A rapid prediction method for stress and fatigue failure of axial-flow propeller turbines

Abstract

The present invention discloses a method for rapidly predicting stress and fatigue damage in an axial-flow propeller turbine. The method comprises the following steps: S1. Extracting torque data from turbine blades through CFD, performing VMD decomposition and main mode selection on the torque data, and screening the main modes; S2. Inputting the main modes into a GMM, extracting amplitude, frequency, and phase, and reconstructing the torque data based on the amplitude, frequency, and phase synthesis; S3. Inputting the reconstructed torque data into an LSTM network, outputting the stresses of the blades and rotor arms, using the reconstructed torque data as FSI load boundary conditions, and combining it with ROM to calculate component stresses; S4. Predicting fatigue damage based on the calculated stresses using the Paris formula and Miner criterion. Through VMD decomposition and parametric reconstruction techniques, the data volume is reduced by over 50% while retaining the physical meaning of the main torque modes. The main mode energy retention rate exceeds 95%, significantly improving signal processing efficiency.

Description

Technical Field

[0001] The invention belongs to the technical field of hydropower generation equipment, and in particular relates to a method for quickly predicting stress and fatigue damage of an axial-flow propeller-type turbine. Background Art

[0002] When an axial-flow propeller turbine operates under variable operating conditions, the unsteady fluid loads on the blades will cause alternating stress concentration and fatigue damage accumulation, directly affecting the safety and life of the unit. In existing technologies, the prediction of dynamic loads mainly relies on high-precision CFD simulations, but the calculations are too time-consuming and difficult to meet the needs of rapid engineering iterations. At the same time, traditional signal processing methods (such as empirical mode decomposition) are susceptible to modal aliasing and noise interference, resulting in insufficient accuracy in the extraction of the main torque modes. In addition, fatigue life analysis is mostly based on the static stress assumption, ignoring the time series characteristics of dynamic loads, and the high computational complexity of the full-order fluid-solid coupling model further limits the analysis efficiency. These defects make it difficult for existing methods to achieve rapid and accurate predictions of dynamic stress and fatigue life in engineering practice. Summary of the Invention

[0003] In response to the above-mentioned deficiencies in the prior art, the present invention provides a method for rapid prediction of stress and fatigue damage of an axial-flow propeller turbine, which solves the problems of low efficiency of dynamic load analysis, insufficient time series modeling, and complex multi-physical field coupling calculations in traditional methods.

[0004] In order to achieve the above-mentioned object of the invention, the technical solution adopted by the present invention is: a method for quickly predicting stress and fatigue failure of an axial-flow propeller turbine, comprising the following steps:

[0005] S1. Extract the torque data of the turbine blades through CFD, perform VMD decomposition and main mode selection on the torque data, and screen the main mode;

[0006] S2. Input the main mode into GMM, extract the amplitude, frequency and phase, and reconstruct the torque data based on the amplitude, frequency and phase;

[0007] S3. Input the reconstructed torque data into the LSTM network, output the stress of the blade and the arm, use the reconstructed torque data as the FSI load boundary condition, and combine it with the ROM to calculate the stress of the component;

[0008] S4. Predict fatigue failure based on the calculated stress using the Paris formula and Miner criterion.

[0009] Further: S1 includes the following sub-steps:

[0010] S11. Simulate the flow field around the turbine blades through computational fluid dynamics and extract the dynamic torque data of the blade surface;

[0011] S12, CFD solution settings: Input dynamic torque data, set the corresponding boundary conditions and time step based on the adaptive grid and SST k-ω model, and output the torque signal. The boundary conditions include the inlet velocity, outlet pressure, and no-slip boundary conditions on the wall;

[0012] S13. Use VMD to decompose torque data into modal components with clear physical meanings and screen key modes related to rotational frequency.

[0013] Furthermore, in S11, the method for extracting the torque data of the turbine blades is specifically as follows:

[0014] The dynamic pressure distribution on the turbine blade surface is calculated through transient CFD simulation based on adaptive mesh and SST k-ω model to generate torque data.

[0015] Further: S13 is specifically:

[0016] Torque data is obtained through VMD solution K IMFs, calculate the energy of each IMF, normalize the energy of each IMF, sort the IMFs in descending order according to the energy size, and select the first m The IMF is selected as the main mode so that the cumulative energy of the selected IMF accounts for more than 80% of the energy of all IMFs.

[0017] Further: S2 includes the following sub-steps:

[0018] S21. Input the main mode into the GMM, and extract the parameters of each Gaussian distribution from the GMM. The parameters of the Gaussian distribution include mean, variance, and weight.

[0019] S22, converting the parameters of the Gaussian distribution into amplitude, frequency and phase;

[0020] S23. Synthesize and reconstruct torque data based on the extracted amplitude, frequency, and phase.

[0021] Further: In S23, the torque data is reconstructed The specific expression is:

[0022]

[0023] Where, A i is the amplitude, f i is the frequency, is the phase, M The number of main modes.

[0024] Furthermore, in S3, the components include a connecting plate, a rotating arm, and a blade, and S3 includes the following sub-steps:

[0025] S31, inputting the reconstructed torque data into the LSTM network, and outputting the stress of the connecting plate, the arm, and the blade through the LSTM network;

[0026] S32. Use ROM to model the FSI problem, input the reconstructed torque data into ROM, and calculate the stress of the connecting plate, the arm, and the blade through ROM;

[0027] S33. Integrate the stresses of the connecting plate, rotating arm, and blade output by the LSTM network and the ROM to obtain the final stresses of the connecting plate, rotating arm, and blade.

[0028] Further: In S4, the paris formula is specifically:

[0029]

[0030] Where, is the fatigue crack growth rate, that is, the rate of change of crack length a with the number of cycles N, C and n are constants related to material properties, environment and other factors, is the stress intensity factor amplitude;

[0031] The Miner criterion formula is as follows:

[0032]

[0033] Where, is the number of cycles of the material under the stress level i, For the material i The criterion is used to estimate the cumulative fatigue damage of a material under cyclic loading at different stress levels. When the cumulative damage reaches 1, the material is considered to have suffered fatigue failure.

[0034] The beneficial effects of the present invention are:

[0035] (1) The present invention provides a method for rapid prediction of stress and fatigue damage of axial-flow propeller turbines. Through VMD decomposition and parametric reconstruction technology, the data volume is reduced by more than 50% while retaining the physical meaning of the main torque mode. The main mode energy retention rate exceeds 95%, significantly improving the signal processing efficiency.

[0036] (2) The present invention adopts LSTM network to learn the time series load characteristics and combines ROM to accelerate fluid-solid coupling calculation, so that the dynamic stress prediction error is less than 5% and the calculation time is reduced by 70%.

[0037] (3) This paper proposes a dynamic fatigue coupling model that integrates the temporal load sequence effect with the multiaxial fatigue criterion, achieving a fatigue life prediction result with a degree of agreement of over 90% with experimental data. This method overcomes the limitations of traditional static assumptions and full-order models, providing an efficient and highly accurate solution for real-time health monitoring and life assessment of key turbine components.

[0038] (4) The method of the present invention shortens the full-process analysis cycle from 2 weeks of the traditional method to 3 days, significantly improving the efficiency of turbine design optimization and the reliability of operation and maintenance decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 The present invention is a flowchart of a method for rapidly predicting stress and fatigue damage of an axial-flow propeller-type turbine. DETAILED DESCRIPTION

[0040] The specific embodiments of the present invention are described below to facilitate understanding of the present invention by those skilled in the art. However, it should be clear that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the appended claims, these changes are obvious, and all inventions and creations utilizing the concepts of the present invention are protected.

[0041] like Figure 1 As shown, in one embodiment of the present invention, a method for quickly predicting stress and fatigue failure of an axial-flow propeller turbine includes the following steps:

[0042] S1. Extract the torque data of the turbine blades through CFD (computational fluid dynamics), perform VMD (variational mode decomposition) on the torque data and select the main mode to screen the main mode;

[0043] S2. Input the main mode into GMM (Gaussian mixture model), extract the amplitude, frequency and phase, and reconstruct the torque data based on the amplitude, frequency and phase;

[0044] S3. Input the reconstructed torque data into the LSTM network, output the stress of the blade and arm, use the reconstructed torque data as the FSI (fluid-structure interaction) load boundary condition, and combine it with the ROM (reduced-order model) to calculate the stress of the component;

[0045] S4. Predict fatigue failure based on the calculated stress using the Paris formula and Miner criterion.

[0046] In S1, the method for extracting the torque data of the turbine blades is specifically as follows:

[0047] The dynamic pressure distribution on the turbine blade surface is calculated through transient CFD simulation based on adaptive mesh and SST k-ω model to generate torque data.

[0048] S1 includes the following steps:

[0049] S11. Use computational fluid dynamics (CFD) to simulate the flow field around the turbine blades and extract dynamic torque data on the blade surface. In this embodiment, ANSYS BladeModeler or SolidWorks is used to build a three-dimensional blade model. Hybrid meshing (boundary layer structured mesh + external unstructured mesh) is employed.

[0050] S12, CFD solution settings: Input dynamic torque data, set the corresponding boundary conditions and time step based on the adaptive grid and SST k-ω model, and output the torque signal. The boundary conditions include the inlet velocity, outlet pressure, and no-slip boundary conditions on the wall;

[0051]

[0052] Where, is the torque data, is the position vector, is the normal vector, is the direction of the rotation axis, For the i The fluid shear stress vector at the infinitesimal point is, For the i The area of ​​a microelement, N is the total number of discrete elements on the blade surface, i is the index number of the blade surface mesh, For the i The static pressure of the fluid at each infinitesimal point;

[0053] S13. Decompose the torque data into modal components with clear physical meanings, screen the key modes related to the rotational frequency, and perform VMD decomposition and main mode selection on the torque data as follows:

[0054] The torque data is decomposed into several intrinsic mode functions using VMD. The main modes are selected based on energy proportion (>80%) and frequency domain correlation (matching the rotation frequency and its harmonics). The specific method is as follows:

[0055] Torque data is obtained through VMD solution K IMFs (Intrinsic Mode Functions), calculate the energy of each IMF, normalize the energy of each IMF, sort the IMFs in descending order according to the energy size, and select the top m The IMF is selected as the main mode so that the cumulative energy of the selected IMF accounts for more than 80% of the energy of all IMFs.

[0056] In this embodiment, VMD decomposition is an adaptive decomposition method for non-stationary signals, which can search for the optimal variational model through multiple iterations. It is assumed that the wind turbine active power signal to be decomposed is composed of several IMF components, each of which can be regarded as an amplitude-frequency modulation signal.

[0057] First, by performing Hilbert transform on each modal function, the analytical values ​​of different modal functions can be obtained:

[0058]

[0059] Where uk and ωk are the modal functions and center frequencies (k∈1,2,...,K), K is the number of modes obtained by decomposition; ∂t is the gradient operation, d ( t ) is the unit pulse function, t is the time, j is the imaginary number, and e is the natural exponent.

[0060] After that, the center frequency of the above formula is estimated and multiplied by complex exponential functions to transfer the spectrum distribution of each modal function to its respective frequency baseband, which can be obtained:

[0061]

[0062] By using Hilbert transform in the frequency domain to process the collected wind turbine active power, we can obtain the constrained variational optimization problem:

[0063]

[0064] By introducing the combination of quadratic penalty function α and Lagrange multiplier λ, the above optimization problem is transformed into an unconstrained optimization problem:

[0065]

[0066] In the above formula, the alternating direction multiplier method is introduced into the unconstrained optimization problem, and the initialization formulas of {uk}, {ωk}, and λ of each sub-signal are:

[0067]

[0068]

[0069]

[0070] Where τ is the update factor of the Lagrange multiplier and n is the number of iterations n = {1, ..., N}. The iteration stops when the following equation is satisfied:

[0071]

[0072] S2 includes the following sub-steps:

[0073] S21. Input the main mode into the GMM, and extract the parameters of each Gaussian distribution from the GMM. The parameters of the Gaussian distribution include mean, variance, and weight.

[0074] mean m : Indicates the center position of the mode and can be used to estimate the amplitude.

[0075] variance s 2 : Indicates the width of the mode and can be used to estimate the frequency.

[0076] Weight π : Represents the weight of each Gaussian distribution, which can be used to evaluate the importance of each mode.

[0077] S22, converting the parameters of the Gaussian distribution into amplitude, frequency and phase;

[0078] Amplitude: can be obtained from the mean of Gaussian distribution m express.

[0079] Frequency: can be calculated by the variance of the Gaussian distribution s 2 To estimate. The relationship between frequency and variance can be approximated by the following formula:

[0080]

[0081] Phase: can be expressed by phase angle. Phase angle can be estimated by the initial phase of the main mode

[0082] S23. Synthesize and reconstruct the torque data based on the extracted amplitude, frequency, and phase. The following formula can be used for reconstruction:

[0083]

[0084] Where, A i is the amplitude, f i is the frequency, is the phase, M is the number of main modes;

[0085] In S3, the components include a connecting plate, a rotating arm, and a blade. S3 includes the following sub-steps:

[0086] S31, inputting the reconstructed torque data into the LSTM network, and outputting the stress of the connecting plate, the arm, and the blade through the LSTM network;

[0087] S32. Use ROM to model the FSI problem, input the reconstructed torque data into ROM, and calculate the stress of the connecting plate, the arm, and the blade through ROM;

[0088] S33. Integrate the stresses of the connecting plate, rotating arm, and blade output by the LSTM network and the ROM to obtain the final stresses of the connecting plate, rotating arm, and blade.

[0089] In S4, the Paris formula is an important formula for describing the fatigue crack growth rate, and its expression is:

[0090]

[0091] in, is the fatigue crack growth rate, that is, the rate of change of crack length a with the number of cycles N, C and n are constants related to material properties, environment and other factors, is the stress intensity factor amplitude.

[0092] The Miner criterion formula is also called the linear cumulative damage criterion, and the formula is:

[0093]

[0094] Where, is the number of cycles of the material under the stress level i, For the material i The criterion is used to estimate the cumulative fatigue damage of a material under cyclic loading at different stress levels. When the cumulative damage reaches 1, the material is considered to have suffered fatigue failure.

[0095] In the description of the present invention, it should be understood that the terms "center", "thickness", "upper", "lower", "horizontal", "top", "bottom", "inner", "outer", "radial", etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", and "third" are used for descriptive purposes only and cannot be understood as indicating or implying the relative importance or the number of technical features implicitly specified. Therefore, the features defined by "first", "second", and "third" may explicitly or implicitly include one or more of such features.

Claims

1. A method for rapid prediction of stress and fatigue failure of an axial-flow propeller turbine, characterized in that: The following steps are involved: S1. Extract the torque data of the turbine blades through CFD, perform VMD decomposition and main mode selection on the torque data, and screen the main mode; S2. Input the main mode into GMM, extract the amplitude, frequency and phase, and reconstruct the torque data based on the amplitude, frequency and phase; S2 includes the following sub-steps: S21. Input the main mode into the GMM, and extract the parameters of each Gaussian distribution from the GMM. The parameters of the Gaussian distribution include mean, variance, and weight. S22, converting the parameters of the Gaussian distribution into amplitude, frequency and phase; S23, synthesizing and reconstructing torque data according to the extracted amplitude, frequency, and phase; S3. Input the reconstructed torque data into the LSTM network, output the stress of the blade and the arm, use the reconstructed torque data as the FSI load boundary condition, and combine it with the ROM to calculate the stress of the component; S4. Predict fatigue failure based on the calculated stress using the Paris formula and Miner criterion.

2. The method for rapid prediction of stress and fatigue failure of an axial-flow propeller turbine according to claim 1, characterized in that: S1 includes the following sub-steps: S11. Simulate the flow field around the turbine blades through computational fluid dynamics and extract the dynamic torque data of the blade surface; S12, CFD solution settings: Input dynamic torque data, set the corresponding boundary conditions and time step based on the adaptive grid and SST k-ω model, and output the torque signal. The boundary conditions include the inlet velocity, outlet pressure, and no-slip boundary conditions on the wall; S13. Use VMD to decompose torque data into modal components with clear physical meanings and screen key modes related to rotation frequency.

3. The method for rapid prediction of stress and fatigue failure of an axial-flow propeller turbine according to claim 2, characterized in that: In S11, the method for extracting the torque data of the turbine blades is specifically as follows: The dynamic pressure distribution on the turbine blade surface is calculated through transient CFD simulation based on adaptive mesh and SST k-ω model to generate torque data.

4. The method for rapid prediction of stress and fatigue failure of an axial-flow propeller turbine according to claim 2, characterized in that: S13 is specifically: Torque data is obtained through VMD solution K IMFs, calculate the energy of each IMF, normalize the energy of each IMF, sort the IMFs in descending order according to the energy size, and select the first m The IMF is selected as the main mode so that the cumulative energy of the selected IMF accounts for more than 80% of the energy of all IMFs.

5. The method for rapid prediction of stress and fatigue failure of an axial-flow propeller turbine according to claim 1, characterized in that: In S23, the torque data is reconstructed The specific expression is: Where, A i is the amplitude, f i is the frequency, is the phase, M is the number of main modes, and t is the time.

6. The method for rapid prediction of stress and fatigue failure of an axial-flow propeller turbine according to claim 1, characterized in that: In S3, the components include a connecting plate, a rotating arm, and a blade. S3 includes the following sub-steps: S31, inputting the reconstructed torque data into the LSTM network, and outputting the stress of the connecting plate, the arm, and the blade through the LSTM network; S32. Use ROM to model the FSI problem, input the reconstructed torque data into ROM, and calculate the stress of the connecting plate, the arm, and the blade through ROM; S33. Integrate the stresses of the connecting plate, rotating arm, and blade output by the LSTM network and the ROM to obtain the final stresses of the connecting plate, rotating arm, and blade.

7. The method for rapid prediction of stress and fatigue failure of an axial-flow propeller turbine according to claim 1, characterized in that: In S4, the paris formula is specifically: Where, is the fatigue crack growth rate, that is, the rate of change of crack length a with the number of cycles N, C and n are constants, is the stress intensity factor amplitude; The Miner criterion formula is as follows: Where, is the number of cycles of the material under the stress level i, For the material i The criterion is used to estimate the cumulative fatigue damage of a material under cyclic loading at different stress levels. When the cumulative damage reaches 1, the material is considered to have suffered fatigue failure.

Images