A method, apparatus, computer equipment, and medium for predicting disc fracture failure.

Abstract

This invention provides a method, apparatus, computer equipment, and medium for predicting the failure of a wheel disc, relating to the field of digital twin technology. The method includes the following steps: conducting a uniaxial quasi-static tensile test on a tensile test specimen to obtain the engineering stress-strain curve of the wheel disc; converting the engineering stress-strain curve into a true stress-strain curve; constructing a wheel disc failure prediction model under deterministic conditions based on the true stress-strain curve; performing sensitivity analysis on the quantified parameters of uncertainty; obtaining sample data under different working conditions through finite element calculation based on key parameters; constructing a maximum strain proxy model for the wheel disc based on the sample data; inputting the key parameters into the maximum strain proxy model to obtain the confidence interval of the wheel disc's rotational speed at the time of failure. This scheme, by constructing a maximum strain proxy model for the wheel disc, achieves high-precision full-domain simulation of the wheel disc's failure process, solving the problems of limited test measurement points and insufficient simulation accuracy.

Description

Technical Field

[0001] This invention relates to the field of digital twin technology, and in particular to a method, apparatus, computer equipment, and medium for predicting wheel breakage failure. Background Technology

[0002] Aero-engine bladed disks are critical components, subjected to high temperatures, high pressures, high speeds, and alternating loads over long periods. During acceleration, the speed may exceed limits, increasing the risk of failure. To prevent failure, it is necessary to study their mechanisms and characteristics. Research methods fall into two categories: experimental over-speed testing and physics-based mechanism analysis. Over-speed testing is costly, has limited measurement points, and struggles to capture overall changes; physics analysis, due to the randomness of processes and operating conditions, cannot fully reflect actual failure. Therefore, for the problem of bladed disk fracture, it is necessary to study digital twin theory and methods to achieve real-time failure analysis and prediction. Summary of the Invention

[0003] In view of this, embodiments of the present invention provide a method for predicting bladed disk fracture failure, to solve the technical problems of limited test measurement points and insufficient simulation accuracy in the prior art when aero-engine bladed disks fail. The method includes:

[0004] A uniaxial quasi-static tensile test was performed on the tensile test specimen to obtain the engineering stress-strain curve of the wheel, and the engineering stress-strain curve was converted into a true stress-strain curve.

[0005] Based on the true stress-strain curve, a wheel failure prediction model under deterministic conditions is constructed. The deterministic wheel failure prediction model is used to define the relationship between model parameters and stress-strain. The model parameters include the wheel's geometric dimensions, material properties, and load conditions.

[0006] Sensitivity analysis is performed on the quantified parameters of uncertainty to determine key parameters. Based on the key parameters, sample data under different working conditions are obtained through finite element calculation. Based on the sample data, a maximum strain proxy model of the wheel disk is constructed. The key parameters are input into the maximum strain proxy model of the wheel disk to obtain the confidence interval of the wheel disk rotation speed at the time of failure.

[0007] This invention also provides a device for predicting turbine disk fracture failure, addressing the technical problems of limited test measurement points and insufficient simulation accuracy in existing technologies for aero-engine bladed disk failure. The device includes:

[0008] The engineering and real data conversion module is used to perform uniaxial quasi-static tensile tests on tensile test specimens, obtain the engineering stress-strain curve of the wheel, and convert the engineering stress-strain curve into a true stress-strain curve.

[0009] The failure prediction model construction module is used to construct a wheel failure prediction model under deterministic conditions based on the true stress-strain curve. The wheel failure prediction model under deterministic conditions is used to define the relationship between model parameters and stress-strain. The model parameters include the wheel's geometric dimensions, material properties, and load conditions.

[0010] The confidence interval determination module is used to perform sensitivity analysis on the uncertainty quantification parameters, determine key parameters, and obtain sample data under different working conditions through finite element calculation based on the key parameters. Based on the sample data, a maximum strain proxy model of the wheel disk is constructed, and the key parameters are input into the maximum strain proxy model of the wheel disk to obtain the confidence interval of the wheel disk rotation speed at the time of failure.

[0011] This invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the above-mentioned method for predicting any of the bladed disk fracture failures, thereby solving the technical problems of limited test measurement points and insufficient simulation accuracy when aero-engine bladed disks fail in the prior art.

[0012] This invention also provides a computer-readable storage medium storing a computer program that executes any of the above-described methods for predicting bladed disk failure, in order to solve the technical problems of limited test measurement points and insufficient simulation accuracy in the prior art when aero-engine bladed disks fail.

[0013] Compared with the prior art, the beneficial effects that at least one technical solution adopted in the embodiments of this specification can achieve include at least:

[0014] A proxy model for the maximum strain of a wheel is constructed to overcome the limitations of traditional methods, achieve high-precision full-domain simulation of the wheel's failure process, and solve the problems of limited experimental measurement points and insufficient simulation accuracy. Attached Figure Description

[0015] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart of a method for predicting wheel rupture failure provided in an embodiment of the present invention;

[0017] Figure 2 This is a first schematic diagram of a tensile test specimen provided in an embodiment of the present invention;

[0018] Figure 3 This is a second schematic diagram of the tensile test specimen provided in an embodiment of the present invention;

[0019] Figure 4 This is a schematic diagram of the tensile testing machine provided in an embodiment of the present invention;

[0020] Figure 5 This is a schematic diagram of 10 sets of stretching curves provided in an embodiment of the present invention;

[0021] Figure 6 This is a schematic diagram of a fractured PMMA sample provided in an embodiment of the present invention;

[0022] Figure 7 This is a schematic diagram of the engineering stress-strain curve provided in an embodiment of the present invention;

[0023] Figure 8 This is a schematic diagram of the actual stress-strain curve provided in the embodiments of the present invention;

[0024] Figure 9 This is a flowchart of the wheel over-rotation fracture analysis provided in an embodiment of the present invention;

[0025] Figure 10 This is a schematic diagram of the dimensional tolerances provided in an embodiment of the present invention;

[0026] Figure 11 This is a schematic diagram of the feature point distribution provided in an embodiment of the present invention;

[0027] Figure 12 This is a schematic diagram of the RBF neural network structure provided in an embodiment of the present invention;

[0028] Figure 13 This is a schematic diagram of the actual and predicted strain field results provided in an embodiment of the present invention;

[0029] Figure 14 This is a schematic diagram of the actual and predicted stress field results provided in the embodiments of the present invention;

[0030] Figure 15 This is a schematic diagram illustrating the error between the true and predicted values ​​of the strain field provided in this embodiment of the invention;

[0031] Figure 16 This is a schematic diagram illustrating the error between the actual and predicted values ​​of the stress field provided in this embodiment of the invention;

[0032] Figure 17 This is a schematic diagram of the method for calculating the confidence interval of the fracture speed provided in an embodiment of the present invention;

[0033] Figure 18 This is a schematic diagram of the fracture speed confidence interval provided in an embodiment of the present invention;

[0034] Figure 19 This is a schematic diagram of the finite element calculation of the true value and the surrogate model prediction value provided in the embodiment of the present invention;

[0035] Figure 20 This is a schematic diagram illustrating the accuracy of the corrected proxy model provided in an embodiment of the present invention;

[0036] Figure 21 This is a schematic diagram of the predicted fracture speed range before and after correction provided in an embodiment of the present invention;

[0037] Figure 22 This is a structural block diagram of a computer device provided in an embodiment of the present invention;

[0038] Figure 23 This is a structural block diagram of a wheel rupture failure prediction device provided in an embodiment of the present invention. Detailed Implementation

[0039] The embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0040] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. This application can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0041] In this embodiment of the invention, a method for predicting wheel breakage failure is provided, such as... Figure 1 As shown, the method includes:

[0042] Step S101: Perform a uniaxial quasi-static tensile test on the tensile test specimen to obtain the engineering stress-strain curve of the wheel, and convert the engineering stress-strain curve into a true stress-strain curve;

[0043] Step S102: Based on the true stress-strain curve, construct a wheel failure prediction model under deterministic conditions, wherein the wheel failure prediction model under deterministic conditions is used to define the relationship between model parameters and stress-strain, and the model parameters include the wheel's geometric dimensions, material properties, and load conditions;

[0044] Step S103: Perform sensitivity analysis on the uncertainty quantification parameters to determine key parameters. Based on the key parameters, obtain sample data under different working conditions through finite element calculation. Based on the sample data, construct a maximum strain proxy model for the wheel disk. Input the key parameters into the maximum strain proxy model for the wheel disk to obtain the confidence interval of the wheel disk rotation speed at the time of fracture.

[0045] In specific implementation, the following steps are used to perform a uniaxial quasi-static tensile test on the tensile test specimen, obtain the engineering stress-strain curve of the wheel, and convert the engineering stress-strain curve into a true stress-strain curve:

[0046] Design and fabricate tensile test specimens, conduct uniaxial quasi-static tensile tests on the tensile test specimens, and obtain the engineering stress-strain curves of the wheel;

[0047] Construct the relationship between true stress and load, and the relationship between true strain and the properties of the test specimen; using the relationship between true stress and load, and the relationship between true strain and the properties of the test specimen, construct the relationship between true strain and engineering strain as the first relationship; using the load and the properties of the test specimen, construct the relationship between true stress and engineering stress as the second relationship; based on the first relationship and the second relationship, convert the engineering stress-strain curve into a true stress-strain curve.

[0048] In specific implementation, the following steps are used to construct a deterministic failure prediction model for the wheel disk based on the true stress-strain curve:

[0049] The yield stress-plastic strain relationship is obtained through the true stress-strain curve; the strain value at fracture is obtained based on the yield stress-plastic strain relationship; the strain value at fracture and the yield stress-plastic strain relationship are set as material properties and input into the finite element analysis method, the rotational speed of the wheel is set, and the finite element analysis method is used to determine whether the solution converges; if it converges, the rotational speed of the wheel at fracture is obtained, the finite element calculation model is obtained and used as the wheel failure prediction model under deterministic conditions; if it does not converge, the rotational speed of the wheel is reset until the solution result is converged.

[0050] In practice, the following steps are used to perform sensitivity analysis on the quantitative parameters of uncertainty, determine key parameters, and obtain sample data under different working conditions through finite element calculation based on the key parameters:

[0051] Through uncertainty analysis, multiple uncertainty quantification parameters are selected from the wheel's geometric dimensions, material properties, and load conditions. Sensitivity analysis is performed on the uncertainty quantification parameters at different rotational speeds. Based on the results of the sensitivity analysis, key parameters affecting the model are determined. Upper and lower limits of the key parameters are set, and the key parameters are randomly sampled to calculate sample data under different working conditions.

[0052] In specific implementation, the maximum strain proxy model of the roulette wheel is constructed based on the sample data through the following steps:

[0053] Feature points are collected and mapped, whereby the feature points are used to characterize the response distribution of the roulette wheel; a maximum strain surrogate model of the roulette wheel is constructed based on an RBF neural network, wherein the activation function of the hidden layer neurons of the RBF neural network is... , x For network input, For the first j The center vector of the Gaussian function of each hidden layer neuron. c ji Indicates the first j The first hidden layer neuron center vector i One portion, i = 1, 2,..., m, m The dimension of the network input. b j For the first j The width of the Gaussian function of each hidden layer neuron. S 1 represents the number of neurons; the feature points are used as input, and the stress-strain values ​​of each feature point are output through the maximum strain surrogate model of the roulette wheel.

[0054] In specific implementation, the key parameters are input into the maximum strain surrogate model of the wheel disk through the following steps to obtain the confidence interval of the wheel disk rotation speed at fracture:

[0055] The key parameters are divided into rotational speed and other key parameters. These other key parameters are input into the maximum strain proxy model of the wheel to obtain the predicted maximum fracture strain. Random sampling is performed on the upper and lower limits of these other key parameters to generate discrete points. A fracture rotational speed calculation model is constructed, wherein the fracture rotational speed calculation model uses a bisection method to minimize the difference between the predicted maximum fracture strain and the actual maximum fracture strain, thus obtaining the rotational speed of the wheel at fracture. The discrete points of these other key parameters are input into the fracture rotational speed calculation model to obtain the confidence interval of the wheel's rotational speed at fracture.

[0056] In specific implementation, the digital twin model is corrected based on experimental data through the following steps to generate the maximum strain proxy model of the wheel:

[0057] Correction coefficients are calculated using the strain values ​​obtained from experimental measurements and the strain values ​​predicted by the maximum strain surrogate model of the wheel disk. The average value of these correction coefficients is then calculated to obtain the final correction coefficients. These final correction coefficients are used to correct the maximum strain surrogate model of the wheel disk once, resulting in the first-corrected model. A modified geometric model is generated using the initial geometric model and the geometric correction amount obtained from experimental measurements. Based on the wheel disk fracture experiment results, strain measurements at actual rotational speeds are obtained, minimizing the mean square error between the measured and predicted strain values ​​to obtain the corrected key parameters. Based on these corrected key parameters and the modified geometric model, the maximum strain surrogate model of the wheel disk is corrected a second time, resulting in the second-corrected model.

[0058] The method for predicting fracture failure according to an embodiment of the present invention is as follows:

[0059] Step 1: To obtain the mechanical property parameters of the PMMA (polymethyl methacrylate) material used in the wheel, a standard tensile test specimen needs to be designed and fabricated for uniaxial quasi-static tensile testing. The dimensions and shape of the test specimen used in the test are shown in [reference needed]. Figure 2 , Figure 3 The dimensions of the test specimens are shown in Table 1.

[0060] Table 1 Sample Size Table

[0061]

[0062] All quasi-static tensile tests were performed on a testing machine, which mainly consists of a loading system, a control system, a measurement system, and an output system. A speed loading mode was used during the test, where a servo motor was controlled to drive the lead screw, thereby causing the crossbeam to move up and down at a uniform speed. (See attached image for details about the testing machine.) Figure 4 By conducting uniaxial tensile tests on tensile specimens, stress-strain curves for each group of engineering specimens and the specimens after fracture were obtained, such as... Figure 5 , Figure 6 As shown.

[0063] The optimal average value was taken from the test results of each group, and the resulting tensile stress-strain curves were as follows: Figure 7 As shown.

[0064] The stress-strain curve obtained through tensile testing is an engineering stress-strain curve. The stress is obtained by dividing the external load during the test by the original cross-sectional area of ​​the measuring section (gauge length), and the strain is obtained by dividing the elongation of the measuring section by its original length. However, experiments show that once the material reaches its yield point and enters plastic deformation, the cross-sectional area decreases as the specimen elongates. This leads to the actual stress borne by the specimen differing from the commonly used engineering stress; that is, the commonly used definition of engineering stress cannot accurately reflect the actual stress borne by the specimen. Most engineering components operate within the elastic or just-entering plastic range, so using engineering stress-strain curves to analyze their working state is sufficient for engineering applications. However, to accurately predict the over-rotation fracture speed of the wheel, a true stress-strain curve is necessary.

[0065] True stress is defined as follows: In a tensile test, true stress is the ratio of the load at a certain moment to the actual cross-sectional area at that moment, and it can be calculated by the following formula:

[0066]

[0067] In the formula: True stress; The load at a certain moment; Let be the cross-sectional area at a certain moment.

[0068] Similarly, true strain is defined as the ratio of the elongation of the specimen at a certain moment to the specimen length at that moment, which can be expressed by the following formula:

[0069]

[0070] Therefore, the true strain at a certain moment can be derived from equation (2). Engineering strain at this time The relationship between them is as follows:

[0071]

[0072] In the formula: To be truly adaptable; The length of the sample at a certain moment; This is the original gauge length of the sample; To adapt to engineering contingencies.

[0073] In a tensile test, the plastic deformation of the specimen is a constant-volume change process. Therefore, as the deformation increases, the cross-sectional area of ​​the specimen gradually decreases. From the definitions of true stress and engineering stress, it is known that the true stress generated under the same load is greater than the engineering stress. To derive the relationship between true stress and engineering stress, let's assume that the engineering stress at a certain moment during the tensile process is... The true stress is The load at this time is Then we have:

[0074]

[0075] In the formula: For engineering stress, The original cross-sectional area of ​​the sample. True stress, Let be the cross-sectional area of ​​the sample at a certain moment. This is the load applied at this time. The original gauge length of the sample. Let C be the length of the sample at a certain moment, and C be a constant.

[0076] Formula (4) can be used to derive the following conversion relationship between true stress and engineering stress:

[0077]

[0078] Therefore, through equations (3) and (5), the engineering stress-strain curves obtained in the experiment can be converted into true stress-strain curves, thus laying the foundation for subsequent calculations and analyses. The converted true stress-strain curves are as follows: Figure 8 As shown.

[0079] Step 2: Conduct over-rotation fracture analysis of the wheel using the ultimate strain method. This method combines the wheel's elastoplastic analysis with the local fracture criterion, predicting the fracture speed by tracking the material damage evolution process. As the wheel's rotational speed increases, internal material damage gradually initiates and expands. When the equivalent strain at any local location on the wheel reaches the material's ultimate strain, the wheel is considered to have failed. The flowchart for the wheel's over-rotation fracture analysis is shown below. Figure 9 As shown.

[0080] A parametric analysis program was established based on secondary development using MATLAB and ABAQUS. Basic modeling is achieved using ABAQUS script files. ABAQUS records all operations (including view adjustments and model creation) in real-time in a .rpy file in the working directory, containing complete modeling instructions. Model parameters can be directly defined by modifying the .rpy file, and then changing its extension to .py allows it to be recognized and executed by ABAQUS as a Python script. MATLAB and ABAQUS work together to achieve parameter iteration. The .py script can be opened and edited using MATLAB, and a loop program can be written to adjust model parameters (such as geometric dimensions, material properties, and load conditions) in batches, achieving automated modeling and calculation under multiple working conditions. Through this process, parametric analysis of wheel over-rotation fracture can be efficiently achieved, providing a flexible and reliable modeling tool for failure prediction under different working conditions.

[0081] Specifically, the first step is to set up a finite element calculation model in ABAQUS. (This involves obtaining the yield stress-plastic strain relationship data from the true stress-strain curve and inputting it into the ABAQUS material property settings; obtaining the strain value at fracture based on the yield stress-plastic strain relationship data, and also inputting this strain value into the ABAQUS material property settings; setting the rotational speed of the wheel and checking for convergence; if converged, obtaining the rotational speed at which the wheel breaks; if not converged, resetting the rotational speed of the wheel until the solution converges.) The second step is parametric calculation: all steps from the first step are converted into Python scripts, involving all eight dimensions, four material properties, and one rotational speed of the wheel, which are then converted into model parameters (geometric dimensions, material properties, and loads).

[0082] Step 3: Construct a roulette wheel failure prediction model under uncertainty conditions.

[0083] Multidimensional error analysis was performed on multi-source data from engine disk operation to identify the randomness of variations in various parameters (size, material, load), construct sample statistics, and analyze sample characteristics. Model uncertainty analysis included: quantification of uncertainties in structural dimensional tolerances, material properties, and external loads. Considering both disk design and over-rotation fracture failure analysis, typical disk dimensions 𝒓 (c1, c2, c3, c4, c5, c6, c7, c8) were selected as geometric uncertainty inputs, such as... Figure 10 As shown. Material selection parameters include density, elastic modulus, Poisson's ratio, fracture stress, and fracture strain (M, E, B, stress, strain). The uncertainty of the load is the loading rotational speed. A total of 13 quantitative uncertainty parameters were obtained {c1, c2, c3, c4, c5, c6, c7, c8, M, E, B, stress, strain}, and sensitivity analysis was performed on these 13 quantitative uncertainty parameters at different rotational speeds.

[0084] Sensitivity analysis refers to establishing the relationship between the index y and various quantitative factors in some experiments. The quantitative expression of the correlation between variables (also known as variables) is called the regression equation, which is used to identify the factors that make the indicators meet the requirements. The range of y. We can assume y is related to... The following relationship exists between them:

[0085]

[0086] here yes A function of is often called a response function, and its graph is also called a response surface; It is a random error, usually assumed to have a mean of 0 and a variance of . The distribution follows a normal distribution. Under the above assumptions, It can be seen as being in a given The mean of the subsequent indicators, i.e.

[0087]

[0088] say The space of possible values ​​is the factor space. Find a point within this space. To ensure that E(y) meets the quality requirements, the following can be obtained through optimization methods using the functional form of f. Therefore, to approximate y using a polynomial, we assume:

[0089]

[0090] here These are unknown parameters, also known as regression coefficients, and they typically need to be estimated using collected data. If... To express the corresponding estimate, it is called:

[0091]

[0092] when When it exists, the least squares estimate is:

[0093]

[0094] Thus, the regression equation can be obtained:

[0095]

[0096] Taking two independent factors as an example, after obtaining the "factor-response" information, a standard second-order polynomial response surface is established to approximate the test points, as shown below:

[0097]

[0098] Differentiating this second-order polynomial, we get:

[0099]

[0100] Therefore, the linear main effect is:

[0101]

[0102] The second-order main effect is:

[0103]

[0104] The interaction effect is:

[0105]

[0106] in, This is the baseline value for the factor.

[0107] The main effect is the average response across all factor experiments at all given levels. The main effect provides an average measure of how a factor influences its effect. Its main principles are as follows:

[0108] (1) Numerical value (absolute value): The larger the main effect value, the greater the influence of the factor on the response.

[0109] (2) Direction: A positive main effect means that the response will increase as the factor increases, while a negative main effect means that the response will decrease as the factor increases.

[0110] (3) Order: The existence of a second-order main effect means that the influence of the factor on the response is not linear.

[0111] (4) Interaction effect: A larger interaction effect means that the impact on the response will be greater when both parameters change at the same time.

[0112] After normalizing the input variables to [-1, +1], a second-order polynomial is obtained by fitting the model using the least squares method. The new model coefficients can then be derived from these coefficients. (i.e., the main effect), this coefficient can more fairly reflect the influence of input variables on the response. After standardization, the result is converted into a percentage contribution rate:

[0113]

[0114] The Monte Carlo distribution sampling method was used to sample 13 uncertainty quantification parameters, and then sensitivity analysis was performed to obtain the key parameters affecting the model. The Pareto plot reflects the percentage contribution of all terms in the model to each response after sample fitting.

[0115] Eight key parameters were identified: inner thickness, outer diameter, density, Poisson's ratio, fracture strain (fitted using P1, P2, and P3), and rotational speed. Random sampling was performed on these eight key parameters with upper and lower limits set, and the data were input into a parameterized program for calculation. Sample data under different working conditions were obtained, and a maximum strain proxy model for the wheel disk was then constructed.

[0116] Step 4: Input the key parameters into the parameterization program to obtain sample data and establish a maximum strain proxy model for the roulette wheel. Research is conducted on model order reduction methods for finite element models with high computational costs in damage models.

[0117] First, feature points are collected and mapped. These feature points are used to characterize the response distribution of the roulette wheel, such as... Figure 11As shown, the collected feature points consist of two parts: 1. Uniformly distributed points used to describe the overall response field distribution; 2. Key locations used to describe the distribution of high-stress strain response related to fracture.

[0118] Then, a reduced-order model corresponding to the maximum stress is established based on RBF. The activation function of the RBF neural network is the radial basis function. The activation ability of a neuron is proportional to the distance of its input from the center of the radial basis function. A weight in an RBF neural network only affects one output point; therefore, RBF exhibits the characteristic of local mapping. The structure of the RBF neural network is as follows: Figure 12 As shown.

[0119] Suppose the input to the RBF neural network is n-dimensional, and the hidden layers have... S One neuron, with the output layer having S 2 neurons. The activation function of the hidden layer neurons in the RBF neural network is shown in Equation (19), which is the Gaussian function.

[0120]

[0121] in, For the first j The center vector of the Gaussian function of each hidden layer neuron, if the hidden layer has n one neuron, then j = 1, 2, ..., n . c ji Indicates the first j The first hidden layer neuron center vector i One portion, i =1, 2, ..., m , m The dimension of the network input. b j For the first j The width of the Gaussian function of each hidden layer neuron. b j The larger the value, the stronger the Gaussian function's ability to map the input. c j This represents the center position of the non-zero output region. The larger the output value, the more sensitive the Gaussian function is to the input. W This is the weight matrix of the output layer. R This is the output matrix of the hidden layer. Studies have shown that... It is the main parameter that determines the performance of RBF neural networks.

[0122] Based on the above method, a reduced-order response field model was constructed. The model's input consists of eight key parameters, and the output is the stress and strain values ​​at each feature point. Inner thickness = 9.621, outer diameter = 324.241, density = 1.179, Poisson's ratio = 0.486, tensile parameters: P1 = 28.463, P2 = -3.798, P3 = -0.070, rotational speed = 324.691. The strain, stress field, and error between the actual and predicted values ​​under the working condition are compared as follows: Figure 13 , Figure 14 , Figure 15 and Figure 16 As shown.

[0123] Step 5: Based on the maximum strain surrogate model of the wheel ( Figure 17 Model 1 (using the input of 7 key parameters and initialization of rotational speed) obtains the predicted maximum fracture strain value. A bisection method is used to minimize the difference between the predicted and actual maximum fracture strain values, thus obtaining the fracture rotational speed. Simultaneously, random sampling is performed based on the upper and lower limits of the 7 key parameters to obtain parameter discrete points, which are then input into Model 2 (…). Figure 17 The confidence interval for the fracture speed is obtained from model 2), and the specific method is as follows: Figure 17 As shown, the specific fracture speed confidence interval is as follows: Figure 18 As shown.

[0124] Step Six: Based on digital twin technology, the maximum strain proxy model of the wheel is corrected by integrating the digital model and the experimental physical model. Specific research content includes: a model parameter pre-correction method based on experimental data, and an online strain field correction technology based on key measurement point data.

[0125] 6.1 A method for pre-correcting model parameters based on experimental data.

[0126] To address the discrepancy between simulation and experimental results, a model parameter pre-correction method based on experimental data is proposed. This method calculates the correction coefficient for each operating condition by comparing the actual experimental value with the predicted value of the surrogate model. The final correction coefficient is then obtained by averaging these coefficients. Subsequently, the original surrogate model is multiplied by the correction coefficient to obtain the corrected surrogate model, thereby significantly improving prediction accuracy. This method, through iterative optimization, effectively reduces prediction errors and enhances the model's applicability under complex operating conditions.

[0127] Given a set of experimental data points , This represents an n-dimensional input vector, and the corresponding true output value is... Original proxy model Used to predict output The pre-correction method first calculates the correction coefficient:

[0128]

[0129] in, The strain value is obtained from experimental measurement; The strain value predicted by the surrogate model; This represents the correction coefficient for the i-th measurement point, reflecting the deviation between the original surrogate model's predicted value and the actual value.

[0130] Based on the set of correction coefficients The final correction factor is obtained by averaging. The modified maximum strain surrogate model for the roulette wheel. By multiplying the maximum strain proxy model of the wheel disk by the correction coefficient, we obtain:

[0131]

[0132] 6.2. Strain field online correction technology based on key measurement point data.

[0133] In the design and manufacturing process of a wheel, the prediction results of the digital model often differ significantly from the actual physical behavior due to factors such as manufacturing tolerances, material inhomogeneity, or loading conditions. To improve the accuracy and reliability of the digital model, this paper proposes an online strain field correction method based on key measurement point data. This method consists of two main steps: geometric model adjustment and correction of material parameters and failure criteria.

[0134] First, the actual geometric dimensions of the wheel test piece are obtained through high-precision coordinate measuring machine (CMM) measurement. Based on these measurement results, the geometric parameters of the digital model are adjusted to eliminate deviations caused by manufacturing tolerances and geometric defects. The adjusted geometric model can be represented as:

[0135]

[0136] in, For the initial geometric model, This is a geometric correction amount obtained based on experimental measurements.

[0137] Then, combining the results of the wheel fracture experiment, the strain measurement value at the actual rotational speed is obtained, and the mean square error between it and the strain value predicted by the model is minimized. This leads to the obtained corrected key parameters, with the optimization objective being:

[0138]

[0139] in: is the corrected key parameter; N is the number of measurement points.

[0140] Corrected key parameters and geometric model Together, they form an optimized digital twin model, whose predictions more closely approximate physical experimental behavior. This method systematically corrects geometric and material parameters by combining experimental data and statistical inference, ensuring consistency between the digital model and the physical model in key performance indicators.

[0141] 6.3 Correction of the digital twin model for super-rotation fracture based on experimental data

[0142] Before the experiment began, a dynamic balancing test was performed on the test piece using a dynamic balancing test device to ensure the smooth progress of the over-rotation fracture test. Then, a coordinate measuring machine was used to measure the test wheel to obtain the key dimensional parameters of the wheel: x1=10, x2=325, and x8=9.621.

[0143] Initialize the elastic modulus x12 = 1.18, Poisson's ratio x13 = 0.49, and tensile parameter x14 = 1. Calculate the maximum strain using finite element analysis (FEM). Simultaneously, predict the maximum strain based on the established RBF surrogate model. The actual FEM values ​​and the surrogate model predictions are shown below. Figure 19 As shown.

[0144] Dynamic strain data is obtained through a data acquisition system, while velocity and vibration data are acquired from the PLC.

[0145] The surrogate model is corrected based on the collected strain values, and the accuracy of the corrected surrogate model is as follows: Figure 20 As shown.

[0146] Based on the modified surrogate model, the strain field at different rotational speeds is predicted, and the fracture rotational speed is predicted by combining the modified failure criterion. Finally, the fracture rotational speed prediction range before and after modification is as follows: Figure 21 As shown in the figure. These methods enable high-precision and high-efficiency prediction of wheel failure, providing strong support for the design and maintenance of wheel.

[0147] In the above embodiments, the prediction method of the present invention is used to predict the failure of the impeller disk. In other embodiments, it can also predict the failure of the bladed disk.

[0148] In this embodiment, a computer device is provided, such as... Figure 22 As shown, it includes a memory 2201, a processor 2202, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the above-described method for predicting any of the wheel breakage failures.

[0149] Specifically, the computer device can be a computer terminal, a server, or a similar computing device.

[0150] In this embodiment, a computer-readable storage medium is provided, which stores a computer program that performs any of the above-described methods for predicting wheel breakage failure.

[0151] Specifically, computer-readable storage media include both permanent and non-permanent, removable and non-removable media, which can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer-readable storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable storage media do not include transient media, such as modulated data signals and carrier waves.

[0152] Based on the same inventive concept, this invention also provides a device for predicting wheel breakage failure, as described in the following embodiments. Since the principle of the device for predicting wheel breakage failure is similar to that of the method for predicting wheel breakage failure, the implementation of the device for predicting wheel breakage failure can refer to the implementation of the method for predicting wheel breakage failure; repeated details will not be elaborated further. As used below, the terms "unit" or "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0153] Figure 23 This is a structural block diagram of a wheel rupture failure prediction device according to an embodiment of the present invention, such as... Figure 23 As shown, it includes: an engineering and real data conversion module 2301, a failure prediction model construction module 2302, and a confidence interval determination module 2303. The structure is described below.

[0154] The engineering and real data conversion module 2301 is used to perform a uniaxial quasi-static tensile test on the tensile test specimen, obtain the engineering stress-strain curve of the wheel, and convert the engineering stress-strain curve into a true stress-strain curve.

[0155] The failure prediction model construction module 2302 is used to construct a wheel failure prediction model under deterministic conditions based on the true stress-strain curve. The wheel failure prediction model under deterministic conditions is used to define the relationship between model parameters and stress-strain. The model parameters include the wheel's geometric dimensions, material properties, and load conditions.

[0156] The confidence interval determination module 2303 is used to perform sensitivity analysis on the uncertainty quantification parameters, determine key parameters, and obtain sample data under different working conditions through finite element calculation based on the key parameters. Based on the sample data, a maximum strain proxy model of the wheel disk is constructed, and the key parameters are input into the maximum strain proxy model of the wheel disk to obtain the confidence interval of the wheel disk rotation speed at the time of failure.

[0157] In one embodiment, the engineering-to-real data conversion module includes:

[0158] The engineering curve generation unit is used to design and process tensile test specimens, perform uniaxial quasi-static tensile tests on the tensile test specimens, and obtain the engineering stress-strain curve of the wheel.

[0159] Load property relationship building unit, used to build the relationship between true stress and load and the relationship between true strain and the properties of the test specimen;

[0160] The first relational construction unit is used to construct the relationship between true strain and engineering strain as the first relational expression through the relationship between true stress and load, and the relationship between true strain and the properties of the test specimen.

[0161] The second relational construction unit is used to construct a relational expression between true stress and engineering stress as a second relational expression based on the load and the properties of the test specimen.

[0162] The true curve generation unit is used to convert the engineering stress-strain curve into a true stress-strain curve based on the first relation and the second relation.

[0163] In one embodiment, the failure prediction model building module includes:

[0164] The stress-strain relationship construction unit is used to obtain the yield stress-plastic strain relationship through the true stress-strain curve;

[0165] The strain value calculation unit is used to obtain the strain value at fracture based on the yield stress-plastic strain relationship.

[0166] The convergence judgment unit is used to input the strain value at fracture and the yield stress-plastic strain relationship as material properties into the finite element analysis method, set the rotation speed of the wheel, and use the finite element analysis method to solve whether convergence has occurred.

[0167] A model building unit is defined, which is used to obtain the rotational speed of the wheel at the time of fracture if convergence is achieved, to obtain a finite element calculation model and use it as a wheel failure prediction model under deterministic conditions. If convergence is not achieved, the rotational speed of the wheel is reset until the solution result is converged.

[0168] In one embodiment, the confidence interval determination module includes:

[0169] The uncertainty parameter determination unit is used to select multiple uncertainty quantification parameters from the disk's geometric dimensions, material properties, and load conditions through uncertainty analysis.

[0170] The key parameter determination unit is used to perform sensitivity analysis on the uncertain quantitative parameters at different rotational speeds, and determine the key parameters affecting the model based on the results of the sensitivity analysis.

[0171] The sample data calculation unit is used to set the upper and lower limits of the key parameters, and to randomly sample the key parameters to calculate sample data under different working conditions.

[0172] In one embodiment, the confidence interval determination module further includes:

[0173] A feature point acquisition unit is used to acquire and map feature points, wherein the feature points are used to characterize the response distribution of the roulette wheel;

[0174] A model unit is constructed to build a maximum strain surrogate model for a roulette wheel based on an RBF neural network, wherein the activation function of the hidden layer neurons of the RBF neural network is... , x For network input, For the first j The center vector of the Gaussian function of each hidden layer neuron. c ji Indicates the first j The first hidden layer neuron center vector i One portion, i = 1, 2, ..., m, m The dimension of the network input. b j For the first j The width of the Gaussian function of each hidden layer neuron. S 1 represents the number of neurons;

[0175] The stress-strain value calculation unit is used to take the feature points as input and output the stress-strain values ​​of each feature point through the maximum strain proxy model of the wheel.

[0176] In one embodiment, the confidence interval determination module further includes:

[0177] The maximum predicted value prediction unit is used to divide the key parameters into rotational speed and other key parameters besides the rotational speed, and input the other key parameters into the maximum strain proxy model of the wheel to obtain the maximum fracture strain prediction value;

[0178] The discrete point generation unit is used to randomly sample the upper and lower limits of the other key parameters to generate discrete points of the other key parameters.

[0179] The rotational speed calculation unit is used to construct a fracture rotational speed calculation model, wherein the fracture rotational speed calculation model uses a bisection method to minimize the difference between the predicted value of the maximum fracture strain and the actual value of the maximum fracture strain, thereby obtaining the rotational speed of the disk at the time of fracture.

[0180] The confidence interval generation unit is used to input the discrete points of the other key parameters into the fracture speed calculation model to obtain the confidence interval of the wheel speed at fracture.

[0181] In one embodiment, the above-described apparatus further includes a correction model module.

[0182] In one embodiment, the model correction module includes:

[0183] The correction coefficient calculation unit is used to calculate the correction coefficient by using the strain value obtained by experimental measurement and the strain value predicted by the maximum strain surrogate model of the wheel disk, and calculate the average value based on the set of correction coefficients to obtain the final correction coefficient;

[0184] The first correction unit is used to correct the maximum strain proxy model of the wheel disk using the final correction coefficient, so as to obtain the maximum strain proxy model of the wheel disk after the first correction.

[0185] The geometric model generation unit is used to generate a modified geometric model from an initial geometric model and geometric corrections based on experimental measurements.

[0186] The key parameter correction unit is used to obtain strain measurement values ​​at actual rotational speed based on the results of the wheel disk fracture experiment, so as to minimize the mean square error between the strain measurement values ​​and the predicted strain values, and obtain the corrected key parameters.

[0187] The second correction unit is used to perform a second correction on the maximum strain proxy model of the wheel disk based on the corrected key parameters and the adjusted geometric model, so as to obtain the maximum strain proxy model of the wheel disk after the second correction.

[0188] The embodiments of the present invention achieve the following technical effects:

[0189] By introducing a maximum strain proxy model for the wheel disk, uncertainty analysis, and a digital twin model of over-rotation fracture based on experimental data, the limitations of traditional methods are overcome, achieving high-precision full-domain simulation of the wheel disk fracture process. This solves the problems of limited experimental measurement points and insufficient simulation accuracy, providing stronger support for engineering applications. Based on real-time data, the maximum strain proxy model for the wheel disk in this embodiment of the invention is corrected. Before correction, the fracture speed was 2930 RPM, and after correction, the fracture speed was 2558 RPM. The actual fracture speed was 2500 RPM, and the accuracy of fracture speed prediction was significantly improved after correction.

[0190] Obviously, those skilled in the art should understand that the modules or steps of the above-described embodiments of the present invention can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those presented here, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the embodiments of the present invention are not limited to any particular hardware and software combination.

[0191] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, various modifications and variations can be made to the embodiments of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method of predicting wheel disc breakage failure, characterized by, The method comprises the following steps: performing uniaxial quasi-static tensile test on a tensile test sample to obtain an engineering stress-strain curve of the disc, and converting the engineering stress-strain curve into a true stress-strain curve; based on the true stress-strain curve, a disc failure prediction model under deterministic conditions is constructed, wherein the disc failure prediction model under deterministic conditions is used to define the relationship between model parameters and stress-strain, and the model parameters include the geometric size, material properties and load conditions of the disc; based on the true stress-strain curve, a disc failure prediction model under deterministic conditions is constructed, comprising: obtaining the yield stress-plastic strain relationship through the true stress-strain curve; obtaining the strain value at the time of fracture based on the yield stress-plastic strain relationship; setting the strain value at the time of fracture and the yield stress-plastic strain relationship as material properties, inputting them into a finite element analysis method, setting the rotating speed value of the disc, and using the finite element analysis method to solve whether it converges; if it converges, obtaining the rotating speed value of the disc at the time of fracture, obtaining the finite element calculation model and taking it as the disc failure prediction model under deterministic conditions, and if it does not converge, resetting the rotating speed value of the disc until the result of the solution is convergence; performing sensitivity analysis on the uncertainty quantification parameters to determine the key parameters, obtaining sample data under different working conditions through finite element calculation based on the key parameters, and constructing a disc maximum strain surrogate model based on the sample data, inputting the key parameters into the disc maximum strain surrogate model to obtain the confidence interval of the rotating speed of the disc at the time of fracture; constructing a disc maximum strain surrogate model, comprising: collecting and mapping feature points, wherein the feature points are used to represent the response distribution of the disc; A maximum strain surrogate model for a roulette wheel is constructed based on an RBF neural network, wherein the activation function of the hidden layer neurons in the RBF neural network is... , x For network input, For the first j The center vector of the Gaussian function of each hidden layer neuron. c ji Indicates the first j The first hidden layer neuron center vector i One portion, i = 1, 2, ..., m, m The dimension of the network input. b j For the first j The width of the Gaussian function of each hidden layer neuron. S 1 represents the number of neurons; inputting the feature points as input, and outputting the stress-strain value of each feature point through the disc maximum strain surrogate model.

2. The wheel disk rupture failure prediction method according to claim 1, wherein performing uniaxial quasi-static tensile test on a tensile test sample to obtain an engineering stress-strain curve of the disc, and converting the engineering stress-strain curve into a true stress-strain curve, comprising: designing and processing a tensile test sample, performing uniaxial quasi-static tensile test on the tensile test sample, and obtaining an engineering stress-strain curve of the disc; constructing a relationship between true stress and load and a relationship between true strain and properties of the test sample; constructing a relationship between true strain and engineering strain as a first relationship through the relationship between true stress and load and the relationship between true strain and properties of the test sample; constructing a relationship between true stress and engineering stress as a second relationship through the load and the properties of the test sample; based on the first relationship and the second relationship, the engineering stress-strain curve is converted into a true stress-strain curve.

3. The wheel disk rupture failure prediction method according to claim 1, characterized by, performing sensitivity analysis on the uncertainty quantification parameters to determine the key parameters, and obtaining sample data under different working conditions through finite element calculation based on the key parameters, comprising: selecting a plurality of uncertainty quantification parameters from the geometric size, the material properties and the load conditions of the disc through uncertainty analysis; Performing sensitivity analysis on the uncertain quantification parameters under different rotating speeds, determining key parameters affecting the model based on the results of the sensitivity analysis; Setting upper and lower limits of the key parameters and performing random sampling on the key parameters to obtain sample data under different working conditions.

4. The wheel disk rupture failure prediction method according to claim 1, wherein Inputting the key parameters into the wheel disc maximum strain surrogate model to obtain a confidence interval of the wheel disc rotating speed at the time of rupture, including: Dividing the key parameters into rotating speed and other key parameters except the rotating speed, inputting the other key parameters into the wheel disc maximum strain surrogate model to obtain a maximum fracture strain prediction value; Randomly sampling upper and lower limits of the other key parameters to generate other key parameter discrete points; Building a rupture rotating speed calculation model, wherein the rupture rotating speed calculation model minimizes the difference between the maximum fracture strain prediction value and the true value of the maximum fracture strain through bisection to obtain the rotating speed of the wheel disc at the time of rupture; Inputting the other key parameter discrete points into the rupture rotating speed calculation model to obtain a confidence interval of the wheel disc rotating speed at the time of rupture.

5. The wheel disk breakage failure prediction method according to any one of claims 1 to 4, characterized in that, Further comprising: Based on digital twinning technology, correcting the built wheel disc maximum strain surrogate model, including: Calculating a correction coefficient based on the strain value measured through experiments and the strain value predicted by the wheel disc maximum strain surrogate model, calculating an average value based on a set of correction coefficients to obtain a final correction coefficient; Performing a correction on the wheel disc maximum strain surrogate model through the final correction coefficient to obtain a first corrected wheel disc maximum strain surrogate model; Generating a corrected geometric model through an initial geometric model and a geometric correction amount obtained based on experimental measurements; Based on the wheel disc rupture experimental results, obtaining strain measurement values under actual rotating speeds, minimizing the mean square error between the strain measurement values and the predicted strain values to obtain corrected key parameters; Based on the corrected key parameters and the corrected geometric model, performing a second correction on the wheel disc maximum strain surrogate model to obtain a second corrected wheel disc maximum strain surrogate model.

6. A wheel disc breakage failure prediction device characterized by comprising: Comprising: An engineering and true data conversion module for performing uniaxial quasi-static tensile test on a tensile test sample to obtain an engineering stress-strain curve of the wheel disc, and converting the engineering stress-strain curve into a true stress-strain curve; A failure prediction model construction module for constructing a wheel disc failure prediction model under deterministic conditions based on the true stress-strain curve, wherein the wheel disc failure prediction model under deterministic conditions is used to define the relationship between model parameters and stress-strain, and the model parameters include the geometric size, material properties, and load conditions of the wheel disc; The failure prediction model construction module comprises: A stress-strain relationship construction unit for obtaining a yield stress-plastic strain relationship through the true stress-strain curve; A strain value calculation unit for obtaining a strain value at the time of rupture based on the yield stress-plastic strain relationship; A convergence judgment unit for inputting the strain value at the time of rupture and the yield stress-plastic strain relationship as material properties into a finite element analysis method, setting a rotating speed value of the wheel disc, and using the finite element analysis method to solve whether to converge; A determination model construction unit is configured to, if the convergence is obtained, obtain a rotating speed value of the wheel disc at the time of the rupture, obtain a finite element calculation model and take the finite element calculation model as a wheel disc failure prediction model under a deterministic condition, and if the convergence is not obtained, reset the rotating speed value of the wheel disc until a result of the solving is the convergence; A determination confidence interval module is configured to perform a sensitivity analysis on an uncertainty quantification parameter, determine a key parameter, obtain sample data under different working conditions through finite element calculation based on the key parameter, and construct a wheel disc maximum strain surrogate model based on the sample data, input the key parameter into the wheel disc maximum strain surrogate model, and obtain a confidence interval of a wheel disc rotating speed at the time of the rupture. The determination confidence interval module comprises: A feature point acquisition unit is configured to acquire and map feature points, wherein the feature points are used to represent a response distribution of the wheel disc. A model unit is constructed to build a maximum strain surrogate model for a roulette wheel based on an RBF neural network, wherein the activation function of the hidden layer neurons of the RBF neural network is... , x For network input, For the first j The center vector of the Gaussian function of each hidden layer neuron. c ji Indicates the first j The first hidden layer neuron center vector i One portion, i = 1, 2, ..., m, m The dimension of the network input. b j For the first j The width of the Gaussian function of each hidden layer neuron. S 1 represents the number of neurons; A stress-strain value calculation unit is configured to take the feature points as inputs, and output stress-strain values of the feature points through the wheel disc maximum strain surrogate model.

7. A computer device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor implements the wheel disc rupture failure prediction method in any one of claims 1 to 5 when executing the computer program.

8. A computer-readable storage medium, characterized in that, The computer readable storage medium stores the computer program for executing the wheel disc rupture failure prediction method in any one of claims 1 to 5.

Images