Online impeller imbalance detection method and system based on video recognition
By using a camera and laser calibrator on a wind turbine to collect image data and calculate the blade clearance curve and rotation data, the problems of high cost of impeller imbalance detection and complex maintenance in the prior art are solved, and real-time and accurate impeller set status is achieved monitor.
Patent Information
- Application Number
- CN202411271970.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-11
- Publication Date
- 2024-11-12
AI Technical Summary
The existing wind turbine impeller imbalance detection method requires the installation of multiple vibration sensors, which is costly and complex to maintain, and it is difficult to monitor the blade clearance curve and rotation data in real time.
Using an online detection method based on video recognition, the camera and laser calibrator are set up at the bottom of the cabin to collect image data, calculate the blade clearance curve and impeller group rotation data, and combine the camera shaking spectrum data to analyze the impeller group operation status in real time.
It reduces inspection costs and maintenance complexity, realizes real-time monitoring without downtime, improves the accuracy and reliability of inspection results, and simplifies equipment wiring and maintenance.
Smart Images

Figure CN118934501A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of wind power generation, and in particular relates to an online impeller imbalance detection method and system based on video recognition. Background Art
[0002] Since wind turbines work for a long time in conditions with large load changes and harsh environments, fatigue damage, impeller eccentricity and other health indicators may decline as the service time increases. Impeller imbalance may also occur due to production accuracy or installation quality. Impeller imbalance may reduce the wind supply efficiency of the fan, or may cause serious blade failure or even collapse of the tower. Impeller imbalance failure is a common type of fan failure. By monitoring impeller imbalance failure, impeller imbalance can be detected at an early stage, the failure rate can be reduced, and the reliable operation of wind turbines can be enhanced.
[0003] At present, the main detection method is to install vibration sensors on the main shaft of the fan to monitor radial and axial vibrations, and to determine whether there is an imbalance by analyzing the vibration data. This method requires the installation of multiple sensors and collectors, and a certain amount of wiring, and then the data is analyzed and judged. The detection investment cost is high, and when a sensor or circuit fails, the maintenance cost is even higher. Summary of the invention
[0004] In view of the above problems, the present invention provides an online impeller imbalance detection method based on video recognition, the detection method comprising:
[0005] S1, the camera on the cabin collects the measured image with the calibration points, periodically collects the first original image and the second original image, and sends the measured image, the first original image and the second original image to the processor;
[0006] S2, the processor obtains the measured image and the first original image, and calculates the blade clearance curve and the impeller group rotation data according to the measured image and the first original image;
[0007] S3, the processor acquires and performs calculation processing on the second original image to obtain camera shake spectrum data;
[0008] S4. The processor determines the unbalanced operating state of the impeller group based on a comprehensive analysis of the camera shaking spectrum data and the impeller group rotation data.
[0009] Further, the measured image is an image including the calibration blade and the calibration point, which is obtained by the camera shooting from top to bottom when the calibration blade rotates to the bottom of the impeller assembly and the axis of the calibration blade is coplanar with the axis of the tower;
[0010] The first original image is an image containing a first rectangular region of interest;
[0011] The second original image is an image containing a second rectangular region of interest.
[0012] Furthermore, the calculation method of the blade clearance curve and the impeller assembly rotation data includes:
[0013] S21, calculating a mapping formula between pixel distance and actual distance according to the measured image;
[0014] S22, identifying and marking effective detection corner points on each first original image, and measuring the second pixel distance between each effective detection corner point and the calibration point;
[0015] S23, calculating the actual distance between each effective detection corner point and the calibration point according to the mapping formula and the second pixel distance, and constructing the corresponding blade clearance curve;
[0016] S24. Obtain the impeller assembly rotation data based on the blade clearance curve.
[0017] Furthermore, the calculation method of the mapping formula includes:
[0018] S211, measuring and calculating the first pixel distance between the tip of the calibrated blade and the calibration point on the image;
[0019] S212, using a laser calibrator to measure and calculate a first actual distance between the tip of the calibrated blade and the calibration point in real space;
[0020] S213: Calculate a mapping formula according to the first pixel distance and the first actual distance.
[0021] Furthermore, the identification method of the effective detection corner point includes:
[0022] S221, using a segmentation model to perform segmentation processing on the first original image to obtain a second image including a leaf foreground contour and calibration points;
[0023] S222, extracting and processing the second image using a corner point detection model to obtain a third image marked with detection corner points and calibration points;
[0024] S223, checking the validity of each detected corner point in the third image, deleting invalid detected corner points, retaining valid detected corner points, and obtaining a fourth image marked with valid detected corner points and calibration points;
[0025] S224. Measure the second pixel distance between the effective detection corner point and the calibration point in the fourth image.
[0026] Furthermore, the method for calculating the camera shake spectrum data includes:
[0027] S31, performing edge extraction processing on the second original image to obtain an edge line image;
[0028] S32, analyzing the edge line image to generate spectrum data.
[0029] Furthermore, the edge extraction process includes:
[0030] S311, performing enhancement processing on the second original image to obtain a fifth image;
[0031] S312, selecting and capturing a second rectangular region of interest in the fifth image to obtain a sixth image;
[0032] S313: Perform edge detection processing on the sixth image or the seventh image to obtain an edge line image.
[0033] Furthermore, analyzing the edge line image includes:
[0034] S321, collecting camera shake amplitude data; wherein the camera shake amplitude data includes camera left-right shake amplitude data and camera front-back shake amplitude data;
[0035] S322, performing time series Fourier transform on the camera shake amplitude data to obtain camera shake spectrum data.
[0036] Furthermore, the detection method also includes:
[0037] S5. The processor compares the blade clearance curves of multiple blades in the same impeller assembly to determine the status of each blade.
[0038] The present invention also provides an online impeller imbalance detection system based on video recognition, the detection system comprising:
[0039] A laser calibrator, used for setting a calibration point on the tower and for measuring a first actual distance;
[0040] A camera is installed at the bottom of the cabin, and is used to collect a measured image with calibration points, periodically collect a first original image and a second original image, and send the measured image, the first original image and the second original image to a processor;
[0041] The processor is used to obtain the measured image, the first original image and the second original image, and calculate the blade clearance curve and the impeller group rotation data according to the measured image and the first original image; to calculate and process the second original image to obtain the camera shake spectrum data; and to determine the operating status of the wind turbine according to the comprehensive analysis of the camera shake spectrum data and the impeller group rotation data.
[0042] The beneficial effects of the present invention are:
[0043] 1. The detection system proposed by the present invention does not need to install a large number of vibration sensors on the wind turbine, but only needs to install a camera at the bottom of the cabin. The number of system devices is small, and the amount of wiring is small. The detection system is low in cost and easy to maintain.
[0044] 2. The calibration point is determined by a laser calibrator, which can be completed during the operation of the wind turbine without stopping the machine, which is convenient and quick. In addition, each type of wind turbine only needs to be calibrated once during the detection process.
[0045] 3. By further inspecting and processing the third image, invalid detection corner points are deleted and only valid detection corner points are retained, which effectively improves the accuracy of the detection results.
[0046] 4. The present invention uses a camera to periodically capture the first original image of each blade, and can analyze the change in the actual distance between each blade and the calibration point during the detection period. The operation method is simple and easy to implement, which is convenient for subsequent analysis and evaluation of the blade structure performance.
[0047] 5. The present invention can obtain blade clearance curve, impeller speed data and edge line shaking spectrum data by setting a camera on the cabin to collect images and an external processor to perform image processing and analysis. A relatively comprehensive operating status data of the wind turbine can be obtained through a simple detection device.
[0048] 6. The present invention obtains data such as blade clearance curve, impeller rotation data and edge line shaking spectrum data through detection, and comprehensively analyzes the operating status of the wind turbine impeller group to ensure that the detection results are accurate and reliable.
[0049] Other features and advantages of the present invention will be described in the following description, and partly become obvious from the description, or be understood by implementing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures pointed out in the description and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS
[0050] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0051] Figure 1 A schematic diagram showing an installation scenario of a detection device according to an embodiment of the present invention is shown;
[0052] Figure 2 A schematic diagram showing a first original image according to an embodiment of the present invention is shown;
[0053] Figure 3 The clearance curves of three blades of an embodiment of the present invention are shown;
[0054] Figure 4 shows a combined headroom curve of an embodiment of the present invention;
[0055] Figure 5 A schematic diagram showing a second original image according to an embodiment of the present invention is shown;
[0056] Figure 6 A schematic diagram showing an amplified second original image according to an embodiment of the present invention is shown;
[0057] Figure 7 A schematic diagram showing a sixth image of an embodiment of the present invention;
[0058] Figure 8 A schematic diagram showing a seventh image of an embodiment of the present invention;
[0059] Fig. 9 A polar coordinate graph of an embodiment of the present invention is shown;
[0060] Fig.10 A schematic diagram showing an edge line image according to an embodiment of the present invention is shown;
[0061] Fig.11 A schematic diagram of time domain data of the front-to-back (axial) shake amplitude of a camera according to an embodiment of the present invention is shown;
[0062] Fig.12 A schematic diagram of frequency domain data of the amplitude of the front-to-back (axial) shake of a camera according to an embodiment of the present invention is shown.
[0063] In the figure: 1-tower; 2-impeller assembly; 201-calibration blade; 3-cabin; 4-camera; 5-laser calibration instrument. DETAILED DESCRIPTION
[0064] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0065] An embodiment of the present invention provides an online impeller imbalance detection method based on video recognition, the detection method comprising:
[0066] S1. A camera on the cabin collects a measured image with calibration points, periodically collects a first original image and a second original image, and sends the measured image, the first original image and the second original image to a processor.
[0067] Specifically, the calibration point is set on the tower of the wind turbine, and the calibration point is the position of the calibration blade tip projected on the tower in the horizontal direction when the calibration blade rotates to the bottom of the impeller assembly and the calibration blade axis is coplanar with the tower axis.
[0068] It should be noted that, generally, the impeller assembly of a wind turbine generator includes multiple blades, for example, three blades. For ease of understanding, during the calibration process, one blade in the impeller assembly is named as the calibration blade.
[0069] For example, Figure 1 As shown, the wind turbine includes a tower 1, an impeller group 2 and a nacelle 3; the nacelle 3 is installed at the top of the tower 1, and the impeller group 2 is installed at the front end of the nacelle 3. The camera 4 is arranged at the bottom of the nacelle 3 and between the impeller group 2 and the tower 1. The laser calibrator 5 is arranged on the ground directly in front of the wind turbine. The laser calibrator 5 shoots a laser beam toward the side of the tower 1 in the horizontal direction, forms a first light spot on the tower 1, and measures the horizontal distance D3 between the laser calibrator 5 and the tower. The actual height of the tower 1 and the actual length of the calibration blade 201 are both factory data and known data. When the calibration blade 201 rotates to the bottom of the impeller group and the axis of the calibration blade 201 is coplanar with the axis of the tower 1, the root of the calibration blade 201 is at the same height as the top of the tower 1. The actual height H1 of the tip of the calibration blade 201 can be calculated by subtracting the actual length of the calibration blade 201 from the actual height of the tower 1. According to the horizontal distance D3 and the actual height H1, the pitch angle of the laser calibrator 5 or the actual length of the calibration line can be calculated. The laser calibrator 5 shoots a laser beam according to the calculated pitch angle or calibration line, forming a light spot on the tower 1, which is the calibration point.
[0070] The calibration point is determined by a laser calibrator, which can be completed during the operation of the wind turbine without stopping the machine, which is convenient and quick. In addition, each type of wind turbine only needs to be calibrated once during the inspection process.
[0071] Furthermore, the measured image is an image including the calibration blade and the calibration point, which is obtained by shooting from top to bottom by the camera when the calibration blade rotates to the bottom of the impeller assembly and the axis of the calibration blade is coplanar with the axis of the tower.
[0072] For example, Figure 1As shown, the camera 4 is arranged at the bottom of the nacelle 3 of the wind turbine generator, and the camera 4 shoots from top to bottom. When the calibration blade 201 rotates to the bottom of the impeller assembly 2 and the axis of the calibration blade 201 is coplanar with the axis of the tower 1, the camera 4 shoots. The measured image obtained by shooting includes the calibration blade 201 and the calibration points on the tower 1.
[0073] Further, the first original image is an image including a first rectangular region of interest. When each blade rotates to the bottom of the impeller assembly, the camera captures an image including the blade and a calibration point on the tower in the first rectangular region of interest.
[0074] like Figure 2 As shown, a rectangular region of interest (ROI) is set in the camera's field of view window, which is recorded as the first rectangular region of interest. When each blade rotates to the bottom of the impeller assembly and enters the first rectangular region of interest, the camera captures the first original image containing the blade and the calibration point on the tower from top to bottom. If there is no blade image in the first rectangular region of interest, the camera is in a waiting state.
[0075] It should be noted that in real scenes, the rotation speed of the blade is not fixed, and the camera periodically shoots at a fixed shooting frequency. For ease of understanding, the time from a blade entering the first rectangular area of interest to leaving the first rectangular area of interest can be recorded as a blade entry time period. In a blade entry time period, the camera may capture multiple first original images, and the position of the first rectangular area of interest in each first original image in the field of view window may be different.
[0076] Preferably, within a blade entry time period, an image with the shortest pixel distance between the blade tip corner point and the calibration point is selected as the first original image.
[0077] The first original image is used to analyze the clearance distance between the blade and the tower and the rotation speed data of the impeller group. The image with the shortest pixel distance between the tip corner point and the calibration point in each blade entry time period is selected as the first original image to ensure that the analysis result is accurate and effective.
[0078] Furthermore, the second original image is an image including a second rectangular region of interest, wherein the second rectangular region of interest includes an image of a section of the tower root edge.
[0079] For example, Figure 5 As shown, a rectangular region of interest (ROI) is set in the visual field window of the camera, which is recorded as the second rectangular region of interest. The position of the second rectangular region of interest is the root edge of a section of the tower.
[0080] It should be noted that in order to ensure that the collected images are clear and accurate, you can zoom in and shoot. Figure 6 The second rectangular region of interest in the second original image is used to analyze the camera shaking, so the position of the second rectangular region of interest in the camera's field of view window remains fixed during the entire detection process.
[0081] The detection method proposed by the present invention does not need to install a large number of vibration sensors on the wind turbine generator set, and only requires to set up a camera at the bottom of the cabin, which is low in cost and easy to maintain.
[0082] S2. The processor obtains the measured image and the first original image, and calculates the blade clearance curve and the impeller assembly rotation data according to the measured image and the first original image.
[0083] Specifically, the calculation method of the blade clearance curve and the impeller assembly rotation data includes:
[0084] S21. Calculate a mapping formula between pixel distance and actual distance based on the measured image.
[0085] Furthermore, the calculation method of the mapping formula includes:
[0086] S211, measuring and calculating the first pixel distance between the tip of the calibrated blade and the calibration point on the image;
[0087] S212, using a laser calibrator to measure and calculate a first actual distance between the tip of the calibrated blade and the calibration point in real space;
[0088] S213: Calculate a mapping formula according to the first pixel distance and the first actual distance.
[0089] Specifically, a pixel point whose pixel distance between the leaf tip position and the calibration point is closest on the measured image is selected as the leaf tip pixel point, and the pixel distance between the leaf tip pixel point and the calibration point is measured and recorded as the first pixel distance.
[0090] Preferably, a laser calibrator is used to mark the tip of the blade, and the measurement image captured by the camera includes the tip laser mark and the calibration point on the tower. The pixel distance between the tip laser mark and the calibration point on the tower in the measurement image is measured and recorded as the first pixel distance.
[0091] Furthermore, the first actual distance between the tip of the calibration blade and the calibration point is measured and calculated by a laser calibrator.
[0092] For example, Figure 1As shown, when the calibration blade 201 rotates to the bottom of the impeller assembly and the axis of the calibration blade is coplanar with the axis of the tower, the laser calibrator 5 shoots a laser beam toward the tip of the calibration blade 201. The laser beam passes through the tip of the calibration blade 201 and is projected onto the tower 1, forming a second light spot on the tower 1, and the actual distance D1 between the laser calibrator 5 and the tip of the calibration blade 201 and the actual distance D4 between the laser calibrator 5 and the second light spot are measured (not shown in the figure). The actual distance D2 between the tip of the calibration blade 201 and the second light spot can be calculated.
[0093] The laser calibrator 5 shoots a laser beam toward the side of the tower 1 in the horizontal direction to form a first light spot on the tower 1, and the horizontal distance D3 between the laser calibrator 4 and the tower is measured.
[0094] According to the actual distance D4 and the horizontal distance D3, the actual height difference H3 between the second light spot and the laser calibrator 4 can be calculated using a trigonometric function calculation formula.
[0095] The actual height H4 of the laser calibrator 5 is known; the actual height of the tower 1 and the actual length of the calibration blade 201 are both known factory data. When the calibration blade 201 rotates to the bottom of the impeller assembly and the axis of the calibration blade is coplanar with the axis of the tower, the root of the calibration blade 201 is at the same height as the top of the tower 1. The actual height H1 of the tip of the calibration blade 201 can be calculated by subtracting the actual length of the calibration blade 201 from the actual height of the tower 1.
[0096] The actual height difference H2=H3+H4-H1 between the second light spot and the tip of the calibration blade 201. According to the actual distance D2 and the actual height difference H2, the first actual distance Y1 between the tip of the calibration blade 201 and the calibration point can be calculated using a trigonometric calculation formula.
[0097] Furthermore, a ratio K of the first actual distance to the first pixel distance is calculated and obtained, and a mapping formula is constructed as Y=K*X, where X is the pixel distance and Y is the actual distance.
[0098] Furthermore, the calculation method of the blade clearance curve and the impeller assembly rotation data also includes:
[0099] S22, identifying and marking effective detection corner points on each first original image, and measuring the second pixel distance between each effective detection corner point and the calibration point.
[0100] Specifically, the identification method of the effective detection corner point includes:
[0101] S221, using a segmentation model to perform segmentation processing on the first original image to obtain a second image including a leaf foreground contour and calibration points;
[0102] S222, extracting and processing the second image using a corner point detection model to obtain a third image marked with detection corner points and calibration points;
[0103] S223, checking the validity of each detected corner point in the third image, deleting invalid detected corner points, retaining valid detected corner points, and obtaining a fourth image marked with valid detected corner points and calibration points;
[0104] S224. Measure the second pixel distance between the effective detection corner point and the calibration point in the fourth image.
[0105] Exemplarily, the segmentation model may adopt but is not limited to the PP-LiteSeg segmentation model. The PP-LiteSeg segmentation model introduces technologies such as the unified attention fusion module (UAFM) and the pyramid pooling module (SPPM), achieving the goal of improving the inference speed while maintaining high accuracy. Its flexible and lightweight decoder (FLD) design further reduces the computational burden and improves the model inference speed.
[0106] The corner detection model adopts but is not limited to Harris corner detection technology and least squares technology.
[0107] By further inspecting and processing the third image, invalid detected corner points are deleted and only valid detected corner points are retained, thereby effectively improving the accuracy of the detection result.
[0108] Exemplarily, the third image is overlapped and compared with the second image, and the leaf foreground contour in the second image is used as a reference. If the detected corner point exceeds the range of the leaf foreground contour, it means that the detected corner point is invalid and the detected corner point is deleted. Then the pixel distance between the remaining detected corner points and the marked point is calculated, and the detected corner point with the shortest pixel distance is taken as the valid detected corner point, and finally a fourth image marked with a valid detected corner point and a calibration point is obtained.
[0109] Furthermore, the method for calculating the blade clearance curve and the impeller assembly rotation data also includes:
[0110] S23. Calculate the actual distance between each effective detection corner point and the calibration point according to the mapping formula and the second pixel distance, and construct a corresponding blade clearance curve.
[0111] Specifically, the second pixel distance between the effective detection corner point and the calibration point in the fourth image is substituted into the mapping formula to obtain the actual distance between the effective detection corner point and the calibration point.
[0112] It should be noted that, within the same blade entry time period, multiple fourth images of the same blade can be acquired, and multiple pixel distances also correspond between the effective detection corner point on the blade and the calibration point. Preferably, within the same blade entry time period, the shortest pixel distance between the effective detection corner point on the blade and the calibration point is selected as the second pixel distance between the effective detection corner point and the calibration point.
[0113] Since the blade is a three-dimensional irregular structure and will be deformed by external forces such as wind resistance during operation, when the blade rotates to the bottom of the impeller group and the blade axis is coplanar with the tower axis, the actual distance between the effective detection corner point and the tower may not be the closest actual distance. By selecting the shortest pixel distance within the same blade entry time period as the second pixel distance corresponding to the effective detection corner point, it can be ensured that the analysis is the closest distance between each effective detection corner point and the tower, ensuring the accuracy of the detection results.
[0114] Furthermore, according to the shooting time sequence, the actual distances between the effective detection corner points and the calibration points on the same blade acquired at different times are counted, and a curve graph of the actual distances between the effective detection corner points and the calibration points is constructed. Each blade corresponds to a blade clearance curve.
[0115] For example, the actual distance data between the effective detection corner points and the calibration points on the same blade are marked in sequence in the figure according to the shooting time sequence, and are connected in sequence with smooth curves to form the blade clearance curve of the blade. Figure 3 As shown, an impeller assembly includes three blades, which correspond to three blade clearance curves. From each blade clearance curve, it can be seen how the actual distance between the corresponding blade and the calibration point changes during the detection period.
[0116] The present invention uses a camera to periodically capture and obtain the first original image of each blade, and can analyze the change of the actual distance between each blade and the calibration point during the detection period. The operation method is simple, easy to implement, and convenient for subsequent analysis and evaluation of the blade structure performance.
[0117] Furthermore, the method for calculating the blade clearance curve and the impeller assembly rotation data also includes:
[0118] S24. Obtain the impeller assembly rotation data based on the blade clearance curve.
[0119] Specifically, the impeller assembly rotation data includes the rotation speed of each blade and the rotation frequency of the impeller assembly.
[0120] For example, an impeller assembly includes three blades, and the clearance data of the three blades are marked in chronological order, such as Figure 4As shown in the figure, the horizontal axis is time and the vertical axis is distance. Each vertex represents a blade clearance data, that is, it represents the actual distance between a blade and the calibration point, and the detection time corresponding to the blade clearance data. For each rotation of the impeller group, four blade clearance data can be obtained, corresponding to the first blade, the second blade, the third blade and the first blade in sequence. Therefore, all clearance data are divided into groups of three in chronological order. For any two adjacent groups of clearance data, the detection time of the first clearance data in the latter group is subtracted from the detection time of the last clearance data in the previous group to obtain the detection duration of the previous group, that is, the speed of the impeller group corresponding to the previous group for one rotation. Then, the rotation frequency of the impeller group over a period of time can be calculated.
[0121] S3. The processor acquires and performs calculation processing on the second original image to obtain camera shake spectrum data.
[0122] Specifically, the method for calculating the camera shake spectrum data includes:
[0123] S31 . Perform edge extraction processing on the second original image to obtain an edge line image.
[0124] Specifically, the edge extraction process includes:
[0125] S311. Perform enhancement processing on the second original image to obtain a fifth image.
[0126] Specifically, the enhancement processing adopts but is not limited to a histogram normalization method to enhance the contrast of the second original image to obtain the fifth image.
[0127] Exemplarily, the second original image is enhanced according to formula (1) to formula (3):
[0128] O r,c =a*I(r,c +b (1)
[0129]
[0130] Where I(r, c) represents the gray value of the rth row and cth column of the second original image I; I min represents the minimum gray value appearing in the second original image I; I max represents the maximum gray value appearing in the second original image I; O(r, c) represents the gray value of the rth row and cth column of the fifth image O; O min represents the minimum gray value in the fifth image O, which is a setting parameter; max Represents the maximum grayscale value in the fifth image O, which is a setting parameter.
[0131] Enhanced processing improves the visual effect of the image, facilitates subsequent processing and analysis, and improves the accuracy of the detection results.
[0132] Furthermore, the edge extraction process further includes:
[0133] S312: Select and capture a second rectangular region of interest in the fifth image to obtain a sixth image.
[0134] like Figure 7 As shown, the sixth image shows an image of a section of the tower root edge enclosed by the second rectangular area of interest.
[0135] It should be noted that the purpose of capturing the second rectangular region of interest in the image is to determine the image content that needs to be analyzed later and to reduce analysis interference.
[0136] Preferably, the edge extraction process further includes: S3121, optimizing the sixth image to obtain a seventh image.
[0137] Specifically, the optimization process adopts but is not limited to a global threshold binarization method.
[0138] Exemplarily, the sixth image is subjected to global threshold binarization processing according to formula (4):
[0139]
[0140] Wherein, T represents the global threshold value, which is a setting parameter; Px,y represents the pixel gray value of any pixel point P in the sixth image; P,(x,y represents the pixel gray value of the pixel point P in the sixth image after global thresholding, that is, the pixel gray value of the corresponding pixel point P in the seventh image. Figure 8 As shown, when the pixel grayscale value is 0, it means that the pixel is black; when the pixel grayscale value is 255, it means that the pixel is white.
[0141] The sixth image is converted into a binary image with only two values through global threshold binarization processing, thereby simplifying the image information and highlighting the edge of the cabin.
[0142] Furthermore, the edge extraction process further includes:
[0143] S313: Perform edge detection processing on the sixth image or the seventh image to obtain an edge line image.
[0144] Specifically, the edge detection process adopts but is not limited to a Sobel operator processing method to detect edge pixel points in the sixth image or the seventh image.
[0145] Exemplarily, the following two sets of matrices are used, which are respectively the operators of horizontal x and vertical y. The two sets of matrices are plane-convolved with the image respectively, so as to obtain the approximate values of the horizontal and vertical brightness differences respectively.
[0146]
[0147] A represents the seventh image, G x and G y Respectively represent the grayscale values of the image after horizontal and vertical edge detection, and their formulas are formula (5) and formula (6) respectively:
[0148]
[0149]
[0150] The gradient |G| represents the approximate horizontal and vertical grayscale values of each pixel, and its calculation formula is:
[0151] |G|=|G x |+|G y | (7)
[0152] If the gradient |G| of a certain pixel point of the seventh image is greater than a preset threshold, the pixel point is considered to be an edge pixel point, and its coordinates on the seventh image are recorded.
[0153] It should be noted that a rectangular coordinate system needs to be constructed on the seventh image to mark the coordinate position of each pixel point. The origin position of the rectangular coordinate system is not limited.
[0154] Furthermore, the edge line is calculated and determined using a linear polar coordinate conversion formula to obtain an image with the edge line drawn.
[0155] Exemplarily, the linear polar coordinate conversion formula is:
[0156] r=xcos(θ)+ysin(θ) (8)
[0157] Among them, r is the vertical distance from the origin of the rectangular coordinate system in the seventh image to the desired edge line; x is the horizontal coordinate of the current pixel point in the rectangular coordinate system; y is the vertical coordinate of the current pixel point in the rectangular coordinate system; θ is the angle between the desired edge line and the positive direction of the horizontal coordinate of the rectangular coordinate system, which is a variable.
[0158] The rectangular coordinates of each edge pixel point on the seventh image are converted into a sine curve in the parameter space r-θ by the linear polar coordinate conversion formula, as shown in Fig. 9 As shown in Figure 1. Each edge pixel corresponds to a sine curve. When two of the sine curves intersect, it means that the two edge pixels corresponding to the two sine curves pass through the same straight line.
[0159] When the number of curves intersecting at the same point exceeds a threshold, the straight line is considered to be an edge line, and an edge line image is drawn.
[0160] Preferably, the edge extraction process further includes: S314, performing edge line validity analysis on the edge line image to retain an edge line.
[0161] For example, when the edge line image contains multiple edge lines, first delete the edge lines whose lengths are not within the length threshold range; then select the edge line with the longest length and delete the other edge lines. If there is only one edge line at this time, then this edge line is the valid edge line; if there are multiple edge lines at this time, you can select any one as the valid edge line and delete the other edge lines. Fig.10 shown.
[0162] Furthermore, the method for calculating the camera shake spectrum data also includes:
[0163] S32, analyzing the edge line image to generate spectrum data.
[0164] Specifically, analyzing the edge line image includes:
[0165] S321, collecting camera shake amplitude data; wherein the camera shake amplitude data includes camera left-right shake amplitude data and camera front-back shake amplitude data.
[0166] Exemplarily, the center point of the edge line image is taken as the origin of the two-dimensional coordinate axis, and the coordinates of the center point of the edge line are recorded. The horizontal coordinate of the center point of the edge line is the amplitude data of the camera shaking left and right, and the vertical coordinate of the center point of the edge line is the amplitude data of the camera shaking forward and backward.
[0167] It should be noted that if Figure 5 The figure shows the camera's shooting angle. When the camera shakes with the cabin, it can be understood that the edge line of the tower root shakes relative to the camera. Taking the tower main axis direction as the reference, the camera's shaking amplitude data in the direction perpendicular to the tower main axis is recorded as the camera's left and right shaking amplitude data; the camera's shaking amplitude data in the direction parallel to the tower main axis is recorded as the camera's forward and backward shaking amplitude data.
[0168] S322, performing time series Fourier transform on the camera shake amplitude data to obtain camera shake spectrum data.
[0169] For example, the camera forward and backward shaking amplitude data collected from each edge line image is statistically analyzed in the order of collection time and plotted as follows: Fig.11 The time domain data of the camera's forward and backward (axial) shaking amplitude is shown. The horizontal axis in the figure is time, and the vertical axis is the camera's forward and backward shaking amplitude at the corresponding time point.
[0170] The time domain data of the camera's forward and backward shaking amplitude is calculated by time series Fourier transform, and the calculation formula is:
[0171]
[0172] Wherein, N is the total number of samples in one detection cycle; n represents the line number sequence number; Δt represents the sampling time interval; k represents the point number sequence number; X represents the frequency domain; and x represents the time domain.
[0173] like Fig.12 As shown, the frequency domain data of the camera's front-to-back (axial) shake amplitude obtained by calculating through time-series Fourier transform.
[0174] The present invention can obtain blade clearance curve, impeller rotation data and edge line shaking spectrum data by setting a camera on the cabin to collect images and an external processor to perform image processing and analysis. A relatively comprehensive operating status data of the wind turbine can be obtained through a simple detection device.
[0175] S4. The processor determines the unbalanced operating state of the impeller group based on a comprehensive analysis of the camera shaking spectrum data and the impeller group rotation data.
[0176] For example, Fig.12 As shown, the frequency domain data is analyzed to see whether there is data consistent with the rotation frequency of the impeller group. If the frequency spectrum data of the camera shaking forward and backward contains the rotation frequency of the impeller group, it is determined that the impeller group has aerodynamic imbalance; if the frequency spectrum data of the camera shaking left and right contains the rotation frequency of the impeller group, it is determined that the impeller group has mass imbalance.
[0177] At the same time, the amplitude of the frequency domain data can be analyzed and compared with the frequency domain data during the entire detection process. If the amplitude of the camera's front-to-back (axial) shaking frequency is too large and exceeds the preset value, it means that the installation angles of the blades in the impeller group are inconsistent or the blades are damaged, resulting in aerodynamic imbalance; if the amplitude of the camera's left-to-right (lateral) shaking frequency is too large and exceeds the preset value, it means that the blades are damaged and cause mass imbalance.
[0178] Preferably, the detection method further comprises:
[0179] S5. The processor compares the blade clearance curves of multiple blades in the same impeller assembly to determine the status of each blade.
[0180] Since multiple blades of the same impeller group have exactly the same structural performance status when they leave the factory; when working, due to the same working environment, if none of the blades are damaged and the structural performance remains good, the change trend of the clearance curves of each blade is the same; if a blade is damaged, the change trend of the blade clearance curve of the blade will be different from the change trend of the blade clearance curves of other blades.
[0181] For example, the impeller assembly includes three blades, and the time for one blade to rotate one circle is recorded as one detection cycle. In one detection cycle, the visual sensor takes pictures of the three blades respectively. Figure 3 As shown in the figure, it is the blade clearance curve of three blades in the same impeller group within a detection cycle. By comparison, it can be seen that the change trend of the blade clearance curve corresponding to the second blade is basically consistent with the change trend of the blade clearance curve corresponding to the third blade; while the change trend of the blade clearance curve corresponding to the first blade is obviously different from the blade clearance curves of the second blade and the third blade. At this time, it can be determined that the first blade is damaged.
[0182] The present invention obtains data such as blade clearance curve, impeller group rotation data and edge line shaking spectrum data through detection, comprehensively analyzes the operating status of the wind turbine impeller group, and ensures that the detection result is accurate and reliable.
[0183] In order to implement the above-mentioned online impeller imbalance detection method based on video recognition, an embodiment of the present invention further provides an online impeller imbalance detection system based on video recognition, the detection system comprising:
[0184] A laser calibrator, used for setting a calibration point on the tower and for measuring a first actual distance;
[0185] A camera is installed at the bottom of the cabin, and is used to collect a measured image with calibration points, periodically collect a first original image and a second original image, and send the measured image, the first original image and the second original image to a processor;
[0186] The processor is used to obtain the measured image, the first original image and the second original image, and calculate the blade clearance curve and the impeller group rotation data according to the measured image and the first original image; to calculate and process the second original image to obtain the camera shake spectrum data; and to determine the operating status of the wind turbine according to the comprehensive analysis of the camera shake spectrum data and the impeller group rotation data.
[0187] For example, Figure 1 As shown, the wind turbine includes a tower 1, an impeller group 2 and a nacelle 3; the nacelle 3 is installed at the top of the tower 1, and the impeller group 2 is installed at the front end of the nacelle 3. The camera 4 is arranged at the bottom of the nacelle 3 and is located between the impeller group 2 and the tower 1.
[0188] The detection system provided by the present invention has a simple structure, and only one camera needs to be installed on the wind turbine, and the ground equipment uses a laser calibrator and a processor. The number of system equipment is small, and the amount of wiring is small.
[0189] Preferably, the processor is also used to compare blade clearance curves of multiple blades in the same impeller assembly to determine the status of each blade.
[0190] The present invention obtains data such as blade clearance curve, impeller group rotation data and edge line shaking spectrum data through detection, comprehensively analyzes the operating status of the wind turbine impeller group, and ensures that the detection result is accurate and reliable.
[0191] Furthermore, the detection system also includes:
[0192] The display is used to display the blade clearance curve and the unbalanced operating status information of the impeller group.
[0193] The detection system provided by the present invention can detect the blade structure during the operation of the wind turbine without stopping the machine, which greatly improves the efficiency of wind turbine blade detection and reduces labor input. By comparing and analyzing the clearance curves of multiple blades, the structural status of each blade can be effectively and comprehensively detected, ensuring the accuracy and timeliness of the detection results.
[0194] Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent substitutions for some of the technical features therein; and these modifications or substitutions 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 present invention.
Claims
1. An online impeller imbalance detection method based on video recognition, characterized in that: The detection method comprises: S1, the camera on the cabin collects the measured image with the calibration points, periodically collects the first original image and the second original image, and sends the measured image, the first original image and the second original image to the processor; S2, the processor obtains the measured image and the first original image, and calculates the blade clearance curve and the impeller group rotation data according to the measured image and the first original image; S3, the processor acquires and performs calculation processing on the second original image to obtain camera shake spectrum data; S4. The processor determines the unbalanced operating state of the impeller group based on a comprehensive analysis of the camera shaking spectrum data and the impeller group rotation data.
2. The method for online impeller imbalance detection based on video recognition according to claim 1 is characterized in that: The measured image is an image including the calibration blade and the calibration point, which is obtained by the camera shooting from top to bottom when the calibration blade rotates to the bottom of the impeller assembly and the axis of the calibration blade is coplanar with the axis of the tower; The first original image is an image containing a first rectangular region of interest; The second original image is an image containing a second rectangular region of interest.
3. The method for online impeller imbalance detection based on video recognition according to claim 1, characterized in that: The calculation method of the blade clearance curve and the impeller group rotation data includes: S21, calculating a mapping formula between pixel distance and actual distance according to the measured image; S22, identifying and marking effective detection corner points on each first original image, and measuring the second pixel distance between each effective detection corner point and the calibration point; S23, calculating the actual distance between each effective detection corner point and the calibration point according to the mapping formula and the second pixel distance, and constructing the corresponding blade clearance curve; S24. Obtain the impeller assembly rotation data based on the blade clearance curve.
4. The method for online impeller imbalance detection based on video recognition according to claim 3 is characterized in that: The calculation method of the mapping formula includes: S211, measuring and calculating the first pixel distance between the tip of the calibrated blade and the calibration point on the image; S212, using a laser calibrator to measure and calculate a first actual distance between the tip of the calibrated blade and the calibration point in real space; S213: Calculate a mapping formula according to the first pixel distance and the first actual distance.
5. The method for online impeller imbalance detection based on video recognition according to claim 3 is characterized in that: The identification method of the effective detection corner point comprises: S221, using a segmentation model to perform segmentation processing on the first original image to obtain a second image including a leaf foreground contour and calibration points; S222, extracting and processing the second image using a corner point detection model to obtain a third image marked with detection corner points and calibration points; S223, checking the validity of each detected corner point in the third image, deleting invalid detected corner points, retaining valid detected corner points, and obtaining a fourth image marked with valid detected corner points and calibration points; S224. Measure the second pixel distance between the effective detection corner point and the calibration point in the fourth image.
6. The method for online impeller imbalance detection based on video recognition according to claim 1, characterized in that: The method for calculating the camera shake spectrum data comprises: S31, performing edge extraction processing on the second original image to obtain an edge line image; S32, analyzing the edge line image and generating spectrum data.
7. The method for online impeller imbalance detection based on video recognition according to claim 6, characterized in that: The edge extraction process comprises: S311, performing enhancement processing on the second original image to obtain a fifth image; S312, selecting and capturing a second rectangular region of interest in the fifth image to obtain a sixth image; S313: Perform edge detection processing on the sixth image or the seventh image to obtain an edge line image.
8. The method for online impeller imbalance detection based on video recognition according to claim 6, characterized in that: The analyzing edge line image comprises: S321, collecting camera shake amplitude data; wherein the camera shake amplitude data includes camera left-right shake amplitude data and camera front-back shake amplitude data; S322, performing time series Fourier transform on the camera shake amplitude data to obtain camera shake spectrum data.
9. An online impeller imbalance detection method based on video recognition according to any one of claims 1 to 8, characterized in that: The detection method further comprises: S5. The processor compares the blade clearance curves of multiple blades in the same impeller assembly to determine the status of each blade.
10. An online impeller imbalance detection system based on video recognition, characterized in that: The detection system comprises: A laser calibrator, used for setting a calibration point on the tower and for measuring a first actual distance; A camera is installed at the bottom of the cabin, and is used to collect a measured image with calibration points, periodically collect a first original image and a second original image, and send the measured image, the first original image and the second original image to a processor; The processor is used to obtain the measured image, the first original image and the second original image, and calculate the blade clearance curve and the impeller group rotation data according to the measured image and the first original image; to calculate and process the second original image to obtain the camera shake spectrum data; and to determine the operating status of the wind turbine according to the comprehensive analysis of the camera shake spectrum data and the impeller group rotation data.
Citation Information
Cited By
Fault diagnosis method and fault diagnosis system based on multi-source data fusion
CN119540234A
Non-contact type fan blade running state detection device and method
CN120487526A
Dynamic balance evaluation method for impeller of axial flow fan based on machine vision
CN120609503A
Balance monitoring method, monitoring equipment and monitoring system of wind generating set
CN121544646A
