Fan blade corrosion evaluation and self-repairing method based on multi-modal fusion

Abstract

The invention discloses a multi-modal fusion-based fan blade corrosion evaluation and self-repairing method, and belongs to the technical field of fan blade state monitoring, and the method comprises the following steps: a microwave radar collects a polarization scattering matrix and meteorological data of a blade surface in real time; calculating a real corrosion index through a polarization-meteorological adaptive compensation algorithm; extracting residual error characteristics of the SCADA power curve; fusing the corrosion index and the residual feature by using a gated attention network; and generating a repair decision based on a deep reinforcement learning DRL model. According to the fan blade corrosion evaluation and self-repairing method based on multi-modal fusion, the influence of rain and fog weather on radar detection is overcome through the polarization scattering compensation model, and the misjudgment rate is reduced; fusing multi-source data to quantify the corrosion depth of the blade, and controlling a detection error; meanwhile, the repair opportunity is intelligently decided based on parameters such as dynamic electricity price and wind speed prediction, excessive maintenance is reduced, and the repair economy is optimized.

Description

Technical Field

[0001] This invention relates to the field of wind turbine blade condition monitoring technology, and in particular to a method for corrosion assessment and self-repair of wind turbine blades based on multimodal fusion. Background Technology

[0002] Wind turbine blades operate in complex natural environments for extended periods, facing erosion from salt spray, rain, and ultraviolet radiation, leading to surface corrosion and material degradation that severely impacts aerodynamic efficiency and operational safety. To achieve proactive blade maintenance, industry and academia have proposed various detection and evaluation methods, but existing technologies still have significant shortcomings.

[0003] At the detection level, current methods primarily rely on a single data source, resulting in limited early damage identification capabilities. For example, image-based methods are susceptible to interference from ambient light and dirt, making quantitative analysis difficult; SCADA data-based methods exhibit a delayed response to performance degradation, failing to identify early corrosion in a timely manner; and while radar-based detection methods can sense structural changes, they lack effective signal compensation mechanisms under meteorological conditions such as rain and fog, leading to a high false positive rate and hindering reliable application in real-world scenarios.

[0004] At the decision-making level, existing maintenance systems generally adopt static strategies based on fixed thresholds, without fully considering dynamic operating parameters such as wind speed and electricity price. This leads to unreasonable selection of maintenance timing, which may increase costs due to over-maintenance or exacerbate damage due to under-maintenance.

[0005] In summary, existing technologies have failed to effectively address the issues of accurate detection and economical repair decisions for blade corrosion in complex environments. There is an urgent need for a comprehensive solution that can integrate multi-source information, adapt to environmental changes, and take into account operational economics. Summary of the Invention

[0006] The purpose of this invention is to provide a method for corrosion assessment and self-repair of wind turbine blades based on multimodal fusion. By using modal fusion and adaptive compensation mechanisms, it overcomes the limitations of single-modal detection and environmental interference problems, and introduces dynamic economic optimization decision-making to improve detection accuracy, reliability and operation and maintenance economy.

[0007] To achieve the above objectives, this invention provides a method for corrosion assessment and self-repair of wind turbine blades based on multimodal fusion, comprising the following steps: S1. Microwave radar collects the polarization scattering matrix and meteorological data on the blade surface in real time; S2. Calculate the true corrosion index using the polarization-meteorological adaptive compensation algorithm; S3. Extract residual features from SCADA power curves; S4. A gated attention network is used to fuse the real corrosion index and residual features; S5. Generate repair decisions based on deep reinforcement learning (DRL) models.

[0008] Preferably, in S1, the polarization scattering matrix is ​​acquired by the radar system and includes the scattering coefficients of the following four polarization channels: HH: Horizontal transmission - horizontal reception; HV: Horizontal transmission - vertical reception; VH: Vertical transmit - horizontal receive; VV: Vertical transmit - vertical receive; Corrosion causes changes in the dielectric constant of the blade surface material, which in turn affects the polarization characteristics of radar echoes. Different polarization channels are sensitive to surface roughness, humidity, and corrosion depth, and can be used to quantitatively assess the corrosion status. By scanning the blade surface, a polarization image is formed, and the corrosion area can be located. Meteorological data includes humidity, rainfall intensity, temperature, salt spray concentration, wind speed, and wind direction. Rain, fog, and high humidity can significantly affect radar echo intensity, leading to misjudgments of corrosion. The polarization scattering matrix provides physical damage signals, while meteorological data provides environmental context. The combination of the two forms an environment-adaptive corrosion detection system.

[0009] Preferably, in S2, the calculation formula is: ; Where HV and HH are the cross-polarization and homopolarization scattering coefficients acquired by the microwave radar, respectively. This is the humidity attenuation factor calculated based on relative humidity (RH). Based on rainfall intensity Calculated rain intensity scattering compensation term; ε To prevent extremely small constants with a denominator of zero; by and Dynamic coupling enables adaptive environmental compensation of radar signals, allowing for the calculation of the true corrosion index; humidity attenuation factor. Rainfall intensity scattering compensation term Dynamic coupling eliminates rain and fog interference and reduces radar monitoring misjudgment rate.

[0010] Preferably, in S3, the specific steps are as follows: S31. First, a large amount of historical data on the wind turbine under healthy conditions is used to establish its baseline wind speed-power curve model. S32. In real-time monitoring, the collected real-time wind speed is input into the model to obtain the theoretical power value, and then the difference between it and the actual measured power is calculated, i.e., the power residual. S33. Extract statistical features from the residual sequence, including mean, standard deviation, and duration of negative values, to quantify the degree of aerodynamic performance degradation caused by blade corrosion. These features serve as temporal evidence of wind turbine functional degradation, complementing the spatial information of physical damage provided by radar, and are jointly input into the subsequent fusion network to achieve a more accurate assessment of corrosion status.

[0011] Preferably, in S4, the gated attention network includes a dual-channel weight generator to achieve accurate alignment between the true corrosion index and the SCADA residual features; the dual-channel weights assign spatial weights to the radar features respectively. W r Temporal weights for SCADA feature assignment W S Spatial weight W r Focusing on the distribution of damaged areas, time-series weights W S The duration of power anomalies is captured, and the dual-channel weights satisfy the following: ; ; in, W r and W S These are the spatial weight vector and temporal weight vector generated by the Sigmoid function, respectively; their calculation depends on... W r , b r and W S , b S The internal parameters of the gated attention network are determined through training; the weights are used to achieve adaptive weighted fusion of the true corrosion index and SCADA residual features; the specific process is as follows: S41. Construct a training dataset containing the true corrosion index, SCADA residual features, and corresponding true blade state labels; S42. Calculate the fusion result using forward propagation, and use the difference between the predicted result and the true label as the loss function; S43. Iteratively update all parameters in the network using the backpropagation algorithm and gradient descent optimizer, including... W r , b r and W S , b S This continues until the model converges, at which point the parameter values ​​are determined. W r, W S It acts as an adaptive information filter, focusing on the damaged area of ​​radar spatial features and capturing the performance significance of SCADA temporal features respectively; the ultimate goal is to improve the accuracy of wind turbine blade corrosion assessment and the robustness of decision-making under different environmental conditions by intelligently fusing multimodal evidence through a dynamic weighting mechanism, overcoming the limitations of fixed threshold models.

[0012] Preferably, the precise alignment of the true corrosion index with the SCADA residual characteristics includes the following steps: S401. Vectorize the spatial distribution corrosion index detected by radar and the temporal power residual features extracted by the SCADA system to construct radar feature vector and SCADA feature vector respectively. S402. Input both into the dual-channel weight generator of the gated attention network to calculate the spatial weight vector and the temporal weight vector respectively. S403. Use the obtained weights to weight each feature vector to highlight key information; The weighted features are spliced ​​and fused to form a unified feature vector that simultaneously contains information on spatial damage distribution and temporal performance degradation, thereby achieving accurate alignment and adaptive fusion of the two modal features at the decision-making level.

[0013] Preferably, step S5 specifically includes the following steps: S51. Constructing the state space The state space includes the corrosion depth. Real-time wind speed Electricity price Residual amount of repair agent The state space serves as the perceptual input to the deep reinforcement learning (DRL) agent, used to dynamically generate the optimal repair strategy. Specifically, the DRL model receives the current state vector at each decision-making time. S t Through its policy network π ( Evaluate all possible actions, namely the long-term value of immediate repair, delayed repair, and no repair, and select the action that maximizes the cumulative reward; erosion depth in the state space. Determine the urgency of repairs, real-time wind speed With real-time electricity price A joint balance between power generation revenue and downtime costs, and the remaining amount of repair agent. This constrains the physical feasibility of repair actions; ultimately, by continuously interacting with the environment to update strategies, autonomous decision-making on the economically optimal repair timing based on real-time system status is achieved. S52. Design the reward function: ; in, This represents the power increase after repair. This is the material cost coefficient per unit corrosion depth. For corrosion depth, Quantify material consumption costs, To account for power loss during repair, For repair time, For real-time electricity prices, Calculate the opportunity cost of downtime; real-time wind speed, an external parameter in the state space. With real-time electricity price Then, through decision rules, when and Time-delay repair directly controls action selection to avoid high opportunity costs. Here, by dynamically coupling all economic parameters, equipment status, and external environment, a breakthrough is achieved in repair decision-making, moving from fixed threshold-based to real-time economic optimization. The reward function serves as the optimization criterion for the deep reinforcement learning model, mathematizing the economic objective of repair decision-making. By quantifying repair benefits, direct costs, and opportunity costs, a unified value measurement scale is provided, driving the model to learn to dynamically weigh power generation revenue, maintenance costs, and downtime losses in complex environments, thereby generating an intelligent repair strategy that aims to maximize long-term cumulative economic benefits.

[0014] Preferably, when the decision is to repair immediately, a microwave trigger command format is generated: [Frequency: 2.45±0.1GHz, Power: 50-100W, Duration: 60-180s, Location Grid ID]. The frequency, power, and duration are preset in the system with the optimal process window based on the dielectric properties of the repair material and the thermodynamic properties of the blade composite material, and are automatically called when the immediate repair command is triggered. The location grid ID comes from the spatial distribution data of the polarization scattering matrix generated by microwave radar scanning the blade surface. After processing, this data accurately locates the specific location where corrosion occurs.

[0015] This invention also provides a multimodal fusion-based system for assessing and self-repairing wind turbine blade corrosion, comprising: Data acquisition module: includes microwave radar unit and meteorological sensor unit, used to collect polarization scattering matrix and environmental meteorological data of the blade surface in real time; Data processing and fusion module: used to execute the polarization-meteorological adaptive compensation algorithm to calculate the true corrosion index and extract the residual features of the power curve in the SCADA system, and then use a gated attention network to fuse the corrosion index and residual features; Intelligent decision-making module: Built-in deep reinforcement learning model, used to generate repair decisions based on fused features and system state space; Repair execution module: Includes a frequency-tunable microwave transmitter, which generates and transmits a microwave trigger signal with specific frequency, power, duration and location information after receiving an immediate repair command, in order to drive the repair process.

[0016] Therefore, this invention adopts the above-mentioned multimodal fusion-based wind turbine blade corrosion assessment and self-repair method, which overcomes the influence of rain and fog on radar detection by using a polarization scattering compensation model to reduce the false judgment rate; it quantifies the blade corrosion depth by fusing multi-source data to control detection errors; and it intelligently decides the repair timing based on parameters such as dynamic electricity price and wind speed prediction to reduce over-maintenance and optimize repair economy.

[0017] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0018] Figure 1 This is a system architecture diagram of an embodiment of a wind turbine blade corrosion assessment and self-repair method based on multimodal fusion according to the present invention; Figure 2 This is a bar chart comparing the effects of polarization-meteorological compensation algorithms in an embodiment of the wind turbine blade corrosion assessment and self-repair method based on multimodal fusion of the present invention. Figure 3 This is a response surface plot comparing the effects of the polarization-meteorological compensation algorithm in an embodiment of the wind turbine blade corrosion assessment and self-repair method based on multimodal fusion of the present invention. Figure 4 This is a diagram illustrating the multimodal fusion mechanism of an embodiment of a wind turbine blade corrosion assessment and self-repair method based on multimodal fusion according to the present invention. Detailed Implementation

[0019] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0020] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0021] Example 1 This invention provides a method for corrosion assessment and self-repair of wind turbine blades based on multimodal fusion. The system architecture and overall process are as follows: Figure 1 As shown, it includes the following steps: S1. Microwave radar acquires the polarization scattering matrix and meteorological data of the blade surface in real time. The polarization scattering matrix is ​​acquired by the radar system and includes the scattering coefficients of the following four polarization channels: HH: Horizontal transmission - horizontal reception; HV: Horizontal transmission - vertical reception; VH: Vertical transmit - horizontal receive; VV: Vertical transmit - vertical receive; Corrosion causes changes in the dielectric constant of the blade surface material, which in turn affects the polarization characteristics of radar echoes. Different polarization channels are sensitive to surface roughness, humidity, and corrosion depth, and can be used to quantitatively assess the corrosion status. By scanning the blade surface, a polarization image is formed, and the corrosion area can be located.

[0022] Meteorological data includes: humidity, rainfall intensity, temperature, salt spray concentration, wind speed, and wind direction; rain, fog, and high humidity can significantly affect radar echo intensity, leading to misjudgment of corrosion; the polarization scattering matrix provides physical damage signals, and meteorological data provides environmental context. The combination of the two forms an environment-adaptive corrosion detection.

[0023] S2. Calculate the true corrosion index using a polarization-meteorological adaptive compensation algorithm; the calculation formula is: ; Where HV and HH are the cross-polarization and homopolarization scattering coefficients acquired by the microwave radar, respectively. This is the humidity attenuation factor calculated based on relative humidity (RH). Based on rainfall intensity Calculated rain intensity scattering compensation term; ε To prevent extremely small constants with a denominator of zero; by and Dynamic coupling enables adaptive environmental compensation of radar signals, allowing for the calculation of the true corrosion index; humidity attenuation factor. Rainfall intensity scattering compensation term Dynamic coupling eliminates rain and fog interference and reduces radar monitoring misjudgment rate.

[0024] It is a variant of the sigmoid function that maps relative humidity (RH) (typically normalized to between 0 and 1) to a smooth attenuation factor in the range of 0 to 1. When humidity is normal, i.e., RH is low, A value close to 1 means almost no degradation. When humidity is extremely high, i.e., RH is close to 1, Approaching 0, the HV signal is significantly enhanced to compensate for atmospheric attenuation.

[0025] Using the hyperbolic tangent function, the rainfall intensity This is mapped to a compensation value with an upper limit. A characteristic of the tanh function is that as the input increases, the output approaches an upper limit (the coefficient 0.2 sets this upper limit). Physical law: In light rain, raindrop scattering interference increases linearly; however, in heavy rain, the interference does not increase indefinitely but tends to saturate. This value is directly subtracted from the signal to eliminate spurious echoes caused by raindrops.

[0026] This embodiment reduces the radar monitoring false alarm rate from 42% to 7.3% under typhoon conditions. , ; Specific results are as follows Figure 2 , Figure 3 As shown.

[0027] S3. Extract the residual features of the SCADA power curve. The specific process is as follows: S31. First, a large amount of historical data on the wind turbine under healthy conditions is used to establish its baseline wind speed-power curve model. S32. In real-time monitoring, the collected real-time wind speed is input into the model to obtain the theoretical power value, and then the difference between it and the actual measured power is calculated, i.e., the power residual. S33. Extract statistical features from the residual sequence, including mean, standard deviation, and duration of negative values, to quantify the degree of aerodynamic performance degradation caused by blade corrosion. These features serve as temporal evidence of wind turbine functional degradation, complementing the spatial information of physical damage provided by radar, and are jointly input into the subsequent fusion network to achieve a more accurate assessment of corrosion status.

[0028] S4. A gated attention network is used to fuse the true erosion index and residual features. The fusion mechanism is as follows: Figure 4 As shown; the gated attention network contains a dual-channel weight generator to achieve accurate alignment between the true corrosion index and SCADA residual features; Precise alignment includes the following steps: S401. Vectorize the spatial distribution corrosion index detected by radar and the temporal power residual features extracted by the SCADA system to construct radar feature vector and SCADA feature vector respectively. S402. Input both into the dual-channel weight generator of the gated attention network to calculate the spatial weight vector and the temporal weight vector respectively. S403. Use the obtained weights to weight each feature vector to highlight key information; The weighted features are spliced ​​and fused to form a unified feature vector that simultaneously contains information on spatial damage distribution and temporal performance degradation, thereby achieving accurate alignment and adaptive fusion of the two modal features at the decision-making level.

[0029] The dual-channel weights are used to assign spatial weights to radar features. W r Temporal weights for SCADA feature assignment W S Spatial weight W r Focusing on the distribution of damaged areas, time-series weights W S The duration of power anomalies is captured, and the dual-channel weights satisfy the following: ; .

[0030] in, W r and W S These are the spatial weight vector and temporal weight vector generated by the Sigmoid function, respectively; their calculation depends on... W r , b r and W S , b S The internal parameters of the gated attention network are determined through training; the weights are used to achieve adaptive weighted fusion of the true corrosion index and SCADA residual features; the specific process is as follows: S41. Construct a training dataset containing the true corrosion index, SCADA residual features, and corresponding true blade state labels; S42. Calculate the fusion result using forward propagation, and use the difference between the predicted result and the true label as the loss function; S43. Iteratively update all parameters in the network using the backpropagation algorithm and gradient descent optimizer, including... W r , b r and W S , b S This continues until the model converges, at which point the parameter values ​​are determined. W r , W SIts function is to act as an adaptive information filter, focusing on the damaged area of ​​radar spatial features and capturing the performance significance of SCADA temporal features respectively; the ultimate goal is to significantly improve the accuracy of wind turbine blade corrosion assessment and decision robustness under different environmental conditions by intelligently fusing multimodal evidence through a dynamic weighting mechanism, breaking through the limitations of fixed threshold models.

[0031] S5. Generate repair decisions based on a deep reinforcement learning (DRL) model. This includes the following steps: S51. Constructing the state space The state space includes the corrosion depth. Real-time wind speed Electricity price Residual amount of repair agent The state space serves as the perceptual input to the deep reinforcement learning (DRL) agent, used to dynamically generate the optimal repair strategy. Specifically, the DRL model receives the current state vector at each decision-making time. S t Through its policy network π ( Evaluate all possible actions, namely the long-term value of immediate repair, delayed repair, and no repair, and select the action that maximizes the cumulative reward; erosion depth in the state space. Determine the urgency of repairs, real-time wind speed With real-time electricity price A joint balance between power generation revenue and downtime costs, and the remaining amount of repair agent. This constrains the physical feasibility of repair actions; ultimately, by continuously interacting with the environment to update strategies, autonomous decision-making on the economically optimal repair timing based on real-time system status is achieved. S52. Design the reward function: ; in, This represents the power increase after repair. This is the material cost coefficient per unit corrosion depth. For corrosion depth, Quantify material consumption costs, To account for power loss during repair, For repair time, For real-time electricity prices, Calculate the opportunity cost of downtime; real-time wind speed, an external parameter in the state space. With real-time electricity price Then, through decision rules, when and Time-delay repair directly controls action selection to avoid high opportunity costs. Here, by dynamically coupling all economic parameters, equipment status, and external environment, a breakthrough is achieved in repair decision-making, moving from fixed threshold-based to real-time economic optimization. The reward function serves as the optimization criterion for the deep reinforcement learning model, mathematizing the economic objective of repair decision-making. By quantifying repair benefits, direct costs, and opportunity costs, it provides a unified value measurement scale, driving the model to learn to dynamically weigh power generation revenue, maintenance expenses, and downtime losses in complex environments, thereby generating an intelligent repair strategy that aims to maximize long-term cumulative economic benefits.

[0032] When the decision is to repair immediately, a microwave trigger command format is generated: [Frequency: 2.45±0.1GHz, Power: 50-100W, Duration: 60-180s, Location Grid ID]. The frequency, power, and duration are preset in the system with an optimal process window based on the dielectric properties of the repair material and the thermodynamic properties of the blade composite material, and are automatically invoked when the immediate repair command is triggered. The location grid ID is derived from the spatial distribution data of the polarization scattering matrix generated by microwave radar scanning the blade surface. After processing, this data accurately locates the specific location where corrosion occurs.

[0033] The method described in this embodiment is based on a multimodal fusion wind turbine blade corrosion assessment and self-healing system, which includes: Data acquisition module: includes microwave radar unit and meteorological sensor unit, used to collect polarization scattering matrix and environmental meteorological data of the blade surface in real time; Data processing and fusion module: used to execute the polarization-meteorological adaptive compensation algorithm to calculate the true corrosion index and extract the residual features of the power curve in the SCADA system, and then use a gated attention network to fuse the corrosion index and residual features; Intelligent decision-making module: Built-in deep reinforcement learning model, used to generate repair decisions based on fused features and system state space; Repair execution module: Includes a frequency-tunable microwave transmitter, which generates and transmits a microwave trigger signal with specific frequency, power, duration and location information after receiving an immediate repair command, in order to drive the repair process.

[0034] Therefore, this invention adopts the above-mentioned multimodal fusion-based wind turbine blade corrosion assessment and self-repair method, which overcomes the influence of rain and fog on radar detection by using a polarization scattering compensation model to reduce the false judgment rate; it quantifies the blade corrosion depth by fusing multi-source data to control detection errors; and it intelligently decides the repair timing based on parameters such as dynamic electricity price and wind speed prediction to reduce over-maintenance and optimize repair economy.

[0035] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for corrosion assessment and self-repair of wind turbine blades based on multimodal fusion, characterized in that, Includes the following steps: S1. Microwave radar collects the polarization scattering matrix and meteorological data on the blade surface in real time; S2. Calculate the true corrosion index using the polarization-meteorological adaptive compensation algorithm; S3. Extract residual features from SCADA power curves; S4. A gated attention network is used to fuse the real corrosion index and residual features; S5. Generate repair decisions based on deep reinforcement learning (DRL) models.

2. The method for corrosion assessment and self-repair of wind turbine blades based on multimodal fusion according to claim 1, characterized in that: In S1, the polarization scattering matrix is ​​acquired by the radar system and includes the scattering coefficients of the following four polarization channels: HH: Horizontal transmission - horizontal reception; HV: Horizontal transmission - vertical reception; VH: Vertical transmit - horizontal receive; VV: Vertical transmit - vertical receive; Corrosion causes changes in the dielectric constant of the blade surface material, which in turn affects the polarization characteristics of radar echoes. Different polarization channels are sensitive to surface roughness, humidity, and corrosion depth, and are used to quantitatively assess the corrosion status. By scanning the blade surface, a polarization image is formed to locate the corrosion area. Meteorological data includes humidity, rainfall intensity, temperature, salt spray concentration, wind speed, and wind direction. Rain, fog, and high humidity can significantly affect radar echo intensity, leading to misjudgments of corrosion. The polarization scattering matrix provides physical damage signals, while meteorological data provides environmental context. The combination of the two forms an environment-adaptive corrosion detection system.

3. The method for corrosion assessment and self-repair of wind turbine blades based on multimodal fusion according to claim 2, characterized in that: In S2, the calculation formula is: ； Where HV and HH are the cross-polarization and homopolarization scattering coefficients acquired by the microwave radar, respectively. This is the humidity attenuation factor calculated based on relative humidity (RH). Based on rainfall intensity Calculated rain intensity scattering compensation term; ε To prevent extremely small constants with a denominator of zero; through the aforementioned and Dynamic coupling enables adaptive environmental compensation of radar signals, thereby calculating the true corrosion index; humidity attenuation factor. Rainfall intensity scattering compensation term Dynamic coupling eliminates rain and fog interference and reduces radar monitoring misjudgment rate.

4. The method for corrosion assessment and self-repair of wind turbine blades based on multimodal fusion according to claim 3, characterized in that: In S3, the specific steps are as follows: S31. Establish a baseline wind speed-power curve model using historical data from the wind turbine's healthy state. S32. In real-time monitoring, the collected real-time wind speed is input into the model to obtain the theoretical power value, and then the difference between it and the actual measured power is calculated, i.e., the power residual. S33. Extract statistical features from the residual sequence, including mean, standard deviation, and duration of negative values, to quantitatively characterize the degree of aerodynamic performance degradation caused by blade corrosion. These features, as temporal evidence reflecting the functional degradation of wind turbines, complement the spatial information on physical damage provided by radar, and are jointly input into the subsequent fusion network to achieve a more accurate assessment of corrosion status.

5. The method for corrosion assessment and self-repair of wind turbine blades based on multimodal fusion according to claim 4, characterized in that: In S4, the gated attention network includes a dual-channel weight generator to achieve accurate alignment between the true corrosion index and SCADA residual features; the dual-channel weights assign spatial weights to the radar features. W r Temporal weights for SCADA feature assignment W S Spatial weight W r Focusing on the distribution of damaged areas, time-series weights W S The duration of power anomalies is captured, and the dual-channel weights satisfy the following: ； ； in, W r and W S These are the spatial weight vector and temporal weight vector generated by the Sigmoid function, respectively; their calculation depends on... W r , b r and W S , b S The internal parameters of the gated attention network are determined through training; the weights are used to achieve adaptive weighted fusion of the true corrosion index and SCADA residual features; the specific process is as follows: S41. Construct a training dataset containing the true corrosion index, SCADA residual features, and corresponding true blade state labels; S42. Calculate the fusion result using forward propagation, and use the difference between the predicted result and the true label as the loss function; S43. Iteratively update all parameters in the network using the backpropagation algorithm and gradient descent optimizer, including... W r , b r and W S , b S This continues until the model converges, at which point the parameter values ​​are determined. W r , W S It acts as an adaptive information filter, focusing on the damaged area of ​​radar spatial features and capturing the performance significance of SCADA temporal features respectively; the ultimate goal is to improve the accuracy of wind turbine blade corrosion assessment and the robustness of decision-making under different environmental conditions by intelligently fusing multimodal evidence through a dynamic weighting mechanism, overcoming the limitations of fixed threshold models.

6. The method for corrosion assessment and self-repair of wind turbine blades based on multimodal fusion according to claim 5, characterized in that: Accurate alignment of the true corrosion index with SCADA residual characteristics includes the following steps: S401. Vectorize the spatial distribution corrosion index detected by radar and the temporal power residual features extracted by the SCADA system to construct radar feature vector and SCADA feature vector respectively. S402. Input both into the dual-channel weight generator of the gated attention network to calculate the spatial weight vector and the temporal weight vector respectively. S403. Use the obtained weights to weight each feature vector to highlight key information; The weighted features are spliced ​​and fused to form a unified feature vector that simultaneously contains information on spatial damage distribution and temporal performance degradation, thereby achieving accurate alignment and adaptive fusion of the two modal features at the decision-making level.

7. The method for corrosion assessment and self-repair of wind turbine blades based on multimodal fusion according to claim 6, characterized in that: S5 specifically includes the following steps: S51. Constructing the state space The state space includes the corrosion depth. Real-time wind speed Electricity price Residual amount of repair agent The state space serves as the perceptual input to the deep reinforcement learning (DRL) agent, used to dynamically generate the optimal repair strategy. Specifically, the DRL model receives the current state vector at each decision-making time. S t Through its policy network π ( Evaluate all possible actions, namely the long-term value of immediate repair, delayed repair, and no repair, and select the action that maximizes the cumulative reward; erosion depth in the state space. Determine the urgency of repairs, real-time wind speed With real-time electricity price A joint balance between power generation revenue and downtime costs, and the remaining amount of repair agent. This constrains the physical feasibility of repair actions; ultimately, by continuously interacting with the environment to update strategies, autonomous decision-making on the economically optimal repair timing based on real-time system status is achieved. S52. Design the reward function: ； in, This represents the power increase after repair. This is the material cost coefficient per unit corrosion depth. For corrosion depth, Quantify material consumption costs, To account for power loss during repair, For repair time, For real-time electricity prices, Calculate the opportunity cost of downtime; real-time wind speed, an external parameter in the state space. With real-time electricity price Then, through decision rules, when and Time-delay repair directly controls action selection to avoid high opportunity costs. Here, by dynamically coupling all economic parameters, equipment status, and external environment, a breakthrough is achieved in repair decision-making, moving from fixed threshold-based to real-time economic optimization. The reward function serves as the optimization criterion for the deep reinforcement learning model, mathematizing the economic objective of repair decision-making. By quantifying repair benefits, direct costs, and opportunity costs, a unified value metric is provided, driving the model to learn to dynamically weigh power generation revenue, maintenance costs, and downtime losses in complex environments, thereby generating an intelligent repair strategy that aims to maximize long-term cumulative economic benefits.

8. The method for corrosion assessment and self-repair of wind turbine blades based on multimodal fusion according to claim 7, characterized in that: When the decision is to repair immediately, a microwave trigger command format is generated: [Frequency: 2.45±0.1GHz, Power: 50-100W, Duration: 60-180s, Location Grid ID]. The frequency, power, and duration are preset in the system with the optimal process window based on the dielectric properties of the repair material and the thermodynamic properties of the blade composite material. When the immediate repair command is triggered, the optimal process window is automatically called. The location grid ID comes from the spatial distribution data of the polarization scattering matrix generated by microwave radar scanning the blade surface. After processing, this data accurately locates the specific location where corrosion occurs.

9. A wind turbine blade corrosion assessment and self-repair system based on multimodal fusion, characterized in that, include: Data acquisition module: includes microwave radar unit and meteorological sensor unit, used to collect polarization scattering matrix and environmental meteorological data of the blade surface in real time; Data processing and fusion module: used to execute the polarization-meteorological adaptive compensation algorithm to calculate the true corrosion index and extract the residual features of the power curve in the SCADA system, and then use a gated attention network to fuse the corrosion index and residual features; Intelligent decision-making module: Built-in deep reinforcement learning model, used to generate repair decisions based on fused features and system state space; Repair execution module: Includes a frequency-tunable microwave transmitter, which generates and transmits a microwave trigger signal with specific frequency, power, duration and location information after receiving an immediate repair command, in order to drive the repair process.

Images