Fan blade fatigue damage prediction method and system

Abstract

The invention relates to the technical field of fan blade fatigue damage prediction. The invention provides a fan blade fatigue damage prediction method and system. The method comprises the following steps: constructing a coupling finite element model based on blade anisotropy parameters; blade surface three-dimensional strain field data, blade vibration acceleration signals, environment temperature and humidity and wind speed and direction data are obtained in real time, and a multi-dimensional monitoring data set is constructed; based on the multi-dimensional monitoring data set, nonlinear coupling features of all the load components are extracted, a multi-dimensional feature tensor is obtained, and a reference stress field matched with the current working condition is generated; inputting the multi-dimensional feature tensor and the reference stress field into a bidirectional long-short-term memory network, and establishing a data-physics combined driven damage evolution model; and positioning a damage area based on a damage probability distribution diagram output by the damage evolution model. The problems that in the prior art, prediction errors are obvious, sensitivity to early damage is insufficient, the false alarm rate is high, and accurate positioning of the damage position and quantitative prediction of the residual life are difficult to achieve are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of fan blade fatigue damage prediction, and in particular to a fan blade fatigue damage prediction method and system. Background Art

[0002] As the core component of a wind power system, the reliability of wind turbine blades directly affects the safe operation and power generation efficiency of the unit. As wind turbines develop towards larger and lighter units, fatigue damage to blades under complex loads (such as aerodynamic loads, centrifugal force, gravity) and harsh environments (such as salt spray, lightning strikes, and extreme temperatures) is becoming increasingly prominent. According to statistics, downtime losses caused by blade failure account for more than 30% of the total operation and maintenance costs of a wind farm, while the timely detection of early fatigue damage can reduce maintenance costs by 60%. Therefore, the development of efficient and accurate fatigue damage prediction methods is of great significance to improving the economy and safety of wind power systems.

[0003] At present, the monitoring and prediction technologies for blade fatigue damage mainly include two categories:

[0004] One approach involves mechanical simulation based on physical models, which simulate blade stress distribution and damage evolution through finite element analysis. However, these methods rely on simplified boundary condition assumptions (such as ignoring the effects of ambient temperature and humidity on material properties) and struggle to reflect the dynamic coupling effects of multiple load sources in real-time during actual operation, leading to significant prediction errors, especially under extreme operating conditions.

[0005] The second approach is online monitoring technology based on vibration signals or strain sensors. This involves collecting vibration spectra or local strain data from the blade surface and combining it with a threshold alarm mechanism to determine the damage status. However, existing methods often rely on a single physical quantity for monitoring, which is insensitive to early-stage damage and susceptible to environmental noise (such as signal distortion caused by lightning electromagnetic pulses), resulting in a high false alarm rate. Furthermore, traditional data-driven models (such as BP neural networks) lack integrated modeling of the anisotropic degradation mechanism of blade composite materials, making it difficult to accurately locate damage and quantitatively predict the remaining life. Summary of the Invention

[0006] The purpose of the present invention is to provide a method and system for predicting fatigue damage of wind blades, aiming to solve the problems of existing wind blade fatigue damage monitoring and prediction technologies, such as significant prediction errors, insufficient sensitivity to early damage, high false alarm rate, and difficulty in accurately locating the damage position and quantitatively predicting the remaining life.

[0007] The present invention is achieved through the following technical solutions:

[0008] A method for predicting fatigue damage of fan blades comprises the following steps:

[0009] A coupled finite element model including aerodynamic load field, structural stress field and thermal field is constructed based on the anisotropic parameters of the blade composite material, wherein the coupled finite element model is provided with a constitutive equation of the composite laminate;

[0010] Acquire blade surface three-dimensional strain field data, blade vibration acceleration signals, ambient temperature and humidity, and wind speed and direction data in real time to construct a multi-dimensional monitoring data set aligned in time and space;

[0011] Based on a multidimensional monitoring data set, the nonlinear coupling characteristics of each load component under the joint frequency-time domain distribution are extracted to obtain a multidimensional characteristic tensor. The constitutive equation parameters of the composite laminate are dynamically modified. The boundary conditions of the coupled finite element model are optimized using an adaptive particle swarm algorithm to generate a benchmark stress field that matches the current working conditions.

[0012] The multidimensional characteristic tensor and the benchmark stress field are input into a bidirectional long short-term memory network, and the anisotropic degradation factor of the composite material is embedded to establish a data-physics-driven damage evolution model.

[0013] Based on the damage probability distribution map output by the damage evolution model, maximum likelihood estimation is used to locate the damaged area. Combined with the material residual stiffness attenuation curve and the load spectrum rain flow counting results, the remaining service life of the blade is iteratively calculated.

[0014] Optionally, the specific process of constructing a coupled finite element model including aerodynamic load field, structural stress field and thermal field based on the anisotropic parameters of the blade composite material, wherein the constitutive equation of the composite laminate is provided in the coupled finite element model, is as follows:

[0015] Through laminate tensile testing and digital image correlation technology, the anisotropic elastic modulus, Poisson's ratio and thermal expansion coefficient of the blade composite material were measured, and a gradient parameter database including fiber orientation and ply sequence was established;

[0016] Based on computational fluid dynamics, the unsteady aerodynamic pressure distribution on the blade surface is simulated, the aerodynamic load field is decomposed into normal pressure component and tangential friction component, and then mapped to the structural grid through the fluid-structure coupling interface;

[0017] The thermal field control equation is introduced into the structural stress field. The equivalent thermal stress caused by the temperature gradient is calculated based on the ambient temperature and humidity data, and vector superposition is performed with the aerodynamic load component.

[0018] The constitutive equation of composite laminates is defined using the variable stiffness method. The material stiffness matrix is ​​dynamically associated with the anisotropic parameters. The failure mode of the single-layer plate under different stress states is determined using the Hashin criterion.

[0019] The first three natural frequencies of the blade are measured through modal tests, and the boundary constraints of the coupled finite element model are optimized with the goal of minimizing the frequency error, completing the verification of the coupled finite element model of multi-physical fields.

[0020] Optionally, the specific process of acquiring the three-dimensional strain field data of the blade surface, as well as the blade vibration acceleration signal, the ambient temperature and humidity, and the wind speed and direction data in real time, and constructing the spatiotemporally aligned multidimensional monitoring data set is as follows:

[0021] A fiber Bragg grating sensor array is arranged in a grid of equal curvature in the key stress areas of the blade surface. Wavelength demodulation technology is used to synchronously collect three-dimensional strain components to generate spatially continuous three-dimensional strain field distribution data.

[0022] Three-axis accelerometers are installed on the blade root flange and the leading edge of the airfoil to capture vibration acceleration signals and eliminate lightning electromagnetic pulse interference through wavelet noise reduction;

[0023] Temperature and humidity sensors and ultrasonic anemometers are deployed on the top of the cabin to record ambient temperature, humidity, wind speed and direction parameters in real time. A timestamp synchronization protocol is used to align multi-source data streams to a unified time sequence.

[0024] The sensor space mapping relationship is established based on the grid node coordinates of the coupled finite element model, and the missing data points are compensated by Kriging interpolation to form a monitoring data matrix that is spatially aligned with the coupled finite element model.

[0025] The monitoring data are sliced ​​at intervals through a sliding time window, and the blade azimuth and speed information at the corresponding moments are superimposed to construct a multidimensional monitoring dataset with synchronous correlation in the time and space dimensions.

[0026] Optionally, the specific process of extracting the nonlinear coupling characteristics of each load component under the frequency domain-time domain joint distribution based on the multidimensional monitoring data set to obtain the multidimensional feature tensor is:

[0027] Perform short-time Fourier transform on the three-dimensional strain field data, extract the strain energy density spectrum of each spatial node in the frequency range of 0.5Hz to 50Hz, and generate the time-frequency energy distribution matrix;

[0028] Perform empirical mode decomposition on the vibration acceleration signal, extract the cross-correlation function amplitudes of the first three intrinsic mode functions and the wind speed time history signal, and construct the load interaction eigenvector;

[0029] The thermal stress transfer function is calculated based on the ambient temperature and humidity data. The temperature gradient and the main frequency component of the strain field are phase-matched through convolution operation to obtain the thermal-mechanical coupling coefficient sequence.

[0030] The time-frequency energy distribution matrix, load interaction eigenvectors, and thermal-mechanical coupling coefficient sequences are tensor-spliced ​​according to the space-time grid, and the kurtosis-skewness joint statistics of each component are calculated by sliding along the time axis to form a multidimensional feature tensor containing frequency domain correlation, time domain mutation, and spatial correlation.

[0031] Optionally, it is characterized in that, optionally, the specific process of dynamically correcting the constitutive equation parameters of the composite laminate and optimizing the boundary conditions of the coupled finite element model by an adaptive particle swarm algorithm to generate a reference stress field matching the current working condition is:

[0032] The anisotropic elastic modulus change rate and Poisson's ratio offset are separated from the multidimensional characteristic tensor, and the stiffness matrix components in the constitutive equation are updated using an online recursive least squares method.

[0033] The fitness function is constructed by using the residual sum of squares between the theoretical strain field and the monitored strain field output by the coupled finite element model, and the ambient temperature gradient and wind speed fluctuation coefficient are encoded as particle swarm dimension variables.

[0034] The inertia weight is dynamically adjusted according to the convergence speed of the particle swarm during the iteration process. When the standard deviation of the particle spacing is less than the threshold, the learning factor reorganization mechanism is triggered to accelerate the convergence of the boundary constraints.

[0035] The optimized aerodynamic load distribution parameters and thermal expansion coefficient are substituted into the coupled finite element model to solve the dynamic stress field under the current rotation speed and azimuth angle. The dynamic stress field is then spatially convolved with the kurtosis-skewness statistics in the multidimensional feature tensor to generate a reference stress field that eliminates noise interference.

[0036] The matching degree of the model is verified based on the mutual information entropy of the benchmark stress field and the monitored strain field. If the mutual information entropy is lower than the set threshold, a secondary parameter correction cycle is triggered until the working condition consistency criterion is met.

[0037] Optionally, the specific process of inputting the multidimensional characteristic tensor and the reference stress field into the bidirectional long short-term memory network, embedding the anisotropic degradation factor of the composite material, and establishing the data-physics jointly driven damage evolution model is as follows:

[0038] The multi-dimensional feature tensor is sliced ​​by time step and expanded into a sequence data stream along the time axis, while the reference stress field is encoded into a static condition vector according to the spatial grid node;

[0039] A parallel channel is set up at the input layer of the bidirectional long short-term memory network to receive the time-frequency feature sequence of the multidimensional feature tensor and the spatial distribution parameters of the reference stress field respectively;

[0040] Dynamically calculate anisotropic degradation factors based on the constitutive equation parameters of composite laminates, including the fiber direction elastic modulus decay rate and the matrix interface shear strength degradation coefficient, and embed them into the gated recurrent weight matrix of the long short-term memory unit;

[0041] A physical constraint module is introduced into the hidden layer to perform tensor dot multiplication between the stress gradient distribution output by the coupled finite element model and the hidden state of the long short-term memory network to generate damage-sensitive features with spatial attention weights.

[0042] Through bidirectional time propagation, the forward causality and backward correlation of damage accumulation are synchronously captured, the damage probability value of each grid node at different time steps is output, and a spatiotemporal continuous damage evolution model is constructed in combination with the material failure threshold.

[0043] Optionally, the specific process of iteratively calculating the remaining service life of the blade based on the damage probability distribution map output by the damage evolution model, locating the damage area using maximum likelihood estimation, and combining the material residual stiffness attenuation curve and the load spectrum rain flow counting result is as follows:

[0044] The damage probability distribution map is gridded, and the blade surface is discretized into several sub-regions. Based on the principle of maximum likelihood estimation, the joint probability density function of the damage probability in each sub-region is calculated. The sub-regions with damage probability values ​​exceeding the preset threshold are screened out and identified as potential damage areas.

[0045] For potential damage areas, a quantitative relationship between residual stiffness and damage degree is established based on the composite material residual stiffness attenuation curve. The stress level of the potential damage area under the current load condition is calculated in combination with the reference stress field to assess the impact of the damage on the blade structural stiffness.

[0046] Perform rain flow counting on the load spectrum to count the number of load cycles and load amplitude distribution in the potential damage area during historical operation. Combined with the material's SN curve, calculate the fatigue life consumed in the potential damage area.

[0047] With the residual stiffness as the constraint condition, based on Miner's linear fatigue cumulative damage theory, the consumed fatigue life and the remaining fatigue life are iteratively calculated until the remaining fatigue life converges. The remaining service life of the potential damage area of ​​the blade is obtained. The remaining service life of the entire blade is determined by combining the remaining service life of each potential damage area.

[0048] Based on the same inventive concept, the present invention further provides a fan blade fatigue damage prediction system for implementing the fan blade fatigue damage prediction method, comprising:

[0049] Data acquisition module, used to collect three-dimensional strain field data, capture vibration acceleration signals, record environmental data, and align multi-source data streams in time and space;

[0050] A model building module is used to construct a coupled finite element model including aerodynamic load field, structural stress field and thermal field based on the anisotropic parameters of the blade composite material, and set the constitutive equation of the composite laminate;

[0051] The feature extraction module is used to process the collected multi-dimensional monitoring data, extract the nonlinear coupling characteristics of each load component under the frequency domain-time domain joint distribution, and obtain a multi-dimensional feature tensor;

[0052] The model optimization module is used to dynamically modify the constitutive equation parameters of the composite laminate based on the multidimensional characteristic tensor, optimize the boundary conditions of the coupled finite element model through the adaptive particle swarm algorithm, and generate a benchmark stress field that matches the current working conditions;

[0053] The prediction and analysis module is used to input the multidimensional feature tensor and the benchmark stress field into a bidirectional long short-term memory network, embed the anisotropic degradation factor of the composite material, establish a damage evolution model, and locate the damaged area based on the damage probability distribution map output by the model, and calculate the remaining service life of the blade.

[0054] Based on the same inventive concept, the present invention also provides an electronic device, including a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the above-mentioned wind turbine blade fatigue damage prediction method.

[0055] Based on the same inventive concept, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which implements the above-mentioned method for predicting fatigue damage of wind turbine blades when the computer program is executed by a processor.

[0056] The technical solution of the present invention has at least the following advantages and beneficial effects:

[0057] Based on the anisotropic parameters of the blade composite material, a coupled finite element model including aerodynamic load field, structural stress field and thermal field is constructed, and the constitutive equation of the composite laminate is set. Compared with the traditional mechanical simulation method based on physical models, it fully considers the anisotropic properties of the material and the multi-field coupling effect, avoids the errors caused by simplified boundary condition assumptions, and can more realistically reflect the stress distribution and damage evolution law of the blade under complex loads and environments in actual operation, greatly improving the accuracy of fatigue damage prediction, especially under extreme working conditions.

[0058] The system acquires three-dimensional strain field data, vibration acceleration signals, ambient temperature and humidity, and wind speed and direction data on the blade surface in real time, constructs a time-space aligned multi-dimensional monitoring data set, and changes the limitation of existing online monitoring technology that mostly uses a single physical quantity for monitoring. Through the comprehensive analysis of multi-source data, it effectively enhances the sensitivity to early fatigue damage of blades, and can capture subtle damage changes in time, thus avoiding increased maintenance costs and more serious failure accidents caused by untimely detection of early damage.

[0059] Based on the multidimensional monitoring data set, the nonlinear coupling characteristics of each load component under the joint distribution of frequency and time domains are extracted to obtain a multidimensional characteristic tensor, and the constitutive equation parameters of the composite laminate are dynamically corrected. At the same time, the boundary conditions of the coupled finite element model are optimized through the adaptive particle swarm algorithm to generate a benchmark stress field that matches the current operating conditions. This enables the model to adapt in real time to the dynamic coupling effects of multi-source loads and environmental changes during blade operation, ensuring that the model is always in the optimal state, further improving the reliability and stability of the prediction.

[0060] The multidimensional feature tensor and the benchmark stress field are input into a bidirectional long short-term memory network, and the anisotropic degradation factor of the composite material is embedded to establish a data-physics jointly driven damage evolution model. This combines the physical model's understanding of the mechanical behavior of the blade and the data-driven model's ability to process complex data. It overcomes the defect of the traditional data-driven model in lacking the integrated modeling of the anisotropic degradation mechanism of blade composite materials. It can not only accurately locate the damage position, but also quantitatively predict the remaining life of the blade, providing more valuable information for the operation and maintenance decision-making of wind farms.

[0061] The construction of a multi-dimensional monitoring data set and the application of a data-physics joint drive model have effectively reduced the false alarm rate caused by environmental noise interference. Accurate damage prediction and location can help operation and maintenance personnel quickly identify faulty blades and damaged areas, reduce unnecessary inspection and maintenance work, significantly improve the operation and maintenance efficiency of wind farms, reduce downtime losses caused by blade failure, and keep the total operation and maintenance costs of wind farms at a lower level. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 Schematic diagram of a flow chart of a method for predicting fatigue damage of a wind turbine blade according to an embodiment of the present invention;

[0063] Figure 2 Schematic diagram of the structure of a fan blade fatigue damage prediction system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0064] The following is a specific implementation method with reference to the accompanying drawings.

[0065] Reference Figure 1 , a method for predicting fatigue damage of fan blades, comprising the following steps:

[0066] Step 1: Based on the anisotropic parameters of the blade composite material, a coupled finite element model including aerodynamic load field, structural stress field and thermal field is constructed. The coupled finite element model is provided with the constitutive equation of the composite laminate.

[0067] In some embodiments, a coupled finite element model including aerodynamic load field, structural stress field, and thermal field is constructed based on the anisotropic parameters of the blade composite material. The specific process of setting the constitutive equation of the composite laminate in the coupled finite element model is as follows:

[0068] Through laminate tensile testing and digital image correlation technology, the anisotropic elastic modulus, Poisson's ratio and thermal expansion coefficient of the blade composite material were measured, and a gradient parameter database including fiber orientation and ply sequence was established;

[0069] Based on computational fluid dynamics, the unsteady aerodynamic pressure distribution on the blade surface is simulated, the aerodynamic load field is decomposed into normal pressure component and tangential friction component, and then mapped to the structural grid through the fluid-structure coupling interface;

[0070] The thermal field control equation is introduced into the structural stress field. The equivalent thermal stress caused by the temperature gradient is calculated based on the ambient temperature and humidity data, and vector superposition is performed with the aerodynamic load component.

[0071] The constitutive equation of composite laminates is defined using the variable stiffness method. The material stiffness matrix is ​​dynamically associated with the anisotropic parameters. The failure mode of the single-layer plate under different stress states is determined using the Hashin criterion.

[0072] The first three natural frequencies of the blade are measured through modal tests, and the boundary constraints of the coupled finite element model are optimized with the goal of minimizing the frequency error, completing the verification of the coupled finite element model of multi-physical fields.

[0073] Step 2: Acquire the three-dimensional strain field data of the blade surface, as well as the blade vibration acceleration signal, ambient temperature and humidity, and wind speed and direction data in real time to construct a multi-dimensional monitoring data set aligned in time and space.

[0074] In some embodiments, the specific process of acquiring the three-dimensional strain field data of the blade surface, as well as the blade vibration acceleration signal, ambient temperature and humidity, and wind speed and direction data in real time, and constructing a spatiotemporally aligned multidimensional monitoring data set is as follows:

[0075] A fiber Bragg grating sensor array is arranged in a grid of equal curvature in the key stress areas of the blade surface. Wavelength demodulation technology is used to synchronously collect three-dimensional strain components to generate spatially continuous three-dimensional strain field distribution data.

[0076] Three-axis accelerometers are installed on the blade root flange and the leading edge of the airfoil to capture vibration acceleration signals and eliminate lightning electromagnetic pulse interference through wavelet noise reduction;

[0077] Temperature and humidity sensors and ultrasonic anemometers are deployed on the top of the cabin to record ambient temperature, humidity, wind speed and direction parameters in real time. A timestamp synchronization protocol is used to align multi-source data streams to a unified time sequence.

[0078] The sensor space mapping relationship is established based on the grid node coordinates of the coupled finite element model, and the missing data points are compensated by Kriging interpolation to form a monitoring data matrix that is spatially aligned with the coupled finite element model.

[0079] The monitoring data are sliced ​​at intervals through a sliding time window, and the blade azimuth and speed information at the corresponding moments are superimposed to construct a multidimensional monitoring dataset with synchronous correlation in the time and space dimensions.

[0080] Step 3: Based on the multidimensional monitoring data set, the nonlinear coupling characteristics of each load component under the joint distribution of the frequency domain and time domain are extracted to obtain the multidimensional characteristic tensor, and the constitutive equation parameters of the composite laminate are dynamically corrected. The boundary conditions of the coupled finite element model are optimized through the adaptive particle swarm algorithm to generate a benchmark stress field that matches the current working conditions.

[0081] In some embodiments, based on a multidimensional monitoring data set, the nonlinear coupling characteristics of each load component under the frequency-time domain joint distribution are extracted to obtain a multidimensional feature tensor in the following specific process:

[0082] Perform short-time Fourier transform on the three-dimensional strain field data, extract the strain energy density spectrum of each spatial node in the frequency range of 0.5Hz to 50Hz, and generate the time-frequency energy distribution matrix;

[0083] Perform empirical mode decomposition on the vibration acceleration signal, extract the cross-correlation function amplitudes of the first three intrinsic mode functions and the wind speed time history signal, and construct the load interaction eigenvector;

[0084] The thermal stress transfer function is calculated based on the ambient temperature and humidity data. The temperature gradient and the main frequency component of the strain field are phase-matched through convolution operation to obtain the thermal-mechanical coupling coefficient sequence.

[0085] The time-frequency energy distribution matrix, load interaction eigenvectors, and thermal-mechanical coupling coefficient sequences are tensor-spliced ​​according to the space-time grid, and the kurtosis-skewness joint statistics of each component are calculated by sliding along the time axis to form a multidimensional feature tensor containing frequency domain correlation, time domain mutation, and spatial correlation.

[0086] In some embodiments, the specific process of dynamically modifying the constitutive equation parameters of the composite laminate and optimizing the boundary conditions of the coupled finite element model using an adaptive particle swarm algorithm to generate a reference stress field matching the current working condition is as follows:

[0087] The anisotropic elastic modulus change rate and Poisson's ratio offset are separated from the multidimensional characteristic tensor, and the stiffness matrix components in the constitutive equation are updated using an online recursive least squares method.

[0088] The fitness function is constructed by using the residual sum of squares between the theoretical strain field and the monitored strain field output by the coupled finite element model, and the ambient temperature gradient and wind speed fluctuation coefficient are encoded as particle swarm dimension variables.

[0089] The inertia weight is dynamically adjusted according to the convergence speed of the particle swarm during the iteration process. When the standard deviation of the particle spacing is less than the threshold, the learning factor reorganization mechanism is triggered to accelerate the convergence of the boundary constraints.

[0090] The optimized aerodynamic load distribution parameters and thermal expansion coefficient are substituted into the coupled finite element model to solve the dynamic stress field under the current rotation speed and azimuth angle. The dynamic stress field is then spatially convolved with the kurtosis-skewness statistics in the multidimensional feature tensor to generate a reference stress field that eliminates noise interference.

[0091] The matching degree of the model is verified based on the mutual information entropy of the benchmark stress field and the monitored strain field. If the mutual information entropy is lower than the set threshold, a secondary parameter correction cycle is triggered until the working condition consistency criterion is met.

[0092] Step 4: Input the multidimensional feature tensor and the benchmark stress field into the bidirectional long short-term memory network, embed the anisotropic degradation factor of the composite material, and establish a data-physics jointly driven damage evolution model.

[0093] In some embodiments, the multidimensional characteristic tensor and the reference stress field are input into a bidirectional long short-term memory network, and the anisotropic degradation factor of the composite material is embedded to establish a data-physics-driven damage evolution model. The specific process is as follows:

[0094] The multidimensional feature tensor is sliced ​​by time step and expanded into a sequence data stream along the time axis. At the same time, the reference stress field is encoded into a static condition vector according to the spatial grid node. Assume that the multidimensional feature tensor is The reference stress field is Among them, N t represents the time step; N s represents the spatial grid; N f Represents the feature dimension. For multi-dimensional feature tensors Slice along the time axis to get a sequence data stream T t represents the multi-physics coupling characteristic matrix of all spatial nodes on the blade surface at the tth time step, and The reference stress field S is mapped into a static condition vector through a fully connected layer That is: c = MLP(S), where MLP stands for multi-layer perceptron.

[0095] A parallel channel is set up in the input layer of the bidirectional long short-term memory network to receive the time-frequency feature sequence of the multidimensional feature tensor and the spatial distribution parameters of the reference stress field respectively. The parallel channels are the time-frequency feature channel and the spatial parameter channel; among them, the time-frequency feature channel directly inputs T t ; The spatial parameter channel copies the static condition vector c to each time step, forming c t The static representation of the spatial distribution parameters of the blade structure stress state in the time series. The input of the parallel channels is merged into

[0096] Based on the constitutive equation parameters of the composite laminate, the anisotropic degradation factor is dynamically calculated, including the fiber direction elastic modulus attenuation rate and the matrix interface shear strength degradation coefficient, and embedded into the gated cyclic weight matrix of the long short-term memory unit. Assume that the fiber direction elastic modulus attenuation rate is η E (t), the degradation coefficient of matrix interface shear strength is η τ (t), is dynamically calculated based on the constitutive equation parameters, as shown in the following formula (1):

[0097]

[0098] Where, E0 represents the initial fiber direction elastic modulus; E(t) represents the elastic modulus at the current moment; τ0 represents the initial matrix interface shear strength; τ(t) represents the shear strength at the current moment;

[0099] The calculation results are embedded in the long short-term memory network gating mechanism to adjust the weight of the forget gate, as shown in the following formula (2):

[0100]

[0101] Among them, f t represents the output of the forget gate at time t; σ represents the sigmoid activation function; W f represents the forget gate weight matrix; h t-1 Indicates the hidden state at the previous moment; b f represents the forget gate bias term; ⊙ represents element-wise multiplication; η(t) represents the degradation factor expansion vector; represents the two-dimensional degradation factor vector [η E (t),η τ (t)] is expanded to a vector with the same dimension as the hidden layer.

[0102] The physical constraint module is introduced into the hidden layer, and the stress gradient distribution output by the coupled finite element model is multiplied by the hidden state of the long short-term memory network to generate damage-sensitive features with spatial attention weights. Assume that the stress gradient output by the coupled finite element model is and long short-term memory network hidden state Perform tensor dot multiplication as shown in the following formula (3):

[0103]

[0104] Among them, α s,t represents the spatial attention weight, i.e., the contribution weight of the sth grid node on the blade surface to the damage-sensitive feature at time step t; h t represents the implicit state vector at time step t; W p Represents the weight matrix in the physical constraint module; Represents the stress gradient vector of the sth spatial node output by the coupled finite element model; Represents the dot product of the hidden state and the mapped stress gradient vector, and obtains the attention score of node s at time step t; softmax represents the normalization of the attention scores of all nodes.

[0105] Generate spatial attention weights, and the weighted damage-sensitive features are shown in the following formula (4):

[0106]

[0107] in, represents the weighted damage-sensitive eigenvector at the t-th time step; Represents the hidden state of the long short-term memory network at the t-th time step and the s-th spatial grid node.

[0108] Through bidirectional time series propagation, the forward causality and backward correlation of damage accumulation are captured synchronously, and the damage probability value of each grid node at different time steps is output. In combination with the material failure threshold, a spatiotemporal continuous damage evolution model is constructed. The forward and backward implicit state joint output of the bidirectional long short-term memory network is shown in the following formula (5):

[0109]

[0110] Among them, d t W represents the damage-related intermediate variable at the tth time step; d represents the output layer weight matrix; b d Represents the output layer bias term; represents the forward hidden state at the tth time step; represents the backward hidden state at the t-th time step.

[0111] The damage probability distribution is Satisfies the following formula (6):

[0112]

[0113] Among them, ps,t represents the damage probability value of the s-th spatial grid node at the t-th time step; represents the damage intermediate variable at the t-th time step and the s-th spatial grid node; θ represents the material failure threshold; and k represents the sensitivity coefficient.

[0114] The spatiotemporal continuous damage evolution model can be expressed as follows:

[0115]

[0116] Wherein, represents the cumulative damage of the s-th spatial grid node at the t-th time step; λ represents the damage accumulation rate factor, and the integral term reflects the inhibitory effect of material performance degradation on damage accumulation; represents the damage probability of the s-th node at the τ-th time step; It indicates the negative feedback effect of elastic modulus attenuation on damage accumulation, that is, the decrease in material stiffness will slow down the accumulation rate of subsequent damage.

[0117] Step 5: Based on the damage probability distribution map output by the damage evolution model, the maximum likelihood estimation is used to locate the damaged area, and the remaining service life of the blade is iteratively calculated by combining the material residual stiffness attenuation curve and the load spectrum rain flow counting results.

[0118] In some embodiments, based on the damage probability distribution map output by the damage evolution model, the maximum likelihood estimation is used to locate the damage area, and the specific process of iteratively calculating the remaining service life of the blade is based on the material residual stiffness attenuation curve and the load spectrum rain flow counting results.

[0119] The damage probability distribution map is gridded, and the blade surface is discretized into several sub-regions. Based on the principle of maximum likelihood estimation, the joint probability density function of the damage probability in each sub-region is calculated. The sub-regions with damage probability values ​​exceeding the preset threshold are screened out and identified as potential damage areas.

[0120] For potential damage areas, a quantitative relationship between residual stiffness and damage degree is established based on the composite material residual stiffness attenuation curve. The stress level of the potential damage area under the current load condition is calculated in combination with the reference stress field to assess the impact of the damage on the blade structural stiffness.

[0121] Perform rain flow counting on the load spectrum to count the number of load cycles and load amplitude distribution in the potential damage area during historical operation. Combined with the material's SN curve, calculate the fatigue life consumed in the potential damage area.

[0122] With the residual stiffness as the constraint condition, based on Miner's linear fatigue cumulative damage theory, the consumed fatigue life and the remaining fatigue life are iteratively calculated until the remaining fatigue life converges. The remaining service life of the potential damage area of ​​the blade is obtained. The remaining service life of the entire blade is determined by combining the remaining service life of each potential damage area.

[0123] Based on the same inventive concept, corresponding to any of the above embodiments, refer to Figure 2 The present invention provides a fan blade fatigue damage prediction system for implementing the aforementioned fan blade fatigue damage prediction method, comprising:

[0124] Data acquisition module, used to collect three-dimensional strain field data, capture vibration acceleration signals, record environmental data, and align multi-source data streams in time and space;

[0125] A model building module is used to construct a coupled finite element model including aerodynamic load field, structural stress field and thermal field based on the anisotropic parameters of the blade composite material, and set the constitutive equation of the composite laminate;

[0126] The feature extraction module is used to process the collected multi-dimensional monitoring data, extract the nonlinear coupling characteristics of each load component under the frequency domain-time domain joint distribution, and obtain a multi-dimensional feature tensor;

[0127] The model optimization module is used to dynamically modify the constitutive equation parameters of the composite laminate based on the multidimensional characteristic tensor, optimize the boundary conditions of the coupled finite element model through the adaptive particle swarm algorithm, and generate a benchmark stress field that matches the current working conditions;

[0128] The prediction and analysis module is used to input the multidimensional feature tensor and the benchmark stress field into a bidirectional long short-term memory network, embed the anisotropic degradation factor of the composite material, establish a damage evolution model, and locate the damaged area based on the damage probability distribution map output by the model, and calculate the remaining service life of the blade.

[0129] Based on the same inventive concept, corresponding to any of the above embodiments, the present invention provides an electronic device, including a memory and a processor, the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the wind turbine blade fatigue damage prediction method of the embodiment.

[0130] Optionally, the above-mentioned electronic device may be a server.

[0131] In addition, this embodiment further provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the method for predicting fatigue damage of wind turbine blades of the embodiment is implemented.

[0132] It is understood that the processor in the embodiments of the present invention may be a central processing unit (CPU), or may be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field programmable gate arrays (FPGA), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. The general-purpose processor may be a microprocessor or any conventional processor.

[0133] The method steps in the embodiments of the present invention can be implemented by hardware or by a processor executing software instructions. The software instructions can be composed of corresponding software modules, which can be stored in random access memory (RAM), flash memory, read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, mobile hard disks, CD-ROMs, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be an integral part of the processor. The processor and the storage medium can be located in an ASIC.

[0134] In the above embodiments, all or part of the embodiments can be implemented by software, hardware, firmware, or any combination thereof. When implemented using software, all or part of the embodiments can be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions according to the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a storage medium or transmitted via a storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that includes one or more available media. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)).

Claims

1. A method for predicting fatigue damage of fan blades, characterized in that: The following steps are involved: A coupled finite element model including aerodynamic load field, structural stress field and thermal field is constructed based on the anisotropic parameters of the blade composite material, wherein the coupled finite element model is provided with a constitutive equation of the composite laminate; Acquire blade surface three-dimensional strain field data, blade vibration acceleration signals, ambient temperature and humidity, and wind speed and direction data in real time to construct a multi-dimensional monitoring data set aligned in time and space; Based on a multidimensional monitoring data set, the nonlinear coupling characteristics of each load component under the joint frequency-time domain distribution are extracted to obtain a multidimensional characteristic tensor. The constitutive equation parameters of the composite laminate are dynamically modified. The boundary conditions of the coupled finite element model are optimized using an adaptive particle swarm algorithm to generate a benchmark stress field that matches the current working conditions. The multidimensional characteristic tensor and the benchmark stress field are input into a bidirectional long short-term memory network, and the anisotropic degradation factor of the composite material is embedded to establish a data-physics-driven damage evolution model. Based on the damage probability distribution map output by the damage evolution model, maximum likelihood estimation is used to locate the damaged area. Combined with the material residual stiffness attenuation curve and the load spectrum rain flow counting results, the remaining service life of the blade is iteratively calculated.

2. The method for predicting fatigue damage of wind turbine blades according to claim 1, wherein: The specific process of constructing a coupled finite element model including aerodynamic load field, structural stress field and thermal field based on the anisotropic parameters of the blade composite material, wherein the constitutive equation of the composite laminate is provided in the coupled finite element model, is as follows: Through laminate tensile testing and digital image correlation technology, the anisotropic elastic modulus, Poisson's ratio and thermal expansion coefficient of the blade composite material were measured, and a gradient parameter database including fiber orientation and ply sequence was established; Based on computational fluid dynamics, the unsteady aerodynamic pressure distribution on the blade surface is simulated, the aerodynamic load field is decomposed into normal pressure component and tangential friction component, and then mapped to the structural grid through the fluid-structure coupling interface; The thermal field control equation is introduced into the structural stress field. The equivalent thermal stress caused by the temperature gradient is calculated based on the ambient temperature and humidity data, and vector superposition is performed with the aerodynamic load component. The constitutive equation of composite laminates is defined using the variable stiffness method. The material stiffness matrix is ​​dynamically associated with the anisotropic parameters. The failure mode of the single-layer plate under different stress states is determined using the Hashin criterion. The first three natural frequencies of the blade are measured through modal tests, and the boundary constraints of the coupled finite element model are optimized with the goal of minimizing the frequency error, completing the verification of the coupled finite element model of multi-physical fields.

3. The method for predicting fatigue damage of wind turbine blades according to claim 1, wherein: The specific process of acquiring the blade surface three-dimensional strain field data, blade vibration acceleration signal, ambient temperature and humidity, and wind speed and direction data in real time to construct a spatiotemporally aligned multidimensional monitoring data set is as follows: A fiber Bragg grating sensor array is arranged in a grid of equal curvature in the key stress areas of the blade surface. Wavelength demodulation technology is used to synchronously collect three-dimensional strain components to generate spatially continuous three-dimensional strain field distribution data. Three-axis accelerometers are installed on the blade root flange and the leading edge of the airfoil to capture vibration acceleration signals and eliminate lightning electromagnetic pulse interference through wavelet noise reduction; Temperature and humidity sensors and ultrasonic anemometers are deployed on the top of the cabin to record ambient temperature, humidity, wind speed and direction parameters in real time. A timestamp synchronization protocol is used to align multi-source data streams to a unified time sequence. The sensor space mapping relationship is established based on the grid node coordinates of the coupled finite element model, and the missing data points are compensated by Kriging interpolation to form a monitoring data matrix that is spatially aligned with the coupled finite element model. The monitoring data are sliced ​​at intervals through a sliding time window, and the blade azimuth and speed information at the corresponding moments are superimposed to construct a multidimensional monitoring dataset with synchronous correlation in the time and space dimensions.

4. The method for predicting fatigue damage of wind turbine blades according to claim 1, wherein: The specific process of extracting the nonlinear coupling characteristics of each load component under the frequency domain-time domain joint distribution based on the multidimensional monitoring data set to obtain the multidimensional feature tensor is as follows: Perform short-time Fourier transform on the three-dimensional strain field data, extract the strain energy density spectrum of each spatial node in the frequency range of 0.5Hz to 50Hz, and generate the time-frequency energy distribution matrix; Perform empirical mode decomposition on the vibration acceleration signal, extract the cross-correlation function amplitudes of the first three intrinsic mode functions and the wind speed time history signal, and construct the load interaction eigenvector; The thermal stress transfer function is calculated based on the ambient temperature and humidity data. The temperature gradient and the main frequency component of the strain field are phase-matched through convolution operation to obtain the thermal-mechanical coupling coefficient sequence. The time-frequency energy distribution matrix, load interaction eigenvectors, and thermal-mechanical coupling coefficient sequences are tensor-spliced ​​according to the space-time grid, and the kurtosis-skewness joint statistics of each component are calculated by sliding along the time axis to form a multidimensional feature tensor containing frequency domain correlation, time domain mutation, and spatial correlation.

5. The method for predicting fatigue damage of wind turbine blades according to claim 4, wherein: The specific process of dynamically correcting the constitutive equation parameters of the composite laminate and optimizing the boundary conditions of the coupled finite element model through the adaptive particle swarm algorithm to generate a reference stress field matching the current working condition is as follows: The anisotropic elastic modulus change rate and Poisson's ratio offset are separated from the multidimensional characteristic tensor, and the stiffness matrix components in the constitutive equation are updated using an online recursive least squares method. The fitness function is constructed by using the residual sum of squares between the theoretical strain field and the monitored strain field output by the coupled finite element model, and the ambient temperature gradient and wind speed fluctuation coefficient are encoded as particle swarm dimension variables. The inertia weight is dynamically adjusted according to the convergence speed of the particle swarm during the iteration process. When the standard deviation of the particle spacing is less than the threshold, the learning factor reorganization mechanism is triggered to accelerate the convergence of the boundary constraints. The optimized aerodynamic load distribution parameters and thermal expansion coefficient are substituted into the coupled finite element model to solve the dynamic stress field under the current rotation speed and azimuth angle. The dynamic stress field is then spatially convolved with the kurtosis-skewness statistics in the multidimensional feature tensor to generate a reference stress field that eliminates noise interference. The matching degree of the model is verified based on the mutual information entropy of the benchmark stress field and the monitored strain field. If the mutual information entropy is lower than the set threshold, a secondary parameter correction cycle is triggered until the working condition consistency criterion is met.

6. The method for predicting fatigue damage of wind turbine blades according to claim 1, wherein: The specific process of inputting the multidimensional characteristic tensor and the reference stress field into the bidirectional long short-term memory network, embedding the anisotropic degradation factor of the composite material, and establishing the data-physics joint driven damage evolution model is as follows: The multi-dimensional feature tensor is sliced ​​by time step and expanded into a sequence data stream along the time axis, while the reference stress field is encoded into a static condition vector according to the spatial grid node; A parallel channel is set up at the input layer of the bidirectional long short-term memory network to receive the time-frequency feature sequence of the multidimensional feature tensor and the spatial distribution parameters of the reference stress field respectively; Dynamically calculate anisotropic degradation factors based on the constitutive equation parameters of composite laminates, including the fiber direction elastic modulus decay rate and the matrix interface shear strength degradation coefficient, and embed them into the gated recurrent weight matrix of the long short-term memory unit; A physical constraint module is introduced into the hidden layer to perform tensor dot multiplication between the stress gradient distribution output by the coupled finite element model and the hidden state of the long short-term memory network to generate damage-sensitive features with spatial attention weights. Through bidirectional time propagation, the forward causality and backward correlation of damage accumulation are synchronously captured, the damage probability value of each grid node at different time steps is output, and a spatiotemporal continuous damage evolution model is constructed in combination with the material failure threshold.

7. The method for predicting fatigue damage of wind turbine blades according to claim 1, wherein: The specific process of iteratively calculating the remaining service life of the blade based on the damage probability distribution map output by the damage evolution model, locating the damage area using maximum likelihood estimation, and combining the material residual stiffness attenuation curve and the load spectrum rain flow counting results is as follows: The damage probability distribution map is gridded, and the blade surface is discretized into several sub-regions. Based on the principle of maximum likelihood estimation, the joint probability density function of the damage probability in each sub-region is calculated. The sub-regions with damage probability values ​​exceeding the preset threshold are screened out and identified as potential damage areas. For potential damage areas, a quantitative relationship between residual stiffness and damage degree is established based on the composite material residual stiffness attenuation curve. The stress level of the potential damage area under the current load condition is calculated in combination with the reference stress field to assess the impact of the damage on the blade structural stiffness. Perform rain flow counting on the load spectrum to count the number of load cycles and load amplitude distribution in the potential damage area during historical operation. Combined with the material's SN curve, calculate the fatigue life consumed in the potential damage area. With the residual stiffness as the constraint condition, based on Miner's linear fatigue cumulative damage theory, the consumed fatigue life and the remaining fatigue life are iteratively calculated until the remaining fatigue life converges. The remaining service life of the potential damage area of ​​the blade is obtained. The remaining service life of the entire blade is determined by combining the remaining service life of each potential damage area.

8. A fan blade fatigue damage prediction system, used to implement the fan blade fatigue damage prediction method according to any one of claims 1 to 7, characterized in that: include: Data acquisition module, used to collect three-dimensional strain field data, capture vibration acceleration signals, record environmental data, and align multi-source data streams in time and space; A model building module is used to construct a coupled finite element model including aerodynamic load field, structural stress field and thermal field based on the anisotropic parameters of the blade composite material, and set the constitutive equation of the composite laminate; The feature extraction module is used to process the collected multi-dimensional monitoring data, extract the nonlinear coupling characteristics of each load component under the frequency domain-time domain joint distribution, and obtain a multi-dimensional feature tensor; The model optimization module is used to dynamically modify the constitutive equation parameters of the composite laminate based on the multidimensional characteristic tensor, optimize the boundary conditions of the coupled finite element model through the adaptive particle swarm algorithm, and generate a benchmark stress field that matches the current working conditions; The prediction and analysis module is used to input the multidimensional feature tensor and the benchmark stress field into a bidirectional long short-term memory network, embed the anisotropic degradation factor of the composite material, establish a damage evolution model, and locate the damaged area based on the damage probability distribution map output by the model, and calculate the remaining service life of the blade.

9. An electronic device, characterized in that: The electronic device comprises a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the wind turbine blade fatigue damage prediction method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for predicting fatigue damage of a wind turbine blade according to any one of claims 1 to 7 is implemented.

Images