A stress field reconstruction method for offshore wind turbine health monitoring

Abstract

The present application relates to the field of offshore wind turbine health monitoring, and specifically relates to a stress field reconstruction method for offshore wind turbine health monitoring, comprising the following steps: step 1: structure modeling and working condition numerical modeling; step 2: data set construction and standardization processing; step 3: full-field stress low-dimensional signature; step 4: candidate monitoring point and observation operator; step 5: sensor layout optimization; step 6: coefficient estimation and full-field reconstruction; step 7: noise injection analysis and robustness evaluation; step 8: result output, visualization and online application. The present application quantitatively evaluates and synchronously optimizes the full-field stress acquisition scheme of the offshore wind turbine jacket structure under the premise of not increasing a large number of measured sensors and offshore experiments, and realizes high-precision and verifiable stress field reconstruction under sparse reachable points and simulation data.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of offshore wind turbine health monitoring, in particular to a stress field reconstruction method for offshore wind turbine health monitoring. BACKGROUND

[0002] In harsh and variable marine environment, offshore wind turbines are subjected to wind, wave, current and other coupled loads for a long time, and the jacket support structure needs to be monitored and evaluated for a long time to ensure safe operation. Therefore, a diversified monitoring system with structural health monitoring as the core is derived, which combines finite element simulation, stress / strain sensing and data analysis, and the monitoring capability and data acquisition accuracy continue to improve. However, the existing monitoring arrangement is often subject to the engineering constraints of difficult access to underwater key parts, high installation and maintenance cost, limited field window period, etc.; the monitoring points often rely on experience selection, and the number is small, and there is a lack of unified theoretical basis and objective evaluation standard for quantitative reconstruction of full-field stress. At the same time, the adaptability and robustness to complex working conditions are insufficient, and it is difficult to obtain high reliability of full-field stress distribution within an acceptable cost. Therefore, it is necessary to establish a full-field stress reconstruction and evaluation method for SHM: under the condition that sensors cannot be fully arranged or only sparse information can be obtained, guided by the model and supported by the data, the monitoring points and information acquisition strategy are quantitatively designed and judged, and the optimal full-field stress acquisition scheme suitable for the jacket is obtained, so as to improve the accuracy and timeliness of fatigue life evaluation and operation safety decision. SUMMARY

[0003] In view of the problem in the prior art that the offshore wind turbine jacket support structure is difficult to access the underwater key parts, the number of sensors is limited, and the points are arranged by experience, and reliable full-field stress distribution cannot be obtained at a reasonable cost, the purpose of the present application is to provide a full-field stress reconstruction and monitoring scheme optimization method suitable for offshore wind turbine jacket support structure health monitoring, which can obtain reliable full-field stress distribution under the condition of sparse observation and installation limitation.

[0004] To achieve the above purpose, the technical scheme adopted by the present application is: a stress field reconstruction method for offshore wind turbine health monitoring, comprising the following steps:

[0005] Step 1: determine the load source, divide the grid, set the convenient condition, establish the parameterized description of the jacket three-dimensional model and the operating condition, and determine the node coordinates and stress output domain;

[0006] Step 2: set the single working condition simulation parameters according to the user's needs, generate multiple working condition simulation snapshots in the parameter domain, form a full-field stress data set, and perform consistency and standardization processing on coordinates, units and scales;

[0007] Step 3: Extract mean field and several basis vectors based on dataset simulation snapshot, establish compact representation of full field stress ≈ mean + basis × coefficient, determine subspace dimension according to interpretability and numerical stability , construct basis using subspace dimension;

[0008] Step 4: Generate feasible monitoring point candidate set combining offshore installation and maintenance constraints, define mapping relationship from full field to sensor observation, clarify available information channel, define observation operator, obtain observation model using basis;

[0009] Step 5: Under budget and engineering constraints, construct comprehensive evaluation model with reconstruction error, cost and robustness as targets, jointly optimize monitoring point combination and training sample set, reduce redundancy and improve distinguishability;

[0010] Step 6: Select regular parameter using fixed value or adaptive selection based on validation set, solve coefficient estimation problem with regularization for sparse observation under given working condition and restore full field stress;

[0011] Step 7: Set noise intensity according to sensor RMS, carry out noiseless and noisy evaluation on validation / test set, output MAE and RMSE indicators of global and hot spot area, test noise resistance and extrapolation ability;

[0012] Step 8: Output monitoring point meta information, form reproducible data package, provide visualization interface, obtain observation at sensor location and reconstruct full field for any test working condition, generate real / predicted side-by-side three-dimensional graph and index label.

[0013] The stress field reconstruction method for offshore wind turbine health monitoring described above, in step 1, establishes a fluid domain in the Fluent module and performs a Boolean operation, the load sources include wind load, wave load, water flow load, and unit operating state, inputs the load sources into the Mechanical module, performs mesh division, sets boundary conditions, and obtains grid nodes.

[0014] The stress field reconstruction method for offshore wind turbine health monitoring described above, in step 2, the single working condition dataset is in the form of , the training set, validation set and test set are set in the proportion of 0.7:0.15:0.15, and samples are obtained, the stress mean and stress standard deviation of each node are calculated using the training dataset, for the stress field sample of any working condition, the full field normalized vector is expanded element by element through the formula , and the normalized vector is obtained, wherein, Represents a node The standardized dimensionless value at that point. Represents a node The original stress value at that location, Represents stress field samples In the The original stress values ​​of each node, For stress field samples In the The dimensionless stress value of each node after standardization.

[0015] The stress field reconstruction method for offshore wind turbine health monitoring described above, step 3 includes: constructing a low-dimensional representation in the normalized snapshot Z of the training set to obtain the basis. With mean field The stress under any working condition is expressed as: ,in, This represents the eigenvector of the stress field under this working condition in a low-dimensional subspace.

[0016] The stress field reconstruction method for offshore wind turbine health monitoring described above, step 4 includes:

[0017] Step 4-1: Based on actual offshore installation and maintenance constraints at all nodes Forming a candidate set Select a set of sensors from them ,in, Represents a set of sensors The number of sensors on board, and the actual constraints for offshore installation and maintenance include accessibility, wiring, corrosion resistance, and minimum distance from existing components;

[0018] Step 4-2: Define the observation operator from the entire field to the sensor. Then the observation model under a certain working condition is:

[0019] Among them, noise , It is the identity matrix. The standard deviation of noise.

[0020] The stress field reconstruction method for offshore wind turbine health monitoring described above, step 5 includes:

[0021] Step 5-1: Under the premise of satisfying engineering constraints, for the sensor set With regularization parameters Construct and minimize the comprehensive objective, denoted as... In order to be in and Below is a sample of working conditions. The entire venue was reconstructed;

[0022] Step 5-2: Define the reconstruction cost of the validation set: ,in, This step indicates that the sample index set is used for verification. Indicates the first The true standardized stress vector of a sample of working conditions Indicates the first The squared norm of the reconstruction error vector of a working condition sample at all nodes;

[0023] Step 5-3: Let the installation and maintenance cost function be... The robustness penalty is ;

[0024] Step 5-4: Given weights The comprehensive optimization problem is written as:

[0025] ,

[0026] ,

[0027] ,in, For the defined comprehensive optimization objective function, Indicates the first Coordinates of each monitoring point Indicates the first Spatial coordinate vectors of monitoring points For minimum spacing constraints, Indicates partition, This indicates the upper limit of component quota statistics. Indicates the lower limit of component quota statistics;

[0028] Step 5-5: Use rank reveal decomposition to obtain a preliminary selection set with a large amount of information, and perform combinatorial optimization under the constraints in steps 5-1 to 5-4.

[0029] In the aforementioned stress field reconstruction method for offshore wind turbine health monitoring, step 6 includes:

[0030] Step 6-1: Observation of the given working condition In the sensor set Order , ,in, Indicating in the sensor set The mean observation vector, which is the average stress value corresponding to each sensor location. Indicating in the sensor set The corresponding basis vector matrix;

[0031] Step 6-2: Obtain low-dimensional coefficients Regularized least squares estimator Its closed-form solution satisfies: ;

[0032] Step 6-3: Solve the stable problem using Cholesky decomposition. The full-field reconstruction result is as follows: ;

[0033] Step 6-4: Use a normalized mesh And the mean absolute error of the validation set Choose the regular expression parameter, where, Indicates the first The first working condition sample was in the... The predicted value of each node, Indicates the first The first working condition sample was in the... The true value of each node.

[0034] The stress field reconstruction method for offshore wind turbine health monitoring described above, step 7 includes:

[0035] Step 7-1: Set the noise intensity based on the sensor RMS during training, so that the training set... The observation matrix on is Its scale is:

[0036] ,

[0037] ,in, Indicates the noise percentage coefficient. For the validation set sample index set, It is the number of samples in the validation set. Indicates the first Individual operating condition samples in the sensor set Standardized observation vectors on;

[0038] Step 7-2: Injection repeat By averaging the results, robust optimal regularization parameters are obtained. With performance evaluation;

[0039] Step 7-3: Use , Conduct indicator testing, among which, Indicates the first The reconstructed value of each node, Indicates the first The actual value of each node.

[0040] The stress field reconstruction method for offshore wind turbine health monitoring described above includes, in step 8, the metadata including monitoring point index and spatial coordinates, model parameters, data partitioning and error summary. Step 8 also includes: during operation, following the same process as steps 1-7, accessing new observation or simulation snapshots, performing incremental updates and re-evaluations to ensure the model remains stable and reliable as operating conditions change.

[0041] The beneficial effects of this invention's stress field reconstruction method for offshore wind turbine health monitoring are as follows: It quantitatively evaluates and simultaneously optimizes the full-field stress acquisition scheme for offshore wind turbine jacket structures without increasing the number of measured sensors and offshore experiments, achieving high-precision and verifiable stress field reconstruction under sparsely accessible points / simulation data; it reduces monitoring and deployment costs, avoiding information redundancy and ineffective deployment; and it improves the accuracy and efficiency of stress assessment under complex operating conditions, providing a practical theoretical and methodological foundation for the health monitoring, integrity assessment, and life prediction of jacket structures. Attached Figure Description

[0042] Figure 1 This is a schematic diagram of marine working conditions and sensor layout in an embodiment of the present invention;

[0043] Figure 2 This is a schematic diagram of the dataset format in an embodiment of the present invention;

[0044] Figure 3 This is a schematic diagram of the dataset selection window in the visualization operation interface when visualizing the operation and maintenance diagram in an embodiment of the present invention;

[0045] Figure 4 This is a schematic diagram of the parameter setting window for the visualization operation interface when visualizing the operation and maintenance diagram in an embodiment of the present invention;

[0046] Figure 5 This is a schematic diagram illustrating the selection of test samples when visualizing the operation and maintenance diagram in this embodiment of the invention;

[0047] Figure 6 This is a schematic diagram of the actual stress results when visualizing the operation and maintenance diagram in an embodiment of the present invention;

[0048] Figure 7 This is a schematic diagram of the reconstruction stress results when visualizing the operation and maintenance diagram in an embodiment of the present invention;

[0049] Figure 8 This is an overall flowchart of an embodiment of the present invention. Detailed Implementation

[0050] To enable those skilled in the art to better understand the technical solution of the present invention, the technical solution of the present invention will be described below in conjunction with specific embodiments and accompanying drawings.

[0051] Example 1

[0052] A stress field reconstruction method for health monitoring of offshore wind turbines includes the following steps.

[0053] Step 1: Determine the load source, divide the mesh, set convenient conditions, establish a three-dimensional model of the jacket and a parametric description of the operating conditions, and clarify the node coordinates and stress output domain.

[0054] Step 2: Set the single-condition simulation parameters according to user needs, generate multi-condition simulation snapshots within the parameter domain, form a full-field stress dataset, and perform consistency and standardization processing of coordinates, units, and scales.

[0055] Step 3: Extract the mean field and several basis vectors from the simulation snapshot of the dataset, and establish a compact representation of the total stress as mean + basis vectors × coefficients. Use the interpretation rate and numerical stability to determine the dimension of the subspace. The basis is constructed using the dimension of the subspace.

[0056] Step 4: Combine the constraints of offshore installation and maintenance to generate a feasible candidate set of monitoring points, define the mapping relationship from the whole field to sensor observation, clarify the available information channels, define the observation operators, and obtain the observation model using the basis.

[0057] Step 5: Under budget and engineering constraints, construct a comprehensive evaluation model with the goals of reconstruction error, cost and robustness, and jointly optimize the combination of monitoring points and training sample set to reduce redundancy and improve identifiability.

[0058] Step 6: Using fixed values ​​or adaptive selection of regularization parameters based on the validation set, solve the regularized coefficient estimation problem and recover the full-field stress for sparse observations under a given working condition.

[0059] Step 7: Set the noise intensity according to the sensor RMS, conduct noise-free and noise-added evaluations on the verification / test set, and output the MAE and RMSE indices for the global and hot spot areas to verify the noise resistance and extrapolation capabilities.

[0060] Step 8: Output monitoring point metadata to form a reproducible data package, provide a visualization interface, acquire observations at sensor locations for any test condition and reconstruct the entire field, and generate a side-by-side 3D map of the real / predicted data and indicator annotations.

[0061] Example 2

[0062] like Figures 1-8 As shown, a stress field reconstruction method for health monitoring of offshore wind turbines includes the following steps.

[0063] S1, Structural modeling and numerical modeling of operating conditions.

[0064] Establish a three-dimensional model of the jacket and a parametric description of its operating conditions, and clarify the node coordinates and stress output domain.

[0065] Taking the jacket support structure of an offshore wind turbine as the research object, a three-dimensional model was established. A fluid domain was created in the Fluent module and Boolean operations were performed. External factors such as wind, waves, current, and turbine operating status were parameterized as load sources for the fluid domain. The load file was imported into the Mechanical module, meshed, and boundary conditions were set. The total results for this step are recorded. Each grid node. The operation procedure for this step is as follows: Figure 1 As shown.

[0066] Figure 1 This is a schematic diagram of the operating environment, used to show the operating environment of the jacket platform and the sensor layout. Figure 1 The diagram in the middle illustrates an offshore wind turbine jacket system in a typical wind-wave-current coupled marine environment. The upper region is the air domain, primarily subjected to wind speed profiles varying with altitude; the interface between the air and water is the wave domain, where waves propagate across the water surface and exert alternating hydrodynamic forces on the jacket structure; the lower region is the water domain, primarily subjected to current velocity profiles varying with depth. Furthermore, in the fluid-structure interaction analysis, a fixed support constraint is applied to the interface between the jacket and the seabed to secure the jacket.

[0067] S2, Dataset Construction and Standardization.

[0068] Multi-condition simulation snapshots are generated within the parameter domain to form a full-field stress dataset; the consistency and standardization of coordinates, units, and scales are completed.

[0069] The single-condition dataset is in the form of ,like Figure 2 As shown. This step can be configured according to user needs, such as a training set, validation set, and test set ratio of 0.7:0.15:0.15. Let the total obtained be... Using a sample dataset, the mean stress of each node is calculated. and stress standard deviation For stress field samples under arbitrary working conditions The standardized vector is obtained by expanding the entire field standardized vector element by element using equation (1). .

[0070] (1),

[0071] in, Represents a node The standardized dimensionless value at that point. Represents a node The original stress value at that location, Represents stress field samples In the The original stress values ​​of each node, For stress field samples In the The dimensionless stress value of each node after standardization.

[0072] Figure 2 This illustrates the sample data obtained from fluid-structure interaction analysis, where the sample data has the following dimensions: The 4 dimensions are represented as ( , , , In this study, a total of 97,660 nodes were obtained through mesh generation, i.e. In this study, 66 sensors were sparsely arranged, and 97,594 nodes needed to be predicted. Therefore, the 97,660 nodes in the same sample data were divided into 66 sensor nodes and 97,594 prediction nodes. Furthermore, each fluid-structure interaction scenario output 33 sample data points, therefore the total number of samples is... Therefore, 594 samples were used for the training and validation of the method.

[0073] S3, low-dimensional characteristic of full-field stress.

[0074] Based on training snapshots, the mean field and several basis vectors are extracted to establish a compact representation of "total stress ≈ mean + basis × coefficients"; the subspace dimension is determined based on the explanatory power and numerical stability. It supports rapid reconstruction.

[0075] Construct a low-dimensional representation in the normalized snapshot Z of the training set to obtain the basis. With mean field , base Given an N-row, r-column matrix, Determined based on the cumulative variance explained rate and numerical stability. Utilizing the subspace dimension. Constructing a base, with different subspace dimensions This could lead to substrates of different dimensions. The stress under any given condition can be expressed by equation (2).

[0076] (2),

[0077] in Determined based on the cumulative variance explained rate and numerical stability, This represents the eigenvector of the stress field under this working condition in a low-dimensional subspace. To accommodate both large-scale samples and streaming data, this step can employ incremental matrix factorization with a mini-batch iterative fitting approach.

[0078] S4, candidate monitoring points and observation operators.

[0079] By combining the constraints of offshore installation and maintenance, a feasible candidate set of monitoring points is generated, and the mapping relationship from the whole field to sensor observation is defined to clarify the available information channels.

[0080] Based on actual offshore installation and maintenance constraints (accessibility, wiring, corrosion resistance, minimum spacing from existing components, etc.), at all nodes Forming a candidate set Select a set of sensors from them ,in, Represents a set of sensors The number of sensors can also be understood as the actual number of monitoring points installed in the current deployment plan.

[0081] Define the observation operator from the entire field to the sensor. Then the observation model under a certain working condition is:

[0082] (3),

[0083] Among the noise ,in, The identity matrix, dimension, and noise vector are given. Consistent, The standard deviation of noise This indicates that the observation error of each sensor follows a mean of 0 and a standard deviation of . Gaussian white noise, The scale is set based on the RMS of the sensor during the training period (see S7).

[0084] S5, sensor layout optimized.

[0085] Under budget and engineering constraints, a comprehensive evaluation model is constructed with the goals of reconstructing error, cost and robustness. The combination of monitoring points and training sample set are jointly optimized to reduce redundancy and improve identifiability.

[0086] Under the premise of meeting engineering constraints, for sensor assemblies With regularization parameters A comprehensive objective is constructed and minimized to reduce spatial redundancy and improve recognizability and reconstruction accuracy. In order to be in and The following sample The full reconstruction (see S6) defines the reconstruction cost of the validation set:

[0087] (4),

[0088] in, This step indicates that the sample index set is used for verification, i.e. It is the number of samples. Indicates the first The true standardized stress vector of a sample of working conditions Indicates the first The squared L2 norm of the reconstruction error vector of a sample at all nodes.

[0089] Let the installation and maintenance cost function be... Robustness penalties (such as the condition number or the reciprocal of the minimum singular value) are: Given weights The comprehensive optimization problem is written as:

[0090] (5),

[0091] (6),

[0092] (7), where, For the defined comprehensive optimization objective function, Indicates the first Coordinates of each monitoring point Indicates the first The spatial coordinate vectors of candidate (or selected) monitoring points. For minimum spacing constraints, Indicates partition, This indicates the upper limit of component quota statistics. This indicates the lower limit for component quota statistics. For minimum spacing constraints, This indicates the zoning and component quota statistics, subject to upper and lower limits. , Constraints. In practice, rank decomposition can be used to obtain an initial selection set with a large amount of information, and then combinatorial optimization can be performed under the above constraints.

[0093] S6, Coefficient Estimation and Full Field Reconstruction.

[0094] For sparse observations under given working conditions, solve the problem of coefficient estimation with regularization and recover the full field stress; the regularization parameter can be a fixed value or an adaptive selection based on the validation set to balance accuracy and stability.

[0095] Observation of a given working condition In the sensor set Order , Then the low-dimensional coefficients The regularized least squares estimate is:

[0096] (8),

[0097] in, Indicating in the sensor set The mean observation vector, which is the average stress value corresponding to each sensor location. Indicating in the sensor set The corresponding basis vector matrix.

[0098] Its closed-form solution satisfies:

[0099] (9).

[0100] The Cholesky decomposition method can be used to stably solve the problem, and the full-field reconstruction result is as follows:

[0101] (10).

[0102] To improve robustness across operating conditions and noise levels, normalized meshes are used for the regularization parameters. And the mean absolute error of the validation set is used for selection:

[0103] (11), among which, Indicates the first The first working condition sample was in the... The predicted value of each node, Indicates the first The first working condition sample was in the... The true value of each node.

[0104] Formulas (8), (9), and (10) give the results for a given set of sensors. and regular expression parameters At that time, from the observation Obtain the reconstructed stress field The complete process of its error. Therefore, formula (11) adopts a normalized grid based on the first three steps. The optimal value is obtained by using the mean absolute error of the validation set as the metric. Thus determining the optimal This enables adaptive selection of regularization parameters.

[0105] In formula (11), It is a dimensionless scalar coefficient used to control the regularization strength in a normalized manner. In selecting the optimal... During the process, different values Each has a separate ,therefore, For all The set, and yes The best one selected from among them. .

[0106] S7, Injection Noise Analysis and Robustness Assessment.

[0107] The noise intensity is set according to the sensor RMS. Noise-free and noise-added evaluations are carried out on the verification / test set. The MAE, RMSE and other indicators of the global and hot spot areas are output to verify the noise resistance and extrapolation capabilities.

[0108] Noise intensity is set based on the sensor RMS during training: The training set is set at... The observation matrix on is Its scale is:

[0109] (12),

[0110] (13)

[0111] in, Indicates the noise percentage coefficient. The validation set is the set of sample indices, which is the set of operating condition numbers that are assigned to the validation set. It is the number of samples in the validation set. Indicates the first Individual operating condition samples in the sensor set The standardized observation vector on.

[0112] Injection during verification repeat Taking the average of the two methods yields a robust result. With performance evaluation. Consistent with the explanation in formula (11), through injecting noise and repetition The optimal regularization parameter is obtained by "sub-average verification". The difference between the two is that in formula (11) Find the optimal state in a noise-free environment. In this step, the optimal solution is found under noisy conditions. .

[0113] The test metrics used are:

[0114] , (14), among which, Indicates the first The reconstructed value of each node, Indicates the first The actual value of each node.

[0115] S8: Results output, visualization, and online applications.

[0116] Export monitoring point index and coordinates, model parameters and error summary; provide side-by-side visualization of true values ​​and predictions for project acceptance; support incremental updates and re-evaluation of new data, facilitating operation and maintenance implementation and long-term use.

[0117] After completing the above process, output the monitoring point index and spatial coordinates, and model parameters. Meta-information such as data partitioning and error summarization is used to form a reproducible data package. A visualization interface is also provided to acquire observations at sensor locations for any test condition and reconstruct the entire field, generating a side-by-side "real / predicted" 3D graph with indicator annotations for engineering acceptance and decision support, such as... Figures 3-7 As shown, during operation, adding new observations or simulation snapshots can be incrementally updated and re-evaluated using the same process, ensuring the model remains stable and reliable as operating conditions change.

[0118] The above embodiments are merely illustrative of the structural concept and features of the present invention, intended to enable those skilled in the art to understand the content of the present invention and implement it accordingly, and should not be construed as limiting the scope of protection of the present invention. All equivalent changes or modifications made based on the essence of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A stress field reconstruction method for health monitoring of offshore wind turbines, characterized in that, Includes the following steps: Step 1: Determine the load source, divide the mesh, set convenient conditions, establish a three-dimensional model of the jacket and a parametric description of the operating conditions, and clarify the node coordinates and stress output domain; Step 2: Set the single-condition simulation parameters according to user needs, generate multi-condition simulation snapshots within the parameter domain, form a full-field stress dataset, and perform consistency and standardization processing of coordinates, units, and scales. Step 3: Extract the mean field and several basis vectors from the simulation snapshot of the dataset, and establish a compact representation of the total stress as mean + basis vectors × coefficients. Use the interpretation rate and numerical stability to determine the dimension of the subspace. The basis is constructed using the dimension of the subspace; Step 4: Combine the constraints of offshore installation and maintenance to generate a feasible candidate set of monitoring points, define the mapping relationship from the whole field to sensor observation, clarify the available information channels, define the observation operators, and obtain the observation model using the basis. Step 5: Under budget and engineering constraints, construct a comprehensive evaluation model with the goals of reconstruction error, cost and robustness, and jointly optimize the combination of monitoring points and training sample set to reduce redundancy and improve identifiability; Step 6: Using fixed values ​​or adaptive selection of regularization parameters based on the validation set, solve the coefficient estimation problem with regularization and recover the full field stress for sparse observations under a given working condition; Step 7: Set the noise intensity according to the sensor RMS, conduct noise-free and noise-added evaluations on the verification / test set, and output the MAE and RMSE indices for the global and hot spot areas to verify the noise resistance and extrapolation capabilities. Step 8: Output monitoring point metadata to form a reproducible data package, provide a visualization interface, acquire observations at sensor locations for any test condition and reconstruct the entire field, and generate a side-by-side 3D map of the real / predicted data and indicator annotations.

2. The stress field reconstruction method for offshore wind turbine health monitoring according to claim 1, characterized in that, In step 1, a fluid domain is established and Boolean operations are performed in the Fluent module. The load sources include wind load, wave load, water flow load, and unit operating status. The load sources are input into the Mechanical module, meshed, and boundary conditions are set to obtain the desired result. 1 grid node.

3. The stress field reconstruction method for offshore wind turbine health monitoring according to claim 1, characterized in that, In step 2, the single-condition dataset is in the form of The training set, validation set, and test set were configured in a ratio of 0.7:0.15:0.15 to obtain... Using a sample dataset, the mean stress of each node is calculated. and stress standard deviation For stress field samples under arbitrary working conditions Through formula Expand the normalized vector element by element to obtain the normalized vector. ,in, Represents a node The standardized dimensionless value at that point. Represents a node The original stress value at that location, Represents stress field samples In the The original stress values ​​of each node, For stress field samples In the The dimensionless stress value of each node after standardization.

4. The stress field reconstruction method for offshore wind turbine health monitoring according to claim 1, characterized in that, Step 3 includes: constructing a low-dimensional representation in the normalized snapshot Z of the training set to obtain the basis. With mean field The stress under any working condition is expressed as: ,in, This represents the eigenvector of the stress field under this working condition in a low-dimensional subspace.

5. The stress field reconstruction method for offshore wind turbine health monitoring according to claim 1, characterized in that, Step 4 includes: Step 4-1: Based on actual offshore installation and maintenance constraints at all nodes Forming a candidate set Select a set of sensors from them ,in, Represents a set of sensors The number of sensors on board, and the actual constraints for offshore installation and maintenance include accessibility, wiring, corrosion resistance, and minimum distance from existing components; Step 4-2: Define the observation operator from the entire field to the sensor. Then the observation model under a certain working condition is: Among them, noise , It is the identity matrix. The standard deviation of noise.

6. The stress field reconstruction method for offshore wind turbine health monitoring according to claim 1, characterized in that, Step 5 includes: Step 5-1: Under the premise of satisfying engineering constraints, for the sensor set With regularization parameters Construct and minimize the integrated objective, denoted as... In order to be in and Below is a sample of working conditions. The entire venue was reconstructed; Step 5-2: Define the reconstruction cost of the validation set: ,in, This step indicates that the sample index set is used for verification. Indicates the first The true standardized stress vector of a sample of working conditions Indicates the first The squared norm of the reconstruction error vector of a working condition sample at all nodes; Step 5-3: Let the installation and maintenance cost function be... The robustness penalty is ; Step 5-4: Given weights The comprehensive optimization problem is written as: ， ， ,in, For the defined comprehensive optimization objective function, Indicates the first Coordinates of each monitoring point Indicates the first Spatial coordinate vectors of monitoring points For minimum spacing constraints, Indicates partition, This indicates the upper limit of component quota statistics. Indicates the lower limit of component quota statistics; Step 5-5: Use rank reveal decomposition to obtain a preliminary selection set with a large amount of information, and perform combinatorial optimization under the constraints in steps 5-1 to 5-4.

7. The stress field reconstruction method for offshore wind turbine health monitoring according to claim 1, characterized in that, Step 6 includes: Step 6-1: Observation of the given working condition In the sensor set Order , ,in, Indicating in the sensor set The mean observation vector, which is the average stress value corresponding to each sensor location. Indicating in the sensor set The corresponding basis vector matrix; Step 6-2: Obtain low-dimensional coefficients Regularized least squares estimator Its closed-form solution satisfies: ; Step 6-3: Solve the stable problem using Cholesky decomposition. The full-field reconstruction result is as follows: ; Step 6-4: Use a normalized mesh And the mean absolute error of the validation set Choose the regular expression parameter, where, Indicates the first The first working condition sample was in the... The predicted value of each node, Indicates the first The first working condition sample was in the... The true value of each node.

8. The stress field reconstruction method for offshore wind turbine health monitoring according to claim 1, characterized in that, Step 7 includes: Step 7-1: Set the noise intensity based on the sensor RMS during training, so that the training set... The observation matrix on is Its scale is: ， ,in, Indicates the noise percentage coefficient. For the validation set sample index set, It is the number of samples in the validation set. Indicates the first Individual operating condition samples in the sensor set Standardized observation vectors on; Step 7-2: Injection repeat By averaging the results, robust optimal regularization parameters are obtained. With performance evaluation; Step 7-3: Use , Conduct indicator testing, among which, Indicates the first The reconstructed value of each node, Indicates the first The actual value of each node.

9. The stress field reconstruction method for offshore wind turbine health monitoring according to claim 1, characterized in that, In step 8, the metadata includes monitoring point index and spatial coordinates, model parameters, data partitioning and error summary. Step 8 also includes: during operation, following the same process as steps 1-7, accessing new observations or simulation snapshots, performing incremental updates and reviews, so that the model remains stable and reliable as the operating conditions change.

Images