Method, system and device for diagnosing structural state of offshore photovoltaic support and medium

Abstract

The invention relates to the technical field of civil engineering structure health monitoring, in particular to an offshore photovoltaic support structure state diagnosis method, system and device and a medium. The offshore photovoltaic support structure state diagnosis method comprises the steps of generating acceleration response data of healthy and damaged working conditions based on a finite element model, and performing time-frequency transformation on acceleration signal data to construct time-frequency matrix data; calculating an auto-power spectrum and a cross-power spectrum of each acceleration response data, and calculating coherence matrix data based on time-frequency transformation through a two-dimensional convolution function; performing damage positioning by comparing a part with maximum time-frequency coordinate change in coherence matrix data before and after damage, and calculating norms of relative change of characteristic values under a healthy working condition and a damaged working condition so as to quantify the damage degree; and a discrete wavelet neural operator is adopted to establish a mapping model of acceleration response data and damage positioning and damage quantification. According to the method, the accuracy and efficiency of health diagnosis of the offshore photovoltaic support are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of civil engineering structure health monitoring technology, and in particular to a method, system, device and computer-readable storage medium for diagnosing the structural status of an offshore photovoltaic support. Background Art

[0002] With the widespread adoption of offshore photovoltaic power generation systems, maintaining the safety and stability of photovoltaic mounts has become a critical issue. However, civil engineering structures such as offshore photovoltaic mounts are becoming increasingly complex and large-scale. During their service life, they are affected by multiple factors such as external loads, material aging, and environmental corrosion. Traditional detection methods (such as manual inspections and single-sensor monitoring) are inefficient, rely on experience, and have poor noise immunity, making it difficult to detect structural damage in complex environments in real time.

[0003] In existing technologies, although methods based on wavelet transform or modal parameters can partially identify damage, they have the following problems: 1. Unable to effectively process the time-frequency dynamic characteristics of non-stationary vibration signals; 2. Lack of sensitivity to minor injuries; 3. Lack of quantitative analysis capability of damage extent.

[0004] Comparative documents (such as CN104458173B) propose damage indicators based on the rate of change of wavelet coefficients. However, these rely on single frequency domain features and lack deep learning methods, resulting in limited generalization capabilities. Therefore, an efficient diagnostic method that integrates signal processing and artificial intelligence is urgently needed. Summary of the Invention

[0005] Embodiments of the present invention provide a method, system, device and storage medium for diagnosing the structural status of an offshore photovoltaic support to address the problems of insufficient extraction of non-stationary signal features, low recognition rate of minor damage and lack of quantitative damage analysis in traditional methods.

[0006] In order to achieve the above object, on the one hand, a method for diagnosing the structural status of an offshore photovoltaic support is provided, comprising the following steps: Generate acceleration response data of healthy and damaged working conditions based on the finite element model, and perform time-frequency transformation on the acceleration response data to construct time-frequency matrix data; Calculating the auto-power spectrum and cross-power spectrum data of each acceleration response data according to the time-frequency matrix data; Calculating coherence matrix data of each acceleration response data by a two-dimensional convolution function according to the auto-power spectrum and cross-power spectrum data; Calculate the coherence matrix data of each acceleration response data under healthy conditions, calculate the coherence matrix data of each acceleration response data under damaged conditions, and determine the maximum area of the time-frequency coordinate change as the damage location by comparing the coherence matrix data before and after damage. Calculate the norm of the relative change of the coherence matrix data under damaged conditions, and the norm is used to quantify the damage degree; According to the time-frequency coordinate change values of the coherence matrix data before and after damage and the norm, use a predetermined discrete wavelet neural operator to establish mapping models of the acceleration response data with damage location and damage quantification respectively. By inputting the measured acceleration response data into the mapping models, obtain the damage location and damage quantification data.

[0007] In some embodiments, after the step of using a discrete wavelet neural operator to establish mapping models of the acceleration response data with damage location and damage quantification respectively according to the time-frequency coordinate change values of the coherence matrix data before and after damage and the norm, and inputting the acceleration response data through the mapping models to obtain the damage location and damage quantification data, it further includes: Fine-tune the model parameters based on experimental data to verify the damage location and quantification results.

[0008] In some embodiments, in the step of generating the acceleration response data of healthy and damaged conditions based on a finite element model and performing time-frequency transformation on the acceleration response data to construct time-frequency matrix data, it includes: Simulate the damaged condition by randomly reducing the node stiffness or removing elements; Simulate the real environmental load by applying random white noise to the finite element model; Wherein, the damaged conditions include local bolt loosening, beam end cracks, column bottom corrosion, support loosening and / or support fracture.

[0009] In some embodiments, the time-frequency transformation includes S transformation, wavelet transformation and short-time Fourier transformation; Wherein, the expression of the S transformation is:

[0010] In the formula, is the acceleration response signal to be analyzed, is the Gaussian window function with a window width of in the time domain, is the complex weight term, t is the time, is the frequency, is the time center.

[0011] In some embodiments, in the step of calculating the coherence matrix data of each acceleration response data through a two-dimensional convolution function according to the auto-power spectrum and cross-power spectrum data, it includes:

[0012] Among them, is the time operator of the two-dimensional convolution, is the scale operator formula of the two-dimensional convolution, where is the convolution operator, is the acceleration response data is the average unbiased estimate of the auto-power spectrum of the time-frequency matrix of ; is the acceleration response data j is the average unbiased estimate of the auto-power spectrum of the time-frequency matrix of ; is the acceleration response data is the cross-power spectrum of the time-frequency matrix of the acceleration response data j with respect to the time-frequency matrix of the acceleration response data is the average unbiased estimate of is the acceleration response data j is the cross-power spectrum of the time-frequency matrix of the acceleration response data with respect to the time-frequency matrix of the acceleration response data is the average unbiased estimate of is the acceleration response data is the auto-power spectrum of the time-frequency matrix of in the time domain and frequency domain translation operations, is the acceleration response data j is the auto-power spectrum of the time-frequency matrix of in the time domain and frequency domain translation operations, is the acceleration response data is the cross-power spectrum of the time-frequency matrix of the acceleration response data j with respect to the time-frequency matrix of the acceleration response data in the time domain and frequency domain translation operations, is the acceleration response data j is the cross-power spectrum of the time-frequency matrix of the acceleration response data with respect to the time-frequency matrix of the acceleration response data in the time domain and frequency domain translation operations, is the two-dimensional convolution mask, and the definition formula is as follows:

[0013] where n is an integer, and 2 ≤ n ≤ 5; The coherence matrix data is as follows: .

[0014] where is the coherence matrix data, is the acceleration response data The time-frequency matrix of the acceleration response data j The cross-power spectrum of the time-frequency matrix of the acceleration response data The square of the absolute value of the mean unbiased estimate of is the acceleration response data The auto-power spectrum of the time-frequency matrix of the acceleration response data The mean unbiased estimate of is the acceleration response data j The auto-power spectrum of the time-frequency matrix of the acceleration response data The mean unbiased estimate of

[0015] In some embodiments, the norm includes the Frobenius norm, the spectral norm, and the nuclear norm.

[0016] In some embodiments, in the steps of calculating the coherence matrix data of each acceleration response data under healthy conditions, calculating the coherence matrix data of each acceleration response data under damaged conditions, determining the maximum position of the time-frequency coordinate change as damage localization by comparing the coherence matrix data before and after damage, and calculating the norm of the relative change of the coherence matrix data under damaged conditions, where the norm is used to quantify the degree of damage, it includes: Calculating the average value of the coherence matrix data of the acceleration response data under healthy conditions, and using the obtained average coherence matrix data as the localization reference, where the subscript 0 represents the healthy condition; Performing eigenvalue decomposition on the average coherence matrix data through the following formula:

[0017] In the formula, is the coherence matrix data, is The matrix composed of the eigenvectors corresponding to the respective eigenvalues in is The transpose matrix of is The eigenvalue matrix obtained after eigenvalue decomposition; The eigenvalues obtained by decomposition are used as the quantitative reference; By calculating the coherence matrix data under each damaged condition , where the subscript d represents the damaged condition, and damage localization is performed through the following formula:

[0018] In the formula, represents the difference matrix of the coherence matrix before and after damage, is used as the positioning reference, is the coherence matrix data under the damage condition; Perform eigenvalue decomposition on the coherence matrix data under each damage condition to obtain the corresponding eigenvalues, and use the relative change values of the eigenvalues under the healthy condition and the damage condition to represent the overall damage degree of the structure, as shown in the following formula:

[0019] where, is the Frobenius norm, that is, the square root of the sum of the squares of all time-frequency coordinates in the matrix. DI represents the overall damage degree of the structure, and 0 ≤ DI ≤ 1. The closer DI is to 1, the greater the damage degree. and are the eigenvalue matrix of the quantitative reference and the damage condition respectively.

[0020] In some embodiments, the steps of establishing a mapping model between the acceleration response and the damage area using a discrete wavelet neural operator include: Input to the discrete wavelet neural operator, is the acceleration response data, t is the time; Use the local transformation to increase the dimension of the input to obtain , is constructed by a convolutional neural network; Input to a series of wavelet kernel integration layers for multi-layer wavelet decomposition: parameterize the wavelet kernel to obtain horizontal, vertical, and diagonal coefficients at different levels to complete the discrete wavelet decomposition , is 's kernel function, is the convolution operator; Reconstruct the convolutional input , is the convolution operator; Use the convolutional neural network W to perform a linear transformation: , is the convolution operator; Input and add them and activate, and reduce the dimension through the local transformation to complete the and the corresponding and the mapping between DI, and obtain the damage location mapping model and the damage quantification mapping model respectively.

[0021] On the other hand, a diagnostic system for the structural state of an offshore photovoltaic support is provided, including: A time-frequency matrix module, configured to generate acceleration response data for healthy and damaged conditions based on a finite element model, and perform time-frequency transformation on the acceleration response data to construct time-frequency matrix data; A power spectrum module, configured to calculate the auto-power spectrum and cross-power spectrum data of each acceleration response data according to the time-frequency matrix data; A coherence data module, configured to calculate the coherence matrix data of each acceleration response data through a two-dimensional convolution function according to the auto-power spectrum and cross-power spectrum data; A damage location and quantification module, configured to calculate the coherence matrix data of each acceleration response data under healthy conditions, calculate the coherence matrix data of each acceleration response data under damaged conditions, determine the location with the largest change in time-frequency coordinates as the damage location by comparing the coherence matrix data before and after damage, and calculate the norm of the relative change in the coherence matrix data under damaged conditions, where the norm is used to quantify the degree of damage; A discrete wavelet module, configured to establish mapping models of acceleration response data with damage location and damage quantification respectively by using discrete wavelet neural operators according to the time-frequency coordinate change values and the norm of the coherence matrix data before and after damage, and input the acceleration response data through the mapping models to obtain damage location and damage quantification data.

[0022] On the other hand, a diagnostic device for the structural state of an offshore photovoltaic support is provided, including a memory and a processor. The memory stores at least one segment of program, and the at least one segment of program is executed by the processor to implement the above-mentioned diagnostic method for the structural state of an offshore photovoltaic support.

[0023] On the other hand, a computer-readable storage medium is provided, in which at least one segment of program is stored, and the at least one segment of program is executed by the processor to implement the above-mentioned diagnostic method for the structural state of an offshore photovoltaic support.

[0024] The above technical solution has the following technical effects: This application utilizes the adaptive time-frequency resolution of time-frequency transformation to accurately capture the local mutation characteristics of non-stationary vibration signals, and can detect minor stiffness damages (such as bolt loosening, crack initiation). Compared with traditional wavelet methods, the sensitivity is significantly improved; based on the norm change of the eigenvalues of the coherence matrix data, the degree of damage is quantified, solving the problem that traditional methods can only make qualitative judgments; through discrete wavelet neural operators, end-to-end learning of the mapping relationship between time-frequency features and damage is realized, without the need for artificial design of indicators, and it is applicable to multi-type damage diagnosis in complex environments (such as sea waves, wind loads), reducing the maintenance cost to a certain extent; this application significantly improves the accuracy and efficiency of the health diagnosis of offshore photovoltaic supports. Description of the Drawings

[0025] Figure 1 It is a schematic flow chart of the diagnostic method for the structural state of an offshore photovoltaic support in an embodiment of the present invention; Figure 2 It is a schematic diagram of a finite element model in an embodiment of the present invention; Figure 3 It is a schematic flow chart of the method for establishing a mapping model between acceleration response and damage area using a discrete wavelet neural operator in another embodiment of the present invention; Figure 4 It is a schematic structural diagram of the diagnostic system for the structural state of an offshore photovoltaic support in yet another embodiment of the present invention. Detailed implementation manners

[0026] To further illustrate each embodiment, the present invention provides accompanying drawings. These drawings are part of the disclosure of the present invention, mainly used to illustrate the embodiments, and can be combined with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these contents, those of ordinary skill in the art should be able to understand other possible implementation manners and the advantages of the present invention. The components in the figures are not drawn to scale, and similar component symbols are usually used to represent similar components.

[0027] The structural diagnosis of offshore photovoltaic supports is of crucial importance, mainly because of their special offshore application scenarios and severe environmental challenges. The offshore environment has characteristics such as high humidity, strong salt spray corrosion, continuous wave impact, strong wind loads, and tidal changes. These factors will accelerate the accumulation of damages such as fatigue of support materials, bolt loosening, weld cracking, and corrosion of support structures. If not detected in time, it may lead to local failure or even overall collapse, not only causing a decrease in power generation efficiency and economic losses, but also potentially causing environmental pollution or safety accidents. Therefore, structural health monitoring is the core requirement for ensuring the long-term stable operation of offshore photovoltaic systems.

[0028] The randomness of waves and wind loads results in non-stationary characteristics of vibration signals, making it difficult for traditional time-domain or frequency-domain methods to capture dynamic changes; at the same time, the interference of ocean environmental noise (such as wave slapping, equipment vibration) is significant. How to effectively suppress high-frequency noise and improve the signal-to-noise ratio are all difficult problems in the structural diagnosis of offshore photovoltaic supports. This application combines time-frequency analysis and deep learning, not only providing a high-precision and anti-interference damage diagnosis solution for offshore photovoltaic supports, but also having the potential for cross-scenario migration in its technical framework. By simply optimizing parameters according to the load characteristics, damage types, and noise environment of specific scenarios, it can be widely applied to the intelligent health monitoring of onshore infrastructure, aerospace, and industrial equipment.

[0029] The technical solution of this application, which includes time-frequency analysis, coherence eigenvalue decomposition, and discrete wavelet neural operator (DWNO), has universality and can be extended to other scenarios, such as: Wind turbine tower: The tower is affected by wind loads and mechanical vibrations. This application can detect bolt pre-tightening force loss or tower inclination.

[0030] Aircraft wing / fuselage: The aerodynamic loads and vibration signals during flight are complex. The S-transform can extract the time-frequency characteristics of fatigue cracks, and the DWNO is suitable for real-time analysis of high-speed data streams.

[0031] The present invention will be further described in conjunction with the accompanying drawings and specific embodiments.

[0032] Referring to Figure 1 , the present invention provides a method for diagnosing the structural state of an offshore photovoltaic support, including the following steps: Step S101, generating acceleration response data for healthy and damaged conditions based on a finite element model, and performing time-frequency transformation on the acceleration response data to construct time-frequency matrix data; Step S102, calculating the auto-power spectrum and cross-power spectrum data of each acceleration response data according to the time-frequency matrix data; Step S103, calculating the coherence matrix data of each acceleration response data through a two-dimensional convolution function according to the auto-power spectrum and cross-power spectrum data; Step S104, calculating the coherence matrix data of each acceleration response data under the healthy condition, calculating the coherence matrix data of each acceleration response data under the damaged condition, determining the maximum part of the time-frequency coordinate change as the damage location by comparing the coherence matrix data before and after damage, and calculating the norm of the relative change of the coherence matrix data under the damaged condition, where the norm is used to quantify the damage degree; Step S105, respectively establishing mapping models of acceleration response data with damage location and damage quantification by using a discrete wavelet neural operator according to the time-frequency coordinate change values and the norm of the coherence matrix data before and after damage, and obtaining damage location and damage quantification data by inputting measured acceleration response data into the mapping models.

[0033] In the above embodiments, the time-frequency transformation includes the S-transform, wavelet transform, and short-time Fourier transform; among them, the S-transform, namely the Stockwell transform, is a time-frequency analysis tool that combines the advantages of the short-time Fourier transform (STFT) and wavelet transform. It can adaptively adjust the time-frequency resolution while retaining phase information and is widely used in the analysis of non-stationary signals.

[0034] In some embodiments, based on the acceleration response data, the acceleration signal is S-transformed according to the following formula (1) to obtain an acceleration response time-frequency matrix database: (1) In the formula, is the acceleration response time-frequency matrix data, is the acceleration response signal to be analyzed, is a Gaussian window function with a window width of in the time domain, is a complex weight term, t is time, is frequency, is the time center.

[0035] In the above embodiments, the Finite Element Model (FEM) is a numerical analysis tool used to discretize complex physical systems such as mechanical structures, fluids, electromagnetic fields, etc. into multiple simple small elements such as triangles, quadrilaterals, hexahedrons, etc. By solving the mathematical equations of each element, an approximate solution of the entire system is finally obtained. Its core idea is to "divide and conquer", transforming continuous problems into discrete problems, and is suitable for analyzing complex geometric shapes and nonlinear behaviors. In this application, the continuous structure of the offshore photovoltaic support is divided into a finite number of simple elements such as beam elements, shell elements, solid elements, etc. The elements are connected by nodes, thus transforming the continuous problem into a discrete system of equations. The finite element model is suitable for the structural stability assessment of the offshore photovoltaic support in this application. Applying it to the structural mechanics analysis of the offshore photovoltaic support can simulate the dynamic responses of the support under healthy conditions and different damage conditions.

[0036] Referring to Figure 2 , as an example of the finite element model, a 4-story steel frame finite element model driven by MATLAB provided by ASCE benchmark model Phase Ⅰ is selected. Of course, other finite element models such as ABAQUS, OpenSees, etc. can also be selected. A large amount of acceleration response data under different working conditions is generated through the finite element model. The working conditions include healthy conditions and various custom damage conditions; in some embodiments, the damage conditions include local bolt loosening, beam end cracks, column bottom corrosion, support loosening, and support fracture. Among them, each damage condition is simulated and realized in the following ways: Local bolt loosening is realized by randomly reducing the local stiffness matrix of specific nodes, which is used to simulate the bolt loosening that may occur to the offshore photovoltaic support under the long-term action of sea breeze and waves in the marine environment; Beam end cracks are realized by randomly reducing the local element stiffness at both ends of the beam, which is used to simulate the beam end cracking that may be caused by the long-term impact of waves or wind loads on the offshore photovoltaic support; Column bottom corrosion is realized by randomly reducing the column bottom stiffness, which is used to simulate the column bottom corrosion caused by the seawater erosion of the offshore photovoltaic support; Support loosening is realized by randomly reducing the stiffness of the connection nodes of the support rods, and support fracture is realized by randomly removing the support rod elements completely.

[0037] By applying random white noise to the finite element model to simulate the real environmental load, acceleration sensors are arranged on each floor slab to collect the x and y two-way vibration response signals with a sampling frequency ≥ 500 Hz and a single sampling duration of 30 seconds, generating an acceleration response database containing healthy conditions and various custom damage conditions.

[0038] In some embodiments, after the step of establishing a mapping model between the acceleration response and the damage area using the discrete wavelet neural operator, it further includes: fine-tuning the model parameters based on experimental data to verify the damage location and damage quantification results.

[0039] Exemplarily, the acceleration response dataset of the laboratory four-story steel frame model provided based on the ASCE benchmark model Phase Ⅱ is selected to simulate the real structural acceleration response of the offshore photovoltaic support. Only the acceleration response under the healthy condition is used for fine-tuning, and the acceleration response under the damage condition is used to verify the damage location DWNO and the damage quantification DWNO.

[0040] After the DWNO model is trained, transfer learning is carried out using the laboratory measured data (ASCE Phase II), and the adaptability of the model to actual noise is optimized by fine-tuning the convolution kernel parameters. By fine-tuning the model parameters based on experimental data, it is beneficial to improve the generalization ability of the mapping model, increase the recognition accuracy of damage-like conditions to a certain extent, and verify that the linear fitting between the damage result and the actual damage degree is good.

[0041] In addition to selecting the laboratory four-story steel frame model provided based on the ASCE benchmark model Phase Ⅱ, it is also possible to select to establish a refined finite element model of the offshore photovoltaic support based on models such as ABAQUS and OpenSees, which will not be elaborated here.

[0042] In some embodiments, on the basis of generating acceleration response data of healthy and damaged conditions from the finite element model and performing time-frequency transformation on the acceleration response data to construct time-frequency matrix data, the auto-power spectrum and cross-power spectrum data of each acceleration response data are calculated according to the time-frequency matrix data; among them, the calculation of the auto-power spectrum is shown in formulas (2) to (3): (2) (3) In the formula, is the auto-power spectrum of the time-frequency matrix of the acceleration response data , is the auto-power spectrum of the time-frequency matrix of the acceleration response data j , is the acceleration response data of the acceleration response time-frequency matrix, is the acceleration response data j of the acceleration response time-frequency matrix, where * represents the complex conjugate, i.e., is the complex conjugate matrix of is the complex conjugate matrix of is the frequency, is the time center.

[0043] Calculate the cross-power spectrum as shown in Eqs. (4) to (5): (4) (5) In the equations, is the cross-power spectrum of the time-frequency matrix of the acceleration response data with respect to the time-frequency matrix of the acceleration response data j , is the cross-power spectrum of the time-frequency matrix of the acceleration response data j with respect to the time-frequency matrix of the acceleration response data , is the acceleration response time-frequency matrix of the acceleration response data , is the acceleration response time-frequency matrix of the acceleration response data j , where * represents the complex conjugate, i.e., is the complex conjugate matrix of is the complex conjugate matrix of is the frequency, is the time center.

[0044] The power spectral density is the density function of the signal power distribution with respect to frequency, used to describe the power magnitude of each frequency component of the signal in the frequency domain. It is an important tool for measuring the signal strength in the frequency domain.

[0045] The coherence based on the S-transform needs to be averaged to obtain a suitable estimated value. Lack of averaging will lead to a biased estimate. To obtain an unbiased average estimate of the coherence based on the S-transform, the process of calculating the coherence based on the S-transform using the two-dimensional convolution function is shown in Eqs. (6)-(9): (6) (7) (8) (9) Among them, is the time operator for two-dimensional convolution, is the scale operator formula for two-dimensional convolution, where is the convolution operator, is the acceleration response data is the average unbiased estimate of the auto-power spectrum of the time-frequency matrix of , is the acceleration response data j is the average unbiased estimate of the auto-power spectrum of the time-frequency matrix of , is the acceleration response data is the cross-power spectrum of the time-frequency matrix of the acceleration response data j with respect to the time-frequency matrix of the acceleration response data is the average unbiased estimate of is the acceleration response data j is the cross-power spectrum of the time-frequency matrix of the acceleration response data with respect to the time-frequency matrix of the acceleration response data is the average unbiased estimate of is the acceleration response data is the auto-power spectrum of the time-frequency matrix of translation operations in the time domain and frequency domain, is the acceleration response data j is the auto-power spectrum of the time-frequency matrix of translation operations in the time domain and frequency domain, is the acceleration response data is the cross-power spectrum of the time-frequency matrix of the acceleration response data j with respect to the time-frequency matrix of the acceleration response data translation operations in the time domain and frequency domain, is the acceleration response data j is the cross-power spectrum of the time-frequency matrix of the acceleration response data with respect to the time-frequency matrix of the acceleration response data translation operations in the time domain and frequency domain, is the two-dimensional convolution mask, and the definition formula is as follows: (10) where n is an integer, and 2 ≤ n ≤ 5.

[0046] The calculation process of the coherence matrix data is as follows: (11) where is the coherence matrix data, is the acceleration response data is the cross-power spectrum of the time-frequency matrix of the acceleration response data jThe cross-power spectrum of the time-frequency matrix The square of the absolute value of the mean unbiased estimate of is the acceleration response data The auto-power spectrum of the time-frequency matrix of The mean unbiased estimate of is the acceleration response data j The auto-power spectrum of the time-frequency matrix of The mean unbiased estimate of

[0047] Coherence is a normalized measure that examines the relationship between two analytical responses. Smoothing the coherence matrix with a two-dimensional convolution mask helps suppress noise interference; coherence analysis helps improve the sensitivity for identifying minor stiffness damages such as bolt looseness.

[0048] In some embodiments, the norm includes the Frobenius norm, the spectral norm, and the nuclear norm. Among them, the Frobenius norm is the square root of the sum of the squares of all values in the matrix, the spectral norm is the maximum value among the singular values of the matrix, and the nuclear norm is the sum of the singular values of the matrix.

[0049] In some embodiments, in the steps of calculating the coherence matrix data of each acceleration response data under healthy conditions, calculating the coherence matrix data of each acceleration response data under damaged conditions, determining the location of the maximum change in time-frequency coordinates as the damage location by comparing the coherence matrix data before and after damage, and calculating the norm of the relative change of the coherence matrix data under damaged conditions, where the norm is used to quantify the degree of damage, it includes: Calculating the average value of the coherence matrix data of the acceleration response data under healthy conditions, and using the obtained average coherence matrix data as the location reference, where the subscript 0 represents the healthy condition; Performing eigenvalue decomposition on the average coherence matrix data through the following formula: (12) In the formula, is the coherence matrix data, is The matrix composed of the eigenvectors corresponding to each eigenvalue in is The transpose matrix of is The eigenvalue matrix obtained after eigenvalue decomposition of

[0050] Substituting into formula (12), the eigenvalues are obtained after decomposition. Taking as the quantitative reference; By calculating the coherence matrix data under each damage condition , where the subscript d represents the damage condition. Damage location is performed through the following formula: (13) In the formula, represents the difference matrix of the coherence matrix before and after damage, is the positioning reference, is the coherence matrix data under the damage condition.

[0051] Perform eigenvalue decomposition on the coherence matrix data under each damage condition to obtain the corresponding eigenvalues. Use the relative change value of the eigenvalues under the healthy condition and the damage condition to represent the overall damage degree of the structure, as shown in the following formula: (14) In the formula, is the Frobenius norm, that is, the square root of the sum of the squares of all time-frequency coordinates in the matrix. This DI value is a number between , and the closer it is to 1, the greater the damage degree; is the quantitative reference, is the eigenvalue matrix under the damage condition.

[0052] In the above embodiments, by calculating the average coherence matrix under the healthy condition and performing eigenvalue decomposition, the reference eigenvalue is obtained; under the damage condition, calculate the damage coherence matrix and perform eigenvalue decomposition to obtain . By comparing the Frobenius norms of the eigenvalue matrices before and after damage, the overall damage degree of the structure is quantified into a normalized index (DI value) of 0-1. Eigenvalue decomposition extracts the principal component information of the coherence matrix, and the Frobenius norm comprehensively characterizes the damage degree through the global sum of squares change of the matrix time-frequency coordinates, avoiding the limitations of single-frequency domain features. Directly quantify the damage severity through the DI value (such as the bolt loosening torque loss rate), solving the problem that traditional methods can only qualitatively judge the existence of damage; the principal component analysis based on eigenvalues suppresses noise interference (such as random vibration of sea waves); the Frobenius norm comprehensively reflects the overall change of the matrix, greatly improving the sensitivity to distributed damage (such as corrosion at the bottom of the column); the algorithm has high calculation efficiency, is suitable for embedded devices, and meets the real-time monitoring requirements of offshore photovoltaic supports.

[0053] Refer to Figure 3 , in some embodiments, the steps of establishing a mapping model between the acceleration response and the damage area using a discrete wavelet neural operator include: Step S1051, taking Input to DWNO, is acceleration response data, t is time; Step S1052, use local transformation to lift the dimension of the input to obtain , is a dimensionality-increasing convolutional neural network, and the convolutional kernel size of this convolutional neural network can be 1 1, 2 2, 3 3, etc., and the lifted dimension n, n 2; Step S1053, pass to a series of wavelet kernel integration layers for multi-level wavelet decomposition, that is, parameterize the wavelet kernel to obtain horizontal, vertical, and diagonal coefficients at different levels to complete discrete wavelet decomposition , is 's kernel function, is a convolution operator; Step S1054, reconstruct the convolutional input , is a convolution operator; Step S1055, perform a linear transformation using the convolutional neural network W: , is a convolution operator; Step S1056, add and and perform activation, and finally reduce the dimension to the dimension consistent with the input acceleration response data through local transformation is a dimensionality-reducing convolutional neural network, where the convolutional kernel size of this convolutional neural network can be 1 1, 2 1, 2 2, 3 3, etc., and further complete the mapping between and DI corresponding to it, and complete the mapping between and DI corresponding to it by reducing the dimension to obtain a damage location mapping model and a damage quantification mapping model. the mapping between and DI corresponding to it, and obtain a damage location mapping model and a damage quantification mapping model.

[0054] The Discrete Wavelet Neural Operator (DWNO) is a new operator learning algorithm in the field of deep learning, which is used to learn the mapping between infinite-dimensional function spaces. This algorithm directly learns the non-linear mapping between two function spaces, so it has stronger generalization ability than traditional deep learning. DWNO makes use of the superiority of wavelets in the time-frequency localization of functions, making it possible to accurately track patterns in the spatial domain and effectively learn function mappings. Since wavelets are localized in time, space, and frequency, DWNO can provide high spatial and frequency resolution. DWNO first decomposes the input function space into high-frequency and low-frequency components through the Discrete Wavelet Transform (DWT). In the DWT, the input function space is decomposed by wavelets of different scales. Usually, the components of the function space at lower levels mainly contain noise, so the key information at these scales may be masked. The wavelet coefficients at higher levels, on the other hand, contain more effective feature information. Since the goal is to extract the most relevant features from the function space, only the wavelet coefficients at higher levels in the discrete wavelet transform are selected to construct the information subset. DWNO makes use of the advantages of wavelet transform and can effectively learn the solution operator of highly non-linear partial differential equations (PDEs) with irregular domains and boundary conditions.

[0055] By establishing the mapping model between the acceleration response and the damage area through the above method, wavelet decomposition can accurately capture the local mutations of non-stationary signals such as the impact of cracks at the beam ends and the frequency-domain characteristics such as the resonance shift of the structure, greatly improving the damage localization accuracy; through high-frequency noise suppression by wavelet threshold denoising and parametric kernel adaptive learning, the model still has a good recognition accuracy for untrained damage types such as support fractures; since the original signal is directly mapped to the damage result, avoiding artificial feature design, the training efficiency is greatly improved compared with the traditional method; by fusing time-frequency analysis and deep learning, the diagnostic error for multiple types of damage such as bolt loosening and column bottom corrosion under the interference of sea waves, salt spray, etc. is small, which is significantly better than the single-modal method.

[0056] On the other hand, the present invention also provides a diagnostic system for the structural state of an offshore photovoltaic support, including: A time-frequency matrix module, which is used to generate acceleration response data for healthy and damaged conditions based on a finite element model, and perform an S-transform on the acceleration response data to construct time-frequency matrix data; A power spectrum module, which is used to calculate the auto-power spectrum and cross-power spectrum data of each acceleration response data according to the time-frequency matrix data; A coherence data module, which is used to calculate the coherence matrix data of each acceleration response data through a two-dimensional convolution function according to the auto-power spectrum and cross-power spectrum data; The damage location and quantification module is used to calculate the coherence matrix data of each acceleration response data under healthy conditions, calculate the coherence matrix data of each acceleration response data under damaged conditions, determine the location with the largest change in time-frequency coordinates as the damage location by comparing the coherence matrix data before and after damage, and calculate the norm of the relative change of the coherence matrix data under damaged conditions, where the norm is used to quantify the degree of damage; The discrete wavelet module is used to establish mapping models of acceleration response data with damage location and damage quantification respectively by using discrete wavelet neural operators according to the time-frequency coordinate change values and the norm of the coherence matrix data before and after damage, and input the acceleration response data through the mapping models to obtain damage location and damage quantification data.

[0057] In some embodiments, the above system further includes a fine-tuning and verification module for fine-tuning model parameters based on experimental data and verifying the damage location and damage quantification results.

[0058] Specific fine-tuning can be achieved by gradually increasing the number of layers during multi-layer wavelet decomposition, different selections of wavelet bases such as db6 (Daubechies 6), sym4 (Symlet 4), coif5 (Coiflet 5), different convolutional kernel sizes and the selected lifted dimensions when using a dimensionality-raising convolutional neural network. In S1056, and When adding and activating, different activation functions can be selected, such as ReLU, LeakyReLU, Sigmoid or Tanh.

[0059] The present invention also provides a diagnostic device for the structural state of an offshore photovoltaic support structure, as Figure 4 shown. The device includes a processor 401, a memory 402, a bus 403, and a computer program stored in the memory 402 and executable on the processor 401. The processor 401 includes one or more processing cores. The memory 402 is connected to the processor 401 through the bus 403. The memory 402 is used to store program instructions. When the processor executes the computer program, the steps in the method embodiments of the present invention are implemented.

[0060] Further, as an executable solution, the diagnostic device for the structural state of the offshore photovoltaic support can be a computer unit, which can be a computing device such as a desktop computer, a notebook, a palm computer, and a cloud server. The computer unit may include, but is not limited to, a processor and a memory. Those skilled in the art can understand that the above-described composition structure of the computer unit is only an example of the computer unit and does not constitute a limitation on the computer unit. It may include more or fewer components than the above, or combine certain components, or different components. For example, the computer unit may further include input / output devices, network access devices, a bus, etc., and the embodiments of the present invention do not make any limitations thereto.

[0061] Further, as an executable solution, the so-called processor may be a central processing unit (CPU), or may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The processor is the control center of the computer unit and connects various parts of the entire computer unit through various interfaces and lines.

[0062] The memory can be used to store the computer programs and / or modules. The processor realizes various functions of the computer unit by running or executing the computer programs and / or modules stored in the memory and calling the data stored in the memory. The memory may mainly include a program storage area and a data storage area. Among them, the program storage area may store an operating system and application programs required for at least one function; the data storage area may store data created according to the use of the mobile phone, etc. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one magnetic disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0063] In some embodiments, the present invention further provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned method in the embodiments of the present invention are implemented.

[0064] If the modules / units integrated in the computer unit are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, to implement all or part of the processes in the above-mentioned method embodiments of the present invention, it can also be completed by instructing relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, the steps of the above-mentioned method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file or some intermediate forms, etc. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disc, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction.

[0065] In some embodiments, the present invention further provides a computer program product including a computer program, and when the computer program is executed by a processor, the steps of the method as described above are implemented.

[0066] Although the present invention has been specifically shown and described in conjunction with the preferred embodiments, those skilled in the art should understand that various changes can be made to the present invention in terms of form and details without departing from the spirit and scope of the present invention defined by the appended claims, and all such changes are within the protection scope of the present invention.

Claims

1. A diagnostic method for the structural state of an offshore photovoltaic support, characterized in that, Including the following steps: Generating acceleration response data for healthy and damaged conditions based on a finite element model, and performing time-frequency transformation on the acceleration response data to construct time-frequency matrix data; Calculating the auto-power spectrum and cross-power spectrum data of each acceleration response data according to the time-frequency matrix data; Calculating the coherence matrix data of each acceleration response data through a two-dimensional convolution function according to the auto-power spectrum and cross-power spectrum data; Calculating the coherence matrix data of each acceleration response data under the healthy condition, calculating the coherence matrix data of each acceleration response data under the damaged condition, determining the maximum part of the change in time-frequency coordinates in the coherence matrix data as the damage location by comparing the coherence matrix data before and after damage, and calculating the norm of the relative change of the coherence matrix data under the damaged condition, where the norm is used to quantify the damage degree; According to the time-frequency coordinate change values of the coherence matrix data before and after damage and the norm, respectively establishing mapping models of acceleration response data with damage location and damage quantification using discrete wavelet neural operators, and obtaining damage location and damage quantification data by inputting measured acceleration response data into the mapping models.

2. The diagnostic method for the structural state of the offshore photovoltaic support according to claim 1, characterized in that After the step of establishing the mapping model of acceleration response and damage area using discrete wavelet neural operators, it further includes: Fine-tuning model parameters based on experimental data to verify the damage location and damage quantification results.

3. The diagnostic method for the structural state of the offshore photovoltaic support according to claim 1, characterized in that, In the step of generating acceleration response data for healthy and damaged conditions based on a finite element model and performing time-frequency transformation on the acceleration response data to construct time-frequency matrix data, it includes: Simulating the damaged condition by randomly reducing the node stiffness or removing elements; Simulating the real environmental load by applying random white noise to the finite element model; Wherein, the damaged condition includes local bolt loosening, beam end crack, column bottom corrosion, support loosening and / or support fracture.

4. The diagnostic method for the structural state of the offshore photovoltaic support according to claim 1, wherein In the step of performing time-frequency transformation on the acceleration response data to construct time-frequency matrix data, the time-frequency transformation includes S transform, wavelet transform and short-time Fourier transform; Wherein, the expression of the S transform is: In the formula, is the time-frequency matrix data, is the acceleration response data, is the Gaussian window function with a window width of in the time domain, is the complex weight term, t is the time, f is the frequency, is the time center.

5. The diagnostic method for the structural state of the offshore photovoltaic support according to claim 1, characterized in that, In the step of calculating the coherence matrix data of each acceleration response data through a two-dimensional convolution function according to the auto-power spectrum and cross-power spectrum data, it includes: In the formula, is the convolution operator, is the auto-power spectrum of the time-frequency matrix of the acceleration response data and is the mean unbiased estimate of , is the auto-power spectrum of the time-frequency matrix of the acceleration response data j and is the mean unbiased estimate of , is the cross-power spectrum of the time-frequency matrix of the acceleration response data with respect to the time-frequency matrix of the acceleration response data j and is the mean unbiased estimate of , is the cross-power spectrum of the time-frequency matrix of the acceleration response data j with respect to the time-frequency matrix of the acceleration response data and is the mean unbiased estimate of , is the auto-power spectrum of the time-frequency matrix of the acceleration response data and is the translation operation in the time domain and frequency domain of , is the auto-power spectrum of the time-frequency matrix of the acceleration response data j and is the translation operation in the time domain and frequency domain of , is the cross-power spectrum of the time-frequency matrix of the acceleration response data with respect to the time-frequency matrix of the acceleration response data j and is the translation operation in the time domain and frequency domain of , is the acceleration response data j and is the cross-power spectrum of the time-frequency matrix of the acceleration response data with respect to the time-frequency matrix of the acceleration response data and is the translation operation in the time domain and frequency domain of is the two-dimensional convolution mask, and its definition formula is as follows: In the formula, n is an integer, and 2 ≤ n ≤ 5; The coherence matrix data is as follows: ； Wherein, is the coherence matrix data, is the acceleration response data i of the time-frequency matrix of the acceleration response data j and the cross-power spectrum of the time-frequency matrix is the square of the absolute value of the mean unbiased estimate of is the acceleration response data and the auto-power spectrum of the time-frequency matrix is the mean unbiased estimate of is the acceleration response data j and the auto-power spectrum of the time-frequency matrix is the mean unbiased estimate of 6. The diagnostic method for the structural state of the offshore photovoltaic support according to claim 1, characterized in that, The norm includes Frobenius norm, spectral norm and nuclear norm.

7. The diagnostic method for the structural state of the offshore photovoltaic support according to claim 1, characterized in that, In the step of calculating the coherence matrix data of each acceleration response data under the healthy condition, calculating the coherence matrix data of each acceleration response data under the damaged condition, determining the maximum part of the change in time-frequency coordinates as the damage location by comparing the coherence matrix data before and after damage, and calculating the norm of the relative change of the coherence matrix data under the damaged condition, where the norm is used to quantify the damage degree, it includes: Calculate the average value of the coherence matrix data of the acceleration response data under healthy conditions, and use the obtained average coherence matrix data as the positioning reference, where the subscript 0 represents the healthy condition; Performing eigenvalue decomposition on the average coherence matrix data through the following formula: In the formula, is the coherence matrix data, is the matrix formed by the eigenvectors corresponding to the respective eigenvalues in is the transpose matrix of is the eigenvalue matrix obtained after eigenvalue decomposition; Substitute into the formula for eigenvalue decomposition of the average coherence matrix data, and after decomposition, the eigenvalues are obtained. Take as the quantitative benchmark; By calculating the coherence matrix data under each damage condition , where the subscript d represents the damage condition. Damage location is carried out through the following formula: In the formula, represents the difference matrix of the coherence matrix before and after damage, is the positioning reference, is the coherence matrix data under the damage condition; Performing eigenvalue decomposition on the coherence matrix data under each damaged condition to obtain the corresponding eigenvalues, and using the relative change value of the eigenvalues under the healthy condition and the damaged condition to represent the overall damage degree of the structure, as shown in the following formula: Among them, is the Frobenius norm, and the Frobenius norm is: the square root of the sum of the squares of all time-frequency coordinates in the matrix. DI represents the overall damage degree of the structure, and 0 ≤ DI ≤ 1. The closer DI is to 1, the greater the damage degree. and are the eigenvalue matrices under the quantitative benchmark and the damage condition, respectively.

8. The diagnostic method for the structural state of the offshore photovoltaic support according to claim 1, wherein The mapping models for establishing the acceleration response data and damage location and damage quantification respectively by using the discrete wavelet neural operator according to the time-frequency coordinate change value and the norm of the coherence matrix data before and after damage include: Input to a predetermined discrete wavelet neural operator, is the acceleration response data, t is the time; Using local transformation The input is lifted to a higher dimension to obtain , built by a convolutional neural network Pass to a series of wavelet kernel integral layers for multi-layer wavelet decomposition: parameterize the wavelet kernel to obtain horizontal, vertical, and diagonal coefficients at different levels to complete the discrete wavelet decomposition , where is the kernel function of and is the convolution operator; Reconstructed Convolution Input , is the convolution operator; Perform a linear transformation using the convolutional neural network W: , is the convolution operator; Add and , activate them, and complete the reduction of dimension through local transformation to complete the mapping between and the corresponding and DI, respectively obtaining a damage location mapping model and a damage quantification mapping model.

9. A diagnostic system for the structural state of an offshore photovoltaic support, characterized in that, Including: A time-frequency matrix module, configured to generate acceleration response data for healthy and damaged conditions based on a finite element model, and perform time-frequency transformation on the acceleration response data to construct time-frequency matrix data; A power spectrum module, configured to calculate the auto-power spectrum and cross-power spectrum data of each acceleration response data according to the time-frequency matrix data; A coherence data module, configured to calculate the coherence matrix data of each acceleration response data through a two-dimensional convolution function according to the auto-power spectrum and cross-power spectrum data; A damage location and quantification module, configured to calculate the coherence matrix data of each acceleration response data under the healthy condition, calculate the coherence matrix data of each acceleration response data under the damaged condition, determine the location with the largest change in time-frequency coordinates as the damage location by comparing the coherence matrix data before and after damage, and calculate the norm of the relative change of the coherence matrix data under the damaged condition, where the norm is used to quantify the damage degree; A discrete wavelet module, configured to establish mapping models for the acceleration response data and damage location and damage quantification respectively by using a predetermined discrete wavelet neural operator according to the time-frequency coordinate change value and the norm of the coherence matrix data before and after damage, and input the acceleration response data through the mapping models to obtain damage location and damage quantification data.

10. A diagnostic device for the structural state of an offshore photovoltaic support, characterized in that, Including a memory and a processor, where the memory stores at least one segment of program, and the at least one segment of program is executed by the processor to implement the diagnostic method for the structural state of the offshore photovoltaic support as described in any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that, At least one segment of program is stored in the storage medium, and the at least one segment of program is executed by the processor to implement the diagnostic method for the structural state of the offshore photovoltaic support as described in any one of claims 1 to 8.

Images