A blade burst speed prediction method based on sequence sampling and agent model

Abstract

This invention relates to the field of aero-engine technology and discloses a method for predicting the fracture speed of an integral bladed disk (IBD) based on sequential sampling and a surrogate model. Utilizing a strain prediction surrogate model for analysis and calculation improves the accuracy and efficiency of predicting the fracture speed of an IBD. Furthermore, in the construction of the strain prediction surrogate model, a sequential sampling method is used to optimize the generation logic of sample points, ensuring that sample points cluster within the critical region of interest in over-speed fracture tests. This improves the accuracy of IBD estimation and lays the foundation for building a high-quality surrogate model. This invention considers the influence of uncertain parameters on IBD behavior and predicts the fracture speed range under the influence of uncertain parameters, improving the accuracy and reliability of IBD prediction and providing a more scientific basis for the design and optimization of integral bladed disks (IBDs).

Description

Technical Field

[0001] This invention relates to the field of aero-engine technology and discloses a method for predicting bladed disk fracture speed based on sequence sampling and surrogate models. Background Technology

[0002] To calculate and predict the fracture speed of an integral bladed disk (IBD) in real time under scenarios such as virtual experiments, it is necessary to obtain the strain response of the disk model. Although the traditional finite element method can provide strain field information of the IBD, the rapid nature of overspeed fracture tests makes it difficult to guarantee real-time calculations if a physical model is used directly for prediction. Furthermore, traditional simulation analysis does not consider uncertainties affecting the fracture speed of the IBD, resulting in simulation results that cannot fully represent the actual fracture state. Summary of the Invention

[0003] The purpose of this invention is to provide a method for predicting the bladed disk fracture speed based on sequence sampling and surrogate model, which can meet the real-time requirements of calculation, and at the same time, introduce fracture speed prediction under uncertainty, which can improve the accuracy and efficiency of bladed disk fracture speed prediction.

[0004] To achieve the above-mentioned technical effects, the technical solution adopted by the present invention is as follows:

[0005] A method for predicting bladed disk fracture speed based on sequence sampling and surrogate models includes:

[0006] Establish a finite element analysis model of the integral bladed disk of the compressor;

[0007] Uncertainty parameters affecting the maximum strain value of the compressor integral bladed disk are obtained. The sensitivity of each uncertainty parameter to the strain value of the compressor integral bladed disk is obtained by simulation using a finite element analysis model. Uncertainty parameters whose sensitivity to the maximum strain value of the compressor integral bladed disk is greater than a preset sensitivity threshold are selected as characterization parameters. The characterization parameters include key dimension parameters, elastic modulus, Poisson's ratio and rotational speed.

[0008] Based on the characterization parameters, initial sample points are randomly generated within the design space of the characterization parameters. Based on the characterization parameters contained in each initial sample point, the characterization parameters contained in each initial sample point are input into the finite element analysis model for simulation to obtain the maximum strain analysis value of the compressor integral bladed disk at the corresponding speed for each initial sample point. Using the characterization parameters of all initial sample points as input and the maximum strain analysis value corresponding to each initial sample point as output, the Kriging model is used to construct the initial strain prediction proxy model of the compressor integral bladed disk.

[0009] An adaptive sequence sampling method based on a space reduction strategy is used to generate training sample points in the design space of the characterization parameters. The characterization parameters contained in each training sample point are input into the finite element analysis model for simulation to obtain the maximum strain analysis value corresponding to each training sample point. The initial strain prediction proxy model is trained using each training sample point and the corresponding maximum strain analysis value to obtain a trained strain prediction proxy model.

[0010] The strain prediction surrogate model is used to predict the fracture speed of the integral bladed disk of the compressor, and the theoretical fracture speed and the fracture speed range under the influence of uncertain parameters are obtained.

[0011] Furthermore, the step of generating training sample points within the design space of the representation parameters using an adaptive sequence sampling method based on a space reduction strategy includes:

[0012] The initial sample points are used as historical sample points, and a set of historical sample points is generated. The maximum and minimum values ​​of the maximum strain prediction values ​​corresponding to all current historical sample points are determined.

[0013] The design space of the characterization parameters is divided into a rejection region and a feasible region. New sample points to be evaluated are randomly generated in the feasible region. The rejection region is the area within a preset radius centered on the historical sample points. The feasible region is the area in the design space outside the rejection region.

[0014] The initial strain prediction surrogate model is used to analyze each new sample point to be evaluated, and the maximum strain prediction value of each new sample point to be evaluated at the corresponding rotational speed is obtained; each new sample point to be evaluated is input into the initial strain prediction surrogate model for analysis, and the maximum expected strain value of each new sample point to be evaluated at the corresponding rotational speed is obtained; at the rotational speed of each new sample point to be evaluated, the Monte Carlo sampling method is used to sample within the design space of the key dimensional parameters, elastic modulus, and Poisson's ratio, and the initial strain prediction surrogate model is used for analysis to obtain the maximum and minimum values ​​of the maximum strain prediction value at the rotational speed of each new sample point to be evaluated;

[0015] Based on the maximum and minimum values ​​of the maximum strain prediction value corresponding to all historical sample points, the maximum strain prediction value, the maximum expected value of the maximum strain, the maximum and minimum values ​​of the maximum strain prediction value at the rotational speed of each new sample point to be evaluated, and the sampling criterion function is used to analyze and obtain the sampling criterion function value of each new sample point to be evaluated.

[0016] Within the design space, determine the minimum distance between each new sample point to be evaluated and all historical sample points;

[0017] Using the location of the new sample point to be evaluated as the variable, and the sampling criterion function value corresponding to the new sample point to be evaluated... With maximizing as the optimization objective, and with the constraint that the distance between the new sample point to be evaluated and the nearest historical sample point is greater than or equal to the preset radius of the rejection region, a genetic algorithm is used to analyze and obtain the position of the new sample point that satisfies the constraint and maximizes the value of the sampling criterion function. The corresponding new sample point is then output as the training sample point.

[0018] Furthermore, the expression for the sampling criterion function is:

[0019] ;

[0020] in: The value of the sampling criterion function; The maximum predicted strain value of the compressor integral bladed disk corresponding to the new sample point to be evaluated; It is the minimum of the maximum predicted strain values ​​corresponding to all historical sample points. It is the maximum value of the maximum predicted strain corresponding to all historical sample points. This represents the expected value of the maximum strain at the current rotational speed of the new sample point; Rotation speed of the new sample points to be evaluated The maximum value of the predicted maximum strain; Rotation speed of the new sample points to be evaluated The minimum value of the predicted maximum strain.

[0021] Furthermore, the expression for the constraint condition is:

[0022] ;

[0023] in, This represents the minimum distance between the new sample point to be evaluated and all historical sample points. The minimum distance between all current historical sample points; To design the maximum number of sample points; This represents the current number of historical sample points.

[0024] Furthermore, the formula for calculating the minimum distance between all current historical sample points is:

[0025] ;

[0026] in, The minimum distance between all current historical sample points. The sample point set consists of all current historical sample points. For current historical sample points Historical sample points The One dimension, The number of dimensions for historical sample points.

[0027] Further, the steps for obtaining a trained strain prediction proxy model include: inputting the characterization parameters contained in the training sample points into the finite element analysis model for analysis, obtaining the maximum strain analysis value of each training sample point at the corresponding rotational speed, dividing the training sample points into a training set and a test set, using the characterization parameters of each training sample point in the training set as input and the maximum strain analysis value corresponding to each training sample point as output, reconstructing the strain prediction proxy model using the Kriging model, then inputting the characterization parameters of the training sample points in the test set into the reconstructed strain prediction proxy model, obtaining the maximum strain prediction value corresponding to the training sample points in the test set, and then calculating the root mean square error between the maximum strain prediction value and the maximum strain analysis value of the training sample points in the test set. If the root mean square error is less than a preset error threshold, it is determined that the prediction accuracy of the reconstructed strain prediction proxy model meets the preset accuracy requirement, and a trained strain prediction proxy model is obtained; otherwise, the existing training sample points are used as historical sample points and added to the historical sample point set, and new training sample points are generated again until a trained strain prediction proxy model is obtained.

[0028] Furthermore, the steps of using the strain prediction surrogate model to predict the fracture speed of the integral bladed disk of the compressor, and obtaining the theoretical fracture speed and the fracture speed range under the influence of uncertainty parameters, include:

[0029] The theoretical values ​​of the geometric characterization parameters, elastic modulus, and Poisson's ratio of the target compressor integral bladed disk are input into the trained strain prediction proxy model, along with the first and second initial test speeds. The first and second initial test speeds are then input into the trained strain prediction proxy model for analysis, obtaining the first and second maximum strain prediction values ​​of the target compressor integral bladed disk. The difference between the minimum of the first and second maximum strain prediction values ​​and the theoretical fracture strain of the target compressor integral bladed disk is calculated. If the difference is less than a preset difference range, the corresponding test speed is used as the theoretical fracture speed prediction value of the target compressor integral bladed disk; otherwise, the test speed is adjusted and updated using a bisection method to obtain the theoretical fracture speed of the target compressor integral bladed disk.

[0030] The Monte Carlo sampling method is used to sample within the design space of the key dimensional parameters, elastic modulus, and Poisson's ratio, generating several sample points. Each sample point and the maximum strain prediction value corresponding to the theoretical fracture speed are input into the trained strain prediction surrogate model for analysis. The fracture speed corresponding to each sample point and the maximum strain prediction value is calculated. Then, the maximum and minimum values ​​of the fracture speed are statistically determined. Based on the maximum and minimum values, the fracture speed range of the target compressor integral bladed disk is determined when considering the uncertainty parameters. The theoretical fracture speed and fracture speed range of the target compressor integral bladed disk are output as prediction results.

[0031] Compared with the prior art, the beneficial effects of this invention are:

[0032] This invention utilizes a strain prediction surrogate model for analysis and calculation, which improves the accuracy and efficiency of predicting the fracture speed of the integral bladed disk (IBD) of a compressor. Furthermore, in the construction of the surrogate model, a sequential sampling method is employed to optimize the generation logic of sample points, ensuring that sample points cluster within the critical region of interest in the over-rotation fracture test. This improves the accuracy of the IBD fracture speed estimation and lays the foundation for building a high-quality surrogate model. This invention replaces the traditional finite element method with a strain prediction surrogate model for strain prediction calculations. It eliminates the need for numerous simulation models simulating IBD fracture, requiring only a small number of sample points for finite element calculations to quickly output strain prediction results under any operating condition. This results in high computational efficiency, low computational cost, and the ability to meet real-time computational requirements. This invention considers the influence of uncertain parameters on IBD fracture behavior, predicting the fracture speed range under the influence of uncertain parameters, thus improving the accuracy and reliability of fracture speed prediction and providing a more scientific basis for the design and optimization of the integral bladed disk of a compressor. Attached Figure Description

[0033] Figure 1 This is a flowchart of the bladed disk fracture speed prediction method based on sequence sampling and surrogate model in the embodiment. Detailed Implementation

[0034] The present invention will now be described in further detail with reference to the embodiments and accompanying drawings. However, this should not be construed as limiting the scope of the above-described subject matter of the present invention to the following embodiments; all technologies implemented based on the content of the present invention fall within the scope of the present invention.

[0035] Example 1

[0036] See Figure 1 A method for predicting bladed disk fracture speed based on sequence sampling and surrogate model, comprising:

[0037] Establish a finite element analysis model of the integral bladed disk of the compressor;

[0038] Uncertainty parameters affecting the maximum strain value of the compressor integral bladed disk are obtained. The sensitivity of each uncertainty parameter to the strain value of the compressor integral bladed disk is obtained by simulation using a finite element analysis model. Uncertainty parameters whose sensitivity to the maximum strain value of the compressor integral bladed disk is greater than a preset sensitivity threshold are selected as characterization parameters. The characterization parameters include key dimension parameters, elastic modulus, Poisson's ratio and rotational speed.

[0039] Based on the characterization parameters, initial sample points are randomly generated within the design space of the characterization parameters. Based on the characterization parameters contained in each initial sample point, the characterization parameters contained in each initial sample point are input into the finite element analysis model for simulation to obtain the maximum strain analysis value of the compressor integral bladed disk at the corresponding speed for each initial sample point. Using the characterization parameters of all initial sample points as input and the maximum strain analysis value corresponding to each initial sample point as output, the Kriging model is used to construct the initial strain prediction proxy model of the compressor integral bladed disk.

[0040] An adaptive sequence sampling method based on a space reduction strategy is used to generate training sample points in the design space of the characterization parameters. The characterization parameters contained in each training sample point are input into the finite element analysis model for simulation to obtain the maximum strain analysis value corresponding to each training sample point. The initial strain prediction proxy model is trained using each training sample point and the corresponding maximum strain analysis value to obtain a trained strain prediction proxy model.

[0041] The strain prediction surrogate model is used to predict the fracture speed of the integral bladed disk of the compressor, and the theoretical fracture speed and the fracture speed range under the influence of uncertain parameters are obtained.

[0042] This invention utilizes a strain prediction surrogate model for analysis and calculation, which improves the accuracy and efficiency of predicting the fracture speed of the integral bladed disk (IBD) of a compressor. Furthermore, in the construction of the surrogate model, a sequential sampling method is employed to optimize the generation logic of sample points, ensuring that sample points cluster within the critical region of interest in the over-rotation fracture test. This improves the accuracy of the IBD fracture speed estimation and lays the foundation for building a high-quality surrogate model. This invention replaces the traditional finite element method with a strain prediction surrogate model for strain prediction calculations. It eliminates the need for numerous simulation models simulating IBD fracture, requiring only a small number of sample points for finite element calculations to quickly output strain prediction results under any operating condition. This results in high computational efficiency, low computational cost, and the ability to meet real-time computational requirements. This invention considers the influence of uncertain parameters on IBD fracture behavior, predicting the fracture speed range under the influence of uncertain parameters, thus improving the accuracy and reliability of fracture speed prediction and providing a more scientific basis for the design and optimization of the integral bladed disk of a compressor.

[0043] Example 2

[0044] To elaborate on this invention, a certain type of integral bladed disk of a compressor is used as the target integral bladed disk of the compressor. To predict the fracture speed of the target integral bladed disk, the following are the specific steps of the bladed disk fracture speed prediction method based on sequential sampling and a surrogate model of this invention, including:

[0045] Step 1: Determine the theoretical fracture strain value of the target compressor integral bladed disk material and the range of the target compressor integral bladed disk material parameters; and establish a finite element analysis model of the target compressor integral bladed disk based on the theoretical values ​​of the geometric dimensions and material parameters of the target compressor integral bladed disk; the material parameters include the density, Poisson's ratio, and elastic modulus of the target compressor integral bladed disk.

[0046] Specifically, firstly, several sample bladed disks from different production batches of the target compressor integral bladed disk are obtained. Specimens are fabricated on each sample bladed disk for material property testing. Tensile tests are conducted to obtain the measured engineering stress and strain values ​​for each sample bladed disk. Then, the average measured engineering stress and strain values ​​are calculated, and based on these, the material engineering stress-strain curve of the target compressor integral bladed disk is plotted. The engineering stress-strain curve is then converted into a true stress-true strain curve using the following formula:

[0047] ;

[0048] in, To be truly adaptable; The length of the sample at a certain moment; This is the original gauge length of the sample; For engineering strain

[0049] ;

[0050] in, For engineering stress, This is the true stress.

[0051] The strain value at fracture of the integral bladed disk of the target compressor is obtained by reading the strain value from the true stress-true strain curve of the integral bladed disk of the target compressor.

[0052] Specimens were prepared on each sample bladed disk, and the Poisson's ratio and elastic modulus of each bladed disk material were tested using a universal testing machine. The maximum and minimum values ​​of the measured Poisson's ratio and elastic modulus were then used as the range of values ​​for the Poisson's ratio and elastic modulus of the overall bladed disk of the target compressor, respectively. At the same time, the density of each sample bladed disk was measured, and the maximum and minimum values ​​of the measured density were used as the range of values ​​for the density of the overall bladed disk of the target compressor.

[0053] Then, based on the design drawings and other relevant design data of the target compressor integral bladed disk, the theoretical values ​​of its geometric dimensions and material parameters are obtained. Based on these theoretical values, a finite element analysis model of the target compressor integral bladed disk is established. It should be noted that the theoretical geometric dimensions are the basic dimensions in the design drawings without considering tolerances, and the theoretical material parameters are the Poisson's ratio and elastic modulus of the target compressor integral bladed disk material obtained by looking up tables.

[0054] Step 2: Obtain the uncertainty parameters that affect the maximum strain value of the compressor integral bladed disk. Use the finite element analysis model to simulate and obtain the sensitivity of each uncertainty parameter to the strain value of the compressor integral bladed disk. Select the uncertainty parameters with sensitivity greater than the preset sensitivity threshold as characterization parameters. The uncertainty parameters include size parameters, material parameters and load parameters. The characterization parameters include key size parameters, elastic modulus, Poisson's ratio and rotational speed.

[0055] Specifically, traditional methods for analyzing the structural performance of integral bladed disks (IBDs) of compressors typically assume that the material properties and geometric parameters of IBDs are deterministic constants, meaning they do not change with time or conditions throughout the analysis. However, during the design and manufacturing stages of IBDs, the materials themselves inevitably exhibit a certain degree of non-uniformity; for example, material parameters such as elastic modulus and density fluctuate between different batches. Furthermore, due to current manufacturing limitations, it is difficult to achieve consistent replication of all IBDs. Even within the same production batch, the structural dimensions and material parameters exhibit a certain range of random distribution. These multi-source uncertainties significantly affect the fracture behavior characteristics of IBDs, causing the critical fracture speed to exhibit significant dispersion and fluctuation range, rather than a single deterministic value. Therefore, this invention considers the uncertainties of IBDs when predicting their fracture speed, systematically reflecting the impact of these uncertainties on the structural response and over-speed fracture behavior of the IBD. The specific steps are as follows:

[0056] Step 2.1: Obtain the uncertainty parameters affecting the maximum strain value of the target compressor integral bladed disk. These uncertainty parameters include the dimensional parameters, material parameters, and load parameters of the target compressor integral bladed disk. This embodiment considers the geometric uncertainty of the compressor integral bladed disk from the perspective of dimensional tolerance. First, a 3D scan is performed on each sample bladed disk. Based on the 3D scan data, the values ​​of each dimensional parameter on different sample bladed disks are statistically analyzed, and the maximum and minimum values ​​are determined. This yields the numerical fluctuation range of each dimensional parameter of the target compressor integral bladed disk. Dimensional parameters with fluctuation ranges greater than a preset threshold are selected, and the statistically obtained numerical fluctuation range is used as the value range for the selected dimensional parameters. For example, based on the 3D scan data, after statistical filtering, the following values ​​are obtained: C 1. C 2. C 3. C 4. C 5. C 6. C 7. C 8 These are 8 size parameters whose fluctuation range exceeds a preset threshold, which is set based on experience.

[0057] Meanwhile, this embodiment will measure the density of the target compressor integral bladed disk. Elastic modulus E Poisson's ratio B Material parameters, including rotational speed, are included as uncertain parameters. S The load parameter is one of the uncertain parameters. Among them, the rotational speed... S The range of values ​​is set manually based on experience and needs to cover the design operating speed and estimated fracture speed of the target compressor's overall bladed disk. The estimated fracture speed can be determined based on experience or experiments.

[0058] Step 2.2: Using a global sensitivity analysis method, characterization parameters with a sensitivity greater than a preset sensitivity threshold to the maximum strain value of the target compressor integral bladed disk are selected from the uncertainty parameters. Specifically, an optimized Latin hypercube sampling method is used to randomly generate 200 sensitivity analysis sample points within the design space. Each sensitivity analysis sample point consists of a set of the aforementioned uncertainty parameters, that is, each sensitivity analysis sample point includes a set of { C 1, C 2, C 3, C 4, C 5, C 6, C 7, C 8, , B , E , SThe design space is the range of values ​​for each uncertainty parameter, and the same applies below. Then, the uncertainty parameter of each sensitivity analysis sample point is used as input, and the finite element analysis model is used for analysis to obtain the rotational speed of the target compressor integral bladed disk at each sample point. S The maximum strain value is obtained for each sensitivity analysis sample point. A high-precision mapping relationship between the uncertainty parameters and the maximum strain value of the roulette wheel is then constructed using the Kriging surrogate model. The main effect sensitivity index of each uncertainty parameter is evaluated using the Sobol variance decomposition method combined with Monte Carlo sampling technology. Finally, uncertainty parameters whose main effect sensitivity index exceeds a preset sensitivity threshold are selected as characterization parameters. The characterization parameters in this embodiment include { C 2, C 3, C 7, C 8, B , E , S},in{ C 2, C 3, C 7, C 8} represents the key dimensional parameters selected. It should be noted that the uncertainty parameters of the sensitivity analysis sample points are input into the finite element analysis model for analysis to obtain the target compressor integral bladed disk at the corresponding rotational speed. S The maximum strain value in the strain field is the maximum strain analysis value.

[0059] Step 3: Based on the selected characterization parameters, an optimized Latin hypercube sampling method is used to select initial sample points in the design space. Each initial sample point includes a set of characterization parameters { C 2, C 3, C 7, C 8, B , E , S The characterization parameters contained in each initial sample point are input into the finite element analysis model for simulation to obtain the value of each initial sample point at the corresponding rotational speed. S The maximum strain analysis value of the target compressor integral bladed disk is obtained. Using the characterization parameters of all initial sample points as input and the maximum strain analysis value corresponding to each initial sample point as output, the Kriging model is used to construct the initial strain prediction surrogate model of the target compressor integral bladed disk.

[0060] Step 4: An adaptive sequence sampling method based on a space reduction strategy is used to generate training sample points within the design space of the characterization parameters. The characterization parameters contained in each training sample point are input into the finite element analysis model for simulation to obtain the maximum strain analysis value corresponding to each training sample point. The initial strain prediction surrogate model is then trained using each training sample point and its corresponding maximum strain analysis value. Specifically, the following steps are included:

[0061] Step 4.1: Use the initial sample points as historical sample points, generate a set of historical sample points, and determine the maximum value of the maximum strain prediction value corresponding to all historical sample points. and minimum value .

[0062] Step 4.2: Divide the design space of the characterization parameters into a rejection region and a feasible region. Randomly generate several new sample points to be evaluated within the feasible region. The rejection region is centered on historical sample points and extends within a preset radius. The area within, with a preset radius , The minimum distance between all historical sample points. To design the maximum number of sample points, The number of historical sample points; the feasible region is the area in the design space outside the rejection region.

[0063] It should be noted that the minimum distance between all historical sample points ,in, The sample point set consists of all historical sample points. Historical sample points Historical sample points The One dimension, This represents the number of dimensions for historical sample points. In this embodiment, each historical sample point includes { C 2, C 3, C 7, C 8, B , E , S There are a total of 7 representation parameters, so the number of dimensions is... .

[0064] Step 4.3: Analyze each new sample point to be evaluated using the initial strain prediction surrogate model to obtain the maximum predicted strain value of each new sample point at the corresponding rotational speed; input each new sample point to be evaluated into the initial strain prediction surrogate model for analysis to obtain the maximum expected strain value of each new sample point at the corresponding rotational speed. Rotation speed at each new sample point to be evaluated When fixed, 1000 sample points are sampled within the design space of the key dimensional parameters, elastic modulus, and Poisson's ratio using the Monte Carlo sampling method. The initial strain prediction surrogate model is then used for analysis to obtain the values ​​of each new sample point to be evaluated at a fixed rotational speed. Maximum value of the predicted maximum strain and minimum value It should be noted that the expected maximum strain value of the new sample point to be evaluated at the corresponding rotational speed is... It is to use the characterization parameters of the new sample points to be evaluated { C 2, C 3, C 7, C 8, B , E The theoretical value and rotational speed of the new sample points to be evaluated. S Input the initial strain prediction surrogate model and obtain the maximum strain analysis value. In addition, at the rotational speed of the new sample points to be evaluated, 1000 sample points are sampled using the Monte Carlo sampling method within the design space of the key dimensional parameters, elastic modulus, and Poisson's ratio, excluding historical sample points.

[0065] Step 4.4: Based on the maximum predicted strain value corresponding to all historical sample points. and minimum value The maximum predicted strain value corresponding to each new sample point to be evaluated. Maximum expected strain Each new sample point to be evaluated is at a fixed rotational speed. Maximum value of the predicted maximum strain and minimum value The sampling criterion function is used for analysis to obtain the sampling criterion function value for each new sample point to be evaluated. The expression is:

[0066] .

[0067] In the sampling criterion function, through This causes newly generated sample points to be evaluated to cluster towards the region with the largest strain.

[0068] Step 4.5: Within the design space, determine the minimum distance between each new sample point to be evaluated and all historical sample points. .

[0069] Step 4.6: Using the location of the new sample point to be evaluated as the variable, and the sampling criterion function value corresponding to the new sample point to be evaluated... Maximizing the sampling criterion function is the optimization objective. The constraint is that the distance between the new sample point to be evaluated and its nearest historical sample point is greater than or equal to the preset radius of the rejection region. A genetic algorithm is used to analyze the data and obtain the position of the new sample point that satisfies the constraint and maximizes the sampling criterion function value. This new sample point is then output as the training sample point. It should be noted that the expression for the constraint is:

[0070] ;

[0071] in, This represents the minimum distance between the new sample point to be evaluated and all historical sample points. The minimum distance between all current historical sample points; To design the maximum number of sample points; This represents the current number of historical sample points.

[0072] Step 4.7: Input the characterization parameters contained in the training sample points into the finite element analysis model for analysis to obtain the maximum strain analysis value of each training sample point at the corresponding rotational speed. Divide the training sample points into a training set and a test set. Using the characterization parameters of each training sample point in the training set as input and the maximum strain analysis value corresponding to each training sample point as output, reconstruct the strain prediction surrogate model using the Kriging model. Then, input the characterization parameters of the training sample points in the test set into the reconstructed strain prediction surrogate model to obtain the maximum strain prediction value corresponding to the training sample points in the test set. Then, calculate the root mean square error between the maximum strain prediction value and the maximum strain analysis value of the training sample points in the test set. If the root mean square error is less than the preset error threshold, the prediction accuracy of the reconstructed strain prediction surrogate model is considered to meet the preset accuracy requirement, and a trained strain prediction surrogate model is obtained. Otherwise, the existing training sample points are used as historical sample points and added to the historical sample point set. Repeat steps 4.1 to 4.6 to obtain a newly generated training sample point, and then reconstruct the strain prediction surrogate model until a trained strain prediction surrogate model is obtained.

[0073] It should be noted that the strain field of the target compressor integral bladed disk output by the finite element analysis model in this invention at the corresponding rotational speed is used to extract the maximum strain value as the maximum strain analysis value. Similarly, the strain prediction proxy model in this invention also outputs the strain field of the target compressor integral bladed disk at the corresponding rotational speed, and extracts the maximum strain value as the maximum strain prediction value.

[0074] Step 5: Utilize the strain prediction surrogate model to predict the fracture speed of the target compressor integral bladed disk, obtaining the predicted fracture speed result. Specifically, this includes:

[0075] Step 5.1: Input the theoretical values ​​of the geometric characterization parameters, elastic modulus, and Poisson's ratio of the target compressor integral bladed disk into the trained strain prediction surrogate model, and input the first initial test speed and the second initial test speed. Both the first initial test speed and the second initial test speed are within the range of the speed... S Within the range of values, the first initial test speed must be less than the predicted fracture speed of the target compressor integral bladed disk based on experience, and the second initial test speed must be greater than the predicted fracture speed of the target compressor integral bladed disk based on experience. Then, the first and second initial test speeds are input into the trained strain prediction proxy model for analysis to obtain the first and second maximum strain prediction values ​​of the target compressor integral bladed disk, respectively. The difference between the minimum of the first and second maximum strain prediction values ​​and the theoretical fracture strain of the target compressor integral bladed disk is calculated. If the difference is less than a preset difference range, the corresponding test speed is used as the predicted theoretical fracture speed of the target compressor integral bladed disk; otherwise, the test speed is adjusted and updated using a bisection method to obtain the theoretical fracture speed of the target compressor integral bladed disk.

[0076] Step 5.2: Use the Monte Carlo sampling method to measure the key dimensional parameters and elastic modulus. E Poisson's ratio B Sampling was conducted within the design space to generate 1000 sample points. Each sample point includes a set of key dimensional parameters, elastic modulus, and Poisson's ratio, i.e., { C 2, C 3, C 7, C 8, B , E The maximum strain prediction value corresponding to each sample point and the theoretical fracture speed is input into the trained strain prediction proxy model for analysis. The fracture speed corresponding to each sample point and the maximum strain prediction value is calculated. Then, the maximum and minimum values ​​of the fracture speed are statistically determined. Based on the maximum and minimum values, the fracture speed range of the target compressor integral bladed disk is determined when considering the uncertainty parameters. The theoretical fracture speed and fracture speed range of the target compressor integral bladed disk are output as the prediction results.

[0077] This invention predicts the theoretical fracture speed based on the theoretical values ​​of relevant parameters of the integral bladed disk (IBD) of the target compressor. It also considers the influence of uncertain parameters on the disk's fracture behavior. By introducing uncertain parameters such as key dimensional parameters, elastic modulus, and Poisson's ratio, it predicts the fracture speed range under the influence of these uncertain parameters. Compared to traditional prediction methods based solely on deterministic parameters, this invention can more comprehensively and realistically reflect the fracture speed conditions that the IBD of the target compressor may encounter during actual operation, improving the accuracy and reliability of fracture speed prediction. This provides a more scientific and reasonable basis for the design, operation, and maintenance of IBDs of compressors.

[0078] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting bladed disk fracture speed based on sequence sampling and surrogate model, characterized in that, include: Establish a finite element analysis model of the integral bladed disk of the compressor; Uncertainty parameters affecting the maximum strain value of the compressor integral bladed disk are obtained. The sensitivity of each uncertainty parameter to the strain value of the compressor integral bladed disk is obtained by simulation using a finite element analysis model. Uncertainty parameters whose sensitivity to the maximum strain value of the compressor integral bladed disk is greater than a preset sensitivity threshold are selected as characterization parameters. The characterization parameters include key dimension parameters, elastic modulus, Poisson's ratio and rotational speed. Based on the characterization parameters, initial sample points are randomly generated within the design space of the characterization parameters; Based on the characterization parameters contained in each initial sample point, the characterization parameters contained in each initial sample point are input into the finite element analysis model for simulation, and the maximum strain analysis value of the compressor integral bladed disk at the corresponding speed is obtained for each initial sample point. Using the characterization parameters of all initial sample points as input and the maximum strain analysis value corresponding to each initial sample point as output, the Kriging model is used to construct an initial strain prediction surrogate model for the integral bladed disk of the compressor. An adaptive sequence sampling method based on a space reduction strategy is used to generate training sample points in the design space of the characterization parameters. The characterization parameters contained in each training sample point are input into the finite element analysis model for simulation to obtain the maximum strain analysis value corresponding to each training sample point. The initial strain prediction proxy model is trained using each training sample point and its corresponding maximum strain analysis value to obtain a trained strain prediction proxy model. The strain prediction surrogate model is used to predict the fracture speed of the integral bladed disk of the compressor, and the theoretical fracture speed and the fracture speed range under the influence of uncertain parameters are obtained.

2. The method for predicting the bladed disk fracture speed according to claim 1, characterized in that, The uncertainty parameters include dimensional parameters, material parameters, and load parameters.

3. The method for predicting the bladed disk fracture speed according to claim 2, characterized in that, The steps of generating training sample points within the design space of the representation parameters using an adaptive sequence sampling method based on a space reduction strategy include: The initial sample points are used as historical sample points, and a set of historical sample points is generated. The maximum and minimum values ​​of the maximum strain prediction values ​​corresponding to all current historical sample points are determined. The design space of the characterization parameters is divided into a rejection region and a feasible region. New sample points to be evaluated are randomly generated in the feasible region. The rejection region is the area within a preset radius centered on the historical sample points. The feasible region is the area in the design space outside the rejection region. The initial strain prediction surrogate model is used to analyze each new sample point to be evaluated, and the maximum strain prediction value of each new sample point to be evaluated at the corresponding rotational speed is obtained; each new sample point to be evaluated is input into the initial strain prediction surrogate model for analysis, and the maximum expected strain value of each new sample point to be evaluated at the corresponding rotational speed is obtained; at the rotational speed of each new sample point to be evaluated, the Monte Carlo sampling method is used to sample within the design space of the key dimensional parameters, elastic modulus, and Poisson's ratio, and the initial strain prediction surrogate model is used for analysis to obtain the maximum and minimum values ​​of the maximum strain prediction value at the rotational speed of each new sample point to be evaluated; Based on the maximum and minimum values ​​of the maximum strain prediction value corresponding to all historical sample points, the maximum strain prediction value, the maximum expected value of the maximum strain, the maximum and minimum values ​​of the maximum strain prediction value at the rotational speed of each new sample point to be evaluated, and the sampling criterion function is used to analyze and obtain the sampling criterion function value of each new sample point to be evaluated. Within the design space, determine the minimum distance between each new sample point to be evaluated and all historical sample points; Using the location of the new sample point to be evaluated as a variable, maximizing the sampling criterion function value corresponding to the new sample point to be evaluated as the optimization objective, and the distance between the new sample point to be evaluated and the nearest historical sample point being greater than or equal to the preset radius of the rejection region as a constraint, a genetic algorithm is used to analyze and obtain the location of the new sample point that satisfies the constraint and maximizes the sampling criterion function value, and the corresponding new sample point is output as the training sample point.

4. The method for predicting the bladed disk fracture speed according to claim 3, characterized in that, The expression for the sampling criterion function is: ； in: The value of the sampling criterion function; The maximum predicted strain value of the compressor integral bladed disk corresponding to the new sample point to be evaluated; It is the minimum of the maximum predicted strain values ​​corresponding to all historical sample points. It is the maximum value of the maximum predicted strain corresponding to all historical sample points. This represents the expected value of the maximum strain at the current rotational speed of the new sample point; Rotation speed of the new sample points to be evaluated The maximum value of the predicted maximum strain; Rotation speed of the new sample points to be evaluated The minimum value of the predicted maximum strain.

5. The method for predicting the bladed disk fracture speed according to claim 4, characterized in that, The expression for the constraint condition is: ； in, This represents the minimum distance between the new sample point to be evaluated and all historical sample points. The minimum distance between all current historical sample points; To design the maximum number of sample points; This represents the current number of historical sample points.

6. The method for predicting the bladed disk fracture speed according to claim 5, characterized in that, The formula for calculating the minimum distance between all current historical sample points is: ； in, The minimum distance between all current historical sample points. The sample point set consists of all current historical sample points. For current historical sample points Historical sample points The One dimension, The number of dimensions for historical sample points.

7. The method for predicting the bladed disk fracture speed according to claim 6, characterized in that, The steps to obtain a trained strain prediction surrogate model include: The characterization parameters of the training sample points are input into the finite element analysis model for analysis to obtain the maximum strain analysis value of each training sample point at the corresponding rotational speed. The training sample points are divided into a training set and a test set. The characterization parameters of each training sample point in the training set are used as input, and the maximum strain analysis value corresponding to each training sample point is used as output. The strain prediction surrogate model is reconstructed using the Kriging model. Then, the characterization parameters of the training sample points in the test set are input into the reconstructed strain prediction surrogate model to obtain the maximum strain prediction value corresponding to the training sample points in the test set. Then, the root mean square error between the maximum strain prediction value and the maximum strain analysis value of the training sample points in the test set is calculated. If the root mean square error is less than the preset error threshold, the prediction accuracy of the reconstructed strain prediction surrogate model is considered to meet the preset accuracy requirement, and a trained strain prediction surrogate model is obtained. Otherwise, the existing training sample points are used as historical sample points and added to the historical sample point set to regenerate new training sample points until a trained strain prediction surrogate model is obtained.

8. The method for predicting the bladed disk fracture speed according to claim 7, characterized in that, The steps for predicting the fracture speed of the integral bladed disk of the compressor using the strain prediction surrogate model, and obtaining the theoretical fracture speed and the fracture speed range under the influence of uncertain parameters, include: The theoretical values ​​of the geometric characterization parameters, elastic modulus, and Poisson's ratio of the target compressor integral bladed disk are input into the trained strain prediction proxy model, along with the first and second initial test speeds. The first and second initial test speeds are then input into the trained strain prediction proxy model for analysis, obtaining the first and second maximum strain prediction values ​​of the target compressor integral bladed disk. The difference between the minimum of the first and second maximum strain prediction values ​​and the theoretical fracture strain of the target compressor integral bladed disk is calculated. If the difference is less than a preset difference range, the corresponding test speed is used as the theoretical fracture speed prediction value of the target compressor integral bladed disk; otherwise, the test speed is adjusted and updated using a bisection method to obtain the theoretical fracture speed of the target compressor integral bladed disk. The Monte Carlo sampling method is used to sample within the design space of the key dimensional parameters, elastic modulus, and Poisson's ratio, generating several sample points. Each sample point and the maximum strain prediction value corresponding to the theoretical fracture speed are input into the trained strain prediction surrogate model for analysis. The fracture speed corresponding to each sample point and the maximum strain prediction value is calculated. Then, the maximum and minimum values ​​of the fracture speed are statistically determined. Based on the maximum and minimum values, the fracture speed range of the target compressor integral bladed disk is determined when considering the uncertainty parameters. The theoretical fracture speed and fracture speed range of the target compressor integral bladed disk are output as prediction results.

Images