Method for determining fatigue degree of photovoltaic support cable

Abstract

The application provides a kind of method for determining the fatigue degree of photovoltaic support cable, belongs to the technical field of photovoltaic support cable detection, and the application establishes the multi-physical field coupling finite element model of integrated electrochemical corrosion kinetics equation, ultraviolet light aging degradation kinetics equation and thermal coupling constitutive relationship to calculate temperature field, humidity field, irradiation field distribution and map as material degradation coefficient, establishes fatigue crack propagation simulation model based on random phase field theory, introduces Gaussian random field to represent the spatial variability of material performance, and adjusts the phase evolution rate parameter according to the damage confidence, adopts the stress path reconstruction algorithm driven by topology optimization to optimize the material distribution of anchoring area, calculates the cumulative fatigue damage degree according to the optimized stress field distribution and crack propagation path, and solves the technical problem that the fatigue damage evolution law of photovoltaic support cable in multi-field coupling environment is difficult to accurately predict.

Description

Technical Field

[0001] This invention belongs to the field of photovoltaic support cable testing technology, and specifically relates to a method for determining the fatigue degree of photovoltaic support cables. Background Technology

[0002] As a critical load-bearing component of photovoltaic power plants, solar cable supports endure complex environmental conditions such as wind loads, temperature cycling, and ultraviolet radiation over long periods. Traditional fatigue life assessment methods primarily rely on stress amplitude statistics combined with SN curves for calculation, or on periodic visual inspection and ultrasonic testing to identify macroscopic damage. However, when applied to the service environment of solar cable supports, these methods struggle to capture real-time damage information during the microcrack initiation stage and cannot comprehensively consider the degradation effects of multiple physical fields, such as electrochemical corrosion, ultraviolet aging, and thermo-coupling, on material properties. In current photovoltaic power plant operation and maintenance practices, the lack of dynamic monitoring methods for the evolution of microscopic damage within the cables often leads to fatigue assessments lagging behind actual damage development, resulting in sudden fractures at stress concentration points in the anchorage region. In other words, existing technologies suffer from the technical challenge of accurately predicting the fatigue damage evolution of solar cable supports under multi-field coupling environments. Summary of the Invention

[0003] In view of this, the present invention provides a method for determining the fatigue degree of photovoltaic support cables, which can solve the technical problem in the prior art that it is difficult to accurately predict the fatigue damage evolution law of photovoltaic support cables under multi-field coupling environment.

[0004] This invention is implemented as follows: This invention provides a method for determining the fatigue degree of photovoltaic support cables. An array of acoustic emission sensors is uniformly arranged axially on the surface of the photovoltaic support cable to collect acoustic emission signals during the service life of the photovoltaic support cable, and wavelet packet decomposition is performed on the acoustic emission signals to extract values ​​from 0 to 50. The energy ratio feature vector of the frequency band is input into the fatigue damage identification model to identify three damage modes inside the photovoltaic support cable: microcrack initiation, crack propagation, and wire fracture. The model outputs damage mode category labels and damage confidence scores. A multi-physics coupled finite element model of the photovoltaic support cable is established, integrating electrochemical corrosion kinetics, ultraviolet aging degradation kinetics, and thermo-coupling constitutive relations. A sequential coupling algorithm is used to calculate the temperature field distribution, humidity field distribution, and irradiation field distribution. These distributions are then mapped to the degradation coefficients of the elastic modulus, yield strength, and fracture toughness of the photovoltaic support cable material. Sub-model techniques are then used to further analyze these degradation coefficients. The local mesh of the photovoltaic support cable anchorage area is refined. A fatigue crack propagation simulation model of the photovoltaic support cable is established based on random phase field theory. The crack interface is blurred into a continuous phase field variable. A Gaussian random field is introduced to characterize the spatial variability of the photovoltaic support cable steel wire material properties. An adaptive time step algorithm is used to calculate the crack propagation path. The phase field evolution rate parameter in the photovoltaic support cable fatigue crack propagation simulation model is adjusted according to the damage confidence. A topology optimization-driven stress path reconstruction algorithm is used to optimize the material distribution in the photovoltaic support cable anchorage area. Based on the optimized stress field distribution and crack propagation path, the cumulative fatigue damage degree of the photovoltaic support cable is calculated and the fatigue state of the photovoltaic support cable is determined.

[0005] The wavelet packet decomposition process involves decomposing the acoustic emission signal into multiple layers according to a binary tree structure, up to the fifth layer, to obtain wavelet packet coefficients for 32 frequency bands. The energy value of each frequency band is calculated as the sum of the squares of all wavelet packet coefficients within the band. After normalizing the energy values ​​of the 32 frequency bands, an energy ratio feature vector is formed.

[0006] Among them, the fatigue injury recognition model adopts a composable reasoning architecture based on neural module networks. It decomposes the injury pattern recognition task into three reusable neural modules: feature extraction module, temporal modeling module, and pattern classification module. The module combination scheme is automatically selected based on the characteristics of the input data through program generation technology.

[0007] Among them, the discrimination model of the fatigue damage identification model adopts a multi-scale classification optimization mechanism based on hierarchical soft maximum value. It constructs a three-layer classifier structure to achieve progressive classification from coarse-grained to fine-grained. Each classifier uses a soft maximum value function to calculate the class probability and output the uncertainty quantification index.

[0008] Among them, the training dataset for the fatigue damage identification model is established by using a generative adversarial network to synthesize failure samples. The generator of the generative adversarial network learns the probability distribution of failure samples in the energy ratio feature vector space, and the discriminator of the generative adversarial network distinguishes between real failure samples and generated failure samples.

[0009] The training dataset is established by combining synthetic nearest neighbor oversampling technology to oversample failed samples. New samples are randomly generated on the connection between each failed sample and its nearest neighbor in the feature space. Tomk link undersampling technology is used to delete samples in normal service that are too close to failed samples.

[0010] Among them, the fatigue damage identification model training adopts a cost-sensitive learning strategy to assign a weight of 5 times that of normal service samples to failed samples, defines the loss function as the product of cross-entropy loss and weight coefficients, and uses an adaptive moment estimation optimization algorithm to update the neural module network parameters.

[0011] The module activation adjustment function is used to adjust the module activation weights of the composable inference architecture based on neural module networks. The module activation adjustment function calculates the module activation adjustment value based on three data: the feature complexity of the current batch input data, the prediction accuracy of the historical batch model, and the rate of change of the validation set loss.

[0012] Among them, the electrochemical corrosion kinetic equation describes the relationship between the corrosion rate and time of the photovoltaic support cable steel wire in an acid rain environment, and the corrosion rate is related to acid rain. The value, temperature, and surface potential difference of the photovoltaic support cable steel wire are related. The mass loss of the surface material of the photovoltaic support cable steel wire per unit time is calculated by using Faraday's law.

[0013] Among them, the ultraviolet light aging degradation kinetic equation describes the performance degradation law of the polymer material of photovoltaic bracket cable sheath under ultraviolet irradiation. The degradation rate is related to the ultraviolet irradiation intensity, cumulative irradiation dose and material temperature. The exponential relationship between degradation rate and material temperature is established in the form of Arrhenius equation.

[0014] Among them, the thermo-coupling constitutive relation describes the stress-strain response of the photovoltaic support cable material under temperature cyclic loading. Considering the thermal expansion effect caused by temperature and the temperature dependence of the elastic modulus of the photovoltaic support cable material, the incremental constitutive equation is used to couple the temperature increment and strain increment to calculate the stress increment.

[0015] The sequential coupling algorithm first solves the temperature field control equation to obtain the temperature field distribution of the photovoltaic support cable system. Then, it substitutes the temperature field distribution as a heat source term into the humidity field control equation to solve for the humidity field distribution. Finally, it substitutes the temperature field distribution and humidity field distribution as boundary conditions into the electrochemical corrosion kinetic equation to calculate the corrosion depth distribution.

[0016] Among them, the sub-model technology establishes a full-scale coarse mesh model of the photovoltaic support cable for preliminary stress analysis, extracts the boundary displacement field of the anchorage area of ​​the photovoltaic support cable as the boundary condition of the sub-model, and establishes a fine mesh sub-model for the anchorage area of ​​the photovoltaic support cable. The mesh size of the fine mesh sub-model is one-tenth of the mesh size of the coarse mesh model.

[0017] Among them, the stochastic phase field theory characterizes cracks as continuous phase field variables distributed in space. A value of 0 for the continuous phase field variable represents intact material, a value of 1 for the continuous phase field variable represents complete fracture, and a value between 0 and 1 for the continuous phase field variable represents the crack transition zone. The phase field evolution equation couples elastic energy and fracture energy.

[0018] Among them, the Gaussian random field characterizes the spatial variability of the photovoltaic support cable steel wire material properties. The elastic modulus and yield strength of the photovoltaic support cable steel wire material parameters are regarded as spatial random fields. The Gaussian probability distribution is used to describe the values ​​of the material parameters at each point in space. The correlation of the material parameter values ​​at different spatial locations is defined by the covariance function.

[0019] Specifically, the phase field evolution rate parameter is adjusted according to the damage confidence level. When the damage confidence level is between 0.8 and 1.0, the phase field evolution rate parameter is increased to 1.5 times the baseline value. When the damage confidence level is between 0.5 and 0.8, the baseline value is maintained. When the damage confidence level is less than 0.5, the phase field evolution rate parameter is decreased to 0.6 times the baseline value.

[0020] This invention achieves intelligent identification of three damage modes—microcrack initiation, crack propagation, and wire breakage—by combining wavelet packet energy ratio feature extraction of acoustic emission signals with a composable reasoning architecture of neural module networks, thus overcoming the limitation of traditional detection methods in failing to capture early damage in real time. By establishing a multi-physics coupled finite element model integrating electrochemical corrosion kinetics, ultraviolet aging degradation kinetics, and thermo-coupling constitutive relations, and employing a sequential coupling algorithm to calculate the distributions of temperature, humidity, and irradiation fields and map them to material degradation coefficients, this invention addresses the problem of traditional methods neglecting the coupling effect of environmental factors. Furthermore, by introducing a Gaussian random field to characterize the spatial variability of material properties through a crack propagation simulation model based on random phase field theory, and adaptively adjusting the phase field evolution rate parameter according to the damage confidence level, combined with a topology-optimized stress path reconstruction algorithm to optimize the stress distribution in the anchorage region, this invention achieves accurate simulation of the entire fatigue damage evolution process. In summary, this invention solves the technical problem mentioned in the background art of the difficulty in accurately predicting the fatigue damage evolution law of photovoltaic support cables under multi-field coupling environments. Attached Figure Description

[0021] Figure 1 This is a flowchart of the method of the present invention.

[0022] Figure 2 The temperature field and corrosion depth distribution are shown in the multiphysics coupling analysis of photovoltaic support cables.

[0023] Figure 3 This is a diagram showing the evolution of fatigue crack propagation paths based on random phase field theory. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below.

[0025] like Figure 1 The diagram shown is a flowchart of a method for determining the fatigue degree of photovoltaic support cables provided by the present invention. This method includes the following steps:

[0026] S1. An array of acoustic emission sensors is uniformly arranged along the axial direction on the surface of the photovoltaic support cable to collect acoustic emission signals during the service process of the photovoltaic support cable, and the acoustic emission signals are decomposed by wavelet packet to extract the energy ratio feature vector of the frequency band from 0 to 50 kHz.

[0027] S2. Input the energy ratio feature vector into the fatigue damage identification model for processing, identify three damage modes: microcrack initiation, crack propagation, and wire breakage inside the photovoltaic support cable, and output the damage mode category label and damage confidence.

[0028] S3. Establish a multi-physics field coupled finite element model of photovoltaic support cables, integrate electrochemical corrosion kinetic equation, ultraviolet light aging degradation kinetic equation and thermo-coupling constitutive relation, and use sequential coupling algorithm to calculate temperature field distribution, humidity field distribution and irradiation field distribution.

[0029] S4. Map the temperature field distribution, humidity field distribution, and irradiation field distribution to the elastic modulus degradation coefficient, yield strength degradation coefficient, and fracture toughness degradation coefficient of the photovoltaic support cable material, and refine the local mesh of the photovoltaic support cable anchorage area using sub-model technology.

[0030] S5. Based on the random phase field theory, a simulation model for fatigue crack propagation of photovoltaic support cable is established. The crack interface is fuzzified into a continuous phase field variable. A Gaussian random field is introduced to characterize the spatial variability of the steel wire material properties of the photovoltaic support cable. An adaptive time step algorithm is used to calculate the crack propagation path.

[0031] S6. Adjust the phase field evolution rate parameter in the photovoltaic support cable fatigue crack propagation simulation model according to the damage confidence level. When the damage confidence level is [0.8, 1.0], increase the phase field evolution rate parameter to 1.5 times the baseline value of the phase field evolution rate parameter. When the damage confidence level is [0.5, 0.8), maintain the baseline value of the phase field evolution rate parameter. When the damage confidence level is less than 0.5, decrease the phase field evolution rate parameter to 0.6 times the baseline value of the phase field evolution rate parameter.

[0032] S7. The stress path reconstruction algorithm driven by topology optimization is used to optimize the material distribution in the anchorage area of ​​the photovoltaic support cable. The objective function is to minimize the stress concentration factor. The gradient of the design variable is calculated by adjoint sensitivity analysis, and the material density distribution is iteratively optimized to homogenize the stress field.

[0033] S8. Calculate the cumulative fatigue damage degree of the photovoltaic support cable based on the optimized stress field distribution and the crack propagation path. When the cumulative fatigue damage degree is greater than or equal to 0.85, the photovoltaic support cable is determined to be in the critical state of fatigue failure. When the cumulative fatigue damage degree is within [0.6, 0.85), the photovoltaic support cable is determined to be in the state of fatigue damage development. When the cumulative fatigue damage degree is less than 0.6, the photovoltaic support cable is determined to be in the state of safe service.

[0034] The wavelet packet decomposition steps specifically include: decomposing the acoustic emission signal into multiple layers according to a binary tree structure; the first layer decomposes the acoustic emission signal into low-frequency approximation components and high-frequency detail components; each subsequent layer continues to decompose the low-frequency approximation components and high-frequency detail components of the previous layer, until the fifth layer is reached to obtain wavelet packet coefficients of 32 frequency bands; the energy value of each frequency band is calculated as the sum of the squares of all wavelet packet coefficients in the frequency band; and the energy values ​​of the 32 frequency bands are normalized to form the energy ratio feature vector.

[0035] The fatigue damage recognition model is structured as follows: the generative model employs a composable reasoning architecture based on neural module networks, decomposing the damage pattern recognition task into three reusable neural modules: a feature extraction module, a temporal modeling module, and a pattern classification module. The feature extraction module uses one-dimensional convolutional layers to extract local features of the energy ratio feature vector. The temporal modeling module uses a long short-term memory network to capture the time dependence of the acoustic emission signal. The pattern classification module uses fully connected layers to output the probability distribution of damage patterns. The module combination scheme is automatically selected based on the characteristics of the input data using procedural generation technology, and the policy gradient algorithm in reinforcement learning is utilized. The optimization module selection strategy improves learning efficiency by sharing convolution kernel parameters and recurrent unit parameters among different neural modules. The discrimination model adopts a multi-scale classification optimization mechanism based on hierarchical soft maximum values, constructing a three-layer classifier structure. The first layer coarse classifier distinguishes between damaged and undamaged states, the second layer medium classifier distinguishes between crack-type damage and wire-broken damage, and the third layer fine classifier identifies the specific damage patterns of microcrack initiation, crack propagation, and wire breakage. Each classifier uses a soft maximum function to calculate the class probability and outputs an uncertainty quantification index. Through hierarchical decision-making, progressive classification from coarse to fine granular is achieved.

[0036] The steps for establishing the training dataset of the fatigue damage identification model specifically include: collecting acoustic emission signals from 100 photovoltaic support cables during a cyclic loading test on a fatigue testing machine as normal service samples, and collecting acoustic emission signals from 20 photovoltaic support cables during fatigue fracture failure as failure samples; synthesizing the failure samples using a generative adversarial network (GAN), where the generator of the GAN learns the probability distribution of the failure samples in the energy ratio feature vector space, generates adversarial samples in the boundary region of the probability distribution, and the discriminator of the GAN distinguishes between real failure samples and generated failure samples; oversampling the failure samples using synthetic nearest neighbor oversampling technology, randomly generating new samples on the connection line between each failure sample and its nearest neighbor in the feature space, and deleting samples that are too close to the failure samples in the normal service samples using Tomk link undersampling technology, so that the ratio of normal service samples to failure samples in the training dataset reaches 3:1.

[0037] The specific steps for training the fatigue damage recognition model include: dividing the training dataset into a training set and a validation set in a 7:3 ratio; assigning a weight of 5 times that of the normal service samples to the failed samples using a cost-sensitive learning strategy; defining the loss function as the product of cross-entropy loss and weight coefficients; updating the neural module network parameters using an adaptive moment estimation optimization algorithm; setting the learning rate to 0.001 and the batch size to 32; evaluating the model performance on the validation set after every 10 batches during training; stopping training when the validation set loss does not decrease for 20 consecutive batches; optimizing the module selection strategy using reinforcement learning; defining the reward function as the model's classification accuracy on the validation set; and updating the module selection probability distribution using the policy gradient algorithm.

[0038] The fatigue damage identification model achieves adaptive identification of acoustic emission features of different damage modes through the composable reasoning architecture based on neural module networks, avoiding the problem of insufficient adaptation of traditional fixed network structures to the complexity of acoustic emission signals. The multi-scale classification optimization mechanism based on hierarchical soft maximum values ​​provides uncertainty quantification indicators while ensuring identification accuracy, providing reliable damage mode input for subsequent fatigue damage assessment. The generative adversarial network synthesizes the failure samples and the cost-sensitive learning strategy solves the sample imbalance problem caused by the scarcity of failure samples in actual engineering, enabling the model to effectively identify fatigue failure modes.

[0039] The module activation adjustment function is used to adjust the module activation weights of the composable inference architecture based on neural module networks. The module activation adjustment function calculates the module activation adjustment value based on three data: the feature complexity of the current batch input data, the prediction accuracy of the historical batch model, and the rate of change of the validation set loss. When the module activation adjustment value is greater than or equal to 0.75, an enhanced module combination scheme is adopted to simultaneously activate the feature extraction module, the temporal modeling module, and the pattern classification module, and the connection weights between the neural modules are increased to 1.3 times the baseline value of the connection weights. When the module activation adjustment value is in the range [0.4, 0.75), a standard module combination scheme is adopted to activate each neural module in sequence while maintaining the baseline value of the connection weights. When the module activation adjustment value is less than 0.4, a simplified module combination scheme is adopted to activate only the feature extraction module and the pattern classification module, and the connection weights between the neural modules are reduced to 0.7 times the baseline value of the connection weights.

[0040] The electrochemical corrosion kinetic equation describes the relationship between the corrosion rate of the photovoltaic support cable wire and time in an acid rain environment. The corrosion rate is related to the pH value of the acid rain, temperature, and the surface potential difference of the photovoltaic support cable wire. The mass loss of the surface material of the photovoltaic support cable wire per unit time is calculated by Faraday's law, and the mass loss is converted into the reduction of the cross-sectional area of ​​the photovoltaic support cable wire.

[0041] The ultraviolet light aging degradation kinetic equation describes the performance degradation law of the polymer material of the photovoltaic support cable sheath under ultraviolet irradiation. The degradation rate is related to the ultraviolet irradiation intensity, cumulative irradiation dose, and material temperature. The exponential relationship between the degradation rate and the material temperature is established in the form of the Arrhenius equation, and the decay curve of the mechanical properties of the polymer material over time is calculated by integration.

[0042] The thermo-coupling constitutive relation describes the stress-strain response of the photovoltaic support cable material under temperature cyclic loading. Considering the thermal expansion effect caused by temperature and the temperature dependence of the elastic modulus of the photovoltaic support cable material, the incremental constitutive equation is used to couple the temperature increment and strain increment to calculate the stress increment.

[0043] The calculation steps of the sequential coupling algorithm are as follows: First, solve the temperature field control equation to obtain the temperature field distribution of the photovoltaic support cable system. Then, substitute the temperature field distribution as a heat source term into the humidity field control equation to solve for the humidity field distribution. Next, substitute the temperature field distribution and the humidity field distribution as boundary conditions into the electrochemical corrosion kinetic equation to calculate the corrosion depth distribution. Finally, substitute the irradiation field distribution into the ultraviolet light aging degradation kinetic equation to calculate the material performance degradation. Finally, input the temperature field distribution, the humidity field distribution, the corrosion depth distribution, and the material performance degradation as material parameters into the mechanical analysis model.

[0044] The implementation steps of the sub-model technology are as follows: a full-scale coarse mesh model of the photovoltaic support cable is established for preliminary stress analysis; the displacement field of the boundary of the anchorage area of ​​the photovoltaic support cable is extracted as the boundary condition of the sub-model; a fine mesh sub-model is established for the anchorage area of ​​the photovoltaic support cable, the mesh size of the fine mesh sub-model is one-tenth of the mesh size of the coarse mesh model; and high-precision stress calculation is performed in the fine mesh sub-model to obtain details of local stress concentration.

[0045] The stochastic phase field theory characterizes cracks as a continuous spatial distribution of continuous phase field variables. A value of 0 for a continuous phase field variable represents intact material, a value of 1 for a continuous phase field variable represents complete fracture, and a value between 0 and 1 represents the crack transition zone. The phase field evolution equation couples elastic energy and fracture energy, driving crack propagation by minimizing the total energy of the system.

[0046] The method for characterizing the spatial variability of photovoltaic support cable steel wire material properties using Gaussian random fields is as follows: the elastic modulus and yield strength of the photovoltaic support cable steel wire material parameters are regarded as spatial random fields, the Gaussian probability distribution is used to describe the values ​​of the material parameters at each point in space, the correlation of the material parameter values ​​at different spatial locations is defined by the covariance function, and random field samples that satisfy the given statistical characteristics are generated using spectral representation methods.

[0047] The adaptive time step algorithm dynamically adjusts the time step based on the rate of change of the continuous phase field variables. When the crack propagates rapidly, the rate of change of the continuous phase field variables is large, so the time step is reduced to improve the calculation accuracy. When the crack propagates slowly, the rate of change of the continuous phase field variables is small, so the time step is increased to improve the calculation efficiency.

[0048] The topology optimization-driven stress path reconstruction algorithm uses a variable density method with material density as the design variable. A material density value of 0 represents no material, and a material density value of 1 represents solid material. The relationship between the material density and the elastic modulus is established through an interpolation model. The stress concentration factor is defined as the ratio of the maximum stress to the average stress. The adjoint sensitivity analysis efficiently calculates the gradient of the objective function with respect to the design variables by solving the adjoint equation. The moving asymptote algorithm is used to iteratively update the material density distribution. The level set method uses implicit functions to represent the material boundary and achieves a clear representation of the topological boundary by evolving contour lines with a level set function value of zero.

[0049] The cumulative fatigue damage is calculated using the linear damage accumulation theory. The service process of the photovoltaic support cable is divided into multiple stress cycle segments. The damage degree of each stress cycle segment is the ratio of the number of cycles in the stress cycle segment to the fatigue life under the corresponding stress amplitude. The cumulative fatigue damage degree is the sum of the damage degrees of all stress cycle segments.

[0050] The feature complexity is quantified by calculating the information entropy of the energy-to-feature vector. The larger the information entropy, the more dispersed the feature distribution, and the higher the feature complexity.

[0051] The historical batch model prediction accuracy is the ratio of the number of correctly predicted samples in the most recent 50 batches on the validation set to the total number of samples.

[0052] The rate of change of the validation set loss is the difference between the current batch validation set loss and the previous batch validation set loss divided by the previous batch validation set loss.

[0053] The formula for calculating the module activation adjustment value is: the normalized value of the feature complexity multiplied by 0.5, the historical batch model prediction accuracy multiplied by 0.3, and the normalized value of the validation set loss change rate multiplied by 0.2. The normalized value of the feature complexity is the feature complexity divided by the maximum value of the feature complexity during training, and the normalized value of the validation set loss change rate is the validation set loss change rate divided by the maximum value of the validation set loss change rate during training.

[0054] The elastic modulus degradation coefficient is the ratio of the current elastic modulus to the initial elastic modulus, the yield strength degradation coefficient is the ratio of the current yield strength to the initial yield strength, and the fracture toughness degradation coefficient is the ratio of the current fracture toughness to the initial fracture toughness.

[0055] The phase field evolution rate parameter controls the speed of crack propagation; the larger the value of the phase field evolution rate parameter, the faster the crack propagates.

[0056] The baseline value of the phase field evolution rate parameter is the value of the phase field evolution rate parameter in the photovoltaic support cable fatigue crack propagation simulation model under standard working conditions.

[0057] The synthetic nearest neighbor oversampling technique selects the five nearest neighbors of each minority class sample in the feature space, and randomly selects a position on the line connecting the minority class sample and each of the nearest neighbors to generate the new sample.

[0058] The Tomk link is a pair of samples in the feature space, wherein the two samples in the pair belong to different categories and are each other's nearest neighbors.

[0059] The connection weight baseline value is the value of the connection weight between neural modules in the composable inference architecture based on neural module networks under standard training mode.

[0060] Optionally, the present invention also provides a method for determining the fatigue degree of photovoltaic support cables by means of a computer, wherein the computer is provided with a readable storage medium, the readable storage medium stores program instructions, and the program instructions execute the above-described method for determining the fatigue degree of photovoltaic support cables when the computer is run.

[0061] The specific implementation methods of the above steps are described in detail below.

[0062] The specific implementation of step S1 is as follows: First, an acoustic emission sensor is arranged every 2m along the axial direction of the photovoltaic support cable to form an acoustic emission sensor array. The acoustic emission sensor array covers the entire length of the photovoltaic support cable to achieve spatial positioning. The sensor sampling frequency is set to 1MHz to capture high-frequency acoustic emission signals during the microcrack initiation stage. The collected acoustic emission signals are then imported into a wavelet packet decomposition algorithm for frequency domain feature extraction. The wavelet packet decomposition uses the Daubechies4 wavelet basis function. Through recursive decomposition, the acoustic emission signal is subdivided into 32 sub-bands in the frequency domain. Each sub-band corresponds to the energy distribution of a different frequency range. The ratio of the energy of each sub-band to the total energy is calculated to form a 32-dimensional energy ratio feature vector. The energy ratio feature vector can effectively distinguish the acoustic emission spectrum characteristics of different damage modes, providing input data for the subsequent fatigue damage identification model.

[0063] The specific implementation of step S2 involves using the energy ratio feature vector obtained in step S1 as input to the fatigue damage identification model. The fatigue damage identification model first extracts local features from the energy ratio feature vector using a one-dimensional convolutional layer in the feature extraction module. The convolutional kernel size is set to 3, extracting the combination patterns of adjacent frequency band energies. Then, the extracted local features are input into the time-series modeling module. This module uses a long short-term memory network to establish temporal dependencies on acoustic emission signals within a continuous time window, capturing the temporal characteristics of the damage evolution process. Finally, the fully connected layer of the pattern classification module outputs the probability distributions of three damage modes: microcrack initiation, crack propagation, and wire breakage. The damage mode with the highest probability is selected as the damage mode category label, and the highest probability value is output as the damage confidence score. The discrimination model achieves hierarchical decision-making through a three-layer classifier structure. The first layer, a coarse classifier, sets a threshold of 0.3 to determine the presence of damage; the second layer, a medium classifier, sets a threshold of 0.5 to distinguish damage types; and the third layer, a fine classifier, sets a threshold of 0.7 to identify specific damage modes. The damage confidence score reflects the model's certainty regarding the identification results, providing a basis for adjusting the phase field evolution rate parameter in step S6.

[0064] The specific implementation of step S3 involves establishing a finite element model that includes the geometric structure, material properties, and boundary conditions of the photovoltaic support cable. The electrochemical corrosion kinetics equation, the ultraviolet aging degradation kinetics equation, and the thermo-coupling constitutive relation are embedded into the finite element solver. A sequential coupling algorithm is used to solve the model in the order of temperature field, humidity field, corrosion field, and irradiation field. First, the heat conduction equation is solved based on historical ambient temperature data to obtain the temperature field distribution. The temperature field boundary conditions are set as a periodic function of ambient temperature over time. Then, the temperature field distribution is used as the input condition for solving the humidity field. The humidity diffusion coefficient is expressed using the Arrhenius equation to represent its temperature dependence. The humidity diffusion equation is solved to obtain the humidity field distribution. Next, the temperature field distribution and humidity field distribution are substituted into the electrochemical corrosion kinetics equation. The corrosion rate has an exponential relationship with temperature and a linear relationship with humidity, and the corrosion depth distribution is calculated. Finally, the irradiation field distribution is solved based on irradiation intensity data. The irradiation field only affects the sheath material and not the internal steel wires. This multi-physics coupled finite element model can accurately describe the multi-field synergistic process of the photovoltaic support cable in the real service environment.

[0065] The specific implementation of step S4 involves extracting the values ​​of the temperature field distribution, humidity field distribution, and irradiance field distribution calculated in step S3 at each node of the photovoltaic support cable. Based on the material degradation constitutive relationship, the elastic modulus degradation coefficient, yield strength degradation coefficient, and fracture toughness degradation coefficient of each node are calculated. The elastic modulus degradation coefficient decreases linearly with increasing temperature, decreasing by 2% for every 10°C increase in temperature. The yield strength degradation coefficient is affected by the corrosion depth, decreasing by 5% for every 0.1mm increase in corrosion depth. The fracture toughness degradation coefficient is affected by the cumulative ultraviolet radiation dose, decreasing by 5% for every 0.1mm increase in cumulative radiation dose. The fracture toughness decreased by 3%. Then, a fine mesh sub-model was established in the anchorage area of ​​the photovoltaic support cable. The mesh size of the sub-model was set to 0.5 mm, and the mesh size of the coarse mesh model was set to 5 mm. The boundary displacement field of the anchorage area calculated by the coarse mesh model was used as the displacement boundary condition of the sub-model. The stress distribution was recalculated in the sub-model to obtain the stress concentration factor and stress gradient of the anchorage area. The material degradation coefficient was used to update the material parameters of the photovoltaic support cable fatigue crack propagation simulation model in step S5.

[0066] The specific implementation of step S5 involves introducing continuous phase field variables to characterize the crack state in the fatigue crack propagation simulation model of photovoltaic support cables. The phase field evolution equation describes the change of continuous phase field variables over time. The driving force for evolution originates from the difference between the elastic strain energy release rate and the fracture energy dissipation rate. Crack propagation occurs when the driving force is positive, and crack cessation occurs when the driving force is negative. The elastic modulus degradation coefficient, yield strength degradation coefficient, and fracture toughness degradation coefficient obtained in step S4 are substituted into the phase field evolution equation to update the material parameters. A Gaussian random field is used to describe the spatial distribution of the photovoltaic support cable steel wire material properties. The correlation length of the random field is set to 100 μm, and the variance is set to 10% of the average value of the material parameters. A random material parameter field satisfying a Gaussian distribution is generated using a spectral representation method. An adaptive time step algorithm is used to solve the phase field evolution equation, with the initial time step set to... When the rate of change of the continuous phase field variable is greater than 0.01, the time step is halved; when the rate of change of the continuous phase field variable is less than 0.001, the time step is doubled. The iterative calculation continues until the phase field evolution reaches a steady state. The contour lines where the continuous phase field variable is equal to 0.5 are extracted as crack propagation paths. The crack propagation paths reflect the random bifurcation characteristics of cracks under the influence of the spatial variability of material properties.

[0067] The specific implementation of step S6 involves reading the damage confidence value output in step S2, adjusting the phase field evolution rate parameter according to the interval to which the damage confidence value belongs, and controlling the coefficient of the time derivative term in the phase field evolution equation. When the damage confidence value belongs to [0.8, 1.0], it indicates that the model has a high degree of certainty in damage identification. At this time, the phase field evolution rate parameter is adjusted from the baseline value. Increase to To accelerate the crack propagation simulation and reflect the rapid propagation trend of real cracks, when the damage confidence level is within [0.5, 0.8), it indicates that the model has moderate certainty in damage identification. In this case, the phase field evolution rate parameter is kept as the baseline value. When the damage confidence level is less than 0.5, it indicates that the model is uncertain about damage identification or has not detected obvious damage. In this case, the phase field evolution rate parameter should be reduced to [value missing]. To reduce the crack propagation simulation rate and avoid over-prediction, the phase field evolution rate parameter adjustment mechanism achieves closed-loop feedback between damage monitoring information and crack propagation simulation.

[0068] The specific implementation of step S7 involves establishing a topology optimization model using the variable density method. The anchorage region of the photovoltaic support cable is discretized into a finite element mesh, and each element is assigned a material density design variable. The material density ranges from 0 to 1. The objective function is defined as minimizing the stress concentration factor, with the constraint that the material volume fraction does not exceed 80% of the initial volume. The adjoint sensitivity analysis method is used to calculate the gradient of the objective function with respect to the material density of each element. The right-hand side of the adjoint equation is the partial derivative of the objective function with respect to the stress field. Solving the adjoint equation yields the adjoint variable. Multiplying the adjoint variable by the derivative of the element stiffness matrix with respect to the material density yields the sensitivity. The moving asymptote algorithm is used to update the material density distribution based on sensitivity information. In each iteration, the material density in high-stress areas is increased by 0.05, and the material density in low-stress areas is decreased by 0.05. After 50 iterations, the material density distribution tends to stabilize. The level set method is used to post-process the optimized material density field, extracting contour lines with a material density of 0.5 as material boundaries, eliminating intermediate density elements, and obtaining a clear topological structure. The optimized material distribution is then substituted into the finite element model to recalculate the stress field distribution. The optimized stress field distribution has a more uniform stress distribution and a lower stress concentration factor.

[0069] The specific implementation of step S8 involves extracting the optimized stress field distribution obtained in step S7 and the crack propagation path obtained in step S5. The service load history of the photovoltaic support cable is decomposed into multiple stress cycle segments using the rainflow counting method. Each stress cycle segment has an average stress and stress amplitude. The fatigue life corresponding to each stress cycle segment is determined according to the material fatigue curve. The fatigue curve uses the Basquin formula to represent the relationship between stress amplitude and fatigue life. The damage degree of each stress cycle segment is calculated by dividing the actual number of cycles by the fatigue life. The damage degrees of all stress cycle segments are summed to obtain the cumulative fatigue damage degree. When the cumulative fatigue damage degree is greater than or equal to 0.85, it is determined that the photovoltaic support cable is about to fail due to fatigue, and it is recommended to replace it immediately. When the cumulative fatigue damage degree is within [0.6, 0.85), it is determined that the photovoltaic support cable is in the fatigue damage development stage, and it is recommended to increase the monitoring frequency and formulate a maintenance plan. When the cumulative fatigue damage degree is less than 0.6, it is determined that the photovoltaic support cable is in a safe service state and can be inspected according to the regular cycle. The cumulative fatigue damage degree comprehensively considers the influence of environmental coupling, random crack propagation, and stress field optimization, and can accurately assess the fatigue state of the photovoltaic support cable.

[0070] It should be noted that one of the key technical ideas of this invention is to combine acoustic emission monitoring with machine learning to achieve early identification of microcracks. By extracting the energy ratio feature vector of the acoustic emission signal in the 0 to 50 kHz frequency band through wavelet packet decomposition, different damage modes are adaptively identified using the composable inference architecture based on neural module networks in the fatigue damage identification model. This solves the problem of insufficient resolution of micron-level cracks in conventional non-destructive testing techniques. Compared with traditional ultrasonic testing, which can only detect millimeter-level cracks, acoustic emission monitoring combined with machine learning can provide early warning in the early stage of crack initiation, significantly extending the monitoring window. At the same time, the problem of scarce failure samples in actual engineering is solved by synthesizing failure samples through generative adversarial networks and cost-sensitive learning strategies, so that the model still has reliable identification capabilities under extremely imbalanced sample conditions.

[0071] The second key technical approach is to construct a multi-physics coupled finite element model to accurately describe the environmental synergistic degradation process. By integrating the electrochemical corrosion kinetic equation, the ultraviolet light aging degradation kinetic equation, and the thermo-mechanical coupled constitutive relation, a sequential coupling algorithm is used to calculate the coupled effects of temperature field, humidity field, and irradiation field on material properties. The environmental field distribution is mapped to the material degradation coefficient. Compared with the traditional empirical correction coefficient based on a single environmental factor, the multi-physics coupled model can reflect the interaction and time-varying evolution law between environmental factors, avoid the huge error generated when extrapolating experimental data of a single environmental factor to actual working conditions, and improve the accuracy of fatigue life prediction.

[0072] The third key technical approach is to use stochastic phase-field theory to simulate the stochastic bifurcation and propagation behavior of cracks. By introducing a Gaussian random field to characterize the spatial variability of material properties, the deterministic fracture mechanics model is extended into a stochastic phase-field model capable of capturing stochastic behavior. Compared to traditional Monte Carlo simulations, this approach requires... With sampling more than once, the stochastic phase field method characterizes the crack interface through continuous phase field variables. The adaptive time step algorithm significantly reduces the computational cost while maintaining an accurate description of the random bifurcation path of the crack, thus meeting the needs of rapid engineering evaluation.

[0073] The synergistic effect of the three key technical approaches mentioned above establishes a complete chain from microscopic damage monitoring to macroscopic fatigue life prediction. The damage confidence provided by acoustic emission monitoring is used to adjust the phase field evolution rate parameter, realizing a closed-loop feedback between monitoring information and simulation calculation. The material degradation coefficient calculated by the multiphysics coupling model is used to update the material parameters of the stochastic phase field model, enabling crack propagation simulation to reflect the effects of real environmental degradation. The topology optimization algorithm further improves the accuracy of fatigue life prediction by reconstructing the stress field. The synergistic effect of the three approaches enables the fatigue degree determination method to comprehensively consider multiple factors such as microscopic crack initiation, environmental coupling degradation, stochastic crack propagation, and stress field optimization. Compared with traditional empirical methods based solely on macroscopic crack propagation data, the method of this invention has earlier warning capabilities, more accurate life prediction, and more reliable safety assessment.

[0074] It should be noted that this invention also solves the following technical problem: In the process of fatigue damage identification of photovoltaic support cables, the scarcity of fatigue failure samples in actual engineering projects leads to an imbalance in the training dataset, resulting in low accuracy of traditional machine learning models in identifying minority failure samples. This invention addresses this problem by employing a generative adversarial network (GAN) to synthesize failure samples. The generator learns the probability distribution of failure samples in the energy ratio feature vector space and generates adversarial samples in the boundary region of the probability distribution. The discriminator distinguishes between real failure samples and generated failure samples to improve generation quality. Combined with synthetic nearest neighbor oversampling, new samples are randomly generated on the connection lines between each failure sample and its nearest neighbor in the feature space. Tomk link undersampling is used to remove samples that are too close to failure samples from the normal service samples, achieving a 3:1 ratio of normal service samples to failure samples in the training dataset. Furthermore, a cost-sensitive learning strategy is used to assign five times the weight of normal service samples to failure samples. This addresses the sample imbalance problem from two aspects: data augmentation and loss function optimization, enabling the fatigue damage identification model to effectively identify scarce failure modes.

[0075] Specifically, the principle of this invention is as follows: This invention solves the technical problem of accurately predicting the fatigue damage evolution of photovoltaic support cables under multi-field coupling environments. The principle lies in the ability of acoustic emission technology to capture stress wave signals released by microscopic fracture events within the material in real time. By extracting energy ratio features of different frequency bands through wavelet packet decomposition, the differences in acoustic emission characteristics of different damage modes can be distinguished. The composable reasoning architecture of the neural module network automatically selects the combination scheme of feature extraction, temporal modeling, and pattern classification modules based on the characteristics of the input data, avoiding the problem of insufficient adaptability of fixed network structures to complex acoustic emission signals. The multi-physics coupled finite element model quantifies the degradation effects of temperature, humidity, and irradiation fields on material properties into degradation coefficients of elastic modulus, yield strength, and fracture toughness through a sequential coupling algorithm, reflecting the influence of the actual service environment on fatigue performance. The stochastic phase field theory fuzzifies the discrete crack interface into a continuous phase field variable, introduces a Gaussian random field to characterize the spatial variability of material properties, and makes the crack propagation path simulation more consistent with the real fracture process. The phase field evolution rate parameter is adjusted according to the damage confidence to realize the closed-loop feedback between the monitoring data and the simulation model. The topology optimization algorithm reduces the stress concentration in the anchoring area by reconstructing the stress path, thereby inhibiting the initiation of fatigue cracks from the source of damage.

[0076] The following provides a specific embodiment 1 of the present invention, and the specific implementation of each step in this embodiment 1 is described in detail below.

[0077] The specific implementation of step S1 is as follows: An array of acoustic emission sensors is uniformly arranged axially on the surface of the photovoltaic support cable, with a sensor spacing of 500–800 mm. Acoustic emission signals from the photovoltaic support cable during its service life are collected, with a sampling frequency set to 2 MHz. Wavelet packet decomposition is performed on the acoustic emission signals to extract the energy ratio feature vector in the 0–50 kHz frequency band. Wavelet packet decomposition decomposes the acoustic emission signal into multiple layers according to a binary tree structure. The first layer decomposes the acoustic emission signal into low-frequency approximate components and high-frequency detail components. Subsequent layers further decompose the low-frequency approximate components and high-frequency detail components of the previous layer, until the fifth layer yields wavelet packet coefficients for 32 frequency bands. The energy value of each frequency band is calculated as the sum of the squares of all wavelet packet coefficients within that band. After normalizing the energy values ​​of the 32 frequency bands, an energy ratio feature vector is constructed. The formula for calculating the energy ratio feature vector is as follows:

[0078] .

[0079] In the formula, For the first The energy ratio of each frequency band; For the first The first frequency band Each wavelet packet coefficient is represented in millivolts. For the first The total number of wavelet packet coefficients in each frequency band; This represents the maximum energy value across all frequency bands, expressed in square millivolts. The value range is from 1 to 32; The wavelet packet coefficient index; Frequency band number; For the first The total number of wavelet packet coefficients in each frequency band.

[0080] The specific implementation of step S2 is as follows: The energy ratio feature vector is input into the fatigue damage identification model for processing to identify three damage modes: microcrack initiation, crack propagation, and wire breakage within the photovoltaic support cable, and the damage mode category label and damage confidence score are output. The fatigue damage identification model adopts a composable reasoning architecture based on neural module networks, decomposing the damage pattern identification task into three reusable neural modules: a feature extraction module, a temporal modeling module, and a pattern classification module. The module combination scheme is automatically selected based on the characteristics of the input data using program generation technology, and the module selection strategy is optimized using the policy gradient algorithm in reinforcement learning. The discrimination model adopts a multi-scale classification optimization mechanism based on hierarchical soft maximum values, constructing a three-layer classifier structure. The first layer, a coarse classifier, distinguishes between damaged and undamaged states; the second layer, a middle classifier, distinguishes between crack-type damage and wire-breakage damage; and the third layer, a fine classifier, identifies the specific damage modes of microcrack initiation, crack propagation, and wire breakage. Each classifier uses a soft maximum function to calculate the category probability and outputs an uncertainty quantification index.

[0081] The specific implementation of step S3 is as follows: A multi-physics coupled finite element model of the photovoltaic support cable is established, integrating the electrochemical corrosion kinetic equation, the ultraviolet light aging degradation kinetic equation, and the thermo-coupling constitutive relation. A sequential coupling algorithm is used to calculate the temperature field distribution, humidity field distribution, and irradiation field distribution. The electrochemical corrosion kinetic equation describes the relationship between the corrosion rate of the photovoltaic support cable steel wire and time in an acid rain environment. The corrosion rate calculation formula is expressed as follows:

[0082] .

[0083] In the formula, Corrosion rate, expressed in millimeters per year; The corrosion rate constant is expressed in millimeters per year, with an empirical value of 0.0125 millimeters per year. The activation energy of the corrosion reaction is expressed in joules per mole, and is typically taken as a value of [value missing]. Joules per mole; The gas constant is 8.314 joules per mole per Kelvin. This refers to ambient temperature, expressed in Kelvin. For reference temperature, the value is taken as 298 Kelvin; This refers to the acidity / alkalinity value of acid rain. For reference pH value, a value of 7 is used; This represents the surface potential difference of the steel wire, measured in volts. The reference potential difference is set to 0.5 volts. The mass loss of the surface material of the photovoltaic support cable wire per unit time is calculated using Faraday's law, converting the mass loss into a decrease in the cross-sectional area of ​​the wire. The formula for calculating the mass loss is as follows:

[0084] .

[0085] In the formula, This refers to mass loss, expressed in grams. The molar mass of steel is 56 grams per mole. This is the corrosion current, measured in amperes. Corrosion time, in seconds; The electron transfer number has a value of 2; This is the Faraday constant, with a value of 96,485 coulombs per mole; ρ represents the density of steel, expressed in grams per cubic centimeter, typically taken as 7.85 grams per cubic centimeter. The ultraviolet (UV) light aging degradation kinetic equation describes the performance degradation of the polymer material in the photovoltaic support cable sheath under UV irradiation. The degradation rate calculation formula is as follows:

[0086] .

[0087] In the formula, Degradation rate, expressed in years; The degradation frequency factor is expressed in annually, with an empirical value of 0.036 per year. The activation energy of the degradation reaction is expressed in joules per mole, and is typically taken as a value of [value missing]. Joules per mole; Ultraviolet radiation intensity, expressed in watts per square meter; For reference irradiance, the value is taken as 100 watts per square meter; Cumulative irradiation dose, expressed in joules per square meter; For reference irradiation dose, the value is taken as follows: Joules per square meter; This is the radiation intensity index, typically taken as 1.2; The cumulative dose exponent is typically set to 0.8. The thermo-coupled constitutive relation describes the stress-strain response of the photovoltaic support cable material under cyclic temperature loading. The formula for calculating the stress increment is as follows:

[0088] .

[0089] In the formula, This represents the stress increment, measured in megapascals (MPA). This is the temperature-dependent elastic modulus, measured in gigapascals (GPa), and its relationship with temperature is as follows: ,in This is the temperature sensitivity coefficient, typically taken as 0.0003; The elastic modulus at the reference temperature is taken as 200 gigapascals; For strain increment; The coefficient of thermal expansion is expressed in Kelvin and is typically taken as a value of [value missing]. Every Kelvin; This represents the temperature increment, measured in Kelvin.

[0090] The specific implementation of step S4 is as follows: The temperature field distribution, humidity field distribution, and irradiation field distribution are mapped to the elastic modulus degradation coefficient, yield strength degradation coefficient, and fracture toughness degradation coefficient of the photovoltaic support cable material. The formula for calculating the elastic modulus degradation coefficient is expressed as follows:

[0091] .

[0092] In the formula, It is the coefficient of elastic modulus degradation; The current temperature is expressed in Kelvin. This is the humidity value, expressed as a percentage. This represents the maximum humidity value, which is 100%; This is the maximum temperature value, in Kelvin, typically taken as 350 Kelvin. The maximum cumulative radiation dose, expressed in joules per square meter, is typically taken as a value of [value missing]. Joules per square meter. The formula for calculating the yield strength degradation coefficient is as follows:

[0093] .

[0094] In the formula, The yield strength degradation coefficient; The maximum corrosion rate is expressed in millimeters per year, typically taken as 0.5 millimeters per year. The formula for calculating the fracture toughness degradation coefficient is as follows:

[0095] .

[0096] In the formula, The coefficient for fracture toughness degradation; The maximum degradation rate is expressed in years, typically 0.15 per year. Sub-model techniques are used to refine the local mesh in the photovoltaic support cable anchorage area, establishing a full-scale coarse-mesh model for preliminary stress analysis. The displacement field at the boundary of the photovoltaic support cable anchorage area is extracted as the boundary condition of the sub-model. A fine-mesh sub-model is then established for the photovoltaic support cable anchorage area, with the mesh size being one-tenth that of the coarse-mesh model.

[0097] The specific implementation methods of steps S5 and S7 are the same as those described above, and will not be repeated in detail here.

[0098] The specific implementation of step S6 is as follows: Adjust the phase field evolution rate parameter in the photovoltaic support cable fatigue crack propagation simulation model according to the damage confidence level. When the damage confidence level is between 0.8 and 1.0, increase the phase field evolution rate parameter to 1.5 times the baseline value; when the damage confidence level is between 0.5 and 0.8, maintain the baseline value; and when the damage confidence level is less than 0.5, decrease the phase field evolution rate parameter to 0.6 times the baseline value. The formula for adjusting the phase field evolution rate parameter is as follows:

[0099] .

[0100] In the formula, The adjusted phase field evolution rate parameter is expressed in seconds. This is the baseline value for the phase field evolution rate parameter, measured in seconds, and is typically taken as [value missing]. per second; The damage confidence level.

[0101] The specific implementation of step S8 is as follows: The cumulative fatigue damage of the photovoltaic support cable is calculated based on the optimized stress field distribution and crack propagation path. The cumulative fatigue damage is calculated using linear damage accumulation theory. The service life of the photovoltaic support cable is divided into multiple stress cycle segments. The formula for calculating the cumulative fatigue damage is as follows:

[0102] .

[0103] In the formula, To accumulate fatigue damage; This represents the total number of stress cycle segments; This refers to the stress cycle segment number; For the first The number of cycles in each stress cycle segment; For the first Each stress cycle segment corresponds to the fatigue life under the stress amplitude. When the cumulative fatigue damage degree is greater than or equal to 0.85, the photovoltaic support cable is determined to be in the critical state of fatigue failure. When the cumulative fatigue damage degree is between 0.6 and 0.85, the photovoltaic support cable is determined to be in the state of fatigue damage development. When the cumulative fatigue damage degree is less than 0.6, the photovoltaic support cable is determined to be in a safe service state.

[0104] The specific implementation method of the explanation section is as follows: Feature complexity is quantified by calculating the information entropy of the energy-to-feature vector. The larger the information entropy, the more dispersed the feature distribution, and the higher the feature complexity. The formula for calculating feature complexity is expressed as follows:

[0105] .

[0106] In the formula, For feature complexity; For the first The energy ratio of each frequency band. The historical batch model prediction accuracy is the ratio of the number of correctly predicted samples in the validation set over the most recent 50 batches to the total number of samples. The formula for calculating the historical batch model prediction accuracy is as follows:

[0107] .

[0108] In the formula, The accuracy of the historical batch model prediction; Batch number; For the first The number of correctly predicted samples in each batch; For the first The total number of samples in each batch. The rate of change of validation set loss is the difference between the validation set loss of the current batch and the validation set loss of the previous batch, divided by the validation set loss of the previous batch. The formula for calculating the rate of change of validation set loss is as follows:

[0109] .

[0110] In the formula, To verify the rate of change of the set loss; The loss is the validation set loss for the current batch. This represents the loss from the previous batch of validation sets. The formula for calculating the module activation modifier is as follows:

[0111] .

[0112] In the formula, Activate the adjustment value for the module; This represents the maximum value of the feature complexity during the training process; To verify the maximum value of the absolute value of the rate of change of the set loss during the training process; To verify the absolute value of the rate of change of the set loss.

[0113] Additional explanation of location: In step S3, the location of the... The complete expression is supplemented in After its first appearance, that is, in " This is the temperature-dependent elastic modulus, in gigapascals (GPa). The relationship between this modulus and temperature is as follows: ,in This is the temperature sensitivity coefficient, typically taken as 0.0003; "The elastic modulus at the reference temperature is taken as 200 gigapascals." In step S4, and The explanation is supplemented after its first appearance in the corresponding formula. In the explanation section, it will... The explanation is supplemented in the formula for calculating the module activation adjustment value.

[0114] To better understand and implement this invention, the following is a specific application scenario of this invention, Example 2:

[0115] To verify the effectiveness of this invention, technicians set up a test environment and conducted a fatigue assessment on the cable system of a photovoltaic power station that had been in service for three years. The photovoltaic power station has 156 support cables. Technicians selected the 12 cables with the highest stress as monitoring targets. Eight acoustic emission sensors were evenly arranged axially on the surface of each cable, with a sensor spacing of 450 mm. The sampling frequency was set to 500 kHz, and acoustic emission signal data was continuously collected for 30 days. The collected acoustic emission signals were decomposed using wavelet packet decomposition. A five-level decomposition was performed using the Db4 wavelet basis function, resulting in wavelet packet coefficients for 32 frequency bands. The energy values ​​of each frequency band were calculated and normalized to form an energy ratio eigenvector, as shown in Table 1.

[0116] Table 1. Distribution of Eigenvectors of Acoustic Emission Signal Energy Ratio at Typical Moments

[0117]

[0118] Technicians input the energy ratio feature vector into a pre-trained fatigue damage recognition model for processing. This model employs a composable inference architecture based on neural network modules, including 16 one-dimensional convolutional kernels in the feature extraction module, 128 long short-term memory units in the temporal modeling module, and 3 fully connected layers in the pattern classification module. The model output showed that 3 out of 12 cables exhibited crack propagation damage patterns, with damage confidence scores of 0.87, 0.76, and 0.62, respectively. The remaining 9 cables were in normal service condition. Based on the current batch data, the technicians calculated the feature complexity to be 1.68. The model prediction accuracy over 50 historical batches was 0.91, and the validation set loss rate was -0.03. Substituting these values ​​into the module activation adjustment value calculation formula yielded a module activation adjustment value of 0.68. A standard module combination scheme was then used for subsequent analysis.

[0119] like Figure 2 As shown, technicians established a multiphysics coupled finite element model of the photovoltaic support cable. The total cable length is 6800 mm, the wire diameter is 5 mm, and the sheath thickness is 2.5 mm. The temperature field analysis considered diurnal temperature variation and seasonal changes, setting the ambient temperature to fluctuate between -15℃ and 65℃, and the peak solar irradiance to be 1000 ppm. In the humidity field analysis, the relative humidity was set to vary between 40% and 95%, with the surface humidity reaching saturation during rainy weather. In the electrochemical corrosion kinetic equation, the acid rain pH was set to 4.2, and the surface potential difference of the steel wire was -0.35V. In the ultraviolet light aging degradation kinetic equation, the cumulative irradiation dose reached [amount missing] after 3 years of service. The elastic modulus of the sheath polymer material degraded from the initial 850 MPa to 612 MPa. The coefficient of thermal expansion of the steel wire is considered in the thermo-coupling constitutive relation. The temperature dependence coefficient of the elastic modulus is negative. The sequential coupling algorithm was used for calculation. The temperature field distribution results showed that the temperature in the middle of the cable reached the highest of 62℃, while the temperature in the anchorage area remained at 48℃ due to faster heat dissipation. The humidity field distribution results showed that the peak of the surface humidity concentration field was located on the lower surface of the cable. The corrosion depth distribution showed that the maximum corrosion depth in the anchorage area reached 0.18mm.

[0120] Technicians mapped the temperature, humidity, and irradiation field distributions to material property degradation coefficients. The elastic modulus degradation coefficient decreased from 0.94 in the anchorage region to 0.72 in the central region, the yield strength degradation coefficient decreased from 0.91 to 0.68, and the fracture toughness degradation coefficient decreased from 0.88 to 0.65. Sub-model technology was used to refine the local mesh in the anchorage region. The full-scale coarse mesh model had a mesh size of 20 mm, while the fine mesh sub-model had a mesh size of 2 mm. The boundary displacement field was extracted as the boundary condition for the sub-model. The maximum stress concentration factor in the anchorage region was calculated to be 3.47 in the fine mesh. Figure 3 As shown, technicians established a fatigue crack propagation simulation model based on stochastic phase-field theory. The crack interface was fuzzified into continuous phase-field variables, and a Gaussian random field was introduced to characterize the spatial variability of the steel wire material properties. The coefficient of variation for the elastic modulus random field was set to 0.15, the coefficient of variation for the yield strength random field was set to 0.12, and the correlation length of the covariance function was set to 8 mm. An adaptive time-step algorithm was used to calculate the crack propagation path. The initial time step was set to 0.001 s. When the rate of change of the phase-field variables was greater than 0.05, the time step was reduced to 0.0003 s; when the rate of change of the phase-field variables was less than 0.01, the time step was increased to 0.005 s.

[0121] Technicians adjusted the phase field evolution rate parameters based on the damage confidence level. For cables with a damage confidence level of 0.87, the phase field evolution rate parameters were adjusted from the baseline value. Increase to For the baseline value of the cable-maintaining phase field evolution rate parameter with a damage confidence level of 0.76, and for the baseline value of the cable-maintaining phase field evolution rate parameter with a damage confidence level of 0.62, technicians used a topology optimization-driven stress path reconstruction algorithm to optimize the material distribution in the anchorage region. The initial value of the material density design variable was set to 0.5. The gradient of the objective function was calculated through adjoint sensitivity analysis. After 50 iterations using the moving asymptote algorithm, the material density distribution converged, and the stress concentration factor decreased from 3.47 before optimization to 2.18, significantly improving the homogenization of the stress field. Based on the optimized stress field distribution and crack propagation path, the cumulative fatigue damage was calculated. The 3-year service life was divided into 156 stress cycle segments, each containing a 7-day load history. The stress amplitude in each cycle segment varied between 85 MPa and 245 MPa, corresponding to a fatigue life of... Next Between these values, the cumulative fatigue damage of the cable with a damage confidence level of 0.87 was calculated to be 0.79, which is considered a state of fatigue damage development. The cumulative fatigue damage of the cable with a damage confidence level of 0.76 was 0.68, which is also considered a state of fatigue damage development. The cumulative fatigue damage of the cable with a damage confidence level of 0.62 was 0.54, which is considered a state of safe service.

[0122] This invention extracts multi-band energy features from acoustic emission signals through wavelet packet decomposition. Compared to traditional single-band analysis, it can more comprehensively capture the acoustic emission characteristics during crack initiation and propagation. The composable reasoning architecture based on neural network modules can adaptively adjust the module combination scheme according to the complexity of the input data, avoiding the limitations of a fixed network structure. A hierarchical soft-maximum multi-scale classification mechanism achieves progressive identification from coarse-grained to fine-grained and provides uncertainty quantification indicators for reliable input in subsequent analysis. The multi-physics coupled finite element model integrates multiple degradation mechanisms such as electrochemical corrosion, UV aging, and thermo-mechanical coupling. Compared to traditional single mechanical analysis, it can realistically reflect the performance evolution of photovoltaic support cables under complex environments. The sequential coupling algorithm progressively transmits the influence of temperature, humidity, and irradiation fields to material parameters, avoiding the numerical difficulties of direct coupling solutions. Random phase field theory, by introducing Gaussian random fields to characterize the spatial variability of material properties, can reflect the influence of material inhomogeneity on crack paths in actual engineering compared to deterministic crack propagation models. The adaptive time step algorithm dynamically adjusts the calculation accuracy and efficiency according to the rate of change of phase field variables, achieving accurate capture of the rapid propagation stage. The topology-driven stress path reconstruction algorithm achieves stress field homogenization by adjusting the material distribution in the anchoring region. Compared with traditional uniform material design, it can effectively reduce stress concentration and avoid premature local failure. The accompanying sensitivity analysis method efficiently calculates the gradient of design variables, accelerating the optimization convergence process.

[0123] It should be noted that the variables involved in this invention are explained in detail in Tables 2 and 3.

[0124] Table 2. Variable Explanation Table (Part 1)

[0125]

[0126] Table 3. Variable Explanation Table (Part Two)

[0127]

[0128] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for determining the fatigue degree of photovoltaic support cables, characterized in that, An array of acoustic emission sensors is uniformly arranged axially on the surface of the photovoltaic (PV) support cable to collect acoustic emission signals during its service life. Wavelet packet decomposition is performed on the acoustic emission signals to extract the energy ratio feature vector in the 0-50kHz frequency band. The energy ratio feature vector is input into a fatigue damage identification model to identify three damage modes: microcrack initiation, crack propagation, and wire fracture within the PV support cable. Damage mode category labels and damage confidence scores are output. A multi-physics coupled finite element model of the PV support cable is established, integrating electrochemical corrosion kinetics, ultraviolet aging degradation kinetics, and thermo-coupling constitutive relations. A sequential coupling algorithm is used to calculate the temperature field distribution, humidity field distribution, and irradiation field distribution, mapping these distributions to the elastic modulus of the PV support cable material. The degradation coefficients of quantity, yield strength, and fracture toughness were calculated, and the local mesh of the anchorage area of ​​the photovoltaic support cable was refined using sub-model technology. A fatigue crack propagation simulation model of the photovoltaic support cable was established based on random phase field theory. The crack interface was fuzzified into continuous phase field variables, and a Gaussian random field was introduced to characterize the spatial variability of the photovoltaic support cable steel wire material properties. An adaptive time step algorithm was used to calculate the crack propagation path. The phase field evolution rate parameter in the photovoltaic support cable fatigue crack propagation simulation model was adjusted according to the damage confidence. A topology optimization-driven stress path reconstruction algorithm was used to optimize the material distribution in the anchorage area of ​​the photovoltaic support cable. Based on the optimized stress field distribution and crack propagation path, the cumulative fatigue damage degree of the photovoltaic support cable was calculated, and the fatigue state of the photovoltaic support cable was determined.

2. The method for determining the fatigue degree of photovoltaic support cables according to claim 1, characterized in that, The steps of wavelet packet decomposition are as follows: the acoustic emission signal is decomposed into multiple layers according to the binary tree structure up to the fifth layer to obtain wavelet packet coefficients of 32 frequency bands. The energy value of each frequency band is calculated as the sum of the squares of all wavelet packet coefficients in the frequency band. After normalizing the energy values ​​of the 32 frequency bands, an energy ratio feature vector is formed.

3. The method for determining the fatigue degree of photovoltaic support cables according to claim 2, characterized in that, The generative model of the fatigue injury recognition model adopts a composable reasoning architecture based on neural module networks. It decomposes the injury pattern recognition task into three reusable neural modules: feature extraction module, temporal modeling module, and pattern classification module. The module combination scheme is automatically selected based on the characteristics of the input data through program generation technology.

4. The method for determining the fatigue degree of photovoltaic support cables according to claim 3, characterized in that, The discrimination model of the fatigue damage identification model adopts a multi-scale classification optimization mechanism based on hierarchical soft maximum value. A three-layer classifier structure is constructed to achieve progressive classification from coarse-grained to fine-grained. Each classifier uses a soft maximum value function to calculate the class probability and output the uncertainty quantification index.

5. The method for determining the fatigue degree of photovoltaic support cables according to claim 4, characterized in that, The training dataset for the fatigue damage identification model is established by using a generative adversarial network (GAN) to synthesize failure samples. The generator of the GAN learns the probability distribution of failure samples in the energy-ratio feature vector space, and the discriminator of the GAN distinguishes between real failure samples and generated failure samples.

6. The method for determining the fatigue degree of photovoltaic support cables according to claim 5, characterized in that, The training dataset is established by combining synthetic nearest neighbor oversampling technology to oversample failed samples. New samples are randomly generated on the connection between each failed sample and its nearest neighbor in the feature space. Tomk link undersampling technology is used to delete samples in normal service that are too close to failed samples.

7. The method for determining the fatigue degree of photovoltaic support cables according to claim 6, characterized in that, The fatigue damage identification model training adopts a cost-sensitive learning strategy to assign 5 times the weight of normal service samples to failed samples. The loss function is defined as the product of cross-entropy loss and weight coefficients. An adaptive moment estimation optimization algorithm is used to update the neural module network parameters.

8. The method for determining the fatigue degree of photovoltaic support cables according to claim 7, characterized in that, The module activation adjustment function is used to adjust the module activation weights of the composable inference architecture based on neural module networks. The module activation adjustment function calculates the module activation adjustment value based on three data: the feature complexity of the current batch input data, the prediction accuracy of the model in the historical batches, and the rate of change of the validation set loss.

9. The method for determining the fatigue degree of photovoltaic support cables according to claim 8, characterized in that, The electrochemical corrosion kinetic equation describes the relationship between the corrosion rate and time of the photovoltaic support cable steel wire in an acid rain environment. The corrosion rate is related to acid rain. The value, temperature, and surface potential difference of the photovoltaic support cable steel wire are related. The mass loss of the surface material of the photovoltaic support cable steel wire per unit time is calculated by using Faraday's law.

10. The method for determining the fatigue degree of photovoltaic support cables according to claim 9, characterized in that, The kinetic equation for ultraviolet (UV) aging degradation describes the performance degradation of polymer materials for photovoltaic support cable sheaths under UV irradiation. The degradation rate is related to UV irradiation intensity, cumulative irradiation dose, and material temperature. An exponential relationship between degradation rate and material temperature is established using the Arrhenius equation.

Images