Wind turbine generator cluster blade detection system and detection method

Abstract

The invention discloses a wind turbine generator cluster blade detection system and method, and the system comprises an excitation unmanned plane which is provided with an excitation device for generating an excitation signal, and is used for flying to a detection point of a to-be-detected blade according to a preset path; the receiving unmanned aerial vehicle is provided with a response signal acquisition device which is used for acquiring a structural response signal corresponding to the excitation signal and is used for keeping a preset relative position with the excitation unmanned aerial vehicle according to a preset path; and the controller is used for acquiring a three-dimensional model of the to-be-detected blade, planning a cooperative flight path of the excitation unmanned aerial vehicle and the receiving unmanned aerial vehicle, controlling a relative position in real time, performing feature extraction on a structure response signal when the structure response signal is received, inputting features into a preset deep learning defect classification model, and identifying defect information of the to-be-detected blade. According to the invention, automatic, full-coverage, high-efficiency and accurate detection of internal defects of a large-scale structure is realized.

Description

Technical Field

[0001] This invention relates to the field of wind turbine technology, and in particular to a wind turbine cluster blade detection system and detection method. Background Technology

[0002] Large composite material structures such as wind turbine blades are prone to internal defects such as delamination, cracks, debonding, and honeycomb core damage during manufacturing, transportation, and long-term operation. If these defects are not detected in time, they may lead to catastrophic structural failure, causing huge economic losses and safety hazards.

[0003] Currently, the main methods for detecting internal defects in blades are as follows: 1. Manual "tapping" inspection: Inspectors identify defects by tapping the blade surface and judging the difference in echoes based on experience and hearing. This method is highly dependent on personnel experience, inefficient, and difficult to implement, posing extremely high safety risks, especially for large, high-altitude wind turbine blades.

[0004] 2. Ultrasonic testing: While offering high accuracy, it typically requires the use of a coupling agent and demands high flatness and accessibility of the surface being tested. This makes ultrasonic testing difficult to apply to complex curved surfaces and rapid on-site inspections, and it also prevents high-altitude operations.

[0005] 3. Acoustic cameras / acoustic imaging: These devices are expensive and their performance is limited in complex noise environments such as wind fields. They also face the challenge of working at heights.

[0006] 4. Single UAV inspection: mainly relies on optical cameras and thermal imaging cameras, which can only detect surface damage to blades and cannot effectively detect internal defects.

[0007] Therefore, there is an urgent need in this field for a detection solution that can be automated, comprehensive, efficient, and accurate in detecting internal defects in large structures. Summary of the Invention

[0008] The purpose of this invention is to overcome the shortcomings of the prior art and provide a wind turbine cluster blade inspection system and method. By coordinating the operation of excitation and reception drones, the traditional "knocking" inspection process is simulated. Combined with deep learning defect classification model analysis, the system achieves rapid, accurate and full-coverage inspection of internal defects in large structures.

[0009] The technical solution of the present invention provides a wind turbine cluster blade detection system, comprising: An excitation drone is provided, which is equipped with an excitation device for generating an excitation signal and flying to the detection point of the blade to be tested according to a preset path. A receiving drone is provided with a response signal acquisition device for acquiring structural response signals corresponding to the excitation signal, and for maintaining a preset relative position with the excitation drone according to the preset path; The controller is used to acquire a three-dimensional model of the blade under test, plan the cooperative flight path of the excitation UAV and the receiving UAV, control the relative position in real time, and extract features from the structural response signal when the structural response signal is received. The features are then input into a preset deep learning defect classification model to identify the defect information of the blade under test. The defect information includes the defect location, defect size and defect type.

[0010] In one of the alternative technical solutions, acquiring the three-dimensional model of the blade to be tested and planning the cooperative flight path of the excitation UAV and the receiving UAV includes: Obtain a three-dimensional model of the blade to be tested; Based on the three-dimensional model and the preset excitation-receiver geometry, the detection point grid covering the surface of the blade under test is calculated; Based on the detection point grid, the cooperative flight path is generated, wherein at each detection point, the distance and angle between the excitation UAV and the receiving UAV satisfy the excitation-reception geometric relationship.

[0011] In one of the alternative technical solutions, the deep learning defect classification model is obtained using the following method: Construct a deep learning network, the deep learning network comprising an input layer, at least one hidden layer, and an output layer; The historical defect information and the historical features corresponding to the historical defect information are divided into a training dataset and a validation dataset. The training dataset is input into the deep learning network, and the weights and biases of the deep learning network are adjusted using the backpropagation algorithm to minimize the error between the predicted defect information and the historical defect information. The deep learning network is then trained to obtain the deep learning defect classification model to be tested. The deep learning defect classification model to be tested is validated using the validation dataset to obtain the deep learning defect classification model.

[0012] In one of the alternative technical solutions, the features include time-domain features, frequency-domain features, and / or time-frequency-domain features.

[0013] In one of the alternative technical solutions, the excitation device includes at least one of an acoustic exciter, an electromagnetic excitation head, or a micro-impact device.

[0014] In one of the alternative technical solutions, the response signal acquisition device includes an acoustic sensor array and / or a laser vibrometer.

[0015] The technical solution of the present invention also provides a detection method for the wind turbine cluster blade detection system as described above, comprising: Step S1: Obtain the three-dimensional model of the blade to be tested, and plan the cooperative flight path of the excitation drone and the receiving drone so that the excitation drone and the receiving drone maintain a preset relative position. Step S2: Control the excitation drone and the receiving drone to fly together to the detection point, and control the excitation drone and the receiving drone to maintain the relative position in real time; Step S3: Control the excitation drone to generate an excitation signal, and the receiving drone collects the structural response signal corresponding to the excitation signal; Step S4: Extract features from the structural response signal; Step S5: Input the features into a preset deep learning defect classification model to identify the defect information of the blade to be tested. The defect information includes the defect location, defect size and defect type.

[0016] In one of the alternative technical solutions, step S5 is followed by: Step S6: Control the excitation drone and the receiving drone to fly together to the next detection point, and repeat steps S2 to S5 until all the blades to be tested in the wind turbine are scanned.

[0017] In one of the alternative technical solutions, step S1 includes: Obtain a three-dimensional model of the blade to be tested; Based on the three-dimensional model and the preset excitation-receiver geometry, the detection point grid covering the surface of the blade under test is calculated; Based on the detection point grid, the cooperative flight path is generated, wherein at each detection point, the distance and angle between the excitation UAV and the receiving UAV satisfy the excitation-reception geometric relationship.

[0018] In one of the alternative technical solutions, the deep learning defect classification model is obtained using the following method: Construct a deep learning network, the deep learning network comprising an input layer, at least one hidden layer, and an output layer; The historical defect information and the historical features corresponding to the historical defect information are divided into a training dataset and a validation dataset. The training dataset is input into the deep learning network, and the weights and biases of the deep learning network are adjusted using the backpropagation algorithm to minimize the error between the predicted defect information and the historical defect information. The deep learning network is then trained to obtain the deep learning defect classification model to be tested. The deep learning defect classification model to be tested is validated using the validation dataset to obtain the deep learning defect classification model.

[0019] The above technical solution has the following beneficial effects: By coordinating the operation of the excitation drone and the receiving drone, the traditional "tapping" inspection process is simulated, which can keenly capture the differences in weak signals. The detection rate and accuracy of internal defects are much higher than those of traditional inspection methods. Moreover, the drone swarm composed of the excitation drone and the receiving drone can reach any high altitude and complex area, realizing the detection of large structures without blind spots. The structural response signals collected by the receiving drone are used to extract features, and the extracted features are input into a deep learning defect classification model to identify the defect information of the blades. This realizes automated, fast, accurate and full-coverage non-contact inspection of internal defects of large structures, avoiding the damage to the surface caused by contact inspection. Attached Figure Description

[0020] The disclosure of this invention will become more readily understood by referring to the accompanying drawings. It should be understood that these drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings: Figure 1 This is a schematic diagram of a wind turbine cluster blade detection system according to an embodiment of the present invention; Figure 2 This is another structural schematic diagram of the wind turbine cluster blade detection system of the present invention; Figure 3 for Figure 1 A schematic diagram of the structure of the induction drone; Figure 4 for Figure 1 A schematic diagram of the receiving drone structure in the diagram; Figure 5 A schematic diagram illustrating how to incentivize collaborative operations between a drone and a receiving drone; Figure 6 A flowchart illustrating the detection method of a wind turbine cluster blade detection system according to an embodiment of the present invention; Figure 7 This is a flowchart illustrating the steps of collaborative flight path planning in one embodiment of the present invention; Figure 8 This is a flowchart illustrating the steps involved in obtaining a deep learning defect classification model in one embodiment of the present invention. Figure 9This is a flowchart illustrating the steps of defect information identification in one embodiment of the present invention; Figure 10 The flowchart illustrates the detection method of a wind turbine cluster blade detection system, which is a preferred embodiment of the present invention. Detailed Implementation

[0021] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0022] It is readily understood that, based on the technical solution of this invention, various structural and implementation methods can be interchanged by those skilled in the art without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of the invention.

[0023] The directional terms such as up, down, left, right, front, back, front, back, top, and bottom mentioned or possibly used in this specification are defined relative to the structures shown in the accompanying drawings. They are relative concepts and may therefore vary depending on their location and usage. Therefore, these or other directional terms should not be interpreted as restrictive.

[0024] like Figures 1-5 As shown, an embodiment of the present invention provides a wind turbine cluster blade detection system, comprising: The excitation drone 11 is equipped with an excitation device 111 for generating excitation signals, which is used to fly to the detection point of the blade 12 to be tested according to a preset path. The receiving drone 13 is equipped with a response signal acquisition device 131 for acquiring structural response signals corresponding to the excitation signal, and for maintaining a preset relative position with the excitation drone 11 according to the preset path. The controller 14 is used to acquire the three-dimensional model of the blade 12 to be tested, plan the cooperative flight path of the excitation UAV 11 and the receiving UAV 13, control the relative position in real time, and extract features from the structural response signal when the structural response signal is received, input the features into a preset deep learning defect classification model, and identify the defect information of the blade 12 to be tested. The defect information includes defect location, defect size and defect type.

[0025] Specifically, the wind turbine cluster blade detection system of the present invention mainly includes an excitation drone 11, a receiving drone 13, and a controller 14.

[0026] The excitation drone 11 is used to fly to the detection point of the blade 12 under test according to a preset path. The excitation drone 11 is equipped with a high-precision positioning module, an obstacle avoidance module, and a flight control unit to determine flight stability and positioning accuracy. The excitation drone 11 is equipped with an excitation device 111, which is used to generate an excitation signal to the surface of the blade 12 under test. The excitation device 111 can be selected according to user needs to adapt to different detection requirements.

[0027] In another embodiment, the excitation device 111 includes at least one of an acoustic exciter, an electromagnetic exciter head, or a micro-impact device. The acoustic exciter is a highly directional loudspeaker used to emit high-frequency acoustic pulses with an adjustable frequency (e.g., in the range of 1kHz-20kHz) to achieve non-contact excitation. The electromagnetic exciter head or micro-impact device is used to simulate a physical "knocking" on the surface of the blade 12 under test, generating stress waves.

[0028] The receiving drone 13 is also equipped with a high-precision positioning module, obstacle avoidance module, and flight control unit to ensure that it maintains a preset relative position with the excitation drone 11. The relative position refers to the stable and optimal excitation-reception geometric relationship between the excitation drone 11 and the receiving drone 13, such as maintaining a certain angle and distance. The receiving drone 13 carries a response signal acquisition device 131, which is used to acquire the structural response signal generated by the excitation point. In another embodiment, the response signal acquisition device 131 includes an acoustic sensor array and / or a laser vibrometer. The acoustic sensor array is used to acquire the echo generated by the excitation point. The laser vibrometer is used to acquire the vibration response signal generated by the excitation point.

[0029] The controller 14 acquires the 3D model data of the blade 12 under test and automatically generates a full-coverage detection path point grid based on this data. According to this grid, it generates a cooperative flight path for the excitation drone 11 and the receiving drone 13. When the blade 12 needs to be inspected, the controller controls the excitation drone 11 and the receiving drone 13 to fly to the detection point according to the cooperative flight path. It also controls the relative position and attitude of the excitation drone 11 and the receiving drone 13 in real time to ensure that the geometric relationship (such as distance and angle) between the excitation point and the acquisition point is always optimal. Figure 4As shown. After the excitation drone 11 and the receiving drone 13 fly to the detection point, the controller 14 controls the excitation device 111 to generate an excitation signal. When the structural response signal (such as an acoustic signal or vibration signal) transmitted by the receiving drone 13 is received, the structural response signal can be preprocessed, such as by filtering, noise reduction, and normalization, to improve the signal-to-noise ratio. Then, feature extraction is performed on the preprocessed structural response signal. In another embodiment, the features include time-domain features, frequency-domain features, and / or time-frequency-domain features. Finally, the extracted features are input into a preset deep defect classification model. The deep defect classification model identifies the defect information of the blade 12 under test, including the defect location, defect size, and defect type. The defect types include delamination, cracks, debonding, and honeycomb core pressure loss. The deep defect classification model is a complex nonlinear mapping relationship between features and defect types. Based on the nonlinear mapping relationship between the learned features and defect types, the deep defect classification model determines which defect type feature pattern the feature is closest to, thereby outputting the probability of a defect at the detection point and the specific defect type. For example, a certain time-frequency feature may be identified by the model as a "delamination defect". The relationship between features and defect types can be shown in the following table:

[0030] Preferably, the controller 14 is an electronic device with independent processing capabilities, such as a programmable logic controller (PLC) installed in a ground control station.

[0031] In this embodiment, by coordinating the operation of excitation and receiving drones, the traditional "tapping" inspection process is simulated. This approach can keenly capture differences in weak signals, and the detection rate and accuracy of internal defects are far higher than traditional inspection methods. Furthermore, the drone swarm composed of excitation and receiving drones can reach any high altitude and complex area, achieving comprehensive inspection of large structures. Feature extraction is performed on the structural response signals collected by the receiving drones, and the extracted features are input into a deep learning defect classification model to identify the defect information of the blades. This achieves automated, rapid, accurate, and comprehensive non-contact inspection of internal defects in large structures, avoiding surface damage caused by contact inspection.

[0032] In one embodiment, acquiring the three-dimensional model of the blade under test and planning the cooperative flight path of the excitation UAV and the receiving UAV includes: Obtain a three-dimensional model of the blade to be tested; Based on the three-dimensional model and the preset excitation-receiver geometry, the detection point grid covering the surface of the blade under test is calculated; Based on the detection point grid, the cooperative flight path is generated, wherein at each detection point, the distance and angle between the excitation UAV and the receiving UAV satisfy the excitation-reception geometric relationship.

[0033] Specifically, when the controller 14 detects the blade 12 under test, it first acquires the three-dimensional model data of the blade 12 under test, automatically plans a detection path point grid covering the entire blade surface of the wind turbine, and then generates a cooperative flight path between the excitation drone 11 and the receiving drone 13 based on the detection point grid. This ensures that at each detection point, the distance and angle between the excitation drone 11 and the receiving drone 13 meet the optimal excitation-reception geometric relationship, thereby achieving blind-angle detection of large structures and improving accuracy.

[0034] In one embodiment, the deep learning defect classification model is obtained using the following method: Construct a deep learning network, the deep learning network comprising an input layer, at least one hidden layer, and an output layer; The historical defect information and the historical features corresponding to the historical defect information are divided into a training dataset and a validation dataset. The training dataset is input into the deep learning network, and the weights and biases of the deep learning network are adjusted using the backpropagation algorithm to minimize the error between the predicted defect information and the historical defect information. The deep learning network is then trained to obtain the deep learning defect classification model to be tested. The deep learning defect classification model to be tested is validated using the validation dataset to obtain the deep learning defect classification model.

[0035] Specifically, when training the deep learning defect classification model, the controller 14 first constructs a deep learning network, such as a convolutional neural network, a recurrent neural network, or a long short-term memory network. This deep learning network includes an input layer, at least one hidden layer, and an output layer (outputting defect information). Then, historical defect information and features are divided into training and validation datasets. Historical defect information and features are obtained by performing high-frequency tapping detection on blades with known defect types and locations, collecting corresponding historical structural response signals, and preprocessing and extracting features from the collected historical structural response signals. Next, the training dataset is input into the constructed deep learning network. The weights and biases in the network are continuously adjusted using the backpropagation algorithm to minimize the loss function between predicted defect information and historical defect information, thus training the deep learning network to obtain the deep learning defect classification model to be tested. Finally, the validation dataset is used to evaluate the performance of the deep learning defect classification model to be tested. Based on the validation results, hyperparameters or network structure are adjusted until the deep learning defect classification model to be tested reaches the preset accuracy requirement, thereby obtaining the final deep learning defect classification model and improving the detection rate and accuracy of internal defects.

[0036] In one embodiment, a data management and visualization interface is also included for storing detection data and analysis results, and displaying defect distribution in a graphical manner, such as overlaying color images on a three-dimensional model of the blade to intuitively display the distribution, size and type of defects.

[0037] like Figure 6 As shown, an embodiment of the present invention provides a detection method for a wind turbine cluster blade detection system, comprising: Step S1: Obtain the three-dimensional model of the blade to be tested, and plan the cooperative flight path of the excitation drone and the receiving drone so that the excitation drone and the receiving drone maintain a preset relative position. Step S2: Control the excitation drone and the receiving drone to fly together to the detection point, and control the excitation drone and the receiving drone to maintain the relative position in real time; Step S3: Control the excitation drone to generate an excitation signal, and the receiving drone collects the structural response signal corresponding to the excitation signal; Step S4: Extract features from the structural response signal; Step S5: Input the features into a preset deep learning defect classification model to identify the defect information of the blade to be tested. The defect information includes the defect location, defect size and defect type.

[0038] Specifically, when it is necessary to inspect the blades of a wind turbine for defects, the controller first executes step S1 to acquire a 3D model of the blade to be tested, and automatically generates a full-coverage detection path point grid based on the 3D model data. Based on this detection path point grid, a coordinated flight path for the excitation drone and the receiving drone is generated. Then, step S2 is executed to control the excitation drone and the receiving drone to fly to the detection point according to the coordinated flight path, and to control the relative position and attitude of the excitation drone and the receiving drone in real time to ensure that the geometric relationship (such as distance and angle) between the excitation point and the acquisition point is always in an optimal state. After the excitation drone and the receiving drone fly to the detection point, step S3 is executed to control the excitation device on the excitation drone to generate an excitation signal on the surface of the blade to be tested. Upon receiving the structural response signal (such as an acoustic signal or vibration signal) transmitted by the receiving drone, the structural response signal can be preprocessed, such as through filtering, noise reduction, and normalization, to improve the signal-to-noise ratio. Then, step S4 is executed to extract features from the preprocessed structural response signal, including time-domain features, frequency-domain features, and / or time-frequency domain features. Finally, step S5 inputs the extracted features into a preset deep defect classification model. This model identifies the defect information of the blade under test, including defect location, size, and type. Defect types include delamination, cracks, debonding, and honeycomb core damage. The deep defect classification model represents a complex nonlinear mapping relationship between features and defect types. Based on this learned nonlinear mapping, the model determines which defect type's feature pattern a feature most closely resembles, thus outputting the probability of a defect at that detection point and the specific defect type. For example, a particular time-frequency feature might be identified by the model as a "delamination defect."

[0039] In this embodiment, by coordinating the operation of excitation and receiving drones, the traditional "tapping" inspection process is simulated. This approach can keenly capture differences in weak signals, and the detection rate and accuracy of internal defects are far higher than traditional inspection methods. Furthermore, the drone swarm composed of excitation and receiving drones can reach any high altitude and complex area, achieving comprehensive inspection of large structures. Feature extraction is performed on the structural response signals collected by the receiving drones, and the extracted features are input into a deep learning defect classification model to identify the defect information of the blades. This achieves automated, rapid, accurate, and comprehensive non-contact inspection of internal defects in large structures, avoiding surface damage caused by contact inspection.

[0040] In one embodiment, step S5 is followed by: The excitation drone and the receiving drone are controlled to fly together to the next detection point, and steps S2 to S5 are repeated until all the blades to be tested in the wind turbine are scanned. The blades to be tested in the entire wind turbine are detected, thereby achieving blind spot detection of large structures and improving accuracy.

[0041] In one embodiment, step S6 is followed by: It stores test data, analyzes results, generates test reports, and displays defect distribution graphically, such as overlaying color images on a 3D model of a blade to visually display the distribution, size, and type of defects.

[0042] like Figure 7 As shown, the step of acquiring a 3D model of the blade under test and planning a cooperative flight path between the excitation drone and the receiving drone to maintain a preset relative position includes: Step S701: Obtain the three-dimensional model of the blade to be tested; Step S702: Calculate the detection point grid covering the surface of the blade under test based on the three-dimensional model and the preset excitation-receiver geometry. Step S703: Based on the detection point grid, generate the cooperative flight path, wherein at each detection point, the distance and angle between the excitation UAV and the receiving UAV satisfy the excitation-reception geometric relationship.

[0043] Specifically, when the controller detects the blade under test, it first executes step S701 to obtain the three-dimensional model data of the blade under test; then it executes step S702 to automatically plan a detection path point grid covering the entire blade surface of the wind turbine based on the three-dimensional model; finally, it executes step S703 to generate a cooperative flight path between the excitation drone and the receiving drone based on the detection point grid, ensuring that at each detection point, the distance and angle between the excitation drone and the receiving drone meet the optimal excitation-reception geometric relationship, thereby achieving blind-spot-free detection of large structures and improving accuracy.

[0044] like Figure 8 As shown, the deep learning defect classification model is obtained using the following method: Step S801: Construct a deep learning network, the deep learning network including an input layer, at least one hidden layer and an output layer; Step S802: Divide the historical defect information and the historical features corresponding to the historical defect information into a training dataset and a validation dataset; Step S803: Input the training dataset into the deep learning network, use the backpropagation algorithm to adjust the weights and biases of the deep learning network to minimize the error between the predicted defect information and the historical defect information, train the deep learning network, and obtain the deep learning defect classification model to be tested. Step S804: Use the validation dataset to validate the deep learning defect classification model to be tested, and obtain the deep learning defect classification model.

[0045] Specifically, when training the deep learning defect classification model, the controller first executes step S801 to construct a deep learning network, such as a convolutional neural network, recurrent neural network, or long short-term memory network. The deep learning network includes an input layer, at least one hidden layer, and an output layer (outputting defect information). Then, step S802 divides the historical defect information and historical features into training datasets and validation datasets. The historical defect information and historical features can be obtained by performing high-frequency tapping detection on blades with known defect types and locations, collecting corresponding historical structural response signals, and preprocessing and extracting features from the collected historical structural response signals to obtain historical features. Next, step S803 inputs the training dataset into the constructed deep learning network, continuously adjusting the weights and biases in the network through the backpropagation algorithm to minimize the loss function between the predicted defect information and the historical defect information, thereby training the deep learning network and obtaining the deep learning defect classification model to be tested. Finally, step S804 evaluates the performance of the deep learning defect classification model under test using the validation dataset. Based on the validation results, the hyperparameters or network structure are adjusted until the deep learning defect classification model under test reaches the preset accuracy requirement, thereby obtaining the final deep learning defect classification model and improving the detection rate and accuracy of internal defects.

[0046] like Figure 9 As shown, in one embodiment of the present invention, the defect information identification step includes: Step S901: First stage: Offline model training; Specifically, the process involves collecting a large number of acoustic / vibration signals from labeled blades, preprocessing and extracting features from the labeled blades' acoustic / vibration signals, constructing and training a deep learning model, and validating and saving the optimal model (i.e., a deep defect classification model).

[0047] Step S902: Second stage: Online real-time detection.

[0048] Specifically, the process involves: receiving the raw structural response signal transmitted from the receiving UAV → preprocessing the structural response signal (including filtering, noise reduction, and normalization) → determining the detection mode → if it is a fast mode, extracting key features (including frequency domain features or time domain features) → inputting the features into the deep defect classification model; if it is a high-precision mode, generating a time domain map (such as a spectrum map) → inputting the time domain map into the deep defect classification model → model calculation → determining whether the output confidence score is greater than the threshold → if so, determining it as a "defect" and outputting the type and location → feeding back the detection result to the 3D visualization map in real time; otherwise, determining it as "normal" → completing the detection of this detection point and proceeding to the next detection point, repeating steps S802 and S803 until all blades in the entire wind turbine have been detected.

[0049] like Figure 10 As shown, the preferred embodiment of the present invention provides a detection method for a wind turbine cluster blade detection system, comprising: Step S1001: Phase 1: Task Planning and Initialization; Specifically, the process involves inputting a 3D model of the blade to be tested, planning a full-coverage detection path and a UAV collaborative flight route, and then having the UAV swarm take off and fly to the starting detection point.

[0050] Step S1002: Phase Two: Collaborative Detection and Cruise; Specifically, determine whether all detection points have been traversed → if not, excite the UAV to locate and emit high-frequency acoustic / mechanical excitation pulses → receive the UAV's synchronous positioning and collect echo / vibration response signals → transmit the signals to the ground control station in real time; otherwise, proceed to stage four.

[0051] Step S1003: Stage Three: Model Defect Analysis; Specifically, the process involves signal preprocessing (filtering, noise reduction, normalization) and feature extraction → model analysis to determine if the signal is normal. If normal, the status of the detection point is recorded as normal; otherwise, defect information (location, size, type) is recorded. The detection results are then annotated on the 3D model map in real time. The drone swarm then flies synchronously to the next detection point.

[0052] Step S1004: Stage Four: Generate a test report.

[0053] Specifically, the process involves integrating all detection data to generate a detection report, the drone swarm automatically returning to base and landing, and the mission ending.

[0054] The above embodiments are only used to illustrate the technical solutions of the embodiments of the present invention, and are not intended to limit them. Although the embodiments of the present invention have been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A wind turbine cluster blade detection system, characterized in that, include: An excitation drone is provided, which is equipped with an excitation device for generating an excitation signal and flying to the detection point of the blade to be tested according to a preset path. A receiving drone is provided with a response signal acquisition device for acquiring structural response signals corresponding to the excitation signal, and for maintaining a preset relative position with the excitation drone according to the preset path; The controller is used to acquire a three-dimensional model of the blade under test, plan the cooperative flight path of the excitation UAV and the receiving UAV, control the relative position in real time, and extract features from the structural response signal when the structural response signal is received. The features are then input into a preset deep learning defect classification model to identify the defect information of the blade under test. The defect information includes the defect location, defect size and defect type.

2. The wind turbine cluster blade detection system as described in claim 1, characterized in that, The process of acquiring a 3D model of the blade under test and planning the cooperative flight path of the excitation UAV and the receiving UAV includes: Obtain a three-dimensional model of the blade to be tested; Based on the three-dimensional model and the preset excitation-receiver geometry, the detection point grid covering the surface of the blade under test is calculated; Based on the detection point grid, the cooperative flight path is generated, wherein at each detection point, the distance and angle between the excitation UAV and the receiving UAV satisfy the excitation-reception geometric relationship.

3. The wind turbine cluster blade detection system as described in claim 1, characterized in that, The deep learning defect classification model was obtained using the following method: Construct a deep learning network, the deep learning network comprising an input layer, at least one hidden layer, and an output layer; The historical defect information and the historical features corresponding to the historical defect information are divided into a training dataset and a validation dataset. The training dataset is input into the deep learning network, and the weights and biases of the deep learning network are adjusted using the backpropagation algorithm to minimize the error between the predicted defect information and the historical defect information. The deep learning network is then trained to obtain the deep learning defect classification model to be tested. The deep learning defect classification model to be tested is validated using the validation dataset to obtain the deep learning defect classification model.

4. The wind turbine cluster blade detection system as described in claim 1, characterized in that, The features include time-domain features, frequency-domain features, and / or time-frequency-domain features.

5. The wind turbine cluster blade detection system as described in any one of claims 1-4, characterized in that, The excitation device includes at least one of an acoustic exciter, an electromagnetic exciter head, or a micro-impact device.

6. The wind turbine cluster blade detection system as described in claim 5, characterized in that, The response signal acquisition device includes an acoustic sensor array and / or a laser vibrometer.

7. A detection method for a wind turbine cluster blade detection system as described in any one of claims 1-6, characterized in that, include: Step S1: Obtain the three-dimensional model of the blade to be tested, and plan the cooperative flight path of the excitation drone and the receiving drone so that the excitation drone and the receiving drone maintain a preset relative position. Step S2: Control the excitation drone and the receiving drone to fly together to the detection point, and control the excitation drone and the receiving drone to maintain the relative position in real time; Step S3: Control the excitation drone to generate an excitation signal, and the receiving drone collects the structural response signal corresponding to the excitation signal; Step S4: Extract features from the structural response signal; Step S5: Input the features into a preset deep learning defect classification model to identify the defect information of the blade to be tested. The defect information includes the defect location, defect size and defect type.

8. The detection method as described in claim 7, characterized in that, Step S5 is followed by: Step S6: Control the excitation drone and the receiving drone to fly together to the next detection point, and repeat steps S2 to S5 until all the blades to be tested in the wind turbine are scanned.

9. The detection method as described in claim 7, characterized in that, Step S1 includes: Obtain a three-dimensional model of the blade to be tested; Based on the three-dimensional model and the preset excitation-receiver geometry, the detection point grid covering the surface of the blade under test is calculated; Based on the detection point grid, the cooperative flight path is generated, wherein at each detection point, the distance and angle between the excitation UAV and the receiving UAV satisfy the excitation-reception geometric relationship.

10. The detection method as described in claim 7, characterized in that, The deep learning defect classification model was obtained using the following method: Construct a deep learning network, the deep learning network comprising an input layer, at least one hidden layer, and an output layer; The historical defect information and the historical features corresponding to the historical defect information are divided into a training dataset and a validation dataset. The training dataset is input into the deep learning network, and the weights and biases of the deep learning network are adjusted using the backpropagation algorithm to minimize the error between the predicted defect information and the historical defect information. The deep learning network is then trained to obtain the deep learning defect classification model to be tested. The deep learning defect classification model to be tested is validated using the validation dataset to obtain the deep learning defect classification model.

Images