Wind turbine generator blade surface defect intelligent detection method based on image processing

Abstract

The invention relates to the field of image processing, and discloses a wind turbine generator blade surface defect intelligent detection method based on image processing, which is used for improving the accuracy of wind turbine generator blade surface defect detection. Comprising the steps that pulse laser thermal excitation infrared thermal imaging and a multi-angle polarization light source array polarization imaging technology are fused, a bimodal data set is constructed, and thermal anomaly features of a dynamic thermal image sequence are extracted; and photoelastic stress characteristics are obtained through phase analysis and local energy filtering. Generating a fusion defect indication diagram by utilizing heat conduction blocking and stress concentration space coincidence strengthening mechanism fusion features, obtaining a defect probability distribution diagram, calculating a predicted risk value, matching a maintenance strategy library according to the risk value, outputting a maintenance scheme code, and generating a three-dimensional maintenance guide diagram. According to the invention, intelligent and accurate detection and maintenance guidance of blade surface defects are realized, and the comprehensiveness and accuracy of defect detection are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of image processing, and particularly relates to an intelligent detection method for surface defects of wind turbine blades based on image processing. BACKGROUND

[0002] With the transformation of global energy structure, wind energy as a clean and renewable energy, its proportion in energy supply is increasing. Wind turbine as the core equipment of wind energy development and utilization, its safe and stable operation is directly related to the efficiency and reliability of wind power generation. Wind turbine blades as the key components of capturing wind energy, long-term exposed to complex natural environment, bearing the action of aerodynamic load, gravity, temperature change and other factors, prone to surface cracks, wear, corrosion and other defects. These defects not only reduce the aerodynamic performance of the blade, affect the power generation efficiency, but also may cause blade fracture and other serious accidents, threaten the safe operation of the whole wind turbine. Therefore, timely and accurate detection and evaluation of surface defects of wind turbine blades is of great significance to ensure the reliable operation of wind turbine, prolong its service life and reduce the operation and maintenance cost.

[0003] The traditional detection method of surface defects of wind turbine blades mainly relies on manual visual inspection and regular shutdown detection. Manual visual inspection is limited by the experience, vision and detection environment of the detector, and is prone to missed detection and false detection, and the detection efficiency is low, which is difficult to meet the demand of large-scale wind farm rapid detection. Although regular shutdown detection can comprehensively check the blade condition, it will increase the downtime of wind turbine, cause power loss, and frequent start-stop operation will also cause damage to the equipment.

[0004] Therefore, we propose an intelligent detection method for surface defects of wind turbine blades based on image processing to solve the above problems. SUMMARY

[0005] The present application provides an intelligent detection method for surface defects of wind turbine blades based on image processing, which can improve the accuracy of detection of surface defects of wind turbine blades.

[0006] The first aspect of the present application provides an image processing-based intelligent detection method for surface defects of a wind turbine blade, comprising: generating a transient thermal excitation on the blade surface, synchronously triggering the acquisition of a dynamic thermal image sequence, simultaneously starting a polarized light source array to irradiate the blade surface, acquiring a polarized image group, and establishing a dual-modal data set; after anisotropic diffusion filtering of the dynamic thermal image sequence, generating a thermal conduction gradient tensor matrix, extracting the principal curvature extreme value to form a thermal anomaly feature map, performing photoelastic phase analysis on the polarized image group to obtain an optical elastic phase distribution, and combining local energy filtering to generate an optical elastic stress feature map; based on the spatial overlap reinforcement mechanism of the thermal conduction blocked area and the stress concentration area, generating a fusion defect indication map according to the pixel intensity of the thermal anomaly feature map and the gradient modulus value of the optical elastic stress feature map; in a logarithmic continuous scale space, performing phase consistency fluctuation analysis on the fusion defect indication map to generate a defect probability distribution map; according to the phase distribution data analyzed from the polarized image group and the spatial coordinates of the defect probability distribution map, reconstructing a defect three-dimensional depth map, and obtaining a predicted risk value based on the connected area of the defect probability distribution map and the maximum depth change rate of the three-dimensional depth map.

[0007] Optionally, in the first implementation manner of the first aspect of the present application, the method comprises: emitting a laser pulse to act on a target area of the blade surface, and synchronously sending a hardware-level trigger signal to an infrared thermal imager; starting acquisition after the laser pulse acts based on the trigger signal, and generating a time-encoded dynamic thermal image sequence; starting 0°, 45°, 90° and 135° polarization state LED light sources of a multi-angle polarized light source array at the same time as the laser pulse is emitted, and capturing four groups of polarized images through single exposure; time-axis registering the dynamic thermal image sequence and the polarized image group based on the time stamp of the trigger signal, and completing pixel-level spatial alignment by using a pre-labeled spatial transformation matrix, and outputting a dual-modal data set.

[0008] Optionally, in the second implementation manner of the first aspect of the present application, the method comprises: performing anisotropic diffusion filtering operation on the dynamic thermal image sequence to generate a denoised thermal conduction image sequence; calculating the second-order partial derivative of the time-space domain of the denoised thermal conduction image sequence to construct a thermal conduction gradient tensor matrix; solving the eigenvalue of the thermal conduction gradient tensor matrix to extract the maximum principal curvature extreme value point set and generate a thermal anomaly feature map; performing photoelastic phase analysis on the polarized image group to calculate the inverse tangent function value through four-way polarization intensity and generate an original phase distribution map; performing local energy filtering processing on the original phase distribution map to extract a stress concentration area and output an optical elastic stress feature map.

[0009] Optionally, in the third implementation manner of the first aspect of the present application, the phase angle of the original phase distribution map is: ; wherein, is the 0° polarization direction intensity, is the 45° polarization direction intensity, is the 90° polarization direction intensity, is the 135° polarization direction intensity, the coefficient is the birefringence effect due to the phase angle and stress difference.

[0010] Optionally, in a fourth implementation manner of the first aspect of the present application, the method comprises: identifying a heat conduction blocked area in the heat anomaly feature map, and extracting a pixel intensity distribution thereof as a first physical field input; calculating a gradient modulus value of the photoelastic stress feature map, and marking a stress concentration area as a second physical field input; performing Hadamard product operation on the pixel intensity distribution and the gradient modulus value to generate an initial fusion response map; and performing a physical field synergistic reinforcement operation on the initial fusion response map to enhance the response intensity of a spatially overlapped area of the heat conduction blocked area and the stress concentration area, and output a fusion defect indication map.

[0011] Optionally, in a fifth implementation manner of the first aspect of the present application, the method comprises: constructing a continuous scale space of a logarithmic distribution, the scale range covering 0.5 pixels to 8.0 pixels, and generating a multi-scale filter bank; performing phase consistency fluctuation analysis on the fusion defect indication map in the continuous scale space to obtain a local phase response under each scale; performing scale normalization processing on the local phase response to generate a normalized phase response map set; and integrating the normalized phase response map set along the scale dimension to generate a defect probability distribution map.

[0012] Optionally, in a sixth implementation manner of the first aspect of the present application, the method comprises: extracting a phase gradient modulus value of a defect area according to phase distribution data analyzed from the polarization image group and in combination with a spatial coordinate of the defect probability distribution map; converting the phase gradient modulus value into a depth value through a phase-depth mapping model to generate a defect three-dimensional depth map; marking a connected area in the defect probability distribution map to obtain a geometric area; extracting a maximum depth change rate from the defect three-dimensional depth map; and obtaining a predicted risk value based on the product of the geometric area and the maximum depth change rate.

[0013] Optionally, in a seventh implementation manner of the first aspect of the present application, the method further comprises a repair decision parameter generation: matching a pre-stored repair strategy library according to the predicted risk value to output a repair scheme code; generating a material removal thickness distribution map based on a spatial distribution feature of the defect three-dimensional depth map; and combining the repair scheme code and the material removal thickness distribution map to output a three-dimensional repair guide map.

[0014] The mechanism of the present application is as follows: through the spatio-temporal synchronous coupling of the thermal-optical dual physical fields and the physical mechanism driven image processing chain, intelligent detection and quantitative decision of the blade defects are realized without machine learning. Beneficial effects: The pulse laser thermal excitation infrared thermal imaging technology can capture the abnormal heat conduction on the surface of the blade, and the multi-angle polarized light source array polarization imaging technology can obtain the stress distribution information. The combination of the two can provide a binocular detection, which can more comprehensively and accurately find different types of defects, whether the defects are caused by heat conduction obstruction or stress concentration, and greatly improves the coverage and accuracy of defect detection. Based on the spatial coincidence strengthening mechanism of the heat conduction obstruction area and the stress concentration area, the pixel intensity of the thermal anomaly feature map and the gradient modulus value of the photoelastic stress feature map are subjected to Hadamard product operation to generate a fusion defect indication map, which fully considers the coupling relationship between different physical fields, strengthens the feature response of the defect area, and makes the fused feature map more clearly indicate the defect position. Compared with the traditional simple fusion method, the sensitivity and accuracy of defect detection are greatly improved. In the logarithmic continuous scale space, phase consistency fluctuation analysis is performed on the fusion defect indication map, and a defect probability distribution map is generated by scale normalization response integration. From the perspective of multi-scale analysis of defect features, the performance of defects at different scales can be more comprehensively considered, and the generated defect probability distribution map can more accurately reflect the possibility of defects, providing more detailed data support for risk assessment. According to the predicted risk value, a pre-stored repair strategy library is matched, and a repair scheme code is output, which provides targeted repair suggestions for each defect. Different risk levels of defects correspond to different repair strategies, realizing personalized customization of repair schemes and improving the pertinence and effectiveness of repair. Based on the spatial distribution characteristics of the three-dimensional depth map of the defect, a material removal thickness distribution map is generated. Combined with the repair scheme code and the material removal thickness distribution map, a three-dimensional repair guide map is output. The repair personnel can intuitively see the position, depth and specific requirements of the repair operation of the defect, and obtain an accurate navigation map, greatly reducing the difficulty and uncertainty of repair, improving the repair efficiency and quality, and reducing secondary damage caused by improper repair. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 An embodiment of the wind turbine blade surface defect intelligent detection method based on image processing in the embodiment of the application is shown in the figure. Figure 2 Another embodiment of the wind turbine blade surface defect intelligent detection method based on image processing in the embodiment of the application is shown in the figure. Figure 3 An embodiment of the wind turbine blade surface defect intelligent detection device based on image processing in the embodiment of the application is shown in the figure. DETAILED DESCRIPTION

[0016] The embodiment of the present application provides a wind turbine blade surface defect intelligent detection method based on image processing, and is used for improving the accuracy of wind turbine blade surface defect detection. The terms "first", "second", "third", "fourth" and the like (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not have to be used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "include" or "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0017] For ease of understanding, the specific flow of the embodiment of the present application is described below. Please refer to Figure 1 One embodiment of the wind turbine blade surface defect intelligent detection method based on image processing in the embodiment of the present application includes: 101, dual-physical-field image synchronous acquisition: a transient thermal excitation is generated on the blade surface by a pulsed laser thermal excitation device, and a dynamic thermal image sequence with a time resolution of 500Hz or more is collected by synchronously triggering an infrared thermal imager; at the same time, a multi-angle polarized light source array is started to irradiate the blade surface, and a polarization image group containing 0°, 45°, 90° and 135° polarization states is synchronously collected by a four-way polarization camera, and a dual-modality data set is established; It can be understood that the execution subject of the present application can be a wind turbine blade surface defect intelligent detection device based on image processing, and can also be a terminal or a server, and the specific place is not limited. The embodiment of the present application takes the server as the execution subject for example.

[0018] It should be noted that the device selection and configuration, the pulsed laser thermal excitation device: a pulsed laser with a wavelength of 1064nm (peak power of 5kW) is selected, a transient thermal excitation is generated on the blade surface within 0.5ms, the spot diameter is 10mm, the energy density is 5J / cm 2 , and the thermal penetration depth reaches 2mm of the blade composite material surface layer; the laser trigger signal is output through a BNC interface to synchronize the pulse (rise time <10ns), and the infrared thermal imager and the polarization camera are triggered to start.

[0019] Infrared thermal camera acquisition system: high-speed mid-wave infrared camera (FLIR X8500sc) was used with a resolution of 640x512 pixels and a sampling rate of more than 500 Hz (520 Hz in this case) and a noise equivalent temperature difference (NETD) of less than 20 mK. The dynamic thermal image sequence acquisition time was 1.5 seconds, generating 780 thermal images with a time resolution of 1.92 ms / frame, recording the leaf surface temperature decay process.

[0020] Polarization optical system: light source: annular LED polarization light source array (wavelength 850 nm), irradiating the leaf at 0°, 45°, 90°, and 135° polarization angles, with an illuminance of 100 klux and a coverage area of 1 m 2 . Camera: four-way polarization CMOS camera (LUCID PHX050S) with a resolution of 2448x2048 and a pixel size of 3.45 μm, capturing four sets of polarization state images simultaneously through time-sharing exposure (exposure time 200 μs / state).

[0021] Spatial registration and synchronization control: four high-reflectivity targets (5 mm in diameter) were pre-set on the leaf surface, with a laser thermal excitation point 20 cm away from the center of the target. The infrared and polarization cameras were spatially aligned through target coordinate mapping, with a registration error of less than 0.1 pixels.

[0022] Timing synchronization: the laser trigger signal synchronously starts the infrared thermal camera (internal clock) and the polarization camera (external trigger mode). The time deviation between the first frame of the infrared sequence and the first frame of the polarization image group is less than 50 μs, which is verified by the time stamp of the acquisition card (Linghua PCIe-CPL64).

[0023] Data acquisition process: the laser emits a pulse at leaf position S1, synchronously triggering the infrared thermal camera to record the thermal sequence (780 frames, 520 Hz) and the polarization camera to capture four-state image groups (4 images per group, with a 1 ms interval). The translation device moves to the adjacent position S2 (15 cm apart), and the above steps are repeated to cover the full length of a single leaf (60 m, a total of 400 detection points).

[0024] Single acquisition data volume: thermal sequence: MB; polarization group: MB; total volume of dual-modality data set: GB.

[0025] Output dual-modality data set structure:

[0026] 102. Synergistic extraction of physical field features: After anisotropic diffusion filtering of the dynamic thermographic sequence, a heat conduction gradient tensor matrix is calculated by computing the second-order partial derivatives in the space-time domain to extract the principal curvature extrema and form a thermal anomaly feature map. The polarization image set is subjected to phase analysis, and the photoelastic phase distribution is calculated by the inverse tangent function to generate a photoelastic stress feature map in combination with local energy filtering; It should be noted that the thermal physical field feature extraction (dynamic thermographic sequence processing) is anisotropic diffusion filtering: the dynamic thermographic sequence (780 frames, 640x512 pixels) collected at 520 Hz is filtered. The thermal conductivity coefficient , the smoothing coefficient , and the iteration number 20 are set. The heat conduction suppresses the uniform area noise while retaining the defect edges (regions with gradient mutation > 30℃ / px). Heat conduction gradient tensor generation: the second-order partial derivatives in the space-time domain of the filtered sequence are calculated to generate a heat conduction gradient tensor matrix. At time point s, the spatial gradient components , and the time gradient are extracted to form a 3x3 tensor matrix. Thermal anomaly feature map generation: the principal curvature extrema (curvature radius <0.5mm-1) of the tensor matrix are calculated and marked as thermal anomaly regions. The measured results show that the curvature extrema of the crack region are 12.8, which are significantly higher than the background value (0.3~1.2).

[0027] Optical physical field feature extraction (polarization image set processing), phase analysis: photoelastic phase calculation is performed on the four-way polarization image set (2448x2048 pixels, 0° / 45° / 90° / 135°). The inverse tangent function is used: ; to generate a phase distribution map (range 0~π), in which the phase jump of the crack region is >1.2 rad.

[0028] Photoelastic stress feature map generation: local energy filtering (window size 5x5) is performed on the phase map to enhance the stress concentration area. The measured gradient modulus value of the bubble defect is 85 MPa / px, while the normal area is <5 MPa / px. The output example is:

[0029] 103. Thermal-optical physical feature fusion: based on the spatial coincidence reinforcement mechanism of the heat conduction blocked area and the stress concentration area, the pixel intensity of the thermal anomaly feature map and the gradient modulus value of the photoelastic stress feature map are subjected to Hadamard product operation to generate a fusion defect indication map; It should be noted that the input feature map alignment: the thermal anomaly feature map (640x512 pixels): generated by step 102, the principal curvature extrema of the crack region are ≥12.8mm -1(normalized intensity 0.85), background area ≤ 1.2mm -1 (Intensity 0.05-0.15). Photoelastic stress signature map (2448×2048 pixels): Bubble defect gradient modulus ≥85 MPa / px (normalized value 0.92), normal area <5 MPa / px (normalized value 0.08). The thermal anomaly signature map was downsampled to 2448×2048 resolution using bicubic interpolation using four preset high-reflectivity target coordinates and spatially aligned with the photoelastic stress signature map, with a registration error of <0.3 pixel.

[0030] Hadamard product fusion operation, spatial coincidence strengthening mechanism: In the crack region, due to internal structural fractures, heat conduction is hindered (high thermal anomaly intensity), and stress is concentrated (high gradient modulus). The spatial coincidence between the two is greater than 90%. Surface stains only cause optical scattering (medium-to-high gradient modulus) but no thermal conduction anomaly (low thermal anomaly intensity), and the coincidence is less than 15%.

[0031] Pixel-level fusion calculation: Perform Hadamard product (i.e., multiply the corresponding pixel intensities) on each pixel of the two registered feature maps: Example of crack area: thermal anomaly intensity 0.85 × stress gradient modulus 0.92 → fusion value 0.782 Example of background area: thermal anomaly intensity 0.12 × stress gradient modulus 0.10 → fusion value 0.012; Output fused defect indication image (2448×2048, float32 format), the intensity of the real defect area is increased to above 0.75, and the background noise is suppressed to below 0.05. Output:

[0032] 104. Multi-scale defect boundary segmentation: Perform phase consistency fluctuation analysis on the fused defect indication image in a logarithmic continuous scale space of 0.5-8.0 pixels, and generate a defect probability distribution map by scale-normalized response integration; It should be noted that the logarithmic continuous scale space is constructed with a scale range of 0.5 to 8.0 pixels on the fused defect indication image (2448×2048 pixels), with a total of 12 scale levels (scale parameters ). Spatial sampling: Each scale layer is convolved with a Gaussian kernel function to generate a multi-scale image pyramid. At this scale, the gradient response intensity of the crack edge reaches a peak value (about 0.78), while the background noise response is <0.05.

[0033] Phase coherence fluctuation analysis, directional filtering: 8-direction Gabor filter bank (angle interval 22.5°) is used to extract phase coherence features at each scale. Take the crack region as an example, the phase coherence value in the 45° direction (crack main direction) is 0.92, while in the vertical direction (135°) it is only 0.15. Multi-scale response integration: calculate the phase coherence fluctuation energy of each pixel point in all scales and directions: crack region: the energy peak value is concentrated in the scale, and the average fluctuation energy is 0.85; background region: energy is dispersed and the average is <0.10.

[0034] Scale normalized response integration, weight distribution: according to scale sensitivity, weights are distributed, small scale (0-2°) weight accounts for 60% (capture fine cracks), large scale (8-12°) accounts for 20% (suppress structural texture interference). Probability map generation: weighted integration of phase coherence responses of 12 scales, output defect probability distribution map (2448x2048, float32 format). Measured data: crack region probability value ≥0.90 (confidence interval); stain / texture region probability value ≤0.15; defect-background signal-to-noise ratio is improved to 28:1 (15:1 for fusion image input). Output verification:

[0035] 105, three-dimensional topography quantification and risk assessment: according to the phase distribution data analyzed from the polarization image set and the spatial coordinates of the defect probability distribution map, the three-dimensional depth map of the defect is reconstructed through the phase-depth mapping model, and based on the connected area of the defect probability distribution map and the maximum depth change rate of the three-dimensional depth map, the predicted risk value is calculated.

[0036] It should be noted that the input data of the three-dimensional depth reconstruction is aligned: the photoelastic phase distribution is analyzed from the four-way polarization image set (0° / 45° / 90° / 135°), and the phase jump value of the crack region is ≥1.2 rad (spatial resolution 2448x2048). Defect probability distribution map: generated by step 104, crack region probability value ≥0.90, coordinate range (x:1200-1250, y:800-850).

[0037] Phase-depth mapping: Establish the linear relationship between phase difference and depth: the phase jump increases by 1 rad, the depth increases by 0.8 mm. The phase jump of the crack center point (x: 1225, y: 825) is 1.5 rad, and the depth is 1.2 mm; the phase jump of the edge point (x: 1200, y: 800) is 0.3 rad, and the depth is 0.24 mm. Output three-dimensional depth map (2448x2048, float32), the maximum depth of the crack is 1.2 mm, and the minimum depth is 0.2 mm (background noise <0.05 mm).

[0038] Risk quantification calculation, parameter extraction: connected area: the connected area of the defect probability map above the threshold value of 0.85 is 800 mm 2 (equivalent rectangle 40mmx20mm). Maximum depth rate: the peak value of the depth rate along the main direction of the crack (45°) is 0.25 (calculation formula: ).

[0039] Risk value calculation: risk model: ( , ).

[0040] This example: (risk threshold: >500 is high risk). Output report:

[0041] In the embodiment of the present application, the defect detection is carried out in combination with the thermophysical field and the optical physical field. The thermophysical field can reflect the change of heat conduction caused by the internal structure of the blade due to defects, and the optical physical field can reflect the surface stress distribution. The combination of the two overcomes the limitations of single physical field detection. The surface stains are abnormal in the optical physical field but normal in the thermophysical field. The defects and interference can be accurately distinguished by the combination, and the accuracy and reliability of the defect detection are significantly improved. A unique method is used in the feature extraction link. In the thermophysical field feature extraction, the anisotropic diffusion filter can suppress the noise in the uniform area while retaining the defect edge. In the optical physical field feature extraction, the optical phase distribution is calculated by the arctangent function, the phase jump in the crack area is accurately captured, and the extracted features can more accurately reflect the defect characteristics, providing a solid foundation for subsequent accurate analysis. The feature fusion is carried out based on the space overlap strengthening mechanism of the heat conduction blocked area and the stress concentration area. The fusion defect indication map is generated by Hadamard product operation, the correlation of the features of the two physical fields in the defect area is fully utilized, the strength of the real defect area is greatly improved, the background noise is significantly suppressed, and the contrast between the defect and the background is greatly enhanced, which facilitates the subsequent accurate segmentation of the defect boundary. The phase consistency fluctuation analysis is carried out in the logarithmic continuous scale space to realize the multi-scale defect boundary segmentation. The multi-scale image pyramid is constructed, the phase consistency features are extracted by using the multi-direction Gabor filter set, and the scale normalized response integral is generated to generate the defect probability distribution map. The defect features of different scales can be captured. The small scale captures the fine cracks, and the large scale suppresses the structural texture interference, effectively improves the defect boundary positioning accuracy, and reduces the edge positioning error. Not only the defect detection is realized, but also a three-dimensional morphology quantization and risk assessment system is established. According to the phase distribution data of the polarization image group and the defect probability distribution map, the three-dimensional depth map of the defect is reconstructed, and the risk value is calculated and predicted in combination with the connected area and the maximum depth change rate. The system can comprehensively evaluate the risk from the geometric shape and physical characteristics of the defect, etc., provide a scientific basis for the blade maintenance decision, help to find high-risk defects in time, and ensure the safe operation of the wind turbine.

[0042] Please refer to Figure 2 Another embodiment of the image processing-based intelligent detection method for the surface defects of the wind turbine blade in the embodiment of the present application includes: 201. A transient thermal excitation is generated on the surface of the blade, and a dynamic thermal image sequence is collected synchronously. At the same time, a polarization light source array is started to irradiate the surface of the blade, and a polarization image group is collected to establish a dual-mode data set; It should be noted that the control pulse laser thermal excitation device emits a laser pulse with a duration of 10μs to act on the target area on the surface of the blade, and a hardware level trigger signal is sent to the infrared thermal imager synchronously; the infrared thermal imager starts to collect within 1ms after the laser pulse acts based on the trigger signal, and generates a time-encoded dynamic thermal image sequence at a frame rate of not less than 500Hz; at the same time of laser pulse emission, the 0°, 45°, 90° and 135° polarization state LED light sources of the multi-angle polarization light source array are started, and four sets of polarization images are captured by the four-way polarization camera through a single exposure of the light splitting prism; the dynamic thermal image sequence and the polarization image group are time-axis registered based on the timestamp of the trigger signal, and pixel-level spatial alignment is completed by using a pre-calibrated spatial transformation matrix, and a spatio-temporally unified bimodal data set is output; It should be noted that the hardware configuration is as follows: pulse laser excitation: a pulse laser with a wavelength of 1064nm and a peak power of 2kW is used to apply a laser pulse with a duration of 10μs to the target area (50mm×50mm) on the surface of the blade, and the energy density is 5J / cm 2 .

[0043] Infrared thermal imager: a PT series flagship thermal imager (resolution 1280×1024 pixels, temperature measurement accuracy ±2%) of Goootech is selected, and the collection is started within 1ms after the laser pulse ends through a hardware trigger signal (TTL level), and 1000 frames of dynamic thermal image sequences (time encoding accuracy 0.1ms) are generated by continuously recording for 2 seconds at a frame rate of 500Hz.

[0044] Polarization imaging system: light source: ring-shaped polarization LED array (0°, 45°, 90° and 135° polarization state), wavelength 520nm, illuminance 5000lux, and the light is turned on synchronously with the laser pulse (delay <1μs). Camera: four-way light splitting prism polarization camera (IMX250MYR sensor), which synchronously captures four sets of polarization images (resolution 2048×1536) in a single exposure (100μs).

[0045] Synchronization and alignment process: timing control: the trigger signal is sent from the laser controller to the thermal imager and the polarization light source controller, and the timestamp synchronization error is ≤50ns. The starting frame of the thermal image sequence is marked as ms, and the polarization image group is marked as μs (laser action time). Spatial calibration: pre-calibration stage: a chessboard calibration plate is pasted on the surface of the blade, and the thermal imager and the polarization camera collect images respectively, and a spatial transformation matrix (affine transformation, accuracy ±0.5 pixels) is calculated. Alignment operation: the polarization image group is registered to the thermal image sequence coordinate system by bilinear interpolation, and a 512×512 pixel bimodal data set (containing thermal image AD value, polarization phase angle and intensity) is output.

[0046] Output dataset example: Thermal image sequence: Each frame contains a temperature field matrix (unit: AD value), with a time axis interval of 2ms and a spatial resolution of 0.1mm / pixel. Polarization image group: Four-channel polarization intensity matrix (I0°, I 45 °、I 90 °、I 135 °), and after registration, it shares the same coordinate system with the thermal image sequence. Metadata: timestamp sequence, ambient temperature (25°C), emissivity (0.95), and laser energy parameters.

[0047] 202. After performing anisotropic diffusion filtering on the dynamic thermal image sequence, a thermal conduction gradient tensor matrix is ​​generated, and the principal curvature extreme values ​​are extracted to form a thermal anomaly feature map. Phase analysis is performed on the polarization image group to obtain a photoelastic phase distribution, and a photoelastic stress feature map is generated by combining local energy filtering. Specifically, an anisotropic diffusion filtering operation is performed on a dynamic thermal image sequence to generate a denoised thermal conduction image sequence; the second-order partial derivatives in the spatiotemporal domain of the denoised thermal conduction image sequence are calculated to construct a thermal conduction gradient tensor matrix; the eigenvalues ​​of the thermal conduction gradient tensor matrix are solved, the set of maximum principal curvature extreme points is extracted, and a thermal anomaly feature map is generated; photoelastic phase analysis is performed on the polarization image group, and the inverse tangent function value is calculated based on the four-directional polarization intensity to generate an original phase distribution map; local energy filtering is performed on the original phase distribution map to extract the stress concentration area, and a photoelastic stress feature map is output; It should be noted that for the 50mm×50mm inspection area on the surface of the wind turbine blade (including artificial prefabricated defects: 0.2mm deep microcracks and 3mm diameter debonding areas), dynamic thermal image sequences (500Hz frame rate, 2 seconds duration) and polarization image groups (0°, 45°, 90°, 135° four-directional polarization states) are collected simultaneously.

[0048] Thermal anomaly feature map generation process, thermal conduction dynamic noise reduction: Anisotropic diffusion filtering (conduction coefficient 0.15, 10 iterations) is performed on a 1000-frame thermal image sequence to suppress noise while retaining defect edges, and output a denoised thermal image sequence (with a signal-to-noise ratio improved to 35dB).

[0049] Heat conduction gradient tensor construction: Calculate the second-order partial derivatives in the spatiotemporal domain of the denoised sequence (time step 2 ms, spatial step 0.1 mm / pixel) to generate a 5×5×5 heat conduction gradient tensor matrix (dimensions: spatial x×y×temporal t).

[0050] Principal curvature extreme value extraction: Solve the eigenvalues ​​of the tensor matrix, extract the maximum principal curvature extreme point (threshold > 0.8), mark the area where heat conduction is blocked (debonding area), and output the thermal anomaly feature map (resolution 512×512 pixels, defect area intensity value 120-255).

[0051] Optical stress signature generation flow, optical phase analysis: based on four-way polarization intensity (I0°, I 45 °, I 90 °, I 135 °), calculate the original phase distribution map according to the formula: ; Coefficient is due to the birefringence effect of the phase angle and stress difference; The phase angle of the micro-crack region jumps ±60° (within ±10° of the background).

[0052] Local energy filter enhancement: use a 5x5 local energy filter (weight kernel: Gaussian-Laplacian hybrid) to enhance the stress concentration area, output the optical stress signature (gradient modulus >0.6 area marked as stress anomaly, corresponding to the crack position).

[0053] Thermal anomaly feature map: debonding area shows high temperature retention (average pixel intensity 180), micro-cracks show low temperature dark bands due to heat flow obstruction (intensity value 50). Optical stress signature: micro-crack edge stress concentration is significant (gradient modulus peak value 0.92), debonding area stress distribution is uniform (gradient modulus <0.3).

[0054] 203、Based on the spatial overlap enhancement mechanism of the heat conduction blocked area and the stress concentration area, a fusion defect indication map is generated according to the pixel intensity of the thermal anomaly feature map and the gradient modulus of the optical stress signature; Specifically, identify the heat conduction blocked area in the thermal anomaly feature map, extract its pixel intensity distribution as the first physical field input; calculate the gradient modulus of the optical stress signature, mark the stress concentration area as the second physical field input; based on the heat-stress spatial coupling mechanism, the pixel intensity distribution and the gradient modulus are multiplied by Hadamard to generate an initial fusion response map; Perform physical field synergistic enhancement operation on the initial fusion response map to enhance the response intensity of the spatial overlap area of heat conduction obstruction and stress concentration, and output the fusion defect indication map; It should be noted that for a 50mmx50mm detection area on the surface of the fan blade (including artificial prefabricated defects: 0.2mm deep micro-cracks and 3mm diameter debonding area), the thermal anomaly feature map (resolution 512x512 pixels) and the optical stress signature (same resolution) have been generated through step 202.

[0055] Fusion defect indication map generation process, physical field input extraction: thermal anomaly feature map: extract the heat conduction blocked area (pixel intensity of debonding area 180±10, micro crack area low intensity 50±5 due to thermal blockage), as the first physical field input. Photoelastic stress feature map: calculate gradient modulus (Sobel operator), stress concentration area (micro crack edge gradient modulus 0.92±0.05, debonding area background gradient modulus <0.3), marked as the second physical field input.

[0056] Hadamard product operation: pixel-by-pixel multiplication operation is performed on the two feature maps: debonding area: thermal intensity (180) x stress gradient modulus (0.25) → initial response value 45; micro crack: thermal intensity (50) x stress gradient modulus (0.92) → initial response value 46; Generate initial fusion response map (pixel value range 0~255), at this time the response intensity of micro crack and debonding area is close and difficult to distinguish.

[0057] Physical field synergistic reinforcement: spatial coincidence area identification: locate the heat conduction blocked and stress concentrated overlapping area (micro crack area overlap degree >85%, debonding area <30%). Strengthening operation: apply 3 times weight enhancement to the overlapping area: micro crack response value from 46 to 138; debonding area response value remains 45 (not enhanced due to low overlap degree); non-overlapping area response value decays by 50%, suppressing background noise.

[0058] Output result, fusion defect indication map: micro crack area is significantly highlighted after reinforcement (average pixel intensity 138±12), debonding area weak response (intensity 45±8), background noise intensity <10.

[0059] 204、In the logarithmic continuous scale space, the fusion defect indication map is subjected to phase consistency fluctuation analysis to generate a defect probability distribution map; Specifically, a continuous scale space of logarithmic distribution is constructed, the scale range covers 0.5 pixels to 8.0 pixels, and a multi-scale filter bank is generated; in the continuous scale space, the fusion defect indication map is subjected to phase consistency fluctuation analysis, and the local phase response under each scale is calculated; the local phase response is subjected to scale normalization processing to generate a normalized phase response map set; the normalized phase response map set is integrated along the scale dimension to generate a defect probability distribution map.

[0060] It should be noted that the input data: fusion defect indication map (resolution 512x512 pixels), containing 0.2mm deep micro crack (response intensity 138±12) and 3mm diameter debonding area (response intensity 45±8), background noise intensity <10.

[0061] Step implementation details, multi-scale filter bank construction: construct a logarithmic scale space, scale range 0.5~8.0 pixels, generate 6 scale layers according to exponential interval (0.5, 1.0, 2.0, 4.0, 6.4, 8.0 pixels). The filter size corresponding to each scale: the minimum scale 0.5 pixel (3x3 Gaussian kernel), the maximum scale 8.0 pixel (41x41 Gaussian kernel), used to capture micro cracks to macro debonding defects.

[0062] Phase consistency fluctuation analysis: at each scale, calculate the local phase response of the fusion map: 0.5 pixel scale: sensitive to micro crack edges, local phase response peak value 0.85 (debonding area only 0.15); 8.0 pixel scale: capture the overall shape of the debonding area, response peak value 0.70 (micro crack response 0.25). The phase response calculation uses a Log-Gabor odd-symmetry filter bank with a directional resolution of 22.5° (a total of 8 directions).

[0063] Scale normalization processing: energy normalization is performed on the local phase response of each scale: micro crack area: 0.5 pixel scale weight 0.6, 8.0 pixel scale weight 0.15; debonding area: 0.5 pixel scale weight 0.1, 8.0 pixel scale weight 0.5. Output the normalized response map set to eliminate scale sensitivity differences.

[0064] Defect probability distribution generation: integrate the normalized response map set along the scale dimension: micro crack area: integral value 0.92 (close to 1.0 indicates high defect probability); debonding area: integral value 0.35 (medium-low probability); background area: integral value <0.1. The defect probability distribution map is output in the form of a heat map, with a probability threshold of 0.6 (>0.6 for high-risk defects). Scale parameter table:

[0065] 205、According to the phase distribution data analyzed from the polarization image group and the spatial coordinates of the defect probability distribution map, a three-dimensional depth map of the defect is reconstructed, and a predicted risk value is obtained based on the connected area of the defect probability distribution map and the maximum depth change rate of the three-dimensional depth map; Specifically, according to the phase distribution data analyzed from the polarization image group, the phase gradient modulus value of the defect area is extracted in combination with the spatial coordinates of the defect probability distribution map; the phase gradient modulus value is converted into a depth value through a phase-depth mapping model to generate a three-dimensional depth map of the defect; the connected area is marked in the defect probability distribution map, and its geometric area is calculated; the maximum depth change rate is extracted from the three-dimensional depth map of the defect; and the predicted risk value is calculated based on the product of the geometric area and the maximum depth change rate.

[0066] It should be noted that the input data: polarization image group: four-way polarization state (0°, 45°, 90°, 135°) intensity matrix (resolution 2048x1536), microcrack edge phase gradient modulus 0.95±0.03, debonding area center phase gradient modulus 0.15±0.05. Defect probability distribution map: 512x512 pixels, microcrack area probability value 0.92 (high risk), debonding area probability value 0.35 (medium-low risk), background area probability <0.1.

[0067] Step implementation flow, phase-depth mapping: extract the phase gradient modulus of the defect probability>0.6 area (microcrack: 0.95, debonding area: 0.15). By calibrating the phase-depth model (empirical formula: , mm), generate defect three-dimensional depth map: microcrack maximum depth 0.20mm (error ±0.02mm), debonding area depth 0.03mm (error ±0.01mm).

[0068] Defect area quantification: label the connected area in the probability map: microcrack: linear area, area 2.5mm 2 (length x width: 5mm x 0.5mm); debonding area: circular area, area 7.1mm 2 (diameter 3mm).

[0069] Extract the maximum depth rate of the three-dimensional depth map (unit: mm / pixel): microcrack edge: 0.08 (abrupt area); debonding area center: 0.005 (gentle area).

[0070] Predict risk value calculation: : microcrack: ; debonding area: ; normalized processing (0~1 range): microcrack risk level 0.82 (high risk), debonding area 0.15 (low risk). Parameter comparison table:

[0071] 206、Maintenance decision parameter generation: match the pre-stored maintenance strategy library according to the predicted risk value, output the maintenance scheme code; based on the spatial distribution characteristics of the defect three-dimensional depth map, generate a material removal thickness distribution map; combine the maintenance scheme code and the material removal thickness distribution map, output a three-dimensional maintenance guide map.

[0072] It should be noted that the input data: predicted risk value: microcrack area risk value 0.82 (high risk), debonding area risk value 0.15 (low risk). Defect three-dimensional depth map: microcrack maximum depth 0.20mm, debonding area depth 0.03mm (spatial resolution 0.1mm / pixel).

[0073] Step implementation flow, repair scheme matching: pre-stored repair strategy library defines three types of schemes: high-risk scheme (H): immediate shutdown repair, code "H001" (including carbon fiber reinforcement + epoxy resin filling process). Medium-risk scheme (M): planned repair, code "M002" (only resin filling). Low-risk scheme (L): handle during annual maintenance, code "L003" (surface grinding). Microcrack risk value 0.82 matches scheme code H001; debonding area risk value 0.15 matches scheme code L003.

[0074] Material removal thickness distribution map generation: based on the spatial distribution characteristics of the three-dimensional depth map: microcrack area: depth 0.20mm→ mm (blade tip area). Debonding area: depth 0.03mm→ mm (blade root area), output thickness distribution heat map.

[0075] Three-dimensional repair guide map synthesis: superimpose repair scheme code and thickness distribution map: microcrack area labeled H001, 0.24mm layered grinding path displayed synchronously, debonding area labeled L003. Output three-dimensional guide map, support AR device visualization operation (positioning accuracy ±2mm).

[0076] Repair scheme and parameter correspondence table:

[0077] In the embodiment of the application, the transient thermal excitation and the multi-angle polarization imaging are synchronously triggered, the space-time alignment is realized through the hardware level timestamp, the dual-modal data set unified in space-time is constructed, the Hadamard product operation and the physical field cooperative reinforcement algorithm are proposed, the response weight is dynamically adjusted through the spatial overlap degree of the heat conduction blocked area and the stress concentration area, the 0.5-8.0 pixel logarithmic scale space is constructed, the multi-scale features from micro cracks to macro debonding are captured through the Log-Gabor filter set, the scale normalization processing is combined to eliminate the scale sensitive difference, the phase gradient modulus is converted into the depth value based on the phase-depth mapping model, the prediction risk value is calculated combined with the connected area and the maximum depth change rate, the pre-stored repair strategy library is combined to automatically output the scheme code according to the risk value, and the three-dimensional guide map supported by the AR device is generated by superimposing the repair scheme code and the material removal thickness distribution map.

[0078] Figure 3 It is a kind of structure schematic view of wind turbine blade surface defect intelligent detection equipment based on image processing provided in the embodiment of the application, and the based on image processing wind turbine blade surface defect intelligent detection equipment 300 can be different due to configuration or performance, equipment 300 includes transmitter 301, receiver 302 and processor 303.Processor 303 can also be controller, Figure 3The device 300 can also include a modem processor 305 that can include a coder 306, a modulator 307, a demodulator 309, and a decoder 308. The modem processor 305 can be a baseband processor. In some aspects, the modem processor 305 can be integrated with a wireless modem and / or an application processor. The modem processor 305 can process traffic data based on the coding, modulation, demodulation, and decoding described in the embodiments of the present disclosure. The modem processor 305 can process signaling messages based on the coding, modulation, demodulation, and decoding described in the embodiments of the present disclosure.

[0079] On the uplink, a transmitter 301 can condition (e.g., amplify, filter, and modulate) the output samples and generate an uplink signal, which can be transmitted via an antenna to the access network device. On the downlink, the antenna can receive downlink signals transmitted by the access network device. A receiver 302 can condition (e.g., filter, amplify, and frequency downconvert) signals received from the antenna and provide input samples. In the modem processor 305, the coder 306 can receive traffic data and signaling messages to be sent on the uplink and process (e.g., format, encode, and interleave) the traffic data and signaling messages. The modulator 307 can further process (e.g., symbol map and modulate) the coded traffic data and signaling messages and provide output samples. The demodulator 309 can process (e.g., demodulate) the input samples and provide symbol estimates. The decoder 308 can process (e.g., deinterleave and decode) the symbol estimates and provide decoded data and signaling messages transmitted to the device 300. The coder 306, the modulator 307, the demodulator 309, and the decoder 308 can be implemented by a synthetic modem processor 305. These units can process the traffic data and signaling messages based on the radio access technology employed by the wireless access network (e.g., LTE and other evolved systems). It is noted that when the device 300 does not include the modem processor 305, the functions of the modem processor 305 can be performed by the processor 303.

[0080] The processor 303 can manage and control the overall operations of the device 300, for performing processes of the embodiments of the present disclosure described above by the device 300. For example, the processor 303 can be configured to perform the steps of the transmitting device or the receiving device in the method embodiments described above, and / or other steps of the technical solutions described in the embodiments of the present disclosure.

[0081] Further, the device 300 can also include a memory 304, configured to store program codes and data for the device 300.

[0082] It can be understood that, Figure 3 The device 300 is merely illustrated as a simplified design. In actual applications, the device 300 can include any number of transmitters, receivers, processors, modem processors, memories, etc., and all devices that can implement the embodiments of the present disclosure are within the protection scope of the embodiments of the present disclosure.

[0083] The application further provides a wind turbine blade surface defect intelligent detection device based on image processing, which comprises a memory and a processor, the memory stores computer readable instructions, and the computer readable instructions are executed by the processor to make the processor execute the steps of the wind turbine blade surface defect intelligent detection method based on image processing in each of the above embodiments.

[0084] The application further provides a computer readable storage medium, which can be a non-volatile computer readable storage medium or a volatile computer readable storage medium, the computer readable storage medium stores instructions, and the instructions make the computer execute the steps of the wind turbine blade surface defect intelligent detection method based on image processing when the instructions are run on the computer.

[0085] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described system, device and unit can refer to the corresponding processes in the foregoing method embodiments, which will not be described here.

[0086] The integrated unit, if realized in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the application or the whole or part of the technical solutions that essentially contribute to the prior art can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the application. The foregoing storage medium includes a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various program code storage media.

[0087] The above-described embodiments are only used to illustrate the technical solutions of the application, rather than limit them; although the application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the application.

Claims

1. An image processing-based intelligent detection method for surface defects of a wind turbine blade, characterized in that, The image processing-based wind turbine blade surface defect intelligent detection method comprises: Synchronously triggering the acquisition of dynamic thermal image sequence and starting the polarized light source array to irradiate the blade surface to acquire the polarized image group and establish the dual-mode data set; After anisotropic diffusion filtering of the dynamic thermal image sequence, a thermal conduction gradient tensor matrix is generated, the principal curvature extreme value is extracted to form a thermal anomaly feature map, and the polarized image group is phase-analyzed to obtain an optical elastic phase distribution, combined with local energy filtering to generate an optical elastic stress feature map; Based on the spatial overlap reinforcement mechanism of the thermal conduction blocked area and the stress concentration area, a fusion defect indication map is generated according to the pixel intensity of the thermal anomaly feature map and the gradient modulus value of the optical elastic stress feature map; In the logarithmic continuous scale space, phase-consistent fluctuation analysis is performed on the fusion defect indication map to generate a defect probability distribution map; According to the phase distribution data analyzed from the polarized image group and the spatial coordinates of the defect probability distribution map, a defect three-dimensional depth map is reconstructed, and a predicted risk value is obtained based on the connected area of the defect probability distribution map and the maximum depth change rate of the three-dimensional depth map.

2. The image processing based wind turbine blade surface defect intelligent detection method according to claim 1, characterized in that, It comprises: The laser pulse acts on the target area of the blade surface, and a hardware-level trigger signal is sent to the infrared thermal imager synchronously; Based on the trigger signal, the acquisition is started after the laser pulse action to generate a time-encoded dynamic thermal image sequence; At the same time of laser pulse emission, the 0°, 45°, 90° and 135° polarization state LED light sources of the multi-angle polarized light source array are started, and four groups of polarized images are captured through single exposure; Based on the timestamp of the trigger signal, the dynamic thermal image sequence and the polarized image group are time-axis registered, and pixel-level spatial alignment is completed by using a pre-calibrated spatial transformation matrix, and a dual-mode data set is output.

3. The image processing based wind turbine blade surface defect intelligent detection method according to claim 2, characterized in that, It comprises: Anisotropic diffusion filtering operation is performed on the dynamic thermal image sequence to generate a denoised thermal conduction image sequence; The second-order partial derivative of the denoised thermal conduction image sequence in the space-time domain is calculated to construct a thermal conduction gradient tensor matrix; The eigenvalues of the thermal conduction gradient tensor matrix are solved to extract the maximum principal curvature extreme value point set to generate a thermal anomaly feature map; The optical elastic phase of the polarized image group is analyzed, the inverse tangent function value is calculated through four-way polarization intensity, and an original phase distribution map is generated; Local energy filtering is performed on the original phase distribution map to extract the stress concentration area and output the optical elastic stress feature map.

4. The image processing based wind turbine blade surface defect intelligent detection method according to claim 3, characterized in that, The phase angle of the original phase distribution map is: ； wherein is the 0° polarization direction intensity, is the 45° polarization direction intensity, is the 90° polarization direction intensity, is the 135° polarization direction intensity, and the coefficient is due to the birefringence effect of the phase angle and the stress difference.

5. The image processing based wind turbine blade surface defect intelligent detection method according to claim 3, characterized in that, It comprises: The thermal conduction blocked area in the thermal anomaly feature map is identified, and its pixel intensity distribution is extracted as the first physical field input; The gradient modulus value of the optical elastic stress feature map is calculated to mark the stress concentration area as the second physical field input; The pixel intensity distribution and the gradient modulus value are subjected to Hadamard product operation to generate an initial fusion response map; Physical field synergistic reinforcement operation is performed on the initial fusion response map to enhance the response intensity of the spatial overlap area of the thermal conduction blocked and stress concentration, and a fusion defect indication map is output.

6. The image processing based wind turbine blade surface defect intelligent detection method according to claim 5, characterized in that, It comprises: A logarithmic distribution continuous scale space is constructed, the scale range covers 0.5 pixels to 8.0 pixels, and a multi-scale filter group is generated; In the continuous scale space, phase consistency fluctuation analysis is performed on the fusion defect indication map to obtain local phase responses at each scale; Scale normalization processing is performed on the local phase responses to generate a normalized phase response map set; The normalized phase response map set is integrated along the scale dimension to generate a defect probability distribution map.

7. The image processing based wind turbine blade surface defect intelligent detection method according to claim 6, characterized in that, It includes: According to the phase distribution data analyzed from the polarization image group, combined with the spatial coordinates of the defect probability distribution map, the phase gradient modulus value of the defect area is extracted; The phase gradient modulus value is converted into a depth value through a phase-depth mapping model to generate a defect three-dimensional depth map; Labeling the connected regions in the defect probability distribution map to obtain the geometric area; The maximum depth change rate is extracted from the defect three-dimensional depth map; Based on the product of the geometric area and the maximum depth change rate, a predicted risk value is obtained.

8. The image processing based wind turbine blade surface defect intelligent detection method according to claim 1, characterized in that, It also includes maintenance decision parameter generation: According to the predicted risk value, match the pre-stored maintenance strategy library, output the maintenance scheme code; Based on the spatial distribution characteristics of the defect three-dimensional depth map, a material removal thickness distribution map is generated; Combined with the maintenance scheme code and the material removal thickness distribution map, a three-dimensional maintenance guide map is output.

Images