Method and System for Monitoring the Stiffness of Rotary Blades of Wind Turbines
By acquiring multi-view images and operational data, combined with pose observation, modal recognition and stiffness monitoring are performed, solving the stability and reliability issues of rotating blade stiffness assessment and achieving high-precision rotating blade stiffness monitoring.
Patent Information
- Application Number
- CN202611358733.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-03
- Publication Date
- 2026-11-17
AI Technical Summary
Existing technologies struggle to suppress the effects of complex coordinate transformations, observation noise, and structural modal coupling in dynamic monitoring scenarios of rotating blades. This results in insufficient stability and reliability of quantitative assessment results for the stiffness of rotating blades, and makes it easy to misjudge their health status.
By acquiring multi-view image sequences, wind turbine operating condition data, and observation equipment pose data, the composite pose prior, pose covariance, and actual visible observation set are determined. Modal coordinate time series and posterior covariance are solved by combining the joint objective function to perform modal identification, output target modal frequency and modal confidence index, and calculate stiffness monitoring results under the condition of satisfying pre-constraints.
It effectively suppresses the adverse interference of coordinate transformation, observation noise and modal coupling, improves the stability and reliability of quantitative assessment of blade stiffness under rotating conditions, and reduces the risk of misjudgment of health status.
Smart Images

Figure CN123109969A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of wind turbine blade monitoring technology, and in particular to a method and system for monitoring the stiffness of rotating wind turbine blades. Background Technology
[0002] In the field of wind power generation, wind turbines typically consist of a tower, nacelle, hub, and multiple rotating turbine blades. As the core load-bearing component of the wind turbine, the rotating turbine blades are susceptible to stiffness decay and localized damage during long-term service due to aerodynamic loads, centrifugal loads, and alternating fatigue. Changes in blade stiffness directly reflect the structural health of the blade. Therefore, it is necessary to conduct structural monitoring on target rotating blades in rotating operation to achieve quantitative assessment of blade stiffness, thereby identifying potential early damage to the blades and ensuring the safe operation of the wind turbine.
[0003] In the existing technology, for the modal and stiffness monitoring of rotating blades, one type of solution uses contact sensors to collect blade vibration response data; the other type of non-contact monitoring solution relies on UAV visual imaging to measure the blade displacement, frequency and mode shape.
[0004] However, existing related technical solutions still have shortcomings. In the dynamic monitoring scenario of rotating blades, due to the combined effects of complex coordinate transformations caused by continuous blade rotation, observation noise, and structural modal coupling, existing non-contact visual monitoring technologies are unable to suppress the adverse effects of various interference factors on the monitoring process. The quantitative assessment results of blade structural characteristics are not stable enough, and the monitoring output accuracy is easily constrained by on-site working conditions. It is difficult to guarantee the reliability of the quantitative assessment results of blade stiffness under rotating conditions, and it is easy to misjudge the health status of blade structure. Summary of the Invention
[0005] Based on this, the purpose of the present invention is to overcome the defects or deficiencies of the prior art and provide a method for monitoring the stiffness of rotating blades of a wind turbine.
[0006] In a first aspect, the present invention provides a method for monitoring the stiffness of rotating blades of a wind turbine, comprising: Acquire a sequence of images of the target rotating blades from different perspectives, wind turbine operating data, and pose observation data of the observation equipment; Based on the image sequence, the operating condition data, and the pose observation data, the composite pose prior, pose covariance, and actual visible observation set are determined respectively; the composite pose prior represents the rigid body transformation from the blade rotation coordinate system corresponding to the target rotating blade to the camera coordinate system corresponding to the observation device; the actual visible observation set represents the pairing relationship between the feature reference points on the target rotating blade and the corresponding effective pixel observations in the image sequence; Based on the composite pose prior, the pose covariance, and the actual visible observation set, determine the modal coordinate temporal sequence and posterior covariance corresponding to each target mode. Modal identification is performed based on the modal coordinate time sequence to obtain the target modal frequency and modal confidence index of the target rotating blade. If the modal coordinate timing, the modal confidence index, and the configuration parameters of the target modal basis all meet the pre-constraint conditions, the target stiffness monitoring result is determined based on the modal quality corresponding to the target mode and the target modal frequency.
[0007] Compared to existing technologies, this invention acquires multi-view image sequences, wind turbine operating condition data, and observation equipment pose observation data to determine the composite pose prior, pose covariance, and actual visible observation set representing the rigid body transformation of the observation equipment relative to the blade rotating coordinate system, as well as the pairing relationship between feature reference points and effective pixel observations. It then solves for the modal coordinate time series and posterior covariance of each target mode using a joint objective function. Based on the modal coordinate time series, it performs modal recognition to obtain the target modal frequency and modal confidence index. The modal coordinate time series, modal confidence index, and target modal basis configuration parameters are used as pre-constraints. Only when all conditions are met is the stiffness monitoring result output in conjunction with modal quality and target modal frequency. This effectively suppresses adverse interference from field factors such as coordinate transformation, observation noise, and modal coupling, quantifies the estimation uncertainty in the modal solution process, and sets multiple admission constraints to filter out unqualified calculation windows. This effectively improves the stability and reliability of the quantitative assessment results of blade stiffness under rotating conditions and reduces the risk of misjudging the health status of the blade structure.
[0008] In one embodiment, determining the composite pose prior, pose covariance, and actual visible observation set based on the image sequence, the operating condition data, and the pose observation data includes: Image features are extracted from the image sequence, and relative pose data are calculated based on the image features; Based on the pose observation data and the relative pose data, and combined with a multi-source fusion algorithm, the first pose transformation relationship is obtained; the first pose transformation relationship represents the rigid body transformation between the camera coordinate system and the ground coordinate system. Based on the impeller azimuth angle in the operating condition data and the pre-calibrated hub geometry parameters, a second pose transformation relationship is obtained; the second pose transformation relationship characterizes the rigid body transformation between the ground coordinate system and the blade rotation coordinate system. Based on the first pose transformation relationship and the second pose transformation relationship, determine the composite pose prior and the pose covariance; Pair the pixel observations of the same feature reference point in the image sequence to obtain a candidate pairing set; Based on the actual working conditions of the image, valid observation pairs are selected from the candidate pairing set to obtain the actual visible observation set.
[0009] This scheme extracts image features from image sequences to calculate relative pose data. It then combines this with pose observation data from the observation equipment and a multi-source fusion algorithm to solve for composite pose priors and pose covariance. This achieves accurate fusion and uncertainty quantification of multi-source pose information, effectively mitigating the error interference of single pose observation data and accurately constructing the rigid body transformation relationship between the observation equipment and the blade rotation coordinate system. Simultaneously, it generates a candidate pairing set through feature pixel observation pairing and filters effective observation pairs based on actual image operating conditions to construct a real visible observation set. This accurately identifies effective observation information and eliminates invalid and interfering observation data, significantly improving the accuracy and effectiveness of observation pairing. This provides high-precision, high-reliability basic observation data for subsequent accurate solution of modal coordinate time series, modal parameter identification, and stiffness monitoring calculations. It reduces monitoring deviations caused by operating condition interference, observation errors, and pose deviations from the source, significantly improving the accuracy and stability of rotating blade stiffness monitoring results.
[0010] In one embodiment, the step of performing modal identification based on the modal coordinate time series to obtain the target modal frequency and modal confidence index of the target rotating blade includes: Based on the impeller azimuth angle in the operating condition data, phase grid resampling processing is performed on the modal coordinate time series, and the blade rotation synchronization component is obtained after multi-cycle phase averaging. The difference between the modal coordinate timing sequence and the blade rotation synchronization component is calculated to obtain the asynchronous residual sequence. Preliminary mode identification is performed based on the asynchronous residual sequence to obtain candidate modal poles and corresponding candidate modal frequencies; The frequency interval is determined based on the candidate modal frequencies and the impeller rotor frequency multiples. The coherence coefficient is determined based on the modal coordinate timing sequence and the corresponding rotor order harmonic reference timing sequence; the rotor order harmonic reference timing sequence is determined based on the impeller azimuth angle. The frequency interval is compared with a preset bandwidth, and / or the coherence coefficient is compared with a preset threshold to obtain a comparison result; Based on the comparison results, the target modal frequency and the modal confidence index are determined.
[0011] This scheme effectively separates the blade rotation synchronization component through phase resampling and multi-cycle phase averaging, and uses residual sequences to eliminate rotational order interference, achieving accurate extraction of the blade structure's asynchronous modal response. A dual discrimination mechanism using frequency intervals and coherence coefficients adaptively distinguishes between mode and rotor order coupling states, effectively reducing identification errors caused by order harmonic interference and modal coupling. Finally, it adaptively outputs reliable target modal frequencies and modal confidence indices, significantly improving the accuracy and robustness of blade modal parameter identification under rotating conditions.
[0012] In one embodiment, determining the target modal frequency and the modal confidence index based on the comparison result includes: When the comparison result is that the frequency interval is less than or equal to the preset bandwidth or the coherence coefficient is greater than the preset threshold, the harmonic basis function and modal pole joint estimation algorithm is adopted to jointly decouple and identify the order forcing term and the structural modal pole based on the modal coordinate time series, so as to obtain the target modal pole and the corresponding target modal frequency and modal confidence index. When the comparison result is that the frequency interval is greater than the preset bandwidth or the coherence coefficient is less than or equal to the preset threshold, the candidate modal pole is determined as the target modal pole and the corresponding target modal frequency and modal confidence index are obtained.
[0013] This scheme adaptively switches the corresponding identification strategy based on the comparison results of frequency intervals and coherence coefficients. When coupling interference occurs between the mode and the rotor order, a joint estimation algorithm is used to decouple and distinguish the order forcing term from the structural modal poles, reducing the contamination of modal parameters by order harmonics. When the coupling degree is low, the candidate modal poles obtained by the solution are directly reused, taking into account the efficiency of identification computation. Thus, it is adapted to modal identification scenarios under different operating conditions of rotating blades, ensuring the reliability of the output results of target modal frequencies and modal confidence indices under complex operating conditions.
[0014] In one embodiment, determining the target stiffness monitoring result based on the modal quality corresponding to the target mode and the target modal frequency includes: Calculate the equivalent modal stiffness index based on the modal mass and the target modal frequency; Based on the aforementioned operating condition data, the equivalent modal stiffness index is corrected using centrifugal stiffening correction term and aerodynamic stiffness correction term to obtain the operating condition corrected equivalent modal stiffness index. The equivalent modal stiffness index or the working condition modified equivalent modal stiffness index is compared with the healthy stiffness reference index under the same working condition to obtain the equivalent modal stiffness change index. The target stiffness monitoring result is determined based on the equivalent modal stiffness index, the working condition-corrected equivalent modal stiffness index, and / or the equivalent modal stiffness change index.
[0015] This scheme calculates the equivalent modal stiffness index based on modal mass and target modal frequency, introduces centrifugal stiffening and aerodynamic stiffness correction terms, and combines these with operating conditions to complete the conditional correction of stiffness, reducing the impact of dynamic parameter deviations caused by rotational conditions. By comparing with healthy stiffness reference indices under the same operating conditions, stiffness change indices are obtained. The monitoring results are output by comprehensively considering the original stiffness, the stiffness after conditional correction, and / or the stiffness change indices. This approach can adapt to different actual operating conditions of the wind turbine, reduce evaluation bias caused by operating condition factors, and achieve a quantitative and reliable assessment of the stiffness state of the rotating blades.
[0016] In one embodiment, determining the temporal sequence of modal coordinates and the posterior covariance corresponding to each order of target mode based on the composite pose prior, the pose covariance, and the actual visible observation set includes: Based on the composite pose prior, the pose covariance, and the actual visible observation set, the modal coordinate temporal series and the posterior covariance are calculated using a joint objective function; wherein, the expression of the joint objective function is: ; in, For robust loss function; For pixel observation covariance; For pose covariance; It is the actual set of visible observations; For pose prior weight coefficients; For time smoothing or modal dynamics operators; For time-series constraint weights; This represents the frame-by-frame pose correction amount corresponding to time t; For the first Modal coordinates of the target mode.
[0017] In one embodiment, the method further includes: The first modality edge information matrix is obtained by reconstructing the edge information matrix based on the robust weights of the convergence points of the joint objective function and the actual visible observation set. Based on the minimum eigenvalue of the first modal edge information matrix and the posterior covariance, a preliminary quality check is performed on the target stiffness monitoring results; If the initial quality verification is satisfactory, the stiffness monitoring results are further verified based on the modal confidence index. If the auxiliary quality verification is qualified, the stiffness monitoring results are output.
[0018] This scheme reconstructs the first modal edge information matrix based on the robust weights of the convergence point of the joint objective function and the actual visible observation set. Preliminary quality verification is then performed using the minimum eigenvalue and posterior covariance of this matrix. This filters out computationally degraded results from the perspectives of observation constraints and solution uncertainties. Further auxiliary quality verification is then conducted using modal confidence indices to further validate the modal matching effect. Through a hierarchical, multi-level quality verification mechanism, unreliable stiffness monitoring conclusions influenced by observational anomalies and modal mismatches are reduced, thereby improving the credibility of the final output results.
[0019] In one embodiment, acquiring a sequence of images of the target rotating blade from different viewpoints includes: When multiple observation devices are deployed in a target observation array, the image sequence is acquired; wherein, the design method of the target observation array includes the following steps: The reference coordinates of the characteristic reference points of the reference blade, the target mode basis and modal mass corresponding to the target mode are obtained; the target mode includes at least one of flapping, oscillation and torsion; the target mode basis represents the displacement distribution of each order of target mode at each characteristic reference point; For each candidate observation array, geometric visibility verification and imaging quality verification are performed on an azimuth grid covering at least one complete rotation cycle, based on the reference coordinates, to filter and obtain a set of predicted visible observations at each azimuth angle; the candidate observation array is used to characterize the spatial position and orientation relationship between each observation device; Based on the predicted visible observation set and combined with the target modality basis, a second modality edge information matrix is constructed; Calculate the minimum eigenvalue of the second-mode edge information matrix corresponding to each candidate observation array; The candidate observation array with the largest minimum eigenvalue is selected as the target observation array.
[0020] This scheme pre-obtains the reference coordinates of the reference blade's characteristic reference points, the target modal basis, and modal quality. It then verifies the geometric visibility and imaging quality of each candidate observation array across the entire rotation cycle to obtain a predicted visible observation set. Combined with the target modal basis, a second modal edge information matrix is constructed. The target observation array is then selected based on the minimum eigenvalue of the matrix, and observation equipment is deployed according to this array to complete data acquisition. By performing simulation and optimization of the array before the acquisition phase, the scheme fully considers the observation constraints throughout the blade's entire rotation cycle, reducing the adverse effects of observation occlusion and poor imaging conditions from the outset. This provides a superior observation configuration foundation for subsequent modal solving and stiffness monitoring, thus improving the overall stability of the entire monitoring process.
[0021] In one embodiment, constructing a second modality edge information matrix based on the predicted visible observation set and the target modality basis includes: For each azimuth angle, the Jacobian matrix is calculated for the predicted visible observation set; the Jacobian matrix includes the modal Jacobian and the pose Jacobian. The Jacobian matrix is weighted and encapsulated, and the pose prior information terms corresponding to the pose prior covariance are superimposed to obtain the joint information submatrix. The joint information submatrices corresponding to all azimuth angles within the complete rotation cycle are weighted and accumulated to obtain a comprehensive joint information matrix covering the entire rotation process; The pose blocks of the integrated joint information matrix are subjected to Schuler complement elimination to obtain the second modality edge information matrix.
[0022] This scheme constructs Jacobian matrices containing modal and pose Jacobian values for the predicted visible observation sets at each azimuth angle. Prior pose information is introduced to form a joint information submatrix. The submatrices for all azimuth angles throughout the entire rotation cycle are accumulated, and then the interference from pose variables is eliminated through Shure complement elimination to obtain the second modal edge information matrix. This approach comprehensively considers the observation constraints and pose uncertainties at different azimuth angles throughout the entire blade rotation process, objectively quantifies the comprehensive observation capabilities of different candidate observation arrays for the target mode, and ensures that the evaluation criteria for array selection closely align with the actual blade rotation conditions, guaranteeing that the selected target observation array possesses full-cycle modal resolution capability.
[0023] Secondly, the present invention also provides a wind turbine rotating blade stiffness monitoring system, comprising: The acquisition module is used to acquire a set of image sequences of the target rotating blade from different perspectives, wind turbine operating condition data, and pose observation data of the observation equipment; The pose and observation set determination module is used to determine the composite pose prior, pose covariance, and actual visible observation set based on the image sequence, the operating condition data, and the pose observation data, respectively. The composite pose prior represents the rigid body transformation from the blade rotation coordinate system corresponding to the target rotating blade to the camera coordinate system corresponding to the observation device. The actual visible observation set represents the pairing relationship between the feature reference points on the target rotating blade and the corresponding effective pixel observations in the image sequence. The joint target calculation module is used to determine the modal coordinate temporal sequence and posterior covariance corresponding to each target mode based on the composite pose prior, the pose covariance and the actual visible observation set; The modal recognition module is used to perform modal recognition based on the modal coordinate time sequence to obtain the target modal frequency and modal confidence index of the target rotating blade. The stiffness calculation module is used to determine the target stiffness monitoring result based on the modal quality and the target modal frequency, provided that the modal coordinate time series, the modal confidence index, and the configuration parameters of the target modal basis all meet the pre-constraint conditions.
[0024] Thirdly, the present invention also provides a computer device, comprising: At least one memory for storing one or more programs; At least one processor, when the one or more programs are executed by the at least one processor, such that the at least one processor performs the steps of the method described in any embodiment of the present invention.
[0025] Fourthly, the present invention also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements the steps of the method described in any embodiment of the present invention.
[0026] To better understand and implement this invention, the following detailed description is provided in conjunction with the accompanying drawings. Attached Figure Description
[0027] Figure 1 This is a schematic diagram of the on-site deployment of a wind turbine rotating blade monitoring system according to an exemplary embodiment; Figure 2 This is a structural block diagram of a wind turbine rotating blade stiffness monitoring system according to an exemplary embodiment; Figure 3 This is a schematic diagram illustrating the execution process of a wind turbine rotating blade stiffness monitoring system according to an exemplary embodiment; Figure 4 This is an internal structural diagram of an electronic device according to an exemplary embodiment. Detailed Implementation
[0028] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.
[0029] Please see Figure 1 , Figure 1This is a schematic diagram of the on-site deployment for monitoring the rotating blades of a wind turbine according to the present invention. The present invention uses a group of unmanned aerial vehicles (UAVs) C, each equipped with an image acquisition unit C1 and a pose sensing sensor. Multiple UAVs C are deployed in a pre-defined observation array. On one hand, they acquire image sequences of the target rotating blade B1 of the wind turbine B from multiple perspectives; on the other hand, the pose sensing sensors output their own pose observation data at the same time reference. Simultaneously, the operating condition sensing unit on the wind turbine B can output operating condition data such as impeller azimuth angle and rotational speed. A wind turbine rotating blade stiffness monitoring system D communicates with the multiple groups of UAVs C and the operating condition sensing unit of the wind turbine B via a communication link, receiving image sequences captured by the target rotating blade B1 and data from the UAVs C. The system collects pose observation data and wind turbine operating condition data, and determines the composite pose prior, pose covariance, and actual visible observation set based on the received multi-source data. It uses a joint objective function to solve for the modal coordinate time series and posterior covariance, completes modal identification to obtain the target modal frequency and modal confidence index, calculates the equivalent modal stiffness index after verifying the pre-constraint conditions, and obtains the operating condition corrected equivalent modal stiffness by combining the centrifugal stiffening correction term and the aerodynamic stiffness correction term. It compares the equivalent modal stiffness change index with the healthy stiffness reference index, and outputs the target stiffness monitoring results after multi-level quality verification, thereby improving the accuracy and reliability of the target rotating blade stiffness monitoring under rotating conditions.
[0030] Specifically, the operating condition data of a wind turbine is a set of state parameters collected during the operation of the wind turbine. It includes at least the impeller azimuth angle parameter, and may also include data that can characterize the real-time operating status of the wind turbine, such as impeller speed, output power, and wind speed.
[0031] Specifically, pose observation data refers to the raw positioning and attitude measurement data output by the observation equipment itself, which may include position information and attitude angle information, providing raw observation input for solving the composite pose prior. For example, pose observation data may include, but is not limited to, real-time kinematic (RTK) data, inertial measurement unit (IMU) data, and relative ranging data between UAVs.
[0032] When designing an observation array to collect image sequences of the target rotating blades of a wind turbine, the number and type of observation equipment can be configured according to requirements. For example, at least two drones and one fixed camera can be configured to form an observation array, or three or more drones can be configured to form an observation array. After determining the number and type of observation equipment to be deployed, multiple sets of alternative observation arrays are generated within the permitted operating space for wind turbine blade monitoring. For example, with three drones, the alternative observation array is as follows: the first drone provides an axial observation view close to the hub, the second and third drones provide a lateral oblique view and a vertical oblique view, respectively. All drones are deployed outside the impeller sweep area, and geofencing, drone collision avoidance distances, and return directions in case of loss of contact are configured. Then, based on preset optimization constraints, a preliminary feasibility screening of multiple candidate deployment schemes is conducted. These preset constraints include, but are not limited to, the UAV not entering the impeller sweep area, the distance between the UAV and the impeller, and the distance between UAVs themselves, exceeding the minimum safe distance, maintaining a predetermined number of feature reference points at each azimuth angle, and ensuring that the intersection angle, image resolution, communication link, and remaining endurance meet preset requirements. Candidate observation formations that do not meet any of these preset optimization constraints are eliminated, resulting in candidate observation formations. Finally, the target observation formation is selected from the candidate formations.
[0033] Specifically, the design method for target observation arrays includes the following steps: A1: Obtain the reference coordinates of the feature reference points of the reference blade, the target modal basis corresponding to the target mode, and the modal mass. The reference blade is a blade model entity or finite element simulation model used for pre-simulation analysis, providing prior modal parameters, and its structure is consistent with the target rotating blade to be monitored. Feature reference points are pre-defined discrete spatial points on the reference blade, used to characterize the spatial position of the blade surface, corresponding to observable features such as natural texture, contour, coded markings, and speckle. Reference coordinates refer to the spatial coordinate values of the feature reference points in the blade's rotating coordinate system. For example, reference coordinates can be obtained through various methods: they can be directly extracted from the finite element model or parametric beam model; the corresponding coordinates can be obtained by reading the design model from the blade manufacturing stage; they can also be obtained through photogrammetry when the blade is stopped; or the coordinates of the feature reference points can be obtained by collecting observation data of a healthy wind turbine blade during normal operation and solving it through 3D reconstruction. All feature reference point coordinates obtained from these sources are uniformly mapped to the blade's rotating coordinate system. The target mode includes at least one of flapping, oscillation, and torsion. The target modal basis characterizes the displacement distribution of each target mode at various characteristic reference points. Modal mass refers to the generalized modal mass calculated for each target mode, which is an intrinsic parameter obtained from blade dynamics simulation or modal testing.
[0034] In some embodiments, a ground coordinate system can be pre-established. Hub coordinate system Rotating coordinate system of blades and camera coordinate system And determine the first rotation matrix of the hub coordinate system relative to the ground coordinate system. The position of the hub center in the ground coordinate system is The second rotation matrix of the blade rotating coordinate system relative to the hub coordinate system and the composite rotation matrix from the blade rotating coordinate system to the ground coordinate system ;in, This represents the impeller azimuth angle. Reference coordinates of the characteristic reference point. Defined in the blade rotating coordinate system, the target modal basis contains the modal displacement vectors of each target mode at each characteristic reference point. Solving for modal quality can be achieved by performing modal normalization on the modal basis vectors. This normalization can be performed using the reference quality matrix M, satisfying the following conditions: Corresponding modal mass ;in, Let be the transpose of the modal basis vectors (mode shape vectors) corresponding to the j-th order target mode. This represents the modal basis vector corresponding to the j-th order target mode; alternatively, the modal displacement of the reference feature point along a set direction can be specified as 1, and then... The modal quality is calculated. The baseline state and the subsequent actual monitoring state need to use the same set of reference modal basis, modal quality and normalization method, and the current blade response can still be stably characterized by the reference modal basis through the Modal Assurance Criterion (MAC).
[0035] A2: For each candidate observation array, geometric visibility verification and imaging quality verification are performed on an azimuth grid covering at least one complete rotation cycle, based on reference coordinates, to obtain the predicted visible observation set for each azimuth angle. The candidate observation array characterizes the spatial position and orientation relationship between each observation device. The azimuth grid is a set of discrete azimuth sampling points obtained by discretely sampling the impeller's complete rotation cycle, covering at least one complete rotation cycle of the blade. Geometric visibility verification refers to the simulation verification process of determining whether each feature reference point on the reference blade is within the field of view of the observation device and is not obstructed by the blade itself, given an azimuth angle and candidate observation array. Imaging quality verification refers to the simulation evaluation of the imaging conditions of the feature reference points on the image plane of the observation device, including the verification process of constraints such as observation distance, imaging angle, and projection resolution, given an azimuth angle and candidate observation array. The predicted visible observation set refers to the theoretical pairing set between feature reference points and observation devices obtained after geometric visibility verification and imaging quality verification at each impeller azimuth angle during the simulation phase.
[0036] In some embodiments, for each candidate observation array, an azimuth grid covering at least one complete rotation cycle is traversed, and a composite rotation matrix from the blade rotation coordinate system to the ground coordinate system is combined. The position vector of the hub center in the ground coordinate system The system predicts the position of each feature reference point in the ground coordinate system based on the first formula. Further coordinate transformation yields the depth, pixel position, and intersection of the line of sight with the blade surface of the feature reference point in the observation equipment coordinate system. Simultaneously, the feature imaging quality is evaluated, and geometric visibility and imaging quality checks are performed sequentially. Geometric visibility check determines whether the feature reference point is in front of the observation equipment, falls within the effective image plane of the observation equipment, and whether the line of sight from the observation equipment to the feature reference point is not obstructed by the blade itself, hub, tower, or other components. Imaging quality check verifies whether the imaging scale, incident angle, blurriness, and texture quality all meet the target thresholds. Only when both geometric visibility and imaging quality checks pass are the corresponding observation pairs included in the predicted visible observation set. Among them, the observation pair represents the imaging pairing relationship between a single feature reference point and a single observation device.
[0037] For example, the expression for the first formula can be: ;in, Feature reference point Position vector in the ground coordinate system; This is the composite rotation matrix from the blade rotating coordinate system to the ground coordinate system; Feature reference point Reference coordinates in the blade rotating coordinate system; For the first The target mode at the feature reference point Modal displacement vector at; For the first Modal coordinates of the target mode; This is the position vector of the hub center in the ground coordinate system.
[0038] A3: Based on the predicted visible observation set and combined with the target mode basis, a second mode edge information matrix is constructed. The second mode edge information matrix indicates the matrix obtained after the simulation optimization stage, which is constructed based on the predicted visible observation set and combined with the target mode basis, and is encapsulated by the Jacobian matrix, accumulated by the full-period joint information submatrix, and processed by the Shure complement elimination. It is used to evaluate the observation resolution capability of the candidate observation array for the target mode.
[0039] In some embodiments, constructing a second modality edge information matrix based on the predicted set of visible observations and in conjunction with the target modality basis includes: A31: For each azimuth angle, calculate the Jacobian matrix corresponding to the set of predicted visible observations; the Jacobian matrix includes the modal Jacobian and the pose Jacobian.
[0040] In some embodiments, for each azimuth angle corresponding to the predicted visible observation set, the Jacobian matrix is solved one by one for each effective observation feature point by combining the preset target mode basis with the second formula. The Jacobian matrix is divided into two parts, namely the modal Jacobian and the pose Jacobian. The modal Jacobian is obtained by mapping the target mode basis and is used to establish the mapping relationship between pixel observation and blade modal vibration parameters. The pose Jacobian is used to establish the mapping relationship between pixel observation and the pose correction of the observation equipment, and fully characterizes the differential correlation characteristics between the observed variables and the parameters to be solved.
[0041] For example, the second formula can be expressed as follows: ;in, For the predicted observation set Determined selection matrix; superscript This means that after filtering by the selection matrix, only the matrix version of the visible observations is retained; Modal Jacobian indicates the Jacobian of the image plane pixel observation with respect to the target modal coordinates; For pose Jacobian, it indicates the Jacobian of the frame-by-frame pose correction amount for image plane pixel observations; The original pixel observation covariance matrix contains only pixel localization noise, feature reference point matching error, and the correlation between the above errors, but does not include the uncertainty of the observation device pose. When performing visibility screening for each feature reference point, the imaging quality of the (observation device-feature reference point) observation pair is evaluated, such as blurriness, matching confidence, and image signal-to-noise ratio. Based on the imaging quality, a noise level is assigned to each observation, and the original pixel observation covariance matrix is assembled.
[0042] A32: Weight the Jacobian matrix and add the pose prior information terms corresponding to the pose prior covariance to obtain the joint information submatrix.
[0043] In some embodiments, the Jacobian matrices calculated for each azimuth angle are weighted and encapsulated. The observation noise is weighted using pixel observation covariance, and a pose prior information term constructed from the pose prior covariance is superimposed to achieve the fusion of modal parameters, pose parameters, observation noise, and prior constraints. Finally, a joint information submatrix corresponding to a single azimuth angle is obtained. The specific construction method is described in Formula 3. During matrix solving, if the pose-related matrix is close to singular and exhibits instability in inversion, a small normal number ε that does not change the candidate observation array order is introduced to participate in the calculation, ensuring the stability and reliability of the matrix inversion operation.
[0044] For example, the expression for the third formula can be as follows: ; in, , The instruction specifies the total Jacobian matrix after filtering by the selection matrix, and concatenates them column by column. This is the pixel observation covariance matrix after being filtered by the selection matrix; It is a block diagonal matrix. For the pose prior information item, the prior information block corresponding to the modal coordinates is a zero matrix; For the prior covariance of pose The resulting pose prior accuracy matrix , Indicates the prior weights of the pose; Represents the joint information submatrix, with subscripts. , , and Represent modality-modality, modality-pose, pose-modality, and pose-pose blocks respectively. According to the target modal coordinates and pose correction amount Divided into blocks , , and ;in, The pose prior accuracy matrix is already included. Therefore, pose covariance is no longer included. This avoids the uncertainty of the same pose being repeatedly weighted.
[0045] A33: Weighted accumulation of the joint information submatrices corresponding to all azimuth angles within the complete rotation cycle yields a comprehensive joint information matrix covering the entire rotation process.
[0046] In some embodiments, the joint information submatrices corresponding to all azimuth angles within the complete rotation cycle can be weighted and accumulated along the azimuth angle grid according to the azimuth angle weights to obtain a comprehensive joint information matrix covering the entire rotation process.
[0047] A34: Perform Schur complement elimination on the pose block of the integrated joint information matrix to obtain the second modality edge information matrix.
[0048] In some embodiments, Schur complement elimination is performed on the pose blocks of the integrated joint information matrix to obtain the second modality edge information matrix. The calculation method is detailed in the fourth formula, which is expressed as follows: ; in, The second modal edge information matrix represents the edge information provided by pixel observation to the target modal coordinates when the pose correction is unknown and subject to pose prior constraints. This refers to either azimuth weighting or effective observation quality weighting. Indicates the acquisition window Summing of all sampling times within the range; azimuth weighting It can be set according to the phase grid equalization and satisfy the requirements. Alternatively, adjustments can be made based on the number of visible feature reference points and the quality of predicted features, while maintaining the lowest weight for each phase; during calculation, adjustments can be made based on... Additional Perform eigenvalue decomposition stabilization, and... Additional Perform Shure-Supplement stabilization; and Represents a small positive constant with stable numerical values, where the subscript... and These represent the modal coordinate block and the pose block, respectively. Represents the identity matrix of the corresponding dimension.
[0049] A4: Calculate the minimum eigenvalue of the edge information matrix of the second mode corresponding to each candidate observation array; where the minimum eigenvalue represents the degree of discernibility of the weakest mode in each target mode under pose interference conditions. The larger the minimum eigenvalue, the stronger the observation and identification capability of the candidate observation array for the target mode.
[0050] A5: Select the candidate observation formation with the largest minimum eigenvalue as the target observation formation.
[0051] In some embodiments, for the edge information matrix of the second mode corresponding to each candidate observation array, at least one of the following can be used as the array evaluation index: minimum eigenvalue, determinant, or logarithmic determinant increment brought about by the addition of a new candidate observation pose. The determinant represents the overall comprehensive observation information of all target modes; the logarithmic determinant increment brought about by the addition of a new candidate observation pose is used to represent the improvement in modal observation information after adding a new observation path, and is suitable for scenarios involving local array adjustment and reconstruction. The preferred target observation array is obtained by screening from the candidate observation arrays using the evaluation index. When the eigenvector corresponding to the minimum eigenvalue indicates insufficient directional information for a certain target mode, the increment of the evaluation index after moving at least one UAV or changing the camera orientation can be calculated in the candidate observation poses. The feasible adjustment scheme with the largest increment is selected, and array reconstruction is performed only between adjacent acquisition windows.
[0052] For example, one way to filter target observation formations based on the minimum eigenvalue can be as follows: ; in, For target observation formation; This indicates that the formation that maximizes the subsequent evaluation quantity is selected from the candidate observation formations; Represents the smallest eigenvalue of a matrix; Used to evaluate the discriminability of the weakest direction of the target modality after pose disturbance elimination; This is the edge information matrix for the second mode.
[0053] In one embodiment, to achieve data synchronization among multiple observation devices and simultaneously align the time of image sequences, pose observation data, and operational condition data, the multiple observation devices can establish a unified time base and impeller azimuth reference based on any one or more combinations of Global Navigation Satellite System Pulse Per Second (GNSS PPS), Institute of Electrical and Electronics Engineers (IEEE) Precision Time Protocol, and wireless hardware triggering methods. They can continuously acquire image sequences from different perspectives at least two complete blade rotation cycles, with each image frame carrying a hardware timestamp. The image sequences carrying hardware timestamps are then correlated and matched with at least one operational condition data point of the wind turbine, ensuring the standard deviation of the time synchronization error meets the phase error constraint, and simultaneously outputting the pose observation data of each observation device.
[0054] For example, the formula for the standard deviation of time synchronization error to satisfy the phase error constraint can be: ;in, The standard deviation of the time synchronization error; The maximum permissible phase error (in radians); The highest target modal frequency.
[0055] The acquisition frame rate of the observation equipment can be set to be greater than twice the highest target modal frequency to meet the Nyquist sampling criterion and prevent modal frequency aliasing distortion. Furthermore, it can be configured to complete image sampling at least a number of times within each highest target modal vibration cycle, such as nine, ten, or eleven times, to increase the sampling density of the modal coordinate time-series data. Simultaneously, the exposure time of the observation equipment is adaptively adjusted according to the image plane motion speed generated by the blade rotation and vibration, controlling the displacement of the rotating target blade on the imaging plane within the exposure time period to not exceed a preset pixel threshold, suppressing image blurring caused by blade motion, and ensuring clear and complete blade feature information in the image sequence.
[0056] Please also refer to Figure 2 and Figure 3 , Figure 2 This is a schematic diagram of the wind turbine rotating blade stiffness monitoring system of the present invention. Figure 3 for Figure 2 The diagram shows the execution flow of a wind turbine rotating blade stiffness monitoring system. This system includes an acquisition module 10, a pose and observation set determination module 20, a joint target calculation module 30, a modal recognition module 40, and a stiffness calculation module 50.
[0057] The acquisition module 10 is used to perform step S1: acquire a set of image sequences of the target rotating blade from different perspectives, wind turbine operating condition data, and pose observation data of the observation equipment.
[0058] The pose and observation set determination module 20 is used to execute step S2: based on the image sequence, operating condition data and pose observation data, determine the composite pose prior, pose covariance and actual visible observation set respectively; the composite pose prior represents the rigid body transformation from the blade rotation coordinate system corresponding to the target rotating blade to the camera coordinate system corresponding to the observation device; the actual visible observation set represents the pairing relationship between the feature reference points on the target rotating blade and the corresponding effective pixel observations in the image sequence.
[0059] The joint objective calculation module 30 is used to execute step S3: based on the composite pose prior, pose covariance and actual visible observation set, the joint objective function is used to calculate the modal coordinate time series and posterior covariance corresponding to each target mode.
[0060] The modal recognition module 40 is used to perform step S4: perform modal recognition based on the modal coordinate time sequence to obtain the target modal frequency and modal confidence index of the target rotating blade.
[0061] The stiffness calculation module 50 is used to execute step S5: under the condition that the modal coordinate time series, modal confidence index and the configuration parameters of the target modal basis all meet the pre-constraint conditions, the target stiffness monitoring result is determined according to the modal quality and target modal frequency corresponding to the target mode.
[0062] Specifically, the acquisition module 10 includes a first transmission unit, a second transmission unit, and a third transmission unit.
[0063] The first transmission unit is used to perform step S11: acquire a set of image sequences of the target rotating blade from different perspectives.
[0064] The second transmission unit is used to perform step S12: acquiring the operating condition data of the wind turbine.
[0065] The third transmission unit is used to perform step S13: acquire pose observation data of the observation equipment.
[0066] Specifically, the pose and observation set determination module 20 includes a visual pose calculation unit, a pose fusion unit, a pairing unit, and a filtering unit.
[0067] The visual pose calculation unit is used to perform step S21: extracting image features from the image sequence and calculating relative pose data based on the image features.
[0068] The image features can include, but are not limited to, any fixed visual feature located on the stationary structure of the wind turbine, such as the nacelle, tower, or hub. Relative pose data indicates the pose transformation relationship between the observation equipment and the stationary structure of the wind turbine, or between multiple observation equipment.
[0069] In some embodiments, the visual pose calculation unit can sequentially perform grayscale conversion, filtering and noise reduction, and contrast enhancement preprocessing on a single frame image in an image sequence, and then call the corresponding feature detection algorithm to extract image features of fixed static structures such as nacelles, towers, or hubs; perform feature matching on image features at the same location in image sequences from different perspectives collected by multiple observation devices, and calculate the relative pose data between the observation device and the fixed structure of the wind turbine, and between multiple observation devices, based on the matching results.
[0070] The pose fusion unit is used to execute step S22: based on pose observation data, relative pose data, and operational condition data, to solve for the composite pose prior and pose covariance. Specifically, this includes: S221: Based on pose observation data and relative pose data, and combined with a multi-source fusion algorithm, the first pose transformation relationship is obtained; the first pose transformation relationship represents the rigid body transformation between the camera coordinate system and the ground coordinate system.
[0071] In some embodiments, at least two types of pose data are selected for multi-source information fusion. For example, fusion is performed based on at least two types of data in pose observation data, or based on at least two types of data in relative pose data, or based on at least one type of data in both pose observation data and relative pose data. The rigid body transformation of the camera coordinate system of each observation device relative to the ground coordinate system, i.e., the first pose transformation relationship, is then obtained.
[0072] S222: Based on the impeller azimuth angle in the operating condition data and the pre-calibrated hub geometry parameters, the second pose transformation relationship is obtained; the second pose transformation relationship characterizes the rigid body transformation between the ground coordinate system and the blade rotation coordinate system.
[0073] In some embodiments, the real-time impeller azimuth angle from the operating condition data and pre-stored hub geometric calibration parameters are read. These hub geometric calibration parameters include fixed geometric transformation parameters obtained from factory calibration, such as hub installation eccentricity, initial zero-position offset of the hub, and the nacelle-hub rigid body installation transformation matrix. First, using the hub geometric calibration parameters, a fixed rigid body transformation from the ground coordinate system to the hub-machine body coordinate system is constructed. Then, based on the real-time impeller azimuth angle, a rotation matrix around the impeller's main rotation axis is constructed to characterize the attitude rotation of the impeller as the rotor rotates. The above fixed rigid body transformation is then multiplied sequentially by the rotation matrix corresponding to the azimuth angle to obtain the second attitude transformation relationship.
[0074] S223: Determine the composite pose prior and pose covariance based on the first pose transformation relationship and the second pose transformation relationship.
[0075] In some embodiments, the first pose transformation relation and the second pose transformation relation are combined to obtain the frame-by-frame composite pose prior and the uncertainty statistic corresponding to the pose estimation result of the observation device, i.e., the pose covariance. The pose covariance can be used to construct a frame-by-frame block diagonal covariance matrix according to the pose correction amount. The inverse of this matrix is scaled to obtain the pose prior accuracy matrix. For example, a calculation method for determining the composite pose prior can be as follows: ; in, This is the first pose transformation relationship; This is the second pose transformation relationship.
[0076] The pairing unit is used to perform step S23: pairing the pixel observations of the same feature reference point in the image sequence to obtain a candidate pairing set.
[0077] Among them, the pixel observation value indicates the pixel coordinate measurement value corresponding to the feature reference point in a single frame image.
[0078] In some embodiments, for each time t of the image sequence within the current acquisition window, the impeller azimuth angle corresponding to that time is obtained, the predicted visible observation set under the corresponding azimuth angle is retrieved, the range of feature reference points predicted to be visible by the observation equipment is determined, feature matching and sub-pixel localization are performed on the feature reference points within the feature reference point range, and the pixel observation value corresponding to the i-th observation equipment and the a-th feature reference point at time t is extracted. Using "observation equipment - feature reference point - time" as the basic unit for a single observation pairing, the observation pairs that have completed the initial matching together constitute a candidate pairing set.
[0079] The filtering unit is used to perform step S24: based on the actual working conditions of the image, filter out the effective observation pairs from the candidate pairing set to obtain the actual visible observation set.
[0080] The actual image working conditions indicate the set of constraints used to filter valid observation pairs, which may include, but are not limited to, at least one of the following: image plane imaging range constraints, occlusion constraints, image blur threshold, and matching confidence threshold.
[0081] In some embodiments, for each observation pair in the candidate pairing set, validity verification is performed sequentially through multiple preset image actual working conditions. The image actual working conditions may include, but are not limited to, image plane imaging range constraints, occlusion constraints, image blur thresholds, and matching confidence thresholds. Among them, the image plane imaging range constraints are used to verify whether the projected pixel coordinates of the feature reference point are within the effective imaging area of the image; the occlusion constraints are used to verify whether the feature reference point is occluded by objects such as the blade structure itself, the nacelle, and the tower based on the geometric projection relationship; the image blur threshold is used to verify whether the image clarity of the neighborhood of the feature reference point meets the observation accuracy requirements; and the matching confidence threshold is used to verify whether the reliability of the feature matching result meets the preset standard. Candidate observation pairs that do not meet any of the verification conditions are eliminated, and all valid observation pairs that pass the verification are retained. All valid observation pairs together constitute the actual visible observation set.
[0082] In one embodiment, step S3 includes: for each time t within the acquisition window, incorporating the effective observation pairs contained in the actual visible observation set into the inversion calculation, and constructing a pixel observation equation based on the camera perspective projection function, as shown below: ; in, Let be the pixel observation value of the i-th observation device and the a-th feature reference point at time t; • This is the camera perspective projection function; For the first Composite pose priors of multiple cameras; This means that the six-dimensional pose correction is converted into a rigid body transformation by the exponential mapping of SE(3) (Special Euclidean Group) and then multiplied by the composite pose prior on the right; SE(3) is a Lie group composed of three-dimensional rigid body transformations; This represents the frame-by-frame pose correction amount corresponding to time t. Indicates increment; This refers to pixel observation error; Feature reference point Reference coordinates in the blade rotating coordinate system; For the first The target mode at the feature reference point Modal displacement vector at the location (target mode basis); For the first Modal coordinates of the target mode.
[0083] Let the pixel prediction value be the result calculated on the right-hand side of the pixel observation equation without error terms. The pixel observation residual is defined as the difference between the observed pixel value and the predicted pixel value. Based on pixel observation residuals, pose prior constraints, and modal smoothness constraints, a joint objective function is constructed. The unknowns are solved by minimizing this objective function. The expression of the joint objective function is as follows: ; in, For robust loss functions, at least one of Huber loss, Cauchy loss, or other robust loss methods may be used; Pixel observation covariance, which characterizes the pixel localization and matching covariance without including the uncertainty of the observation device pose; For pose covariance; It is the actual set of visible observations; These are the pose prior weight coefficients, defined in the same way as the scaling factor for pose prior accuracy; These are time-smoothing or modal dynamics operators used to constrain the temporal variation of modal coordinates; For time-series constraint weights; Let be the frame-by-frame pose correction amount corresponding to time t, and be the unknown quantity to be determined. For the first The modal coordinates of the target mode are unknowns to be determined. The modal coordinates of the last term in the formula are stacked as a column vector according to the time intervals within the acquisition window. This term is equivalent to the squared L2 norm of the time smoothing or dynamic residual.
[0084] During the solution process, the local normal matrix corresponding to the joint objective function is determined according to... and The solution is divided into blocks, and Schur complement elimination is performed on the pose blocks to reduce the solution dimensionality. This Schur complement operation is of the same origin as the pose block Schur complement in the formation planning stage. When the operator L is a dynamic operator with unknown modal frequencies, the theoretical natural frequencies of each target mode output by the finite element model of the reference blade are used as the initial substitution frequencies. If it is a rolling calculation with a continuous acquisition window, or multiple iterations within the same acquisition window, the target modal frequencies identified in the previous acquisition window or the previous iteration are used as the initial substitution frequencies. The solution is gradually optimized through iteration. After the joint objective function iterates to the convergence point, the normal solution matrix is reconstructed based on the robust weights corresponding to each effective observation pair in the convergence state, combined with the actual visible observation set, and the final accurate solution is performed to ensure the numerical robustness of the solution results. By directly inverting the pixel domain, the core unknowns are set as low-dimensional modal coordinates. The solution process is optimized directly based on the pixel observation residuals. There is no need to reconstruct the dense three-dimensional point cloud of the blade surface frame by frame before performing modal decomposition. This can reduce the accumulation of errors in intermediate steps and improve computational efficiency.
[0085] After solving, the posterior covariance corresponding to the modal coordinates is calculated. The pose block matrices at each time step are arranged according to the state order of the acquisition window. After performing pose block Shur complement elimination, the following is added: The regularization term is then inverted to obtain the posterior covariance. The regularization term corresponding to the modal temporal constraint is only included in the posterior covariance calculation and is not included in the evaluation of the optical observation information of the array. Furthermore, the pose covariance is only introduced through the pose prior constraint term and will not be repeatedly included in the pixel observation covariance, thus avoiding the error being repeatedly weighted.
[0086] Specifically, the modality recognition module 40 includes a phase resampling processing unit, a residual solving unit, a preliminary modality recognition unit, a frequency interval calculation unit, a coherence coefficient calculation unit, a threshold comparison unit, and a modality parameter determination unit.
[0087] The phase resampling processing unit is used to execute step S41: based on the impeller azimuth angle in the operating condition data, perform phase grid resampling processing on the modal coordinate time sequence, and obtain the blade rotation synchronization component after multi-cycle phase averaging.
[0088] Among them, phase grid resampling is a signal preprocessing method for order tracking. It converts the time-series signal sampled at equal time intervals into an angular domain sequence with equal rotational phase intervals based on the rotor phase information, thereby improving the problem of uneven time axis sampling caused by speed fluctuations and achieving data alignment between different rotational cycles.
[0089] Among them, the blade rotation synchronization component is a well-known synchronous vibration component in rotating machinery signals, which strictly corresponds to the impeller rotation period. It mainly includes the vibration response generated by impeller rotation and periodic aerodynamic excitation, and can be extracted by multi-period phase domain averaging.
[0090] In one embodiment, based on the impeller azimuth angle in the operating condition data, phase grid resampling processing is performed on the modal coordinate time series, mapping the time-sampled modal coordinate time series onto a unified impeller rotation phase grid, for the k-th phase grid point. ,use Multi-cycle phase averaging calculations are performed over a complete rotation cycle to obtain the blade rotation synchronization component. The calculation formula is as follows: ;in, For the first The first-order target mode in the 1st order Phase grid points The corresponding rotational synchronization component is indicated by the horizontal line above, which represents the multi-cycle phase averaging. The number of complete rotation cycles used in the average calculation; The rotation period number; For the first azimuth angle during one rotation cycle The corresponding modal coordinates.
[0091] The residual solving unit is used to execute step S42: to calculate the difference between the modal coordinate timing sequence and the blade rotation synchronization component to obtain the asynchronous residual sequence.
[0092] Among them, the asynchronous residual sequence is the residual signal obtained after removing the rotational synchronization component from the original response sequence. It belongs to the asynchronous vibration component in the field of rotating machinery. This sequence removes the periodic forced response caused by rotor frequency harmonics and mainly retains the vibration information corresponding to the structure's own modes.
[0093] In one embodiment, based on the impeller azimuth angle corresponding to time t, the value is taken from the blade rotation synchronization component obtained by averaging the phase grid. The modal coordinate time sequence is subtracted from the blade rotation synchronization component corresponding to that azimuth angle to remove the periodic synchronous excitation response caused by impeller rotation, and the asynchronous residual sequence is output. The calculation formula is as follows: ;in, Let be the asynchronous residual corresponding to the j-th order target mode at time t, with the superscript apostrophe indicating the residual; For a moment Modal coordinates; According to time The impeller azimuth angle is the synchronous component value obtained by interpolation from the blade rotation synchronous component obtained by multi-cycle phase averaging. This represents the phase synchronization response, which includes the effects of gravity, wind shear, and other rotor synchronization excitations.
[0094] The preliminary mode identification unit is used to perform step S43: perform preliminary mode identification based on the asynchronous residual sequence to obtain candidate mode poles and corresponding candidate mode frequencies.
[0095] In some embodiments, preliminary identification of operating modes is carried out based on the obtained asynchronous residual sequence. Time-domain modal parameter identification is performed on the asynchronous residual sequence, the system eigenvalues in complex domain form are solved, and a set of candidate modal poles are output. The imaginary part of the pole corresponds to the vibration angular frequency, and the real part represents the modal damping. The corresponding candidate modal frequency is obtained by converting the imaginary part of the candidate modal pole, and the modal damping information is output simultaneously from the real part of the pole. The candidate modal poles obtained in this stage include the actual blade structure modal poles, but may also contain false poles caused by noise and residual harmonics. Further screening and verification are required in combination with frequency interval and coherence coefficient.
[0096] The frequency interval calculation unit is used to perform step S44: determine the frequency interval based on the candidate modal frequency and the impeller rotor frequency multiple.
[0097] Among them, the impeller rotor frequency multiplication characterizes the integer multiple harmonic frequencies obtained by taking the impeller rotor rotation frequency as the fundamental frequency. The impeller rotor rotation frequency can be calculated from the impeller azimuth angle time sequence in the operating condition data. The impeller rotor frequency multiplication is used to distinguish whether the candidate modal frequencies obtained are the natural vibration frequencies of the blades or the rotational forced harmonic components induced by the impeller rotation.
[0098] In some embodiments, the frequency interval is determined by calculating the absolute value of the difference between the candidate modal frequency and the impeller rotor's r-times.
[0099] The coherence coefficient calculation unit is used to execute step S45: determine the coherence coefficient based on the modal coordinate timing and the corresponding rotor order harmonic reference timing; the rotor order harmonic reference timing is determined based on the impeller azimuth angle.
[0100] Among them, the rotor order harmonic reference time sequence is a standard harmonic time sequence constructed based on the impeller azimuth angle, which corresponds one-to-one with the impeller rotation order and is used as a reference signal for correlation comparison with the actual vibration response.
[0101] The coherence coefficient is a coherence function derived from signal processing, and its value is 0. 1. Used to characterize the correlation strength between the modal coordinate timing and the rotor order harmonic reference timing. The higher the value, the stronger the response is affected by rotor rotation harmonic interference.
[0102] In some embodiments, a rotor order harmonic reference timing sequence is constructed based on the impeller azimuth angle, and the coherence function between the modal coordinate timing sequence and the rotor order harmonic reference timing sequence is calculated to obtain the coherence coefficient.
[0103] The threshold comparison unit is used to perform step S46: comparing the frequency interval with the preset bandwidth, and / or comparing the coherence coefficient with the preset threshold to obtain the comparison result.
[0104] The modal parameter determination unit is used to perform step S47: determine the target modal frequency and modal confidence index based on the comparison results.
[0105] The target modal frequency is the vibration frequency obtained by converting the imaginary part of the target modal poles.
[0106] Among them, the modal confidence index is a comprehensive quantitative index used to quantify the credibility of modal identification results. It reflects the degree of matching between the identified modal parameters and the actual vibration modes of the blade. It can integrate multiple information such as frequency isolation degree, signal coherence, mode shape matching degree, and residual level. The higher the value, the higher the reliability of the current modal identification result.
[0107] In some embodiments, the corresponding target modal poles are calculated based on the comparison results. The undamped operating modal angular frequency in the rotating coordinate system is obtained by converting the imaginary part of the target modal poles and used as the target modal frequency. The modal confidence index is then calculated by combining at least two of the following: mode shape similarity, frequency interval, coherence coefficient, and inversion residual level. The mode shape similarity can be obtained by calculating the modal confidence criterion.
[0108] In some embodiments, step S47 includes: S471: When the comparison result is that the frequency interval is less than or equal to the preset bandwidth or the coherence coefficient is greater than the preset threshold, the joint estimation algorithm of harmonic basis function and modal pole is adopted. Based on the modal coordinate time series, the order forcing term and the structural modal pole are jointly decoupled and identified to obtain the target modal pole and the corresponding target modal frequency and modal confidence index. S472: When the comparison result is that the frequency interval is greater than the preset bandwidth or the coherence coefficient is less than or equal to the preset threshold, the candidate modal pole is determined as the target modal pole and the corresponding target modal frequency and modal confidence index are obtained.
[0109] In some embodiments, when the frequency interval between the candidate modal frequency and the impeller rotor's r-fold frequency is less than or equal to a preset bandwidth threshold, or when the coherence coefficient between the modal coordinate time series and the corresponding order rotor harmonic reference time series exceeds a preset threshold, instead of directly eliminating all synchronous components, at least one algorithm among harmonic basis function and modal state joint estimation, phase conditional random subspace identification, order tracking combined with operational modal analysis, Bayesian operational modal analysis, or linear periodic time-varying system identification can be used to synchronously estimate the order forcing term and structural modal poles, avoiding the erroneous deletion of the actual structural vibration response locked by the order; when the frequency interval between the candidate modal frequency and the impeller rotor's r-fold frequency is greater than a preset bandwidth threshold, or when the coherence coefficient is less than or equal to a preset threshold, the candidate modal poles obtained based on the asynchronous residual sequence are directly determined as the target modal poles.
[0110] In some embodiments, during the rotation of the target rotating blade, gravity applies periodic excitation to the blade as the impeller azimuth angle changes. The generalized gravitational force corresponding to each target mode at different azimuth angles can be calculated. For example, one method for calculating the generalized gravitational force is: ;in, For the first The target mode in the azimuth angle The corresponding generalized gravitational force, subscript Represents gravity; The ground coordinate system gravity acceleration vector is a stack of feature reference points, with subscripts... Indicates stacking; for An identity matrix of order 1. This represents the total number of stacked feature reference points; For Kronecker product; Indicates the rotation matrix Diagonal dimension increase is performed based on each feature reference point to transform the gravity vector in the ground coordinate system to the blade rotating coordinate system.
[0111] Furthermore, both gravity excitation and aerodynamic loads introduced by wind shear tend to generate responses near the impeller rotational frequency 1P. The superposition of these two types of effects can interfere with the solution of modal parameters and equivalent modal stiffness. Therefore, under multiple wind speed, rotational speed, and pitch angle operating condition windows, the gravity term is defined as a known basis function that varies only with the impeller azimuth angle, and the aerodynamic load is characterized as an unidentified basis function that varies with the operating condition parameters. By using joint least squares, Bayesian estimation, or state-space parameter estimation, the two types of effects of gravity excitation and aerodynamic load can be separated, reducing the coupling interference caused by aerodynamic load.
[0112] Specifically, the stiffness calculation module 50 includes a pre-constraint unit, a stiffness index calculation unit, a working condition correction unit, a stiffness comparison unit, and a stiffness monitoring result determination unit.
[0113] The pre-constraint unit is used to execute step S51: determine whether the modal coordinate timing, modal confidence index and the configuration parameters of the target modal basis all satisfy the pre-constraint conditions.
[0114] Among them, the pre-constraint conditions are used to verify whether the current acquisition window meets the direct calculation assumptions of the equivalent modal stiffness index. The pre-constraint conditions may include, but are not limited to: the target modal response within the current acquisition window is approximately linear, the target mode is separable or the influence of modal coupling is negligible, the target modal basis can stably characterize the current vibration response of the blade, and the reference state and the current state use the same target modal basis, modal quality and normalization method.
[0115] In some embodiments, for the modal coordinate time series obtained under the current acquisition window, the magnitude of blade vibration deformation within the window is evaluated to determine whether the vibration response corresponding to the modal coordinate time series is within the small deformation range, and to verify whether the condition of approximate linearity of the target modal response is met; candidate modal poles of the modal recognition output are extracted, the discriminability between target modal poles of each order is evaluated, and it is determined whether the target mode is separable or the modal coupling effect is negligible. The modal confidence index is read and compared with a preset confidence threshold. Only when the modal confidence index is greater than or equal to the preset confidence threshold is the current modal recognition result considered reliable. The configuration parameters of the target modal basis are verified. The target modal basis currently in use is compared with the target modal basis stored in the reference state to verify that the target modal basis has not undergone parameter shift and can stably characterize the current blade vibration response; at the same time, the current state and the reference state are verified to confirm that they use completely consistent target modal basis, modal mass, and mode shape normalization processing rules. Only when all the above verification items pass simultaneously is it determined that the modal coordinate timing, modal confidence index, and configuration parameters of the target modal basis all meet the preconditions. If any one of the verification items fails, it is determined that the preconditions are not met, and the direct calculation process of equivalent modal stiffness will no longer be executed. Instead, the entire stiffness monitoring process can be repeated.
[0116] The stiffness index calculation unit is used to perform step S52: calculate the equivalent modal stiffness index based on the modal mass and the target modal frequency.
[0117] The equivalent modal stiffness index is a global dynamic characterization quantity, which is only used to make relative comparisons of stiffness states between the same target mode and the same type of operating conditions. It does not represent the local true structural stiffness at the blade material, cross section or finite element location.
[0118] In some embodiments, the modal mass corresponding to the j-th order target mode and the target modal frequency identified in the rotating coordinate system are read; if the target modal frequency obtained by solving the target modal poles is the damped oscillation angular frequency, the estimated value of the undamped operating modal angular frequency is obtained by combining the modal damping ratio output by synchronous identification, and then the equivalent modal stiffness index is calculated based on the estimated value of the undamped operating modal angular frequency.
[0119] For example, one way to calculate the equivalent modal stiffness index is as follows: ;in, For the first The equivalent modal stiffness index of the target mode; For the first Modal quality of the first-order target mode in the target mode basis; This is the estimated angular frequency of the undamped operating mode.
[0120] The operating condition correction unit is used to execute step S53: combining the operating condition data, using the centrifugal stiffening correction term and the aerodynamic stiffness correction term to correct the equivalent modal stiffness index, and obtain the operating condition corrected equivalent modal stiffness index.
[0121] In some embodiments, it is determined whether the aerodynamic stiffness correction term can be reliably obtained; when the absolute value of the aerodynamic stiffness correction term can be reliably obtained, the operating condition correction calculation is performed to obtain the operating condition correction equivalent modal stiffness index, and the operating condition correction equivalent modal stiffness index is selected to participate in the subsequent comparison; when the absolute value of the aerodynamic stiffness correction term cannot be reliably obtained, the equivalent modal stiffness index is directly selected to participate in the subsequent comparison.
[0122] In some embodiments, operating condition data is read to obtain information on rotational speed, wind speed, blade pitch angle, and impeller azimuth angle. The centrifugal stiffening correction term and aerodynamic stiffness correction term under the corresponding operating condition are retrieved. The aerodynamic stiffness correction term is integrally averaged along the complete impeller rotation cycle to obtain the aerodynamic correction term after the cycle is averaged. Then, the equivalent modal stiffness index is corrected according to the subtraction operation relationship to obtain the equivalent modal stiffness index with operating condition correction.
[0123] For example, one method for calculating the equivalent modal stiffness index under modified operating conditions is as follows: ;in, The equivalent modal stiffness index is corrected for the operating conditions; It is the equivalent modal stiffness index; Rotational speed The corresponding centrifugal stiffening correction term; Wind speed Pitch angle The aerodynamic correction term is averaged over the complete rotational period under operating conditions. It should be noted that the centrifugal stiffening correction term, the aerodynamic correction term after period averaging, and the equivalent modal stiffness index use the same set of target modal basis, normalization method, and dimensions.
[0124] The stiffness comparison unit is used to perform step S54: compare the equivalent modal stiffness index or the working condition modified equivalent modal stiffness index with the healthy stiffness reference index under the same working condition to obtain the equivalent modal stiffness change index.
[0125] In some embodiments, a healthy stiffness reference index is retrieved from the healthy operating condition database. This index matches the current rotational speed, wind speed, pitch angle, and temperature conditions and is pre-constructed using the same target modal basis and normalization method. The selected equivalent modal stiffness index or the operating condition-corrected equivalent modal stiffness index is compared with the healthy stiffness reference index to obtain the equivalent modal stiffness change index.
[0126] For example, one method for calculating the equivalent modal stiffness variation index can be: ;in, For the first The equivalent modal stiffness variation index of the target mode; For equivalent modal stiffness indices or working condition-corrected equivalent modal stiffness indices; As a health reference indicator.
[0127] The stiffness monitoring result determination unit is used to perform step S55: determine the target stiffness monitoring result based on the equivalent modal stiffness index, the working condition modified equivalent modal stiffness index, and / or the equivalent modal stiffness change index.
[0128] In some embodiments, if the equivalent modal stiffness index, the working condition-corrected equivalent modal stiffness index, and the equivalent modal stiffness change index are all effective, then the working condition-corrected equivalent modal stiffness index can be used as the core judgment criterion. The equivalent modal stiffness index and the equivalent modal stiffness change index can be combined for comprehensive judgment to finally determine the target stiffness monitoring result. Alternatively, if only equivalent modal stiffness indices and operating condition-corrected equivalent modal stiffness indices exist, but no equivalent modal stiffness change index is generated, the blade stiffness state can be jointly determined based on the equivalent modal stiffness indices and the corrected operating condition-corrected equivalent modal stiffness indices to determine the target stiffness monitoring result. Alternatively, if only equivalent modal stiffness indices and equivalent modal stiffness change indexes exist, but no effective operating condition-corrected equivalent modal stiffness indices exist, the target stiffness monitoring result can be determined based on the equivalent modal stiffness indices, combined with the equivalent modal stiffness change index that characterizes the degree of stiffness degradation. Alternatively, if only equivalent modal stiffness indices are generated, but no operating condition-corrected equivalent modal stiffness indices and equivalent modal stiffness change indexes exist, the blade stiffness state can be determined directly based on the equivalent modal stiffness indices, and the target stiffness monitoring result can be output.
[0129] In one embodiment, for a three-bladed wind turbine, the three blades have a fixed rotor azimuth angle deviation during rotation. Therefore, the rotor azimuth angle difference is first compensated for by timing the modal coordinates output by each of the three blades, mapping the vibration response of the three blades to the same phase reference. Within the same operating condition sub-bins defined by rotational speed, wind speed, pitch angle, and temperature, the equivalent modal stiffness indices of each of the three blades are calculated. Robust statistics of the equivalent modal stiffness indices of the three blades are selected, or the indices of two other blades confirmed to be in a healthy state are selected as common mode references. The equivalent modal stiffness indices of the single blade to be evaluated are compared with these common mode references to obtain the relative difference index of the single blade. The relative difference index of the single blade and the equivalent modal stiffness change index complement each other. The equivalent modal stiffness change index is the longitudinal evaluation result of the blade relative to its own historical health reference, while the relative difference index of the single blade is the lateral evaluation result among multiple blades in the same unit. At least one of the two can be included in the target stiffness monitoring results.
[0130] Specifically, the wind turbine rotating blade stiffness monitoring system also includes a quality verification module; the quality verification module is used to execute step S6: verify the quality of the target stiffness monitoring results, and determine whether to output based on the quality verification results. Specifically, it includes: S61: Reconstruct the first mode edge information matrix based on the robust weights of the convergence point of the joint objective function and the actual set of visible observations.
[0131] In some embodiments, the robust weights corresponding to the convergence points of the joint objective function are read, and the actual visible observation set obtained by online inversion is combined to complete the matrix reconstruction operation to obtain the first modal edge information matrix. The first modal edge information matrix is constructed based on real observation data in the online inversion stage, which is different from the second modal edge information matrix constructed by offline simulation in the array planning stage. The matrix incorporates the robust weights of real pixel observations and actual effective observation constraints, reflecting the degree of actual observation identifiability of modal coordinates under the current acquisition window.
[0132] S62: Perform a preliminary quality check on the target stiffness monitoring results based on the minimum eigenvalue of the first mode edge information matrix and the posterior covariance.
[0133] In some embodiments, the minimum eigenvalue calculated from the first modal edge information matrix is extracted, and the modal coordinate posterior covariance output from the joint objective function is read. These two are set as two mandatory indicators for preliminary quality verification. The minimum eigenvalue of the first modal edge information matrix is compared with a preset eigenvalue threshold, and the modal coordinate posterior covariance is compared with a preset covariance threshold. The preliminary quality verification is deemed qualified only when both mandatory indicators meet their respective preset threshold conditions. If the minimum eigenvalue is lower than the preset threshold, or the modal coordinate posterior covariance is greater than the preset covariance threshold, the preliminary quality verification is deemed unqualified, the target stiffness monitoring result is not output, and corresponding handling actions are triggered based on the unqualified indicators. The handling actions include at least one of the following: extending the sampling period, adjusting the image frame rate or exposure parameters, adjusting the material point selection area, reconfiguring the observation equipment array, and waiting for the wind turbine operating conditions to meet the conditions.
[0134] S63: If the initial quality verification is qualified, the stiffness monitoring results shall be verified by means of modal confidence index.
[0135] In some embodiments, an auxiliary quality verification is performed on the premise that the initial quality verification is qualified. The modal confidence index output by the modal recognition link is read and compared with a preset confidence threshold. If the modal confidence index is greater than or equal to the preset confidence threshold, the auxiliary quality verification is deemed qualified. If the modal confidence index is lower than the preset confidence threshold, the auxiliary quality verification is deemed unqualified. Similarly, no target stiffness monitoring result is output, and the corresponding handling action is triggered.
[0136] S64: Output stiffness monitoring results if the auxiliary quality verification is qualified.
[0137] In some embodiments, after both the preliminary quality verification and the auxiliary quality verification are deemed qualified, the target stiffness monitoring results are integrated and output. The output content may include, but is not limited to, at least one of the following: target mode number, mode normalization method, operating mode frequency in the rotating coordinate system, equivalent modal stiffness index, operating condition corrected equivalent modal stiffness index, equivalent modal stiffness change index, confidence interval, and quality mark. The quality mark is used to indicate that all quality verification items have passed.
[0138] In this embodiment, a specific example is provided below in conjunction with any of the above embodiments: A finite element method or parametric beam model of a horizontal axis wind turbine with a blade length of 63m was selected as the reference model. The rotational speed was set to 8-12 r / min. The first three target modes were extracted and uniformly normalized. No less than 30 material points were arranged along the blade. At the same time, triangular meshes of the blade, hub, and tower were constructed to realize projection and ray occlusion judgment. Three virtual UAVs were configured as observation equipment. The candidate positions were arranged in the wind-side safe area on the rotor plane, 80-120m away from the center of the hub. The horizontal or vertical baseline was 20-50m. The camera resolution was set to 3840×2160, the frame rate was 120 frames / s, and the exposure time was no more than 1 / 2000s. At no less than 180 rotor azimuth angle mesh points per revolution, a predicted visible observation set was generated based on positive depth, image plane boundary, the first intersection point of the ray from the camera to the material point and the structural mesh, the projection scale, and the incident angle to complete the selection of the observation array.
[0139] In the reference model, the elastic parameters corresponding to the target mode are scaled by 1.00, 0.98, 0.95, and 0.90 respectively. The blade response under different wind speed, rotational speed, and pitch angle conditions is obtained by aeroelasticity solver or multibody dynamics solver. Pixel coordinates are generated based on the camera projection model. Pixel noise of 0.1, 0.2, and 0.3 pixels, three different levels of pose prior covariance, 10%, 30%, and 50% random or continuous occlusion, and time synchronization error that meets the constraints are introduced respectively. Each set of working conditions is repeatedly calculated using a preset random seed.
[0140] Based on the obtained actual visible observation set, joint objective function inversion, phase grid resampling, and multi-period phase averaging are sequentially performed to complete modal identification. After verifying that the pre-constraint conditions are met, the equivalent modal stiffness index is calculated. Centrifugal stiffening correction terms and aerodynamic stiffness correction terms are introduced as needed to obtain the working condition corrected equivalent modal stiffness, and the equivalent modal stiffness change index is further calculated. The minimum eigenvalue of the first mode edge information matrix, the visible observation ratio, the root mean square error of modal coordinates, the relative error of the operating mode frequency, the error of the equivalent modal stiffness change index, the confidence interval coverage, and the false alarm rate are recorded synchronously throughout the acquisition process. The target stiffness monitoring results are quality verified, and the target stiffness monitoring results are output after the quality verification is deemed qualified.
[0141] Compared to existing technologies, this invention acquires multi-view image sequences, wind turbine operating condition data, and observation equipment pose observation data to determine the composite pose prior, pose covariance, and actual visible observation set representing the rigid body transformation of the observation equipment relative to the blade rotating coordinate system, as well as the pairing relationship between feature reference points and effective pixel observations. It then solves for the modal coordinate time series and posterior covariance of each target mode using a joint objective function. Based on the modal coordinate time series, it performs modal recognition to obtain the target modal frequency and modal confidence index. The modal coordinate time series, modal confidence index, and target modal basis configuration parameters are used as pre-constraints. Only when all conditions are met is the stiffness monitoring result output in conjunction with modal quality and target modal frequency. This effectively suppresses adverse interference from field factors such as coordinate transformation, observation noise, and modal coupling, quantifies the estimation uncertainty in the modal solution process, and sets multiple admission constraints to filter out unqualified calculation windows. This effectively improves the stability and reliability of the quantitative assessment results of blade stiffness under rotating conditions and reduces the risk of misjudging the health status of the blade structure.
[0142] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.
[0143] Based on the same inventive concept, this invention provides a computer device, which can be a terminal, providing voice and / or data connectivity to a user. Exemplarily, the terminal can be an Internet of Things (IoT) terminal, such as a sensor device, a mobile phone, and a computer with an IoT terminal; for example, it can be a fixed, portable, pocket-sized, handheld, or computer-embedded device. Alternatively, the terminal can also be a device from an unmanned aerial vehicle (UAV). Its internal structure diagram can be as follows... Figure 4 As shown. The computer device includes a processor, memory, communication interface, display screen, and input device connected via a method bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores operating methods and computer programs. The internal memory provides an environment for the operation of the operating methods and computer programs in the non-volatile storage medium. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements the wind turbine blade stiffness monitoring method mentioned in any of the above embodiments. The display screen can be a liquid crystal display (LCD) or an e-ink display. The input device can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the computer device casing, or an external keyboard, touchpad, or mouse.
[0144] Based on the same inventive concept, the present invention also provides a computer-readable storage medium corresponding to the aforementioned embodiments of the wind turbine rotating blade stiffness monitoring method. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements the steps of the wind turbine rotating blade stiffness monitoring method described in any of the above embodiments.
[0145] This invention can take the form of a computer program product implemented on one or more storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing program code. Computer-usable storage media include permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to: phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device.
[0146] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0147] The embodiments described above are merely examples of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and the present invention also intends to include these modifications and variations.
Claims
1. A method of monitoring the stiffness of a fan rotating blade, characterized by, include: Acquire a sequence of images of the target rotating blades from different perspectives, wind turbine operating data, and pose observation data of the observation equipment; Based on the image sequence, the operating condition data, and the pose observation data, the composite pose prior, pose covariance, and actual visible observation set are determined respectively; the composite pose prior represents the rigid body transformation from the blade rotation coordinate system corresponding to the target rotating blade to the camera coordinate system corresponding to the observation device; the actual visible observation set represents the pairing relationship between the feature reference points on the target rotating blade and the corresponding effective pixel observations in the image sequence; Based on the composite pose prior, the pose covariance, and the actual visible observation set, determine the modal coordinate temporal sequence and posterior covariance corresponding to each target mode. Modal identification is performed based on the modal coordinate time sequence to obtain the target modal frequency and modal confidence index of the target rotating blade. If the modal coordinate timing, the modal confidence index, and the configuration parameters of the target modal basis all meet the pre-constraint conditions, the target stiffness monitoring result is determined based on the modal quality corresponding to the target mode and the target modal frequency.
2. The method of claim 1, wherein, The step of determining the composite pose prior, pose covariance, and actual visible observation set based on the image sequence, the operating condition data, and the pose observation data includes: Image features are extracted from the image sequence, and relative pose data are calculated based on the image features; Based on the pose observation data and the relative pose data, and combined with a multi-source fusion algorithm, the first pose transformation relationship is obtained; the first pose transformation relationship represents the rigid body transformation between the camera coordinate system and the ground coordinate system. Based on the impeller azimuth angle in the operating condition data and the pre-calibrated hub geometry parameters, a second pose transformation relationship is obtained; the second pose transformation relationship characterizes the rigid body transformation between the ground coordinate system and the blade rotation coordinate system. Based on the first pose transformation relationship and the second pose transformation relationship, determine the composite pose prior and the pose covariance; Pair the pixel observations of the same feature reference point in the image sequence to obtain a candidate pairing set; Based on the actual working conditions of the image, valid observation pairs are selected from the candidate pairing set to obtain the actual visible observation set.
3. The method according to claim 1, characterized in that, The step of performing modal identification based on the modal coordinate time series to obtain the target modal frequency and modal confidence index of the target rotating blade includes: Based on the impeller azimuth angle in the operating condition data, phase grid resampling processing is performed on the modal coordinate time series, and the blade rotation synchronization component is obtained after multi-cycle phase averaging. The difference between the modal coordinate timing sequence and the blade rotation synchronization component is calculated to obtain the asynchronous residual sequence. Preliminary mode identification is performed based on the asynchronous residual sequence to obtain candidate modal poles and corresponding candidate modal frequencies; The frequency interval is determined based on the candidate modal frequencies and the impeller rotor frequency multiples. The coherence coefficient is determined based on the modal coordinate timing sequence and the corresponding rotor order harmonic reference timing sequence; the rotor order harmonic reference timing sequence is determined based on the impeller azimuth angle. The frequency interval is compared with a preset bandwidth, and / or the coherence coefficient is compared with a preset threshold to obtain a comparison result; Based on the comparison results, the target modal frequency and the modal confidence index are determined.
4. The method according to claim 3, characterized in that, Determining the target modal frequency and the modal confidence index based on the comparison results includes: When the comparison result is that the frequency interval is less than or equal to the preset bandwidth or the coherence coefficient is greater than the preset threshold, the harmonic basis function and modal pole joint estimation algorithm is adopted to jointly decouple and identify the order forcing term and the structural modal pole based on the modal coordinate time series, so as to obtain the target modal pole and the corresponding target modal frequency and modal confidence index. When the comparison result is that the frequency interval is greater than the preset bandwidth or the coherence coefficient is less than or equal to the preset threshold, the candidate modal pole is determined as the target modal pole and the corresponding target modal frequency and modal confidence index are obtained.
5. The method according to claim 1, characterized in that, The step of determining the target stiffness monitoring result based on the modal quality corresponding to the target mode and the target modal frequency includes: Calculate the equivalent modal stiffness index based on the modal mass and the target modal frequency; Based on the aforementioned operating condition data, the equivalent modal stiffness index is corrected using centrifugal stiffening correction term and aerodynamic stiffness correction term to obtain the operating condition corrected equivalent modal stiffness index. The equivalent modal stiffness index or the working condition modified equivalent modal stiffness index is compared with the healthy stiffness reference index under the same working condition to obtain the equivalent modal stiffness change index. The target stiffness monitoring result is determined based on the equivalent modal stiffness index, the working condition-corrected equivalent modal stiffness index, and / or the equivalent modal stiffness change index.
6. The method according to claim 1, characterized in that, The step of determining the temporal coordinates and posterior covariance of each target mode based on the composite pose prior, the pose covariance, and the actual visible observation set includes: Based on the composite pose prior, the pose covariance, and the actual visible observation set, the modal coordinate temporal series and the posterior covariance are calculated using a joint objective function; wherein, the expression of the joint objective function is: ; in, For robust loss function; For pixel observation covariance; For pose covariance; It is the actual set of visible observations; For pose prior weight coefficients; For time smoothing or modal dynamics operators; For time-series constraint weights; This represents the frame-by-frame pose correction amount corresponding to time t; For the first Modal coordinates of the target mode.
7. The method according to claim 1, characterized in that, The method further includes: The first modality edge information matrix is obtained by reconstructing the edge information matrix based on the robust weights of the convergence points of the joint objective function and the actual visible observation set. Based on the minimum eigenvalue of the first modal edge information matrix and the posterior covariance, a preliminary quality check is performed on the target stiffness monitoring results; If the initial quality verification is satisfactory, the stiffness monitoring results are further verified based on the modal confidence index. If the auxiliary quality verification is qualified, the stiffness monitoring results are output.
8. The method according to claim 1, characterized in that, The acquisition of a sequence of images of the target rotating blade from different viewpoints includes: When multiple observation devices are deployed in a target observation array, the image sequence is acquired; wherein, the design method of the target observation array includes the following steps: The reference coordinates of the characteristic reference points of the reference blade, the target mode basis and modal mass corresponding to the target mode are obtained; the target mode includes at least one of flapping, oscillation and torsion; the target mode basis represents the displacement distribution of each order of target mode at each characteristic reference point; For each candidate observation array, geometric visibility verification and imaging quality verification are performed on an azimuth grid covering at least one complete rotation cycle, based on the reference coordinates, to filter and obtain a set of predicted visible observations at each azimuth angle; the candidate observation array is used to characterize the spatial position and orientation relationship between each observation device; Based on the predicted visible observation set and combined with the target modality basis, a second modality edge information matrix is constructed; Calculate the minimum eigenvalue of the second-mode edge information matrix corresponding to each candidate observation array; The candidate observation array with the largest minimum eigenvalue is selected as the target observation array.
9. The method according to claim 8, characterized in that, The construction of a second modality edge information matrix based on the predicted visible observation set and the target modality basis includes: For each azimuth angle, the Jacobian matrix is calculated for the predicted visible observation set; the Jacobian matrix includes the modal Jacobian and the pose Jacobian. The Jacobian matrix is weighted and encapsulated, and the pose prior information terms corresponding to the pose prior covariance are superimposed to obtain the joint information submatrix. The joint information submatrices corresponding to all azimuth angles within the complete rotation cycle are weighted and accumulated to obtain a comprehensive joint information matrix covering the entire rotation process; The pose blocks of the integrated joint information matrix are subjected to Schuler complement elimination to obtain the second modality edge information matrix.
10. A wind turbine rotating blade stiffness monitoring system, characterized in that, include: The acquisition module is used to acquire a set of image sequences of the target rotating blade from different perspectives, wind turbine operating condition data, and pose observation data of the observation equipment; The pose and observation set determination module is used to determine the composite pose prior, pose covariance, and actual visible observation set based on the image sequence, the operating condition data, and the pose observation data, respectively. The composite pose prior represents the rigid body transformation from the blade rotation coordinate system corresponding to the target rotating blade to the camera coordinate system corresponding to the observation device. The actual visible observation set represents the pairing relationship between the feature reference points on the target rotating blade and the corresponding effective pixel observations in the image sequence. The joint target calculation module is used to determine the modal coordinate temporal sequence and posterior covariance corresponding to each target mode based on the composite pose prior, the pose covariance and the actual visible observation set; The modal recognition module is used to perform modal recognition based on the modal coordinate time sequence to obtain the target modal frequency and modal confidence index of the target rotating blade. The stiffness calculation module is used to determine the target stiffness monitoring result based on the modal quality and the target modal frequency, provided that the modal coordinate time series, the modal confidence index, and the configuration parameters of the target modal basis all meet the pre-constraint conditions.
