A method and system for detecting the verticality of wind turbine towers
By processing 3D laser point cloud data and combining gravity alignment and cross-height joint optimization, the noise and multi-source error problems in wind turbine tower verticality detection were solved, and reliable verticality detection and automated result output were achieved.
Patent Information
- Application Number
- CN202611503108.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-24
- Publication Date
- 2026-11-17
AI Technical Summary
In existing technologies for detecting the verticality of wind turbine towers, noise, obstruction, and multi-source acquisition errors make it difficult to reconstruct the centerline stably, distinguish between overall tilt and local deflection, and are prone to misjudgment near the critical threshold.
By employing a three-dimensional laser point cloud data processing method, and through gravity alignment, cross-height joint optimization, and identifiable decoupling, combined with uncertainty-driven three-state conclusion gating, a traceable and repeatable verticality detection result is output.
This improved the stability and accuracy of wind turbine tower verticality detection, reduced the risk of misjudgment, and enhanced the reliability and automation of the detection results.
Smart Images

Figure CN123111202A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of wind power generation technology, specifically relating to a method and system for detecting the verticality of wind turbine towers. Background Technology
[0002] As wind turbine towers become taller and more flexible, they are susceptible to tilting or bending deformation due to factors such as foundation settlement, wind load, temperature differences, installation errors, and long-term fatigue. Three-dimensional laser scanning can rapidly and non-contactly acquire high-density three-dimensional point clouds of the tower surface, providing a data foundation for full-height, traceable geometric inspection. However, how to stably reconstruct the tower centerline from point clouds containing noise, obstructions, and multi-source acquisition errors, and based on this, provide interpretable, traceable, and repeatable verticality conclusions, remains a pressing problem to be solved in this field.
[0003] Existing technologies have methods for determining the tilt of wind turbine towers based on three-dimensional laser scanning. Generally, after obtaining the point cloud of the wind turbine tower through three-dimensional laser scanning, the points are layered according to a preset height. Planar circular fitting is performed on each layer of the point cloud to obtain the center point. Then, the target straight line and target center point are determined from the center point using the idea of random sampling consistency. Finally, the tilt is calculated based on the highest and lowest target center points. However, the following problems still exist: (1) Discrete measuring points cannot fully reflect the local curvature of the tower at its full height, and it is difficult to deploy the measurement in complex sites, strong light, obstruction, or high-altitude areas; (2) Flanges, ladders, cables, blade projections, reflected noise, and overlapping areas of multiple stations will interfere with the section fitting, and the single least squares fitting is not strong enough to resist abnormal points; (3) Point clouds of different dates lack a unified benchmark and a comparison mechanism at the same elevation, making it difficult to form comparable trend indicators.
[0004] Therefore, providing an accurate detection method that can stably reconstruct the center sequence and centerline of the tower section and output traceable and repeatable wind turbine tower verticality detection results under the presence of noise, obstruction and multi-source acquisition errors is an urgent technical problem to be solved in this field. Summary of the Invention
[0005] This invention provides a method and system for detecting the verticality of wind turbine towers, in order to solve the problems in existing wind turbine tower verticality detection methods, such as difficulty in stably reconstructing the centerline under conditions of noise, obstruction, and multi-source acquisition errors, difficulty in distinguishing between overall tilt and local deflection, and easy misjudgment near the critical threshold.
[0006] To achieve the above objectives, the present invention provides a method for detecting the verticality of a wind turbine tower, comprising: S1: Collect the three-dimensional laser point cloud of the wind turbine tower under test and the calibration data for gravity direction determination and inter-station registration. Perform preprocessing, multi-station registration and gravity alignment on the three-dimensional laser point cloud to obtain the gravity-aligned point cloud. S2: Segment the tower surface point cloud from the gravity-aligned point cloud and mark the protrusions on the wind turbine tower under test as candidate points of anomalies; S3: In a coordinate system where the Z-axis is aligned with the vertical direction of gravity on site, the effective height of the tower is divided into layers along the Z-axis with a preset layer thickness. Cross-sectional fitting is performed on the point set of each layer of the point cloud on the surface of the tower. The fitting center and fitting radius of each layer are used as parameters to be estimated for joint optimization across the height. The objective function of the joint optimization across the height includes at least a single-point radial residual term, a continuity constraint term for the center of adjacent cross sections, and a continuity constraint term for the radius of adjacent cross sections. Solving the objective function yields the fitting center, fitting radius, center covariance, and observation weight of each effective cross section. S4: Sort the fitted circle centers of each effective section by height to form a center sequence, and perform identifiable decoupling of the overall tilt and local deflection on the center sequence to obtain the overall tilt parameter, local deflection parameter and decoupling parameter covariance. The identifiable decoupling includes: representing the horizontal coordinate of the center sequence as the sum of the overall linear term and the local smoothing margin, and applying a weighted zero mean constraint and a weighted zero linear trend constraint to the local smoothing margin to uniquely separate the overall tilt component and the local deflection component. S5: Based on the center covariance and observation weights obtained in step S3, and the overall tilt parameter, local deflection parameter, and decoupling parameter covariance obtained in step S4, determine the uncertainty of the detection index. The detection index includes at least one of the following: overall tilt angle, top horizontal offset, maximum horizontal offset, maximum local deflection, and direction. Compare the measurement interval formed by the detection index and the uncertainty with a preset threshold, and output a detection result in one of three states: normal, alarm, or retest / manual verification.
[0007] In one embodiment, the tower geometry and design information includes the nominal height of the tower. Diameter of the bottom of the tower Diameter of the top of the tower The tower design drawings or design section parameters, the wind turbine asset information including wind turbine asset coordinates, known spatial locations and design dimensions of the nacelle and blades, and the scan registration information including scan site number and registration residuals in the inter-site overlap area. Step S2 includes: The region of interest is established based on the coordinates of the wind turbine assets and the nominal dimensions of the tower, and the bottom radius, top radius and effective height of the tower of the wind turbine to be measured are obtained. Set a ground elevation threshold based on the site elevation or the elevation of the tower foundation cap. The elevation is lower than The point cloud data was identified as ground points and excluded. To pre-determine the elevation margin, and based on the known spatial location, design dimensions, and installation location of the nacelle and blades, a spatial bounding box or bounding cylinder for the nacelle and blades is established. Point clouds falling into the bounding box or bounding cylinder are identified as nacelle or blade point clouds and excluded, thus obtaining the exclusion results of ground, nacelle, and blade point clouds and the remaining point clouds. Within the region of interest, a random sampling consistency cylinder or frustum model is used to fit the initial surface of the tower from the point cloud of the region of interest. Outliers are eliminated according to preset constraints on the normal vector direction and the radius changing slowly with height. Isolated noise clusters are removed through connected component analysis, and the point cloud connected to the tower surface is retained to obtain the point cloud of the tower surface and the initial surface model of the tower. For at least one protrusion among ladders, cables, flanges, platforms, and reflected noise, calculate the distance from the point to the initial tower surface, the local curvature, and the continuous height range. Points that meet the preset conditions are marked as abnormal candidate points instead of being directly deleted, thus obtaining abnormal candidate point markings. Candidate flange areas are then marked according to the tower design drawing or design section parameters.
[0008] In one embodiment, step S3, which involves dividing the effective height of the tower along the Z-axis into layers with a preset layer thickness, performing cross-sectional fitting on each layer of the point cloud on the tower surface, and using the center and radius of the fitted cross-section of each layer as parameters to be estimated, includes: along The shaft is based on a preset layer thickness. The effective height of the tower is layered, and the point cloud on the surface of the tower is divided into multiple height section point sets, the first... The set of points on the cross section falls from the height The tower-shaped clouds are composed of points; Project the point set of each cross section onto the horizontal plane, and independently fit each layer to obtain the initial fitting center, fitting radius and cross section fitting residual of each layer; Based on the preset cross-section validity judgment conditions, the cross-sections obtained by independent fitting layer by layer are judged for validity. Layers that meet the conditions are marked as valid cross-sections, and layers that do not meet the conditions are marked as invalid cross-sections and the reasons for invalidity are recorded. Flange area processing is performed based on the radius continuity constraint according to the flange indication, and a radius continuity penalty is applied to the flange area. For illegal orchid areas, the cross-layer radius continuity term is removed and replaced with a constraint. The flange indication is established based on the flange height given in the tower design drawing. The difference in radius between adjacent layers. For the rate of change of the design radius, for The turning point value; Using the initial center and radius of each layer as parameters to be estimated, the joint optimization solution is applied to all effective height layers according to the preset convergence conditions, yielding the jointly optimized center ci and radius ri of each layer. The objective function of the joint optimization is:
[0009] in, For the first Layer From the point to the center of the circle to be estimated Horizontal distance, It is the observation weight. It is a robust loss function. and To constrain the weights, It is a radius continuity penalty function; according to Calculate the covariance matrix of all parameters. Then, the covariance of the center of each layer is calculated. ,in, The covariance of each circle center is estimated by dividing the weighted sum of squared residuals by the effective degrees of freedom. for Chinese correspondence , A 2×2 submatrix.
[0010] In one embodiment, the observation weight for: ; in, Circumferential coverage factor; The incident angle factor; Point density factor; Register residual factors for the sites; Candidate markers for anomalies; The observation weight Normalize within each layer so that the sum of the weights for that layer is 1.
[0011] In one embodiment, the overall tilt parameters include the overall tilt angle, top horizontal offset, maximum horizontal offset, and layer weight; the local deflection parameters include local deflection at each height, maximum local deflection and its height, orientation angle, and decoupling parameter covariance; and the overall tilt and local deflection are identifiablely decoupled from the central sequence to obtain the overall tilt parameters and local deflection parameters, including: Arrange all effective cross-section centers according to height to form a center sequence. For the central sequence Perform weighted three-dimensional line fitting to obtain the overall fitted axis. and the overall axial direction from bottom to top , If the bottom area is obstructed by the foundation or flange, use the minimum effective height and the maximum effective height and extrapolate to the nominal detection range according to the model; A local flexure model is constructed using low-order polynomials, piecewise splines, or smooth curves that satisfy the continuity of the tower structure. The horizontal centerline is represented as ,in, For height The horizontal center coordinates of the location For constant terms, The coefficient of the overall sloping first-order term. For local smoothing margin; using cubic B-spline basis matrix express , The coefficient vector is a B-spline vector; a weighted zero-mean constraint is applied. and weighted zero linear trend constraint ; Determine layer weights ,in, For coverage and residual gating factors, For the first Layer center covariance traces, Pick To avoid division by zero, normalization is then performed. ; The B-spline uses a cubic basis function, with repeated nodes at the endpoints; the initial spacing of the internal nodes is taken as... Each node interval must contain at least three valid sections; otherwise, adjacent intervals are merged, and the smoothing parameter is adjusted. exist to The minimum value is selected from the logarithmic candidate set using generalized cross-validation. Solve the constrained least squares problem: and apply simultaneously and ;in, This is the layer weight matrix. A vector consisting of the horizontal center coordinates of all layers. To design the matrix and the first Action is , For smoothing parameters, Given a second-order difference matrix; solve using the KKT equations in one step. If the condition number of the KKT matrix is greater than... If the intervals of adjacent nodes with the fewest valid points are merged sequentially and the calculation is repeated, and if the condition is still not met, the process is transferred to manual review. The same layer weights are used in both the X and Y directions. B-spline basis matrix and smoothing parameters Solve separately to keep the directional components comparable; Coefficient of the overall inclined linear term Calculate the overall tilt angle ,Depend on Calculate the local deflection and obtain the decoupling parameter covariance from the corresponding sub-block of the KKT system inverse matrix. Calculate height Horizontal offset at location and local deflection Result in top horizontal offset Maximum horizontal offset Maximum local deflection and direction angle.
[0012] In one embodiment, step S5 specifically includes: Before performing uncertainty calculations, check the effective tower height, effective cross-section ratio, convergence status of joint optimization across height, and covariance condition number. If any of these conditions are not met, output the retest / manual verification directly. For the parameter vector Calculated indicators Calculate the gradient at the convergent solution. , and according to Calculate the uncertainty components of the model, where The parameter covariance matrix obtained by joint optimization across height for S3; The 6-DOF pose covariance of control point or station registration Jacobi through coordinate transformation Propagation as registration uncertainty components ; Gravity direction covariance Through the tilt angle to the unit vector of gravity Jacobian propagation as a component of gravitational uncertainty ; Let the quantization resolution independent components be denoted as , and according to Calculate the combined standard uncertainty; Pick This results in an expanded uncertainty with approximately 95% coverage. When the effective degrees of freedom are less than the preset threshold, calculate veff according to the Welch-Satterthwaite formula and take t0.975, veff instead of 2; To verify the linear propagation results, a two-tiered stratified bootstrap resampling method was used, with 200 iterations at both the site and cross-section levels. First, sites were sampled with replacement, then effective cross-sections were sampled with replacement within each site. This process was repeated sequentially across heights for joint optimization and identifiable decoupling. A fixed random seed was used, and the seed, effective resampling times, and index sequences were saved. If the effective number of iterations was less than 180, or the ratio of the bootstrap standard deviation to the linear propagation standard deviation was not between 0.8 and 1.25, the larger value was taken, and the conclusion was downgraded to a verification result. For the test indicators that have passed the pre-test quality inspection, construct the measurement interval. The measurement interval is compared with a preset threshold. Comparison: When the upper bound of the measurement interval is less than The output is normal; when the lower bound of the measurement interval is greater than An alarm will be output in real time when the measurement range is touched or crossed. Output for retesting / manual verification; for directional bilateral limits, apply the same rule to both boundaries; among which... Limits given for corporate procedures, design documents, or acceptance criteria.
[0013] In one embodiment, step S1 further includes performing a quality gating check on the gravity-aligned point cloud, including: Perform control point residual checks, assuming the first... The known coordinates of the control points are: After registration, the coordinates are calculated as follows: Then the first The residuals at each control point are: The residual size is , The root mean square of the residuals at each control point is ,when If the residual error exceeds the preset control point threshold, it is considered unqualified. Perform registration residual checks in the inter-station overlap area, and check the first... For each station, take corresponding point pairs within the overlapping area, and let the th station be... In a pair of corresponding points, point The unit normal vector to the local plane of the point cloud at adjacent stations is A point on the plane is The point-to-surface residual is: , No. There are a total of [number] stations in the overlapping area. When there are 1 corresponding point pair, the registration residual in the overlapping area between stations is: ,when If the error exceeds the preset registration residual threshold, it is considered unqualified. Perform point cloud coverage checks, including height coverage checks and circumferential coverage checks. Height coverage is... ,in, and These represent the lowest and highest altitudes actually covered by the point cloud, respectively. The nominal height of the tower; circumferential coverage is calculated based on preset azimuth sectors. For a given height section, the number of sectors covered by the measured point cloud is assumed to be... The measured circumferential coverage angle is: The circumferential coverage rate is: If the height coverage is lower than the preset height coverage threshold, or the circumferential coverage is lower than the preset circumferential coverage threshold, or the single-layer circumferential coverage angle is lower than the preset coverage angle threshold, it is deemed unqualified. If any indicator in the quality gate inspection does not meet the preset quality threshold, the process returns to step S1 for supplementary sampling or proceeds to manual review, and does not proceed to automatic qualification determination.
[0014] In one embodiment, the detection method further includes: connecting the detection result, the original point cloud index, the algorithm version, the threshold version, and the three states to the inspection platform, and generating a supplementary sampling, retesting, or alarm work order based on the three states.
[0015] Based on the same inventive concept, this invention also proposes a wind turbine tower verticality detection system, comprising: The point cloud acquisition and registration module is used to acquire the three-dimensional laser point cloud of the wind turbine tower under test and the calibration data for gravity direction determination and inter-station registration. The three-dimensional laser point cloud is preprocessed, multi-station registered and gravity aligned to obtain the gravity aligned point cloud. The tower segmentation and anomaly marking module is used to segment the tower surface point cloud from the gravity-aligned point cloud and mark the protrusions on the wind turbine tower under test as anomaly candidate points. The cross-height joint optimization module is used to divide the effective height of the tower into layers with a preset layer thickness along the Z-axis in a coordinate system where the Z-axis is consistent with the vertical direction of gravity on site. It performs cross-sectional fitting on the point cloud of the tower surface for each layer, and performs cross-height joint optimization with the fitting center and fitting radius of each layer as parameters to be estimated. The objective function of the cross-height joint optimization includes at least a single-point radial residual term, a continuity constraint term for the center of adjacent cross sections, and a continuity constraint term for the radius of adjacent cross sections. Solving the objective function yields the fitting center, fitting radius, center covariance, and observation weight of each effective cross section. The identifiable decoupling module is used to sort the fitted circles of each effective section by height to form a center sequence, and to perform identifiable decoupling of the center sequence for overall tilt and local deflection to obtain the overall tilt parameter, local deflection parameter and decoupling parameter covariance. The identifiable decoupling includes: representing the horizontal coordinate of the center sequence as the sum of the overall linear term and the local smoothing margin, and applying a weighted zero mean constraint and a weighted zero linear trend constraint to the local smoothing margin to uniquely separate the overall tilt component and the local deflection component. The uncertainty three-state gating module is used to determine the uncertainty of the detection index based on the obtained center covariance and observation weight, as well as the obtained overall tilt parameter, local deflection parameter and decoupling parameter covariance. The detection index includes at least one of the following: overall tilt angle, top horizontal offset, maximum horizontal offset, maximum local deflection and its direction. The measurement interval formed by the detection index and the uncertainty is compared with a preset threshold, and the detection result is output as one of three states: normal, alarm or retest / manual verification.
[0016] Based on the same inventive concept, the present invention also proposes an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the detection method as described in any of the preceding claims.
[0017] Based on the same inventive concept, embodiments of the present invention also propose a computer storage medium storing at least one executable instruction that causes a processor to execute the detection method described in any of the preceding claims.
[0018] Compared with existing technologies, the beneficial effects of this invention are as follows: Addressing the problems in existing wind turbine tower verticality detection methods, such as difficulty in stably reconstructing the centerline under conditions of noise, obstruction, and multi-source acquisition errors; difficulty in distinguishing between overall tilt and local deflection; and susceptibility to misjudgment near critical thresholds, a wind turbine tower verticality detection method based on three-dimensional laser point clouds is proposed. Through cross-height coupled joint optimization, identifiable decoupling of overall tilt and local deflection, and uncertainty-driven three-state conclusion gating, these methods are applied sequentially to the geometric reconstruction, deformation characterization, and conclusion output stages, sharing weights, residuals, and covariance. Adjacent reliable cross-sections form structural constraints on weakly observed layers, improving the stability of the cross-section center sequence while preserving the traceability of flange mutations and invalid layers. Through unique and interpretable decomposition constraints, physically meaningful overall tilt and local deflection are output separately, reducing the risk of misjudging local bending as basic tilt. Furthermore, the aforementioned covariance-driven three-state conclusion gating directly expands the judgment interval for low-quality observations, reducing false alarms and false negatives near critical values. Meanwhile, by calibrating the gravity direction and unifying coordinate transformation, the risk of instrument attitude error being misjudged as tower tilt is reduced; by explicitly handling flange mutations and invalid cross sections, the stability of center extraction under conditions of accessories, obstruction, and outliers is improved; cross-period comparability is achieved through common benchmarks and resampling at the same elevation, while retaining the original point cloud, algorithm version, and parameters for easy traceability and verification; finally, a closed loop of supplementary sampling, retesting, alarm, and work orders is formed with the new energy inspection platform to improve the automation level of batch wind turbine inspection. Attached Figure Description
[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0020] Figure 1 This is a flowchart illustrating the wind turbine tower verticality detection method in an embodiment of the present invention. Figure 2 This is a schematic diagram illustrating the geometric relationship between the center of the cross-section fitting circle, the overall tilt angle, and the local deflection in an embodiment of the present invention. Figure 3 This is a schematic diagram of the structure of the wind turbine tower verticality detection system in an embodiment of the present invention; Figure 4 This is a schematic diagram of an electronic device in an embodiment of the present invention. Detailed Implementation
[0021] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to specific embodiments and accompanying drawings.
[0022] It should be noted that, unless otherwise defined, the technical or scientific terms used in the embodiments of this invention should have the ordinary meaning understood by those skilled in the art. The terms "first," "second," and similar terms used in the embodiments of this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word covers the element or object listed after the word and its equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are only used to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.
[0023] like Figure 1 As shown in the figure, this embodiment of the invention provides a method for detecting the verticality of a wind turbine tower, including the following steps: S1: Collect the three-dimensional laser point cloud of the wind turbine tower under test and the calibration data for gravity direction determination and inter-station registration. Perform preprocessing, multi-station registration and gravity alignment on the three-dimensional laser point cloud to obtain the gravity-aligned point cloud.
[0024] S2: Segment the tower surface point cloud from the gravity-aligned point cloud and mark the protrusions on the wind turbine tower under test as candidate points of anomalies.
[0025] S3: In a coordinate system where the Z-axis is aligned with the vertical direction of gravity on site, the effective height of the tower is divided into layers along the Z-axis with a preset layer thickness. Cross-section fitting is performed on the point set of each layer of the point cloud on the tower surface. The fitting center and fitting radius of each layer are used as parameters to be estimated for joint optimization across the height. The objective function of the joint optimization across the height includes at least a single-point radial residual term, a continuity constraint term for the center of adjacent cross sections, and a continuity constraint term for the radius of adjacent cross sections. Solving the objective function yields the fitting center, fitting radius, center covariance, and observation weight of each effective cross section.
[0026] S4: Sort the fitted circles of each effective section by height to form a center sequence, and perform identifiable decoupling of the overall tilt and local deflection on the center sequence to obtain the overall tilt parameter, local deflection parameter and decoupling parameter covariance. Identifiable decoupling includes: representing the horizontal coordinate of the center sequence as the sum of the overall linear term and the local smoothing margin, and applying weighted zero mean constraint and weighted zero linear trend constraint to the local smoothing margin to uniquely separate the overall tilt component and the local deflection component.
[0027] S5: Based on the covariance of the center and the observation weight obtained in step S3, and the covariance of the overall tilt parameter, local deflection parameter and decoupling parameter obtained in step S4, determine the uncertainty of the detection index. The detection index includes at least one of the following: overall tilt angle, top horizontal offset, maximum horizontal offset, maximum local deflection and its direction. Compare the measurement interval formed by the detection index and uncertainty with the preset threshold, and output the detection result in one of three states: normal, alarm or retest / manual verification.
[0028] During implementation, operators select multiple visible stations around the wind turbine tower and deploy control targets, using a ground-based 3D laser scanner to complete multi-station scanning. For leeward sides of the tower or high-altitude obstructed areas, a UAV equipped with a 3D laser scanning system can be used for supplementary sampling. The system reads the control points and equipment attitude, registers the point clouds of each station, and converts them into a gravity-aligned coordinate system. Subsequently, the region of interest of the tower is clipped, large-scale point clouds on the ground and nacelle are removed, and the tower surface is extracted based on the body-of-rotation constraint. First, a horizontal section is established according to the set layer thickness. Robust circle fitting is performed on each layer of point sets to obtain the fitted circle center, fitted radius, cross-sectional fitting residual, and circumferential coverage. Invalid sections are marked instead of being replaced by interpolation. Weighted straight line and spline fitting are performed on the effective center sequence to obtain the overall axis and local centerline, respectively, and the overall tilt angle is calculated. Top horizontal offset Top offset e(H), maximum offset e(max), maximum deflection f (max) and orientation angle. The system then compares the current curve with the historical centerline under a common benchmark to generate a change curve and trend level. All parameters, quality indicators, and results are written to the inspection platform, which generates normal, retest, or alarm status based on the configured thresholds. Specific layer thickness, filtering neighborhood, robust loss parameters, and thresholds should be calibrated based on equipment accuracy, tower size, and enterprise testing procedures.
[0029] In this embodiment, the input data is a three-dimensional laser point cloud of the wind turbine tower under test. Each point is represented as... Each point also carries at least one of the following attributes: echo intensity, echo count, timestamp, site number, normal vector, incident angle, and point cloud density. In addition to the 3D laser point cloud, the input data also includes the wind turbine number, tower type, nominal tower height, tower design drawing or design section parameters, detection time, wind speed, temperature, unit operating or shutdown status, scanning equipment information, control point coordinates or control target coordinates, gravity direction calibration data, and allowable deviations and alarm thresholds from the company's testing procedures.
[0030] For data from multiple UAV sorties or multiple ground stations, the data is first converted to the same on-site coordinate system before precise registration. This ensures that all subsequent calculations of tilt angles, horizontal offsets, and deflections are performed under the same gravity reference, preventing misjudgments of tower tilt due to instrument attitude or multi-station coordinate errors. Specifically, a detection coordinate system is first established. ,in, The axis is aligned with the vertical direction of gravity on site. The plane is horizontal, with the origin of the coordinate system. It can be determined based on on-site control points, the center of the tower base, or the site reference point. The direction of gravity can be determined through at least one of the following methods: the scanner's built-in tilt compensation results, a calibrated inertial measurement unit (IMU), known leveling control points, on-site control targets, or a total station control network.
[0031] In this embodiment, data acquisition and on-site calibration include: recording the wind turbine number, tower type, nominal tower height, detection time, wind speed, temperature, unit operating or shutdown status, and scanning equipment information before acquisition. At least three ground scanning stations are deployed around the wind turbine tower under test, or a UAV flight path is planned around the tower to ensure sufficient overlap between adjacent scanning areas, with at least 30% overlap between adjacent stations, and coverage of the tower's circumference and full height as much as possible. At least three non-collinear control targets are deployed around the tower, distributed in different directions around the tower and not obstructed by the tower itself. In other embodiments, existing measurement control points can be used to establish an on-site benchmark. During scanning, point clouds of the tower surface are acquired, and calibration data for gravity direction determination and inter-station registration are acquired simultaneously. For the leeward side of the tower or high-altitude obstructed areas, a UAV flight path is planned around the tower, with the flight altitude covering the full height of the tower and adjacent flight paths having sufficient overlap. When collecting data while the wind turbine is running, the single-station scanning time can be shortened to a preset value, or multi-frame time synchronization or point cloud timing filtering can be used to remove motion distortion points caused by tower vibration, thus reducing motion distortion caused by tower vibration. When obstructed areas are present, additional stations or UAV perspectives can be added to supplement coverage. For obstructed areas that cannot be supplemented, the obstruction mask should be recorded, and interpolated data should not be used to falsely identify measurement points. Control points or stable ground references can be reused for each detection of the same wind turbine to improve the comparability of cross-period data.
[0032] In this embodiment, when performing point cloud preprocessing and coordinate unification processing, the input data includes the acquired raw point cloud. This refers to a 3D laser point cloud, which also includes synchronously acquired control target coordinates, scanning equipment attitude data, inertial measurement unit data, gravity direction calibration data, station numbers and location information for each scanning station, scanner nominal ranging error, and wind turbine asset information such as turbine number, tower type, and nominal tower height. (Original point cloud) Each point in the array is denoted as It also carries attributes such as echo intensity, echo count, timestamp, station number, normal vector, and angle of incidence.
[0033] In this embodiment, the original point cloud is first processed. Preprocessing is performed, including range clipping, invalid echo removal, statistical outlier filtering or radius filtering, density equalization sampling, and normal vector estimation. Then, multi-site cloud coarse registration, multi-site cloud fine registration, and gravity alignment are performed.
[0034] Specifically, region clipping is used to remove point clouds outside the region of interest (ROI) of the wind turbine tower. Based on the wind turbine asset coordinates, the nominal height of the tower, and the nominal radius of the tower, a spatial ROI is established. Point clouds within this ROI are retained, while ground, distant buildings, vegetation, and other irrelevant features outside the ROI are removed. The boundaries of the region clipping can be set according to the tower design drawings and actual site conditions. For example, using the center of the tower's bottom as a reference, the horizontal direction can be a preset multiple of the tower's maximum radius, and the vertical direction can be a preset range of the tower's nominal height.
[0035] Invalid echo rejection is used to remove points whose reflection intensity is lower than a preset threshold, whose echo count is abnormal, or whose ranging confidence is insufficient. Specifically, it can remove at least one of the following: points whose echo intensity is lower than a preset intensity threshold, points whose echo count is greater than a preset echo count threshold, points whose ranging confidence is lower than a preset confidence threshold, and points whose coordinates significantly exceed the scanner's effective range.
[0036] The preset intensity threshold, preset echo number threshold, and preset confidence threshold can be calibrated according to the scanning equipment model, on-site lighting conditions, and tower surface material, and the thresholds used are written into the result package.
[0037] Statistical outlier filtering removes points whose number of neighboring points is less than a preset value. Specifically, for each point, the number of neighboring points within a certain radius is counted. If the number of neighboring points is less than the preset minimum number of neighboring points, it is identified as an outlier and removed. Radius filtering removes points whose number of neighboring points within a given radius is insufficient. Statistical outlier filtering and radius filtering can be used individually or in combination. The neighborhood radius and minimum number of neighboring points can be calibrated based on the point cloud density and tower size.
[0038] Density equalization sampling is used to reduce the impact of point cloud density inhomogeneity on subsequent cross-sectional fitting. Since the point cloud density acquired by the scanner varies significantly at different distances and incident angles, direct equal-weighted fitting would result in denser regions dominating the fitting outcome. Therefore, density equalization sampling is performed on the point cloud to make the point cloud density more uniform across different regions of the tower surface. Density equalization sampling can employ voxel downsampling, random downsampling, or adaptive sampling based on local point spacing.
[0039] Normal vector estimation is used to provide a basis for subsequent observation weight calculation. For each point, the normal vector of that point is obtained by fitting a local plane based on its neighborhood point set. The neighborhood size can be determined based on the point cloud density and the curvature of the tower surface. The normal vector estimation results can also be used for subsequent tower point cloud segmentation and incident angle factor calculation. The calculation.
[0040] First, coarse registration is performed on the point clouds from multiple sites. Coarse registration is used to initially unify the point clouds of each site under the same field coordinate system, providing initial values for subsequent fine registration. Coarse registration can be performed using at least one of the following methods: matching the center of the controlled target, matching stable feature points manually, initial pose values based on satellite / inertial combined positioning results, and coarse alignment based on the approximate axis of rotation of the tower.
[0041] After coarse registration is completed, fine registration is performed. Fine registration is used to further reduce the registration residuals between stations, so that the point clouds of each station are accurately aligned in the overlapping areas.
[0042] Fine registration employs point-to-surface ICP (Iterative Closest Point) method, with the iteration termination condition being that the difference between two adjacent transformation matrices is less than a preset threshold, or the number of iterations reaches a preset maximum. After fine registration, the registration residuals in the inter-station overlap area are calculated. This serves as the input for subsequent observation weights.
[0043] Then, the registered point cloud is aligned according to the gravity vector. Rotate to Axis, to obtain gravity-aligned point cloud : .in, For original or pieced-together point clouds, To rotate the on-site gravity vector to A rotation matrix that unifies the axis and direction. Let be the coordinate translation vector. The unit vector of gravity is denoted as . When the coordinate system is aligned by gravity, . From the on-site gravity vector and The rotational relationship between the axes is determined. Specifically, the on-site gravity vector is first determined based on the scanner's built-in tilt compensation results, the calibrated inertial measurement unit, known leveling control points, or control targets set up on-site. Then calculate Rotate to Axial direction rotation matrix . The translation vector is determined by the control point coordinate translation. Specifically, the translation vector is determined based on the difference between the coordinates of the control points in the field coordinate system and the coordinates in the target coordinate system. . and The rotation matrix is determined jointly by the control points, equipment attitude, and registration results. Translation vector The gravity direction calibration residual is written into the result package.
[0044] In this embodiment, after obtaining the gravity-aligned point cloud Afterwards, quality gating is performed, including calculation of control point residuals, registration residuals in inter-station overlap areas, and point cloud coverage.
[0045] Specifically, control point residuals are used to evaluate the accuracy of point cloud registration and coordinate unification. Control point residuals can be calculated based on the difference between the coordinates of the control points in the registered point cloud and their known coordinates. Let the first... The known coordinates of the control points are: After registration, the coordinates are calculated as follows: Then the first The residuals at each control point are: .
[0046] The magnitude of its residual is: .
[0047] If there is If there are 1 control point, then the root mean square of the control point residuals is: .
[0048] An exemplary threshold for control point residuals can be derived from the scanner's nominal ranging error. Set, for example, not greater than Scanner nominal ranging error The quality threshold can be estimated by the scanner's nominal accuracy and repeated scanning of the target on site. The quality threshold is configured based on the equipment accuracy, tower type, and company testing procedures; no fixed value is assumed in this embodiment. If the equipment accuracy or tower type differs, the 95th percentile of at least three repeated scans at the same location can be used instead of the default threshold, and the used threshold should be written into the results package.
[0049] Registration residuals in overlapping areas between stations are used to evaluate the registration consistency of multi-site cloud systems in overlapping regions. For the first... For each station, corresponding point pairs are taken within the overlapping area, and the registration residuals of the inter-station overlapping area are calculated based on the root mean square value of the point-to-area residuals. .
[0050] Let the first In a pair of corresponding points, point The unit normal vector to the local plane of the point cloud at adjacent stations is A point on the plane is The point-to-surface residual is: .
[0051] If the first There are a total of [number] stations in the overlapping area. For each corresponding point pair, the registration residual in the inter-station overlap area is: .
[0052] Registration residuals in inter-station overlap areas The exemplary threshold can be determined based on the scanner's nominal ranging error. Set, for example, not greater than If the equipment accuracy or tower type is different, the 95th percentile of at least three repeated scans at the same location can be used instead of the default threshold.
[0053] Point cloud coverage is used to evaluate the effective coverage of point clouds on the tower surface, including height coverage and circumferential coverage. Height coverage can be defined as the ratio of the actual covered height within the effective height range of the tower to the nominal height of the tower: ,in, and These represent the lowest and highest altitudes actually covered by the point cloud, respectively. This refers to the nominal height of the tower.
[0054] Circumferential coverage is statistically calculated based on 5° azimuth sectors. For a given height section, let the number of sectors covered by the measured point cloud be... The measured circumferential coverage angle is: The circumferential coverage rate is: .
[0055] For the entire tower, the average or minimum circumferential coverage of each effective cross-section can be used as the evaluation index. An exemplary threshold for point cloud coverage can be set as follows: height coverage not less than 0.9, circumferential coverage not less than 0.8, or a single-layer circumferential coverage angle not less than 120°. Specific thresholds can be configured based on equipment accuracy, tower type, and company testing procedures.
[0056] If any of the following indicators—control point residual, inter-station overlap registration residual, and point cloud coverage—fail to meet the preset quality threshold, the process returns to step S1 for supplementary sampling or proceeds to manual review; it does not proceed to automatic qualification determination. The quality gating results, the thresholds used, and their versions are written into the result package to ensure that the same conclusion can be traced back to the corresponding original point cloud and calculation configuration.
[0057] In this embodiment, when performing tower point cloud segmentation, the input data includes gravity-aligned point clouds. This includes auxiliary information such as tower geometry and design information, wind turbine asset information, and scan registration information. Tower geometry and design information includes the nominal tower height. Diameter of the bottom of the tower Diameter of the top of the tower Tower design drawings or design section parameters. These are used for flange height indication and candidate flange area determination. Wind turbine asset information includes wind turbine asset coordinates, known spatial locations and design dimensions of the nacelle and blades. Scan registration information includes scan site number and registration residuals in the inter-site overlap area. , On-site ground elevation or foundation elevation, data collection time, unit operating status, and blade position code.
[0058] Specifically, tower point cloud segmentation includes: (1) Establish Region of Interest. Based on the wind turbine asset coordinates and the nominal tower dimensions, establish a region of interest to initially limit the search range of the tower point cloud. Specifically, this can be based on the center coordinates of the tower bottom and the nominal tower height. Diameter of the bottom of the tower and the diameter of the top of the tower Establish a cylindrical or frustum-shaped region of interest. The method for establishing the region of interest can be adjusted according to the tower design drawings and the actual site conditions.
[0059] For a tower with a constant diameter, a cylindrical region of interest can be established with a base radius of: ,in, The preset radius margin is used to cover the point cloud on the tower surface and a small number of auxiliary facilities.
[0060] For a conical tower, a frustum-shaped region of interest can be established, with a base radius of: The top radius is: The height range is: ,in, and These are the elevations of the bottom and top of the tower, respectively. This is the preset elevation margin.
[0061] (2) Exclude point clouds from the ground, nacelle, and blades. Exclude ground points based on elevation thresholds. Specifically, set ground elevation thresholds based on the site ground elevation or tower foundation elevation. The elevation is lower than Points in the point cloud are identified as ground points and excluded. Ground elevation threshold. It can be determined based on the on-site leveling control points, the design elevation of the foundation cap, or the point cloud elevation histogram. A preset elevation margin is used to avoid accidentally deleting point clouds at the bottom of the tower. Point clouds of the nacelle and blades are excluded based on their known spatial positions. Specifically, a spatial bounding box or cylinder for the nacelle and blades can be established based on their design dimensions, installation location, and operating status. Point clouds falling within this bounding box or cylinder are identified as nacelle or blade point clouds and excluded. For wind turbines in operation, blade positions change over time. The spatial position of the blades can be determined based on the acquisition time, unit operating status, and blade position coding, or the impact of blade movement on tower point cloud segmentation can be reduced by shortening the single-station scanning time.
[0062] (3) Tower surface extraction. Within the region of interest, the tower surface is fitted using a randomly sampled uniform cylindrical or frustum model. Combined with constraints on the normal vector direction and the radius changing slowly with height, the tower surface point cloud is extracted. Specifically, this can be achieved by: fitting the initial tower surface from the point cloud within the region of interest using a randomly sampled uniform cylindrical or frustum model; retaining points whose normal vectors are approximately perpendicular to the tower axis and removing points with significantly abnormal normal vector directions based on the normal vector direction constraint; removing points whose radii deviate significantly from the design value or exhibit significant jumps based on the radius changing slowly with height constraint; and removing isolated noise clusters through connected component analysis, retaining the point cloud connected to the tower surface to obtain the tower surface point cloud and the initial tower surface model. For conical towers, a frustum model can be fitted piecewise, or a frustum model with a radius that changes linearly with height can be used.
[0063] (4) Marking of Abnormal Candidate Points. Protrusions such as ladders, cables, flanges, platforms, and reflected noise are not directly deleted but marked as abnormal candidate points. Candidate flange areas are then marked according to the tower design drawings or design section parameters. Specifically, the distance from the point to the initial tower surface, local curvature, and continuous height range are calculated. Points meeting preset conditions are marked as abnormal candidate points. These preset conditions include: the distance from the point to the initial tower surface is greater than a preset distance threshold; the local curvature is greater than a preset curvature threshold; the point appears continuously within a continuous height range, and the height range is less than a preset height threshold; the point is located near the flange design height, and the radius abrupt change exceeds a preset threshold. Abnormal candidate points are assigned lower weights in subsequent robust fitting rather than being directly deleted, to preserve the traceability of evidence.
[0064] For known structures such as flanges, the flange indicator can be established based on the flange height given in the tower design drawings. In other embodiments, when there is no tower design drawing, if the deviation of the difference between adjacent radii relative to its local median exceeds a preset threshold, and the anomaly only occurs within a preset continuous height range, it is marked as a candidate flange area and manual verification is required.
[0065] By segmenting the tower point cloud, we can obtain the tower surface point cloud, anomaly candidate point markers, candidate flange area markers, initial tower surface model or initial axis, region of interest parameters, and exclusion results for ground, nacelle and blades, as well as the effective tower height.
[0066] In this embodiment, the effective height of the tower refers to the height range within which the tower surface can be effectively covered by the point cloud and used for cross-section fitting. If the bottom area is obstructed by the foundation or flange, and the top area is obstructed by the nacelle, the lowest and highest effective heights can be taken as the start and end ranges of the effective height, respectively, and extrapolated to the nominal detection interval according to the model.
[0067] Specifically, the construction of layered sections and robust fitting of section centers include: along The shaft is based on a preset layer thickness. The effective height of the tower is divided into layers, resulting in multiple sets of height section points. Layer thickness. The setting can be set from 0.5m to 2m depending on the tower height and required testing accuracy. The height of each layered section is denoted as , No. Layer section point set Falling from height The tower-shaped clouds are composed of point clouds. After projecting onto a horizontal plane, a circular or slightly elliptical cross-sectional model is fitted. For conical towers, the radii of each layer are estimated independently, and the center sequence is used for centerline calculation; therefore, the radii of the upper and lower sections are not required to be the same. Weighted least squares with robust loss functions, random sample consistency, or iterative reweighting methods are employed to reduce the impact of appendages, occlusion edges, and sparse outliers.
[0068] No. The center of the layer fitting circle is denoted as: , No. The layer fitting radius is denoted as . No. The objective function for layered circle fitting is: Among them, the parameter to be determined is the first... Layer fitting center and the first Layer fitting radius. For the th layer... Each point in the layer Calculation points The horizontal distance to the center of the circle to be estimated, i.e. Subtract the radius to be estimated This yields the radial residuals. If the point is on the circle, the residual is 0; if the point is outside the circle, the residual is positive; and if the point is inside the circle, the residual is negative. For the first Layer The weight of each point; This is a robust loss function, meaning it is insensitive to outliers. It means looking for a group This minimizes the sum of the weighted robust losses for all points. Use Huber, Tukey, or other robust loss functions. The settings are based on point cloud density, incident angle, normal vector, site origin, and anomaly candidate markers. For example, points in dense areas are more reliable; lasers hitting the surface perpendicularly are more reliable, while grazing incidence is unreliable; the direction of the normal vector can help determine whether a point is on the tower surface; registration quality varies between different scanning stations; points marked as appendages or occluded edges have reduced weights. The output of layer-by-layer independent fitting is the fitting circle center, fitting radius, and cross-sectional validity for each layer.
[0069] In this embodiment, the effectiveness of the cross-section is determined at least based on the number of effective points, the circumferential coverage angle, the cross-section fitting residual, and the radius continuity. Specifically, the first... A layer can be marked as a valid section only if it meets the following conditions: (1) Number of valid points Not less than 200; (2) Measured circumferential coverage angle based on 5° azimuth sector statistics Not less than 120°; (3) Cross-section fitting residual Let be the root mean square value of the radial residuals from each point to the fitted circle, and be the th . Cross-sectional fitting residuals of robust layer fitting (4) Central standard uncertainty Not greater than .in, Estimated by the scanner's nominal ranging error and repeated scanning of the target on site; For the first The center of the layer The covariance matrix describes where the center of the circle is located. direction and Variance in direction and the correlation between the two; To obtain the trace of this covariance matrix, i.e. Directional variance addition Directional variance; For the first The standard uncertainty of the center of the circle obtained from layer fitting within the horizontal plane, expressed in units of length. (Center standard uncertainty) The smaller the value, the more accurate and reliable the center of the circle fitted by this layer is.
[0070] If the equipment accuracy or tower type is different, the 95th percentile of at least three repeated scans at the same location can be used to replace the above default threshold, and the threshold used should be written into the result package to ensure traceability.
[0071] In the radius continuity judgment, when the continuous invalid height reaches Automatic judgment is stopped and re-sampling or manual verification is triggered; the original observation and invalidity reason of a single invalid layer are retained, and the interpolated value is not used to replace the measured value in the solution.
[0072] Meanwhile, abrupt changes in flange radius are addressed through a combination of structural priors and data evidence. An indicator is established based on the flange height given in the tower design drawings. Without a design drawing, if the deviation of the difference between adjacent radii relative to their local median exceeds... And this anomaly only occurs when the number of cases does not exceed a certain threshold. If it appears within a continuous height range, it is marked as a candidate flange area and manual verification is required. Among these, This represents three times the standard deviation of the radius difference. This represents 0.2% of the median radius. When (i.e., in the non-flag zone), a continuity penalty is applied, and the radius continuity penalty function is: Radius difference between adjacent layers It should be close to the rate of change of the design radius. Multiply by layer thickness . For deviation Loss penalty yes The turning point value. The radius change of the non-flange zone should be smooth and conform to the design taper. When (i.e., in the flange area), cancel the cross-layer continuity term and replace it with a constraint. This means that the flange area is allowed to have a sudden change in radius, but the magnitude of the change must be close to the design value. The error does not exceed If there is no design value, the requirement cannot be automatically relaxed and will be transferred to manual review.
[0073] Furthermore, if the design drawing does not specify the allowable mutation amount, the constraints should not be automatically relaxed, and the candidate region section will be subject to manual review. Therefore, controlled mutations only affect the radius continuity term and do not cancel the center continuity, observation residuals, or quality gating.
[0074] In this embodiment, the cross-section center fitting in step S3 adopts joint optimization coupled across height, rather than independent layer-by-layer fitting. Conventional independent layer-by-layer fitting schemes are prone to isolated jump points when a certain layer is locally obstructed or interfered with by accessories. In this embodiment, the initial center and radius of each layer are used as the parameters to be estimated, and single-point radial residuals, second-order difference constraints of the centerline, and constraints on the continuous change of tower radius with height are simultaneously added to the same objective function to jointly solve for all effective heights.
[0075] Specifically, the objective function for joint optimization is: ; Among them, the first item Used to fit the measured cross section, so that the center and radius of the circle are as close as possible to the actual point; For the first Layer From the point to the center of the circle to be estimated Horizontal distance, For the first Layer Radial residuals at each point; It is the observation weight. It is a robust loss function.
[0076] Second item Used to suppress center jumps that do not conform to the continuity of the tower structure, so that the centers of adjacent layers do not suddenly shift; The second difference of the circle center sequence is the largest; the smaller the second difference, the smoother the center line. To determine the constraint weights based on interlayer spacing, equipment accuracy, and tower type, The larger the diameter, the smoother the center line.
[0077] Third item It is used to constrain the changes in adjacent radii, allowing for a smooth transition between the radii of adjacent layers, but not to allow controlled abrupt changes in known structural locations such as flanges; It is a radius continuity penalty function. To determine the constraint weights based on interlayer spacing, equipment accuracy, and tower type, The larger the radius, the smoother the change. This mechanism allows well-informed adjacent sections to provide structural constraints on locally weak observation layers, but does not replace the original observations with artificial interpolation.
[0078] The observation weights are not a single constant, but are jointly determined by circumferential coverage, incident angle, point density, station registration residuals, and outlier candidate markers. We can set: ;in, Circumferential coverage factor; The incident angle factor; Point density factor; Register residual factors for the sites; These are candidate anomaly markers. They are normalized at each layer, with the sum of the layer weights being 1.
[0079] Therefore, stations with poor registration quality, grazing incidence points, sparse regions, and candidate points for attachments in multi-station overlapping areas are automatically downweighted to avoid systematic shifts caused by simple stitching and equal-weight fitting.
[0080] Each quality factor is calculated using the following formula and normalized at each level: , ,in, For the first Peripheral coverage angle of the layer For the ideal coverage angle, For surface normal, The direction of laser incidence; ,in, The local average point spacing, The distance between midpoints in the same layer; ,in, The registration root mean square residual of the point is calculated from the registration residual of the overlapping area between the point and the adjacent points. Take each station median and The larger value in; The weight of abnormal points is significantly reduced by assigning a value of 1 to normal points and 0.1 to candidate points of attachments or occluded edges.
[0081] Based on this, the normalized weights are: .
[0082] The inflection point of the loss is taken as: ; ; in, Estimate using the absolute deviation of the median; For radial residuals, ; It is the median of the residuals. It is the absolute value of the residual deviation from the median. The above definition allows the weights to be calculated directly from the original point attributes and registration results, rather than being subjectively specified by the operator.
[0083] Joint optimization employs the following steps: (1) Initialize with the RANSAC circle fitting results of each layer , That is, Random Sample Consensus (RANSAC) is used to fit a relatively robust initial circle center and radius layer by layer as the starting point for joint optimization; (2) Use the radius median Interlayer spacing and Dimensionless treatment is applied to the radial residual, the central second-order difference, and the radius difference. (3) In the candidate set , … Search in each and The selection criterion is to minimize the prediction residuals obtained by three repeated scans or one valid layer validation; when there is no duplicate data, the inflection point of the L-curve is used, and the candidate values, scores and selected values are saved. (4) Fix the current weights, use Gauss-Newton or Levenberg-Marquardt to solve for the increments, and then update the Huber weights according to the new residuals; (5) When the relative change of the objective function between two consecutive times is less than 10 -6 And the parameter increment norm is less than 10 -5 If the rmed function converges, it can be iterated up to 100 times; if it does not converge or the Hessian matrix condition number is greater than 10¹², it is switched to the complex kernel.
[0084] To avoid local solutions, the algorithm is started twice, using the results of layer-by-layer circle fitting and the results of layer-by-layer circle fitting smoothed along the height. The algorithm that satisfies the constraints and has a smaller objective function is selected.
[0085] The covariance of the central parameter is obtained from the weighted Jacobian matrix at the convergence point: ; in, The unit weighted variance is estimated by dividing the weighted sum of squared residuals by the effective degrees of freedom. For a weighted Jacobian matrix, This is the weight matrix. It is a central second-order difference matrix. This is the radius difference matrix. A sandwich covariance correction can be used for the Huber weights to make the covariance estimation more robust. Let be the covariance matrix of all parameters, and be the standard uncertainty of all parameters to be estimated. The covariance of the center of each layer is denoted as Covariance of each layer's center From Extract the corresponding , 2×2 submatrix, covariance of the center Fitting the center of each layer ,radius and It is then passed on to the subsequent decoupling and gating steps.
[0086] Using the above method, step S3 ultimately outputs not only the center of the fitted circle for each layer, but also... and radius It also outputs the covariance of the center of each layer. This covariance will be used for the subsequent identifiable decoupling of overall tilt and local deflection, as well as for gating three-state conclusions driven by measurement uncertainty.
[0087] Unlike independent circle fitting for each cross section, the objective function of joint optimization simultaneously considers radial fitting residuals, central second-order continuity, and radius continuity. Observation weights are generated using formulas for determining circumferential coverage, incident angle, point density, station registration residuals, and anomaly markers. Constraint weights are selected using repeated scanning, leave-one-out verification, or L-curves. The central covariance is obtained through double initial value iteration and convergence / condition number gating. Structural constraints are formed on weak observation layers using adjacent reliable cross sections, while preserving the traceability of flange mutations and invalid layers.
[0088] In step S4 of this embodiment, the construction of the centerline model and the calculation of perpendicularity, tilt angle, and deflection specifically include: Center of all effective cross sections The central sequence is formed by sorting by height. Based on the testing objectives, an overall axis model and a local deflection model are constructed. The overall axis model can be fitted using a weighted three-dimensional straight line to reflect the overall tilt direction of the tower; the local deflection model can use a low-order polynomial, piecewise splines, or a smooth curve that satisfies the continuity of the tower structure to reflect the local bending of the tower. The fitting weights are combined with the number of effective points. Circumferential coverage angle Cross-section fitting residuals Sure.
[0089] Specifically, fitting the center line for ,in, Indicates a specific height position of the tower; Indicates height The fitted center coordinates at the location; Indicates at altitude The fitted circle center Axis coordinates Indicates at altitude The fitted circle center Axis coordinates; fitted centerline The actual centerline (curve) connecting the effective cross-section centers of each layer includes both overall tilt (the angular deviation of the overall fitting axis of the tower relative to the vertical direction of gravity) and local deflection (the local lateral deviation of the fitting centerline of the tower relative to the overall fitting axis).
[0090] like Figure 2 As shown, the overall axial direction The overall axial direction vector from bottom to top. From the top center Subtract the bottom center Then normalize to obtain the overall tilt direction of the tower. If the bottom area is obstructed by the foundation or flange, the minimum and maximum effective heights can be used and extrapolated to the nominal detection range according to the model.
[0091] Overall tilt angle of the tower It is calculated from the angle between the overall axial direction vector and the unit vector of gravity. ,in, This is the unit vector of gravity, when the coordinate system is aligned with gravity. Overall tilt angle That is, the overall axis and The included angle of the axis.
[0092] high The horizontal offset at point is denoted as , .in, Indicates height Fitting center line Relative to the reference plumb line or reference center line Horizontal offset. It can be taken as passing through the reference center at the bottom of the tower and parallel to the direction of gravity. The plumb line can be used, or an approved as-built baseline can be used as a comparison benchmark. The top horizontal offset is denoted as... The maximum horizontal offset is denoted as The verticality ratio can be expressed as .in, This is the projection matrix (horizontal projection operator). We obtain a three-dimensional vector representing the height. The offset of the current center in space relative to the reference center.
[0093] The overall fitting axis is denoted as To fit using a straight line The overall centerline of the tower obtained by fitting the center of the fitted circle of all effective sections only reflects the overall tilt trend and does not reflect local bending.
[0094] high The local deflection at point is denoted as , .in, Indicates the fitted center line Relative to the overall fitting axis Local deflection (local lateral deviation) is used to separate local bending from overall tilt. The orientation angle is determined by the horizontal offset vector. Orientation calculation in a plane. The projection matrix (lateral projection operator) represents projecting a vector onto a plane perpendicular to the global axis. On the plane, remove the component along the overall axis direction; We obtain a three-dimensional vector representing the height. The current center relative to the overall axis The offset, where the offset perpendicular to the overall axis is the local deflection. The maximum local deflection is denoted as... .
[0095] In this embodiment, the overall tilt and local deflection in step S4 are calculated using a recognizable decoupling method. When fitting independently, fitting with only a straight line will lose the local curvature, while fitting with only a smooth curve may mix the overall tilt into the local deflection. This embodiment writes the horizontal centerline as the sum of the overall linear term and the local smoothing margin, and applies zero mean and zero linear trend constraints to the local margin, thereby avoiding arbitrary transfer between the two components and obtaining a unique decomposition with clear physical meaning.
[0096] Specifically, let the horizontal centerline be represented as And impose the following constraints: The weighted mean of the local surplus is zero. ; The weighted linear trend of the local surplus is zero. .
[0097] in, For height The horizontal center coordinates of the location; For constant terms; The coefficient of the overall sloping linear term; This is a local smoothing margin; For the first Layer weights are assigned to each cross section, with higher-quality layers having larger weights. The horizontal centerline is the fitted centerline. The projection trajectory in the horizontal plane.
[0098] make Design matrix The Behavior Using cubic B-spline basis matrix express , Let be the B-spline coefficient vector, and let represent the local smoothing margin using B-splines. Layer weights are determined by the following formula:
[0099] in, For coverage and residual gating factors, For the first Layer center covariance traces, Pick To avoid division by zero, normalization is then performed. .
[0100] By solving the following constrained least squares problem, the global linear term and local residual parameters are obtained:
[0101] And apply simultaneously:
[0102]
[0103] in, This is the layer weight matrix. A vector consisting of the horizontal center coordinates of all layers. To design the matrix and the first Action is This corresponds to the first-order term of the whole; For smoothing parameters, It is a second-order difference matrix. Two constraints eliminate constant translation and linear tilt in the local residuals, respectively, making... , and They cannot be arbitrarily transferred to each other. The overall tilt angle is determined by... Calculations show that the local deflection is caused by... calculate.
[0104] The B-spline uses a cubic basis function, with repeated nodes at the endpoints; the initial spacing of the internal nodes is taken as... Each node interval must contain at least three valid sections; otherwise, adjacent intervals are merged. Smoothing parameters. exist to The minimum value is selected from the logarithmic candidate set using generalized cross-validation. The constrained problem is solved in one step using the KKT equations; if the condition number of the KKT matrix is greater than... If the condition is still not met, the adjacent node intervals with the fewest valid points are merged sequentially and recalculated. If the condition is still not met, manual review is required. The same layer weights are used in both the X and Y directions. B-spline basis matrix and smoothing parameters Solve separately to keep the directional components comparable.
[0105] In this embodiment, the data transmission path is the output of S3. sum and difference Sure and ; Determining the effective layer label for joint optimization The line; decoupling the solution output , and its covariance It is obtained from the corresponding sub-block of the inverse matrix of the KKT system. This is then used for the propagation of uncertainties in tilt angle, maximum local deflection, and orientation angle. Therefore, any low-quality section will affect geometric reconstruction, component decoupling, and the final conclusion through the same weight chain, ensuring that errors are not locally masked but are traceable throughout the entire process. It corresponds to the inverse matrix of the KKT system. The sub-blocks are then multiplied by the unit weight variance.
[0106] In the above manner, S4 not only outputs the overall tilt angle Top horizontal offset and local deflection at various heights It also outputs the covariance of the decoupling parameters. This covariance will be used for gating the three-state conclusions driven by subsequent measurement uncertainty.
[0107] This embodiment also includes historical data alignment and deformation trend assessment.
[0108] Specifically, for the test results of the same wind turbine at different times, the point cloud of each period is first transformed to a common reference by using control points that do not deform with the tower, stable areas of the foundation platform, or fixed objects in the site; then the centerline is sampled at a uniform height to avoid misalignment caused by different layer starting points and layer thicknesses.
[0109] Let the center line of this period be The base period is Then the intertemporal level change is: ; right , , , A time series analysis is established based on the direction of change, and trend classification can be performed using the rate of change, continuous multi-period monotonicity, and quality confidence level. When historical changes exceed the thresholds specified in the enterprise's procedures, a retest recommendation or alarm is generated, but the final safety judgment is still made by qualified personnel in combination with design limits, environmental conditions, and other structural test results.
[0110] In this embodiment, the trend judgment and conclusion output adopt a three-state gating driven by measurement uncertainty, instead of directly comparing the calculated value with the threshold.
[0111] Current automatic detection methods often directly compare calculated values with thresholds, ignoring differences in occlusion levels, registration residuals, and model stability. This implementation method obtains the index from the cross-sectional center covariance, registration residual propagation, or repeated resampling. Expanded uncertainty The system automatically categorizes conclusions into three states: "normal, alarm, and retest / manual review," instead of forcing binary output.
[0112] Specifically, if If it is normal; if If the result is positive, it will be an alarm; otherwise, it will be a retest / manual review.
[0113] in, Limits given by enterprise procedures, design documents, or acceptance criteria are not learned by the algorithm itself; This is the coverage factor.
[0114] Before automatic judgment, the system checks the effective height, effective cross-sectional ratio, joint optimization convergence state, and covariance condition number. If any item fails to meet the requirements, a retest / manual review is directly output. For indicators that pass the pre-qualification quality check, a closed interval is used. determination: The upper bound is strictly less than Normal; lower bound strictly greater than For alarm purposes; contact or crossing of sections. For retesting / manual verification. For directional bilateral limits, the same rule is applied to both boundaries.
[0115] Uncertainty is calculated along a traceable path. For parameters... Calculated indicators First, calculate the gradient at the convergent solution: Then calculate the model components: .
[0116] 6-DOF pose covariance for control point or station registration Jacobi through coordinate transformation The spread is as follows: .
[0117] Gravity direction calibration covariance By tilt angle pair Jacobi propagation is as follows: .
[0118] Quantization resolution and other independent components are denoted as The combined standard uncertainty is: .
[0119] Generally taken This results in an expanded uncertainty with approximately 95% coverage: .
[0120] When the effective degrees of freedom are less than 30, calculate veff according to the Welch-Satterthwaite formula and take t0.975, veff instead of 2.
[0121] To verify the linear propagation results, a two-tiered bootstrap resampling process was conducted 200 times, categorized by site and cross-section: first, sites were sampled with replacement, and then effective cross-sections were sampled with replacement within each site, with joint optimization and decoupling repeated sequentially. A fixed random seed was used, and the seed, effective resampling counts, and index sequences were saved. If the number of effective resampling counts was less than 180, or the ratio of the bootstrap standard deviation to the linear propagation standard deviation was not between 0.8 and 1.25, the larger value was taken, and the conclusion was downgraded to a verification result. In this way, cross-section covariance, registration error, and model nonlinearity are all subject to the same gating.
[0122] The algorithm sequence for three-state gating is as follows: B1. Read the algorithm version corresponding to this time. , and quality threshold; B2. Calculation and And record each uncertainty component; B3. Construct the measurement interval; B4. Based on intervals and The location output indicates normal status, alarm status, or verification status. B5. Write the input point cloud index, weights, parameter covariance, threshold version, and decision reason into the result package.
[0123] When thresholds are missing, versions are inconsistent, or input conditions exceed the calibration range, the default safety conclusions are not allowed to be used.
[0124] In this embodiment, for the detection results of the same wind turbine at different times, a common baseline is first established in the point cloud of each period at the stable control point or foundation abutment area, and then the center line is resampled according to a uniform height grid; the change in the current period and the baseline period simultaneously carries the uncertainty of both periods. Specifically, the combined uncertainty of the horizontal change across periods can be taken as:
[0125] in, This represents the uncertainty for the current period. The uncertainty is set at the baseline period, and the same three-state rule as described above is used to determine whether the change is significant. This avoids misinterpreting different stratification starting points, differences in environmental conditions, or low-quality data from a single period as long-term deformation.
[0126] In this embodiment, the covariance of the center is... Together with the effective gating factor, the layer weights in the decoupling step are determined. The covariance of the overall tilt parameters and local deflection parameters output from the decoupling step is further passed to the uncertainty gating. The uncertainty gating, combined with the registration error, gravity direction calibration error, and resolution error, outputs a normal, alarm, or retest / manual verification conclusion.
[0127] In the above manner, step S5 not only outputs the inter-period change amount and trend classification, but also combines the uncertainty of the current period and the base period into the uncertainty of the change amount, and uses three-state gating to determine whether to output a normal, alarm or retest / manual verification conclusion.
[0128] In this embodiment, the result output and platform integration specifically include: The detection indicators, quality levels, and alarm statuses output by the three-state gating are generated into a structured detection result package. The result package includes at least: fan and task identifier, acquisition equipment and time, coordinate reference, environment and operating status, original point cloud index, processing parameters, effective height range, center and radius of each layer, cross-sectional residual and coverage, overall tilt angle, orientation angle, top offset, maximum horizontal offset, maximum local deflection and its height, historical variation, quality level, and alarm status.
[0129] The system simultaneously generates a centerline diagram, various height offset curves, cross-section fitting sampling diagrams, and an interactive 3D point cloud view.
[0130] The inspection platform interface module receives result packets via file upload, message queue, or application programming interface (API), and associates them with the wind turbine asset ledger and inspection tasks. When a quality level fails to meet the requirements, a supplementary data collection task is created; when an indicator exceeds a threshold, a retest or defect work order is created. The platform saves algorithm versions, threshold versions, and manual review records, ensuring that the same conclusion can be traced back to the corresponding original point cloud and computational configuration.
[0131] In this embodiment, the method further includes quality assessment and anomaly handling, performing quality checks on acquisition, registration, cross-section, model, working condition and conclusion respectively, and outputting automatic detection conclusion only when all quality parameters simultaneously reach the threshold.
[0132] Specifically, quality assessment and anomaly handling include at least one of the following: Data Acquisition Quality Check: Check the effective height coverage of the tower, circumferential coverage angle, point density, scanning blind zone, and the number and distribution of control points. If the effective height coverage is insufficient, the circumferential coverage angle is too small, the point density is too low, there are unrecorded scanning blind zones, or the number and distribution of control points do not meet the requirements, return to S1 for supplementary acquisition or transfer to manual review.
[0133] Registration quality check: Check the consistency of control point residuals, overlapping area point-to-area residuals, and independent results of each station; if the limits are exceeded, re-register or re-sample. When the consistency of control point residuals, overlapping area point-to-area residuals, or independent results of each station does not meet the preset threshold, return to S1 for re-registration or re-sample.
[0134] Section quality check: Check the number of valid points, coverage angle, circle / ellipse fitting residual, radius continuity with height, and adjacent center jumps. When the number of valid points is insufficient, the coverage angle is too small, the fitting residual is too large, the radius is discontinuous with height, or the adjacent center jump exceeds the threshold, the section is marked as invalid, and its original observations and invalid reasons are retained according to the rules in S3.
[0135] Model quality check: The stability of the centerline parameters is assessed using leave-one-out validation, repeated fitting, or autopilot resampling, and a confidence interval or quality score is output. When the stability of the centerline parameters does not meet the requirements, a lower quality score or a wider confidence interval is output, and the conclusion is downgraded to retesting or manual verification.
[0136] Operating condition control: Operating conditions are marked when there is strong wind, significant tower vibration, rain / fog, or abnormal equipment posture, and dynamic oscillation snapshots are not directly interpreted as permanent deformation. In a preferred embodiment, for data collected during operation, motion distortion caused by tower vibration is reduced by shortening single-station scan time, multi-frame time synchronization, or point cloud timing filtering, and operating condition information such as wind speed, temperature, and unit operation / shutdown status is recorded in the results package.
[0137] Conclusion Gating: Automatic detection conclusions are output only when the quality of acquisition, registration, cross-section, and model all reach the threshold simultaneously; otherwise, the status of supplementary acquisition, retesting, or manual verification is output.
[0138] Through the above quality assessment and anomaly handling steps, if the quality of any step fails to meet the threshold, an automatic qualified conclusion will not be output. Instead, the process will return to supplementary sampling, retesting, or manual verification, thereby avoiding giving misleading conclusions when there is insufficient coverage, large registration error, unreliable cross-sectional fitting, or unstable model.
[0139] The wind turbine tower verticality detection method in this embodiment acquires high-density three-dimensional geometric data of the entire tower height in a non-contact manner, reducing the need for high-altitude sampling and manual reading. Simultaneously, gravity orientation calibration and unified coordinate transformation reduce the risk of instrument attitude or multi-station coordinate errors being misjudged as tower tilt. Based on this, robust fitting of layered sections and quality gating improve the stability of center extraction under conditions of accessories, obstructions, flanges, and outliers, while simultaneously distinguishing between overall tilt and local deflection. It outputs a centerline and indicators that vary with height. First, the cross-sectional center is jointly solved using multi-source observation quality and tower height continuity. Then, identifiable constraints decompose the centerline into overall tilt and local deflection. Finally, model uncertainty is introduced into threshold judgment, providing a more complete solution than single-point or single-angle tilt analysis. These three mechanisms transmit cross-sectional weights and parameter covariance before and after, preventing outliers, obstructions, and registration errors from being amplified into erroneous alarms in subsequent stages. Comparability across periods is achieved through a common benchmark and resampling at the same elevation, while retaining the original point cloud, algorithm version, and parameters for easy traceability and verification. Ultimately, it forms a closed loop with the new energy inspection platform for supplementary sampling, retesting, alarms, and work orders, improving the automation level of batch wind turbine testing.
[0140] The foregoing has described specific embodiments of the present invention. In some cases, the actions or steps described in the embodiments of the present invention may be performed in a different order than that shown in the embodiments and the desired results may still be achieved. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0141] Based on the same concept, embodiments of the present invention also provide a wind turbine tower verticality detection system. This system is applied to servers. For example... Figure 3 As shown, the wind turbine tower verticality detection system includes: a point cloud acquisition and registration module 100, a tower segmentation and anomaly marking module 200, a cross-height joint optimization module 300, an identifiable decoupling module 400, and an uncertainty three-state gating module 500. Among them, The point cloud acquisition and registration module 100 is used to acquire the three-dimensional laser point cloud of the wind turbine tower under test and the calibration data for gravity direction determination and inter-station registration. The three-dimensional laser point cloud is preprocessed, multi-station registered and gravity aligned to obtain the gravity aligned point cloud.
[0142] The tower segmentation and anomaly marking module 200 is used to segment the tower surface point cloud from the gravity-aligned point cloud and mark the protrusions on the wind turbine tower under test as anomaly candidate points.
[0143] The cross-height joint optimization module 300 is used to divide the effective height of the tower into layers along the Z-axis with a preset layer thickness in a coordinate system where the Z-axis is consistent with the vertical direction of gravity on site. It performs cross-sectional fitting on the point set of each layer of the point cloud on the surface of the tower, and performs cross-height joint optimization with the fitting center and fitting radius of each layer as parameters to be estimated. The objective function of the cross-height joint optimization includes at least a single-point radial residual term, a continuity constraint term for the center of adjacent cross sections, and a continuity constraint term for the radius of adjacent cross sections. Solving the objective function yields the fitting center, fitting radius, center covariance, and observation weight of each effective cross section.
[0144] The identifiable decoupling module 400 is used to sort the fitted circles of each effective section by height to form a center sequence, and to perform identifiable decoupling of the center sequence for overall tilt and local deflection to obtain the overall tilt parameter, local deflection parameter and decoupling parameter covariance. The identifiable decoupling includes: representing the horizontal coordinate of the center sequence as the sum of the overall linear term and the local smoothing margin, and applying a weighted zero mean constraint and a weighted zero linear trend constraint to the local smoothing margin to uniquely separate the overall tilt component and the local deflection component.
[0145] The uncertainty three-state gating module 500 is used to determine the uncertainty of the detection index based on the center covariance and observation weight obtained in S3, and the overall tilt parameter, local deflection parameter and decoupling parameter covariance obtained in S4. The detection index includes at least one of the following: overall tilt angle, top horizontal offset, maximum horizontal offset, maximum local deflection and its direction. The measurement interval formed by the detection index and the uncertainty is compared with a preset threshold, and the detection result is output as one of three states: normal, alarm or retest / manual verification.
[0146] For ease of description, the above system is described by dividing it into various modules based on their functions. Of course, in implementing the embodiments of the present invention, the functions of each module can be implemented in one or more software and / or hardware.
[0147] The system described in the above embodiments is applied to the corresponding method in the foregoing embodiments and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.
[0148] Based on the same inventive concept, embodiments of the present invention also provide an electronic device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the method described in any of the above embodiments.
[0149] This invention provides a non-volatile computer storage medium storing at least one executable instruction that can execute the method described in any of the above embodiments.
[0150] Figure 4 This embodiment illustrates a more specific hardware structure of an electronic device, which may include a processor 501, a memory 502, an input / output interface 503, a communication interface 504, and a bus 505. The processor 501, memory 502, input / output interface 503, and communication interface 504 are interconnected internally via the bus 505.
[0151] The processor 501 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present invention.
[0152] The memory 502 can be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 502 can store the operating system and other application programs. When the technical solution provided by the method embodiment of the present invention is implemented by software or firmware, the relevant program code is stored in the memory 502 and is called and executed by the processor 501.
[0153] Input / output interface 503 is used to connect input / output modules to realize information input and output. Input / output modules can be configured as components in the device (not shown in the figure) or externally connected to the device to provide corresponding functions. Input devices may include keyboards, mice, touch screens, microphones, various sensors, etc., and output devices may include displays, speakers, vibrators, indicator lights, etc.
[0154] Communication interface 504 is used to connect a communication module (not shown in the figure) to enable communication between this device and other devices. The communication module can communicate via wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).
[0155] Bus 505 includes a pathway for transmitting information between various components of the device (e.g., processor 501, memory 502, input / output interface 503, and communication interface 504).
[0156] It should be noted that although the above-described device only shows the processor 501, memory 502, input / output interface 503, communication interface 504, and bus 505, in specific implementations, the device may also include other components necessary for normal operation. Furthermore, those skilled in the art will understand that the above-described device may only include the components necessary for implementing the embodiments of the present invention, and does not necessarily include all the components shown in the figures.
[0157] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of this application is limited to these examples; within the framework of this application, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of the different aspects of this application as described above, which are not provided in detail for the sake of brevity.
[0158] This application is intended to cover all such substitutions, modifications, and variations that fall within the broad scope of the embodiments of this invention. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the embodiments of this invention should be included within the protection scope of this application.
Claims
1. A method for detecting the verticality of a wind turbine tower, characterized in that, include: S1: Collect the three-dimensional laser point cloud of the wind turbine tower under test and the calibration data for gravity direction determination and inter-station registration. Perform preprocessing, multi-station registration and gravity alignment on the three-dimensional laser point cloud to obtain the gravity-aligned point cloud. S2: Segment the tower surface point cloud from the gravity-aligned point cloud and mark the protrusions on the wind turbine tower under test as candidate points of anomalies; S3: In a coordinate system where the Z-axis is aligned with the vertical direction of gravity on site, the effective height of the tower is divided into layers along the Z-axis with a preset layer thickness. Cross-sectional fitting is performed on the point set of each layer of the point cloud on the surface of the tower. The fitting center and fitting radius of each layer are used as parameters to be estimated for joint optimization across the height. The objective function of the joint optimization across the height includes at least a single-point radial residual term, a continuity constraint term for the center of adjacent cross sections, and a continuity constraint term for the radius of adjacent cross sections. Solving the objective function yields the fitting center, fitting radius, center covariance, and observation weight of each effective cross section. S4: Sort the fitted circle centers of each effective section by height to form a center sequence, and perform identifiable decoupling of the overall tilt and local deflection on the center sequence to obtain the overall tilt parameter, local deflection parameter and decoupling parameter covariance. The identifiable decoupling includes: representing the horizontal coordinate of the center sequence as the sum of the overall linear term and the local smoothing margin, and applying a weighted zero mean constraint and a weighted zero linear trend constraint to the local smoothing margin to uniquely separate the overall tilt component and the local deflection component. S5: Based on the center covariance and observation weights obtained in step S3, and the overall tilt parameter, local deflection parameter, and decoupling parameter covariance obtained in step S4, determine the uncertainty of the detection index. The detection index includes at least one of the following: overall tilt angle, top horizontal offset, maximum horizontal offset, maximum local deflection, and direction. Compare the measurement interval formed by the detection index and the uncertainty with a preset threshold, and output a detection result in one of three states: normal, alarm, or retest / manual verification.
2. The method for detecting the verticality of a wind turbine tower as described in claim 1, characterized in that, The tower geometry and design information includes the nominal height of the tower. Diameter of the bottom of the tower Diameter of the top of the tower The tower design drawings or design section parameters, the wind turbine asset information including wind turbine asset coordinates, known spatial locations and design dimensions of the nacelle and blades, and the scan registration information including scan site number and registration residuals in the inter-site overlap area. Step S2 includes: The region of interest is established based on the coordinates of the wind turbine assets and the nominal dimensions of the tower, and the bottom radius, top radius and effective height of the tower of the wind turbine to be measured are obtained. Set a ground elevation threshold based on the site elevation or the elevation of the tower foundation cap. The elevation is lower than The point cloud data was identified as ground points and excluded. To pre-determine the elevation margin, and based on the known spatial location, design dimensions, and installation location of the nacelle and blades, a spatial bounding box or bounding cylinder for the nacelle and blades is established. Point clouds falling into the bounding box or bounding cylinder are identified as nacelle or blade point clouds and excluded, thus obtaining the exclusion results of ground, nacelle, and blade point clouds and the remaining point clouds. Within the region of interest, a random sampling consistency cylinder or frustum model is used to fit the initial surface of the tower from the point cloud of the region of interest. Outliers are eliminated according to preset constraints on the normal vector direction and the radius changing slowly with height. Isolated noise clusters are removed through connected component analysis, and the point cloud connected to the tower surface is retained to obtain the point cloud of the tower surface and the initial surface model of the tower. For at least one protrusion among ladders, cables, flanges, platforms, and reflected noise, calculate the distance from the point to the initial tower surface, the local curvature, and the continuous height range. Points that meet the preset conditions are marked as abnormal candidate points instead of being directly deleted, thus obtaining abnormal candidate point markings. Candidate flange areas are then marked according to the tower design drawing or design section parameters.
3. The method for detecting the verticality of a wind turbine tower as described in claim 2, characterized in that, Step S3, which involves dividing the effective height of the tower along the Z-axis into layers with a preset layer thickness, performing cross-sectional fitting on each layer of the point cloud on the tower surface, and using the center and radius of the fitted cross-section of each layer as parameters to be estimated, includes: along The shaft is based on a preset layer thickness. The effective height of the tower is layered, and the point cloud on the surface of the tower is divided into multiple height section point sets, the first... The set of points on the cross section falls from the height The tower-shaped clouds are composed of points; Project the point set of each cross section onto the horizontal plane, and independently fit each layer to obtain the initial fitting center, fitting radius and cross section fitting residual of each layer; Based on the preset cross-section validity judgment conditions, the cross-sections obtained by independent fitting layer by layer are judged for validity. Layers that meet the conditions are marked as valid cross-sections, and layers that do not meet the conditions are marked as invalid cross-sections and the reasons for invalidity are recorded. Flange area processing is performed based on the radius continuity constraint according to the flange indication, and a radius continuity penalty is applied to the flange area. For illegal orchid areas, the cross-layer radius continuity term is removed and replaced with a constraint. The flange indication is established based on the flange height given in the tower design drawing. The difference in radius between adjacent layers. For the rate of change of the design radius, for The turning point value; Using the initial center and radius of each layer as parameters to be estimated, the joint optimization solution is applied to all effective height layers according to the preset convergence conditions, yielding the jointly optimized center ci and radius ri of each layer. The objective function of the joint optimization is: in, For the first Layer From the point to the center of the circle to be estimated Horizontal distance, It is the observation weight. It is a robust loss function. and To constrain the weights, It is a radius continuity penalty function; according to Calculate the covariance matrix of all parameters. Then, the covariance of the center of each layer is calculated. ,in, The covariance of each circle center is estimated by dividing the weighted sum of squared residuals by the effective degrees of freedom. for Chinese correspondence , A 2×2 submatrix.
4. The method for detecting the verticality of a wind turbine tower as described in claim 3, characterized in that, The observation weight for: ; in, Circumferential coverage factor; The incident angle factor; Point density factor; Register residual factors for the sites; Candidate markers for anomalies; The observation weight Normalize within each layer so that the sum of the weights for that layer is 1.
5. The method for detecting the verticality of a wind turbine tower as described in claim 3 or 4, characterized in that, The overall tilt parameters include the overall tilt angle, top horizontal offset, maximum horizontal offset, and layer weight. The local deflection parameters include local deflection at each height, maximum local deflection and its height, orientation angle, and decoupling parameter covariance. The overall tilt and local deflection are then identifiablely decoupled from the central sequence to obtain the overall tilt parameters and local deflection parameters, including: Arrange all effective cross-section centers according to height to form a center sequence. For the central sequence Perform weighted three-dimensional line fitting to obtain the overall fitted axis. and the overall axial direction from bottom to top , If the bottom area is obstructed by the foundation or flange, use the minimum effective height and the maximum effective height and extrapolate to the nominal detection range according to the model; A local flexure model is constructed using low-order polynomials, piecewise splines, or smooth curves that satisfy the continuity of the tower structure. The horizontal centerline is represented as ,in, For height The horizontal center coordinates of the location For constant terms, The coefficient of the overall sloping first-order term. For local smoothing margin; using cubic B-spline basis matrix express , The coefficient vector is a B-spline vector; a weighted zero-mean constraint is applied. and weighted zero linear trend constraint ; Determine layer weights ,in, For coverage and residual gating factors, For the first Layer center covariance traces, Pick To avoid division by zero, normalization is then performed. ; The B-spline uses a cubic basis function, with repeated nodes at the endpoints; the initial spacing of the internal nodes is taken as... Each node interval must contain at least three valid sections; otherwise, adjacent intervals are merged, and the smoothing parameter is adjusted. exist to The minimum value is selected from the logarithmic candidate set using generalized cross-validation. Solve the constrained least squares problem: and apply simultaneously and ;in, This is the layer weight matrix. A vector consisting of the horizontal center coordinates of all layers. To design the matrix and the first Action is , For smoothing parameters, Given a second-order difference matrix; solve using the KKT equations in one step. If the condition number of the KKT matrix is greater than... If the intervals of adjacent nodes with the fewest valid points are merged sequentially and the calculation is repeated, and if the condition is still not met, the process is transferred to manual review. The same layer weights are used in both the X and Y directions. B-spline basis matrix and smoothing parameters Solve separately to keep the directional components comparable; Coefficient of the overall inclined linear term Calculate the overall tilt angle ,Depend on Calculate the local deflection and obtain the decoupling parameter covariance from the corresponding sub-block of the KKT system inverse matrix. Calculate height Horizontal offset at location and local deflection Result in top horizontal offset Maximum horizontal offset Maximum local deflection and direction angle.
6. The method for detecting the verticality of a wind turbine tower as described in claim 5, characterized in that, Step S5 specifically includes: Before performing uncertainty calculations, check the effective tower height, effective cross-section ratio, convergence status of joint optimization across height, and covariance condition number. If any of these conditions are not met, output the retest / manual verification directly. For the parameter vector Calculated indicators Calculate the gradient at the convergent solution. , and according to Calculate the uncertainty components of the model, where The parameter covariance matrix is obtained from the joint optimization across heights of S3; The 6-DOF pose covariance of control point or station registration Jacobi through coordinate transformation Propagation as registration uncertainty components ; Gravity direction covariance Through the tilt angle to the unit vector of gravity Jacobian propagation as a component of gravitational uncertainty ; Let the quantization resolution independent components be denoted as , and according to Calculate the combined standard uncertainty; Pick This results in an expanded uncertainty with approximately 95% coverage. When the effective degrees of freedom are less than the preset threshold, calculate veff according to the Welch-Satterthwaite formula and take t0.975, veff instead of 2; To verify the linear propagation results, a two-tiered stratified bootstrap resampling method was used, with 200 iterations at both the site and cross-section levels. First, sites were sampled with replacement, then effective cross-sections were sampled with replacement within each site. This process was repeated sequentially across heights for joint optimization and identifiable decoupling. A fixed random seed was used, and the seed, effective resampling times, and index sequences were saved. If the effective number of iterations was less than 180, or the ratio of the bootstrap standard deviation to the linear propagation standard deviation was not between 0.8 and 1.25, the larger value was taken, and the conclusion was downgraded to a verification result. For the test indicators that have passed the pre-test quality inspection, construct the measurement interval. The measurement interval is compared with a preset threshold. Comparison: When the upper bound of the measurement interval is less than The output is normal; when the lower bound of the measurement interval is greater than An alarm will be output in real time when the measurement range is touched or crossed. Output for retesting / manual verification; for directional bilateral limits, apply the same rule to both boundaries; among which... Limits given for enterprise procedures, design documents, or acceptance criteria.
7. The method for detecting the verticality of a wind turbine tower as described in claim 1, characterized in that, Step S1 further includes performing a quality gating check on the gravity-aligned point cloud, including: Perform control point residual checks, assuming the first... The known coordinates of the control points are: After registration, the coordinates are calculated as follows: Then the first The residuals at each control point are: The residual size is , The root mean square of the residuals at each control point is ,when If the residual error exceeds the preset control point threshold, it is considered unqualified. Perform registration residual checks in the inter-station overlap area, and check the first... For each station, take corresponding point pairs within the overlapping area, and let the th station be... In a pair of corresponding points, point The unit normal vector to the local plane of the point cloud at adjacent stations is A point on the plane is The point-to-surface residual is: , No. There are a total of [number] stations in the overlapping area. When there are 1 corresponding point pair, the registration residual in the overlapping area between stations is: ,when If the error exceeds the preset registration residual threshold, it is considered unqualified. Perform point cloud coverage checks, including height coverage checks and circumferential coverage checks. Height coverage is... ,in, and These represent the lowest and highest altitudes actually covered by the point cloud, respectively. The nominal height of the tower; circumferential coverage is calculated based on preset azimuth sectors. For a given height section, the number of sectors covered by the measured point cloud is assumed to be... The measured circumferential coverage angle is: The circumferential coverage rate is: If the height coverage is lower than the preset height coverage threshold, or the circumferential coverage is lower than the preset circumferential coverage threshold, or the single-layer circumferential coverage angle is lower than the preset coverage angle threshold, it is deemed unqualified. If any indicator in the quality gate inspection does not meet the preset quality threshold, the process returns to step S1 for supplementary sampling or proceeds to manual review, and does not proceed to automatic qualification determination.
8. A wind turbine tower verticality detection system, characterized in that, include: The point cloud acquisition and registration module is used to acquire the three-dimensional laser point cloud of the wind turbine tower under test and the calibration data for gravity direction determination and inter-station registration. The three-dimensional laser point cloud is preprocessed, multi-station registered and gravity aligned to obtain the gravity aligned point cloud. The tower segmentation and anomaly marking module is used to segment the tower surface point cloud from the gravity-aligned point cloud and mark the protrusions on the wind turbine tower under test as anomaly candidate points. The cross-height joint optimization module is used to divide the effective height of the tower into layers with a preset layer thickness along the Z-axis in a coordinate system where the Z-axis is consistent with the vertical direction of gravity on site. It performs cross-sectional fitting on the point cloud of the tower surface for each layer, and performs cross-height joint optimization with the fitting center and fitting radius of each layer as parameters to be estimated. The objective function of the cross-height joint optimization includes at least a single-point radial residual term, a continuity constraint term for the center of adjacent cross sections, and a continuity constraint term for the radius of adjacent cross sections. Solving the objective function yields the fitting center, fitting radius, center covariance, and observation weight of each effective cross section. The identifiable decoupling module is used to sort the fitted circles of each effective section by height to form a center sequence, and to perform identifiable decoupling of the center sequence for overall tilt and local deflection to obtain the overall tilt parameter, local deflection parameter and decoupling parameter covariance. The identifiable decoupling includes: representing the horizontal coordinate of the center sequence as the sum of the overall linear term and the local smoothing margin, and applying a weighted zero mean constraint and a weighted zero linear trend constraint to the local smoothing margin to uniquely separate the overall tilt component and the local deflection component. The uncertainty three-state gating module is used to determine the uncertainty of the detection index based on the obtained center covariance and observation weight, as well as the obtained overall tilt parameter, local deflection parameter and decoupling parameter covariance. The detection index includes at least one of the following: overall tilt angle, top horizontal offset, maximum horizontal offset, maximum local deflection and its direction. The measurement interval formed by the detection index and the uncertainty is compared with a preset threshold, and the detection result is output as one of three states: normal, alarm or retest / manual verification.
9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the detection method as described in any one of claims 1-7.
10. A computer storage medium, characterized in that, The storage medium stores at least one executable instruction that causes the processor to perform the detection method as described in any one of claims 1-7.
