A method, device and equipment for optimizing the arrangement of sensors on wind turbine blades

Abstract

The present invention provides a method, device and equipment for optimizing the arrangement of sensors on wind turbine blades, which belongs to the field of sensor arrangement. The method includes the following steps: establishing a finite element model of wind turbine blades, using the K-means classification method to divide the area to be measured in the finite element model into multiple partitions; using the generalized modal kinetic energy method to assign weights to each partition, and determining the number of sensors in each partition according to the weight assignment result; defining the modal matrix of the finite element model as a Fisher information matrix, using the trace and determinant values ​​of the Fisher information matrix as a quantization standard, using the effective independence method to solve the eigenvalues ​​of the Fisher information matrix of each partition, and obtaining the eigenvalue contribution matrix; determining the sensor arrangement position of each partition through the eigenvalue contribution matrix. The present invention can effectively reduce the problem of sensor concentration, thereby reducing the information redundancy caused by the concentrated arrangement of sensors as much as possible.

Description

Technical Field

[0001] The present invention belongs to the field of sensor arrangement, and in particular relates to a method, device and equipment for optimizing the arrangement of sensors on wind turbine blades. Background Art

[0002] The arrangement of sensors directly affects the signal components and quality, which is of great significance to structural health monitoring. How to collect the maximum real response information of the structure with a limited number of sensors is becoming a prominent issue in this field.

[0003] The existing sensor arrangement method based on modal recognition does not consider the model structure, but only maximizes the modal matrix recognition ability and energy coefficient when the sensor is arranged. This will lead to the concentration of sensors in local areas, especially when facing large models and highly flexible structures. This problem is more likely to occur. This problem is particularly obvious on wind turbine blades. For large-area and ultra-long wind turbine blades, the over-concentration of sensors often has the following disadvantages: First, when the sensors are concentrated in one area, the local reliability deteriorates. If the area fails, the entire system will fail. Second, the centralized arrangement of sensors will increase the complexity of space and wiring, which is not conducive to later maintenance and replacement. Third, it is impossible to monitor the blades in different areas. Fourth, due to the different stress and strain states of different positions of the blade, the centralized arrangement cannot detect the slight changes in each section of the blade. Fifth, the weight distribution is uneven. The centralized arrangement will cause the weight of the sensor area to be too large, affecting the weight balance of the blade. Sixth, data redundancy and correlation. The sensors densely arranged on the blades may cause the data measured by some sensors to be highly correlated, increase the complexity of data processing and analysis, and may mislead the monitoring results.

[0004] In summary, for large-area and extra-long wind turbine blades, the existing sensor layout method will cause the sensors to be arranged too concentratedly, affecting the reliability of the monitoring results, and thus failing to detect blade structural damage in a timely and accurate manner. Summary of the invention

[0005] In order to overcome the above-mentioned shortcomings of the prior art, the present invention provides a method for optimizing the arrangement of wind turbine blade sensors, comprising the following steps:

[0006] Establish a finite element model of wind turbine blades, and use K-means classification method to divide the area to be tested in the finite element model into multiple partitions;

[0007] Using the generalized modal kinetic energy method to assign weights to each of the partitions, and determining the number of sensors required for each partition according to the weight assignment result;

[0008] The modal matrix is ​​defined as the Fisher information matrix. The trace and determinant of the Fisher information matrix are used as the quantization standard. The eigenvalues ​​of the Fisher information matrix of each partition are solved using the effective independence method to obtain the eigenvalue contribution matrix.

[0009] The arrangement position of the sensor in each partition is determined by the eigenvalue contribution matrix.

[0010] Preferably, the K-means classification method is used to divide the area to be measured in the finite element model into multiple partitions, including the following steps:

[0011] Obtain the three-dimensional point matrix D of the finite element model of the wind turbine blade;

[0012] The three-dimensional dot matrix D of the finite element model of the wind turbine blade is input into K-means, and k center points are randomly generated;

[0013] Calculate the distance between the three-dimensional dot matrix D and the center point, determine the partition of the three-dimensional dot matrix D according to the center point closest to the three-dimensional dot matrix D, add the three-dimensional dot matrix D to the corresponding partition, calculate the centroid of each partition, and use K-means to iterate the center point to obtain multiple partitions.

[0014] Preferably, the generalized modal kinetic energy method is used to assign weights to each partition, comprising the following steps:

[0015] The contribution of the modal kinetic energy of the i nodes corresponding to the mode of the r-th partition is arithmetic averaged to obtain the average modal kinetic energy MKE(r) of the r-th partition;

[0016] A weight score S(r) of the rth partition is calculated based on the average modal kinetic energy MKE(r) of the rth partition, and the weight of the rth partition is determined by the weight score S(r).

[0017] Preferably, the calculation formula of the average modal kinetic energy MKE(r) is:

[0018]

[0019] In the formula, MKE ij It represents the response degree of the j-order mode corresponding to the i-th node.

[0020] Preferably, the calculation formula of the weight score S(r) is as follows:

[0021]

[0022] Where MKE(r) is the average modal kinetic energy, m is the node that the sensor can select, and n is the modal order of the modal matrix of the finite element model.

[0023] Preferably, the calculation formula for the total number of sensors is:

[0024] N i =S(r)×t

[0025] Where t is the total number of sensors in the area to be measured, and S(r) is the rescore S(r) of each partition.

[0026] Preferably, the Fisher information matrix is:

[0027] R=Φ T Φ

[0028] Where Φ is the modal matrix of the finite element model.

[0029] Preferably, the calculation formula of the characteristic value is:

[0030]

[0031] Where Φ is the modal matrix of the finite element model, A im ΦΨ i The value of the mth element in the column vector, For the jth node, i , Ψ represents the eigenvector of the Fisher information matrix, and m is the node that the sensor can select.

[0032] The present invention also provides a sensor position optimization arrangement device, comprising:

[0033] The model segmentation module is used to establish a finite element model of a wind turbine blade and to segment the area to be tested in the finite element model into multiple partitions using the K-means classification method;

[0034] A weight allocation module, used for using the generalized modal kinetic energy method to allocate weights to each of the partitions, and determining the number of sensors required for each partition according to the weight allocation result;

[0035] The eigenvalue calculation module is used to define the modal matrix as the Fisher information matrix, use the trace and determinant of the Fisher information matrix as the quantization standard, and use the effective independence method to solve the eigenvalue of the Fisher information matrix of each partition to obtain the eigenvalue contribution matrix;

[0036] The sensor arrangement position module is used to determine the arrangement position of the sensor in each partition through the eigenvalue contribution matrix.

[0037] The present invention also provides a computer device, comprising a memory and a processor; the memory stores a computer program, and the processor is used to run the computer program in the memory to execute the wind turbine blade sensor arrangement optimization method.

[0038] The wind turbine blade sensor arrangement optimization method, device and equipment provided by the present invention have the following beneficial effects:

[0039] The present invention can cut the finite element model of a wind turbine blade into multiple partitions by using the K-means algorithm, and can effectively reduce the influence of grid division on sensor allocation by using the generalized modal kinetic energy method to distribute weights to the partitions; and then use the effective independence method in the partitioned small area to effectively reduce the problem of sensor concentration generated in the large model structure, thereby reducing the information redundancy caused by the concentrated arrangement of sensors as much as possible, so that the wind turbine blades can be accurately detected and the damage of the wind turbine blades can be discovered in time. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the embodiment of the present invention and its design scheme, the following briefly introduces the drawings required for this embodiment. The drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0041] Figure 1 A flow chart of a method for optimizing the arrangement of sensors on wind turbine blades according to an embodiment of the present invention;

[0042] Figure 2 It is the sensor arrangement scheme obtained based on the effective independence method;

[0043] Figure 3 The sensor arrangement scheme is obtained based on the energy coefficient-effective independence method;

[0044] Figure 4 The sensor placement scheme is derived based on the partition weight-effective independence method;

[0045] Figure 5 is the MAC matrix corresponding to the effective independence method;

[0046] Figure 6 is the MAC matrix corresponding to the energy coefficient-effective independence method;

[0047] Figure 7 is the MAC matrix corresponding to the partition weight-effective independence method. DETAILED DESCRIPTION

[0048] In order to enable those skilled in the art to better understand the technical solution of the present invention and implement it, the present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the scope of protection of the present invention.

[0049] In the description of the present invention, it is to be understood that the terms “center”, “longitudinal”, “lateral”, “length”, “width”, “thickness”, “up”, “down”, “front”, “back”, “left”, “right”, “vertical”, “horizontal”, “top”, “bottom”, “inside”, “outside”, “axial”, “radial”, “circumferential”, etc., indicating orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, are only for the convenience of describing the technical solutions of the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention.

[0050] In addition, the terms "first", "second", etc. are used for descriptive purposes only and are not to be understood as indicating or implying relative importance. In the description of the present invention, it should be noted that, unless otherwise clearly specified or limited, the terms "connected" and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to the specific circumstances. In the description of the present invention, unless otherwise specified, "plurality" means two or more, which will not be described in detail here.

[0051] Example 1

[0052] The present invention provides a method for optimizing the arrangement of wind turbine blade sensors, specifically: Figure 1 As shown, the following steps are included:

[0053] Step 1: Establish a finite element model of the wind turbine blade, and use the K-means classification method to divide the area to be tested in the finite element model into multiple partitions according to the geometric spatial distribution of the area to be tested.

[0054] K-means classification is a commonly used clustering analysis algorithm. The advantages of this method are simple calculation and fast operation speed. It is suitable for processing large data sets, especially for the regional division of wind turbine blades with complex structures. K-means classification is sensitive to the selection of the initial center point, which may lead to a local optimal solution. However, in large-scale, multi-node models, the sensitivity to the center point is greatly reduced because the nodes are evenly arranged and there is no obvious irregularity.

[0055] Use K-means classification method to divide the area to be tested in the finite element model into multiple partitions. The specific process is as follows:

[0056] Obtain a three-dimensional point matrix D of a finite element model of a wind turbine blade;

[0057] The three-dimensional dot matrix D of the finite element model of the wind turbine blade is input into K-means, and k center points are randomly generated;

[0058] Calculate the distance between the three-dimensional dot matrix D and the center point, determine the partition of the three-dimensional dot matrix D according to the center point closest to the three-dimensional dot matrix D, add the three-dimensional dot matrix D to the corresponding partition, calculate the centroid of each partition, and use K-means to iterate the center point to obtain multiple partitions.

[0059] Step 2: Use the generalized modal kinetic energy method to assign weights to each partition, and determine the number of sensors in each partition based on the weight assignment results.

[0060] The generalized modal kinetic energy (MKE) selects the sensor location with the goal of maximizing the measured modal kinetic energy. This method can effectively improve the signal-to-noise ratio of the structural dynamic response measurement. The calculation formula is as follows:

[0061] MKE=Φ T MΦ

[0062] In the formula, Φ represents the modal matrix of the structure, and M is the mass matrix of the structure. After ranking the contribution of each node according to the generalized modal kinetic energy matrix, the first n nodes with the largest contribution are selected for sensor arrangement.

[0063] The modal kinetic energy criterion effectively solves the problem of placing sensors at locations with large structural vibrations, which used to rely on the subjective judgment of engineers. It is a more scientific quantitative method in sensor placement theory. The basic idea is that at locations with large modal kinetic energy, the vibration data is also large. Placing sensors at this location is conducive to improving the signal-to-noise ratio of the test signal and the accuracy of modal parameter identification.

[0064] However, this method also has obvious defects: once the quality of mesh division is poor, the accuracy of modal parameters will be significantly reduced, which will have a negative impact on the final effect of sensor position optimization.

[0065] Therefore, in the present invention, after using the K-means classification method to divide the area to be measured in the finite element model into multiple partitions, the generalized modal kinetic energy method is used to assign weights to the partitions, thereby reducing the impact of grid division on sensor allocation. The specific weight allocation is based on the following formula. First, the contribution of the mode of the rth region to the modal kinetic energy of the i-th degree of freedom is arithmetic averaged to obtain the average modal kinetic energy MKE(r) of the region:

[0066]

[0067] In the formula, MKE ijIt represents the response degree of the j-order mode corresponding to the i-th node.

[0068] The average modal kinetic energy can directly measure the contribution of each degree of freedom to the overall vibration response data of the structure. Then, the weight score S(r) of the rth region can be calculated by the following formula. Due to the modal concentration effect, the obtained score will have a difference in magnitude, thus biased towards the modal sensitive area to cause the sensor concentration effect. Logarithmic transformation can reduce this impact.

[0069]

[0070] Next, the number of regional sensors N is determined according to the average modal kinetic energy i =S(r)×t, where t is the total number of sensors in the area to be measured.

[0071] Step 3: Define the modal matrix of the finite element model as the Fisher information matrix. Use the trace and determinant of the Fisher information matrix as the quantification standard. Use the effective independence method to solve the eigenvalues ​​of the Fisher information matrix of each partition to obtain the eigenvalue contribution matrix.

[0072] The modal matrix of a finite element model is a matrix consisting of all the modal vectors (also called eigenvectors or eigenvectors) of the system. In dynamics and vibration analysis, these modal vectors represent the shape or mode of the system when it vibrates at different natural frequencies. Specifically, in the modal matrix, each column represents a different modal order, and each row represents each node.

[0073] In the field of sensor placement, various iterative algorithms and optimization methods emerge in an endless stream. Among them, a classic sensor placement method is the effective independence method (EI) proposed by Kammer. The effective independence method is a mature iterative algorithm. Each iteration will delete the node with the worst contribution. For example, there are 200 nodes initially, and one will be deleted in each iteration, leaving a certain number of nodes. In the present invention, the number of nodes is determined by the generalized modal kinetic energy method (MKE). The purpose of the effective independence method is to maximize the Fisher information matrix. Each iteration will delete a node with the smallest contribution to the Fisher information matrix (represented as the corresponding row in the matrix).

[0074] The effective independence method is one of the most efficient and widely used methods in the field of sensor optimization. Its basic principle is to arrange sensors as much as possible in positions that can make the modal space linearly independent, sort the contribution of each sampling position to the independence of the target modal matrix, eliminate the points to be tested with small linear contribution in the modal space, and achieve modal space completeness and information maximization under the condition of a limited number of sensors. The key step is to calculate the effective independent components by solving the characteristic equation of the equivalent matrix of the Fisher information matrix.

[0075] According to the modal superposition principle, the sensor output can be expressed as:

[0076]

[0077] Among them, u=[u1,u2,…,u m ] T is the displacement column vector output by the sensor, Φ=[φ ij ] m×n is the modal matrix of the finite element model, where n is the modal order, m is the degree of freedom (in the field of sensor placement, m is the selectable node for placing the sensor), q = [q1, q2, …, q m ] T is the modal coordinate, i.e., the combined participation coefficient of each vibration mode.

[0078] According to dynamic control theory, the target modes of the structural system must have excellent observability while ensuring linear independence. In order to ensure good correlation between model testing and analysis in structural dynamic testing, the structural modes fed back by the candidate sensor nodes must be linearly independent of each other. Otherwise, it will be difficult to distinguish between the test modes and the modes of the corresponding finite element model (the test modes are the modes obtained by experimental testing. The test modes are measured by different methods on the entity (such as the hammer method), while the modes of the finite element model are theoretical modes obtained by building a model on a computer and then calculating). This will cause the modes fed back by the sensor points to be confused during the identification process, reducing the efficiency of collecting effective structural information. In the above formula, only the modal coordinate q is the quantity that needs to be estimated. Therefore, as long as the optimal estimate of the modal coordinate q is obtained, the linear independence of the measured modes can be ensured, and the best estimate of the mode can be obtained. Substitute the estimated value of the modal coordinate q Defined as:

[0079]

[0080] In order to ensure that the modes measured by the sensor points have the highest possible linear independence, the validity of the modal coordinate estimates must be optimized, and the optimal estimate usually means that the covariance matrix of the estimation error is minimized. In order to obtain an effective unbiased estimate, the estimated error covariance matrix of the modal coordinate q is expressed as follows:

[0081]

[0082] Where E is the expected function. For the convenience of calculation, it is assumed that the Gaussian white noise in the above formula has no correlation and has the same statistical characteristics, that is, the covariance vector is a constant. Define R = Φ T Φ, because R and Q differ only by multiples For the convenience of calculation and discussion, the present invention refers to the matrix R as the Fisher information matrix (FIM).

[0083] Through the derivation of the above formula, the effective independence method transforms the sensor position optimization problem into the problem of maximizing the Fisher information matrix. Maximizing the Fisher information matrix can minimize the estimated error covariance matrix of the modal coordinate q to provide the best estimate of q. This optimal estimate ensures the linear independence between the modes fed back by the candidate nodes of the sensor, ensuring the effect in identification. In order to maximize the Fisher information matrix, Kammer chose the trace and determinant of the Fisher information matrix as the quantification standard. Changes in the corresponding degrees of freedom of the candidate nodes of the structural system will affect the maximum quantification of the Fisher information matrix. In order to remove the redundancy of modal information and retain effective information, it is necessary to quantify the contribution of the candidate nodes to the Fisher information matrix. To this end, it is necessary to solve the eigenvalue problem of the Fisher information matrix:

[0084] [FIM-λI]Ψ=0

[0085] Among them, λ represents the eigenvalue of the Fisher information matrix, Ψ represents the corresponding eigenvector, and I is the unit matrix. Since FIM is a positive definite matrix, the following relationship can be obtained:

[0086] Ψ T FIMΨ=λ

[0087] Ψ T Ψ=I

[0088] The following expression can be obtained through matrix transformation:

[0089]

[0090] In the above formula, A im ΦΨ iThe value of the mth element in the column vector. From the above formula, we can see that when the selected mode number m is determined, For the jth node, i 's contribution, thus obtaining the eigenvalue contribution matrix G:

[0091]

[0092] Among them, the symbol It means that the two matrices perform point-to-point multiplication of corresponding elements. The sum of each column of G is equal to the eigenvalue corresponding to FIM. Therefore, the i-th sensor position in the column corresponds to its contribution to the eigenvalue. The contribution matrix of different degrees of freedom to the eigenvalue of the Fisher information matrix can be calculated according to the following formula, so that all directions have the same importance in the absolute recognition space:

[0093] F E =Gλ -1

[0094] F E Adding up the rows, we get:

[0095]

[0096] Among them, F Eij Indicates F E The i-th row and j-th column in the column vector E D is an effectively independent distribution of the sensor set.

[0097] The above introduces the solution principle of the EI method, and now an example is given as follows:

[0098] (1) This modal matrix gives a 3-DOF, first 2-order modal vibration shape, through FIM = Φ T Φ can be obtained

[0099] (2) Calculate the eigenvector of the Fisher information matrix Eigenvalue

[0100] (3) have to

[0101] (4)F E =Gλ -1 ,calculate Add each row to get

[0102] (5) In E DThe minimum value is 0.46 in the first row, indicating that the first node has the lowest contribution to the entire mode. Deleting the first row in the Fisher information matrix yields Then the second iteration is performed.

[0103] Through the above E D The contribution of each degree of freedom can be seen. The node with the smallest contribution is deleted in each iteration until the number of remaining nodes is the same as the predetermined number. Since the entire Fisher information matrix changes after deleting the node, it is impossible to obtain all sensor positions in one calculation.

[0104] In the present invention, the effective independence method is used to optimize the sensor layout in each partition: the input nodes are all nodes in the partition, and the output number of sensor positions is the score obtained in the previous step. To ensure the uniformity of the modal analysis of each partition, the modal matrix still uses the results of the overall modal analysis.

[0105] The algorithm of the present invention is as follows:

[0106]

[0107]

[0108] In the above algorithm, clusters represent partitions, and each row of Ed corresponds to the corresponding node number.

[0109] The present invention also provides a sensor position optimization arrangement device, comprising:

[0110] The model segmentation module is used to establish a finite element model of a wind turbine blade and to segment the area to be tested in the finite element model into multiple partitions using the K-means classification method;

[0111] A weight allocation module, used for using the generalized modal kinetic energy method to allocate weights to each of the partitions, and determining the number of sensors required for each partition according to the weight allocation result;

[0112] The eigenvalue calculation module is used to define the modal matrix as the Fisher information matrix, use the trace and determinant of the Fisher information matrix as the quantization standard, and use the effective independence method to solve the eigenvalue of the Fisher information matrix of each partition to obtain the eigenvalue contribution matrix;

[0113] The sensor arrangement position module is used to determine the arrangement position of the sensor in each partition through the eigenvalue contribution matrix.

[0114] The present invention also provides a computer device, comprising a memory and a processor; the memory stores a computer program, and the processor is used to run the computer program in the memory to execute a method for optimizing the arrangement of wind turbine blade sensors.

[0115] The present invention also provides a computer-readable storage medium, which stores a computer program. The computer program is suitable for being loaded by a processor to execute a method for optimizing the arrangement of sensors on wind turbine blades.

[0116] Example 2

[0117] In this embodiment, the wind turbine blade structure is modeled by ANSYS software and modal analysis is performed. The blade parameters are: density 1550kg / m 3 , Young's modulus 42.97Gpa, Poisson's ratio 0.35, shear modulus 3.78Gpa, maximum length, width and height are 86.4m, 6.2m and 3.97m, and the cross-sectional area gradually widens and then narrows as the length of the blade increases. The model uses eight-node three-dimensional block SOLID45 elements, with a total of 5082 nodes and 5241 elements. Among them, the candidate nodes are 5061 nodes on the blade surface, and the first six orders are selected. Considering the construction of matrices in the effective independence method and other methods, the modal observability should be met, that is, the number of sensors finally selected must be greater than the selected modal order.

[0118] Finite element modal analysis is used to calculate the natural frequency and vibration mode of the wind turbine blade. In order to verify the feasibility of this method, the first 6 modes are selected for simulation calculation. The calculation results are shown in Table 1.

[0119] Table 1 Simulation results of the first 6 modal modes

[0120]

[0121] Choice of mode shapes and number of sensors

[0122] Research shows that the number of sensors required for structural modal identification should not be less than the number of selected modes. In order to accurately identify the various modes of wind turbine blades without loss of generality, the present invention sets the number of sensors to be arranged to 100. Too few sensors will lead to reduced accuracy in modal identification, while too many sensors will lead to excessive calculations, and the positions of the arranged sensors will inevitably be concentrated. In this example, 100 sensors are arranged in 5082 nodes.

[0123] The effective independence method, energy coefficient-effective independence method and partition weight-effective independence method proposed in this invention are used to optimize the sensor layout. The obtained sensor layout schemes are shown in Figure 2 to Figure 4 .from Figure 2It can be clearly seen that this method can fully consider the importance of different areas and make the node distribution more in line with actual needs. Compared with the traditional method, this solution not only arranges enough sensors in the strong node area, but also maintains a certain distance between nodes, avoiding the information redundancy problem caused by node concentration while effectively reducing the mutual interference between nodes, improving measurement accuracy and reliability. Such node distribution is also conducive to modal expansion, so as to provide more accurate modal parameter identification results. Figure 3 and Figure 4 It can be seen that the layout schemes obtained by the effective independence method and the energy coefficient-effective independence method have very serious information redundancy problems and are overly concentrated in certain strong node areas, resulting in more serious information redundancy. This is very unfavorable for mode expansion and subsequent modal parameter identification.

[0124] Results Analysis

[0125] After optimizing the placement of sensors, it is necessary to develop evaluation criteria. Different usage scenarios affect the formulation of evaluation criteria. A good evaluation criterion can clearly show the advantages and disadvantages of the placement location.

[0126] (1) Farthest Neighbor Distance Criterion

[0127] Furthest Nearest Distance (FND) is a method used to evaluate the uniformity of spatial point distribution. It calculates the distance between each point and its farthest neighbor, and based on the distribution of these distances, it can effectively determine the concentration of points in space. The formula is as follows:

[0128]

[0129]

[0130] In the formula, d represents the Euclidean distance between points, N is the total number of distributed points, x, y, and z are the coordinates of a point in three-dimensional space, and the Euclidean distance between each point in the space is calculated and the largest set of relative distances is found. The average value is calculated to get the F coefficient. If the F coefficient is too large, it means that the distribution of points in the space is relatively concentrated, otherwise it means that the distribution is uniform. The F coefficient can be used very effectively to evaluate the density of sensor layout. Specifically, if the distance between a point in the space and its nearest neighbor is small, it means that these points are relatively concentrated because the distance between them is short.

[0131] The values ​​of the farthest neighbor distance criteria of the three methods are 33.27, 32.17 and 34.28 respectively. It can be seen from the farthest neighbor distance criterion that the node distribution arranged by the partition weight-effective independence method is more dispersed, avoiding information redundancy.

[0132] The node layout scheme is designed according to the partition weight-effective independence method proposed in the present invention, and the resulting spatial distribution is more reasonable. Figure 2 It can be clearly seen that this method can fully consider the importance of different areas and make the node distribution more in line with actual needs. Compared with the traditional method, this solution not only arranges enough sensors in the strong node area (strong nodes are nodes that reflect different order modes relatively large), but also maintains a certain distance between nodes, avoiding the information redundancy problem caused by node concentration while effectively reducing mutual interference between nodes, improving measurement accuracy and reliability. Such node distribution is also conducive to modal expansion, so as to provide more accurate modal parameter identification results. Figure 3 and Figure 4 It can be seen that the layout schemes obtained by the effective independence method and the energy coefficient-effective independence method have very serious information redundancy problems and are overly concentrated in certain strong node areas, resulting in more serious information redundancy. This is very unfavorable for mode expansion and subsequent modal parameter identification.

[0133] (2) Fisher Information Matrix Criterion

[0134] The Fisher information matrix can characterize the amount of modal information contained in the test response, and it has different measurement indicators, including determinant, F-norm, trace and minimum singular value. The larger the determinant and trace of the Fisher information matrix, the better; increasing the minimum singular value is conducive to increasing the amount of information, while reducing the inaccuracy of the identified parameters. The larger the value of the determinant of the Fisher information matrix, the lower its entropy value, the greater the amount of information, and the higher the reliability of the identified parameters. Therefore, in the process of modal identification, in order to maximize the modal recognition of the measured data, the determinant value of the Fisher information matrix should be increased as much as possible when optimizing the sensor layout. The larger the determinant of the Fisher information matrix, the more modal information it contains.

[0135] The determinant values ​​of the Fisher information matrix obtained by the three sensor optimization layout methods, namely the effective independence method, the energy coefficient-effective independence method and the partition weight-effective independence method, are 1.55e-19, 4.26e-15 and 7.07e-10, respectively. It can be seen that the partition weight-effective independence method can obtain more modal information while ensuring sufficient sensor spacing.

[0136] (3) Modal assurance criteria

[0137] Modal Assurance Criteria (MAC) is a tool for evaluating the intersection angle of modal vector space, which is used to measure the quality of modal discrimination. The idea is to require the angle between each modal vector to be as large as possible, that is, the dot product of each order modal vector is as small as possible. The calculation formula is as follows:

[0138]

[0139] In the formula, MAC ij is the i-th row and j-th column element of the modal assurance criterion matrix, representing the cosine of the spatial angle between the i-th modal vector and the j-th modal vector. The smaller the value of the off-diagonal element of the modal assurance criterion matrix, the smaller the cosine of the spatial angle between the i-th modal vector and the j-th modal vector, the larger the spatial angle between the i-th modal vector and the j-th modal vector, and the smaller the correlation. Accordingly, the i-th and j-th modes are easier to distinguish. When MAC ij When the value of is 0, it indicates that the i-th order modal vector and the j-th order modal vector are completely orthogonal, and the modal distinction effect is optimal.

[0140] The MAC matrix obtained by the sensor configuration method corresponds to Figure 5 to Figure 7 , where the two horizontal axes represent the modal order, and the vertical axis represents the non-diagonal element value of the MAC matrix. Since the diagonal elements are of the same order, the corresponding modes are all 1. From the MAC matrix diagram, it is shown that the corresponding non-diagonal elements should be as close to 0 as possible. Generally, the average value and maximum value of the non-diagonal elements of the MAC matrix are used for judgment. For actual engineering, the average value of the non-diagonal elements of MAC is usually less than 0.25, which means that the two modes are perfectly orthogonal, that is, easier to distinguish.

[0141] Figure 5 This is the MAC matrix diagram after optimization using the traditional method; at this time, the maximum non-diagonal element of MAC is 0.999, and the average value of the non-diagonal element of MAC is 0.116;

[0142] Figure 6 This is the MAC matrix diagram after optimization using the energy coefficient-effective independence method; at this time, the maximum non-diagonal element of MAC is 0.849, and the average value of the non-diagonal element of MAC is 0.137;

[0143] Figure 7 This is the MAC matrix diagram after optimization using the partition weight-effective independence method; at this time, the maximum non-diagonal element of MAC is 0.758, and the average value of the non-diagonal element of MAC is 0.058;

[0144] Under the same sensor arrangement, the partition weight-effective independence method also has obvious advantages in modal differentiation.

[0145] From the above description, it can be seen that in order to solve the problem of high information redundancy when arranging sensors on the wind turbine blade structure by the traditional effective independence method, a sensor space optimization arrangement method that avoids information redundancy is proposed under the premise of ensuring the modal differentiation effect: the partition weight-effective independence method. The main conclusions are as follows:

[0146] (1) For the problem of optimizing the placement of sensors in wind turbine blade structures in fine grid division, the traditional effective independence method has information redundancy. The present invention proposes a partition weight-effective independence method. This method can more accurately determine the optimal placement of sensors.

[0147] (2) The present invention uses the nearest neighbor distance, Fisher information matrix and modal assurance criterion method to compare three layout schemes (effective independence method, energy coefficient-effective independence method and partition weight-effective independence method). The research results show that the method proposed in the present invention can improve the rationality of the spatial layout of measurement nodes, ensure the independence of test data, and improve the modal differentiation effect when solving the problem of sensor optimization layout in wind turbine blade structure.

[0148] (3) When dealing with the sensor layout problem of wind turbine blade structure, the solution proposed by the present invention shows satisfactory results, can effectively avoid information redundancy, and ensure the accuracy of modal identification. Compared with the traditional method, the sensor optimization layout solution proposed by the present invention has better applicability, rationality and operability when dealing with complex and fine grid division problems.

[0149] (4) In actual engineering, the sensor arrangement of large and complex models often ignores some objective influences when simply relying on theoretical algorithm selection. For example, the locations where bird strikes, lightning strikes, icing, and other working conditions occur on wind turbine blades will not be integrated into the algorithm selection. However, the present invention can control the allocation of sensor detection locations by manually adjusting the weight score S(r) evaluation of the partition, which has certain guiding significance in actual engineering.

[0150] The embodiments described above are only preferred specific implementation modes of the present invention, and the protection scope of the present invention is not limited thereto. Any simple changes or equivalent replacements of the technical solutions that can be obviously obtained by any technician familiar with the field within the technical scope disclosed in the present invention belong to the protection scope of the present invention.

Claims

1. A method for optimizing the arrangement of wind turbine blade sensors, characterized in that: The steps include: Establish a finite element model of wind turbine blades, and use K-means classification method to divide the area to be tested in the finite element model into multiple partitions; Using the generalized modal kinetic energy method to assign weights to each of the partitions, and determining the number of sensors in each partition according to the weight assignment result; The modal matrix of the finite element model is defined as the Fisher information matrix. The trace and determinant of the Fisher information matrix are used as the quantization standard. The eigenvalues ​​of the Fisher information matrix of each partition are solved using the effective independence method to obtain the eigenvalue contribution matrix. Determine the arrangement position of the sensor in each partition through the eigenvalue contribution matrix; The K-means classification method is used to divide the area to be tested in the finite element model into multiple partitions, including the following steps: Obtain a three-dimensional point matrix D of a finite element model of a wind turbine blade; The three-dimensional dot matrix D of the finite element model of the wind turbine blade is input into K-means, and k center points are randomly generated; Calculate the distance between the three-dimensional dot matrix D and the center point, determine the partition of the three-dimensional dot matrix D according to the center point closest to the three-dimensional dot matrix D, add the three-dimensional dot matrix D to the corresponding partition, calculate the centroid of each partition, and use K-means to iterate the center point to obtain multiple partitions; The generalized modal kinetic energy method is used to assign weights to each partition, comprising the following steps: The contribution of the modal kinetic energy of the i nodes corresponding to the mode of the r-th partition is arithmetic averaged to obtain the average modal kinetic energy MKE(r) of the r-th partition; Calculating a weight score S(r) of the rth partition based on the average modal kinetic energy MKE(r) of the rth partition, and determining the weight of the rth partition by the weight score S(r); The calculation formula for the number of sensors in each partition is: N i =S(r)×t Where t is the total number of sensors in the area to be measured, and S(r) is the weight score S(r) of each partition.

2. The wind turbine blade sensor arrangement optimization method according to claim 1, characterized in that: The calculation formula of the average modal kinetic energy MKe(r) is: In the formula, MKE ij It represents the response degree of the j-order mode corresponding to the i-th node.

3. The wind turbine blade sensor arrangement optimization method according to claim 1, characterized in that: The calculation formula of the weight score S(r) is as follows: Where MKE(r) is the average modal kinetic energy, m is the node that the sensor can select, and n is the modal order of the modal matrix of the finite element model.

4. The wind turbine blade sensor arrangement optimization method according to claim 1, characterized in that: The Fisher information matrix is: R=Φ T f Where Φ is the modal matrix of the finite element model.

5. The wind turbine blade sensor arrangement optimization method according to claim 1, characterized in that: The calculation formula of the characteristic value is: Where Φ is the modal matrix of the finite element model, A im ΦΨ i The value of the mth element in the column vector, For the jth node, i , Ψ represents the eigenvector of the Fisher information matrix, and m is the node that the sensor can select.

6. A sensor position optimization arrangement device, characterized in that: include: The model segmentation module is used to establish a finite element model of a wind turbine blade and to segment the area to be tested in the finite element model into multiple partitions using the K-means classification method; A weight allocation module, used for using the generalized modal kinetic energy method to allocate weights to each of the partitions, and determining the number of sensors required for each partition according to the weight allocation result; The eigenvalue calculation module is used to define the modal matrix as the Fisher information matrix, use the trace and determinant of the Fisher information matrix as the quantization standard, and use the effective independence method to solve the eigenvalue of the Fisher information matrix of each partition to obtain the eigenvalue contribution matrix; A sensor placement module, used to determine the placement of the sensor in each partition through an eigenvalue contribution matrix; Among them, the K-means classification method is used to divide the area to be tested in the finite element model into multiple partitions, specifically: Obtain a three-dimensional point matrix D of a finite element model of a wind turbine blade; The three-dimensional dot matrix D of the finite element model of the wind turbine blade is input into K-means, and k center points are randomly generated; Calculate the distance between the three-dimensional dot matrix D and the center point, determine the partition of the three-dimensional dot matrix D according to the center point closest to the three-dimensional dot matrix D, add the three-dimensional dot matrix D to the corresponding partition, calculate the centroid of each partition, and use K-means to iterate the center point to obtain multiple partitions; The generalized modal kinetic energy method is used to assign weights to each partition, specifically: The contribution of the modal kinetic energy of the i nodes corresponding to the mode of the r-th partition is arithmetic averaged to obtain the average modal kinetic energy NKE(r) of the r-th partition; Calculating a weight score S(r) of the rth partition based on the average modal kinetic energy MKE(r) of the rth partition, and determining the weight of the rth partition by the weight score S(r); The calculation formula for the number of sensors in each partition is: N i =S(r)×t Where t is the total number of sensors in the area to be measured, and S(r) is the weight score S(r) of each partition.

7. A computer device, characterized in that: The method comprises a memory and a processor; the memory stores a computer program, and the processor is used to run the computer program in the memory to execute the method according to any one of claims 1 to 5.

Images