Methods for predicting the crack propagation fatigue life of key components in wind turbine drivetrain

Abstract

This invention discloses a method for predicting the crack propagation fatigue life of key components in wind turbine drivetrains, belonging to the field of fatigue analysis and remaining life calculation of generator components. The invention includes: calculating the global stress field under six sets of orthogonal unit loads based on the finite element method; establishing a nonlinear mapping relationship between real-time load and stress correction coefficients using a neural network; rapidly calculating a high-precision stress field in real time through linear stress superposition and neural network correction coefficients; extracting characteristic stress components in vulnerable areas of key components to construct a combined equivalent stress intensity factor; establishing a crack propagation rate model based on an improved Paris formula; and calculating the crack propagation fatigue life using fixed-step numerical integration. This method can be used for defect evolution analysis and crack propagation fatigue life calculation of key components such as gearbox shafts and bearing housing connectors in wind turbine drivetrains.

Description

Technical Field

[0001] This invention relates to the field of fatigue analysis and remaining life calculation of generator components, and in particular to a method for predicting the fatigue life of crack propagation in key components of wind turbine drive trains. Background Technology

[0002] Wind turbine generators are the core equipment for converting wind energy into electrical energy, and their long-term reliable operation is crucial for ensuring energy supply. The drivetrain, as a key subsystem for transmitting turbine torque and achieving speed changes, comprises complex structural components such as the main shaft, gearbox shaft system, and bearing housing connectors. It is subjected to various alternating loads, including random wind loads, gravity, and centrifugal force, over extended periods. Among these, key components such as the main shaft and gearbox shaft system have complex structures and geometric features that easily lead to stress concentration, such as shoulders, unloading grooves, and splines. These features make them highly susceptible to fatigue crack initiation and propagation under alternating loads, and are one of the main causes of drivetrain and even entire turbine failure. Therefore, establishing accurate crack propagation fatigue life prediction methods and implementing damage tolerance design and condition-based maintenance for key drivetrain components has significant engineering and economic value.

[0003] Currently, the prediction of crack propagation fatigue life of key components in the transmission chain of wind turbine generators mainly relies on methods based on damage tolerance theory. Traditional methods usually require obtaining long-term full-condition load spectra, performing load statistics and equivalence through rainflow counting, and calculating the stress intensity factor (SIF) at the crack tip using the extended finite element method (XFEM) or sub-model technology. Although this method is relatively mature, it faces significant challenges in practical applications: (1) High computational cost: Using XFEM or fine sub-models to perform high-precision analysis of the local field at the crack tip, especially when considering complex three-dimensional crack configurations and multiaxial stress states, the computational load is huge, making it difficult to meet the needs of massive design point evaluation or real-time online analysis. (2) Insufficient real-time performance and versatility: Existing prediction models are often built offline for specific load spectra and specific crack configurations, lacking the ability to quickly respond to the current operating state in real-time prediction; and when dealing with multiple operating conditions of variable speed and load, as well as dynamic changes in crack configuration, the model's generalization ability and accuracy are limited.

[0004] To address the computational efficiency bottleneck, some studies have attempted to introduce model reduction techniques to accelerate the calculation of stress intensity factors or to approximate crack propagation behavior using simplified analytical formulas. However, existing reduction strategies often lack sufficient accuracy in characterizing the nonlinear features of the crack tip field under complex load paths; while simplified models (such as crack propagation analysis based on nominal or static stress) struggle to accurately reflect the local stress concentration effects caused by structural geometry and the real-time impact of random dynamic loads, leading to significant discrepancies between predicted and actual lifespans. Therefore, a real-time prediction method for the crack propagation fatigue life of key components in the wind turbine drivetrain that can balance computational efficiency, prediction accuracy, and adaptability to operating conditions is urgently needed. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and propose a method for predicting the crack propagation fatigue life of key components in the wind turbine drive chain.

[0006] The objective of this invention is achieved through the following technical solution: a method for predicting the crack propagation fatigue life of key components in a wind turbine drivetrain, comprising the following steps:

[0007] S1: Define a standard unit load and calculate the global stress field generated by the target component under each set of unit loads using the finite element method;

[0008] S2: Define the nonlinear modified neural network structure and initialize the network parameters;

[0009] S3: Calculate the linear superposition coefficient based on the global stress field generated by the unit load, predict the nonlinear correction coefficient through a neural network, and calculate and correct the global stress field under the actual load based on the linear superposition coefficient and the nonlinear correction coefficient.

[0010] S4: Determine the convergence of the neural network based on the network loss function. If it has not converged, update the parameters and repeat steps S3-S4.

[0011] S5: Extract stress components in the vulnerable areas of the target component and solve for the equivalent stress intensity factor;

[0012] S6: Based on the Paris formula calculated by optimizing the stress intensity factor, a crack propagation rate model applicable to different working conditions is established.

[0013] S7: Based on the crack propagation rate model, calculate the crack propagation fatigue life of the target component under different working conditions.

[0014] Further, step S1 includes:

[0015] Define an orthogonal unit load, which includes concentrated forces and bending moments in the x, y, and z directions of the target component;

[0016] The global stress field for each unit load was calculated using the finite element method.

[0017] Furthermore, step S2 includes: constructing a fully connected neural network prediction architecture, with the network input being the actual load and the network output being nonlinear correction parameters.

[0018] Further, step S3 includes:

[0019] Calculate the linear superposition coefficient, and based on the ratio of the actual load to the unit load, calculate the stress effect generated by the equivalent actual load on the target component;

[0020] The nonlinear correction coefficients are predicted by a network model. The actual load corresponding to the network input is used as the network input, and a set of correction parameters is output.

[0021] Based on the linear superposition coefficient and the nonlinear correction coefficient, and combined with the global stress field of the target component under a unit load, the global stress field of the target component under the actual load is calculated by means of the superposition principle.

[0022] Further, step S4 includes:

[0023] The actual load is uniformly and randomly sampled, and the global stress response of the target component under different loads is calculated using the finite element method.

[0024] Define the mean squared error network loss function and set the convergence condition for network training.

[0025] Further, step S5 includes:

[0026] For each node of the target component, for nodes where the crack propagation effect is greater than a preset value, calculate the normal principal stress of the crack surface of the node and the Von Mises stress of the corresponding region;

[0027] By setting weight parameters for normal and tangential stresses, the control effect of the normal principal stress on the crack surface and the Von Mises stress in the corresponding region on the crack propagation path can be adjusted.

[0028] Further, step S6 includes:

[0029] Based on the Paris formula, the crack propagation rate is calculated in a preset number of cycles;

[0030] Crack propagation fatigue life under a single working condition was calculated using the explicit Euler iterative method; a crack propagation rate model was constructed based on the crack propagation fatigue life under different working conditions.

[0031] Further, step S7 includes:

[0032] For different operating conditions of wind turbines (such as normal power generation, extreme load, etc.), calculate the fatigue life of the target components under each operating condition;

[0033] Analyze the actual operating data of wind turbines to determine the cumulative operating time of each operating condition within one year and its proportion of the total operating time.

[0034] Based on the operating characteristics of wind turbines, the average time required to complete one full rotation cycle under various operating conditions is obtained;

[0035] Based on the life prediction results, time proportions, and rotation cycles for each working condition, the overall fatigue fracture life of the spindle during its service life is calculated.

[0036] The present invention also provides an electronic device, including a memory and a processor, wherein the memory is coupled to the processor; wherein the memory is used to store program data, and the processor is used to execute the program data to implement the method for predicting the crack propagation fatigue life of key components in the wind turbine drivetrain.

[0037] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for predicting the crack propagation fatigue life of key components in the wind turbine drivetrain.

[0038] The beneficial effects of this invention are as follows: Addressing the problems of incomplete multiaxial load characterization, difficulty in quantifying mixed-mode driving forces, and insufficient adaptability to dynamic operating conditions in wind turbine drivetrain crack propagation assessment, this invention innovatively integrates the critical plane method and distortion energy density theory by constructing a normal-shear dual-drive coupled model. First, it decomposes the crack driving force components based on fracture mechanics mechanisms, overcoming the limitations of traditional single-parameter stress intensity factors under complex loads. Then, it constructs a mixed-mode equivalent intensity factor using verifiable weighting coefficients α and β, establishing a bidirectional quantitative mapping relationship between stress state and crack propagation mode, effectively solving the engineering challenge of dynamically balancing modal contributions under time-varying loads. Combined with SEM fracture analysis verification mechanisms and frequency response correction algorithms, it endows the system with precise adaptability to non-proportional loads and variable-speed conditions. This hybrid assessment method, combining physical mechanisms and data-driven approaches, can simulate the entire life cycle crack evolution under different material parameters and load spectra after a single calibration. This technology provides a breakthrough solution for multiaxial fatigue design, in-service damage tolerance assessment, and condition-based maintenance decision-making for key components of wind turbine drivetrains. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 This is a flowchart of a method according to an embodiment of the present invention;

[0041] Figure 2 The following is a schematic diagram of the case setup and test results of an embodiment of the present invention, wherein (a) is a schematic diagram of the finite element mesh of the main shaft of the transmission chain, and (b) is a schematic diagram of the crack propagation fatigue life distribution of the vulnerable area of ​​the main shaft calculated using the method of the present invention.

[0042] Figure 3This is a graph showing the relationship between crack propagation length and number of cycles predicted by the method of this invention. Detailed Implementation

[0043] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Rather, they are merely examples of apparatuses and methods consistent with some aspects of the invention as detailed in the appended claims.

[0044] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The singular forms “a,” “the,” and “the” used in this invention and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0045] It should be understood that although the terms first, second, third, etc., may be used in this invention to describe various information, this information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, first information may also be referred to as second information without departing from the scope of this invention, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to a determination."

[0046] The present invention will now be described in detail with reference to the accompanying drawings. Unless otherwise specified, the features of the following embodiments and implementations can be combined with each other.

[0047] like Figure 1 As shown in the figure, this invention provides a method for predicting the crack propagation fatigue life of key components in a wind turbine drivetrain, comprising the following steps:

[0048] S1: Define a standard unit load, and calculate the global stress field generated by the target component under each set of unit loads using the finite element method. Figure 2 Figure (a) shows the mesh generation of the wind turbine drive train spindle in the finite element software.

[0049] S2: Define the nonlinear modified neural network structure and initialize the network parameters.

[0050] S3: Calculate the linear superposition coefficient based on the global stress field generated by the unit load, predict the nonlinear correction coefficient through a neural network, and calculate and correct the global stress field under the actual load based on the linear superposition coefficient and the nonlinear correction coefficient.

[0051] S4: Determine the convergence of the neural network based on the network loss function. If it does not converge, update the parameters and repeat S3 to S4.

[0052] S5: Extract the stress components of the vulnerable area of ​​the target component and solve for the equivalent stress intensity factor.

[0053] S6: Construct a crack propagation rate model applicable to different working conditions based on the improved Paris formula.

[0054] S7: Based on the crack propagation rate model, calculate the crack propagation fatigue life of the target component under complex working conditions. Figure 2 Figure (b) shows the crack propagation fatigue life distribution of the vulnerable region of the spindle calculated using the inventive method.

[0055] In one embodiment, step S1 includes the following steps:

[0056] S1.1: Define orthogonal unit load, select Concentrated forces and bending moments in three directions are used to describe the loads acting on a structure. Represented as:

[0057] .

[0058] in, They represent Bending moment and concentrated force in three directions.

[0059] The six sets of unit loads applied Defined as:

[0060] .

[0061] S1.2: The global stress field for each unit load calculated using the finite element method is expressed as follows:

[0062] .

[0063] in, Indicates the first Structural stress calculated using the finite element method under a unit load. Let be the spatial coordinates of any point on the structure. For the finite element method solver, this invention selects ABAQUS as the solver.

[0064] In one embodiment, step S2 includes the following steps:

[0065] S2.1: Construct a fully connected neural network prediction architecture. The network input is the actual load, and the network output is the nonlinear correction parameters. The network model... The structure is defined as follows:

[0066] .

[0067] in, This represents the actual load applied to the structure. For nonlinear correction parameters; Let be the trainable parameters of the encoder neural network, denoted as . ; Indicates the first The activation function of the layer. The number of layers, including network input / output layers, is set in the method proposed in this invention. Each hidden layer has 30 neurons.

[0068] In one embodiment, step S3 includes the following steps:

[0069] S3.1: Calculate the linear superposition factor. Based on the ratio of the actual load to the unit load, the equivalent actual load produces a stress effect on the structure. The superposition factor is then calculated. Represented as:

[0070] .

[0071] S3.2: Predict nonlinear correction coefficients using a network model. Input the corresponding actual load, and output a set of correction parameters, expressed as follows:

[0072] .

[0073] S3.3: The synthesized and corrected stress field, based on the calculated linear superposition coefficient and the nonlinear correction coefficient predicted by the network, combined with the global stress field of the structure under unit load obtained through finite element calculation, can be calculated using the superposition principle to calculate the stress field of the structure under actual load. , represented as:

[0074] .

[0075] in, and These represent the superposition coefficient and correction coefficient for each component of the actual load, respectively.

[0076] In one embodiment, step S4 includes the following steps:

[0077] S4.1: Generate training dataset For actual loads, uniform random sampling is performed, and the finite element method is used to calculate the global stress response of the structure under different loads. Specifically, it is expressed as:

[0078] ;

[0079] .

[0080] S4.2: Loss Function and Convergence Condition. Define the mean squared error network loss function as follows:

[0081] .

[0082] The convergence criterion for network training is set as follows: Or online training rounds .

[0083] In one embodiment, step S5 includes the following steps:

[0084] S5.1: Stress component extraction in vulnerable areas. For each node of the component, the normal principal stress of the crack surface that has a significant impact on crack propagation can be calculated. and the corresponding Von Mises stress Specifically, it is expressed as:

[0085] ;

[0086] .

[0087] in, Let be the unit vector normal to the crack propagation plane.

[0088] S5.2: Calculation of equivalent stress intensity factor, using a set of weighting parameters. , The effect of adjusting the normal principal stress on the crack surface and the Von Mises stress in the corresponding region on crack propagation is specifically expressed as follows:

[0089] .

[0090] in, , These represent the dominant factors of normal stress and tangential stress, respectively. Given the current crack length, normal stress drives crack opening (crack separation mode, Type I crack), and tangential stress drives crack slippage (in-plane / out-of-plane shear mode, Type II and III cracks). In the method proposed in this invention, the following settings are provided: , It is important to note the parameters , Numerical constraints exist: .

[0091] In one embodiment, step S6 includes the following steps:

[0092] S6.1: In optimizing the stress intensity factor Based on this, and combined with the Paris formula, a crack propagation rate equation is constructed, which is expressed as:

[0093] .

[0094] in, and These are material parameters. The length of the crack. This refers to the cycle number.

[0095] S6.2: Calculate the crack propagation fatigue life under a single working condition using the explicit Euler iteration method. The expression is written as:

[0096] .

[0097] in The initial crack length is... This is the set critical crack length.

[0098] In crack propagation analysis, it is typically assumed that the crack propagation rate remains constant over a small number of cycles. Based on this assumption, the crack propagation rate over a certain number of cycles is calculated using the Paris formula. If the crack propagation rate remains constant within a certain number of cycles, then in this... Crack propagation length per cycle Represented as:

[0099] .

[0100] Updated crack length Represented as:

[0101] .

[0102] in Let be the crack propagation length after the previous cycle ends. Assuming the crack size is not larger than expected, the appearance of the crack is not considered to affect the stress distribution in this region. Therefore, the updated stress intensity factor after crack propagation can be calculated. :

[0103] .

[0104] In the method proposed in this invention, it is assumed that crack propagation does not affect the stress distribution within the response region, and thus the stress can ultimately be increased by accumulating each... Cycle cycles within The expression for obtaining the final crack propagation fatigue life under the current working condition is:

[0105] .

[0106] Figure 3 The relationship between stress cycle number and crack propagation length under a certain working condition in the embodiment is shown.

[0107] In one embodiment, step S7 includes the following steps:

[0108] S7.1: Weighted life calculation under various operating conditions. The above method can calculate the crack propagation fatigue life under any operating condition (corresponding to a series of load time sequences). For actual wind turbines in operation, the fatigue fracture life of the main shaft under normal operating conditions can be estimated based on the statistical distribution of each operating condition. :

[0109] .

[0110] in, Indicates the total number of operating conditions. This indicates the crack propagation fatigue life of a critical component under a certain operating condition, predicted using the above process. This indicates the number of times a certain working condition occurs within a year in the statistics. This represents the sum of the time all operating conditions occur in a year. For working conditions The corresponding cycle period (the time it takes for the fan to rotate once).

[0111] In one embodiment, this method was implemented and verified on the main shaft of a 10MW wind turbine drive train. The crack propagation fatigue life was calculated. The wind turbine was simulated under DLC1.2, DLC2.4, and DLC4.1 conditions with wind speeds ranging from 3 to 25 meters per second, with a total of 260 load time sequences. The occurrence time of the corresponding conditions over one year and the corresponding wind turbine operating parameters were obtained based on the measured data. The crack propagation fatigue life of the wind turbine drive train main shaft was calculated based on these data. The calculation results are shown in Table 1 below.

[0112] Table 1

[0113] The present invention also provides an electronic device, including a memory and a processor, wherein the memory is coupled to the processor; wherein the memory is used to store program data, and the processor is used to execute the program data to implement the method for predicting the crack propagation fatigue life of key components in the wind turbine drivetrain.

[0114] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for predicting the crack propagation fatigue life of key components in the wind turbine drivetrain.

[0115] The computer-readable storage medium can be an internal storage unit of any data processing device described in any of the foregoing embodiments, such as a hard disk or memory. The computer-readable storage medium can also be any data processing device, such as a plug-in hard disk, smart media card (SMC), SD card, flash card, etc., equipped on the device. Furthermore, the computer-readable storage medium can include both internal storage units of any data processing device and external storage devices. The computer-readable storage medium is used to store the computer program and other programs and data required by the data processing device, and can also be used to temporarily store data that has been output or will be output.

[0116] It will be understood by those skilled in the art that the above descriptions are merely preferred examples of the invention and are not intended to limit the invention. Although the invention has been described in detail with reference to the foregoing examples, those skilled in the art can still modify the technical solutions described in the foregoing examples or make equivalent substitutions for some of the technical features. All modifications and equivalent substitutions made within the spirit and principles of the invention should be included within the scope of protection of the invention.

Claims

1. A method for predicting the fatigue life of crack propagation in key components of a wind turbine drivetrain, characterized in that, Includes the following steps: S1: Define a standard unit load and calculate the global stress field generated by the target component under each set of unit loads using the finite element method; S2: Define the nonlinear modified neural network structure and initialize the network parameters; S3: Calculate the linear superposition coefficient based on the global stress field generated by the unit load, predict the nonlinear correction coefficient through a neural network, and calculate and correct the global stress field under the actual load based on the linear superposition coefficient and the nonlinear correction coefficient. S4: Determine the convergence of the neural network based on the network loss function. If it has not converged, update the parameters and repeat steps S3-S4. S5: Extract stress components in the vulnerable areas of the target component and solve for the equivalent stress intensity factor; S6: Based on the Paris formula calculated by optimizing the stress intensity factor, a crack propagation rate model applicable to different working conditions is established. S7: Based on the crack propagation rate model, calculate the crack propagation fatigue life of the target component under different working conditions.

2. The method for predicting the crack propagation fatigue life of key components in a wind turbine drivetrain according to claim 1, characterized in that, Step S1 includes: Define an orthogonal unit load, which includes concentrated forces and bending moments in the x, y, and z directions of the target component; The global stress field for each unit load was calculated using the finite element method.

3. The method for predicting the crack propagation fatigue life of key components in a wind turbine drivetrain according to claim 1, characterized in that, Step S2 includes: constructing a fully connected neural network prediction architecture, with the actual load as the network input and the nonlinear correction parameters as the network output.

4. The method for predicting the crack propagation fatigue life of key components in a wind turbine drivetrain according to claim 1, characterized in that, Step S3 includes: Calculate the linear superposition coefficient, and based on the ratio of the actual load to the unit load, calculate the stress effect generated by the equivalent actual load on the target component; The nonlinear correction coefficients are predicted by a network model. The actual load corresponding to the network input is used as the network input, and a set of correction parameters is output. Based on the linear superposition coefficient and the nonlinear correction coefficient, and combined with the global stress field of the target component under a unit load, the global stress field of the target component under the actual load is calculated by means of the superposition principle.

5. The method for predicting the crack propagation fatigue life of key components in a wind turbine drivetrain according to claim 1, characterized in that, Step S4 includes: The actual load is uniformly and randomly sampled, and the global stress response of the target component under different loads is calculated using the finite element method. Define the mean squared error network loss function and set the convergence condition for network training.

6. The method for predicting the crack propagation fatigue life of key components in a wind turbine drivetrain according to claim 1, characterized in that, Step S5 includes: For each node of the target component, for nodes where the crack propagation effect is greater than a preset value, calculate the normal principal stress of the crack surface of the node and the Von Mises stress of the corresponding region; By setting weight parameters for normal and tangential stresses, the control effect of the normal principal stress on the crack surface and the Von Mises stress in the corresponding region on the crack propagation path can be adjusted.

7. The method for predicting the crack propagation fatigue life of key components in a wind turbine drivetrain according to claim 1, characterized in that, Step S6 includes: Based on the Paris formula, the crack propagation rate is calculated in a preset number of cycles; Crack propagation fatigue life under a single working condition was calculated using the explicit Euler iterative method; a crack propagation rate model was constructed based on the crack propagation fatigue life under different working conditions.

8. The method for predicting the crack propagation fatigue life of key components in a wind turbine drivetrain according to claim 1, characterized in that, Step S7 includes: For different operating conditions of wind turbines, the fatigue life of the target components under each operating condition is calculated; Analyze the actual operating data of wind turbines to determine the cumulative operating time of each operating condition within one year and its proportion of the total operating time. Based on the operating characteristics of wind turbines, the average time required to complete one full rotation cycle under various operating conditions is obtained; Based on the life prediction results, time proportions, and rotation cycles for each working condition, the overall fatigue fracture life of the spindle during its service life is calculated.

9. An electronic device comprising a memory and a processor, characterized in that, The memory is coupled to the processor; wherein the memory is used to store program data, and the processor is used to execute the program data to implement the method for predicting the crack propagation fatigue life of key components in the wind turbine drivetrain as described in any one of claims 1-8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method for predicting the crack propagation fatigue life of key components in the wind turbine drivetrain as described in any one of claims 1-8.

Images