Data-driven method for assessing the fatigue life of wind turbine blades

Abstract

This invention relates to the field of wind turbine technology, specifically to a data-driven method for assessing the fatigue life of wind turbine blades. The method includes the following steps: acquiring raw data of the wind turbine blades; extracting features from the raw data based on a blade fatigue prediction model to obtain a predicted fatigue value, and simultaneously calculating the cumulative damage value of the stress state in the raw data; combining the predicted fatigue value, the cumulative damage value, and an environmental correction factor, with the optimization objective of minimizing the error between the predicted fatigue value and the cumulative damage value, to calculate the final fatigue value of the wind turbine blades; and completing the life assessment of the wind turbine blades based on the final fatigue value. This invention enables accurate matching between fatigue prediction and measured damage values, providing a reliable decision-making basis for wind power equipment maintenance.

Description

Technical Field

[0001] This invention relates to the field of wind turbine technology, and more specifically, to a data-driven method for assessing the fatigue life of wind turbine blades. Background Technology

[0002] As a key component in wind energy conversion, wind turbine blades endure complex load environments during long-term operation, including the combined effects of various alternating stresses such as wind load, gravity load, and centrifugal force. These complex loads lead to fatigue damage in the blades, potentially causing structural failure. Accurately assessing the fatigue life of the blades is crucial for ensuring the safe operation of wind turbines and developing maintenance strategies. Currently, traditional blade fatigue life assessment methods have significant limitations in practical applications. On the one hand, traditional methods rely excessively on SN curves and Miner's linear cumulative damage theory. This simplified theoretical model cannot accurately describe the nonlinear fatigue behavior of blade materials under complex operating conditions, especially under multiaxial stress states and variable amplitude loads, where linear cumulative theory often underestimates the actual extent of fatigue damage. On the other hand, traditional methods typically employ a single stress analysis approach when processing monitoring data, neglecting the influence of environmental factors such as temperature changes and wind speed fluctuations on the fatigue damage process. Furthermore, they lack the ability to deeply mine and extract features from multi-source sensor data, resulting in the underutilization of a large amount of monitoring data containing information about the blade's health status, ultimately affecting the accuracy and reliability of fatigue life assessment. Summary of the Invention

[0003] The purpose of this invention is to provide a data-driven method for assessing the fatigue life of wind turbine blades. It uses a deep learning model to extract fatigue characteristics from multi-source monitoring data of the blades, combines it with a stress state cumulative damage calculation method to accurately assess the fatigue state of the blades, and introduces environmental correction factors for temperature and wind speed to improve the adaptability of the assessment results in complex environments. It can achieve accurate matching between fatigue prediction and measured damage values, making the assessment results more in line with actual operating conditions and providing a reliable decision-making basis for wind power equipment maintenance.

[0004] The present invention is achieved through the following technical solution:

[0005] A data-driven method for assessing the fatigue life of wind turbine blades, comprising the following steps:

[0006] Obtain raw data of wind turbine blades;

[0007] Based on the blade fatigue prediction model, feature extraction is performed on the original data of wind turbine blades to obtain the blade fatigue prediction results, and at the same time, the cumulative damage value of stress state in the original data of wind turbine blades is solved.

[0008] By combining the blade fatigue prediction results, cumulative damage value under stress, and environmental correction factors, and with the optimization objective of minimizing the error of the blade fatigue prediction results and cumulative damage value under stress, the final fatigue of the wind turbine blade is calculated, and the life assessment of the wind turbine blade is completed based on the final fatigue of the wind turbine blade.

[0009] Optionally, acquiring the raw data of the wind turbine blades specifically includes:

[0010] Collect blade strain data for wind turbines;

[0011] Collect blade load data for wind turbines;

[0012] Collect environmental data on wind turbine blades;

[0013] The blade input vector of the wind turbine is constructed based on the blade strain data, blade load data and blade environmental data of the wind turbine, and the blade input vector of the wind turbine is standardized to serve as the input of the blade fatigue prediction model.

[0014] Optionally, the blade fatigue prediction model is built using a convolutional neural network model, which specifically consists of an input layer, a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a flattening layer, a fully connected layer, and an output layer.

[0015] Optionally, the input layer is calculated using the following formula:

[0016]

[0017] Where X(t) is the blade input vector of the wind turbine, and L is the sequence length;

[0018] The calculation formula for the first convolutional layer is as follows:

[0019]

[0020] Where C1(t) is the feature map output of the first convolutional layer, k1 is the time dimension kernel of the first convolutional kernel, b1 is the bias term of the first convolutional layer, K1(i,j) is the convolutional kernel weight of the j-th channel at time i of the first convolutional layer, i is the sliding index of the time dimension, j is the index of the feature dimension, and X(ti) is the wind turbine blade input vector at time ti.

[0021] The calculation formula for the first pooling layer is as follows:

[0022] P1(t) = max{C1(t:t+p1)}

[0023] Where P1(t) is the feature map output of the first pooling layer, t is the current time point, p1 is the window size of the first pooling layer, and C1(t:t+p1) is the value of the output of the first convolutional layer from time t to t+p1.

[0024] The second convolutional layer is calculated using the following formula:

[0025]

[0026] Where C2(t) is the feature map output of the second convolutional layer, k2 is the time dimension kernel of the second convolutional layer, d1 is the number of output channels of the first convolutional layer, K2(i,j) is the kernel weight of the j-th channel at time i of the second convolutional layer, and b2 is the bias term of the second convolutional layer.

[0027] The second pooling layer is calculated using the following formula:

[0028] P2(t) = max{C2(t:t+p2)}

[0029] Where P2(t) is the feature map output of the second pooling layer, and C2(t:t+p2) is the value of the output of the second convolutional layer from time t to t+p2;

[0030] The calculation formula for the flattened layer is as follows:

[0031] F(t) = flatten(P2(t))

[0032] Where F(t) is the output of the flattened layer;

[0033] The fully connected layer is calculated using the following formula:

[0034] FC(t)=σ(F(t)W+b)

[0035] Where FC(t) represents the predicted fatigue level of the wind turbine blades, σ is the ReLU activation function, and W is the weight matrix. b is the bias vector. d in The input dimension after flattening.

[0036] Optionally, the specific process for solving the cumulative damage value of the stress state in the original data of the wind turbine blade is as follows:

[0037] Collect blade strain data and blade load data of wind turbine generators;

[0038] Calculate the principal stress, shear stress, and equivalent stress of the wind turbine blades based on the blade strain data and blade load data.

[0039] Based on the equivalent stress of wind turbine blades, the cyclic range of equivalent stress is determined, and effective cyclic statistics are performed on the stress cyclic range of wind turbine blades to solve for the cumulative damage value of stress state.

[0040] Optionally, the specific calculation formula for solving the cumulative damage value of the stress state in the original data of the wind turbine blade is as follows:

[0041]

[0042]

[0043] Δσ(t)=max(σ eq (t))-min(σ eq (t))

[0044]

[0045]

[0046] Where σ1(t) and σ2(t) are the first and second principal stresses of the blade, respectively, and τ xy (t) represents the blade shear stress, E represents the elastic modulus of the blade material, ν represents the Poisson's ratio of the blade material, and ε represents the shear stress of the blade. x (t), ε y (t) represents the spanwise and tangential strains of the blade, respectively, F x (t), F y (t) represents the spanwise and chordal forces of the blade, respectively; h is the blade section height; I is the moment of inertia of the blade section; G is the shear modulus of the blade material; γ xy (t) represents the blade shear strain, M z (t) represents the torsional moment of the blade, r represents the radius of the blade section, J represents the polar moment of inertia of the blade section, and σ eq (t) represents the equivalent stress of the blade, Δσ(t) represents the cyclic range of the equivalent stress of the blade, N(t) represents the effective number of stress cycles of the blade, and σ th denoted as the blade fatigue limit threshold, I() as the blade stress effective cycle judgment function, k as the blade stress level division number, D(t) as the cumulative damage value of the blade under the current stress level, which is characterized as the cumulative damage value under stress state, C as the fatigue strength coefficient of the blade material, and m as the negative slope of the SN curve of the blade material.

[0047] Optionally, the final fatigue degree of the wind turbine blade is calculated with the optimization objective of minimizing the error between the predicted fatigue degree of the blade and the cumulative damage value under stress state. The calculation formula is as follows:

[0048]

[0049]

[0050] Where Fatigue(t) is the final fatigue degree of the blade at time t, v(t) is the real-time wind speed at time t, v0 is the reference wind speed, T0 is the reference temperature, T(t) is the blade surface temperature at time t, T is the total length of the time series, and min(L) is the fatigue value of the blade. actual (t) represents the measured fatigue degree of the blade at time t, D actual (t) represents the measured cumulative damage value.

[0051] Optionally, the constraints on the optimization objective are:

[0052]

[0053] Where, σ y σ is the yield strength of the blade material. u s is the ultimate strength of the blade material. f D is the safety factor for the blades. cr v is the critical fatigue damage value. cut-in For the cut-in wind speed of the fan, v cut-out To cut off the wind speed of the fan, T min For the minimum permissible operating temperature, T max The maximum permissible operating temperature.

[0054] The technical solution of the present invention has at least the following advantages and beneficial effects:

[0055] This invention uses a deep learning model to extract fatigue characteristics from multi-source monitoring data of blades, and combines it with a stress state cumulative damage calculation method to accurately assess the fatigue state of the blades. Furthermore, it introduces environmental correction factors for temperature and wind speed, which improves the adaptability of the assessment results in complex environments. This enables precise matching between fatigue prediction and measured damage values, making the assessment results more consistent with actual operating conditions and providing a reliable decision-making basis for wind power equipment maintenance. Attached Figure Description

[0056] Figure 1 A schematic flowchart of the data-driven wind turbine blade fatigue life assessment method provided by the present invention. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, what is described is only a part of this invention, not all of it. The components of this invention, typically described and shown in the accompanying drawings, can be arranged and designed in various different configurations.

[0058] like Figure 1As shown, this invention provides one embodiment: a data-driven method for assessing the fatigue life of wind turbine blades, the method comprising the following steps:

[0059] Obtain raw data of wind turbine blades;

[0060] Based on the blade fatigue prediction model, feature extraction is performed on the original data of wind turbine blades to obtain the blade fatigue prediction results, and at the same time, the cumulative damage value of stress state in the original data of wind turbine blades is solved.

[0061] By combining the blade fatigue prediction results, cumulative damage value under stress, and environmental correction factors, and with the optimization objective of minimizing the error of the blade fatigue prediction results and cumulative damage value under stress, the final fatigue of the wind turbine blade is calculated, and the life assessment of the wind turbine blade is completed based on the final fatigue of the wind turbine blade.

[0062] In the specific implementation of this embodiment, obtaining the raw data of the wind turbine blades specifically includes:

[0063] Collect blade strain data for wind turbines;

[0064] Collect blade load data for wind turbines;

[0065] Collect environmental data on wind turbine blades;

[0066] The blade input vector of the wind turbine is constructed based on the blade strain data, blade load data and blade environmental data of the wind turbine, and the blade input vector of the wind turbine is standardized to serve as the input of the blade fatigue prediction model.

[0067] Specifically, the blade strain data for wind turbines includes: ε x (t), ε y (t) and γ xy (t), the blade load data of the wind turbine includes: F x (t), F y (t) and M z (t); The environmental data for wind turbine blades include: v(t), T(t) and ρ(t).

[0068] In this embodiment, the blade fatigue prediction model is built using a convolutional neural network model, which specifically consists of an input layer, a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a flattening layer, a fully connected layer, and an output layer.

[0069] Optionally, the input layer is calculated using the following formula:

[0070]

[0071] Where X(t) is the blade input vector of the wind turbine, and L is the sequence length;

[0072] The calculation formula for the first convolutional layer is as follows:

[0073]

[0074] Where C1(t) is the feature map output of the first convolutional layer, k1 is the time dimension kernel of the first convolutional kernel, b1 is the bias term of the first convolutional layer, K1(i,j) is the convolutional kernel weight of the j-th channel at time i of the first convolutional layer, i is the sliding index of the time dimension, j is the index of the feature dimension, and X(ti) is the wind turbine blade input vector at time ti.

[0075] The calculation formula for the first pooling layer is as follows:

[0076] P1(t) = max{C1(t:t+p1)}

[0077] Where P1(t) is the feature map output of the first pooling layer, t is the current time point, p1 is the window size of the first pooling layer, and C1(t:t+p1) is the value of the output of the first convolutional layer from time t to t+p1.

[0078] The second convolutional layer is calculated using the following formula:

[0079]

[0080] Where C2(t) is the feature map output of the second convolutional layer, k2 is the time dimension kernel of the second convolutional layer, d1 is the number of output channels of the first convolutional layer, K2(i,j) is the kernel weight of the j-th channel at time i of the second convolutional layer, and b2 is the bias term of the second convolutional layer.

[0081] The second pooling layer is calculated using the following formula:

[0082] P2(t) = max{C2(t:t+p2)}

[0083] Where P2(t) is the feature map output of the second pooling layer, and C2(t:t+p2) is the value of the output of the second convolutional layer from time t to t+p2;

[0084] The calculation formula for the flattened layer is as follows:

[0085] F(t) = flatten(P2(t))

[0086] Where F(t) is the output of the flattened layer;

[0087] The fully connected layer is calculated using the following formula:

[0088] FC(t)=σ(F(t)W+b)

[0089] Where FC(t) represents the predicted fatigue level of the wind turbine blades, σ is the ReLU activation function, and W is the weight matrix. b is the bias vector. d in The input dimension after flattening.

[0090] In implementation, based on the raw monitoring data of wind turbine blades, including strain data, load data, and environmental data, a normalization preprocessing was first performed to standardize the data to a uniform scale. Then, a sliding time window method was used to segment the data with a window length L and a step size s, forming a time-series sample set. Feature extraction was performed through a two-stage convolutional-pooling structure: the first stage used 32 3×9 convolutional kernels for feature extraction, followed by tanh activation and dimensionality reduction through a 2×1 max-pooling layer; the second stage used 64 3×32 convolutional kernels to further extract high-level features, also undergoing activation and pooling operations. Finally, the feature map was flattened and mapped to a 128-dimensional feature space through a fully connected layer to obtain the blade fatigue feature representation FC(t). The entire process achieved automatic extraction of fatigue features from raw multi-source data, providing a reliable feature foundation for subsequent fatigue calculation.

[0091] In this embodiment, the specific process of solving the cumulative damage value of the stress state in the original data of the wind turbine blade is as follows:

[0092] Collect blade strain data and blade load data of wind turbine generators;

[0093] Calculate the principal stress, shear stress, and equivalent stress of the wind turbine blades based on the blade strain data and blade load data.

[0094] Based on the equivalent stress of wind turbine blades, the cyclic range of equivalent stress is determined, and effective cyclic statistics are performed on the stress cyclic range of wind turbine blades to solve for the cumulative damage value of stress state.

[0095] The specific formula for calculating the cumulative damage value of the stress state in the original data of the wind turbine blade is as follows:

[0096]

[0097]

[0098] Δσ(t)=max(σ eq (t))-min(σ eq (t))

[0099]

[0100] Where σ1(t) and σ2(t) are the first and second principal stresses of the blade, respectively, and τ xy (t) represents the blade shear stress, E represents the elastic modulus of the blade material, ν represents the Poisson's ratio of the blade material, and ε represents the shear stress of the blade. x (t), ε y (t) represents the spanwise and tangential strains of the blade, respectively, F x (t), F y (t) represents the spanwise and chordal forces of the blade, respectively; h is the blade section height; I is the moment of inertia of the blade section; G is the shear modulus of the blade material; γ xy (t) represents the blade shear strain, M z (t) represents the torsional moment of the blade, r represents the radius of the blade section, J represents the polar moment of inertia of the blade section, and σ eq (t) represents the equivalent stress of the blade, Δσ(t) represents the cyclic range of the equivalent stress of the blade, N(t) represents the effective number of stress cycles of the blade, and σ th denoted as the blade fatigue limit threshold, I() as the blade stress effective cycle judgment function, k as the blade stress level division number, D(t) as the cumulative damage value of the blade under the current stress level, which is characterized as the cumulative damage value under stress state, C as the fatigue strength coefficient of the blade material, and m as the negative slope of the SN curve of the blade material.

[0101] In practice, based on the blade strain and load data of the wind turbine, Hooke's law is used to establish the strain-stress relationship. Combined with the geometric characteristics of the blade section, the principal stresses σ1(t), σ2(t) and shear stress τ are calculated. xy (t). Then, the equivalent stress σ is calculated based on the von Mises criterion. eq The stress cycles were statistically analyzed using the rainflow counting method to obtain the stress amplitude Δσ(t) and the effective number of cycles N(t). Based on this, Miner's linear cumulative damage theory was applied, combined with the material's SN curve characteristics, namely the slope m and the material constant C, to calculate the cumulative damage value D(t) at each time step. This cumulative damage calculation method based on continuous damage mechanics effectively combines strain-stress analysis with fatigue theory, enabling a quantitative assessment of the blade's fatigue state.

[0102] Optionally, the final fatigue degree of the wind turbine blade is calculated with the optimization objective of minimizing the error between the predicted fatigue degree of the blade and the cumulative damage value under stress state. The calculation formula is as follows:

[0103]

[0104]

[0105] Where Fatigue(t) is the final fatigue degree of the blade at time t, v(t) is the real-time wind speed at time t, v0 is the reference wind speed, T0 is the reference temperature, T(t) is the blade surface temperature at time t, T is the total length of the time series, and min(L) is the fatigue value of the blade. actual (t) represents the measured fatigue degree of the blade at time t, D actual (t) represents the measured cumulative damage value.

[0106] In this specific implementation, the objective function for minimizing the error is constructed, which includes an absolute error term between the predicted fatigue degree and the actual fatigue degree, and a logarithmic error term between the cumulative damage value and the measured damage value. During the optimization process, constraints such as the blade stress safety factor, the critical fatigue damage value, the wind speed operating range, and the temperature operating range are introduced to ensure that the optimization results meet the actual engineering requirements. The final fatigue degree calculation formula comprehensively considers the influence of model output, cumulative damage value, wind speed correction factor, and temperature correction factor. Based on the calculated final fatigue degree of the blade, the damage increment ΔD = k'·Fatigue(t) per unit time is calculated, where k' is the proportionality coefficient. Based on the known critical damage value D of the blade material... cr The remaining blade life is obtained by combining the current cumulative damage value.

[0107]

[0108] More specifically, the constraints on the optimization objective are:

[0109]

[0110] Where, σ y σ is the yield strength of the blade material. u s is the ultimate strength of the blade material. f D is the safety factor for the blades. cr v is the critical fatigue damage value. cut-in For the cut-in wind speed of the fan, v cut-out To cut off the wind speed of the fan, T min For the minimum permissible operating temperature, T max The maximum permissible operating temperature.

[0111] The above are merely preferred embodiments of the present invention and are not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A data-driven method for assessing the fatigue life of wind turbine blades, characterized in that, The steps of this method include: Obtain raw data of wind turbine blades; Based on the blade fatigue prediction model, feature extraction is performed on the original data of wind turbine blades to obtain the blade fatigue prediction results, and at the same time, the cumulative damage value of stress state in the original data of wind turbine blades is solved. Combining the blade fatigue prediction results, stress state cumulative damage value and environmental correction factor of wind turbine, and taking the minimization of the error of blade fatigue prediction results and stress state cumulative damage value as the optimization objective, the final fatigue of wind turbine blade is calculated, and the life assessment of wind turbine blade is completed based on the final fatigue of wind turbine blade. The optimization objective is to minimize the error between the predicted fatigue degree of the blade and the cumulative damage value under stress state. The final fatigue degree of the wind turbine blade is calculated using the following formula: in, for The final fatigue level of the blade at any given moment. The results show the predicted fatigue level of wind turbine blades. This represents the cumulative damage value of the blade under the current stress level, which is characterized by the cumulative damage value under stress state. for Real-time wind speed at any given moment For reference wind speed, For reference temperature, for The blade surface temperature at any given time. The total length of the time series. To minimize the error objective function, for The measured fatigue level of the blade at any given time. This represents the measured cumulative damage value. The constraints of the optimization objective are: in, The yield strength of the blade material. The ultimate strength of the blade material. For the safety factor of the blade, This is the critical fatigue damage value. The cut-in wind speed for the fan. To cut off the wind speed for the fan. The minimum permissible operating temperature, The maximum permissible operating temperature.

2. The data-driven method for assessing the fatigue life of wind turbine blades according to claim 1, characterized in that, The acquisition of raw data of wind turbine blades specifically includes: Collect blade strain data for wind turbine generators; Collect blade load data for wind turbines; Collect environmental data on wind turbine blades; The blade input vector of the wind turbine is constructed based on the blade strain data, blade load data and blade environmental data of the wind turbine, and the blade input vector of the wind turbine is standardized to serve as the input of the blade fatigue prediction model.

3. The data-driven method for assessing the fatigue life of wind turbine blades according to claim 2, characterized in that, The blade fatigue prediction model is built using a convolutional neural network model, which consists of an input layer, a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a flattening layer, a fully connected layer, and an output layer.

4. The data-driven method for assessing the fatigue life of wind turbine blades according to claim 3, characterized in that, The input layer is calculated using the following formula: in, This is the input vector for the wind turbine blades. The sequence length; The calculation formula for the first convolutional layer is as follows: in, The output is the feature map of the first convolutional layer. The time dimension kernel of the first convolution kernel. This refers to the bias term of the first convolutional layer. For the first convolutional layer Time of the first Channel convolution kernel weights, A sliding index along the time dimension. For the index of the feature dimension, for The wind turbine blade input vector at time 1; The calculation formula for the first pooling layer is as follows: in, The output is the feature map of the first pooling layer. At the current time point, The window size of the first pooling layer. The output of the first convolutional layer arrive The value at time; The second convolutional layer is calculated using the following formula: in, This is the feature map output for the second convolutional layer. The time dimension kernel of the second convolutional layer, This represents the number of output channels for the first convolutional layer. For the second convolutional layer Time of the first Channel convolution kernel weights, This refers to the bias term of the second convolutional layer; The second pooling layer is calculated using the following formula: in, The output is the feature map of the second pooling layer. For the output of the second convolutional layer arrive The value at time; The calculation formula for the flattened layer is as follows: in, For output of the flattened layer; The fully connected layer is calculated using the following formula: in, The results show the predicted fatigue level of wind turbine blades. For ReLU activation functions, This is the weight matrix. , For bias vectors, , The input dimension after flattening.

5. The data-driven method for assessing the fatigue life of wind turbine blades according to claim 4, characterized in that, The specific process for solving the cumulative damage value of the stress state in the original data of the wind turbine blade is as follows: Collect blade strain data and blade load data of wind turbine generators; Calculate the principal stress, shear stress, and equivalent stress of the wind turbine blades based on the blade strain data and blade load data. Based on the equivalent stress of wind turbine blades, the cyclic range of equivalent stress is determined, and effective cyclic statistics are performed on the stress cyclic range of wind turbine blades to solve for the cumulative damage value of stress state.

6. The data-driven method for assessing the fatigue life of wind turbine blades according to claim 5, characterized in that, The specific formula for calculating the cumulative damage value of the stress state in the original data of the wind turbine blade is as follows: in, , These are the first and second principal stresses of the blade, respectively. For blade shear stress, The elastic modulus of the blade material. Poisson's ratio for the blade material , These are the spanwise and tangential strains of the blade, respectively. , These are the spanwise and tangential forces of the blade, respectively. The blade cross-section height, Let the moment of inertia of the blade section be... Shear modulus of blade material For the shear strain of the blade. For the torsional bending moment of the blade, Where is the radius of the blade cross section. The polar moment of inertia of the blade cross section, For the equivalent stress of the blade, The cyclic range of equivalent stress in the blade. The effective number of stress cycles for the blade. This represents the fatigue limit threshold of the blade. This is the function for determining the effective cycle of blade stress. The number of stress levels is determined by the number of blade segments. This represents the cumulative damage value of the blade under the current stress level, which is characterized by the cumulative damage value under stress state. The fatigue strength coefficient of the blade material. The negative slope of the SN curve for the blade material.

Images