013] Step 5: Conduct tensile tests on the blade root rim to obtain stress values ​​at key locations and stress field distribution at the blade root rim, and verify the reliability of the field reconstruction results.

[0014] A further improvement of this invention is that, in step 1, the specific implementation method is as follows:

[0015] During the sampling process, the number of additional sample points N is first determined. Since each sample point contains m parameters, and each parameter has a specific range of values, N values ​​are extracted stratified within the specific range of each parameter. For the first pair of tooth curvature parameter h at the leaf root... l The range of values ​​is [l min ,l max This range of values ​​is evenly divided into N sample spaces, each with a length of [missing information]. This creates N equally probable partitions; then, a value is randomly selected from each of these equally probable partitions to complete the adjustment of parameter h. l The selection of representative values ​​for each partition; when combining samples, the representative values ​​of each parameter partition are rearranged, and the i-th representative value of all parameters is selected in turn to form the i-th sample point; after sampling, the sample set contains a total of N sample points, and the data dimension is an N×m matrix. The size of N is selected according to the training data requirements of the reconstructed model.

[0016] A further improvement of this invention is that, in step 2, the specific implementation method is as follows:

[0017] Each parameter sample point contains information about the distance or angle between adjacent points on the outer contour of the blade root. The blade root is modeled by connecting and stretching the points with straight lines or arcs. The rim is modeled based on the blade root profile and positioning information. Then, the blade length is calculated based on the overall radial length of the blade and the radial position of the blade root. A cuboid of the corresponding length is added to the blade root to replace the actual blade body, generating an equivalent centrifugal force. The final blade root and rim assembly structure is used for finite element calculation.

[0018] A further improvement of this invention is that, in step 4, the specific implementation method is as follows:

[0019] The input needs to include the stress values ​​at key locations on the blade root rim; the output is a series of stress values ​​on the blade root rim, with the number K selected to ensure that these stress values ​​can reflect the stress field of the blade root rim, and strength analysis is performed accordingly; in the subsequent tensile test of the blade root rim, it is ensured that measuring points are arranged at key locations to obtain K stress values.

[0020] A further improvement of this invention is that, in step 4, the reconstruction model used is a neural network model based on a two-dimensional deconvolution algorithm, specifically including one fully connected layer and multiple deconvolution layers; the dimensions of the input and output are [N,1,1,L] and [N,1,H,W], respectively, where N is the number of sample points in the sample set, L is the number of input stress values, and the number of stress values ​​in the output reconstructed stress field is H×W; the batch size and the size and stride of the deconvolution kernel in each layer are selected according to the specific sizes of N, L, H, and W, satisfying that the kernel size is smaller than the input size, and the stride is less than or equal to the kernel size in the corresponding dimension; the input is first converted by the fully connected layer. A suitable two-dimensional model is used, with a single input channel. The input is then passed through multiple deconvolutional layers, and the output is the reconstructed stress field. The loss function is the cross-entropy loss function, and the accuracy evaluation criterion is the mean absolute deviation. The input and its corresponding output are divided into training and test sets in an 8:2 ratio. The initial learning rate is set to a range of 0.00001 to 0.001 based on the model size and sample size, and is dynamically adjusted during training using a stochastic gradient descent optimizer. When the accuracy of the reconstructed results from 90% of the inputs in the test set is greater than 90% of the actual output, the reconstructed model is considered to have converged, and the training process ends.

[0021] A further improvement of this invention is that, in step 5, the blade root rim tensile test specifically includes the following steps:

[0022] Step 5.1: Randomly select a parameter sample point from the validation set of the training data, and obtain the blade root rim model accordingly;

[0023] Step 5.2: Model the model from Step 5.1;

[0024] Step 5.3: Strain gauges are placed at key locations on one side of the modeled blade root rim, and paint is sprayed on the other side to form a speckle surface;

[0025] Step 5.4: Perform a tensile test on the blade root rim, and simultaneously obtain the strain values ​​of the strain gauges and the overall strain distribution measured by the non-contact speckle full-field strain gauge. Then, combine the strain-stress relationship σ=Eε to convert the obtained strain into stress.

[0026] A further improvement of the present invention is that, in step 5.2, the modeling principle is: (1) the test piece and the prototype are geometrically similar and similar in stress conditions; (2) a predetermined number of strain gauges are arranged in the blade root throat; (3) the size is reduced with reference to the load capacity of the tensile testing machine used in the test.

[0027] A further improvement of this invention is that the results obtained by strain gauge measurement are contact measurement results, while the results obtained by non-contact speckle full-field strain gauge are non-contact measurement results; the non-contact measurement results are calibrated based on the results obtained by contact measurement to obtain an accurate stress field distribution.

[0028] A further improvement of this invention lies in the fact that the calibration method involves extracting the values ​​of the strain gauges at corresponding positions from the non-contact measurement results and establishing a mapping between non-contact and contact measurement results at corresponding positions.

[0029] ε 接触 =k*ε 非接触 +b

[0030] Where k is the mapping coefficient and b is the offset;

[0031] Map all non-contact measurement results to obtain the overall stress field of the blade root rim using the above formula; compare the calibrated stress field with the results reconstructed by the model to verify the accuracy of the stress field reconstruction results; if the error exceeds the engineering acceptable range, i.e. the relative error exceeds 10%, repeat steps 1-4, increase the number of parameter sample points in the sample set to expand the input and output dataset of the reconstruction model, and adjust the parameters of the reconstruction model for retraining.

[0032] The present invention has at least the following beneficial technical effects:

[0033] 1. This invention provides a method for reconstructing and experimentally verifying the blade root stress field based on multiple measurement points. Unlike traditional finite element strength analysis methods, after the reconstruction model is established, it can quickly and accurately obtain the overall stress distribution of the blade root rim using only a few measurement values ​​at the blade root rim. It is not limited by the model and mesh, has a short response time, and realizes digital twins in industrial scenarios.

[0034] 2. This invention provides a method for reconstructing and experimentally verifying the blade root stress field based on multiple measurement points. The blade root rim tensile test can simulate the stress situation of the blade in actual operation. It adopts a combination of contact and non-contact measurement methods. The non-contact measurement results are calibrated with the measurement results of the contact strain gauges, which can obtain relatively accurate strain values, thereby calculating the stress values ​​for subsequent model verification and strength analysis. Attached Figure Description

[0035] Figure 1This is a schematic diagram of the process for reconstructing and experimentally verifying the leaf root stress field based on multiple measurement points according to the present invention.

[0036] Figure 2 A parameterized schematic diagram of the geometric shape of a swallowtail-shaped leaf root;

[0037] Figure 3 This is a schematic diagram of the arrangement of strain gauges at key positions on the root rim of a swallowtail-shaped blade.

[0038] Explanation of reference numerals in the attached figures:

[0039] 1-Strain gauge. Detailed Implementation

[0040] To enable those skilled in the art to better understand the present invention, the present invention will be further described clearly and completely below in conjunction with the accompanying drawings and specific embodiments.

[0041] Please see Figure 1 This invention provides a flowchart of a method for reconstructing and experimentally verifying the blade root stress field based on multiple measuring points. The main idea of ​​this method is as follows: First, obtain a sample set of blade root parameters and establish a blade root rim model accordingly, and use the finite element method to simulate and calculate the stress field; second, establish a field reconstruction model, with the stress values ​​of key points of the blade root rim as input and the overall stress field as output, and train the model using the input and output data; finally, conduct a tensile test on the blade root rim to obtain the stress and stress field values ​​at the measuring points, and compare the reconstructed stress field with the calibrated test results to verify the reliability of the reconstruction method.

[0042] Please see Figure 1 This invention provides a method for reconstructing and experimentally verifying the leaf root stress field based on multiple measurement points. This method can be divided into the following 5 steps:

[0043] Step 1: Use the Latin hypercube sampling method to sample the leaf root parameter sample set.

[0044] To ensure the uniformity of the sample set, the Latin hypercube sampling method was used to sample the leaf root parameters. A complete description of the geometry of a leaf root requires m parameters. Figure 2 Taking a swallowtail-shaped leaf root as an example, a total of 9 parameters are needed to describe its cross-sectional geometry.

[0045] During the sampling process, the number of additional sample points N is first determined. Since each sample point contains the aforementioned m parameters, and each parameter has a specific range of values, it is necessary to extract N values ​​stratified within the specific range of each parameter. For example, the parameter h describing the curvature of the first pair of teeth at the leaf root. l The range of values ​​is [l min ,l max This range of values ​​is evenly divided into N sample spaces, each with a length of [missing information]. This creates N equally probable partitions. Then, a value is randomly selected from each of these equally probable partitions to adjust the parameter h. l Selection of representative values ​​for each partition. During sample combination, the representative values ​​for each parameter's partitions are rearranged, and the i-th representative value of each parameter is selected sequentially to form the i-th sample point. After sampling, the sample set contains N sample points, forming an N×m matrix. The size of N is appropriately chosen based on the training data requirements of the reconstructed model.

[0046] Step 2: Model the blade root and rim sections based on the blade root parameter samples obtained in Step 1.

[0047] Each parameter sample point contains information about the distance or angle between adjacent points on the outer contour of the blade root. The blade root is modeled by connecting these points with straight lines or arcs and then extruding. Since this method studies the relationship between the blade root shape and stress distribution, the parameters such as the blade root positioning information required to establish a complete blade and rim model are known and remain unchanged. The rim is modeled based on the blade root profile and positioning information. Then, the blade length is calculated based on the overall radial length of the blade and the radial position of the blade root. A cuboid of the appropriate length is added to the blade root to replace the actual blade body, generating an equivalent centrifugal force. The final blade root and rim assembly structure can be used for finite element analysis.

[0048] Step 3: Calculate the above blade root rim model using the finite element method to obtain the stress distribution.

[0049] Step 4: Based on the stress distribution in Step 3, divide the input and output of the reconstruction model, determine the basic structure and parameters of the field reconstruction model, and complete the training of the reconstruction model according to the input and output.

[0050] The input needs to include stress values ​​at key locations on the blade root rim; the output is a series of stress values ​​on the blade root rim. The number K selected must be sufficient to reflect the stress field of the blade root rim, and strength analysis should be performed accordingly. In the subsequent tensile tests on the blade root rim, it is necessary to ensure that measuring points are arranged at the key locations to obtain K stress values. Figure 3 In the dovetail-shaped blade root rim structure shown in the experiment, strain gauges were evenly distributed in the throat region and the profile region, with multiple strain gauges arranged. The measured strain was then converted into stress using the strain-stress relationship σ=Eε. The stresses at these key locations were used as inputs to obtain the desired stress. A K-dimensional vector.

[0051] The reconstruction model used in this invention is a neural network model based on a two-dimensional deconvolution algorithm, specifically comprising one fully connected layer and multiple deconvolution layers. The input and output dimensions are [N, 1, 1, L] and [N, 1, H, W], respectively, where N is the number of sample points in the sample set, L is the number of input stress values, and the number of stress values ​​in the reconstructed stress field is H × W. The batch size and the size and stride of the deconvolution kernel in each layer are selected based on the specific values ​​of N, L, H, and W, ensuring that the kernel size is smaller than the input size and the stride is less than or equal to the kernel size in the corresponding dimension. The input is first converted to a suitable two-dimensional size through a fully connected layer, maintaining a single input channel. Subsequently, the input passes through multiple deconvolution layers sequentially, and the output after computation is the reconstructed stress field. The loss function is the cross-entropy loss function, and the accuracy evaluation criterion is the mean absolute deviation. The input and its corresponding output are divided into training and test sets in an 8:2 ratio. The initial learning rate is set to a range of 0.00001 to 0.001 based on the model size and sample size. The stochastic gradient descent optimizer is used to dynamically adjust the learning rate during training. When the accuracy of the reconstructed results from 90% of the inputs in the test set is greater than 90% of the actual output, the reconstructed model is considered to have converged, and the training process ends.

[0052] Step 5: Conduct tensile tests on the blade root rim to obtain stress values ​​at key locations and stress field distribution at the blade root rim, and verify the reliability of the field reconstruction results.

[0053] The blade root rim tensile test specifically includes the following steps:

[0054] Step 5.1: Randomly select a parameter sample point from the validation set of the training data, and obtain the blade root rim model accordingly. The blade root model does not include structures other than the blade body and blade crown; it only extends the blade root platform radially to facilitate the application of loads, and the rim model only includes the portion near the profile.

[0055] Step 5.2, the model in step 5.1 is modeled. The modeling principles are: (1) the test piece and the prototype are geometrically similar and have similar stress conditions; (2) a certain number of strain gauges are arranged in the blade root throat; (3) the size is reduced according to the load capacity of the tensile testing machine used in the test.

[0056] Step 5.3: Strain gauges are placed at key positions on one side of the modeled blade root rim, and paint is sprayed on the other side to form a speckle surface.

[0057] Step 5.4: Perform a tensile test on the blade root rim, and simultaneously obtain the strain values ​​of the strain gauges and the overall strain distribution measured by the non-contact speckle full-field strain gauge. Then, combine the strain-stress relationship σ=Eε to convert the obtained strain into stress.

[0058] The results obtained from strain gauge measurements are contact measurements, while the results obtained from a non-contact speckle full-field strain gauge are non-contact measurements. The absolute accuracy (numerical accuracy) of non-contact measurement results is not very high and is easily affected by ambient light and stray light. However, under the same test conditions, the relative accuracy (distribution accuracy) of the measurement results can be guaranteed. In this method, the non-contact measurement results are calibrated based on the results obtained from contact measurements to obtain a more accurate stress field distribution.

[0059] The calibration method involves extracting the strain gauge values ​​at corresponding locations from the non-contact measurement results and establishing a mapping between non-contact and contact measurement results at corresponding locations. Since the non-contact measurement results have relatively high accuracy, while the absolute values ​​are significantly affected by the experimental environment, a linear mapping method is adopted.

[0060] ε 接触 =k*ε 非接触 +b

[0061] Where k is the mapping coefficient and b is the offset. All non-contact measurement results are mapped using the above formula to obtain the overall stress field of the blade root rim. The calibrated stress field is compared with the result reconstructed by the model to verify the accuracy of the stress field reconstruction. If the error exceeds the acceptable engineering range, i.e., the relative error exceeds 10%, steps 1-4 are repeated to increase the number of parameter sample points in the sample set, thereby expanding the input and output dataset of the reconstruction model, and the reconstruction model parameters are adjusted for retraining.

[0062] The above content provides a further detailed description of the present invention. It should not be construed that the specific embodiments of the present invention are limited to this. For those skilled in the art, several simple deductions or substitutions can be made without departing from the concept of the present invention, and all such deductions or substitutions should be considered to fall within the protection scope defined by the submitted claims.

Claims

1. A method for reconstructing and experimentally verifying the leaf root stress field based on multiple measurement points, characterized in that, include: First, a sample set of blade root parameters is obtained and a blade root rim model is established based on it. The stress field is then simulated and calculated using the finite element method. Secondly, a field reconstruction model is established, with the input being the stress values ​​at key points of the blade root rim and the output being the overall stress field. The model is trained using the input and output data. Finally, a tensile test is conducted on the blade root rim to obtain the stress and stress field values ​​at the measurement points. The reconstructed stress field is compared with the calibrated test results to verify the reliability of the reconstruction method. This method specifically includes the following steps: Step 1: Use the Latin hypercube sampling method to sample the leaf root parameter sample set; Step 2: Model the blade root and rim sections based on the blade root parameter samples obtained in Step 1; Step 3: Calculate the above blade root rim model using the finite element method to obtain the stress distribution; Step 4: Based on the stress distribution in Step 3, divide the input and output of the reconstruction model, determine the basic structure and parameters of the field reconstruction model, and complete the training of the reconstruction model according to the input and output; the specific implementation method is as follows: The reconstruction model used is a neural network model based on a two-dimensional deconvolution algorithm; the input needs to include stress values ​​at key locations on the blade root rim; the output is a series of stress values ​​on the blade root rim, with the selected number... These stress values ​​are designed to reflect the stress field at the blade root rim, enabling strength analysis. During subsequent tensile tests of the blade root rim, measuring points were strategically placed at key locations to obtain the required data. One stress value; Step 5: Conduct tensile tests on the blade root rim to obtain stress values ​​at key locations and the stress field distribution at the blade root rim, verifying the reliability of the field reconstruction results; the blade root rim tensile test specifically includes the following steps: Step 5.1: Randomly select a parameter sample point from the validation set of the training data, and obtain the blade root rim model accordingly; Step 5.2: Model the model from Step 5.1; Step 5.3: Strain gauges are placed at key locations on one side of the modeled blade root rim, and paint is sprayed on the other side to form a speckle surface; Step 5.4: Conduct a tensile test on the blade root rim, simultaneously obtaining the strain values ​​from the strain gauges and the overall strain distribution measured by a non-contact speckle full-field strain gauge. Then, combine this with the strain-stress relationship... The obtained strain is converted into stress; The results obtained by strain gauge measurement are contact measurement results, while the results obtained by non-contact speckle full-field strain gauge measurement are non-contact measurement results. The non-contact measurement results are calibrated based on the results obtained by contact measurement to obtain an accurate stress field distribution. The specific calibration method involves extracting the values ​​at the corresponding positions of the strain gauges from the non-contact measurement results and establishing a mapping between the non-contact measurement results and the contact measurement results at corresponding positions. in For mapping coefficients, This is the offset; Map all non-contact measurement results to obtain the overall stress field of the blade root rim using the above formula; compare the calibrated stress field with the results reconstructed by the model to verify the accuracy of the stress field reconstruction results; if the error exceeds the engineering acceptable range, i.e. the relative error exceeds 10%, repeat steps 1-4, increase the number of parameter sample points in the sample set to expand the input and output dataset of the reconstruction model, and adjust the parameters of the reconstruction model for retraining.

2. The method for reconstructing and experimentally verifying the leaf root stress field based on multiple measurement points according to claim 1, characterized in that, In step 1, the specific implementation method is as follows: The first step in the sampling process is to determine the number of additional sample points. Because each sample point contains There are several parameters, and each parameter has a specific range of values. Therefore, stratified extraction is performed within the specific range of each parameter. One value; for the first pair of tooth radius parameters at the leaf root The range of values ​​is Divide this range of values ​​evenly into Each sample space has a length of [number] units. That is, it formed Divide the parameters into equal probability partitions; then randomly select one value from each of these equal probability partitions to complete the parameter optimization. The selection of representative values ​​for each partition; during sample combination, the representative values ​​for each parameter's partitions are rearranged, and the first value of each parameter is selected sequentially. The first representative value constitutes the second 10 sample points; after sampling, the sample set contains a total of 100 sample points. There are 10 sample points, and the data dimension is 1. The matrix is ​​selected based on the training data requirements of the reconstructed model. Size.

3. The method for reconstructing and experimentally verifying the leaf root stress field based on multiple measurement points according to claim 1, characterized in that, In step 2, the specific implementation method is as follows: Each parameter sample point contains information about the distance or angle between adjacent points on the outer contour of the blade root. The blade root is modeled by connecting and stretching the points with straight lines or arcs. The rim is modeled based on the blade root profile and positioning information. Then, the blade length is calculated based on the overall radial length of the blade and the radial position of the blade root. A cuboid of the corresponding length is added to the blade root to replace the actual blade body, generating an equivalent centrifugal force. The final blade root and rim assembly structure is used for finite element calculation.

4. The method for reconstructing and experimentally verifying the leaf root stress field based on multiple measurement points according to claim 1, characterized in that, In step 4, the neural network model based on the two-dimensional deconvolution algorithm specifically includes one fully connected layer and multiple deconvolution layers; the dimensions of the input and output are respectively... and ,in The above refers to the number of sample points in the sample set. The number of input stress values ​​corresponds to the number of stress values ​​in the output reconstructed stress field. ;according to The specific size is selected by choosing the batch size and the size and stride of the deconvolution kernel in each layer. The kernel size is smaller than the input size, and the stride is less than or equal to the kernel size in the corresponding dimension. The input is first converted to a suitable two-dimensional size through a fully connected layer, maintaining a single input channel. Then, the input sequentially passes through multiple deconvolution layers, and the output is the reconstructed stress field. The loss function is the cross-entropy loss function, and the accuracy evaluation criterion is the mean absolute deviation. The input and its corresponding output are divided into training and test sets in an 8:2 ratio. The initial learning rate is set to a range of 0.00001–0.001 based on the model size and sample size, and is dynamically adjusted during training using a stochastic gradient descent optimizer. When the accuracy of the reconstructed results from 90% of the inputs in the test set is greater than 90% of the actual output, the reconstructed model is considered converged, and the training process ends.

5. The method for reconstructing and experimentally verifying the leaf root stress field based on multiple measurement points according to claim 1, characterized in that, In step 5.2, the modeling principles are: (1) the test piece and the prototype are geometrically similar and similar in stress conditions; (2) a predetermined number of strain gauges are arranged at the blade root throat; (3) the size is reduced with reference to the load capacity of the tensile testing machine.

Images