CN-121539450-B - Fan blade clearance monitoring system and method based on AI image recognition

CN121539450BCN 121539450 BCN121539450 BCN 121539450BCN-121539450-B

Abstract

The invention discloses a fan blade clearance monitoring system and method based on AI image recognition, and relates to the technical field of wind power generation. According to the invention, by fusing visual AI and computational imaging technology, high-definition full-color video images of fan blades in a motion state can be acquired under an extremely low-illumination and no-light supplementing environment, motion smear is eliminated, the problem of fuzzy imaging at night or light supplementing requirement of traditional monitoring equipment is solved, the outer contours of the tower and the blades are accurately divided based on a semantic segmentation algorithm, and the high-precision double-shaft inclination sensor of the attitude compensation module is combined, so that the clearance distance calculation precision is improved, meanwhile, the dynamic monitoring of key motion parts of the wind driven generator is realized by tracking the blade tip track and analyzing the pneumatic balance state of the blades, the shutdown loss caused by equipment collision is reduced for the wind energy industry, the blade defect is accurately detected, the health grade is output, the safety risk and the shutdown frequency of manual inspection of the wind energy industry are reduced, and the tiny defect is found in advance to avoid fault expansion.

Inventors

REN CHUANMING
LI YUE
REN HONGCHANG
SHEN ZHAOLIANG
DU HAOQING
YU WANQING
ZHENG HAO

Assignees

华能国际电力股份有限公司大连电厂

Dates

Publication Date: 20260512
Application Date: 20251106

Claims (10)

1. A fan blade clearance monitoring system based on AI image recognition is characterized by comprising an image acquisition and imaging module, a blade recognition analysis module, a blade damage detection module, a gesture compensation module and an orientation monitoring module, wherein the modules are used for carrying out data interaction based on a data synchronous verification mechanism to realize fan blade clearance distance monitoring, blade tip track tracking, tower-shaped state monitoring and blade health assessment: The image acquisition and imaging module is used for acquiring video data of the fan blade and the tower, processing the video data by adopting fusion vision AI and calculation imaging, and generating video image data of the fan blade and the tower; The blade identification and analysis module is used for carrying out single-frame segmentation processing on the generated video image data based on a semantic segmentation algorithm, identifying and segmenting the outer contours of the tower barrel and the fan blade, calculating the clearance distance of the blade, simultaneously tracking the blade tip movement track and analyzing the pneumatic balance states of the three blades; the blade damage detection module is used for carrying out model training on the image data after single-frame segmentation processing based on a multi-mode feature fusion deep learning method, detecting the surface defects of the blade and evaluating the health state of the blade; The attitude compensation module is used for measuring the pitch angle and the roll angle of the image acquisition assembly related to the image acquisition and imaging module in real time and carrying out attitude compensation on the calculation result of the clearance distance of the blade; the directional monitoring module is used for measuring the central displacement of the tower barrel, the azimuth angle of the engine room and the inclination angle of the tower barrel in real time based on the Beidou antenna arranged at the top of the fan tower.
2. The AI image recognition-based fan blade clearance monitoring system of claim 1, wherein the image acquisition and imaging module comprises: The original data acquisition unit is used for acquiring video data of the fan blade and the tower barrel in a motion state in real time based on the image acquisition component; The fusion processing unit is used for processing the video data by adopting a technical scheme of fusing visual AI and computational imaging to generate video image data of continuous frames; And the target data output unit is used for synchronously outputting the video image data of the continuous frames to the blade identification and analysis module based on the data synchronous checking mechanism.
3. The fan blade headroom monitoring system based on AI image recognition of claim 2, wherein the visual AI algorithm eliminates temporal noise of video data at very low illumination through multi-frame noise reduction, and combines with the calculated imaging technology video data to perform dynamic scene brightness enhancement and motion compensation, thereby realizing picture optimization of the video data.
4. The fan blade clearance monitoring system based on AI image recognition of claim 3, wherein the process of the blade recognition analysis module recognizing and segmenting the outer contours of the tower and the fan blade comprises: carrying out single-frame segmentation processing on continuous frame video image data synchronously transmitted by a target data output unit, and generating a standardized single-frame image matched with semantic segmentation algorithm processing; Performing feature recognition on the standardized single-frame image based on a preset semantic segmentation model, classifying and marking each pixel in the standardized single-frame image, distinguishing a tower region from a blade region, and respectively outputting an initial external contour of the tower and an initial external contour of a fan blade; The method comprises the steps of combining characteristic association of adjacent frame images, and correcting dynamic boundaries generated by fan blade movement in a standardized single frame image; optimizing the initial external contour, eliminating noise interference in the initial external contour, generating tower external contour data and fan blade external contour data, and extracting key characteristic parameters of the tower and fan blade external contour data; And carrying out association correction on the outer contour data of the tower barrel and the outer contour data of the fan blade and pitch angle and roll angle data transmitted by the attitude compensation module in real time, carrying out blade clearance distance calculation after correction, and synchronously inputting the blade clearance distance calculation to the blade damage detection module.
5. The AI image recognition-based fan blade headroom monitoring system of claim 4, wherein the preset semantic segmentation model comprises a target tower segmentation model and a target fan blade segmentation model, and the constructing process comprises: Marking the standard single-frame image as an original sample, marking the pixel ranges of the tower barrel, the fan blade and the background, generating a marked sample set, and performing data enhancement processing on the marked sample set; Based on the fan monitoring scene characteristics, respectively constructing a segmentation model frame adapting to the columnar structure of the tower and the dynamic form of the blade; Calculating the matching degree of the tower profile output by the target tower segmentation model and the labeling profile based on the tower label in the labeling sample set, and iteratively adjusting the model parameters of the target tower segmentation model until the identification accuracy of the tower region reaches a preset accuracy threshold; Optimizing the identification accuracy of the blade edge in a motion state based on the blade labels in the labeling sample set and the motion track of the blades in the adjacent frames until the identification accuracy of the blade region reaches a preset accuracy threshold; and establishing a characteristic interaction mechanism of the target tower barrel segmentation model and the target fan blade segmentation model based on the spatial association relation between the fan blades and the tower barrel, and forming a preset semantic segmentation model based on the characteristic interaction mechanism.
6. The AI image recognition-based fan blade clearance monitoring system of claim 5 wherein the specific process of calculating the blade clearance distance is as follows: Extracting the center coordinates and the radius of the tower barrel based on the corrected external contour data of the tower barrel and the external contour data of the fan blade, and screening an edge pixel point set in the rotating process of the fan blade; Respectively calculating the linear distance between each pixel in the fan blade edge pixel point set and the center of the tower cylinder, acquiring the actual distance between each part of the fan blade and the surface of the tower cylinder by combining the radius of the tower cylinder, and extracting the minimum value as the blade clearance distance in a single frame image; extracting tip coordinates of each frame of external contour data of the fan blade in the continuous frame images, and correlating tip positions of adjacent frames to generate a motion trail curve of a single blade; and comparing motion track curve parameters of the three fan blades, analyzing the synchronicity and amplitude difference of the motion of each blade, judging the pneumatic balance state of the blade, and outputting a balance early warning signal based on the judgment result.
7. The AI image recognition-based fan blade clearance monitoring system of claim 6, wherein the blade damage detection module comprises: The data receiving unit is used for receiving the standardized single-frame image, screening samples with the image definition meeting a preset image definition threshold value and constructing a blade surface image data set; The multi-modal model training unit is used for carrying out model training on the blade surface image dataset by adopting a multi-modal feature fusion deep learning method and outputting a trained multi-modal model; The defect detection unit is used for inputting the standardized single-frame image acquired in real time and subjected to segmentation and recognition into a trained multi-mode model, traversing the blade area, detecting the position and the range of the defect in the blade area, and outputting the defect type and the quantization parameter; the health state evaluation unit is used for establishing a blade health state evaluation standard according to the defect type, the quantization parameter and the distribution condition and evaluating the health grade of the fan blade.
8. The fan blade clearance monitoring system based on AI image recognition of claim 7, wherein the attitude compensation module is implemented by: The double-shaft inclination angle sensor is integrated in the clearance monitoring box body, the double-shaft inclination angle sensor and the image acquisition assembly keep fixed relative positions, and zero calibration is carried out on the sensor through the horizontal reference instrument during initial installation; The double-shaft inclination sensor acquires pitch angle data and roll angle data of the image acquisition component in real time, and synchronously transmits the pitch angle data and the roll angle data to the blade identification and analysis module, and keeps consistent with the contour processing time sequence; the blade identification and analysis module establishes a space attitude error compensation model based on the received pitch angle data and roll angle data and combined with fixed parameters of the image acquisition assembly, and corrects the calculation result of the clearance distance of the blade based on the space attitude error compensation model.
9. The system of claim 8, wherein the orientation monitor module further comprises a monitor terminal for transmitting monitor data of the tower center displacement, the nacelle azimuth angle and the tower inclination angle to the monitor terminal in real time, generating a tower top displacement trajectory graph, triggering an early warning signal when the monitor data exceeds a preset parameter threshold, and synchronizing the monitor data to a fan control system for adjusting an operation parameter based on the tower center displacement and the tower inclination angle.
10. The fan blade clearance monitoring method based on AI image recognition, which is applied to the fan blade clearance monitoring system based on AI image recognition as claimed in claim 1, is characterized by comprising the following steps: Collecting video data of a fan blade and tower drum, generating continuous frame video image data by fusing visual AI and computational imaging technology, and simultaneously, collecting the attitude data of an image collecting assembly by utilizing a high-precision double-shaft inclination sensor; carrying out single frame segmentation processing on continuous frame video image data to generate a standardized single frame image, segmenting and outputting external contour data of the tower barrel and the fan blade based on a preset semantic segmentation model, and correcting the external contour data based on the posture data of the image acquisition component; Calculating the clearance distance between the fan blade and the tower barrel based on the corrected external contour data, extracting the tip coordinates of continuous frames to generate a motion track curve of the fan blade, and analyzing the motion synchronism and amplitude difference of the fan blade; blade damage detection and health state assessment, namely constructing a blade surface image dataset based on a segmented single frame image, detecting fan blade defects in real time through a multi-mode model, and assessing the health grade of the fan blade; and the tower cylinder state monitoring and operation control support is used for calculating the central displacement of the tower cylinder, the azimuth angle of the engine room and the inclination angle of the tower cylinder in a three-dimensional coordinate system through the Beidou double antennas, generating a tower top displacement track diagram, and synchronizing the tower top displacement track diagram to a fan control system for adjusting the rotating speed and the yaw angle.

Description

Fan blade clearance monitoring system and method based on AI image recognition Technical Field The invention relates to the technical field of wind power generation, in particular to a fan blade clearance monitoring system and method based on AI image recognition. Background The invention discloses a fan blade clearance automatic monitoring method and system based on multiple laser heads, which is disclosed in China patent publication No. CN111878319B, wherein a point is fixed at a position of a ground clearance value from the wall of a tower, a laser range finder with multiple laser heads is arranged on a cabin, one laser head beam strikes a critical position of lower blade tip triggering safety clearance, other laser head beams strike a region near the lower blade tip, under the action of the multiple laser heads, the blades enter the non-safety clearance region to monitor multiple points of the region near the lower blade tip, so that clearance false report caused by non-blade behaviors is avoided, the range finder is provided with a visible positioning laser for determining the position of the laser when adjusting the angle of the range finder, if the blades run in the safety clearance region, the range finder measures the distance between the distance finder and the ground laser, if the blades enter the non-safety clearance region, the distance between the distance finder and the blade tip is measured, the running of the fan can be controlled, and the safety judgment system can be made, and the safety judgment can be made. The existing monitoring means rely on a laser ranging principle, are easily affected by severe environments such as low illumination, rain, snow and fog, and the like, laser penetrability and ranging accuracy are reduced at night or in complex weather, stable monitoring is difficult, particularly, under complex weather conditions, collision risks of blades, towers and obstacles are difficult to effectively avoid, the safety of fans is threatened, serious economic loss is possibly caused, the influence of attitude deviation of monitoring equipment caused by cabin vibration on a measuring result is not considered, only the local area of a lower blade tip is monitored, the blade tip track of a full-motion period of the blade cannot be tracked, and the dynamic running state of the blade is difficult to comprehensively grasp. Disclosure of Invention The invention aims to provide a fan blade clearance monitoring system and method based on AI image recognition, which are used for realizing stable monitoring by fusing visual AI and computational imaging technology to break through the limit of severe environment on monitoring, improving clearance distance calculation accuracy by combining a semantic segmentation algorithm and an attitude compensation module, eliminating equipment attitude deviation interference and covering full-period track tracking of blades so as to solve the problems in the background technology. In order to achieve the above purpose, the present invention provides the following technical solutions: a fan blade clearance monitoring system based on AI image recognition comprises an image acquisition and imaging module, a blade recognition analysis module, a gesture compensation module, a directional monitoring module and a blade damage detection module, wherein each module carries out data interaction based on a data synchronous verification mechanism, so that fan blade clearance distance monitoring, blade tip track tracking, tower-shaped state monitoring and blade health assessment are realized: The image acquisition and imaging module is used for acquiring video data of the fan blade and the tower, processing the video data by adopting fusion vision AI and calculation imaging, and generating video image data of the fan blade and the tower; The blade identification and analysis module is used for carrying out single-frame segmentation processing on the generated video image data based on a semantic segmentation algorithm, identifying and segmenting the outer contours of the tower barrel and the fan blade, calculating the clearance distance of the blade, simultaneously tracking the blade tip movement track and analyzing the pneumatic balance states of the three blades; the blade damage detection module is used for carrying out model training on the image data after single-frame segmentation processing based on a multi-mode feature fusion deep learning method, detecting the surface defects of the blade and evaluating the health state of the blade; The attitude compensation module is used for measuring the pitch angle and the roll angle of the image acquisition assembly related to the image acquisition and imaging module in real time and carrying out attitude compensation on the calculation result of the clearance distance of the blade; the directional monitoring module is used for measuring the central displacement of the tower barrel, the azimuth angle of the engine room