A method for judging the health state of a mooring line of a floating offshore structure

Abstract

The application provides a kind of offshore floating structure mooring health state determination method, belong to the ocean engineering structure monitoring technical field based on computer data processing;Collect health state and the time history response signal of mooring tension to be detected state and pretreatment, carry out multiscale decomposition using CEEMDAN algorithm, reconstruct to obtain first high-frequency response signal, carry out secondary fine frequency division using VMD algorithm, reconstruct to obtain enhanced high-order response signal, preset sliding window subsection extraction time domain feature, generate health feature matrix and to be detected feature matrix, use KPCA to process health feature matrix, establish health benchmark model, calculate the kernel square prediction error of health and to be detected feature matrix, compare the statistical indicators of the kernel square prediction error of both, determine the state of mooring, judge sea condition according to the mean offset of kernel square prediction error;The application is suitable for extreme sea conditions, suppresses ocean load interference, and enhances the damage identification ability of mooring.

Description

Technical Field

[0001] This invention belongs to the field of marine engineering structure monitoring technology based on computer data processing, and particularly relates to a method for determining the health status of moorings for floating marine structures. Background Technology

[0002] The mooring lines of semi-submersible platforms, floating wind turbines, and offshore photovoltaic systems are subjected to time-varying loads coupled with waves and ocean currents for extended periods, making them prone to early degradation damage such as cable corrosion, internal wire breakage, joint slippage, and anchor chain slack. Early damage only produces weak pulse impact signals, which are easily masked by large-scale low-frequency wind and wave loads and high-frequency ocean clutter, making it difficult for traditional monitoring methods to identify minute damage.

[0003] To address the aforementioned monitoring challenges, this field faces two core technological bottlenecks: First, marine environmental loads are highly nonlinear, time-varying, and sudden, such as sudden storm changes and sea ice impacts. Existing signal decomposition methods cannot achieve a balance between "adaptation to extreme sea states" and "damage feature extraction." Second, mooring tension response is a coupled superposition of environmental load excitation and mooring structure damage response. Traditional fault diagnosis models have not established a mechanism to distinguish between the two. When encountering extreme sea states, the amplitude of tension fluctuations caused by environmental loads increases dramatically, leading to a significant shift in the health assessment threshold and a significant increase in the false alarm rate, making it impossible to achieve unified and accurate monitoring under multiple sea state conditions.

[0004] Existing studies have directly input the coupling characteristics of environmental loads and structural damage into the model without effectively decoupling and distinguishing between the two. When sea conditions change drastically, the model's judgment threshold drifts accordingly, making it impossible to determine whether the tension anomaly is caused by severe sea conditions or actual mooring damage, resulting in both false alarms and missed alarms.

[0005] Therefore, there is an urgent need for a mooring monitoring method that can simultaneously achieve adaptive decomposition under extreme sea conditions, accurate purification of damage characteristics, and unified health assessment across multiple sea conditions. Summary of the Invention

[0006] To address the aforementioned issues, this invention proposes a method for determining the health status of mooring lines based on the adaptive noise complete set empirical mode decomposition algorithm (CEEMDAN) - variational mode decomposition (VMD) second-level signal decomposition and joint kernel principal component analysis (KPCA).

[0007] This invention provides a method for determining the health status of mooring lines for floating marine structures, comprising the following steps: S1, collects the time history response signals of the tether tension of the floating system in the healthy state and the state to be tested, and performs preprocessing; S2. The CEEMDAN algorithm is used to perform multi-scale decomposition on the preprocessed mooring tension time history response signal, extract the intrinsic mode function components at different frequency scales, screen the intrinsic mode function components, and reconstruct the first high-frequency response signal. S3, VMD is used to perform secondary fine frequency division on the first high-frequency response signal to obtain multiple variational mode functions with different center frequencies and frequency band characteristics. The variational mode functions are selected and reconstructed to obtain the enhanced high-order response signal. S4. The enhanced high-order response signals of the healthy state and the state to be detected are segmented by a preset sliding window. The temporal feature vectors representing the structural health state in each window are extracted and combined in chronological order to obtain the health state feature matrix and the state to be detected feature matrix, respectively. S5. Based on the KPCA method, perform kernel mapping and dimensionality reduction on the health feature matrix to establish a health benchmark model and calculate the kernel squared prediction error of each feature vector in the health feature matrix; map the feature matrix to be detected to the kernel feature space of the health benchmark model and calculate the kernel squared prediction error corresponding to each sliding window feature vector. S6. Calculate and compare the nuclear squared prediction error statistical index between the test state and the healthy state to determine the mooring status; if the determination result is a healthy state, then distinguish the wave height level according to the degree of deviation of the mean nuclear squared prediction error under the healthy state.

[0008] Preferably, in step S1, the preprocessing of the mooring tension time history response signal specifically includes: processing the mooring tension time history response signals in the healthy state and the state to be detected. Trend term separation is performed, and the long-term tension drift term is obtained using the least squares linear fitting method. The destressed tether tension time history response signal is obtained by subtracting it from the original signal. .

[0009] Preferably, the S2 process specifically includes: The CEEMDAN algorithm is used to perform multi-scale decomposition on the detrended mooring tension time history signal by setting adaptive noise amplitude and ensemble averaging degree. This enables preliminary separation of environmental load response and structural damage response at different frequency scales. The expression is as follows: ; In the formula, This is the time history response signal of the mooring tension after trend elimination. For time variables, For intrinsic mode functions, For the first i eigenmode functions of order 1 This is the residual trend term; The low-frequency intrinsic mode functions reflect long-period waves and slow-changing motion of the floating body, while the high-frequency intrinsic mode functions reflect the characteristics of frictional impact, local stiffness changes, wire breakage impact, and joint relaxation. Mode selection is performed based on frequency domain prior constraints, kurtosis index optimization, and energy proportion threshold. The first three high-frequency intrinsic mode functions are retained, and the retained high-frequency intrinsic mode functions are reconstructed to obtain the first high-frequency response signal.

[0010] Preferably, in step S3, the first high-frequency response signal output from step S2 is input to variational mode decomposition (VMD) for secondary fine frequency division, and the mode number parameter is set. K and balance parameters α It is decomposed into multiple narrowband modal components with different center frequencies, as expressed by: ; In the formula, The first high-frequency response signal output by S2, For time variables, These are the narrowband modes obtained after VMD decomposition; Calculate each variational mode function separately. The kurtosis and information entropy based on the Hilbert envelope amplitude probability distribution are obtained, and a comprehensive index for each variational mode function is constructed based on the kurtosis and information entropy. The variational mode functions are sorted in descending order according to the comprehensive index, and the variational mode functions with a preset proportion of the highest ranking are selected for reconstruction to obtain an enhanced higher-order response signal.

[0011] Preferably, in step S4, a window length L and a window movement step S are set to overlap and divide the enhanced high-order response signal, so that the continuous response signal is converted into multiple local state samples; within each sliding window, the mean, standard deviation, root mean square, kurtosis, skewness, sample entropy, fractal dimension, and fluctuation index features are extracted and arranged in a fixed order as column vectors to form the feature column vector of the window; the feature column vectors of each window are combined in chronological order to obtain the health state feature matrix and the feature matrix to be detected, respectively.

[0012] Preferably, the health feature matrix is ​​mapped to the kernel feature space using a KPCA model, and the mapped principal component space is used as the health baseline model; wherein, KPCA uses a Gaussian radial basis kernel function, the expression of which is: ; In the formula, For the sample x i With sample x j The inner product mapping value in kernel space. x i andx j For the first i , j The feature vectors corresponding to each sliding window σ The Gaussian kernel width parameter controls the sensitivity of KPCA to differences.

[0013] Preferably, in step S5, the kernel squared prediction error is used to characterize the degree of deviation of the detected feature sample from the feature space in the KPCA health baseline model, and its calculation formula is as follows: ; In the formula, Let be the feature vector of the sample to be detected. for With its own kernel function value, This is the implicit mapping vector that maps to the high-dimensional feature space. Let its projection onto the principal component subspace be , The kernel crossover vector is the vector between the sample to be tested and the healthy sample set. This is the transpose of the kernel crossover vector between the sample to be tested and the healthy sample set. This is the kernel matrix in the health benchmark model. It is the principal component projection transformation matrix.

[0014] Preferably, step S6 specifically includes: using the statistical index of the kernel square prediction error under healthy conditions as benchmark data, comparing the statistical index of the kernel square prediction error of the data to be tested with the benchmark data; when the statistical index of the kernel square prediction error of the data to be tested is higher than the benchmark data, it is determined that the mooring cable has not suffered abnormal damage; under this premise, the greater the deviation of the statistical index of the kernel square prediction error under the test state from the statistical index of the kernel square prediction error under healthy conditions, the farther the sea state deviates from the benchmark operating condition, and the lower the wave height; if it is lower than the benchmark data, it can be determined that the mooring cable structure has broken or other abnormalities.

[0015] Preferably, the statistical index of the kernel square prediction error under the healthy state is the mean of the statistical distribution characteristics of the kernel square prediction error corresponding to multiple sliding windows under the healthy state, and the statistical index of the kernel square prediction error of the data to be detected is the mean of the statistical distribution characteristics of the kernel square prediction error under the state to be detected.

[0016] Compared with the prior art, the present invention has the following beneficial effects: 1. Adapting to the non-stationary characteristics of extreme sea states: CEEMDAN adaptively processes strong time-varying excitations of storms and swells, without losing instantaneous weak damage impacts; it compensates for the poor adaptability of single VMD and the distortion of abrupt signal decomposition. 2. Highly efficient suppression of marine load interference: Two-level frequency division layer-by-layer stripping of low-frequency wave loads solves industry pain points: damage pulses are submerged by load signals under strong sea conditions, and early minor damage cannot be identified; 3. Strong ability to amplify weak damage: VMD narrowband segmentation filters high-frequency ocean clutter, greatly improving the signal-to-noise ratio of damage characteristics, and can identify early-stage wire breakage and early degradation of micro-joint slippage. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the following description is only one embodiment of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of the overall process of the present invention.

[0019] Figure 2 This is a schematic diagram of the mooring system layout for a deep-water semi-submersible platform according to an embodiment of the present invention. The model is based on a right-handed coordinate system, with the platform center as the origin, OX pointing to the east side of the platform, and OY pointing to the north side of the platform. Specifically, it represents the mooring layout under the baseline working conditions and when the significant wave heights are 11.6m and 7.3m.

[0020] Figure 3 This is a schematic diagram of the mooring system layout of a deep-water semi-submersible platform according to an embodiment of the present invention. It is modeled using a right-handed coordinate system, with the platform center as the origin, OX pointing to the east side of the platform, and OY pointing to the north side of the platform. Specifically, it shows the mooring arrangement when mooring cable No. 12 breaks.

[0021] Figure 4 This is a characteristic diagram of the kernel squared prediction error distribution under a 0° wave direction according to an embodiment of the present invention.

[0022] Figure 5 This is a characteristic diagram of the kernel squared prediction error distribution under a 15° wave direction according to an embodiment of the present invention.

[0023] Figure 6 This is a characteristic diagram of the kernel squared prediction error distribution under a 30° wave direction according to an embodiment of the present invention.

[0024] Figure 7 This is a characteristic diagram of the kernel squared prediction error distribution under a 45° wave direction according to an embodiment of the present invention. Detailed Implementation

[0025] This invention proposes a method for determining the health status of mooring lines for floating marine structures. The overall process is as follows: Figure 1 As shown, it includes the following steps: S1, collects the time history response signals of the tether tension of the floating system in the healthy state and the state to be tested, and performs preprocessing; S2, the CEEMDAN algorithm is used to perform multi-scale decomposition on the preprocessed mooring tension time history response signal, extract the intrinsic mode functions at different frequency scales, screen the intrinsic mode functions, and reconstruct the first high-frequency response signal; S3, VMD is used to perform secondary fine frequency division on the first high-frequency response signal to obtain multiple variational mode functions with different center frequencies and frequency band characteristics. The variational mode functions are selected and reconstructed to obtain the enhanced high-order response signal. S4. The enhanced high-order response signals of the healthy state and the state to be detected are segmented by a preset sliding window. The temporal feature vectors representing the structural health state in each window are extracted and combined in chronological order to obtain the health state feature matrix and the state to be detected feature matrix, respectively. S5. Based on KPCA, perform kernel mapping and dimensionality reduction on the health feature matrix to establish a health benchmark model and calculate the kernel squared prediction error of each feature vector in the health feature matrix; map the feature matrix to be detected to the kernel feature space of the health benchmark model and calculate the kernel squared prediction error corresponding to each sliding window feature vector. S6. Calculate and compare the nuclear squared prediction error statistical index between the test state and the healthy state to determine the mooring status; if the determination result is a healthy state, then distinguish the wave height level according to the degree of deviation of the mean nuclear squared prediction error under the healthy state.

[0026] The present invention will be further described below with reference to embodiments. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0027] This embodiment uses a carbon fiber composite mooring system of a deep-water semi-submersible platform as the monitoring object. The semi-submersible platform adopts a 4-group, 16-point mooring system, with each group containing 4 mooring cables. The carbon fiber composite mooring cables are 2003m long, have a dry weight of 14.1kg / m, and an axial stiffness of 1254.8MN.

[0028] S1, the tension time history response signal of the floating system mooring is acquired in real time by a tension sensor arranged on the mooring line, and the specific steps for preprocessing the tension time history response signal of the mooring line are as follows: Time history response signals of mooring tension in healthy and under-detection states Trend term separation is performed, and the long-term tension drift term is obtained using the least squares linear fitting method. The destressed tether tension time history response signal is obtained by subtracting it from the original signal. , ; In the formula, This refers to the time history response signal of the mooring tension during the operation of the floating system. This is the tension time history response signal of the mooring after trend elimination and detrending. This is the long-term tension drift term.

[0029] S2, the detrended mooring tension time history response signal obtained from S1 is decomposed into multiple scales using the CEEMDAN algorithm. By setting the adaptive noise amplitude and ensemble averaging degree, the mooring tension time history response signal is decomposed into several intrinsic mode components, so that the environmental load response and structural damage response at different frequency scales can be initially separated. The specific mathematical expression is as follows: ; In the formula, in the formula, This is the time history response signal of the mooring tension after trend elimination. For time variables, For intrinsic mode functions, For the first i eigenmode functions of order 1 This represents the residual trend term.

[0030] Eigenmode functions (EMFs) are extracted at different frequency scales. Low-frequency EMFs reflect long-period waves and slow-varying motion of the floating body, while high-frequency EMFs reflect frictional impact, local stiffness changes, wire breakage impact, and joint relaxation characteristics of the structural health. Mode selection is performed based on frequency domain prior constraints, kurtosis index optimization, and energy proportion thresholds. The first three high-frequency EMFs are retained. The retained high-frequency EMFs are then reconstructed to obtain the first high-frequency response signal. ; In the formula, The first high-frequency response signal output by S2, For time variables, These are high-frequency intrinsic mode functions.

[0031] S3, the first high-frequency response signal obtained in S2 is subjected to secondary fine frequency division using the VMD algorithm: ; In the formula, The first high-frequency response signal output by S2, For time variables, These are the narrowband modes obtained after VMD decomposition.

[0032] By setting the modal number parameter Kand balance parameters α The first high-frequency response signal is decomposed into multiple narrowband mode components with different center frequencies to reduce mode aliasing and improve the ability to separate the damage feature frequency band, thereby obtaining multiple variational mode functions with different center frequencies and frequency band characteristics.

[0033] Furthermore, the kurtosis and information entropy based on the Hilbert envelope amplitude probability distribution of each variational mode function are calculated respectively. A comprehensive index for each variational mode function is constructed based on the kurtosis and information entropy. The multiple variational mode functions are then subjected to a secondary screening based on the comprehensive index. The specific calculation steps are as follows: For the k Variational mode functions Its kurtosis is: ; In the formula, It is the first The kurtosis of a variational mode function, The index of the variational mode function. The number of sampling points. These are the narrowband modes obtained after VMD decomposition. These are the narrowband modes obtained after VMD decomposition. It is a time variable.

[0034] right Perform Hilbert transform to construct analytic signal And take the envelope amplitude: ; In the formula, For the first The Hilbert envelope amplitude of each component, For time variables, The index of the variational mode function. For the first Analytical signal of each component, These are the narrowband modes obtained after VMD decomposition. The imaginary unit, For the first The Hilbert transform results for each component.

[0035] Normalize the envelope amplitude into a probability distribution: ; In the formula, For the first The variational mode functions at time... The normalized envelope probability, The index of the variational mode function. For the first The Hilbert envelope amplitude of each component, For numerically stable terms, It is a time variable.

[0036] The corresponding information entropy is: ; In the formula, For the first The information entropy index of the envelope amplitude of a variational mode function. The index of the variational mode function. For the first The variational mode functions at time... The normalized envelope probability, The number of sampling points. For time variables, It is a numerically stable term.

[0037] The comprehensive indicators are: ; In the formula, For the first A comprehensive damage sensitivity index for a variety of mode functions. It is the first The kurtosis of a variational mode function, For the first The information entropy index of the envelope amplitude of a variational mode function. It is a numerically stable term.

[0038] according to Sort in descending order and select the first few. Proportional modality, index set is Reconstructing damage-sensitive signals: ; In the formula, To reconstruct the enhanced higher-order response signal, These are the narrowband modes obtained after VMD decomposition. For indexed sets, It is a time variable.

[0039] The variational mode functions are sorted in descending order according to the comprehensive index, and the variational mode functions with the highest sorting ratio are selected for reconstruction to obtain the enhanced higher-order response signal.

[0040] S4, segment the enhanced higher-order response signals of the healthy state and the state to be detected by a preset sliding window. The window length is set to 2000, and the window movement step size is set to 1000. Extract the mean within each sliding window. Standard deviation Root mean square , cliff skewness Sample entropy fractal dimension and volatility index characteristics The feature column vectors that represent the structural health status : ; In the formula, For each sliding window, there are feature column vectors representing the structural health status. For the number of windows, The mean, Standard deviation, It is the root mean square. For ravine, For skewness, For sample entropy, For fractal dimension, It is a volatility index.

[0041] Then, the transposes of the multiple feature column vectors corresponding to each sliding window are combined in chronological order to form a feature matrix: ; In the formula, The characteristic matrix, For the first Transpose of eigenvectors For the number of windows, This represents the total number of sliding windows. Let be a real matrix space.

[0042] The health status feature matrix and the feature matrix to be detected are obtained respectively.

[0043] S5. Based on KPCA, kernel mapping and dimensionality reduction are performed on the health feature matrix to establish a health benchmark model. The specific steps are as follows: The health feature matrix is ​​input into the KPCA model for kernel feature space mapping, and the mapped principal component space is used as the health baseline model; wherein, KPCA uses the Gaussian radial basis kernel function, the expression of which is: ; In the formula, For the sample x i With sample x j The inner product mapping value in kernel space. x i and x j For the first i , jThe feature vectors corresponding to each sliding window σ The Gaussian kernel width parameter controls the sensitivity of KPCA to differences. After the health baseline model is established, the kernel squared prediction error of each feature vector in the health feature matrix is ​​calculated. The kernel squared prediction error is used to characterize the degree of deviation of the detected feature sample from the feature space in the KPCA health baseline model, and its calculation formula is as follows: ; In the formula, Let be the feature vector of the sample to be detected. for With its own kernel function value, This is the implicit mapping vector that maps to the high-dimensional feature space. Let its projection onto the principal component subspace be , The kernel crossover vector is the vector between the sample to be tested and the healthy sample set. This is the transpose of the kernel crossover vector between the sample to be tested and the healthy sample set. The kernel matrix as a health benchmark, It is the principal component projection transformation matrix.

[0044] The feature matrix to be detected is mapped to the kernel feature space of the health benchmark model, and the kernel squared prediction error corresponding to each sliding window feature vector is calculated.

[0045] S6, calculate the statistical mean of the kernel squared prediction error for both the healthy state and the state to be detected: ; In the formula, The mean of the kernel squared prediction error under healthy conditions. The kernel squared error for each window of health data. For the number of health status windows, This represents the number of windows.

[0046] ; In the formula, The mean of the kernel squared prediction error under the condition to be detected is given. The kernel squared error of each window of the data to be tested. The number of state windows to be detected. This represents the number of windows.

[0047] The mean of the kernel squared prediction error under healthy conditions is used as the benchmark data. The mean of the kernel squared prediction error of the data to be tested is compared with the benchmark data. When the mean of the kernel squared prediction error of the data to be tested is lower than the benchmark data, it is determined that the tether has suffered abnormal damage, that is: ; In the formula, The mean of the kernel squared prediction error under the condition to be detected is given. The mean of the squared prediction error under healthy conditions is given.

[0048] When the statistical mean of the kernel squared prediction error of the data to be tested is higher than the baseline data, the mooring is determined to be healthy, but the sea state in which it is located has changed, that is: ; In the formula, The mean of the kernel squared prediction error under the condition to be detected is given. The mean of the squared prediction error under healthy conditions is given.

[0049] Under this premise, sea state conditions are determined based on the offset of the kernel squared prediction error statistical index: ; pass The magnitude of the error indicates the degree to which the sea state deviates from the baseline operating condition. The greater the deviation of the nuclear square prediction error statistical index under the test state from the nuclear square prediction error statistical index under the healthy state, the further the sea state deviates from the baseline operating condition, and the lower the wave height.

[0050] Numerical simulation and result analysis: To verify the accuracy and engineering applicability of this invention in identifying mooring damage and determining changes in sea state conditions, a dynamic response model of the mooring system was established. This model, based on OrcaFlex software, constructs a coupled dynamic analysis model of a floating platform and mooring system. It simulates and acquires tension time-history data under different operating conditions and extracts and analyzes the mooring tension response at the connection between the CFRP cable and the top steel chain. Specific operating condition settings are as follows: This invention uses a once-in-a-century extreme sea state as the benchmark operating condition, which employs the JONSWAP spectrum and significant wave height. =14.6m, spectral peak period = 12.6 s, peak factor =2.4. For other operating conditions, the significant wave height is taken sequentially. =11.6m =7.3m, No. 12 mooring cable broke ( =14.6m), and the above three working conditions are all adjusted and set based on the baseline working conditions. Among them =11.6m The 7.3m condition only changes the significant wave height, while the No. 12 mooring cable breakage condition assumes that the No. 12 mooring cable breaks under the baseline condition, while all other conditions remain unchanged. The specific mooring cable arrangement is as follows: Figure 2 and Figure 3 As shown.

[0051] To effectively identify mooring breakage and significant wave height, this embodiment proposes a two-stage joint signal preprocessing method using CEEMDAN-VMD. First, CEEMDAN adaptively decomposes the non-stationary mooring response time series, eliminating the load component dominated by low-frequency waves. The first three retained intrinsic mode functions are then reconstructed to obtain the first high-frequency response signal. Next, VMD performs narrowband fine-tuning of the high-frequency components, retaining the high-frequency intrinsic mode functions, and reconstructing the enhanced higher-order response signal characterizing the structural health status. The mean, standard deviation, root mean square, kurtosis, skewness, sample entropy, fractal dimension, and wave index features of the enhanced higher-order response signal are extracted using a sliding window. These features are arranged in a fixed order as column vectors, forming the feature column vector of the window. Finally, the feature column vectors of each window are combined in chronological order to construct the health status feature matrix and the feature matrix to be detected. The health feature matrix is ​​input into the KPCA model and mapped to the kernel feature space using a Gaussian radial basis kernel function. The mapped principal component space is used as the health baseline model, and the kernel squared prediction error characterizes the deviation of the sample to be detected from the health feature space. By comparing the kernel squared prediction error statistics and their offsets between the detected state and the healthy state, abnormal damage identification of mooring lines and determination of health status changes under different sea conditions are achieved. This invention utilizes CEEMD-VMD joint enhancement of weak damage features, combined with the KPCA nonlinear health modeling method, to reduce the risk of misjudgment and missed judgment under complex marine load conditions, providing a reliable method for early damage monitoring of floating mooring systems. Based on the above research objectives, this embodiment selects the baseline working condition, =11.6m 、 A comparative analysis was conducted on four typical working conditions: a 7.3m mooring line and a No. 12 mooring line breakage. Each working condition had four wave directions: 0°, 15°, 30°, and 45°.

[0052] The feasibility and effectiveness of the proposed health status determination method were verified by using the tension time history response of cable No. 11 as the input signal.

[0053] Figures 4-7 This is a characteristic map of the kernel squared prediction error distribution under different wave conditions according to an embodiment of the present invention. Figure 4 This is a characteristic diagram of the squared prediction error distribution under a 0° wave. Figure 5 This is a characteristic diagram of the squared prediction error distribution under a 15° wave. Figure 6 This is a characteristic diagram of the kernel squared prediction error distribution under a 30° wave. Figure 7 The distribution characteristics of the squared prediction error under a 45° wave are shown in the diagram.

[0054] Figure 4 The scatter plot distribution and statistical characteristics of kernel squared prediction errors for various working conditions under 0° wave direction are shown. Black scatter plots represent the baseline working condition, and blue represents the significant wave height.H s =11.6m, red indicates significant wave height H s =7.3m, orange indicates the No. 12 mooring cable breakage condition. When environmental conditions change but the mooring cable is in good condition, the mean squared prediction error is higher than that of the baseline condition. =14.6m), and it increases with the decrease of significant wave height, that is, as the deviation of wave height from the reference condition increases, the mean deviation of the nuclear square prediction error also increases; when the adjacent No. 12 mooring cable breaks, the mean nuclear square prediction error is lower than that of the reference condition ( =14.6m), and the dispersion is enhanced, indicating that the mooring line has suffered abnormal damage. By comparing the mean kernel square prediction error of the tested condition with that of the baseline condition, the invention is verified to be able to adapt to changes in sea state and determine whether there is an anomaly in the mooring line's condition. Based on the determination that the mooring line is in a healthy state, the environmental load changes of the mooring line are distinguished by the deviation between the mean kernel square prediction error of the tested condition and that of the baseline condition. Figures 5-7 The distribution and statistical characteristics of kernel squared prediction error at incident wave directions of 15°, 30° and 45° are shown. The sensitivity of the mean kernel squared prediction error to working condition identification is basically the same as that at incident wave direction of 0°.

[0055] In addition, when ocean wave height increases (e.g., wave height is significant) When the wave height is greater than 14.6m, the amplitude of the time history response of the mooring cable will increase directly. Such conditions can usually be determined directly by the time domain characteristics such as the mean of the tension response, without relying on the method based on the kernel square prediction error distribution characteristics proposed in this patent. Therefore, the wave height increase condition is not within the scope of this method.

[0056] Table 1 presents the statistical results of the nuclear squared prediction error for mooring line No. 11 under this segment duration. Under all wave conditions, by comparing the statistical distribution and offset of the nuclear squared prediction error between the detected state and the healthy state, abnormal damage to the mooring line can be identified, and changes in its health status under different sea conditions can be determined.

[0057] Table 1 Statistical results of the nuclear square prediction error corresponding to cable No. 11 ( =120s)

[0058] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0059] While the specific embodiments of the present invention have been described above, they are not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A method for determining the health status of mooring lines for offshore floating structures, characterized in that, Includes the following processes: S1, collects the time history response signals of the tether tension of the floating system in the healthy state and the state to be tested, and performs preprocessing; S2, the CEEMDAN algorithm is used to perform multi-scale decomposition on the preprocessed mooring tension time history response signal, extract the intrinsic mode functions at different frequency scales, screen the intrinsic mode function components and reconstruct the first high-frequency response signal. S3, VMD is used to perform secondary fine frequency division on the first high-frequency response signal to obtain multiple variational mode functions with different center frequencies and frequency band characteristics. The variational mode functions are selected and reconstructed to obtain the enhanced high-order response signal. S4. The enhanced high-order response signals of the healthy state and the state to be detected are segmented by a preset sliding window. The temporal feature vectors representing the structural health state in each window are extracted and combined in chronological order to obtain the health state feature matrix and the state to be detected feature matrix, respectively. S5. Based on the KPCA method, perform kernel mapping and dimensionality reduction on the health feature matrix to establish a health benchmark model and calculate the kernel squared prediction error of each feature vector in the health feature matrix; map the feature matrix to be detected to the kernel feature space of the health benchmark model and calculate the kernel squared prediction error corresponding to each sliding window feature vector. S6. Calculate and compare the nuclear squared prediction error statistical index between the test state and the healthy state to determine the mooring status. If the determination result is a healthy state, the wave height level is distinguished according to the degree of deviation of the mean of the core square prediction error under the healthy state.

2. The method for determining the health status of mooring lines for a floating marine structure as described in claim 1, characterized in that: In step S1, the preprocessing of the mooring tension time history response signal specifically includes: processing the mooring tension time history response signals in the healthy state and the state to be detected. Trend term separation is performed, and the long-term tension drift term is obtained using the least squares linear fitting method. The destressed tether tension time history response signal is obtained by subtracting it from the original signal. .

3. The method for determining the health status of mooring lines for a floating marine structure as described in claim 1, characterized in that: The specific process of S2 includes: The CEEMDAN algorithm is used to perform multi-scale decomposition on the detrended mooring tension time history signal by setting adaptive noise amplitude and ensemble averaging degree. This enables preliminary separation of environmental load response and structural damage response at different frequency scales. The expression is as follows: ； In the formula, This is the time history response signal of the mooring tension after trend elimination. For time variables, Intrinsic mode functions, For the first i eigenmode functions of order 1 This is the residual trend term; The low-frequency intrinsic mode functions reflect long-period waves and slow-changing motion of the floating body, while the high-frequency intrinsic mode functions reflect the characteristics of frictional impact, local stiffness changes, wire breakage impact, and joint relaxation. Mode selection is performed based on frequency domain prior constraints, kurtosis index optimization, and energy proportion threshold. The first three high-frequency intrinsic mode functions are retained, and the retained high-frequency intrinsic mode functions are reconstructed to obtain the first high-frequency response signal.

4. The method for determining the health status of mooring lines for a floating marine structure as described in claim 1, characterized in that: In step S3, the first high-frequency response signal output from S2 is input to Variational Mode Decomposition (VMD) for secondary fine frequency division, and the mode number parameter is set. K and balance parameters α It is decomposed into multiple narrowband modal components with different center frequencies, as expressed by: ； In the formula, The first high-frequency response signal output by S2, For time variables, These are the narrowband modes obtained after VMD decomposition; Calculate each variational mode function separately. The kurtosis and information entropy based on the Hilbert envelope amplitude probability distribution are obtained, and a comprehensive index for each variational mode function is constructed based on the kurtosis and information entropy. The variational mode functions are sorted in descending order according to the comprehensive index, and the variational mode functions with a preset proportion of the highest ranking are selected for reconstruction to obtain an enhanced higher-order response signal.

5. The method for determining the health status of mooring lines for a floating marine structure as described in claim 1, characterized in that: In step S4, the window length L and window movement step S are set to overlap and divide the enhanced high-order response signal, so that the continuous response signal is converted into multiple local state samples. In each sliding window, the mean, standard deviation, root mean square, kurtosis, skewness, sample entropy, fractal dimension and fluctuation index features are extracted and arranged into column vectors in a fixed order to form the feature column vector of the window. The feature column vectors of each window are combined in time order to obtain the health state feature matrix and the feature matrix to be detected.

6. The method for determining the health status of mooring lines for a floating marine structure as described in claim 1, characterized in that: The health feature matrix is ​​mapped to the kernel feature space using the KPCA method, and the mapped principal component space is used as the health baseline model; wherein, KPCA uses the Gaussian radial basis kernel function, the expression of which is: ； In the formula, For the sample x i With sample x j The inner product mapping value in kernel space. x i and x j For the first i , j The feature vectors corresponding to each sliding window σ The Gaussian kernel width parameter controls the sensitivity of KPCA to differences.

7. The method for determining the health status of mooring lines for a floating marine structure as described in claim 1, characterized in that: In S5, the kernel squared prediction error is used to characterize the degree of deviation of the detected feature sample from the feature space in the KPCA health baseline model, and its calculation formula is as follows: ； In the formula, Let be the feature vector of the sample to be detected. for With its own kernel function value, This is the implicit mapping vector that maps to the high-dimensional feature space. Let its projection onto the principal component subspace be , The kernel crossover vector is the vector between the sample to be tested and the healthy sample set. This is the transpose of the kernel crossover vector between the sample to be tested and the healthy sample set. This is the kernel matrix in the health benchmark model. It is the principal component projection transformation matrix.

8. The method for determining the health status of mooring lines for a floating marine structure as described in claim 1, characterized in that: S6 specifically includes: using the statistical index of the kernel square prediction error under healthy conditions as the benchmark data, comparing the statistical index of the kernel square prediction error of the data to be tested with the benchmark data; when the statistical index of the kernel square prediction error of the data to be tested is higher than the benchmark data, it is determined that the mooring cable has not suffered abnormal damage; under this premise, the greater the deviation of the statistical index of the kernel square prediction error under the test state from the statistical index of the kernel square prediction error under healthy conditions, the farther the sea state deviates from the benchmark operating condition, and the lower the wave height; if it is lower than the benchmark data, it can be determined that the mooring cable structure is abnormal.

9. The method for determining the health status of mooring lines for a floating marine structure as described in claim 8, characterized in that: The statistical index of the kernel square prediction error under the healthy state is the mean of the statistical distribution characteristics of the kernel square prediction error corresponding to multiple sliding windows under the healthy state, and the statistical index of the kernel square prediction error of the data to be detected is the mean of the statistical distribution characteristics of the kernel square prediction error under the state to be detected.

Images