Wind turbine monitoring

Abstract

Method for monitoring the operation and / or the geometry of a wind turbine in motion comprising a tower and several blades extending from a rotor hub, the method comprising: (a) acquisition with a three-dimensional acquisition apparatus of a plurality of scans representing at least partially the wind turbine; (b) segmenting the scans to identify at least portions corresponding to the blades; (c) generating a 3D representation by at least aligning and denoising the scans with respect to a common frame of reference associated with the wind turbine and compensating for relative motion between the three-dimensional acquisition apparatus and the wind turbine; (d) computing, based on the 3D representation, at least one geometric or operational parameter and / or computing at least one geometric transformation between at least one representative part of one blade and the corresponding representative part of at least one of the other blades and computing at least one wind turbine geometric or operational parameter based on the at least one computed geometric transformation.

Description

[0001] Description

[0002] Title : Wind turbine monitoring

[0003] Field of the invention

[0004] The present invention relates to the field of wind turbine inspection and more particularly to a method for monitoring a wind turbine, and to a system for implementing such a method.

[0005] Background of the invention

[0006] With the emergence of image processing technologies and the democratization of drones, inspection and maintenance methods have been automated and simplified in various sectors, such as maintenance of dams, railroad lines, electricity pylons, bridges, and all types of static structures.

[0007] However, if the use of drones for inspecting such static structures has proven to be particularly efficient, their use is far more complex in the case of moving structures such as wind turbines. This is specially challenging when 3D data are acquired, since the resulting unstructured data are difficult to process. Therefore, drones are usually used for inspecting wind turbines when they are at a standstill. This necessarily entails a significant loss of production. New solutions start to be developed with turbines in operation but mainly for visual blade inspections. Moreover, inspecting a wind turbine when it is at standstill essentially reveals its structural damages, and it does not allow to detect operating dysfunctions.

[0008] It is also known that using different kinds of imagery enables the analysis of external and internal structures of a wind turbine. Even though such complementary information may increase the quality of the analysis, it does not allow to highlight all operating dysfunctions, such as rotation regularity, resonance effect, or blade bending.

[0009] The publication TOMA SIKORA et al. : « Towards Operating Wind Turbine Inspections using a LiDAR-equipped UAV” ARXIV.ORG, CORNELL UNIVERSITY LIBRARY, 26 June 2023, XP09154781 1 discloses to make numerous LiDaR scans that are compared to a reference 3D model to identify the location of the scanned model. The matching between the scanned model point cloud and the reference 3D model point cloud determines the transformation aligning the two models, which enables the reconstruction of the wind turbine as the set of acquired points, all in the same position and in the same orientation. The blades are still in their acquired position in rotation with respect to the tower and the rotor hub. While it provides the required information allowing the drone to navigate around the whole turning rotor identified as a half disk, it does not allow the reconstruction of the blades themselves as soon as the wind turbine is in motion during the acquisition as presently claimed. In addition, the half disk reconstruction is not realigned with the first scanned frame as all the scans are just accumulated with the blades in their acquired position, without blade alignment, which creates the half disk.

[0010] The publication YANG CONG et al. “Towards accurate image stitching for drone-based wind turbine inspection” RENEWABLE ENERGY, vol 203, 1 February 2023, pages 267-279, XP093030001 , GB, ISSN:0960-1481 , DOI:10.1016 / j.renene.2022.12.063 discloses a method in which portions of the blades are identified based on acquisitions performed with 2D images. All techniques and solutions that are presented in this publication are developed for 2D images do not hold for 3D point clouds.

[0011] There also exist methods based on mono-laser beams enabling the inspection of moving wind turbines. In that case, the mono-laser needs to be static and it can only acquire one dimensional measure along a specific direction. Thus, some parameters such as lateral oscillations, blade tip edgewise bending and absolute blade pitch angle cannot be measured over time. In addition, the acquisition by mono-laser beams requires a static and stable positioning and a clear visibility of the wind turbine. This makes these acquisition methods unusable in certain wind farms, such as offshore wind farms, or those that are in heavily vegetated or hilled fields areas.

[0012] There is thus a need for new methods improving the monitoring of wind turbines in motion. In particular, there is a need for methods that improve the quality of the inspection and the variety of wind turbines that can be monitored, regardless of where they are installed.

[0013] Summary of the invention

[0014] The present invention aims to remedy to all or part of the deficiencies of the prior art mentioned above and exemplary embodiments of the invention relate to a method for monitoring the operation and / or the geometry of a wind turbine comprising a tower and several blades extending from a rotor hub, the method comprising:

[0015] (a) acquisition with a three-dimensional acquisition apparatus of a plurality of scans representing at least partially the wind turbine,

[0016] (b) segmenting the scans to identify at least portions corresponding to the blades;

[0017] (c) generating a 3D representation by at least aligning and denoising the scans with respect to a common frame of reference and compensating for possible relative motion between the three-dimensional acquisition apparatus and the wind turbine,

[0018] (d) computing, based on the 3D representation, at least one geometric or operational parameter and / or computing at least one geometric transformation between at least one representative part of one blade and the corresponding representative part of at least one of the other blades and computing at least one wind turbine geometric or operational parameter based on the at least one computed geometric transformation. The invention does not require the wind turbine to be stopped. The wind turbine may still be in motion during the acquisition. The method according to the invention is more efficient than the state-of-the-art methods as it may take a relatively small amount of time to inspect the wind turbine and to compute its parameters. Moreover, the method provides if desired more geometric parameters than the state-of-the-art methods, thus allowing a global and accurate analysis of the wind turbine.

[0019] Furthermore, not only the wind turbine but also the three-dimensional acquisition apparatus may be in motion. In particular, the height of the acquisition apparatus and the distance between the wind turbine and the acquisition apparatus may vary during the acquisition of the scans.

[0020] By “possible relative motion between the three-dimensional acquisition apparatus and the wind turbine” it is understood the movements induced by the motion of the wind turbine only, the motion of the three-dimensional acquisition apparatus only, or the motion of both the wind turbine and the three-dimensional acquisition apparatus during the acquisition of the scans, and more generally during the implementation of the method.

[0021] The method may use a drone for the acquisition of the scans. The scan processing and other computations may be performed on one or more computers, and more generally on any appropriate processing system.

[0022] In preferred embodiments of the invention, the method comprises at step d) computing at least one geometric transformation between at least one representative part of one blade and the corresponding representative part of at least one of the other blades and computing at least one wind turbine geometric or operational parameter based on the at least one computed geometric transformation.

[0023] The 3D representation may be a cloud of points or a meshed model generated based on such cloud of points.

[0024] Each rotor blade that is identified by segmentation may be realigned in the frame of a first scan in order to keep the first blade position and reference. Each blade may thus be accumulated in the same initial position, which generates a 3D reconstruction as accumulated point clouds of the individual blades as detailed below. The invention enables to align the scans in position and orientation without requiring the use of a reference model, contrary to the publication TOMA SIKORA et al. discussed above. The invention also allows to identify and align the blades in a similar position in rotation around the rotor hub even though they are acquired in different positions during their motion. The compensation for relative motion between the three-dimensional acquisition apparatus and the wind turbine at step c) may be performed through temporal registration of the scans. Step a) Acquisition

[0025] The acquisition is performed such that a plurality of scans, preferably all scans, represent at least a part of the wind turbine, said part being at least partially common for said scans.

[0026] Each scan preferably represents front views of the wind turbine.

[0027] A “scan” may refer to a point cloud, the points being acquired simultaneously or over a lapse of time, the point cloud representing at least one part of the wind turbine. Typically, a scan may be acquired by a scanner, for example a 3D Lidar, the scan being constituted by a plurality of points. At least some points may be acquired at a different time due to the duration of the scan and thus representing a portion of the wind turbine in a different position when it is in motion. Some three-dimensional acquisition apparatus may acquire all points at the same time, as a flash Lidar.

[0028] During the acquisition, the three-dimensional acquisition apparatus is preferably positioned at a height between the wind turbine highest point and the sea or ground level.

[0029] The three-dimensional acquisition apparatus may roughly be positioned at hub height.

[0030] Alternatively, it may be positioned at sea or ground level, or any position between sea or ground level and hub height, and even between the hub height and the wind turbine highest point, for instance, for acquiring the upper parts of the wind turbine.

[0031] The position of the three-dimensional acquisition apparatus may vary during the acquisition. The three-dimensional acquisition apparatus may be a Lidar scanner, in particular a flash Lidar or a conventional rolling shutter acquisition apparatus.

[0032] The wind turbine and / or the three-dimensional acquisition apparatus may be in motion during the acquisition of the scans, with a possibility of both the wind turbine and the three- dimensional acquisition apparatus being in motion. In particular, the three-dimensional acquisition apparatus may be carried by a UAV (unmanned aerial vehicle), for example a drone, or a boat.

[0033] In exemplary embodiments, the wind turbine is an offshore wind turbine.

[0034] The knowledge of the exact position of the acquisition apparatus relatively to the wind turbine during the acquisition is not needed for the acquisition. If provided, it can however improve the accuracy and / or speed of the scan processing.

[0035] The scans are preferably time stamped, which enables temporal registration of the scans, as detailed thereafter.

[0036] Step b) Segmentation

[0037] During the segmentation of a scan, portions of the scan are categorized into different categories. The blades of the wind turbine may be separated from the deformed scans (i.e. the scans that include the blade deformations due for example to the rolling shutter effect) at step b) as a prerequisite to step c) during which the motion deformation is handled, for example by temporal registration as detailed below.

[0038] The data may be segmented according to geometric properties

[0039] Typically, labels are attributed to each point or portion defining the category to which it belongs. The categories may comprise the following: “tower”, “blades” and “rotor hub”. These categories may be subdivided. For example, the category “tower” may be subdivided into “bottom of tower”, “top of tower”, “middle of tower”; the category “blades” may be subdivided into “tip of blades”; “portions of blades near to rotor hub”, “rigid portions of blades”, “blade 1 ”, “blade 2” and “blade 3”, inter alia.

[0040] Preferably, the segmentation is performed to identify portions of the scan corresponding respectively to the blades and the tower, or to the blades and the rotor hub, or to the blades, the rotor hub and the tower.

[0041] The segmentation may be performed by implementing one or several algorithms. For example, one algorithm may be used to label the tower, another may be used to label the rotor hub and another to label the blades.

[0042] By “algorithm”, it is meant a computer program comprising instructions of code that, when executed by a computer (such as a personal computer for example), allows to implement processing steps or more generally calculation steps.

[0043] The segmentation may comprise the identification of a portion of the scan corresponding to the tower and of a portion of the scan corresponding to the tips of the blades, and then the determination of a separation plan separating the portion of the scan corresponding to the tower from a portion of the scan corresponding to the tips of the blades.

[0044] In particular, before performing the blade / tower segmentation, a pre-alignment of all the points of the scan in a reference frame may be performed, this pre-alignment being preferably a rough pre-alignment. The aim of the pre-alignment is to roughly align the points of each scan in order to obtain points observed from the same point of view, facing the wind turbine at a similar height and angle. For example, the reference frame axis is determined by computing the mean normal of all points of the scans.

[0045] Preferably, before performing the blade / tower segmentation, and preferably after performing a pre-alignment, a removal of outlier points of the scan is performed. Such removal step allows avoiding imprecision due for example to acquisition errors.

[0046] The removal of outlier points may be implemented by a Euclidean Maximum Spanning Tree (EMST) algorithm. Of course, all other standard outlier removal methods may be used.

[0047] The blade / tower segmentation may be performed by identifying the tower points first.

[0048] For each scan, the tower points may be segmented by roughly identifying points of the scan corresponding to blade tips and those corresponding to the bottom of the tower, then by computing a separation plan separating the points corresponding to the blade tips and those corresponding to the bottom of tower, and finally by segmenting the other points according to their position to said plan to differentiate the points corresponding to the tower from those corresponding to the blades or to the rotor hub.

[0049] Step c) 3D representation

[0050] A 3D representation is generated by aligning and denoising the scans with respect to a common frame of reference associated with the wind turbine and compensating for possible relative motion between the three-dimensional acquisition apparatus and the wind turbine.

[0051] In other words, the 3D representation may be the cloud of points obtained after aligning and denoising the scans and compensating for possible motion distortions.

[0052] The 3D representation may also be obtained by further processing the aligned scans to generate a meshed digital model of at least the blades of the wind turbine.

[0053] The alignment of the scans is preferably performed relative to a frame of reference associated with the wind turbine that has an axis that is parallel or coincident with the rotation axis of the rotor of the wind turbine.

[0054] The frame of reference may be determined from the plurality of the scans.

[0055] The frame of reference determines a point of view from which the wind turbine is observed, the scans being aligned to allow their observation according to said point of view.

[0056] The alignment of the scans may be performed by pairing scans.

[0057] The alignment of the scans may be done by considering at least points labeled as rotor hub and blades, possibly only points labeled as rotor hub and blades.

[0058] Preferably, when points of a scan are not acquired at the same time but over a time frame, for example when a rolling shutter three-dimensional acquisition apparatus is used to perform the acquisition, step c) comprises a temporal registration of points of the scan with respect to a reference instant.

[0059] The temporal registration may allow to align the blades in a common position.

[0060] The temporal registration may comprise, for each pair of successive scans, pairing points of said scans and computing the transformation from one point of the pair to the other point of the pair for the given instant, by taking into account the individual instant of acquisition of each point of the pair.

[0061] The temporal registration method is particularly efficient to correct deformations and artifacts produced by rolling shutter effect. The rolling shutter effect generally appears when the object to be acquired and / or the acquisition apparatus are moving during the acquisition. When the relative displacement between the acquisition apparatus and the object becomes larger than the acquisition noise, distortions appear in the acquired scans. The movement of the acquisition apparatus may be estimated by sensors acquiring the position of the acquisition apparatus itself, the movement of the acquisition apparatus being used to correct the acquired points.

[0062] Alternatively, or additionally, estimation of the movement of the object may be made by additional devices.

[0063] The temporal registration method disclosed below enables the registration of points acquiring any moving objects by considering the time at which each of them is acquired, without the need for additional devices.

[0064] The temporal registration may comprise both a pairing of corresponding acquired points of moving rotor blades scans in which some points are acquired at different times, and an alignment taking into account these different acquisition times of points in scans.

[0065] The temporal registration method for correcting motion distortions of an object in a plurality of scans may comprise, for each scan Sp, Sq, Sp and Sq being successive scans (immediately or not), respectively acquired over time intervals [to,P; tf;p] and [to,q; tf;q], to,q> tf;p, the following steps:

[0066] - pairing at least part of points of Sp with at least part of points of Sq to obtain pairs of points {qk! Pk}, p and q are corresponding points from scan Sp and scan Sq respectively; if one considers a subset of points p in Sp that have corresponding points q in Sq, k would be the index of the corresponding pair in this subset;

[0067] - projecting all paired points to the given instant tg,

[0068] - determining rigid transformation T(R,t) from Sp to Sq, the rigid transformation being dependent of the difference of time acquisition between the paired points, T(R,t) being the 4x4 matrix representing the rigid transformation T between scan Sp to scan Sq consisting of a 3x3 rotation matrix R and translation vector t, these steps being performed for all pairs of scans such as to obtain plurality of scans registered to the given instant tg.

[0069] For example, the rigid transformation may be calculated by applying the following point-to-point equation: The projection of a point x to the given instant tgbeing defined by : proj(x) = R x + tp, with p =ttgi-ttx0 m being the number of pairs of points, / being the norm of the distance, typically equal to 2, the scans Sq and Sp being acquired over the time interval [to, ti], txbeing the time stamp of the point x from either scans Sp or Sq.

[0070] Since R is the full rotation between a point p at to and it's corresponding point q at ti, and p is the projection parameter that computes the ratio between a given point x at txto the target time tgand the full interval to to ti, Rpis therefore the interpolated portion of the rotation that transforms a point x from either scans Sp or Sq to it's position at the instant tg.

[0071] Determining the rigid transformation may alternatively be obtained by applying a point-to-plan equation.

[0072] In addition, the determination of the rigid transformation may comprise using a quadratic interpolation function f to approximate a non-linear motion of the object.

[0073] For example, the determination of the rigid transformation may be obtained by applying the following point-to-plan equation: tPk is the timestamp of the point pk, same as tqkis the timestamp of point pk

[0074] Such temporal registration method is particularly well adapted to the registration of scans of a wind turbine in motion.

[0075] The temporal registration method is preferably applied before determining center and axis of rotation, when such a step is implemented.

[0076] Classical registration methods, such as standard ICP (Iterative Closest Points), may be applied after the registration of the scans along the centers and axis of rotation, and / or after the temporal registration method, and / or after smooth filtering, to avoid accumulation of errors. Other temporal registration methods may be implemented.

[0077] The alignment of the scans comprises determining the axis of rotation of the blades. This determination may be made by a first rough estimate of the center and the axis of rotation in each scan, the center and axis of rotation are then refined by using the symmetry of the distribution of the blades around the rotor hub, for example, by minimizing the deviation between points of the scan and points of the scan rotated by the theoretical angle between two blades, i.e. 3607(number of blades), generally 120°. Centers and axis of rotation determined for each scan are then compared to each other to accurately align the scans.

[0078] Smooth filtering may be applied to the unified aligned scans, allowing decreasing noise in the scans.

[0079] Step d) Computation of parameters

[0080] The method comprises the computing, based on the 3D representation, of at least one geometric or operational parameter, for example blade tip bending.

[0081] Preferably the method comprises the computing of at least one geometric transformation between at least one representative part of one blade and the corresponding representative part of at least one of the other blades and computing at least one wind turbine geometric or operational parameter based on the at least one computed geometric transformation.

[0082] The geometric transformation may be a rigid geometric transformation. A “rigid transformation” is a geometric transformation of a Euclidean space that preserves the Euclidean distance between every pair of points, wherein deformation is not considered.

[0083] Preferably, the at least one representative part comprises an essentially dimensionally stable part of the blade, in particular the part may be away from the tip of the blade and from the rotor hub. By “dimensionally stable part” it is meant a part which is not deformed or the deformation is negligible when the wind turbine is in motion.

[0084] The at least one representative part of the one blade and the corresponding representative parts for the other blades may be substantially superimposable.

[0085] Computation step d) may comprise the following sub-steps: d1 ) for each aligned scan, segmenting the segmented portion corresponding to the blades so as to identify the at least one representative part of each blade of the wind turbine, d2) aligning in pairs the identified representative parts, and determining the geometric transformations allowing these alignments.

[0086] Sub-step d1 ) may be defined as a sub-segmentation of the points or portions of the scan labeled as “blades”, to differentiate each blade from one another.

[0087] Geometric or operational parameters may be chosen among a lack of symmetry in the distribution of the blades around the axis of rotation of the blades, a deformation of a blade; a pitch at different blade radius, an absolute pitch angle, a blade twist, a rotor axis of rotation, a blade bending during motion, and / or tower oscillations and tower resonance. This list is not exhaustive.

[0088] Knowing the geometric transformations between blades of the wind turbine notably allows to compute the repartition of the plurality of blades around the rotor hub. Typically, a wind turbine comprises three blades evenly distributed around the rotor hub, that is to say an angle of substantially 120° is formed between two successive blades.

[0089] In particular, the geometric transformations allow the measurements of the orientation of each blade along its longitudinal axis, the deviation of the position of the blades from their ideal distribution around the rotation axis of the wind turbine, the detection of dynamic tangential deformations at the tips of each blade.

[0090] The method may comprise a further step e) of evaluation of the operation of the wind turbine including the generation of information concerning the need for adjustments, repairs, the presence of hazard, the loss of profitability due to damage or wear and tear.

[0091] Exemplary embodiments of the invention also relate to a system to implement the method for monitoring the operation of a wind turbine as defined above, the system comprising: a three-dimensional acquisition apparatus, preferably a Lidar scanner; and a computer program comprising code instructions which, when the program is executed by a computer, cause the computer to carry out steps b) to d), and preferably b) to e), of the method according to the invention for monitoring the geometry and / or operation of a wind turbine.

[0092] Preferably, the system comprising a UAV, in particular a drone, carrying the three-dimensional acquisition apparatus.

[0093] Description of the figures

[0094] For a more complete understanding of the present invention, a description will now be given of several examples, taken in conjunction with the accompanying drawings, in which: Fig. 1 illustrates a wind turbine,

[0095] Fig. 2 illustrates a system of the invention,

[0096] Fig. 3 illustrates segmentation step,

[0097] Fig. 4 schematizes pre-alignment step,

[0098] Fig. 5 schematizes outlier removal step,

[0099] Fig. 6 schematizes blades from tower separation step,

[0100] Fig. 7 illustrates temporal registration step,

[0101] Fig. 8 schematizes a scan before the correction of motion distortions (left) and after the correction of motion distortions (right),

[0102] Fig. 9 schematizes a smoothing step,

[0103] Fig. 10 illustrates blade transformation step,

[0104] Fig. 1 1 schematizes rigid parts of blades and their identification,

[0105] Fig. 12 schematizes alignment of two rigid parts of blades along an axis, Fig. 13 schematizes computation of parameters, and

[0106] Fig. 14 schematizes computation of another parameter.

[0107] Detailed description

[0108] An illustration of a wind turbine 1 is represented Fig. 1.

[0109] The wind turbine comprises several blades 10, a nacelle comprising a rotor hub 14 and a tower 12. The blades rotate around an axis of rotation X of the rotor hub.

[0110] An example of a system 2 to implement the method for monitoring the operation of a wind turbine according to the invention is shown in Fig. 2.

[0111] The system 2 comprises a three-dimensional acquisition apparatus 3. The three-dimensional acquisition apparatus may be carried by a drone.

[0112] The three-dimensional acquisition apparatus acquires a plurality of scans 4. The scans may be acquired over a lapse of time from to to t1 .

[0113] A scan may be formed by a plurality of points 40, forming a cloud of points. The points may be acquired at a same time or over a lapse of time.

[0114] The scans are transmitted to a computer program 5 comprising instructions which when the program is run by a computer allow to implement steps b) to d), preferably b) to e) of the monitoring method of the invention.

[0115] Segmentation

[0116] Each scan is first segmented so as to identify points which correspond to the tower, points which correspond to the blades and eventually points which correspond to the rotor hub. The points corresponding to the rotor hub may be included in the category “blades”.

[0117] A label is attributed to each segmented point.

[0118] In embodiments of the invention, the segmentation may comprise all or part of the steps illustrated in Fig. 3 which consist on pre-aligning the scan relatively to a reference frame RO, removing outlier points and defining a separating plan separating the blades from the tower.

[0119] Pre-alignment

[0120] The aim of the pre-alignment is to roughly align the points of each scan in order to obtain points observed from the same point of view, facing the wind turbine at a similar height and angle. Preferably, the reference frame is determined such that the origin of the frame is approximately at the same height as the rotor hub and the z axis of the frame is approximately aligned or parallel to the axis of rotation X, as shown in Fig. 4.

[0121] The reference frame axis may be calculated by computing the mean normal of all points of the scans. Outlier points removal

[0122] The segmentation of a scan may comprise, preferably after pre-aligning the scan, removing outlier points of said scan.

[0123] An example of outlier points detection by using the “Euclidean Maximum Spanning Trees” (EMST) algorithm is represented in Fig. 5.

[0124] In Fig. 5, the sub-Fig. 5a) represents a cloud of points 40 of a scan 4, said scan comprising outlier points 42; 42’.

[0125] Sub-Fig 5b) represents the maximum spanning tree.

[0126] When the length of an edge binding two points 40 of the tree exceed a threshold value for example 5m depending on the scan quality, the edge is removed.

[0127] In the example of Fig. 5, edges 60 determined by the computation of the maximum spanning tree whose length in excess of a said threshold value are removed, as shown in sub-Fig. 5c). After that, only the largest connected cluster may be conserved, as shown in sub-Fig. 5d), other clusters 42; 42’ being considered as outlier points.

[0128] Blades / Tower separation

[0129] The segmentation of a scan may comprise, preferably after pre-aligning the scan, and preferably after removing outlier points, determining a separation plan separating points corresponding to blades from points corresponding to the tower.

[0130] This may be performed by roughly segmenting the blade tips 46 and bottom of the tower 4612, as shown in Fig. 6, see in particular sub-Fig 6c).

[0131] Sub-Fig. 6a) represents a scan 4 after pre-alignment and / or removal of outlier points.

[0132] To determine tips of the blades and bottom of the tower, the method may comprise: computing the barycenter 44 of the points of the scan as shown in sub-Fig 6b), filtering the points located at a predetermined distance from the barycenter, clustering the remaining points, and analyzing depth of the cluster, preferably the depth being defined in the reference frame, so as to identify the cluster corresponding to the bottom of the tower, the other being attributed to clusters of the blades.

[0133] The clustering may be done by applying an EMST, and the edges may be removed when greater than 3m for example.

[0134] The analysis of the depth of the points when observing the wind turbine from front view allows to identify the points corresponding to the tower. Indeed, points corresponding to the tower are farther than the points corresponding to the blades when the scan is a front view acquisition of the wind turbine.

[0135] The predetermined distance may be calculated by computing an average half distance d, in the example of sub-Fig. 6c), the predetermined distance being 1 .2d, with d = ~CH2 mbeing the number of points p and c the barycenter. From the segmentation of the tips of the blades and the bottom of the tower, a separation plan separating the points corresponding to the blades and the points corresponding to the tower may be determined.

[0136] The determination of this separation plan may be made by determining a common plan best fitting the clusters attributed to the blades. The separation plan could be defined by a plan parallel to the common plan and positioned between the blades and the tower.

[0137] The points may be segmented by attributing them a label according to their position, in front or behind the separation plan relatively to a front viewer of the wind turbine, .

[0138] Two categories 4 and 4I2may be identified as illustrated by sub-Fig 6d).

[0139] Alignment of the scans

[0140] In embodiments of the invention, the alignment step comprises all or parts of the sub-steps shown in Fig. 7.

[0141] Correction of motion distortion

[0142] When the acquired scans comprise points acquired at different instants, i.e. each scan being acquired over a lapse of time, a temporal registration is made by taking into account the instant of acquisition of points of the scan.

[0143] In particular, this may be performed by applying the temporal registration method described above allowing to correct motion distortions in scans.

[0144] In Fig. 8, it is illustrated a portion of a scan 4, with blades that appear to be double blades 48a; 48b due to the displacement of the blade which is non negligible. As such, points of one scan are captured for different blade positions.

[0145] By applying a correction of motion distortion, blades 48 can be reconstructed to represent a unique position of the blade.

[0146] Preferably, the correction of motion distortion is applied only to the points of the scans corresponding to the blades and the rotor hub.

[0147] Rotation axis estimation

[0148] The alignment step may comprise, preferably after correcting motion distortions in the scans, estimating the rotation axis X of the rotor hub, the common reference representation being defined by such estimation of said rotation axis X. The axis is determined by the center C and the vector axis N of the rotation axis X : X={C, N}.

[0149] This estimation may be performed by roughly determining the center of the rotor hub for the unified aligned scans, aligning a first cloud of points comprising points of each scan segmented as blades with a second cloud of points corresponding to the first cloud of points rotated by 3607number of blades of the wind turbine around the center. Aligning the first and second cloud of points may comprise minimizing the distance between the first and second clouds, for example by applying: C, N 2;=i||Rc,Nmi>M> M={m0,...,mk } are the points of the first cloud on which the rotation by 0 is applied, and H={ho,...,hk} are the points of the first cloud being the closest points to the rotated points.

[0150] The center determined for a scan may be back propagated through all the scans of the plurality of scans, allowing a more accurate estimation. The back propagation of the centers may comprise applying for each index i the following transformation:

[0151] Where Ci is the center, Ni the vector of the rotation axis Xi for the scan i

[0152] T is a matrix that defines the rotation and the translation to align the scan i with scan i+1 .

[0153] Noise removal

[0154] The alignment step may comprise, preferably after implementing the correction of distortions of the plurality of scans, and / or preferably after estimating the rotation axis X of the rotor hub, and / or preferably before applying an ICP algorithm on the scans, removing noise.

[0155] The noise removal may be applied to the unified aligned scans.

[0156] Fig. 9 illustrates the use of a low pass filter to compute a smoother surface.

[0157] For example, a geometric low pass filter such as described in the article “Algebraic point set surfaces” by Gael Guennebaud and al, ACM Transactions on Graphics, Volume 26, Issue 3pp 23-es, https: / / doi.org / 10-1 145 / 1276377.1276406, may be used.

[0158] ICP (Iterative Closest Point)

[0159] The alignment step may comprise applying a ICP algorithm on the scans, preferably after implementing the correction of distortions of the plurality of scans, and / or preferably after estimating the rotation axis X of the rotor hub, and / or after removing noise.

[0160] Preferably, the ICP algorithm is applied for the blades and for the tower segmented parts of the scans., the result of this ICP on the aligned scans superposition is a 3D representation of the wind turbine.

[0161] Geometric transformations

[0162] In embodiments of the invention, the computation of the geometric transformation is performed by registering the blades in pairs but not exclusively.

[0163] The registration in pairs may comprise the sub-steps detailed Fig. 10.

[0164] Extraction of the ria id parts of the blades

[0165] For each scan, sub-parts 46 i, 46 2, 46 3 of each blade may be isolated as illustrated in Fig

[0166] 1 1. In particular, sub-parts may be isolated by removing the rotor hub segmented points of the scan and the points corresponding to the blade tips, for example by using a mask whose lower and upper limits are predefined depending on the wind turbine.

[0167] The sub-parts may be dimensionally stable parts of the blades.

[0168] Blade identification

[0169] Each blade may be identified from blade sub-parts, for example by using EMST algorithm. Blade angle differences estimations

[0170] The identified blades may be aligned in pairs along an axis Y by applying a rotation of 360° / number of blades, typically 120°, around the rotational axis, as shown in Fig. 12, where sub-part 46 i is aligned with sub-part 46 2, along the axis Y.

[0171] An ICP algorithm may be applied to determine, pair by pair, the transformation differences between the blades, notably to determine radial and axial rotations which may be calculated by using Euler decomposition.

[0172] Computation of parameters

[0173] The computation of parameters may comprise measuring parameters from the 3D representation of the wind turbine and evaluating the operation and the geometry of the wind turbine based on the measures of the parameters.

[0174] Preferably, the parameters comprise at least: the pitch, the roll and yaw and the detection of dynamic tangential deformations at the blade tips.

[0175] The pitch corresponds to the difference in orientation of the blades around their own longitudinal axis.

[0176] The roll and yaw correspond to the lack of symmetry in the position of the blades around the rotation axis, in other words it corresponds to the deviation from a theoretical angle, for example 120° for a wind turbine comprising three blades.

[0177] The parameters may be calculated from transformation matrices estimated between sub-parts of the 3D representation, in particular sub-parts of the blades of the 3D representation.

[0178] The parameters may comprise the twist of at least one blade, preferably all blades. The twist of a blade may be measured by computing pitch variations on several sub-parts of the blade, obtained at different distances from the rotor hub, then comparing the angle formed from each sub-part and a rotor hub plan.

[0179] An illustration of the computation of the twist of a blade is illustrated Fig. 13. Sub-fig 13a) is a front view of a blade and sub-fig. 13b) represents a view of parts 70, 72, and 74 of said blade, observed along the axis Y from its tip.

[0180] From the parts 70, 72 and 74, axis passing through the leading 78 and trailing 76 edges of the blade may be determined. Then, from these axis, twist angles a and may be computed. The parameters may comprise the absolute pitch angle for at least one blade, preferably all the blades, the absolute pitch angle for a blade may be computed based on the knowledge of the 0° blade radius position, and on a sub-part of the blade of the 3D representation of the wind turbine around this position, for example the sub-part is within 1 m radius around this position when 3D representation is shown in full size. Unless otherwise indicated, measurements are given in relation to the actual size of the wind turbine. Of course, the 3D representation may be a reduced representation.

[0181] The parameters may comprise edgewise deformations of at least one blade toward the tip, preferably all the blades, depending on the position of the blade relatively to the ground.

[0182] The parameters may comprise the flapwise bending of at least one blade toward the tip, preferably all the blades. This may be calculated by computing the distance d1 between the blade tip and the rotor hub plan P which is perpendicular to the rotation axis X, as illustrated Fig. 14. The distance may be compared to the distance measured for the other blades of the wind turbine. The distance may be compared to a theoretical distance based on measurements carried out on wind turbines of the same type as the one for which the method is applied. The bending is measured at a predefined rotor rotational speed.

[0183] An evaluation of the operation of the wind turbine based on the parameters may be performed. The evaluation may comprise the furniture of a report including information concerning the need for adjustments, repairs, the presence of hazard, the loss of profitability due to damage or wear and tear.

[0184] The invention may advantageously be implemented without needing the stop of the wind turbine.

[0185] This reduces the time spent by technicians at the wind farm and the loss of energy production during the inspection

[0186] The steps b) to e) may be performed in real time or not.

[0187] The acquisition may be performed and then the computation steps b) to e) may be performed later.

Claims

Claims1. Method for monitoring the operation and / or the geometry of a wind turbine (1 ) in motion, the wind turbine comprising a tower (12) and several blades (10) extending from a rotor hub, the method comprising:(a) acquisition with a three-dimensional acquisition apparatus of a plurality of scans representing at least partially the wind turbine;(b) segmenting the scans to identify at least portions corresponding to the blades;(c) generating a 3D representation by at least aligning and denoising the scans with respect to a common frame of reference associated with the wind turbine and compensating for relative motion between the three-dimensional acquisition apparatus and the wind turbine;(d) computing, based on the 3D representation, at least one geometric or operational parameter and / or computing at least one geometric transformation between at least one representative part of one blade and the corresponding representative part of at least one of the other blades and computing at least one wind turbine geometric or operational parameter based on the at least one computed geometric transformation.

2. Method according to claim 1 , wherein the three-dimensional acquisition apparatus is in motion during the acquisition of the scans.

3. Method according to anyone of claims 1 and 2, wherein at step b) the segmentation is performed to identify portions of the scan corresponding respectively to the blades, and the tower, or to the blades and the rotor hub, or to the blades, the rotor hub and the tower.

4. Method according to anyone of claims 1 to 3, wherein at step b) the segmentation comprises the identification of the portion of the scan corresponding to the tower and of the portion of a scan corresponding to the tips of the blades, and then the determination of a separation plan separating the portion of the scan corresponding to the tower from a portion of the scan corresponding to the tips of the blades.

5. Method according to anyone of claims 1 to 4, wherein step c) comprises temporal registration, which transforms points acquired at different times in the scans to corresponding positions, at a common instant.

6. Method according to claim 5, wherein the temporal registration comprises both a pairing of corresponding acquired points of moving rotor blades scans in which some points are acquired at different times, and an alignment taking into account these different acquisition times of points in scans.

7. Method according to anyone of claims 1 to 6, wherein the alignment at step c) comprises determining centers and axis for each scan, then comparing said centers and axis to align the scans in the common frame of reference.

8. Method according to anyone of claims 1 to 7, wherein step d) comprises the following sub-steps: d1 ) based on the 3D representation, segmenting the segmented portion corresponding to the blades so as to identify the at least one representative part of each blade of the wind turbine, d2) aligning in pairs the identified representative parts, and determining the geometric transformations allowing these alignments and computing blade angle differences based on said transformations.

9. Method according to anyone of claims 1 to 8, the geometric transformations being rigid geometric transformations.

10. Method according to anyone of claims 1 to 9, the representative parts of each blade being selected to be away from the tips of the blades and from the rotor hub.1 1 . Method according to anyone of claims 1 to 10, the geometric parameters being chosen among: a lack of symmetry in the distribution of the blades around the axis of rotation of the blades, a deformation of a blade, the blade angle differences, a pitch at different blade radius, an absolute pitch angle, a blade twist, and a rotor axis of rotation.

12. Method according to any of the preceding claims, the operational parameters being selected among rotor rotational speed, blade bending during motion, tower oscillations and / or a tower resonance.

13. Method according to anyone of claims 1 to 12, wherein the acquisition of the scans lasts between 3 seconds and 30 minutes.

Images