A wind power installation ship structure health prediction method and system based on radar constraint Gaussian structure field

Abstract

The present application relates to the technical field of offshore engineering equipment health monitoring, in particular to a wind power installation ship structure health prediction method and system based on radar constraint Gaussian structure field. The method comprises the following steps: performing semantic division of key structure areas of the wind power installation ship based on compensation results; constructing a three-dimensional Gaussian structure field based on the divided structure point cloud; performing benchmark structure field registration according to the constructed three-dimensional Gaussian structure field; identifying ship structure abnormalities based on registration deviation; extracting structure degradation features based on the ship structure abnormality results; and predicting health risks based on the extracted structure degradation features. The present application maps the health risk index back to the three-dimensional Gaussian structure field to generate a three-dimensional structure risk heat map, so that the monitoring results can directly serve the safety management and maintenance decision of the wind power installation ship field operation.

Description

Technical Field

[0001] This invention relates to the field of marine engineering equipment health monitoring technology, and in particular to a method and system for predicting the structural health of wind turbine installation vessels based on radar-constrained Gaussian structure fields. Background Technology

[0002] Currently, structural health monitoring of wind turbine installation vessels and large marine equipment relies heavily on strain gauges, accelerometers, fiber optic grating sensors, displacement sensors, and manual inspections. While these methods can obtain local response data at specific measurement points, they still have significant limitations. In recent years, lidar, due to its non-contact, high-precision, and large-scale three-dimensional measurement capabilities, has been increasingly applied to spatial mapping and geometric inspection of ships, marine engineering equipment, and large structures. However, traditional lidar detection methods primarily rely on point cloud registration, mesh modeling, or surface ranging, and their results typically remain at the level of static geometric reconstruction and single-shot deviation detection. It is difficult to transform multi-time-lapse scan results into a structural health status expression that can be updated over a long period, quantitatively compared, and predictably evolved. Three-dimensional Gaussian reconstruction technology can continuously represent spatial structures or scenes using a set of three-dimensional Gaussian units. Each Gaussian unit can contain parameters such as center position, scale, rotation, covariance, transparency, color, or reflection intensity. However, existing 3D Gaussian reconstruction technology is mainly geared towards visual scene reconstruction, 3D visualization, or spatial representation, and has not yet fully integrated with the health monitoring needs of critical load-bearing structures on wind turbine installation vessels. It lacks an integrated approach for structural semantic regions, baseline model registration, motion compensation, anomaly identification, and degradation trend prediction. Simply using 3D Gaussian reconstruction as a visualization tool cannot directly solve the problems of anomaly identification and predictive maintenance in the structural health management of wind turbine installation vessels.

[0003] Furthermore, wind turbine installation vessels operate in complex and dynamic environments, subject to low-frequency swaying caused by waves, localized responses from lifting loads, attitude changes due to the movement of the lifting system, and variations in the support status of the pile legs. These factors lead to non-structural shifts in the lidar point cloud in both time and space. Without compensation for the overall rigid body motion of the hull, structural deviations calculated based on the point cloud or 3D model will be mixed with a large amount of non-deformable displacement, thus reducing the reliability of deformation identification, misalignment judgment, and health prediction. Therefore, how to dynamically compensate the lidar point cloud by combining multi-source operational data such as hull attitude, lifting status, crane load, and pile leg pressure, and how to separate the true structural anomalies from it, is a key issue facing structural health monitoring of wind turbine installation vessels.

[0004] Meanwhile, existing monitoring systems mostly rely on threshold alarms or single detection results, lacking predictive health management mechanisms that consider operational phases and temporal evolution. Traditional monitoring methods fail to fully integrate the temporal growth characteristics of structural anomalies, operational phase information, and load condition information, making it difficult to predict the expansion trend of abnormal areas, remaining safe operating windows, and maintenance priorities.

[0005] In summary, existing technologies still struggle to achieve continuous three-dimensional state representation of critical load-bearing structures, separation of actual structural anomalies, prediction of degradation trends, and closed-loop maintenance decision-making under the complex sea conditions, dynamic operations, multi-source load coupling, and large structural spatial constraints of wind turbine installation vessels. Therefore, there is an urgent need for a novel predictive health management method and system for wind turbine installation vessel structures that integrates lidar three-dimensional perception, hull motion compensation, three-dimensional Gaussian representation of structural semantics, reference structural field registration, Gaussian parameter temporal drift analysis, and health risk prediction. This would enable spatial identification and predictive maintenance of key structural deformations, misalignments, settlement, gap anomalies, connection node anomalies, suspected surface degradation areas, and dynamic response anomalies on wind turbine installation vessels. Summary of the Invention

[0006] To address the aforementioned problems, this invention provides a method and system for predicting the structural health of wind turbine installation vessels based on radar-constrained Gaussian structure fields.

[0007] In a first aspect, the present invention provides a method for predicting the structural health of a wind turbine installation vessel based on a radar-constrained Gaussian structure field, which adopts the following technical solution: A method for predicting the structural health of wind turbine installation vessels based on radar-constrained Gaussian structure fields includes: Acquire multi-source heterogeneous data, including lidar point cloud data and ship operation data; Compensation for rigid body motion of point cloud under the coupling effect of complex sea conditions and installation operations using hull operation data and lidar point cloud data. Semantic segmentation of key structural regions of wind turbine installation vessel based on compensation results; A three-dimensional Gaussian structure field is constructed based on the partitioned structural point cloud; The reference structure field is registered based on the constructed three-dimensional Gaussian structure field; Identifying hull structural anomalies based on registration deviation; Structural degradation features are extracted based on the results of ship structure anomalies. Health risk prediction based on extracted structural degradation features; A three-dimensional structural risk heat map and maintenance decisions are generated based on the health risk prediction results.

[0008] Secondly, a structural health prediction system for wind turbine installation vessels based on radar-constrained Gaussian structure fields includes: The data acquisition module is configured to acquire multi-source heterogeneous data, including lidar point cloud data and ship operation data. The motion compensation module is configured to use hull operation data and lidar point cloud data to perform rigid body motion compensation of point cloud under the coupling effect of complex sea conditions and installation operations. The semantic segmentation module is configured to perform semantic segmentation of key structural areas of the wind turbine installation vessel based on the compensation results. The structure field module is configured to construct a three-dimensional Gaussian structure field based on the partitioned structure point cloud. The registration module is configured to register the reference structure field based on the constructed three-dimensional Gaussian structure field. The anomaly detection module is configured to identify hull structural anomalies based on registration deviation. The feature extraction module is configured to extract structural degradation features based on the results of hull structural anomalies. The health prediction module is configured to predict health risks based on extracted structural degradation features. The decision-making module is configured to generate a three-dimensional structural risk heat map and make maintenance decisions based on the health risk prediction results.

[0009] Thirdly, the present invention provides a computer-readable storage medium storing a plurality of instructions adapted for loading and execution by a processor of a terminal device of the method for predicting the structural health of a wind turbine installation vessel based on a radar-constrained Gaussian structure field.

[0010] Fourthly, the present invention provides a terminal device, including a processor and a computer-readable storage medium, wherein the processor is used to implement various instructions; the computer-readable storage medium is used to store multiple instructions, the instructions being adapted to be loaded and executed by the processor as described in the "Wind Turbine Installation Vessel Structural Health Prediction Method Based on Radar-Constrained Gaussian Structure Field".

[0011] In summary, the present invention has the following beneficial technical effects: 1. It realizes the spatial, continuous, and healthy expression of the key load-bearing structure of the wind power installation vessel.

[0012] This invention no longer relies solely on discrete measuring point sensors or single point cloud ranging results. Instead, it utilizes lidar point clouds to construct a three-dimensional Gaussian structural field with structural semantic labels, uniformly expressing key structures such as pile legs, lifting mechanisms, crane bases, deck main beams, connecting nodes, and weld neighborhoods as an updatable, comparable, and predictable structural state field.

[0013] 2. Improved the reliability of structural anomaly identification under complex sea conditions and dynamic operating conditions.

[0014] This invention combines the ship's attitude, lifting operation status, hoisting load, and leg support status to perform rigid body motion compensation on lidar point clouds. It can effectively separate the overall displacement interference caused by wind and wave swaying, platform attitude changes, and equipment operation movements, and reduce the risk of misjudging the ship's motion as structural deformation.

[0015] 3. It expands the structural health monitoring capabilities of traditional 3D reconstruction.

[0016] This invention does not use 3D Gaussian reconstruction merely as a visualization model, but instead utilizes Gaussian center drift, covariance variation, principal axis direction variation, local Gaussian density variation, and reflection intensity variation to construct structural deviation fields and degradation features, thereby identifying various structural state changes such as deformation, misalignment, settlement, gap changes, abnormal connection nodes, suspected areas of surface degradation, and abnormal dynamic response.

[0017] 4. Enhanced predictive health management capabilities.

[0018] This invention extracts structural degradation features based on the Gaussian parameter deviation sequence within a continuous operation cycle, and combines this with the wind power installation vessel's operation stage and load conditions to predict health risks. This can improve upon single anomaly detection to risk trend judgment and remaining safe operation window assessment for future operation cycles.

[0019] 5. A closed loop of perception, reconstruction, identification, prediction, and decision-making has been formed.

[0020] This invention maps the health risk index back to a three-dimensional Gaussian structural field to generate a three-dimensional structural risk heat map, and further outputs maintenance priorities, rescanning strategies, load reduction suggestions, suggestions to reduce lifting speed, suggestions to suspend hoisting or shutdown for maintenance, so that the monitoring results can directly serve the on-site safety management and maintenance decisions of wind turbine installation vessels. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the overall structure of a wind turbine installation vessel structure health prediction method and system based on radar-constrained Gaussian structure field provided by an embodiment of the present invention. Figure 2 This is a schematic diagram of a method and system flow for predicting the structural health of a wind turbine installation vessel based on a radar-constrained Gaussian structure field, provided by an embodiment of the present invention. Figure 3 This is a schematic diagram of the key load-bearing structure and lidar monitoring area layout of the wind power installation vessel provided in this embodiment of the invention; Figure 4 This is a flowchart of the rigid body motion compensation process for lidar point cloud under the coupling effect of complex sea conditions and installation operations, provided in an embodiment of the present invention. Figure 5This is a schematic diagram of the construction of a three-dimensional Gaussian structure field with structural semantic tags and the registration of a reference structure field provided in an embodiment of the present invention; Figure 6 This is a flowchart of structural anomaly identification and degradation feature extraction based on Gaussian parameter deviation provided in an embodiment of the present invention; Figure 7 This is a schematic diagram of structural health risk prediction, risk heat map generation, and maintenance decision output for the wind power installation vessel operation phase provided by an embodiment of the present invention.

[0022] Figure 8 This is a schematic diagram of the experimental verification process provided in the embodiments of the present invention; Figure 9 This is a schematic diagram comparing the deviation distribution of point cloud rigid body motion compensation before and after, provided in an embodiment of the present invention. Figure 10 This is a schematic diagram of a three-dimensional structural risk heat map generated by the health risk index mapping provided in an embodiment of the present invention. Detailed Implementation

[0023] The present invention will be further described in detail below with reference to the accompanying drawings.

[0024] Example 1 Reference Figure 1 This embodiment presents a method for predicting the structural health of wind turbine installation vessels based on radar-constrained Gaussian structural fields. This method utilizes lidar to acquire multi-moment point cloud data of the key load-bearing structures of the wind turbine installation vessel, and combines this data with hull attitude, lifting system status, hoisting load, and leg support status for rigid body motion compensation. A three-dimensional Gaussian structural field with structural semantic labels is constructed in a unified hull coordinate system. By comparing the current three-dimensional Gaussian structural field with a reference three-dimensional Gaussian structural field, Gaussian parameter deviations and structural degradation characteristics are extracted. Furthermore, the method combines the operational stage and load conditions of the wind turbine installation vessel to predict health risks in key areas, and generates a three-dimensional risk heatmap and maintenance decision recommendations.

[0025] like Figure 2 As shown in the figure, this embodiment provides a method and system for predicting the structural health of wind turbine installation vessels based on radar-constrained Gaussian structure fields, which specifically includes the following steps.

[0026] Step 1: Multi-source heterogeneous data acquisition and time synchronization During offshore operations, wind turbine installation vessels are simultaneously subjected to the coupled effects of wind, waves, currents, lifting loads, pile leg reactions, and the movement of the lifting system. To avoid biased structural condition assessments due to reliance on a single sensor, this step involves simultaneously acquiring structural point cloud data and operational status data.

[0027] Step 1.1: LiDAR point cloud acquisition like Figure 3As shown, lidar is deployed on the deck, near the pile legs, near the lifting system, near the crane base, or on the fixed support components of the wind turbine installation vessel. The lidar periodically scans the key load-bearing structures of the wind turbine installation vessel to acquire three-dimensional point cloud data of the key structural areas.

[0028] The key load-bearing structures include, but are not limited to, pile legs, pile leg guide frames, lifting racks, lifting drive mechanisms, locking mechanisms, crane bases, crane slewing bearing areas, deck main beams, reinforcing ribs, support nodes, weld areas, equipment mounting surfaces, and clearance areas for hoisting operations.

[0029] Let the original lidar point cloud at the k-th scan time be: , in, Indicates the first A set of point clouds in the lidar coordinate system at each scanning moment; Indicates the first Point cloud points; , , Represents the three-dimensional coordinates of points in a point cloud; This indicates the lidar reflection intensity at that point; Indicates the first The number of point clouds collected at each scan time.

[0030] Step 1.2: Synchronous Collection of Job Status Data Simultaneously collect data on hull attitude, lifting operation, hoisting load, leg support status, and operation stage.

[0031] The data includes: ship attitude data such as roll angle, pitch angle, bow angle, angular velocity, linear acceleration, or six-degree-of-freedom hull attitude; lifting operation data such as lifting height, lifting speed, locking status, and lifting system displacement; lifting load data such as crane load, crane slewing angle, boom amplitude, and lifting conditions; leg support status data such as leg pressure, leg verticality, leg insertion depth, and support reaction force; and operation phase data such as navigation phase, positioning phase, leg lowering phase, jacking phase, lifting phase, descent phase, and evacuation phase.

[0032] The first The state vector of the multi-source job at each time step is denoted as: , in, Indicates the first The job state vector at each moment; This represents the ship's attitude data; This indicates lifting operation data; This indicates the lifting load data; This indicates the data on the support status of the pile legs; Indicates the task stage label.

[0033] Step 1.3: Multi-source data time synchronization Because the data sampling frequencies of the lidar, IMU, lifting system, crane control system, and pile leg pressure sensor differ, this embodiment uses a unified timestamp mechanism for synchronization. For the lidar scanning time... Select the operation status data at the corresponding time by choosing the one with the smallest time difference or by interpolation.

[0034] Synchronization can be represented as: , in, Indicates the relationship with the first The operational state vector after strict alignment of the frame lidar point cloud; This represents a time synchronization function; This represents multi-source state data over continuous time.

[0035] Step 2: Point cloud rigid body motion compensation under the coupling effect of complex sea conditions and installation operations Wind turbine installation vessels experience overall swaying, localized vibrations, and platform attitude changes under the influence of wind, waves, and lifting loads. Directly comparing point clouds from different times can lead to misinterpretation of the vessel's overall rigid body motion as structural deformation. Therefore, this step compensates for the rigid body motion of the lidar point cloud using vessel attitude data and operational status data, unifying the point cloud into the vessel's structural coordinate system. Figure 4 As shown.

[0036] Step 2.1: Coordinate system establishment and external parameter calibration Establish a world coordinate system hull coordinate system LiDAR coordinate system and the local coordinate system of the key structure .in, Fixed to the hull of the wind turbine installation vessel, used to characterize the structural health status; Fixed to the lidar sensor; Fixed to local monitoring areas such as pile legs, lifting mechanisms, crane bases, or deck main beams.

[0037] The extrinsic parameter matrix from the lidar coordinate system to the ship's coordinate system is expressed as: , in, This represents the homogeneous transformation matrix from the lidar coordinate system to the ship's coordinate system; Represents the rotation matrix; This represents the translation vector.

[0038] Step 2.2: Estimation of Rigid Body Motion of the Hull Based on IMU, GNSS, ship attitude sensors, lifting system status, and crane load data, estimate the... The overall rigid body motion of the ship at a given moment relative to the reference moment can be expressed as: , in, Indicates the first The overall rigid body motion matrix of the ship at each moment; Represents the rotation matrix caused by roll, pitch, and bow. This represents the translation vector caused by waves, lifting operations, or overall displacement.

[0039] Step 2.3: Point Cloud Rigid Body Motion Compensation Original lidar point cloud The coordinate system is transformed from the lidar coordinate system to a unified ship coordinate system, eliminating the influence of the overall rigid body motion of the ship. For any point... The compensated points The calculation is as follows: , in, This represents the point cloud points located in the ship's coordinate system after rigid body motion compensation; Used to counteract the overall motion of the ship; Used to convert lidar point clouds to the ship's coordinate system.

[0040] The compensated point cloud set is denoted as: , Through the above processing, the system removes non-structural displacements caused by sea state and overall attitude changes from the point cloud, providing stable input for subsequent construction of the three-dimensional Gaussian structure field and identification of structural deviations.

[0041] Step 3: Semantic segmentation of key structural areas of the wind turbine installation vessel After rigid body motion compensation, the lidar point cloud has been unified to the ship's coordinate system. Because the key load-bearing structures of the wind turbine installation vessel have significant spatial zoning characteristics and functional load-bearing differences, directly constructing a three-dimensional Gaussian structural field using all compensated point clouds as a unified whole would easily weaken the functional and risk differences between different structural regions, resulting in a lack of clear structural semantic constraints for subsequent structural deviation calculations, anomaly identification, and health risk prediction. Therefore, this step, based on the wind turbine installation vessel's structural topology, design model, and compensated spatial point cloud, divides the key load-bearing structures into semantic regions and assigns structural semantic labels and risk weights to each region. This ensures that the subsequent Gaussian structural field not only has spatial geometric expression capabilities but also structural semantic attributes oriented towards health monitoring.

[0042] Step 3.1: Key Region Partitioning Based on Structural Topology Obtain the design model, structural drawings, finite element model, or digital twin model of the wind turbine installation vessel, extract the spatial boundaries, connection relationships, and functional attributes of the key load-bearing structures of the wind turbine installation vessel, and establish the topology of the key load-bearing structures. The topology of the key load-bearing structures is used to describe the spatial adjacency relationships, force transmission relationships, and functional subordination relationships between different structural regions, including but not limited to the pile legs, pile leg guide frames, lifting racks, lifting drive mechanisms, locking mechanisms, crane bases, crane slewing bearing areas, deck main beams, transverse reinforcing beams, longitudinal reinforcing ribs, connection nodes, weld neighborhoods, equipment mounting surfaces, and the clearance area for lifting operations.

[0043] Specifically, the key structural topology of the wind turbine installation vessel is represented as follows: , in, This indicates the topology of the key load-bearing structure of the wind turbine installation vessel; Represents the set of nodes in the key structural region; Represents the set of connecting edges between structural regions; This represents a set of structural region attributes, which include structural type, geometric boundaries, load-bearing function, operational stage, failure consequences, and historical anomaly information.

[0044] For the first in the topology For each key structural region, its spatial enclosing domain is determined based on the design model or digital twin model. The spatial bounding region can be a cuboid bounding box, a cylindrical bounding region, a curved surface bounding region, a neighborhood bounding region based on a mesh surface, or a composite bounding region formed by combining multiple basic geometric shapes. For the compensated point cloud... any point in Calculate its affiliation with the spatial enclosing domain of each key structural region. When point Located in the region Inside, or to the area The minimum distance is less than the preset distance threshold At that time, this point was initially assigned to the first... Key structural regions.

[0045] Point cloud attribution relationships can be represented as: , in, Indicates the first The first scan time A set of compensation point clouds corresponding to each key structural region; Point To the The spatial enclosed domain of the structural region The minimum distance; This represents the regional affiliation distance threshold.

[0046] Therefore, the compensated point cloud Divided into Key structural regions: , in, Indicates the first The first scan time Point cloud sets corresponding to key monitoring areas; This indicates the number of key monitoring areas.

[0047] During the partitioning process, a smaller spatial partitioning granularity can be used for areas with complex local geometry, significant stress concentration, or severe failure consequences, such as crane bases, lifting racks, locking mechanisms, connection nodes, and weld neighborhoods. A larger spatial partitioning granularity can be used for areas on ordinary deck planes, low-risk support surfaces, or areas far from load transfer paths. This differentiated partitioning approach can improve the monitoring resolution of key sensitive areas while ensuring overall computational efficiency.

[0048] Step 3.2: Structural Region Boundary Correction and Overlapping Point Assignment Due to issues such as occlusion, noise, sparse sampling, and boundary aliasing in lidar point clouds, relying solely on the spatial bounding domain for initial segmentation may lead to incorrect allocation of some boundary points. To improve the stability of region segmentation, this embodiment further refines the initial segmentation results by incorporating local normals, curvature, neighborhood continuity, and topological adjacency relationships.

[0049] For point cloud points that simultaneously fall within two or more candidate bounding regions of structural regions, construct a region attribution scoring function: , in, Point Belonging to the The score value of each structural region; This indicates the distance from a point to the boundary or center plane of the structural region; This indicates the curvature consistency or surface continuity deviation between a point and its neighboring point cloud. This indicates the deviation between the local normal of the point cloud and the normal of the designed structural surface; This represents a structural topology constraint term, used to describe the consistency of the point with the candidate region in terms of structural connectivity. , , and This represents the weighting coefficients for each item.

[0050] Point The structural region with the lowest score is assigned to: , in, point The final key structural region number.

[0051] For areas with missing or severely occluded point clouds, completion can be determined based on adjacent scan frames, design model boundaries, or historical stable point clouds of the same structural region. Isolated noise points that are significantly deviated from the structural surface and do not satisfy any regional topological constraints can be marked as invalid points or temporary anomaly candidates, and will not directly participate in the subsequent initialization of the 3D Gaussian structural field, in order to reduce the impact of noise on the structural health assessment.

[0052] Step 3.3: Assigning structural semantic tags After completing the point cloud segmentation of key structural regions, a structural semantic label is assigned to each key monitoring region. The structural semantic label is used to characterize the functional location, structural category, and health monitoring significance of the region within the wind turbine installation vessel, enabling subsequent 3D Gaussian elements to carry explicit structural attributes.

[0053] The set of structural semantic tags can be represented as: , in, Indicates the first Structural semantic labels for key monitoring areas.

[0054] The structural semantic tags include, but are not limited to, the leg area, the leg guide frame area, the lifting rack area, the lifting drive mechanism area, the locking mechanism area, the crane base area, the crane slewing bearing area, the deck main beam area, the reinforcing rib area, the connection node area, the weld neighborhood area, the equipment mounting surface area, and the hoisting clearance area.

[0055] To enhance the hierarchical nature of structural semantic expression, this embodiment can also set structural semantic tags as multi-level tags. For example, the first-level tags are used to describe the major structural categories, including support structures, lifting structures, hoisting load-bearing structures, deck load-bearing structures, and connection structures; the second-level tags are used to describe specific component types, including pile legs, racks, locking mechanisms, crane bases, main beams, stiffeners, and welds; the third-level tags are used to describe local monitoring locations, including the root, end, transition zone, connection hole neighborhood, weld heat-affected zone, and high stress concentration zone.

[0056] By setting multi-level semantic tags, each Gaussian element in the subsequent Gaussian structure field can correspond to a clear structural region and functional meaning. When Gaussian center drift, covariance variation, principal axis direction change, or reflection intensity anomaly is subsequently identified, the system can directly determine which structural part the anomaly occurs in, thereby improving the interpretability of anomalies and the targeted nature of maintenance decisions.

[0057] Step 3.4: Setting Regional Risk Weights After assigning structural semantic labels, risk weights are set for each structural region based on its load-bearing capacity, failure consequences, load transfer paths, historical anomaly records, finite element stress concentration results, and on-site maintenance experience during wind power installation operations. A higher risk weight indicates a greater impact on the safety of wind power installation vessel operations, structural stability, or maintenance decisions should an anomaly occur in that region.

[0058] The risk weight set is represented as: , in, Indicates the first Risk weights for key structural regions.

[0059] In one alternative implementation, the risk weight can be determined jointly by structural importance, stress concentration level, frequency of historical anomalies, and failure consequence level: , in, Indicates the first The importance score of each structural region is used to characterize the functional importance of that region in hoisting, lifting, supporting or load transfer; The stress concentration score can be determined by finite element analysis, design specifications, or historical test results. This represents the historical anomaly frequency score, used to characterize historical defects, maintenance records, or abnormal alarm situations in this area; The failure consequence score is used to characterize the potential work interruption, structural instability, hoisting risks, or maintenance costs caused by anomalies in the area. , , and This represents the weighting coefficient.

[0060] To facilitate subsequent health risk prediction and risk heatmap generation, the risk weights can be normalized: , in, This represents the normalized regional risk weight, with a value ranging from 0 to 1.

[0061] For example, crane bases, crane slewing bearing areas, pile leg guide frames, lifting racks, locking mechanisms, deck main beams, and critical connection nodes are typically assigned higher risk weights; areas on ordinary deck planes, non-primary load-bearing ancillary structures, or areas far from the load transfer path can be assigned lower risk weights. Through these risk weight settings, the system can give greater attention to load-bearing sensitive areas and high-consequence areas when constructing subsequent anomaly indicators and predicting health risks.

[0062] Step 3.5: Output of semantic segmentation results After the above processing, we obtain a set of point clouds of key structural regions with structural semantic labels and risk weights: , in, Indicates the first Structural semantic partitioning results at each scan time; Indicates the first A set of compensation point clouds for key structural regions; Indicates the first Structural semantic tags for key structural regions; Indicates the first Risk weights for key structural regions.

[0063] The structural semantic partitioning results serve as input for the subsequent construction of the 3D Gaussian structural field. For each key structural region, subsequent steps will initialize 3D Gaussian cells and write the corresponding structural semantic labels and risk weights into the Gaussian cell attributes. Thus, the 3D Gaussian structural field can not only express the continuous spatial morphology of the structural surface but also preserve the structural functional semantics and risk priorities, providing a unified data foundation for subsequent baseline structural field registration, structural deviation field construction, anomaly identification, health risk prediction, and maintenance decisions.

[0064] Step 4: Construction of 3D Gaussian Structure Field with Structural Semantic Tags like Figure 5As shown, traditional lidar point clouds are essentially discrete sets of sampling points, reflecting only the spatial coordinates and reflection intensity of a structural surface at a few scanning positions. They struggle to directly express the continuous morphology, local geometric orientation, spatial scale variations, and long-term degradation trends of the structural surface. For large marine engineering equipment like wind turbine installation vessels, structural anomalies are often not isolated changes at a single measuring point, but rather manifest as center drift, local morphological expansion, changes in the main orientation, attenuation of reflection intensity, or expansion of the anomaly region within a certain spatial range. Therefore, relying solely on point-by-point comparisons based on discrete point clouds is susceptible to the effects of sampling density, occlusion, noise, and changes in viewing angle, leading to unstable representations of the structural state.

[0065] Therefore, this step transforms the semantically segmented point cloud of key structural regions into a 3D Gaussian structure field with structural semantic labels. Each 3D Gaussian cell represents a continuous spatial neighborhood of the local structural surface, rather than just an isolated point. This Gaussian cell simultaneously includes attributes such as center location, covariance matrix, transparency parameter, lidar reflection intensity, structural semantic label, and regional risk weight. Thus, the key structure of the wind turbine installation vessel can be transformed from a discrete point cloud representation into a parameterized, continuous, comparable, and updatable structural state representation.

[0066] By constructing a 3D Gaussian structure field, the system can obtain the scale, orientation, distribution range, and reflection characteristics of local structural surfaces while preserving the spatial geometric information of the point cloud. When the structure undergoes deformation, misalignment, settlement, local bending, loosening of connection nodes, or surface degradation, the Gaussian center, covariance, principal axis direction, or reflection intensity of the corresponding region will exhibit measurable changes. Therefore, the 3D Gaussian structure field is not only used for structural visualization but also for subsequent reference structure field registration, structural deviation field construction, anomaly type identification, and health risk prediction.

[0067] Step 4.1: Initialization of 3D Gaussian Elements For the The first scan time Three key monitoring areas are defined, and their compensated regional point clouds are set as follows: Based on the spatial distribution, sampling density, and structural semantic attributes of the point cloud in this region, Initialize it as a number of three-dimensional Gaussian elements. Each three-dimensional Gaussian element corresponds to a continuous geometric neighborhood of the local structure surface, which is used to approximate the spatial distribution of the point cloud within that neighborhood.

[0068] No. The three-dimensional Gaussian unit set of each key monitoring area is represented as follows: , in, Indicates the first The first scan time A set of three-dimensional Gaussian elements for each structural region; Indicates the first The first structural region One three-dimensional Gaussian unit; This indicates the number of three-dimensional Gaussian elements within the current structural region.

[0069] Each three-dimensional Gaussian element is defined as: , in, This indicates the center position of a three-dimensional Gaussian cell, used to describe the spatial location of the local structural neighborhood; The covariance matrix is ​​used to describe the spatial scale, expansion direction, and morphological distribution of the local structural neighborhood. This represents the transparency parameter or validity weight, used to characterize the contribution of the Gaussian unit to the structural field representation; This indicates the lidar reflection intensity attribute corresponding to the Gaussian unit; The semantic label representing the structural region to which the Gaussian unit belongs; This indicates the regional risk weight inherited by the Gaussian cell.

[0070] The three-dimensional Gaussian distribution function can be expressed as: , in, Indicates the first The scan time, the first The first key structural region A three-dimensional Gaussian unit at a spatial location The resulting structural field response values; Represents the coordinates of any position in three-dimensional space; Indicates the first The scan time, the first The first key structural region The transparency parameter or effectiveness weight of a three-dimensional Gaussian element is used to adjust the contribution of the Gaussian element to the structural field response. This represents the center position vector of the three-dimensional Gaussian element; The covariance matrix represents the three-dimensional Gaussian element, used to describe the scale and orientation distribution of the Gaussian element in three-dimensional space; The inverse matrix of the covariance matrix; This represents the transpose operation of a matrix or vector. This represents the natural exponential function.

[0071] During initialization, the number of 3D Gaussian elements can be determined based on the point cloud sampling density and the importance of the structural region. For structurally sensitive areas such as crane bases, lifting racks, locking mechanisms, connection nodes, weld neighborhoods, and the ends of deck main beams, smaller Gaussian coverage radii and denser Gaussian elements can be set to enhance the ability to identify local anomalies. For flat deck surfaces, ordinary support surfaces, or low-risk areas, larger Gaussian coverage radii can be set to reduce computational load and improve the stability of the structural field representation.

[0072] Step 4.2: Determination of Gaussian element center, scale, and orientation parameters To ensure that the three-dimensional Gaussian element can accurately fit the local geometry of the wind turbine installation vessel's surface, this embodiment determines the center position, scale parameters, and rotation direction of the Gaussian element based on the center, density, curvature, and normal of the local point cloud neighborhood.

[0073] For the Gaussian units are first used in the region point cloud. Select its local neighborhood point cloud This neighborhood can be determined based on a fixed radius, the number of nearest neighbors, or the adjacency relationship of the structural surface mesh. The point cloud center of this neighborhood is then calculated: , in, Indicates the first The local point cloud neighborhood corresponding to each Gaussian unit; This indicates the number of point clouds within that neighborhood; Represents any point within the neighborhood; This indicates the spatial center of the point cloud in that neighborhood.

[0074] Further calculation of the local point cloud covariance matrix: , in, Indicates the first The covariance matrix of the local point cloud corresponding to each Gaussian cell. This matrix reflects the degree of dispersion of the local point cloud in different spatial directions. If the dispersion in a certain direction is large, it indicates that the local structure extends more significantly in that direction; if the dispersion in a certain direction is small, it indicates that the direction may be close to the normal direction of the structure surface.

[0075] Eigenvalue decomposition of the local covariance matrix: , in, Represents the local principal direction matrix; This represents the diagonal matrix of eigenvalues. Eigenvectors are used to characterize the principal, secondary, and normal directions of the local structural surface, while eigenvalues ​​are used to characterize the spatial expansion scale of the point cloud along the corresponding directions.

[0076] The covariance matrix of the three-dimensional Gaussian element can be determined based on the eigenvalues ​​and eigenvectors: , in, , and They represent the first The Gaussian element has three principal directions for its scale parameters. For approximately planar structures such as plates, beams, deck surfaces, and support surfaces, the scale along the structural surface can be made larger than the normal scale so that the Gaussian element is flattened and fits the structural surface. For locally complex structures such as bolt holes, welds, edges, and connection nodes, the Gaussian scale can be appropriately reduced according to the local curvature to enhance the ability to express local geometric abrupt changes.

[0077] Step 4.3: Assigning Gaussian Element Attributes and Semantic Inheritance After determining the spatial parameters of the Gaussian cell, each Gaussian cell is assigned a reflection intensity attribute, a transparency parameter, a structural semantic label, and a risk weight. The reflection intensity attribute can be determined by the average reflection intensity, maximum reflection intensity, or weighted reflection intensity of its local neighborhood point cloud. , in, Indicates the first The reflection intensity properties of a Gaussian unit; Point The reflectance intensity of the lidar. The reflectance intensity attribute can be used to help identify non-pure geometric anomalies such as surface degradation, suspected corrosion areas, coating changes, or localized defects.

[0078] Transparency parameter The transparency parameter can be determined based on local point cloud density, observation reliability, and occlusion levels. For regions with higher point cloud density and more stable observations, the transparency parameter can be set higher; conversely, for regions with sparse point clouds, severe occlusion, or unstable observations, the transparency parameter can be set lower. This reduces the interference of low-quality point clouds on the representation of the structure field.

[0079] Due to the The Gaussian unit is composed of the first... The point cloud of the key structural region is initialized, therefore the Gaussian unit inherits from the first... Semantic tags for each structural region and risk weight : , in, Represents Gaussian unit Structural semantic tags; Represents Gaussian unit The corresponding normalized risk weights; Indicates the first Structural semantic tags for key structural regions; Indicates the first Normalized risk weights for key structural regions.

[0080] Through this attribute inheritance mechanism, each 3D Gaussian element is no longer just a geometric representation unit, but also carries structural functional meaning and risk priority. For example, the Gaussian element located at the crane base will inherit the semantic label of "crane base region" and a higher risk weight; the Gaussian element located in the weld neighborhood will inherit the semantic label of "weld neighborhood region" and will focus on local continuity abrupt changes and reflection intensity changes in subsequent anomaly identification.

[0081] Step 4.4: Generation of 3D Gaussian Structure Field After initializing and assigning properties to Gaussian elements in all key structural regions, the three-dimensional Gaussian elements of each region are combined to obtain the first Gaussian element. Three-dimensional Gaussian structure field of the wind turbine installation vessel at each scan time: , in, Indicates the first The current three-dimensional Gaussian structure field at each scan time; Indicates the first The set of three-dimensional Gaussian elements corresponding to each key structural region.

[0082] The current three-dimensional Gaussian structure field can also be represented as the joint response field of all Gaussian elements: , in, Indicates spatial location The three-dimensional Gaussian structure field response at the location; Indicates the first Within the structural region, the first A Gaussian unit at position The response generated at that location.

[0083] The resulting three-dimensional Gaussian structural field has the following characteristics: First, it expresses the local location of the structural surface through the Gaussian center, which can be used to identify component deformation, misalignment, and settlement. Second, it expresses the local structural scale and orientation through the covariance matrix, which can be used to identify local bending, bulges, depressions, edge damage, and geometric continuity anomalies of connection nodes. Third, it expresses the echo characteristics of the structural surface through the reflection intensity attribute, which can be used to assist in identifying suspected areas of surface degradation or local defects. Fourth, it preserves the functional attributes of components through structural semantic labels and risk weights, enabling subsequent anomaly identification results to directly locate specific structural parts and serve risk classification and maintenance decisions.

[0084] Therefore, the 3D Gaussian structure field generated in this step is not only a smooth representation of the point cloud, but also a parameterized state field for structural health prediction. This state field provides a unified, stable, and interpretable data foundation for subsequent baseline structure field registration, Gaussian parameter deviation calculation, structural anomaly identification, degradation feature extraction, and health risk prediction.

[0085] Step 5: Establishment and registration of the benchmark 3D Gaussian structure field To determine whether the current structural state of the wind turbine installation vessel has undergone abnormal changes, it is necessary to compare the three-dimensional Gaussian structural field constructed at the current scanning moment with the baseline three-dimensional Gaussian structural field under healthy conditions. Unlike traditional point cloud registration, which only compares the distance between discrete points, this embodiment uses a three-dimensional Gaussian structural field with structural semantic labels as the comparison object, measuring structural state changes from multiple dimensions such as Gaussian center position, covariance shape, principal axis direction, reflection intensity, and structural semantic region. This allows structural anomalies to be extended from a single geometric deviation to an interpretable parametric deviation expression, providing a foundation for subsequent anomaly type determination and health risk prediction.

[0086] The baseline 3D Gaussian structural field can be derived from the standard structural field converted from the design CAD model, the initial structural field obtained from the first health scan, the restored structural field obtained from the post-maintenance scan, or a healthy baseline structural field selected from historical stable operation phases. The baseline 3D Gaussian structural field is used to characterize the spatial morphology, local scale, principal direction, and reflection intensity characteristics of the key structures of the wind turbine installation vessel within the healthy state or allowable deviation range. The difference between the current 3D Gaussian structural field and the baseline 3D Gaussian structural field can serve as an important basis for judging suspected areas or local abnormal expansions of structural deformation, misalignment, settlement, loose connections, surface degradation, etc.

[0087] Because the geometry, load-bearing capacity, and allowable deviations of different structural regions on a wind turbine installation vessel vary, this embodiment does not employ global, undifferentiated registration during the registration process. Instead, it uses Gaussian element matching within the same semantic region. The Gaussian elements of the current crane base region are compared only with those of the reference crane base region, and the Gaussian elements of the current deck main beam region are compared only with those of the reference deck main beam region. This semantic constraint avoids mismatches between different structural regions, improving the physical consistency and interpretability of structural deviation calculations.

[0088] Step 5.1: Establishment of the benchmark three-dimensional Gaussian structure field First, a baseline three-dimensional Gaussian structural field under healthy conditions is established. The baseline structural field can be obtained in any of the following ways: First, generate a standard structural point cloud based on the CAD model or digital twin model of the wind turbine installation vessel, and convert it into a baseline three-dimensional Gaussian structural field according to the methods in steps 3 and 4; Second, perform a lidar scan when the wind turbine installation vessel is first put into use, after maintenance is completed, or when it is confirmed to be in a healthy state, and construct the baseline three-dimensional Gaussian structural field from the healthy scan point cloud; Third, select a time period from historical multiple scan data where the structural state is stable, there are no abnormal alarms, and the load conditions are relatively stable, and obtain the healthy baseline structural field through multi-frame fusion.

[0089] The baseline three-dimensional Gaussian structure field is represented as: , in, Represents the reference three-dimensional Gaussian structure field; Indicates the first The set of reference Gaussian elements corresponding to each key structural region; Indicates the number of critical structural regions.

[0090] No. The set of reference Gaussian elements for each key structural region is represented as follows: , in, Indicates the first The first critical structural region One reference Gaussian element; Indicates the first The number of reference Gaussian elements in each key structural region.

[0091] The reference Gaussian element can be represented as: , in, Indicates the center position of the reference Gaussian element; Represents the covariance matrix of the reference Gaussian element; This represents the transparency parameter or validity weight of the reference Gaussian cell; This represents the reflection intensity properties of the reference Gaussian unit; The semantic label represents the structural region to which the reference Gaussian unit belongs; This represents the risk weight of the structural region.

[0092] In practical applications, to improve the stability of the reference structural field, multiple frames of health scan data can be fused. For Gaussian parameters obtained from multiple scans within the same structural region, stable reference parameters can be obtained using the mean, weighted mean, median filtering, or confidence weighting methods. This can reduce the impact of single-scan noise, accidental occlusion, and local reflection anomalies on the reference field.

[0093] Step 5.2: Registration of semantically related regions between the current structural field and the reference structural field In obtaining the current three-dimensional Gaussian structure field and the benchmark three-dimensional Gaussian structure field Next, it is necessary to establish the correspondence between the two. Since both the current structural field and the reference structural field carry structural semantic labels, this embodiment prioritizes matching within the same structural semantic region.

[0094] For the The current three-dimensional Gaussian structure field at the nth scan time, its i-th The key structural regions can be represented as: , in, Indicates the first The first scan time A set of three-dimensional Gaussian elements for each structural region; Indicates the first The first structural region The current Gaussian unit.

[0095] For the current Gaussian unit Only when they have the same semantic tags The set of reference Gaussian elements The best-matching reference Gaussian element is found within the range. The matching relationship can be expressed as: , in, Indicates the relationship with the current Gaussian unit. The most matching reference Gaussian element number; This represents the distance metric function between Gaussian cells.

[0096] This distance metric function considers not only the differences in the Gaussian center positions, but also the differences in covariance shape, principal direction, and reflection intensity, and can be expressed as: , in, This represents the synthesized distance between the current 3D Gaussian element and the reference 3D Gaussian element. Indicates the current scan time. Within the structural region, the first One Gaussian unit, Indicates the first [item] under the health baseline state Within the structural region, the first One reference Gaussian element, and These represent the principal direction matrices of the current Gaussian element and the reference Gaussian element, respectively. and Let these represent the covariance matrices of the current Gaussian cell and the reference Gaussian cell, respectively, used to describe the spatial scale and directional distribution characteristics of the area covered by the Gaussian cell. and These represent the reflection intensity properties of the current Gaussian cell and the reference Gaussian cell, respectively. and These represent the principal direction matrices obtained from the eigendecomposition of the covariance matrices of the current Gaussian element and the reference Gaussian element, respectively, where the subscript 0 indicates the reference state. , , and These represent the weighting coefficients corresponding to the differences in Gaussian center location, covariance structure, reflection intensity, and principal direction, respectively. Represents the L2 norm; Denotes the Frobenius norm. This represents the main direction difference measurement function.

[0097] Through the above matching method, the system can establish the correspondence between the current Gaussian element and the reference Gaussian element within the same structural functional area, avoiding cross-matching between different structural areas such as crane base, deck main beam, weld neighborhood and connection node.

[0098] Step 5.3: Calculation of Gaussian parameter deviation After matching the current Gaussian element with the reference Gaussian element, the Gaussian parameter deviation between them is calculated. For the current Gaussian element... and its matching reference Gaussian element Its center drift is expressed as: , in, Indicates the first The first scan time Within the structural region, the first The center drift of a Gaussian element. This index is used to describe local structural positional changes. When the center drift continuously increases in a certain area, it usually corresponds to anomalies such as component deformation, misalignment, settlement, tilting, or loose connections.

[0099] The change in covariance is expressed as: , in, This represents the change in Gaussian covariance. This index is used to describe the scale and morphological changes in the neighborhood of a local structure. When the covariance changes significantly, it usually indicates that the local structure surface has bending, bulges, depressions, edge damage, or abnormal point cloud distribution patterns.

[0100] The change in reflection intensity is expressed as: , in, This indicates the change in reflection intensity. This index is used to describe changes in the echo characteristics of a structural surface and can help identify areas suspected of surface degradation, coating changes, suspected corrosion areas, or localized defects.

[0101] If the covariance matrix is ​​decomposed into scaled and rotated forms: , in, and These represent the principal direction matrices of the current Gaussian element and the reference Gaussian element, respectively. and These represent the scale matrices of the current Gaussian cell and the reference Gaussian cell, respectively.

[0102] The angle of change of the principal axis direction can then be expressed as: , in, Indicates the angle of change of the principal axis direction of the Gaussian; This represents the matrix trace operation. This index is used to describe changes in the orientation of local structures. When the principal axis direction changes significantly, it usually corresponds to component torsion, bending, local tilting, changes in edge orientation, or abnormal geometric continuity of connection nodes.

[0103] Therefore, each current Gaussian element can obtain a set of parameter deviations: , in, Indicates the first The first scan time Within the structural region, the first A comprehensive description of the deviation of a Gaussian element relative to the reference state.

[0104] Step 5.4: Construction of Structural Deviation Field The parameter deviation of a single Gaussian element can only reflect the state changes in the local structural neighborhood, while structural anomalies of wind turbine installation vessels typically exhibit spatial continuity and regional expansion. Therefore, this embodiment further spatially aggregates the parameter deviations of multiple Gaussian elements within the same structural region to construct a structural deviation field.

[0105] No. The structural region in the first The structural deviation field at each scanning time can be expressed as: , in, Indicates the first The first scan time Structural deviation field of each structural region.

[0106] To further describe the degree of regional structural anomalies, weighted aggregation of Gaussian cell deviations within the region can be performed: , in, Indicates the first The first scan time Regional-level deviation intensity of each structural region; , , and The weighting coefficients represent the deviations of various Gaussian parameters.

[0107] Further combine regional risk weights The risk-weighted bias strength can be obtained as follows: , in, This indicates the intensity of deviation after incorporating structural area risk weights. This indicator allows high-risk areas such as crane bases, lifting racks, locking mechanisms, deck main beams, and critical connection nodes to receive greater attention in subsequent anomaly identification and health prediction.

[0108] The overall structural deviation field is represented as: , in, Indicates the first The overall critical structural deviation field of the wind turbine installation vessel at each scanning moment; Indicates the first Structural deviation field of each structural region; Indicates the number of critical structural regions.

[0109] This structural deviation field not only preserves the deviation information of each local Gaussian unit, but also retains its corresponding structural semantic label and risk weight. Therefore, subsequent steps can further identify structural anomaly types, calculate comprehensive anomaly indicators, and predict health risk levels based on deviation type, deviation amplitude, deviation spatial clustering degree, and structural region importance.

[0110] Step 5.5: Verification of the reliability of the registration results To avoid interference from point cloud occlusion, local missing data, or incorrect matching in the judgment of structural anomalies, this embodiment can also verify the reliability of the registration results between the current structural field and the reference structural field. For each structural region, the proportion of successfully matched Gaussian elements, the mean matching distance, the variance of the matching distance, and the number of unmatched Gaussian elements are statistically analyzed.

[0111] The reliability of region registration can be expressed as: , in, This indicates the registration confidence level of the m-th structural region at the k-th scan time. This indicates the proportion of successfully matched Gaussian cells within the region; This represents the average matching distance of Gaussian cells within this region; Indicates the variance of the matching distance; , and This represents the weighting coefficient.

[0112] When the registration confidence level of a certain region is lower than a preset threshold, the system can mark that region as a low-confidence region and trigger a rescan, manual review, or reduce the weight of the deviation calculation for that region. By verifying the registration confidence level, the robustness of structural anomaly identification can be improved, and the risk of false alarms caused by occlusion, noise, or sparse sampling can be reduced.

[0113] Through the above steps, this embodiment completes the semantic region registration and Gaussian parameter deviation calculation between the current 3D Gaussian structure field and the reference 3D Gaussian structure field. Compared with traditional point-by-point distance comparison, this method can characterize the differences in structural state from multiple perspectives, such as positional drift, local morphological changes, orientation changes, and reflection intensity changes. It also binds the deviation information with structural semantic labels and risk weights, thereby providing a more stable and interpretable input for subsequent structural anomaly identification, degradation feature extraction, and health risk prediction.

[0114] Step 6: Structural anomaly identification based on Gaussian parameter deviation like Figure 6 As shown, this step identifies critical structural anomalies on the wind turbine installation vessel based on the structural deviation field. Unlike traditional single-point sensors, this embodiment can identify continuous spatial anomalies based on three-dimensional Gaussian parameter variations.

[0115] Step 6.1: Construction of Structural Anomaly Indicators For the Construct comprehensive anomaly indicators for key structural regions. : , in, Indicates the first Comprehensive abnormal indicators for key structural regions; This indicates the risk weight of the region; This represents the mean Gaussian center drift within the region; This represents the mean change in covariance within the region; This represents the average change in the principal axis direction within the region; This represents the average value of the variation in reflection intensity; Indicates the spatial clustering degree of anomalous Gaussian cells; to This represents the weighting coefficient.

[0116] The angle of change of the principal axis direction can be calculated by the following formula: , in, Represents the matrix trace operation; Indicates the first The principal axis direction variation angle of a Gaussian element Indicates the first The first scan time Within the structural region, the first The relative principal direction rotation matrix between the current Gaussian element and its matched reference Gaussian element.

[0117] Step 6.2: Determining the Type of Structural Anomaly The structural anomaly type is determined based on the dominant characteristics of different Gaussian parameter deviations.

[0118] When the Gaussian center drift increases continuously in a local area, it is judged as abnormal component deformation, misalignment, settlement, or tilting; when the covariance scale changes significantly and the local principal axis direction changes, it is judged as abnormal local depression, bulge, bending, torsion, or edge damage; when the relative distance between adjacent structural areas changes continuously, it is judged as gap change, coaxiality deviation, parallelism deviation, or perpendicularity deviation; when the Gaussian parameter deviation near the weld or connection node shows local concentration, it is judged as abnormal geometric continuity of the connection node or local abrupt change in the weld neighborhood; when the reflection intensity decreases continuously or the local Gaussian density changes abnormally, it is judged as a suspected area of ​​surface degradation, local defects, or edge damage; when the spatial occupancy relationship of the hoisting path or equipment movement area is abnormal, it is judged as insufficient working clearance or potential collision risk.

[0119] Step 6.3: Regional Risk Level Classification Based on comprehensive abnormal indicators Based on preset risk thresholds, structural risks are classified into five levels: normal, attention, warning, restricted operation, and shutdown for maintenance. , in, Indicates the first Risk levels of key structural areas; , , , This indicates a risk threshold determined by material properties, design specifications, historical data, or expert experience.

[0120] Step 7: Extraction of structural degradation features based on Gaussian parameter temporal drift The core of predictive health management is not just identifying current abnormalities, but also determining whether these abnormalities continue to develop throughout the work cycle. For example... Figure 6 As shown, this step performs time-series modeling of the Gaussian parameter deviations at multiple times to extract structural degradation features.

[0121] Step 7.1: Construction of Gaussian parameter deviation sequence For the Each structural region has a construction length of [number]. Gaussian parameter deviation sequence: , in, Indicates the first The structural regions in recent The sequence of structural deviations within a scan time or work cycle Indicates the first The first scan time Structural deviation fields in key structural regions.

[0122] Step 7.2: Construction of Degenerate Feature Vectors Extracting structural degradation feature vectors from Gaussian parameter deviation sequences: , in, Indicates the first The first scan time Degenerate feature vectors of key structural regions; This represents the growth rate of the Gaussian center drift. This represents the rate of change of the covariance scale. Indicates the cumulative rotation in the spindle direction; Indicates the rate of change of reflection intensity; Indicates the area of ​​the abnormal region; Indicates the expansion rate of the abnormal region; Indicates the spatial clustering degree of abnormal regions; This represents the local maximum deviation value; This represents the dynamic response amplification factor.

[0123] The growth rate of the Gaussian center drift can be expressed as: , in, Indicates the first The first scan time Key structural regions in recent Average growth rate of Gaussian center drift within a scan time or job cycle; Indicates the current scan time; Indicates the key structural region number; Indicates the length of the Gaussian parameter deviation sequence; Indicates the starting time of the Gaussian parameter deviation sequence. Mean Gaussian center drift within each key structural region.

[0124] The expansion rate of the abnormal region can be expressed as: , in, Indicates the first The first scan time Key structural regions in recent Average expansion rate of abnormal areas within a scanning time or job cycle; Indicates the first The first scan time Area of ​​abnormal regions in key structural regions; Indicates the starting time of the Gaussian parameter deviation sequence. The area of ​​the abnormal region in each key structural region.

[0125] Step 8: Health Risk Prediction for Wind Power Installation Phase The structural risks of wind turbine installation vessels are strongly correlated with the operational phase. For example, during the jacking phase, the focus is more on the legs, lifting racks, and locking mechanisms; during the hoisting phase, the focus is more on the crane base, crane slewing bearing area, and deck main beams; and during the descent phase, the focus is more on leg recovery, lifting system resetting, and residual structural deviations. This step inputs structural degradation characteristics, along with the operational phase and load conditions, into the health prediction model.

[0126] Step 8.1: Constructing the Health Prediction Model Input The first Degenerative feature vectors of each structural region , Work stage labels and load case vector Combined into the prediction input: , in, This represents the input to the health prediction model; These include operating parameters such as lifting load, crane slewing angle, lifting height, pile leg support force, wind speed, wave height, current velocity, and ship attitude.

[0127] Step 8.2: Calculation of Health Risk Index Will Input the health prediction model to obtain the health risk index for the current and future work cycles: , in, Indicates the first Key structural regions in the future Health risk index after one work cycle; This represents a health prediction model; Indicates model parameters.

[0128] The health prediction model can be a state-space model, a Bayesian update model, a machine learning model, a deep learning model, or a multi-model fusion prediction model. To avoid overly narrow protection, this embodiment does not limit the specific model structure.

[0129] Step 8.3: Estimation of the Remaining Safe Working Window When the health risk index exceeds the risk threshold, the area is considered to be in a high-risk state during the next work cycle. The remaining safe work window can be represented as: , in, Indicates the first Remaining safe operation windows for key structural areas; This represents the critical risk threshold.

[0130] Step 9: Generation of 3D Structural Risk Heat Map and Output of Maintenance Decisions like Figure 7 As shown, this step maps the health prediction results back to a three-dimensional Gaussian structure field, forming an intuitive spatialized risk heat map, and outputs maintenance decision recommendations.

[0131] Step 9.1: Generation of 3D Structural Risk Heat Map By mapping the health risk index of each key monitoring area to the corresponding three-dimensional Gaussian unit, a three-dimensional structural risk heat map is obtained: , in, Indicates the first A three-dimensional structural risk heat map at any given moment; Indicates the first Health risk index of each structural region; Indicates the risk level.

[0132] Step 9.2: Maintenance Priority Calculation Maintenance priority is calculated based on risk level, risk growth trend, and structural area weight: , in, Indicates the first Maintenance priority scoring for key structural areas; This indicates the rate of change in the health risk index; , , This represents the weighting coefficient.

[0133] Step 9.3: Output of Job Control Recommendations Based on maintenance priority and risk level, the system outputs rescanning strategies, maintenance recommendations, and operation control suggestions. When the risk level is normal or concerning, the system maintains the regular scanning frequency and records the structural degradation trend; when the risk level is warning, the system increases the lidar rescanning frequency and prompts for manual re-inspection; when the risk level is restricted operation, the system outputs suggestions to reduce load, decrease lifting speed, or limit the crane's slewing range; when the risk level is shutdown for maintenance, the system outputs suggestions to suspend hoisting, stop jacking, or initiate maintenance procedures.

[0134] Thus, this embodiment forms a complete closed loop of "LiDAR acquisition—rigid body motion compensation—structural semantic segmentation—3D Gaussian structural field construction—reference structural field registration—structural anomaly identification—degradation feature extraction—health risk prediction—risk heatmap and maintenance decision output".

[0135] Experimental verification To verify the feasibility and effectiveness of the method described in this invention under the coupled effects of complex sea conditions and installation operations, this embodiment constructs a scaled-down test platform for the key load-bearing structure of a wind turbine installation vessel and conducts comparative verification using digital twin simulation data. The test objects include key monitoring areas such as the pile leg guide frame, lifting rack, locking mechanism, crane base, deck main beam, and connection nodes. Several controllable displacement loading points are arranged on the structural surface to simulate local settlement, component misalignment, loose connection nodes, and areas suspected of surface degradation.

[0136] like Figure 8As shown, the verification process includes the following steps: First, a reference lidar point cloud is collected under healthy conditions, and a reference three-dimensional Gaussian structural field is established based on the structural drawings; second, hull roll, pitch, lifting displacement, and hoisting load disturbances are applied at different operational stages, and the attitude, load, lifting status, and leg support status are recorded simultaneously; third, rigid body motion compensation, structural semantic segmentation, Gaussian structural field construction, reference structural field registration, and Gaussian parameter deviation calculation are performed on the collected point cloud; finally, degradation features are extracted based on the deviation sequence, and the health risk index, remaining safe operation window, risk heat map, and maintenance recommendations are output.

[0137] In a set of exemplary verification data, the experiment included 60 sets of scanning sequences: 20 sets for normal conditions, 15 sets for minor local deformation, 15 sets for moderate structural misalignment, and 10 sets for high-risk anomalies. Each sequence contained 10 to 20 consecutive scanning cycles. To verify the rigid body motion compensation effect, equivalent roll ±3°, pitch ±2°, and horizontal translation disturbances were superimposed without changing the actual shape of the structure. To verify the anomaly identification effect, local displacement or gap changes of 10mm, 20mm, and 40mm were respectively set in the crane base, deck main beam, and lifting rack areas.

[0138] Figure 9 A comparison of the deviation distribution of point cloud before and after rigid body motion compensation is presented. Before compensation, the point cloud deviation is mainly affected by the overall attitude change of the hull, and large-scale pseudo-deviations appear on the structural surface. After compensation, the large-scale pseudo-deviations are significantly suppressed, and the abnormal deviations are concentrated in the local structural areas under actual loading. This result shows that the present invention can effectively separate the overall rigid body displacement caused by sea state and operational motion from the actual structural anomalies, thereby reducing the risk of misjudging hull motion as structural damage.

[0139] To further verify the anomaly representation capability of the three-dimensional Gaussian structural field, this embodiment performs semantic region registration between the current structural field and the reference structural field, and calculates Gaussian center drift, covariance variation, principal axis direction variation, and reflection intensity variation. Verification results show that when local settlement occurs in the crane base area, the Gaussian center drift index increases first; when local bending or bulging occurs at the edge of the deck main beam, the covariance scale variation and principal axis direction variation are more sensitive; when changes in reflection characteristics occur at connection nodes or surface degradation areas, the reflection intensity deviation and anomalous Gaussian clustering increase simultaneously. This demonstrates that the present invention can not only detect anomalies but also distinguish anomaly types based on the dominant characteristics of Gaussian parameters.

[0140] Figure 10A 3D structural risk heat map generated by mapping the health risk index is presented. As shown in the figure, high-risk areas are mainly concentrated in load-bearing sensitive areas such as the crane base, connection nodes, and main deck beams. The spatial distribution of risk is consistent with the abnormal loading locations set in the test. The system can further output suggestions based on the risk level, such as re-scanning, manual re-inspection, load reduction, reducing lifting speed, limiting the crane's slewing range, or shutdown for maintenance, so that the test results can directly serve on-site operational safety management.

[0141] To quantify the technical effects of this invention, this embodiment sets up three types of comparison methods: the first type is the traditional threshold method, which directly triggers an alarm based on single-point deviation or sensor threshold; the second type is the ICP point cloud registration deviation method, which calculates the surface distance deviation after conventional point cloud registration; and the third type is the Gaussian method without motion compensation, which constructs a three-dimensional Gaussian structure field without introducing constraints on hull attitude and operational status. Evaluation indicators include anomaly recognition accuracy, F1 score, false alarm rate, average early warning operation cycle, and single-frame processing time. The verification results are shown in Table 1.

[0142] Table 1. Validation results of different methods in the structural anomaly identification and health prediction task. As shown in Table 1, while the traditional threshold method has a fast computation speed, it is easily affected by attitude disturbances and local noise, resulting in a high false alarm rate. The ICP point cloud registration deviation method can reflect geometric deviations, but it lacks structural semantics and temporal degradation features, making it difficult to reliably distinguish between operational disturbances and real anomalies. The Gaussian method without motion compensation has a certain spatial representation capability, but under complex sea conditions, it still misclassifies some overall motion residuals as structural anomalies. In contrast, the method of this invention achieves superior results in terms of anomaly identification accuracy, F1 score, false alarm rate, and early warning capability, indicating that the technical chain of "rigid body motion compensation—three-dimensional Gaussian structural field—structural deviation field—health risk prediction" is practically feasible.

[0143] In terms of computational efficiency, the method of this invention has a single-frame processing time of 0.42s, which can meet the near real-time requirements of periodic structural scanning, re-scanning verification, and health assessment during the operational phase. Since the health risk index output by this invention can be mapped back to the structural semantic region, maintenance personnel can directly locate high-risk components and execute maintenance or operational control measures according to the risk level. Therefore, this embodiment demonstrates that the solution of this invention can be implemented and has significant technical effects from five aspects: data acquisition, motion compensation, anomaly identification, trend prediction, and visual decision-making.

[0144] Example 2 This embodiment provides a structural health prediction system for wind turbine installation vessels based on radar-constrained Gaussian structure fields.

[0145] A computer-readable storage medium storing a plurality of instructions adapted for loading and execution by a processor of a terminal device of the method for predicting the structural health of a wind turbine installation vessel based on a radar-constrained Gaussian structure field.

[0146] A terminal device includes a processor and a computer-readable storage medium, the processor being configured to implement various instructions; the computer-readable storage medium being configured to store multiple instructions adapted for loading and execution by the processor of the aforementioned method for predicting the structural health of a wind turbine installation vessel based on a radar-constrained Gaussian structure field.

[0147] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for predicting the structural health of wind turbine installation vessels based on radar-constrained Gaussian structure fields, characterized in that, include: Acquire multi-source heterogeneous data, including lidar point cloud data and ship operation data; Compensation of rigid body motion of point cloud under the coupling effect of complex sea conditions and installation operations using hull operation data and lidar point cloud data. Semantic segmentation of key structural regions of wind turbine installation vessel based on compensation results; A three-dimensional Gaussian structure field is constructed based on the partitioned structural point cloud; The reference structure field is registered based on the constructed three-dimensional Gaussian structure field; Identifying hull structural anomalies based on registration deviation; Structural degradation features are extracted based on the results of ship structure anomalies. Health risk prediction based on extracted structural degradation features; Generate a three-dimensional structural risk heat map and maintenance decisions based on health risk prediction results; The construction of a three-dimensional Gaussian structure field based on the partitioned structural point cloud includes the following: The first scan time Three key monitoring areas are defined, and their compensated regional point clouds are set as follows: Based on the spatial distribution, sampling density, and structural semantic attributes of the point cloud in this region, the first... The three-dimensional Gaussian unit set of each key monitoring area is represented as follows: in, Indicates the first The first scan time A set of three-dimensional Gaussian elements for each structural region; Indicates the first The first structural region One three-dimensional Gaussian unit; This indicates the number of three-dimensional Gaussian elements within the current structural region. Each three-dimensional Gaussian element is defined as: ,in, Indicates the center position of a three-dimensional Gaussian element; Represent the covariance matrix; This represents the transparency parameter or validity weight; This represents the lidar reflection intensity attribute corresponding to the Gaussian unit; the three-dimensional Gaussian distribution function is expressed as: ,in, Represents any position in three-dimensional space. Indicates the first The scan time, the first The first key structural region The transparency parameter or effectiveness weight of a three-dimensional Gaussian element is used to adjust the contribution of the Gaussian element to the structural field response. This represents the center position vector of the three-dimensional Gaussian element; The covariance matrix represents the three-dimensional Gaussian element, used to describe the scale and orientation distribution of the Gaussian element in three-dimensional space; The inverse matrix of the covariance matrix; This represents the transpose operation of a matrix or vector. This represents the natural exponential function. Indicates position From the first The structural field response generated by the 3D Gaussian element; for the ... Gaussian units are first used in the region point cloud. Select its local neighborhood point cloud Then, the point cloud center of the neighborhood is calculated, followed by the calculation of the local point cloud covariance matrix. Eigenvalue decomposition is then performed on the local covariance matrix, and the covariance matrix of the 3D Gaussian cell is determined based on the eigenvalues ​​and eigenvectors. After determining the spatial parameters of the Gaussian cell, each Gaussian cell is assigned a reflection intensity attribute, transparency parameter, structural semantic label, and risk weight, based on the... The Gaussian unit is composed of the first... The point cloud of the key structural region is initialized, therefore the Gaussian unit inherits from the first... Semantic tags for each structural region and risk weight : ,in, Represents Gaussian unit Structural semantic tags; Represents Gaussian unit The corresponding normalized risk weights; after completing the Gaussian element initialization and attribute assignment for all key structural regions, the three-dimensional Gaussian elements of each region are combined to obtain the first... Three-dimensional Gaussian structure field of the wind turbine installation vessel at each scan time: in, Indicates the first The current three-dimensional Gaussian structure field at each scan time; Indicates the first The set of three-dimensional Gaussian elements corresponding to each key structural region.

2. The method for predicting the structural health of a wind turbine installation vessel based on a radar-constrained Gaussian structure field according to claim 1, characterized in that, The method of using ship operation data and lidar point cloud data to perform rigid body motion compensation for point clouds under the coupled effects of complex sea conditions and installation operations includes performing rigid motion compensation on lidar point clouds using ship attitude data and operation status data, so that the point clouds are unified under the ship structure coordinate system. First, a world coordinate system is established. hull coordinate system LiDAR coordinate system and the local coordinate system of the key structure The extrinsic parameter matrix from the lidar coordinate system to the ship's coordinate system is expressed as: ,in, Represents the transformation matrix; Represents the rotation matrix; Represent the translation vector; then estimate the first translation vector based on the ship's operational data. The overall rigid body motion of the ship at a given moment relative to the reference moment is expressed as: ,in, Indicates the first The overall rigid body motion matrix of the ship at each moment; Represents the rotation matrix; Represent the translation vector; finally, the original lidar point cloud is... Transforming from the lidar coordinate system to a unified ship coordinate system and eliminating the influence of the ship's overall rigid body motion, for any point... The compensated points The calculation is as follows: ,in, This represents the point cloud points located in the ship's coordinate system after rigid body motion compensation; Used to counteract the overall motion of the ship; Used to transform lidar point clouds to the ship's coordinate system, the compensated point cloud set is denoted as: .

3. The method for predicting the structural health of a wind turbine installation vessel based on a radar-constrained Gaussian structure field according to claim 2, characterized in that, The semantic partitioning of key structural regions of the wind turbine installation vessel based on the compensation results includes first obtaining the design model, structural drawings, or digital twin model of the wind turbine installation vessel, and then establishing the topology of the key load-bearing structure. The topology of the key structure of the wind turbine installation vessel is represented as follows: in, This indicates the topology of the key load-bearing structure of the wind turbine installation vessel; Represents the set of nodes in the key structural region; Represents the set of connecting edges between structural regions; Represent the set of structural region attributes; then, the compensated point cloud... Divided into Key structural regions: ,in, Indicates the first The first scan time The point cloud set corresponding to each key monitoring area; for point cloud points that simultaneously fall within the candidate enclosing domains of two or more structural regions, a region attribution scoring function is constructed: in, Point Belonging to the The score value of each structural region; This indicates the distance from a point to the boundary or center plane of the structural region; This indicates the curvature consistency or surface continuity deviation between a point and its neighboring point cloud. This indicates the deviation between the local normal of the point cloud and the normal of the designed structural surface; Represents structural topology constraints; , , and This represents the weight coefficients of each item; after completing the point cloud division of the key structural regions, a structural semantic label is assigned to each key monitoring region, represented as: ,in, Indicates the first Structural semantic labels are assigned to key monitoring areas, and risk weights are set according to different structural areas. The risk weight set is represented as follows: in, Indicates the first The risk weights for each key structural region are expressed as follows: ,in, Indicates the first Importance score of each structural region; Indicates stress concentration score; Indicates the score based on the frequency of historical anomalies; Indicates the failure consequence score; , , and This represents the weighting coefficients, and the risk weights are normalized. ,in, This represents the normalized regional risk weights, ultimately resulting in a set of key structural region point clouds with structural semantic labels and risk weights: ,in, Indicates the first Structural semantic partitioning results at each scan time; Indicates the first A set of compensation point clouds for key structural regions.

4. The method for predicting the structural health of a wind turbine installation vessel based on a radar-constrained Gaussian structure field according to claim 3, characterized in that, The reference structure field registration based on the constructed three-dimensional Gaussian structure field includes representing the reference three-dimensional Gaussian structure field as follows: ,in, Represents the reference three-dimensional Gaussian structure field; Indicates the first The set of reference Gaussian elements corresponding to the key structural regions, the first The set of reference Gaussian elements for each key structural region is represented as follows: in, Indicates the first The first critical structural region One reference Gaussian element; Indicates the first The number of reference Gaussian elements in each key structural region, where the reference Gaussian element is represented as: in, Indicates the center position of the reference Gaussian element; Represents the covariance matrix of the reference Gaussian element; This represents the transparency parameter or validity weight of the reference Gaussian cell; This represents the reflection intensity properties of the reference Gaussian unit; The semantic label represents the structural region to which the reference Gaussian unit belongs; The risk weights of the structural region are then represented; the current three-dimensional Gaussian structural field is then established. and the benchmark three-dimensional Gaussian structure field The correspondence between them, for the first The current three-dimensional Gaussian structure field at the nth scan time, its i-th The key structural regions are represented as follows: in, Indicates the first The first scan time A set of three-dimensional Gaussian elements for each structural region; Indicates the first The first structural region One current Gaussian unit; This indicates the number of Gaussian elements within the current structural region; for the current Gaussian element... Only when they have the same semantic tags The set of reference Gaussian elements Find the best-matching reference Gaussian element; after matching the current Gaussian element with the reference Gaussian element, calculate the Gaussian parameter deviation between the two. and its matching reference Gaussian element Its center drift is expressed as: in, Indicates the first The first scan time Within the structural region, the first The center drift of each Gaussian element is calculated, thus obtaining the parameter deviation of each current Gaussian element. Finally, the parameter deviations of multiple Gaussian elements within the same structural region are spatially aggregated to construct a structural deviation field. The structural region in the first The structural deviation field at each scanning time is represented as: ,in, Indicates the first The first scan time Structural deviation field of each structural region; Indicates the first Within the structural region, the first The parameter deviation of a Gaussian element; This indicates the current number of Gaussian elements within the structural region and aggregates the Gaussian element deviations within the region. ,in, Indicates the first The first scan time Regional-level deviation intensity of each structural region; , , and This represents the weighting coefficients for various Gaussian parameter deviations, which are then combined with regional risk weights. The risk-weighted bias strength is obtained as follows: in, The overall structural deviation field is expressed as follows: This represents the intensity of the deviation after incorporating risk weights for the structural regions. in, Indicates the first The overall critical structural deviation field of the wind turbine installation vessel at each scanning moment; Indicates the first Structural deviation field of each structural region; Indicates the number of critical structural regions.

5. The method for predicting the structural health of a wind turbine installation vessel based on a radar-constrained Gaussian structure field according to claim 4, characterized in that, The identification of hull structural anomalies based on registration deviation includes the identification of the first... Construct comprehensive anomaly indicators for key structural regions. : ,in, Indicates the first Comprehensive abnormal indicators for key structural regions; This indicates the risk weight of the region; This represents the mean Gaussian center drift within the region; This represents the mean change in covariance within the region; This represents the average change in the principal axis direction within the region; This represents the average value of the variation in reflection intensity; Indicates the spatial clustering degree of anomalous Gaussian cells; to The weighting coefficients are represented by the angle of change of the principal axis direction: ,in, Represents the matrix trace operation; Indicates the first The principal axis direction variation angle of each Gaussian element Indicates the first The first scan time Within the structural region, the first The relative principal direction rotation matrix between the current Gaussian element and its matched reference Gaussian element is determined; then, the structural anomaly type is determined based on the dominant characteristics of different Gaussian parameter deviations; finally, the anomaly index is used to determine the structural anomaly type. Based on preset risk thresholds, structural risks are classified into five levels: normal, attention, warning, restricted operation, and shutdown for maintenance.

6. The method for predicting the structural health of a wind turbine installation vessel based on a radar-constrained Gaussian structure field according to claim 5, characterized in that, The structural degradation feature extraction based on the ship's structural anomaly results includes, for the first Each structural region has a construction length of [number]. Gaussian parameter deviation sequence: ,in, Indicates the first The structural regions in recent The sequence of structural deviations within a scan time or work cycle Indicates the first The first scan time The structural deviation fields of key structural regions are then used to extract structural degradation feature vectors based on the Gaussian parameter deviation sequence. ,in, Indicates the first At the [time]th moment Degenerate feature vectors of each structural region; This represents the growth rate of the Gaussian center drift; This represents the rate of change of the covariance scale. Indicates the cumulative rotation in the spindle direction; Indicates the rate of change of reflection intensity; Indicates the area of ​​the abnormal region; Indicates the expansion rate of the abnormal region; Indicates the spatial clustering degree of abnormal regions; This represents the local maximum deviation value; This represents the dynamic response amplification factor; finally, the Gaussian center drift growth rate is expressed as: The expansion rate of the abnormal region is expressed as: .

7. The method for predicting the structural health of a wind turbine installation vessel based on a radar-constrained Gaussian structure field according to claim 6, characterized in that, The health risk prediction based on extracted structural degradation features includes the first Degenerative feature vectors of each structural region , Work stage labels and load case vector Combined into the prediction input: ,in, Indicate the input to the health prediction model; then... Input the health prediction model to obtain the health risk index for the current and future work cycles: ,in, Indicates the first Key structural regions in the future Health risk index after one work cycle; This represents a health prediction model; The model parameters are represented; the remaining safe operation window is represented as follows: ,in, Indicates the first Remaining safe operation windows for key structural areas; This represents the critical risk threshold.

8. The method for predicting the structural health of a wind turbine installation vessel based on a radar-constrained Gaussian structure field according to claim 7, characterized in that, The process of generating a three-dimensional structural risk heat map and maintenance decisions based on health risk prediction results includes mapping the health risk index of key monitoring areas to corresponding three-dimensional Gaussian cells to obtain the three-dimensional structural risk heat map. ,in, Indicates the first A three-dimensional structural risk heat map at any given moment; Indicates the first Health risk index of each structural region; The risk level is indicated; then, maintenance priority is calculated based on the risk level, risk growth trend, and structural area weights. ,in, Indicates the first Maintenance priority scoring for key structural areas; This indicates the rate of change in the health risk index; , , This represents the weighting coefficient.

9. A structural health prediction system for wind turbine installation vessels based on radar-constrained Gaussian structure fields, executing the structural health prediction method for wind turbine installation vessels based on radar-constrained Gaussian structure fields as described in claim 1, characterized in that, include: The data acquisition module is configured to acquire multi-source heterogeneous data, including lidar point cloud data and ship operation data. The motion compensation module is configured to use hull operation data and lidar point cloud data to perform rigid body motion compensation of point cloud under the coupling effect of complex sea conditions and installation operations. The semantic segmentation module is configured to perform semantic segmentation of key structural areas of the wind turbine installation vessel based on the compensation results. The structure field module is configured to construct a three-dimensional Gaussian structure field based on the partitioned structure point cloud. The registration module is configured to register the reference structure field based on the constructed three-dimensional Gaussian structure field. The anomaly detection module is configured to identify hull structural anomalies based on registration deviation. The feature extraction module is configured to extract structural degradation features based on the results of hull structural anomalies. The health prediction module is configured to predict health risks based on extracted structural degradation features. The decision-making module is configured to generate a three-dimensional structural risk heat map and make maintenance decisions based on the health risk prediction results.

Images