A remote, non-contact, photothermal de-icing method and system for the surface of a wind turbine blade

Abstract

The application provides a remote non-contact photo-thermal deicing method for a wind turbine blade surface, which comprises: collecting ice condition data of the wind turbine blade surface to obtain multi-modal data; identifying features of the multi-modal data to obtain ice condition features; constructing a three-dimensional thickness distribution model based on the ice condition features and combining natural light factors to generate a deicing strategy; performing laser deicing according to the deicing strategy and adjusting and controlling the laser beam through a bee swarm focusing algorithm; monitoring the temperature change of the coating on the surface of the wind turbine blade in real time to adjust the laser power; detecting the deicing rate of the deiced wind turbine blade surface, and if the deicing rate does not meet the standard, re-performing the deicing operation until the deicing rate meets the standard. Through the fusion of multispectral imaging, millimeter wave radar, double-flow neural network and dynamic environment modeling technology, the application realizes accurate identification of ice conditions and adaptive laser deicing control, and reduces the dependence on natural light while reducing energy consumption.

Description

Technical Field

[0001] This invention relates to the field of wind power equipment stable operation and maintenance technology, and in particular to a remote non-contact photothermal de-icing method and system for wind turbine blade surfaces. Background Technology

[0002] Wind energy is a clean and renewable energy source with large reserves and wide distribution. However, wind turbines built in high-altitude and cold regions often encounter severe icing problems. Icing on the surface of wind turbine blades alters their aerodynamic performance, leading to decreased power generation efficiency. In severe cases, it can cause unplanned shutdowns, blade breakage, and other accidents. Furthermore, ice fragments falling from the blades pose a safety threat to nearby personnel and equipment. Therefore, solving the problem of icing on wind turbine blades is of great significance for the stable and efficient operation and maintenance of wind power equipment.

[0003] Current active de-icing technologies applied to wind turbine blade surfaces, such as electrothermal de-icing and hot gas de-icing, achieve de-icing by heating the blade surface with external energy input. These technologies suffer from drawbacks such as high energy consumption, complex systems (requiring additional heating, temperature control, and air supply devices), and expensive retrofitting costs, making them unsuitable for long-term application in large-scale wind farms. While passive de-icing technologies based on surface property regulation can achieve de-icing without external energy input, once ice accumulates on the superhydrophobic coating, the material's inherent properties alone are insufficient for rapid and efficient self-removal. Although photothermal conversion materials can efficiently absorb sunlight and convert it into heat energy to prevent or melt ice on the blade surface, their de-icing efficiency is highly dependent on natural light conditions. On cloudy days or at night, the lack of light sources leads to insufficient energy absorption by the coating, resulting in slow ice melting or even de-icing failure. Therefore, designing a remote, non-contact photothermal de-icing method and system for wind turbine blade surfaces is essential. Summary of the Invention

[0004] The purpose of this invention is to provide a remote non-contact photothermal de-icing method and system for wind turbine blade surfaces. By integrating multispectral imaging, millimeter-wave radar, dual-stream neural networks and dynamic environment modeling technology, it can achieve accurate identification of ice conditions and adaptive laser de-icing control, while reducing energy consumption and eliminating dependence on natural light.

[0005] To achieve the above objectives, the present invention provides the following solution:

[0006] A remote, non-contact photothermal de-icing method for wind turbine blade surfaces includes the following steps:

[0007] Ice condition data on the surface of wind turbine blades was collected using a multispectral camera and millimeter-wave radar to obtain multimodal data, which includes multispectral image data and radar echo data.

[0008] Ice condition features are obtained by performing feature recognition on multimodal data using a two-stream convolutional neural network. The ice condition features include: ice type classification results and ice layer thickness.

[0009] Based on ice condition characteristics and natural lighting factors, a three-dimensional thickness distribution model is constructed, and a de-icing strategy is generated.

[0010] Laser de-icing is performed according to the de-icing strategy, and the laser beam is adjusted and controlled by a swarm focusing algorithm.

[0011] Real-time monitoring of coating temperature changes on the surface of wind turbine blades, and adjustment of laser power based on coating temperature;

[0012] After de-icing, the de-icing rate of the wind turbine blades is tested. If the de-icing rate does not meet the preset de-icing standard, the de-icing operation is repeated until the de-icing rate meets the preset de-icing standard.

[0013] Optionally, ice condition data can be collected on the surface of the wind turbine blades using a multispectral camera and millimeter-wave radar to obtain multimodal data, including:

[0014] The surface of the wind turbine blades was scanned in visible light and short-wave infrared bands using a multispectral camera to obtain raw image data.

[0015] Radiometric calibration is performed on the original image data to obtain radiance image data;

[0016] Atmospheric scattering correction is performed on the obtained radiance image data to obtain surface reflectance image data;

[0017] Multispectral image data is obtained by performing multiband spatial registration on surface reflectance image data using the scale-invariant feature transform algorithm.

[0018] Optionally, multi-spectral data on the surface of wind turbine blades is collected using a multispectral camera and millimeter-wave radar to obtain multi-modal data, which also includes:

[0019] The original signal is obtained by transmitting electromagnetic wave signals onto the surface of the wind turbine blades using millimeter-wave radar.

[0020] The transmitted electromagnetic wave signal and the original signal are coherently mixed to obtain the difference frequency signal;

[0021] Radar echo data is obtained by performing time-frequency decomposition on the difference frequency signal using wavelet packet transform.

[0022] Optionally, feature recognition is performed on the multimodal data using a two-stream convolutional neural network to obtain ice condition features, including:

[0023] Ice texture features of multimodal data are extracted using the first sub-network of a dual-stream convolutional neural network;

[0024] Spectral absorption features of multimodal data are extracted using the second sub-network of a two-stream convolutional neural network;

[0025] The ice texture features and spectral absorption features are compared with the preset ice and snow state feature database to obtain the ice type classification results.

[0026] The ice thickness is calculated based on the time delay difference of radar echo data.

[0027] Optionally, based on ice condition characteristics and combined with natural lighting factors, a three-dimensional thickness distribution model is constructed, and a de-icing strategy is generated, including:

[0028] The current ambient light intensity is sampled at multiple points by an irradiance sensor array to obtain a real-time light distribution map;

[0029] An effective illumination intensity map is obtained by performing attenuation compensation on the real-time illumination distribution map using an atmospheric transmittance model.

[0030] Finite element mesh generation is performed on the surface of the wind turbine blades to obtain the element mesh;

[0031] The predicted ice growth rate is obtained by performing time-series analysis on meteorological data provided by the weather station using an LSTM neural network.

[0032] The predicted thickness distribution of the cell grid is obtained based on the predicted values ​​of ice thickness and ice growth rate.

[0033] The thickness prediction distribution is reconstructed in three dimensions by fitting a non-uniform rational B-spline surface to obtain a three-dimensional thickness distribution model.

[0034] Laser action parameters are generated based on a three-dimensional thickness distribution model, and conditions are imposed on these parameters to obtain a de-icing strategy.

[0035] Optionally, laser de-icing is performed according to a de-icing strategy, and the laser beam is adjusted and controlled using a swarm focusing algorithm, including:

[0036] The optimal focusing sequence of the main beam is obtained by using an adaptive particle swarm optimization algorithm to plan the path of the cell grid.

[0037] Real-time temperature monitoring of the ice layer and identification of points of sudden temperature drop;

[0038] An auxiliary beam is generated based on the temperature drop point, and the laser power of the auxiliary beam is adjusted according to the real-time temperature.

[0039] By collecting the vibration spectrum of the wind turbine blade surface, vibration compensation and motion prediction are performed on the wind turbine blade surface to locate the position of the unit grid in real time.

[0040] Optionally, the temperature change of the coating on the wind turbine blade surface is monitored in real time, and the laser power is adjusted according to the coating temperature, including:

[0041] When the coating temperature is lower than the preset first temperature threshold, the laser power is increased to the reference power;

[0042] When the coating temperature reaches the first temperature threshold and is lower than the preset second temperature threshold, the laser power is maintained within the fluctuation range of the reference power, and the laser power is finely adjusted according to the temperature gradient of the coating temperature.

[0043] When the coating temperature exceeds the second temperature threshold, the laser power is reduced to a preset safe power, and the pulse mode is activated.

[0044] Optionally, the de-icing rate of the wind turbine blades is tested after de-icing. If the de-icing rate does not meet the preset de-icing standard, the de-icing operation is repeated until the de-icing rate meets the preset de-icing standard, including:

[0045] Thermal radiation scanning was performed on the surface of the wind turbine blades after de-icing to obtain a temperature distribution map of residual ice.

[0046] The dielectric constant distribution of the wind turbine blades after de-icing was obtained by scanning the dielectric constant of the blade surface using millimeter-wave radar.

[0047] The residual ice volume is obtained by fusing the residual ice temperature distribution map and the complex permittivity distribution using DS evidence theory.

[0048] The de-icing rate is calculated based on the volume of residual ice.

[0049] A remote, non-contact photothermal de-icing system for wind turbine blade surfaces includes:

[0050] The data acquisition module is used to collect ice condition data on the surface of wind turbine blades using a multispectral camera and millimeter-wave radar to obtain multimodal data, including multispectral image data and radar echo data.

[0051] The feature extraction module is used to perform feature recognition on multimodal data through a two-stream convolutional neural network to obtain ice condition features; the ice condition features include: ice type classification results and ice layer thickness;

[0052] The strategy generation module is used to construct a three-dimensional thickness distribution model based on ice condition characteristics and natural lighting factors, and generate de-icing strategies.

[0053] The de-icing execution module is used to perform laser de-icing operations according to the de-icing strategy and to adjust and control the laser beam through a swarm focusing algorithm.

[0054] The laser adjustment module is used to monitor the temperature change of the coating on the surface of the wind turbine blades in real time and adjust the laser power according to the coating temperature.

[0055] The detection module is used to detect the de-icing rate on the surface of the wind turbine blades after de-icing. If the de-icing rate does not meet the preset de-icing standard, the de-icing operation is repeated until the de-icing rate meets the preset de-icing standard.

[0056] According to specific embodiments provided by the present invention, the following technical effects are disclosed: The remote non-contact photothermal de-icing method for wind turbine blade surfaces provided by the present invention includes: acquiring ice condition data of the wind turbine blade surface using a multispectral camera and millimeter-wave radar to obtain multimodal data; the multimodal data includes: multispectral image data and radar echo data; performing feature recognition on the multimodal data using a dual-stream convolutional neural network to obtain ice condition features; the ice condition features include: ice type classification results and ice layer thickness; constructing a three-dimensional thickness distribution model based on the ice condition features and combining natural light factors, and generating a de-icing strategy; performing laser de-icing operation according to the de-icing strategy, and adjusting and controlling the laser beam using a bee swarm focusing algorithm; monitoring the coating temperature change on the wind turbine blade surface in real time, and adjusting the laser power according to the coating temperature; detecting the de-icing rate on the de-iced wind turbine blade surface, and if the de-icing rate does not reach the preset de-icing standard, performing the de-icing operation again until the de-icing rate reaches the preset de-icing standard. This method integrates multispectral imaging, millimeter-wave radar, dual-stream neural networks, and dynamic environment modeling techniques to achieve accurate ice condition identification and adaptive laser de-icing control, while reducing energy consumption and eliminating dependence on natural light. Attached Figure Description

[0057] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0058] Figure 1 This is a flowchart of the remote non-contact photothermal de-icing method of the present invention;

[0059] Figure 2 This is a flowchart of the laser beam adjustment and control process of the present invention. Detailed Implementation

[0060] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0061] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0062] like Figure 1 As shown, the present invention provides a remote non-contact photothermal de-icing method for the surface of wind turbine blades, comprising the following steps:

[0063] Step 100: Collect ice condition data on the surface of the wind turbine blades using a multispectral camera and millimeter-wave radar to obtain multimodal data; the multimodal data includes: multispectral image data and radar echo data;

[0064] Specifically, the multispectral camera is equipped with dual-band sensors for visible light and short-wave infrared, which scan the rotating blade surface line by line using a high-precision servo mechanism. The visible light band is used to capture the contours and color features of the ice layer and the blade base, while the short-wave infrared band responds to the characteristic absorption peaks of ice / water molecules at 1520 nm and 1950 nm, thereby detecting the wet ice hidden on the surface. The raw image data generated by the scan is a digital matrix containing spatial coordinates and band intensity information.

[0065] Next, radiometric calibration was performed on the original image data using a pre-established calibration parameter library to obtain radiance image data. Dark current noise data acquired by the multispectral camera under a standard light source, nonlinear response curves for each band, and sensor gain coefficients were incorporated into the radiative transfer equation to convert the pixel grayscale values ​​of the original image into radiance values ​​with physical dimensions. This eliminated the influence of sensor noise and vignetting effects from the optical system, ensuring the physical consistency of subsequent ice condition inversion.

[0066] Then, atmospheric scattering correction was performed on the radiance image. The MODTRAN atmospheric radiative transfer model was used to solve for the scattering and absorption of downward solar radiation by the atmosphere and the superposition effect of upward path radiation, thereby constructing an atmospheric influence factor lookup table. This table was then used to perform pixel-by-pixel compensation on the radiance image, finally outputting surface reflectance image data to eliminate environmental interference such as clouds and haze, highlighting the spectral characteristics of the leaf surface itself.

[0067] Finally, the Scale Invariant Feature Transform (SIFT) algorithm was used to spatially register the multi-band surface reflectance images. The SIFT algorithm extracts stable features such as blade leading-edge bolts or coating defects from images of different bands as key points, calculates their gradient direction histogram descriptors, uses the RANSAC algorithm to filter matching point pairs, constructs an affine transformation matrix based on the matching results, and accurately aligns the shortwave infrared band images to the visible light band reference coordinate system to generate spatially consistent multispectral image data.

[0068] It should be noted that by combining short-wave infrared characteristic absorption bands, it is possible to penetrate the surface to detect wet ice and layered ice, and to identify the internal ice structure. Radiometric calibration and atmospheric correction eliminate environmental noise and reduce ice reflectivity inversion errors. SIFT registration technology solves motion blur and multi-band misalignment problems, improving the spatial registration accuracy of the image.

[0069] More specifically, the millimeter-wave radar employs a frequency-modulated continuous wave (FMCV) system with a center frequency of 94 GHz. The radar transmitter generates a linearly frequency-modulated signal via a voltage-controlled oscillator (VCO), which is then amplified by a power amplifier and directionally projected onto the blade surface by a conical lens antenna. The electromagnetic waves are reflected at the ice-blade interface and the ice-air interface, and the receiving array antenna captures the raw echo signal containing phase and amplitude information. A dual-balanced diode structure is then used to coherently mix the transmitted and received raw signals, using the transmitted signal as the local oscillator reference signal and nonlinearly superimposing it with the raw echo signal. Due to the Doppler shift caused by blade motion, a difference frequency signal containing range information appears after mixing. This difference frequency signal is then adaptively decomposed into a full subtree using a Daubechies-5 wavelet basis with 6-8 layers, generating radar echo data.

[0070] It should be noted that non-contact millimeter-wave detection has strong penetration capabilities and is not limited by the type of blade coating. Coherent mixing can convert distance information into easily processed frequency domain signals, thereby improving the signal-to-noise ratio. Wavelet packet transform overcomes the time-frequency resolution limitations of FFT, reducing the false positive rate of ice layer identification.

[0071] Step 200: Perform feature recognition on multimodal data using a dual-stream convolutional neural network to obtain ice condition features; ice condition features include: ice type classification results and ice layer thickness;

[0072] Specifically, a heterogeneous dual-stream convolutional neural network is first constructed, consisting of two parallel sub-networks. The first sub-network uses a ResNet-50 backbone network with dilated convolutions. Its input is spatially registered multispectral image data, with an image size of 512×512×6, containing three bands: visible light (RGB) and shortwave infrared. The first sub-network captures local texture features such as ice cracks and bubbles in the image through shallow 7×7 convolutional kernels, and then expands the receptive field through mid-level dilated convolutions with an inflation rate of 2 to identify areas such as frost ice crystal clusters or wet ice seepage zones. Finally, global average pooling outputs a 1024-dimensional high-semantic feature vector to characterize the texture features of the ice layer on the leaf surface.

[0073] The second sub-network is a fully 1D convolutional structure, with its input being the dimension-reduced spectral reflectance curve. This sub-network contains four convolutional blocks. The first layer has a wide convolutional kernel with a width of 15, used to extract the wide absorption valleys of ice / water molecules in the 1420-1520nm and 1900-2050nm ranges. Subsequent layers use progressively smaller convolutional kernels with widths of 7, 3, and 1, respectively, to gradually focus on ice-sensitive feature bands, such as the reflectance abrupt change point at 1310nm in transparent ice, ultimately outputting the spectral absorption characteristics.

[0074] Then, the spectral feature vectors are projected onto the texture feature space using a learnable affine transformation matrix. A gated fusion mechanism is used to obtain the feature weights of the output features of the two sub-networks, and the two features are fused using a sigmoid function. The fused joint features are then compared with a preset ice and snow state feature library, which contains template vectors of typical ice types and frost. The matching score is calculated using cosine similarity, and the category with the highest score is the ice type classification result. In some embodiments, the identified joint features are white granular or feathery, and the classification result is hoarfrost; features with a smooth and hard surface and a glassy appearance are identified as rime; features with a hard outer layer and a loose inner layer are identified as mixed rime; needle-like or scaly features are identified as frost; and classification results for other irregular ice coverings such as wet snow, fan-shaped ice, and icicles are also included.

[0075] Finally, based on radar echo data, the time delay difference sequence generated by wavelet packet transform was used to select the echo time delays of the ice-coating interface and the ice-air interface, and then the formula was applied. The ice thickness was calculated. Where c is the speed of light, Δt is the echo delay, and εᵣ is the dielectric constant of the ice layer. This dielectric constant is dynamically selected based on the ice type classification results, such as 3.17 for transparent ice and 5.3-7.1 for wet ice.

[0076] It should be noted that traditional single-modal detection methods, such as infrared thermal imagers, cannot distinguish visually similar ice patterns. In contrast, the dual-stream network of this invention can simultaneously capture spatial texture and spectral absorption fingerprints, improving the accuracy of mixed ice identification. Furthermore, dynamic dielectric constant correction achieves coupling between radar delay data and optical ice pattern classification results, reducing errors in ice thickness measurement.

[0077] Step 300: Based on ice condition characteristics and natural lighting factors, construct a three-dimensional thickness distribution model and generate a de-icing strategy;

[0078] Specifically, firstly, ambient light field data is collected using an irradiance sensor array. The solar azimuth angle is then calculated using GPS positioning and astronomical algorithms, generating a real-time illumination distribution map with a spatial resolution of 1m × 1m. Next, based on atmospheric pressure, humidity, and aerosol concentration data from a weather station, and combined with the 550nm atmospheric optical thickness obtained from a CE-318 solar photometer, the real-time illumination distribution map is converted into an effective light intensity map received by the leaf surface using a solar spectral model. The expression for the solar spectral model is: ,in For effective irradiance, It is direct radiation. It is scattered radiation. Atmospheric transmittance. The zenith angle of the sun. The blade surface tilt angle.

[0079] Next, an adaptive quadtree algorithm was used to perform finite element mesh generation on the wind turbine blade surface, resulting in a curvature-sensitive triangular mesh. The mesh size at the leading edge was 2cm × 2cm, while the size at the trailing edge was expanded to 5cm × 5cm. Subsequently, an LSTM neural network was used to predict ice growth trends based on meteorological data such as temperature, humidity, wind speed, and liquid water content provided by the weather station. This neural network employed a three-layer gating structure with 128 hidden units and was trained using historical observation data from 2000 sets of icing events. The final output was the predicted ice growth rate for each mesh unit over the next 10 minutes, expressed as: ,in Ice growth prediction values The local collection coefficient of the leaf is , For wind speed, Liquid water content, This is a temperature-dependent phase transition function. Humidity.

[0080] The current ice thickness is then added to the predicted ice growth rate to predict the thickness distribution for each cell grid. Thickness control points are then inserted at the vertices of the cell grid using non-uniform rational B-splines (NURBS). Simultaneously, control point influence factors are set based on radar measurement confidence levels, with regions having a signal-to-noise ratio (SNR) > 20 dB weighted at 1 and regions with an SNR < 10 dB weighted at 0.3. The three-dimensional thickness distribution model is obtained by solving the NURBS basis function tensor product equations.

[0081] Finally, the laser energy density was calculated based on the three-dimensional thickness distribution model. The calculation formula is as follows: ,in For energy density, The density of the ice layer, The latent heat of fusion is 334 kJ / kg in some embodiments. The current grid temperature, For coating absorption rate, The maximum laser power is set. Conditional constraints are added, such as the coating's safe temperature not exceeding 80℃, ultimately yielding de-icing strategies for different cell grids.

[0082] It should be noted that by fusing real-time atmospheric transmission correction with LSTM time-series prediction, the prediction accuracy of ice thickness was improved. The use of NURBS surface fitting overcame the limitations of grid discretization, significantly reducing reconstruction errors.

[0083] Step 400: Perform laser de-icing according to the de-icing strategy, and adjust and control the laser beam using a swarm focusing algorithm; specific steps are as follows. Figure 2 As shown, it includes:

[0084] Step 401: Path planning is performed on the cell grid using an adaptive particle swarm optimization algorithm to obtain the optimal focusing sequence of the main beam;

[0085] Specifically, the main beam path is planned based on an adaptive particle swarm optimization algorithm. The particle swarm state is initialized using the unit mesh divided by the three-dimensional thickness distribution model as the target points. The fitness function expression of this algorithm is: ,in , and These are the weighting coefficients. This is the ratio of theoretical melting energy consumption to actual input laser energy. The ratio of the scanned grid to the total ice layer grid is used to obtain the optimal grid focusing sequence for the main beam during de-icing operations.

[0086] Step 402: Monitor the temperature of the ice layer in real time and identify the point of sudden temperature drop;

[0087] Specifically, the temperature field of the ice layer is monitored in real time using an infrared thermal imager, a temperature distribution heat map is generated at a frame rate of 10Hz, and the point of temperature drop is detected by the Sobel gradient operator.

[0088] Step 403: Generate an auxiliary beam based on the temperature drop point, and adjust the laser power of the auxiliary beam according to the real-time temperature.

[0089] Specifically, 2-4 independently controllable auxiliary beams are generated. These auxiliary beams are positioned to the target area based on the coordinates of the temperature drop point, and their laser power is adjusted according to the real-time temperature. The adjustment formula is as follows:

[0090] ;

[0091] in It is 8W / ℃. It is 0.5 W / (℃·s). This is the baseline melting temperature corresponding to the current ice thickness. This is the current temperature.

[0092] Step 404: Vibration compensation and motion prediction are performed on the wind turbine blade surface by collecting the vibration spectrum of the wind turbine blade surface, so as to locate the position of the unit grid in real time.

[0093] Specifically, the blade vibration spectrum is generated based on the first-order flapping frequency and second-order oscillation frequency extracted from the radar echo data through FFT transformation. Then, using an ARIMA time series model, the blade deformation for the next 0.5 seconds is predicted based on the blade vibration data of the previous 5 seconds, and the target cell mesh is located and tracked in real time based on the predicted deformation to ensure continuous locking of the mesh center.

[0094] It should be noted that the optimal design of the de-icing path was achieved through an adaptive particle swarm optimization algorithm, abandoning the traditional fixed scanning path. Combined with the rapid response of the auxiliary beam, the de-icing efficiency was significantly improved. At the same time, real-time motion compensation for the de-icing process through blade rotation vibration reduced positioning errors.

[0095] Step 500: Monitor the temperature change of the coating on the surface of the wind turbine blades in real time and adjust the laser power according to the coating temperature;

[0096] Specifically, when the coating temperature is below a preset first temperature threshold, the laser power is increased to a reference power; when the coating temperature reaches the first temperature threshold but is below a preset second temperature threshold, the laser power is maintained within the fluctuation range of the reference power, and the laser power is fine-tuned according to the temperature gradient of the coating temperature. The expression for the adjustment process is as follows: ,in The current laser power, The temperature gradient is used; when the coating temperature is higher than the second temperature threshold, the laser power is reduced to the preset safe power and the pulse mode is activated.

[0097] Step 600: Detect the de-icing rate on the surface of the de-iced wind turbine blades. If the de-icing rate does not meet the preset de-icing standard, repeat the de-icing operation until the de-icing rate meets the preset de-icing standard.

[0098] Specifically, when performing thermal radiation scanning on the surface of the wind turbine blades after de-icing to obtain a residual ice temperature distribution map, a high-resolution infrared thermal imager was used to perform a full-area scan of the blade surface in the 8-14μm infrared band at a frame rate of 15Hz. Due to the significant difference in thermal conductivity between residual ice and the blade coating, the residual ice area exhibits different temperature characteristics in the thermal image compared to the surrounding coating. The residual ice area has a lower and more uniform temperature distribution, while the temperature of the ice-free coating area fluctuates more significantly due to environmental factors and residual heat from the laser. During the scanning process, spatial registration was performed with the 3D model of the blade before de-icing, mapping the thermal image data to the 3D coordinate system of the blade surface to generate a residual ice temperature distribution map containing spatial coordinates and temperature values.

[0099] Next, the dielectric constant distribution of the de-iced blade surface was obtained by scanning it with millimeter-wave radar. The radar performed a helical scan of the blade surface with an angular resolution of 0.5°. After the emitted electromagnetic waves were reflected from the blade surface, the receiver extracted the amplitude and phase information of the echo signal through coherent demodulation. Then, combined with the phase change formula of electromagnetic waves propagating in different media, the complex dielectric constant of each scan point was calculated. Regions with a real part of complex dielectric constant in the range of 3.0-3.3 and an imaginary part less than 0.1 were marked as candidate areas for residual ice. These regions were aligned with the spatial coordinate system obtained from the thermal radiation scan to generate a complex dielectric constant distribution map, where each pixel contains the corresponding complex dielectric constant parameter.

[0100] Then, the DS evidence theory was used to fuse the residual ice temperature distribution map and the complex permittivity distribution to determine the residual ice volume. Both maps were divided into identical 1cm × 1cm grid cells, with each grid cell serving as the identification object. For the temperature distribution map, an average temperature ≤ -2℃ within the grid was used as evidence 1, and its basic probability allocation (BPA) for the "existence of residual ice" hypothesis was calculated; the lower the temperature and the greater the temperature difference with the surrounding environment, the higher the BPA value. For the complex permittivity distribution, a real part of the complex permittivity of 3.0-3.3 and an imaginary part ≤ 0.1 was used as evidence 2, and its BPA value was calculated similarly; the closer the permittivity is to the properties of pure ice, the higher the BPA value. Subsequently, the BPA of the two pieces of evidence were fused using the DS synthesis rule. When the confidence level of the fused evidence was ≥ 0.7, the grid cell was determined to be a residual ice region. The number of grid cells in all residual ice regions was counted, and the average thickness of the residual ice was deduced from the grid area and the average thickness of the residual ice using a 3D thickness distribution model. Multiplying these results yielded the total residual ice volume.

[0101] Finally, the ratio of the difference between the total ice volume before de-icing and the residual ice volume to the total ice volume before de-icing is taken as the de-icing rate.

[0102] The present invention also provides a remote non-contact photothermal de-icing system for the surface of wind turbine blades, comprising:

[0103] The data acquisition module is used to collect ice condition data on the surface of wind turbine blades using a multispectral camera and millimeter-wave radar to obtain multimodal data, including multispectral image data and radar echo data.

[0104] The feature extraction module is used to perform feature recognition on multimodal data through a two-stream convolutional neural network to obtain ice condition features; the ice condition features include: ice type classification results and ice layer thickness;

[0105] The strategy generation module is used to construct a three-dimensional thickness distribution model based on ice condition characteristics and natural lighting factors, and generate de-icing strategies.

[0106] The de-icing execution module is used to perform laser de-icing operations according to the de-icing strategy and to adjust and control the laser beam through a swarm focusing algorithm.

[0107] The laser adjustment module is used to monitor the temperature change of the coating on the surface of the wind turbine blades in real time and adjust the laser power according to the coating temperature.

[0108] The detection module is used to detect the de-icing rate on the surface of the wind turbine blades after de-icing. If the de-icing rate does not meet the preset de-icing standard, the de-icing operation is repeated until the de-icing rate meets the preset de-icing standard.

[0109] In some embodiments, all detection and identification related equipment, except for the laser emitting device, can be mounted on a drone for long-distance, non-contact detection. The remote non-contact photothermal de-icing system can also switch between different working modes according to different weather conditions. There are three working modes, and the switching process is as follows:

[0110] Mode 1 (Photothermal Coating + Sunlight): Under sunny conditions, the superhydrophobic photothermal conversion coating applied to the surface of the wind turbine blades can convert sunlight into heat energy. The remote de-icing system can monitor the surface ice shedding and adhesion in real time. If the surface ice melts and falls off, a highly efficient and energy-free de-icing effect can be achieved.

[0111] Mode 2 (Photothermal coating + sunlight + laser assistance): Even under sunlight, surface ice is difficult to melt and adhere to the surface. In this case, a fixed laser is needed as an auxiliary to sunlight. With the assistance of the fixed laser, the photothermal conversion of the coating generates enough heat to melt the surface ice.

[0112] Mode 3 (Photothermal Coating + Laser): On cloudy days with no sunlight, the photothermal conversion coating applied to the wind turbine blades completely loses its photothermal de-icing effect. Therefore, to achieve the de-icing effect, a fixed laser can be directly turned on as a simulated light source. In this case, the photothermal conversion of the coating generates enough heat to melt the surface ice.

[0113] The beneficial effects of this invention are as follows:

[0114] 1) By integrating multispectral imaging and millimeter-wave radar to acquire multimodal data, and combining dual-stream convolutional neural networks to extract ice texture features and spectral absorption features, the ice type classification results are obtained by comparing them with a preset ice and snow state feature database. The ice thickness is calculated based on the time delay difference of radar echo data, which improves the accuracy of ice condition feature identification.

[0115] 2) Based on ice characteristics and natural light factors, a three-dimensional thickness distribution model is constructed to generate a de-icing strategy. The laser beam is adjusted and controlled by a bee swarm focusing algorithm, and the laser power is adjusted by real-time monitoring of coating temperature changes. This achieves adaptive de-icing control, avoids energy waste, and reduces energy consumption while ensuring de-icing effect.

[0116] 3) It does not rely on natural light. In the absence of light or insufficient light, such as on cloudy days or at night, it can use laser as an energy source to remove ice, which solves the problem of traditional photothermal conversion materials relying on natural light.

[0117] 4) The method of combining the main beam and the auxiliary beam is adopted. The main beam operates according to the optimal focusing sequence, and the auxiliary beam dynamically assists at the point of sudden temperature drop. It can also compensate for blade vibration and predict motion, thus improving the de-icing efficiency.

[0118] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0119] Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this invention. Furthermore, those skilled in the art will recognize that, based on the ideas of this invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A remote, non-contact photothermal de-icing method for wind turbine blade surfaces, characterized in that, Includes the following steps: Ice condition data on the surface of wind turbine blades is collected using a multispectral camera and millimeter-wave radar to obtain multimodal data; the multimodal data includes: multispectral image data and radar echo data; The multimodal data is subjected to feature recognition using a dual-stream convolutional neural network to obtain ice condition features, which include: ice type classification results and ice layer thickness. Based on the ice condition characteristics, a three-dimensional thickness distribution model is constructed in conjunction with natural lighting factors, and a de-icing strategy is generated. Laser de-icing is performed according to the aforementioned de-icing strategy, and the laser beam is adjusted and controlled using a swarm focusing algorithm. The temperature change of the coating on the surface of the wind turbine blades is monitored in real time, and the laser power is adjusted according to the coating temperature. The de-icing rate is tested on the surface of the wind turbine blades after de-icing. If the de-icing rate does not meet the preset de-icing standard, the de-icing operation is repeated until the de-icing rate meets the preset de-icing standard. Based on the ice condition characteristics, a three-dimensional thickness distribution model is constructed in conjunction with natural lighting factors, and a de-icing strategy is generated, including: The current ambient light intensity is sampled at multiple points by an irradiance sensor array to obtain a real-time light distribution map; An effective illumination intensity map is obtained by performing attenuation compensation on the real-time illumination distribution map using an atmospheric transmittance model. The surface of the wind turbine blades is subjected to finite element mesh generation to obtain a unit mesh; The predicted ice growth rate is obtained by performing time-series analysis on meteorological data provided by the weather station using an LSTM neural network. The predicted thickness distribution of the cell grid is obtained based on the predicted ice layer thickness and the predicted ice layer growth rate. The predicted thickness distribution is reconstructed in three dimensions by fitting a non-uniform rational B-spline surface to obtain the three-dimensional thickness distribution model. The laser action parameters are generated based on the three-dimensional thickness distribution model, and the laser action parameters are subject to conditional constraints to obtain the de-icing strategy.

2. The remote non-contact photothermal de-icing method for wind turbine blade surfaces according to claim 1, characterized in that, Ice condition data was collected on the surface of wind turbine blades using a multispectral camera and millimeter-wave radar, yielding multimodal data, including: The surface of the wind turbine blades is scanned in visible light and short-wave infrared bands using the multispectral camera to obtain raw image data. Radiometric calibration is performed on the original image data to obtain radiance image data; Atmospheric scattering correction is performed on the obtained radiance image data to obtain surface reflection image data; The multispectral image data is obtained by performing multiband spatial registration on the surface reflectance image data using a scale-invariant feature transform algorithm.

3. The remote non-contact photothermal de-icing method for wind turbine blade surfaces according to claim 2, characterized in that, Ice condition data was collected from the surface of wind turbine blades using a multispectral camera and millimeter-wave radar, yielding multimodal data, including: The original signal is obtained by transmitting electromagnetic wave signals to the surface of the wind turbine blades using the millimeter-wave radar. The transmitted electromagnetic wave signal and the original signal are coherently mixed to obtain a difference frequency signal; The radar echo data is obtained by performing time-frequency decomposition on the difference frequency signal using wavelet packet transform.

4. The remote non-contact photothermal de-icing method for wind turbine blade surfaces according to claim 1, characterized in that, The multimodal data is subjected to feature recognition using a two-stream convolutional neural network to obtain ice condition features, including: Ice texture features of the multimodal data are extracted through the first sub-network of the dual-stream convolutional neural network; The spectral absorption features of the multimodal data are extracted through the second sub-network of the dual-stream convolutional neural network; The ice texture features and the spectral absorption features are compared with a preset ice and snow state feature database to obtain ice type classification results; The thickness of the ice layer is calculated based on the time delay difference of the radar echo data.

5. The remote non-contact photothermal de-icing method for wind turbine blade surfaces according to claim 1, characterized in that, Laser de-icing is performed according to the aforementioned de-icing strategy, and the laser beam is adjusted and controlled using a swarm focusing algorithm, including: The optimal focusing sequence of the main beam is obtained by performing path planning on the unit grid using an adaptive particle swarm optimization algorithm. Real-time temperature monitoring of the ice layer and identification of points of sudden temperature drop; An auxiliary beam is generated based on the temperature drop point, and the laser power of the auxiliary beam is adjusted according to the real-time temperature. The vibration spectrum of the wind turbine blade surface is collected to perform vibration compensation and motion prediction on the wind turbine blade surface, so as to locate the position of the unit grid in real time.

6. The remote non-contact photothermal de-icing method for wind turbine blade surfaces according to claim 1, characterized in that, Real-time monitoring of the coating temperature change on the surface of the wind turbine blades, and adjustment of the laser power according to the coating temperature, including: When the coating temperature is lower than a preset first temperature threshold, the laser power is increased to the reference power; When the coating temperature reaches the first temperature threshold and is lower than the preset second temperature threshold, the laser power is maintained within the fluctuation range of the reference power, and the laser power is finely adjusted according to the temperature gradient of the coating temperature. When the coating temperature is higher than the second temperature threshold, the laser power is reduced to a preset safe power, and the pulse mode is activated.

7. The remote non-contact photothermal de-icing method for wind turbine blade surfaces according to claim 1, characterized in that, The de-icing rate of the wind turbine blades after de-icing is tested. If the de-icing rate does not meet the preset de-icing standard, the de-icing operation is repeated until the de-icing rate meets the preset de-icing standard, including: Thermal radiation scanning was performed on the surface of the wind turbine blades after de-icing to obtain a residual ice temperature distribution map. The dielectric constant distribution of the wind turbine blades after de-icing is obtained by scanning the surface of the de-icing wind turbine blades using the millimeter-wave radar. The residual ice volume is obtained by fusing the residual ice temperature distribution map and the complex permittivity distribution using DS evidence theory. The de-icing rate is calculated based on the residual ice volume.

8. A remote non-contact photothermal de-icing system for wind turbine blade surfaces, characterized in that, include: The data acquisition module is used to collect ice condition data on the surface of wind turbine blades using a multispectral camera and millimeter-wave radar to obtain multimodal data. The multimodal data includes: multispectral image data and radar echo data; The feature extraction module is used to perform feature recognition on the multimodal data through a two-stream convolutional neural network to obtain ice condition features; the ice condition features include: ice type classification results and ice layer thickness; The strategy generation module is used to construct a three-dimensional thickness distribution model based on the ice condition characteristics and natural lighting factors, and to generate a de-icing strategy; including: The current ambient light intensity is sampled at multiple points by an irradiance sensor array to obtain a real-time light distribution map; An effective illumination intensity map is obtained by performing attenuation compensation on the real-time illumination distribution map using an atmospheric transmittance model. The surface of the wind turbine blades is subjected to finite element mesh generation to obtain a unit mesh; The predicted ice growth rate is obtained by performing time-series analysis on meteorological data provided by the weather station using an LSTM neural network. The predicted thickness distribution of the cell grid is obtained based on the predicted ice layer thickness and the predicted ice layer growth rate. The predicted thickness distribution is reconstructed in three dimensions by fitting a non-uniform rational B-spline surface to obtain the three-dimensional thickness distribution model. The laser action parameters are generated based on the three-dimensional thickness distribution model, and the laser action parameters are subject to conditional constraints to obtain the de-icing strategy. The de-icing execution module is used to perform laser de-icing operations according to the de-icing strategy and to adjust and control the laser beam through a swarm focusing algorithm; A laser adjustment module is used to monitor the temperature change of the coating on the surface of the wind turbine blades in real time and adjust the laser power according to the coating temperature. The detection module is used to detect the de-icing rate on the surface of the wind turbine blades after de-icing. If the de-icing rate does not reach the preset de-icing standard, the de-icing operation is repeated until the de-icing rate reaches the preset de-icing standard.

Images