An Unmanned Aerial Vehicle Control Method and System Based on Detection of Wind Turbine Blade Damage

Abstract

The present invention relates to a method and system for controlling an unmanned aerial vehicle (UAV) based on detecting damage to wind turbine blades, belonging to the technical field of UAVs. The present invention includes obtaining the real-time distance between the current position of the UAV and the center of the wind turbine through an RGB-D camera and correcting the position of the UAV; obtaining an image of the wind turbine through the RGB-D camera, establishing a spatial coordinate system, and calculating the recognition result of the wind turbine blades through an image detection and recognition model; presetting a sequence of detection points, obtaining relevant parameters, and calculating the changed coordinates of the detection points through a coordinate calculation model according to the relevant parameters; inputting the relevant parameters into a UAV control network model to obtain the control amount of the motor change of the UAV; the intelligent terminal generates an adaptive signal according to the control amount of the motor change to control the UAV to move to the detection point and implement defect detection; repeating the execution until all blades are detected. High-precision UAV control for detecting wind turbine blades is achieved.

Description

Technical Field

[0001] The present invention belongs to the technical field of unmanned aerial vehicles, and particularly relates to a control method and system for an unmanned aerial vehicle based on detecting damage to wind turbine blades. Background Art

[0002] With the rapid development of renewable energy technology, wind power generation has become an important part of the global energy structure. In the wind power generation industry, the reliability and performance of wind turbine blades are crucial. However, due to their long-term exposure to natural environments and operating conditions, defects often occur on the blades, which may lead to performance degradation or even failures. Current detection methods usually require shutdown maintenance or the use of expensive equipment, relying on manual climbing or using ropes to hang for detection. These methods have problems such as high labor intensity, low efficiency, and high danger, and cannot achieve real-time monitoring and tracking of blade defects.

[0003] Existing problems:

[0004] (1) The damage detection of wind turbine blades mainly relies on manual labor, with problems such as high labor intensity, low efficiency, and poor safety.

[0005] (2) Existing unmanned aerial vehicle control systems cannot meet the requirements of precise detection of wind turbine blades, lacking professional control strategies and algorithms. Summary of the Invention

[0006] To solve the above problems existing in the prior art, the present invention provides a control method and system for an unmanned aerial vehicle based on detecting damage to wind turbine blades.

[0007] The object of the present invention can be achieved by the following technical solutions:

[0008] S1: Control the unmanned aerial vehicle to fly relative to the wind turbine, obtain the real-time distance between the current position of the unmanned aerial vehicle and the center of the wind turbine through the RGB-D camera carried on the unmanned aerial vehicle, and transmit it to the intelligent terminal. The intelligent terminal compares the real-time distance with the preset optimal detection distance to obtain a comparison result, and corrects the position of the unmanned aerial vehicle according to the comparison result.

[0009] S2: Real-time obtain wind turbine images through the RGB-D camera and transmit them to the intelligent terminal. The intelligent terminal establishes a spatial coordinate system according to the wind turbine images, and calculates the first blade, the second blade, and the third blade through an image detection and recognition model, and uses the first blade as the detection blade.

[0010] S3: Preset the detection point sequence of the detection blade, obtain the initial coordinates of each detection point in the detection point sequence in the spatial coordinate system, obtain the rotational angular velocity of the detection blade, the distance data between the detection point and the center point, the instantaneous velocity of the drone, the detection running speed of the drone, and the current coordinates of the drone. Calculate the transformed coordinates of the detection point within a preset time period according to the rotational angular velocity and the initial coordinates of the detection point through a coordinate calculation model, and calculate the coordinate difference between the current coordinates of the drone and the transformed coordinates;

[0011] S4: Input the rotational angular velocity, the detection running speed of the drone, the instantaneous velocity of the drone, the distance data between the detection point and the center point, the motor control amount of the drone, and the coordinate difference as state quantities into the drone control network model to calculate the motor change control amount of the drone through calculation;

[0012] S5: The intelligent terminal generates an adaptive signal according to the motor change control amount to control the running state of the propellers of the drone, so that the drone runs to the position corresponding to the transformed coordinates, and uses the ultrasonic machine carried on the drone to detect the defects in the area corresponding to the transformed coordinates;

[0013] S6: Repeat steps S1 - S5 until all the second blade and the third blade are detected.

[0014] As a preferred technical solution of the present invention, the specific steps for training the drone control network model are as follows:

[0015] S401: Set up a detection virtual environment for the wind turbine blade;

[0016] S402: Obtain the state quantities of all detection points at the current moment in the detection virtual environment, set a reward function according to the state quantities at the current moment, input the state quantities at the current moment into a preset drone control network model to calculate the motor change control amount of the drone through calculation, obtain the state quantities at the next moment, establish an experience quadruple with the state quantities at the current moment, the motor change control amount, the reward function, and the state quantities at the next moment, and store all the experience quadruples in a reinforcement learning dedicated experience replay pool. When all the blades are detected, it indicates the end of this round of interaction;

[0017] S403: Take n groups of experience quadruples with poor correlation from the reinforcement learning dedicated experience replay pool through the prioritized experience replay algorithm as a test group, and continuously update the parameters of the preset drone control network model according to the test group;

[0018] S404: Obtain the reward function curve graph, and determine whether the reward function tends to be stable according to the reward function curve graph. If the reward function tends to be stable, end the loop to obtain the UAV control network model; if the reward function does not tend to be stable, continue to repeat steps S402 - S404 for training until the reward function tends to be stable.

[0019] Specifically, the reward function r t (s t ) has the following expression:

[0020]

[0021] where ω1, ω2, ω3 represent hyperparameters, ω v is the rotational angular velocity, l r is the distance data between the detection point and the center point, v0 is the detection running speed of the UAV, v t is the instantaneous speed of the UAV, Δx, Δy, Δz are the coordinate differences, u1, u2, u3, u4 are the motor control quantities of the first wing, second wing, third wing, and fourth wing of the UAV respectively, Δu1, Δu2, Δu3, Δu4 are the motor change control quantities of the first wing, second wing, third wing, and fourth wing of the UAV.

[0022] Specifically, the UAV control network model consists of an Actor network, a Target Actor network, a Critic1 network, a Target Critic1 network, a Critic2 network, and a Target Critic2 network. The input of the Actor network is the state quantity s t perceived by the UAV at the current moment, and the output quantity of the Actor network is the control quantity a t sent to the UAV motor at the current moment. The input of the Target Actor network is the state quantity s t ' perceived by the UAV at the next moment, and the output quantity of the Target Actor network is the control quantity a t ' sent to the UAV motor at the next moment; the inputs of the Critic1 network and the Critic2 network are the state-action pair (s t , a t ) at the current moment, and the output is the state value evaluation value given by applying the neural network according to the state-action pair (s t , a t ) at the current moment; the inputs of the Target Critic1 network and the Target Critic2 network are the state-action pair (s t ', a t'), the output is the state-value target value given by applying the neural network to the state-action pair (s t ', a t ')

[0023] Specifically, the calculation formula for the motor control quantity is as follows:

[0024] a t = μ(s t |θ μ ) + ∈

[0025] ∈ = clip(N(0, σ), -b, b)

[0026] where s t represents the current state quantity, θ μ represents the Actor network parameters, μ(·) represents the Actor network, ∈ represents the random noise parameter, a t is the motor control quantity, the clip function means that when N(0, σ) < -b, ∈ = -b, when N(0, σ) > b, ∈ = b, ∈ = N(0, σ) means that ∈ follows a normal distribution, and -b and b represent fixed parameters.

[0027] Specifically, the calculation formula for the prioritized experience replay algorithm to screen out experience data from the experience buffer is:

[0028]

[0029] where P(c) represents a set of experience quadruples (s t , a t , r t , s t ') screened out from the experience replay pool; c represents the serial number of the extracted experience data; p k represents the priority of the extracted experience data; α represents a preset parameter for adjusting the priority sampling degree of data samples.

[0030] Specifically, the coordinate calculation model regards the wind turbine as rotating around the Z-axis of the space coordinate system, calculates the rotation angle of the detection blade around the Z-axis according to the rotation angular velocity, and establishes a rotation matrix according to the rotation angle, and then calculates the transformed coordinates according to the rotation matrix and the initial coordinates through the coordinate transformation algorithm. The calculation formula of the coordinate transformation algorithm is:

[0031]

[0032] where (x1, y1, z1) is the transformed coordinate, a is the scale parameter, b is the rotation angle, and (x, y, z) is the initial coordinate.

[0033] A drone control system for detecting damage to wind turbine blades, comprising: a position correction module, an identification module, a data acquisition module, a control calculation module, and a detection module;

[0034] The position correction module is used to control the drone to fly relative to the wind turbine. The real-time distance between the current position of the drone and the center of the wind turbine is obtained by the RGB-D camera carried on the drone and transmitted to the intelligent terminal. The intelligent terminal compares the real-time distance with the preset optimal detection distance to obtain a comparison result, and corrects the position of the drone according to the comparison result;

[0035] The identification module is used to obtain the wind turbine image in real time through the RGB-D camera and transmit it to the intelligent terminal. The intelligent terminal establishes a spatial coordinate system according to the wind turbine image, and calculates the first blade, the second blade, and the third blade through an image detection and recognition model, and takes the first blade as the detection blade;

[0036] The data acquisition module is used to preset the detection point sequence of the detection blade, and obtain the initial coordinates of each detection point in the detection point sequence in the spatial coordinate system, obtain the rotational angular velocity of the detection blade, the distance data between the detection point and the center point, the instantaneous speed of the drone, the detection running speed of the drone, and the current coordinates of the drone. According to the rotational angular velocity and the initial coordinates of the detection point, the transformed coordinates of the detection point within a preset time period are calculated through a coordinate calculation model, and the coordinate difference between the current coordinates of the drone and the transformed coordinates is calculated;

[0037] The control calculation module is used to input the rotational angular velocity, the detection running speed of the drone, the instantaneous speed of the drone, the distance data between the detection point and the center point, the motor control amount of the drone, and the coordinate difference as state quantities into the drone control network model to calculate the motor change control amount of the drone through calculation;

[0038] The detection module is used for the intelligent terminal to generate an adaptive signal according to the motor change control amount to control the running state of the propeller of the drone, so that the drone runs to the transformed coordinates and uses the ultrasonic machine carried on the drone to detect the defects in the area corresponding to the transformed coordinates on the detection blade.

[0039] An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor implements the above-mentioned drone control method for detecting damage to wind turbine blades when executing the program.

[0040] A storage medium containing computer-executable instructions, characterized in that the computer-executable instructions are used to execute the unmanned aerial vehicle control method for detecting damage to wind turbine blades as described above when executed by a computer processor.

[0041] The beneficial effects of the present invention are as follows:

[0042] (1) An unmanned aerial vehicle equipped with an RGB-D camera is used to collect images of a wind turbine, and a spatial coordinate system is established based on the wind turbine images. The recognition result of the wind turbine blades is obtained through an image detection and recognition model, and a sequence of detection points is marked on the spatial coordinate system. The unmanned aerial vehicle tracks in real time according to the detection point coordinates and uses an ultrasonic machine to emit ultrasonic waves to detect the detection point area, solving the problem of the need to stop the wind turbine for detection and avoiding the need for manual detection.

[0043] (2) The rotational angular velocity, the detection running speed of the unmanned aerial vehicle, the instantaneous speed of the unmanned aerial vehicle, the distance data between the detection point and the center point, the motor control amount of the unmanned aerial vehicle, and the coordinate difference are used as state quantities and input into the unmanned aerial vehicle control network model. The motor change control amount of the unmanned aerial vehicle is obtained through calculation, and the heading and flight speed of the unmanned aerial vehicle are changed according to the motor change control amount, improving the tracking and positioning of the unmanned aerial vehicle for the detection point.

[0044] (3) The prioritized experience replay algorithm is introduced into the unmanned aerial vehicle control network model, improving the learning rate of the unmanned aerial vehicle control network model, reducing sample bias, and improving stability, robustness, and generalization ability. Description of the Drawings

[0045] For the convenience of those skilled in the art to understand, the present invention will be further described below with reference to the accompanying drawings.

[0046] Figure 1 It is a schematic flow chart of an unmanned aerial vehicle control method for detecting damage to wind turbine blades according to the present invention.

[0047] Figure 2 It is a flow chart of the training process of the unmanned aerial vehicle control network model according to the present invention.

[0048] Figure 3 It is a schematic diagram of the network architecture of the Actor class network.

[0049] Figure 4 It is a schematic diagram of the network architecture of the Critic class network. Detailed Embodiments

[0050] To further elaborate on the technical means and effects adopted by the present invention to achieve the intended invention purpose, the following describes in detail the specific embodiments, structures, features, and effects of the present invention with reference to the accompanying drawings and preferred embodiments.

[0051] Please refer to Figure 1 , a method for controlling an unmanned aerial vehicle (UAV) based on detecting damage to wind turbine blades,

[0052] S1: Control the UAV to fly relative to the wind turbine. Obtain the real-time distance between the current position of the UAV and the center of the wind turbine through the RGB-D camera carried on the UAV and transmit it to the intelligent terminal. The intelligent terminal compares the real-time distance with the preset optimal detection distance to obtain a comparison result, and corrects the position of the UAV according to the comparison result;

[0053] S2: Real-time obtain the wind turbine image through the RGB-D camera and transmit it to the intelligent terminal. The intelligent terminal establishes a spatial coordinate system based on the wind turbine image, and calculates the first blade, the second blade, and the third blade through an image detection and recognition model, and uses the first blade as the detection blade;

[0054] S3: Preset the detection point sequence of the detection blade, and obtain the initial coordinates of each detection point in the detection point sequence in the spatial coordinate system. Obtain the rotational angular velocity of the detection blade, the distance data between the detection point and the center point, the instantaneous velocity of the UAV, the detection running speed of the UAV, and the current coordinates of the UAV. Calculate the transformed coordinates of the detection point within a preset time period through a coordinate calculation model according to the rotational angular velocity and the initial coordinates of the detection point, and calculate the coordinate difference between the current coordinates of the UAV and the transformed coordinates;

[0055] S4: Input the rotational angular velocity, the detection running speed of the UAV, the instantaneous velocity of the UAV, the distance data between the detection point and the center point, the motor control amount of the UAV, and the coordinate difference as state quantities into the UAV control network model to calculate the motor change control amount of the UAV through calculation;

[0056] S5: The intelligent terminal generates an adaptive signal according to the motor change control amount to control the operating state of the propellers of the UAV, so that the UAV runs to the transformed coordinates and uses the ultrasonic machine carried on the UAV to detect defects in the area corresponding to the transformed coordinates on the detection blade;

[0057] S6. Repeat steps S1 - S5 until all the second blade and the third blade are detected.

[0058] In this embodiment, the image detection and recognition model is an image recognition model based on deep learning, which can effectively learn the features of the image through multiple convolutional layers and pooling layers, and can identify the wind turbine blades from the wind turbine image.

[0059] As a preferred technical solution of the present invention, the specific steps for training the UAV control network model are as follows:

[0060] S401: Set up a detection virtual environment for the wind turbine blade;

[0061] S402: Obtain the state quantities at the current moment of all detection points in the detection virtual environment, set a reward function according to the state quantities at the current moment, input the state quantities at the current moment into a preset UAV control network model to calculate the motor change control quantity of the UAV, obtain the state quantity at the next moment, establish an experience quadruple with the state quantity at the current moment, the motor change control quantity, the reward function, and the state quantity at the next moment, and store all the experience quadruples in a reinforcement learning dedicated experience replay pool. When all blade detections are completed, it indicates the end of this interaction;

[0062] S403: Take n groups of experience quadruples with poor correlation from the reinforcement learning dedicated experience replay pool through the prioritized experience replay algorithm as a test group, and continuously update the parameters of the preset UAV control network model according to the test group;

[0063] S404: Obtain a reward function curve graph, judge whether the reward function tends to be stable according to the reward function curve graph. If the reward function tends to be stable, end the loop and obtain the UAV control network model; if the reward function does not tend to be stable, continue to repeat steps S402 - S404 for training until the reward function tends to be stable.

[0064] In this embodiment, the UAV control network model adopts an improved double - delay deep deterministic policy gradient algorithm.

[0065] Specifically, the reward function r t (s t ) is expressed as:

[0066]

[0067] Among them, ω1, ω2, ω3 represent hyperparameters, ω v is the rotational angular velocity, l r is the distance data between the detection point and the center point, v0 is the detection running speed of the UAV, v t is the instantaneous speed of the UAV, Δx, Δy, Δz are the coordinate differences, are respectively the motor control quantities of the first wing, the second wing, the third wing, and the fourth wing of the UAV, is the motor change control quantity of the first wing, the second wing, the third wing, and the fourth wing of the UAV.

[0068] Specifically, the UAV control network model consists of an Actor network, a Target Actor network, a Critic1 network, a Target Critic1 network, a Critic2 network, and a Target Critic2 network. The input of the Actor network is the state quantity s sensed by the UAV at the current moment t , and the output quantity of the Actor network is the control quantity a sent to the UAV motor at the current moment t , the input of the Target Actor network is the state quantity s' sensed by the UAV at the next moment t , and the output quantity of the Target Actor network is the control quantity a' sent to the UAV motor at the next moment t ; the inputs of the Critic1 network and the Critic2 network are the state-action pair (s t , a t ) at the current moment, and the output is the state value evaluation value given by applying the neural network according to the state-action pair (s t , a t ) at the current moment; the inputs of the Target Critic1 network and the Target Critic2 network are the state-action pair (s t ', a t ) at the next moment, and the output is the state value target value given by applying the neural network according to the state-action pair (s t ', a t ) at the next moment.

[0069] In this embodiment, the process of updating the parameters of the Critic1 and Critic2 networks:

[0070] Use the Target Actor network to calculate the action a t ' under the state quantity s t ' at the next moment. The calculation formula of the action a t ' is as follows:

[0071] a t ′ = μ′(s t ′|θ μ′ ),

[0072] where θ μ' is the network parameter of the Target Actor network, and μ'(·) represents the Target Actor network;

[0073] Then, based on the target policy smoothing regularization, add noise to the target action a t ':

[0074] at = μ(s t |θ μ ) + ∈

[0075] ∈ = clip(N(0, σ), -b, b)

[0076] Next, based on the idea of the dual network, use the Target Critic network to calculate the state-value target y t , a t ) of the state-action pair (s t , and the calculation formula of the state-value target y t is as follows:

[0077]

[0078] where r t is the reward function, γ is the discount return rate, θ i Q' are the network parameters of TargetCritic1 and Target Critic2 networks, and Q i '(·) represents the Target Critic1 and Target Critic2 networks;

[0079] Finally, use the gradient descent algorithm to minimize the error L ci between the evaluation value and the target value, so as to update the parameters in the Critic1 and Critic2 networks:

[0080]

[0081] where Q i (·) represents the Target Critic1 and Target Critic2 networks, s t and a t represent the state quantity and action quantity at the current moment, and θ i Q are the network parameters of the Critic1 and Critic2 networks.

[0082] Update process of Actor network parameters:

[0083] After the Critic1 and Critic2 networks are updated d steps, start the update of the Actor network, and use the Actor network to calculate the action a t under the state s t :

[0084] a t = μ(s t |θ μ ),

[0085] Among them, θ μ is the network parameter of the Actor network, and μ(·) represents the Actor network. It should be noted here that no noise needs to be added after calculating the action because it is expected that the Actor network can be updated in the direction of the maximum value, and adding noise has no meaning;

[0086] Use the Critic1 or Critic2 network to calculate the state value evaluation value q t , a t ) of the state-action pair (s t , and the calculation formula of the state value evaluation value q t is as follows. Here, it is assumed that the Critic1 network is used:

[0087] q t = Q1(s t , a t |θ Q1 ),

[0088] Among them, θ Q1 is the network parameter of the Critic1 network, and Q1(·) represents the Critic1 network;

[0089] Finally, use the gradient ascent algorithm to maximize the state value evaluation value q t , thereby updating the parameter θ μ in the Actor network. The reason why either Critic1 or Critic2 can be used to calculate the Q value here is mainly because the purpose of the Actor network is to maximize the cumulative expected return, and there is no need to use the minimum value.

[0090] The TD3 algorithm adopts a soft update method. Among them, the parameter update process of the Target Actor network is as follows:

[0091] θ μ′ = τθ μ + (1 - τ)θ μ′ ,

[0092] The update processes of the Target Critic1 and Target Critic2 networks are as follows:

[0093]

[0094] Among them, τ is the learning rate (momentum), τ ∈ (0, 1), and usually takes the value of 0.005.

[0095] Specifically, noise needs to be added when calculating the motor control quantity, and the calculation formula of the motor control quantity is as follows:

[0096] at = μ(s t |θ μ ) + ∈

[0097] ∈ = clip(N(0, σ), -b, b)

[0098] where s t represents the current state quantity, θ μ represents the Actor network parameters, μ(·) represents the Actor network, a t is the motor control quantity, ∈ represents the random noise parameter; the clip function means that when N(0, σ) < -b, ∈ = -b, when N(0, σ) > b, ∈ = b, ∈ = N(0, σ) means that ∈ follows a normal distribution, and -b and b are fixed parameters.

[0099] In this embodiment, the Actor network mainly includes 1 input layer, 4 hidden layers and 1 output layer. The input layer contains 11 nodes, corresponding to 11 parameters of the state quantity s t . The hidden layers are composed of linear layers. The four hidden layers respectively contain 512, 512, 256, and 128 nodes. The output layer contains 4 nodes, corresponding to four parameters of the control quantity a t . The Critic network architecture contains two branches. The first branch is similar to the Actor network and includes 1 input layer and 4 hidden layers. The input layer contains 11 nodes, corresponding to 11 parameters of the state quantity s t . The hidden layers are composed of linear layers. The four hidden layers respectively contain 512, 512, 256, and 128 nodes. The first branch is composed of 1 input layer and 2 hidden layers. The input layer is the output layer of the Actor network, that is, the action quantity a t , which contains 4 nodes. The two hidden layers respectively contain 256 and 128 nodes. The two branches have hidden layers of the same size with 128 nodes. Therefore, the outputs of the two hidden layers with 128 nodes are added and normalized and then output to the output layer. The output layer contains 1 node, that is, the evaluation value or target value of the state-action pair.

[0100] Specifically, the formula for the prioritized experience replay algorithm to screen out experience data from the experience buffer is:

[0101]

[0102] where P(c) represents a set of experience quadruples (s t , a t , r t , s t ') screened out from the experience replay pool; c represents the serial number of the extracted experience data; p kIndicates the priority of the extracted empirical data; α represents a preset parameter for adjusting the priority sampling degree of the data sample.

[0103] Specifically, the coordinate calculation model regards the wind turbine as rotating around the Z-axis of the spatial coordinate system, calculates the rotation angle of the detection blade around the Z-axis according to the rotation angular velocity, establishes a rotation matrix according to the rotation angle, and then calculates the transformed coordinates through a coordinate transformation algorithm according to the rotation matrix and the initial coordinates. The calculation formula of the coordinate transformation algorithm is:

[0104]

[0105] Among them, (x1, y1, z1) are the transformed coordinates, a is the scale parameter, b is the rotation angle, and (x, y, z) are the initial coordinates.

[0106] An unmanned aerial vehicle control system based on wind turbine blade detection damage includes: a position correction module, an identification module, a data acquisition module, a control calculation module, and a detection module;

[0107] The position correction module is used to control the flight of the unmanned aerial vehicle relative to the wind turbine, obtain the real-time distance between the current position of the unmanned aerial vehicle and the center of the wind turbine through the RGB-D camera carried on the unmanned aerial vehicle and transmit it to the intelligent terminal. The intelligent terminal compares the real-time distance with the preset optimal detection distance to obtain a comparison result, and corrects the position of the unmanned aerial vehicle according to the comparison result;

[0108] The identification module is used to obtain the wind turbine image in real time through the RGB-D camera and transmit it to the intelligent terminal. The intelligent terminal establishes a spatial coordinate system according to the wind turbine image, and calculates the first blade, the second blade, and the third blade through an image detection and recognition model, and uses the first blade as the detection blade;

[0109] The data acquisition module is used to preset the detection point sequence of the detection blade, obtain the initial coordinates of each detection point in the detection point sequence in the spatial coordinate system, obtain the rotation angular velocity of the detection blade, the distance data between the detection point and the center point, the instantaneous speed of the unmanned aerial vehicle, the detection running speed of the unmanned aerial vehicle, and the current coordinates of the unmanned aerial vehicle. Calculate the transformed coordinates of the detection point within a preset time period according to the rotation angular velocity and the initial coordinates of the detection point through a coordinate calculation model, and calculate the coordinate difference between the current coordinates of the unmanned aerial vehicle and the transformed coordinates;

[0110] The control calculation module is configured to input the rotational angular velocity, the detected operating speed of the UAV, the instantaneous speed of the UAV, the distance data between the detection point and the center point, the motor control quantity of the UAV, and the coordinate difference as state quantities into the UAV control network model, and calculate the motor change control quantity of the UAV through calculation;

[0111] The detection module is configured to generate an adaptive signal by the intelligent terminal according to the motor change control quantity to control the operating state of the propellers of the UAV, so that the UAV runs to the transformed coordinate, and use the ultrasonic machine carried on the UAV to detect the defects in the area corresponding to the transformed coordinate on the detected blade.

[0112] An electronic device includes a memory, a processor, and a computer program stored on the memory and executable on the processor. The electronic device is characterized in that when the processor executes the program, the above-mentioned UAV control method for detecting damage to wind turbine blades is implemented.

[0113] A storage medium containing computer-executable instructions is characterized in that the computer-executable instructions are used to execute the above-mentioned UAV control method for detecting damage to wind turbine blades when executed by a computer processor.

[0114] The program code included in the method in the embodiments of the present invention can be transmitted by any suitable medium, including but not limited to wireless, wire, optical cable, etc., or any suitable combination of the above. The computer program code for performing the operations of the present invention can be written in one or more programming languages or combinations thereof. The programming languages include object-oriented programming languages such as Java, Smalltalk, C++, and also include conventional procedural programming languages such as the "C" language or similar programming languages. The program code can be executed completely on the user computer, partially on the user computer, executed as an independent software package, partially on the user computer and partially on a remote computer, or completely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (for example, by using an Internet service provider to connect through the Internet).

[0115] The above are only the preferred embodiments of the present invention, and do not impose any form of limitation on the present invention. Although the present invention has been disclosed above with the preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some changes or modifications to equivalent embodiments by using the above-disclosed technical content without departing from the technical solution of the present invention. However, as long as it does not depart from the technical solution content of the present invention, any simple modification, equivalent change and modification made to the above embodiments according to the technical essence of the present invention still belong to the scope of the technical solution of the present invention.

Claims

1. A UAV control method based on wind turbine blade damage detection, characterized in that: The following steps are involved: S1: Control the drone to fly relative to the wind turbine, obtain the real-time distance between the current position of the drone and the center of the wind turbine through the RGB-D camera carried by the drone and transmit it to the smart terminal, the smart terminal compares the real-time distance with the preset optimal detection distance to obtain a comparison result, and corrects the drone position according to the comparison result; S2: The wind turbine image is acquired in real time by the RGB-D camera and transmitted to the smart terminal. The smart terminal establishes a spatial coordinate system according to the wind turbine image, and calculates blade No. 1, blade No. 2, and blade No. 3 by using an image detection and recognition model, and uses blade No. 1 as the detection blade. S3: Preset a detection point sequence for the detection blade, and obtain the initial coordinates of each detection point in the detection point sequence in the spatial coordinate system, obtain the rotation angular velocity of the detection blade, the distance data between the detection point and the center point, the instantaneous speed of the drone, the detection running speed of the drone and the current coordinates of the drone, calculate the transformed coordinates of the detection point within a preset time length according to the rotation angular velocity and the initial coordinates of the detection point through a coordinate calculation model, and calculate the coordinate difference between the current coordinates of the drone and the transformed coordinates; S4: Input the rotation angular velocity, the UAV detection running speed, the UAV instantaneous speed, the distance data between the detection point and the center point, the motor control amount of the UAV, and the coordinate difference as state quantities into the UAV control network model to obtain the motor change control amount of the UAV through calculation; The specific steps of training the UAV control network model are as follows: S401: Setting a virtual environment for detecting wind turbine blades; S402: obtaining the current state quantity of all detection points in the detection virtual environment, setting a reward function according to the current state quantity, inputting the current state quantity into the preset UAV control network model to obtain the motor change control quantity of the UAV by calculation, obtaining the state quantity at the next moment, establishing an experience quadruple with the current state quantity, the motor change control quantity, the reward function, and the next moment state quantity, and storing all the experience quadruple in the reinforcement learning dedicated experience playback pool. When all blades are detected, it indicates that the interaction of this game is over; S403: taking n groups of experience quadruple groups with poor correlation from the reinforcement learning dedicated experience replay pool as test groups through a priority experience replay algorithm, and continuously updating the preset UAV control network model parameters according to the test groups; S404: Obtain a reward function curve graph, and determine whether the reward function tends to be stable according to the reward function curve graph. If the reward function tends to be stable, the loop is terminated to obtain the UAV control network model; if the reward function does not tend to be stable, continue to repeat steps S402-S404 for training until the reward function tends to be stable; S5: the intelligent terminal generates an adaptive signal according to the motor change control amount to control the propeller operation state of the drone, so that the drone runs to the transformed coordinates, and uses the ultrasonic machine carried by the drone to perform defect detection on the area corresponding to the transformed coordinates; S6. Repeat steps S1-S5 until the second blade and the third blade are all detected.

2. The UAV control method based on wind turbine blade damage detection according to claim 1 is characterized in that: The reward function r t (s t ) is expressed as: Among them, ω1, ω2, ω3 represent hyperparameters, ω v is the rotation angular velocity, l r is the distance data between the detection point and the center point, v0 is the detection running speed of the drone, and v t is the instantaneous speed of the drone, Δx, Δy, Δz are the coordinate differences, are the motor control quantities of the first wing, the second wing, the third wing, and the fourth wing of the UAV, respectively. It is the motor change control amount of wing No. 1, wing No. 2, wing No. 3 and wing No. 4 of the UAV.

3. The UAV control method based on wind turbine blade damage detection according to claim 1 is characterized in that: The UAV control network model consists of Actor network, Target Actor network, Critic1 network, Target Critic1 network, Critic2 network, and Target Critic2 network. The input of the Actor network is the state quantity s perceived by the UAV at the current moment. t The output of the Actor network is the control amount a sent to the drone motor at the current moment. t The input of the TargetActor network is the state s perceived by the drone at the next moment t ', the output of the Target Actor network is the control amount a sent to the drone motor at the next moment t '; The input of the Critic1 network and the Critic2 network is the state-action pair at the current moment (s t ,a t ), the output is the action pair according to the current state (s t ,a t ) The state value evaluation value given by the application neural network; the input of the Target Critic1 network and the Target Critic2 network is the state action pair (s t ',a t '), the output is the state action pair (s t ',a t ') Apply the state value target value given by the neural network.

4. The UAV control method based on wind turbine blade damage detection according to claim 1 is characterized in that: The calculation formula of the motor control quantity is as follows: a t =μ(s t |θ μ )+∈ ∈=clip(N(0,σ),-b,b) where s t Indicates the current state, a t is the motor control quantity, θ μ represents the Actor network parameters, μ(·) represents the Actor network, ∈ represents the random noise parameters, the clip function represents that ∈=-b when N(0,σ)<-b, and ∈=b when N(0,σ)>b, ∈=N(0,σ) represents that ∈ satisfies the normal distribution, -b and b represent fixed parameters.

5. The UAV control method based on wind turbine blade damage detection according to claim 1 is characterized in that: The priority experience replay algorithm selects the experience data from the experience buffer pool and the calculation formula is: Wherein, P(c) represents a set of experience quadruplets (s t ,a t ,r t ,s t '); c represents the sequence number of the extracted empirical data; p k represents the priority of the extracted empirical data; α represents the preset parameter used to adjust the priority sampling degree of data samples.

6. The UAV control method based on wind turbine blade damage detection according to claim 1 is characterized in that: The coordinate calculation model regards the wind turbine as rotating about the Z axis of the spatial coordinate system, calculates the rotation angle of the detection blade around the Z axis according to the rotation angular velocity, establishes a rotation matrix according to the rotation angle, and then calculates the transformed coordinates according to the rotation matrix and the initial coordinates through a coordinate conversion algorithm. The coordinate conversion algorithm calculation formula is: Among them, (x1, y1, z1) is the transformed coordinate, a is the scale parameter, b is the rotation angle, and (x, y, z) is the initial coordinate.

7. A UAV control system based on wind turbine blade damage detection, characterized in that: The method for controlling a UAV based on wind turbine blade damage detection according to any one of claims 1 to 6 comprises: a position correction module, an identification module, a data acquisition module, a control calculation module, and a detection module; The position correction module is used to control the UAV to fly relative to the wind turbine. The real-time distance between the current position of the UAV and the center of the wind turbine is obtained through the RGB-D camera carried by the UAV and transmitted to the smart terminal. The smart terminal compares the real-time distance with the preset optimal detection distance to obtain a comparison result, and corrects the UAV position according to the comparison result. The recognition module is used to obtain the wind turbine image in real time through the RGB-D camera and transmit it to the smart terminal. The smart terminal establishes a spatial coordinate system according to the wind turbine image, and calculates blade No. 1, blade No. 2, and blade No. 3 through the image detection and recognition model, and uses blade No. 1 as the detection blade; The data acquisition module is used to preset the detection point sequence of the detection blade, and obtain the initial coordinates of each detection point in the detection point sequence in the spatial coordinate system, obtain the rotation angular velocity of the detection blade, the distance data between the detection point and the center point, the instantaneous speed of the drone, the detection running speed of the drone and the current coordinates of the drone, calculate the transformed coordinates of the detection point within a preset time length according to the rotation angular velocity and the initial coordinates of the detection point through a coordinate calculation model, and calculate the coordinate difference between the current coordinates of the drone and the transformed coordinates; The control calculation module is used to input the rotation angular velocity, the UAV detection running speed, the UAV instantaneous speed, the distance data between the detection point and the center point, the motor control amount of the UAV, and the coordinate difference as state quantities into the UAV control network model to obtain the motor change control amount of the UAV through calculation; The detection module is used for the intelligent terminal to generate an adaptive signal according to the motor change control quantity to control the propeller operation state of the drone, so that the drone runs to the transformed coordinates, and uses the ultrasonic machine carried on the drone to perform defect detection on the area corresponding to the transformed coordinates.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the drone control method based on wind turbine blade damage detection as described in any one of claims 1-6 is implemented.

9. A storage medium containing computer executable instructions, characterized in that: The computer executable instructions, when executed by a computer processor, are used to execute the drone control method based on wind turbine blade damage detection as described in any one of claims 1-6.

Images