A detection method suitable for detecting tilt defects of a grading ring
By constructing radial and axial lines and combining convolutional neural networks and edge detection algorithms, the tilt state of the equalizing ring can be accurately determined, solving the problem that existing technologies cannot identify tilt faults in the equalizing ring. This improves the recognition rate and accuracy, reduces the influence of the shooting angle, and detects potential weld cracking risks.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SKILL TRAINING CENT STATE GRID JIBEI ELECTRONICS POWER COMPANY
- Filing Date
- 2022-12-20
- Publication Date
- 2026-07-24
AI Technical Summary
Existing deep learning neural networks cannot accurately identify whether there is a tilting fault in the equalizing ring, especially for minor differences where the deformation is small or needs to be judged based on the degree of deformation.
By acquiring images of the equalizing ring and insulator, radial and axial lines are constructed, the tilt angle is calculated, and the tilt state of the equalizing ring is determined by combining convolutional neural networks and edge detection algorithms, and the determination is made by setting a threshold.
It improves the identification rate and accuracy of tilting defects in equalizing rings, reduces the impact of shooting angle on calculation results, enables early detection of potential weld cracking risks, and improves the accuracy and efficiency of inspection.
Smart Images
Figure CN115953368B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of power line inspection technology, and specifically to a method for detecting tilting defects in equalizing rings. Background Technology
[0002] An equipotential bonding ring is a ring-shaped hardware component used in transmission lines to improve the voltage distribution of insulator strings. Its function is to protect against side-strike lightning strikes. It is suitable for alternating current (AC) voltage and can evenly distribute high voltage around the object, ensuring no potential difference between different parts of the ring. This results in a uniform electric field strength at the installation location, reducing corona discharge and achieving the effect of voltage equalization, thus protecting the insulator surface from high-voltage burns. Equalizing rings can be classified according to their application, such as surge arrester equipotential bonding rings, lightning protection equipotential bonding rings, insulator equipotential bonding rings, instrument transformer equipotential bonding rings, high-voltage testing equipment equipotential bonding rings, and transmission and transformation line equipotential bonding rings.
[0003] When installing equipotential bonding rings, their installation direction must be consistent with the axial direction of the insulators and core rods. Regular inspection of equipotential bonding rings is a crucial measure to ensure the safe operation of power systems. In the past, transmission lines relied on manual on-site inspections to detect faults in equipotential bonding rings. With the development and application of technology, technologies for automated inspection of equipotential bonding ring defects using drones have emerged. Currently, equipotential bonding ring fault identification relies on drone aerial photography to acquire a large number of inspection photos, followed by manual review of these photos and supplemented by machine image recognition. This reduces manual workload and improves inspection efficiency to some extent. However, current machine image recognition cannot completely replace manual identification. Defect identification algorithms based on deep learning neural networks have their limitations. These algorithms primarily identify defects in target equipment based on shape and color features, and can only identify defects with obvious deformation characteristics. For cases with minor deformation or where the degree of deformation is crucial for determining defect presence, such as subtle differences like equipotential bonding ring tilting, existing deep learning neural networks cannot accurately determine whether the equipotential bonding ring has tilted. Summary of the Invention
[0004] The technical problem to be solved by the present invention is to provide a method for detecting tilt defects in equalizing rings, so as to improve the recognition rate and accuracy and make the judgment results more accurate.
[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows.
[0006] A method for detecting tilting defects in equalizing rings mainly includes the following steps:
[0007] S1: Obtain an image of the area where the equalizing ring is located, and perform image recognition on the equalizing ring and insulator in the area image;
[0008] S2: Obtain the contours of the equalizing ring and the insulator, obtain two first reference points at the location of the equalizing ring and multiple second reference points at the location of the insulator, and construct the radial line of the equalizing ring based on the first reference points;
[0009] S3: Construct multiple third reference points based on multiple second reference points, and construct the axial line of the mandrel through multiple third reference points;
[0010] S4: Obtain the slope of the radial line and the slope of the axial line, calculate the tilt angle between the equalizing ring and the mandrel, and determine the tilt state of the equalizing ring based on the set threshold.
[0011] In a preferred embodiment, the step prior to obtaining the image of the area where the equalizing ring is located further includes:
[0012] S11. Obtain the 3D model of the power pole / tower;
[0013] S12. Generate UAV flight path and image acquisition points based on the coordinates of the equipotential rings on the power poles / towers;
[0014] S13. After the UAV flies to the image acquisition point, it acquires images of the area where the equalization ring is located, obtaining images of the area where the equalization ring is located.
[0015] In a preferred embodiment, the location of the image acquisition point includes at least two of the following directions: east, west, south, north, southeast, southwest, northeast, and northwest of the equalization ring.
[0016] In a preferred embodiment, the step prior to image recognition of the equalizing ring and insulator in the regional image further includes: constructing a neural network model based on a convolutional neural network, the main steps of which include:
[0017] S21: Acquire images of the equalizing ring sample to obtain a first original image set, and use the first original image set to train a convolutional neural network to obtain a first training model;
[0018] S22: Acquire images of insulator samples to obtain a second original image set, use the second original image set to train a convolutional neural network to obtain a second training model, and merge the data of the first training model and the second training model to obtain an initial model;
[0019] S23: Collect third-party data to enhance the initial model and obtain a neural network model.
[0020] In a preferred embodiment, the specific steps for constructing the radial line of the equalizing ring based on the first reference point include:
[0021] S24: Establish two first reference points based on the contour of the equalizing ring. The two first reference points include: the highest vertical point and the lowest horizontal point.
[0022] S25: Construct a straight line based on the highest vertical point and the farthest horizontal point to obtain the radial line of the equalizing ring.
[0023] In a preferred embodiment, the main steps of constructing multiple third reference points based on multiple second reference points and constructing the axial line of the mandrel through the third reference points include:
[0024] S31. Construct radial lines of the insulator using two reference points located on the same insulator, and obtain the midpoints of multiple radial lines;
[0025] S32. Perform curve fitting on multiple midpoints to obtain the axial line of the mandrel.
[0026] In a preferred embodiment, the main steps of determining the tilt state of the equalizing ring based on a set threshold include:
[0027] S41: Obtain the tilt threshold;
[0028] S42: Calculate the tilt angle of the equalizing ring;
[0029] S43: Obtain the tilt angle determination value through the tilt angle. If the tilt angle determination value is less than the tilt threshold, it is determined that the equalizing ring is in a balanced state. If the tilt angle determination value is greater than the tilt threshold, it is determined that the equalizing ring is in a tilted state.
[0030] In a preferred embodiment, the main steps of identifying the region image of the equalizing ring in a balanced state and determining whether there is a potential risk include:
[0031] S51: Identify the region image of the equalizing ring in equilibrium;
[0032] S52: Record the equipotential rings with weld cracks, assess their potential risks, and generate an analysis report.
[0033] Due to the adoption of the above technical solutions, the technical progress achieved by this invention is as follows.
[0034] After training a convolutional neural network to obtain a neural model, this invention acquires a regional image of the equalizing ring and identifies the contours of the equalizing ring and the insulator. It then establishes the radial line of the equalizing ring and the axial line of the core rod, calculates the angle between the radial line and the axial line, and obtains the angle between the equalizing ring and the core rod. This allows the determination of whether the equalizing ring is tilted, thus improving the recognition rate and accuracy.
[0035] This invention can accurately obtain the radial line of the equalizing ring by modifying the algorithm, reducing the influence of contour deviation caused by different orientations of regional image acquisition, thereby facilitating the accurate calculation of the angle between the equalizing ring and the mandrel, reducing the influence of the shooting angle on the calculation results, and making the judgment results more accurate.
[0036] This invention uses a neural network model to identify regional images of the equalizing ring, thereby identifying whether the equalizing ring has cracks. Based on the corrosion rate of the equalizing ring material and the initial length of the crack, the safe service life of the equalizing ring is calculated. This facilitates the early detection of weld cracks in the equalizing ring caused by environmental corrosion, welding factors, and material factors, and helps to identify potential risks to the equalizing ring. Attached Figure Description
[0037] Figure 1 This is a flowchart of the present invention;
[0038] Figure 2 This is a schematic diagram of the equalizing ring described in this invention. Detailed Implementation
[0039] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.
[0040] A method for detecting tilting defects in equalizing rings mainly includes the following steps:
[0041] S1: Obtain an image of the area where the equalizing ring is located, and perform image recognition on the equalizing ring and insulator in the area image;
[0042] S2: Obtain the contours of the equalizing ring and the insulator through the edge detection algorithm, obtain two first reference points at the location of the equalizing ring and multiple second reference points at the location of the insulator, and construct the radial line of the equalizing ring based on the first reference points;
[0043] S3: Construct multiple third reference points based on multiple second reference points, and construct the axial line of the mandrel through multiple third reference points;
[0044] S4: Obtain the slope of the radial line and the slope of the axial line, calculate the tilt angle between the equalizing ring and the mandrel, and determine the tilt state of the equalizing ring based on the set threshold.
[0045] Furthermore, the steps prior to obtaining the image of the area where the equalizing ring is located also include:
[0046] S11. Obtain the 3D model of the power pole / tower, generate the coordinates of all power fittings based on the geodetic coordinate system, and classify and sort them according to the type of power fittings.
[0047] S12. Based on the classification of power fittings, select the voltage equalization ring target to be inspected, and generate the UAV flight path and image acquisition points based on the coordinates of the voltage equalization ring to be inspected. Each voltage equalization ring shall be image acquired at least twice. The location of the image acquisition points includes at least two of the following directions: east, west, south, north, southeast, southwest, northeast, and northwest of the voltage equalization ring.
[0048] Furthermore, in S12, two acquisition points are used to establish straight lines with the coordinate points of the equalization ring, respectively, denoted as L1 and L2. The angle between L1 and L2 on the horizontal plane is less than 180 degrees. In this embodiment, the angle between L1 and L2 is preferably 45 degrees to 135 degrees. For example, when acquiring regional images of the equalization ring, the first image acquisition point is located due east of the equalization ring, and the second image acquisition point is located in one of the directions of due south, due north, southeast, southwest, northeast, and northwest. As another example, when acquiring regional images of the equalization ring, the first image acquisition point is located in the southeast of the equalization ring, and the second image acquisition point is located in one of the directions of due east, due west, due south, due north, southwest, and northeast.
[0049] S13. The UAV performs inspection operations. After the UAV flies to the image acquisition point, it acquires images of the area where the equalizing ring is located.
[0050] It should be noted that a matching drone inspection system is used in conjunction with drones for inspection tasks, through which the drone's patrol path can be planned.
[0051] In one specific implementation, a geodetic coordinate system is established in 3D software, and a 3D model of the power pole / tower is created based on the geodetic coordinate system. Simultaneously, the coordinates of all power fittings (such as equipotential rings, insulators, crossarms, suspension fittings, connecting fittings, etc.) are generated. The power fittings are classified and numbered, and a model database is established. When an equipotential ring needs to be inspected, the target equipotential ring to be inspected is selected based on the model database. The UAV inspection system uses the Bat Optimization (BA), Differential Evolutionary Algorithm, Ant Colony Algorithm (ACO), or other algorithms to generate an inspection path. Image acquisition points are generated on the inspection path based on the coordinates of the equipotential ring to be inspected. After the inspection path is generated, the UAV performs the inspection operation. After the UAV flies to the image acquisition point according to the inspection path, it acquires images of the area where the equipotential ring is located and numbers the images according to the number of the equipotential ring. By processing and calculating the images of this area, the tilt angle of the equipotential ring is obtained.
[0052] Furthermore, the steps prior to image recognition of the equalizing ring and insulator in the regional image also include: constructing a neural network model based on a convolutional neural network, the main steps of which include:
[0053] S21: Acquire images of the equalizing ring sample to obtain a first original image set, and use the first original image set to train a convolutional neural network to obtain a first training model;
[0054] S22: Acquire images of insulator samples to obtain a second original image set, use the second original image set to train a convolutional neural network to obtain a second training model, and merge the data of the first training model and the second training model to obtain an initial model;
[0055] S23: Collect network data from third-party platforms, train convolutional neural networks, enhance the initial model, and obtain a neural network model.
[0056] In one specific implementation, a convolutional neural network (CNN) is constructed, comprising at least convolutional layers, pooling layers, fully connected layers, and activation functions. After the CNN is constructed, a first image set is input into it. The convolutional layers use a convolutional kernel to slide across the images in the set, extracting image features (such as horizontal edges) from each image. Then, an activation function is used to suppress gradient vanishing. The extracted feature data is further extracted using another convolutional kernel. Pooling layers are then used to shrink the image, reducing pixel information while retaining important information. Finally, a fully connected layer performs recognition and classification on the results. At this point, the first training model is obtained. The same operation is performed on a second original image set to obtain a second training model. The first training model is then used as the main database, and the second training model... As a database, two databases are merged to obtain an initial model. Equalizing ring graphics and insulator images are collected from a third-party platform via the network and used for neural network training, thereby enhancing the initial model and ultimately obtaining a neural network model. Specifically, the activation function can be any one or more of the following functions: Stpigmoid activation function, Tanh activation function, ReLU activation function, LReLU activation function, PReLU activation function, StpwiStph activation function, or other nonlinear activation functions. Among them, the ReLU activation function is preferred because it is computationally simple, has sparse representativeness, and can avoid gradient vanishing. In this embodiment, the activation function of the convolutional layer is preferably the ReLU activation function, and the activation function of the fully connected layer is preferably the Stpigmoid activation function.
[0057] Furthermore, the specific steps for constructing the radial line of the equalizing ring based on the first reference point include:
[0058] S24: Establish two first reference points based on the contour of the equalizing ring. The two first reference points include: the highest vertical point and the lowest horizontal point.
[0059] S25: Construct a straight line based on the highest vertical point and the farthest horizontal point to obtain the radial line of the equalizing ring.
[0060] In one specific implementation, the contour of the equalizing ring is obtained through an edge algorithm. Based on the contour of the equalizing ring, the vertical highest point P1 and the horizontal farthest lowest point P2 are identified. A straight line is constructed based on P1 and P2 to obtain the radial line of the equalizing ring.
[0061] Furthermore, the main steps of constructing multiple third reference points based on multiple second reference points, and constructing the axial line of the mandrel through the third reference points, include:
[0062] S31. Construct radial lines of the insulator using two reference points located on the same insulator, and obtain the midpoints of multiple radial lines;
[0063] S32. Perform curve fitting on multiple midpoints to obtain the axial line of the mandrel.
[0064] Furthermore, the main steps for determining the tilt state of the equalizing ring based on the set threshold include:
[0065] S41: Obtain the tilt threshold;
[0066] S42: Calculate the tilt angle of the equalizing ring. The formula is as follows:
[0067]
[0068] Where α is the angle between the radial line and the axial line, K1 is the slope of the radial line, and K2 is the slope of the axial line;
[0069] S43: Determine the tilt angle and obtain the tilt angle determination value. The determination formula is as follows:
[0070]
[0071] Wherein, θ is the tilt angle determination value, and α is the absolute value of the tilt angle. If the tilt angle determination value θ is less than the tilt threshold, the equalizing ring is determined to be in a balanced state. If the tilt angle determination value θ is greater than the tilt threshold, the equalizing ring is determined to be in a tilted state.
[0072] In one specific implementation, the slopes K1 of the radial line and K2 of the axial line are obtained, the tilt angle α of the equalizing ring is calculated, the absolute value of α is taken, and substituted into the determination formula for judgment. If α is less than or equal to 90 degrees, the tilt angle determination value θ is the same as α; if α is greater than 90 degrees, the tilt angle determination value θ is 180-α; if the tilt angle determination value θ is less than the tilt threshold, the equalizing ring is in a balanced state; if the tilt angle determination value θ is greater than the tilt threshold, the equalizing ring is in a tilted state. For example, the tilt threshold... If the value is set to 5°, and α is -35°, then the value of α is 35°. Substituting this into the judgment formula, since the value of α is less than 90°, the value of θ is 35°. Since the value of θ is greater than the tilt threshold, the equalizing ring is in a tilted state. For example, if the tilt threshold is set to 5°, and α is 178°, then the value of α is 178°. Substituting this into the judgment formula, since the value of α is greater than 90°, the value of θ is 2°. Since the value of θ is less than the tilt threshold, the equalizing ring is in a balanced state.
[0073] It should be noted that since the equalizing ring is fitted onto the outside of the insulator, the tilt angle of the equalizing ring will not be 90°.
[0074] Furthermore, the main steps for identifying the region image of the equalizing ring in a balanced state and determining whether there is a potential risk include:
[0075] S51: The neural network model is used to identify the area image of the equalizing ring in equilibrium state, and to identify whether there is weld cracking in the equalizing ring in equilibrium state. The equalizing ring without weld cracking is judged as normal state, and the equalizing ring with weld cracking is judged as risk state.
[0076] S52: Record the initial crack length L1 of the equalizing ring with cracks and the number of the equalizing ring through image recognition;
[0077] S53: Set the safety threshold L for the equalizing ring, and calculate the safe service time y of the equalizing ring based on the corrosion rate η of the equalizing ring material. The formula for calculating the safe service time is:
[0078] Specifically, the corrosion rate η can be obtained by conducting an electrochemical corrosion test on the material of the equalizing ring. The corrosion rate calculation formula is as follows: Where m0 is the weight before the corrosion test and m1 is the weight after the corrosion test. It should be noted that the electrochemical corrosion test can simulate the corrosion environment in nature and thus obtain the corrosion rate of the material.
[0079] In one specific embodiment, the neural network model obtained in S23 is used to identify the area image of the equalizing ring in a balanced state to identify whether there are cracks in the equalizing ring. If no cracks are identified in the equalizing ring, it is determined to be in a normal state; if cracks are found, it is determined to be in a risk state. The length L1 of the crack and the number of the equalizing ring are obtained from the area image of the equalizing ring. Substitute these into the calculation formula to obtain the safe service life of the equalizing ring and generate an analysis report. For example, if the corrosion rate of low carbon steel in the atmosphere is 0.35 mm / year, and the neural network model identifies an 8.86 mm crack in a low carbon steel equalizing ring, and the safety threshold of the equalizing ring is set to 10 mm, the safe service life of the equalizing ring can be obtained as 3.2 years by substituting into the calculation formula, and an analysis report is generated. The report includes at least the number of the equalizing ring, the area image, and the safe service life. By checking the equalizing ring in a balanced state, the possibility of weld cracking caused by environmental corrosion, welding factors, and material factors can be detected in advance.
[0080] The above description is merely a preferred embodiment of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention. Structures, devices, and operating methods not specifically described or explained in this invention are implemented according to conventional methods in the art unless otherwise specified or limited.
Claims
1. A method for detecting tilting defects in equalizing rings, characterized in that: The main steps include: S1: Obtain an image of the area where the equalizing ring is located, and perform image recognition on the equalizing ring and insulator in the area image; S2: Obtain the contours of the equalizing ring and the insulator, obtain two first reference points at the location of the equalizing ring and multiple second reference points at the location of the insulator, and construct the radial line of the equalizing ring based on the first reference points; The specific steps for constructing the radial line of the equalizing ring based on the first reference point include: S24: Establish two first reference points based on the contour of the equalizing ring. The two first reference points include: the highest vertical point and the farthest and lowest horizontal point. S25: Construct a straight line based on the highest vertical point and the farthest and lowest horizontal point to obtain the radial line of the equalizing ring; S3: Construct multiple third reference points based on multiple second reference points, and construct the axial line of the mandrel through these third reference points; the main steps include: S31. Construct radial lines of the insulator using two reference points located on the same insulator, and obtain the midpoints of multiple radial lines; S32. Perform curve fitting on multiple midpoints to obtain the axial line of the mandrel; S4: Obtain the slope of the radial line and the slope of the axial line, calculate the tilt angle between the equalizing ring and the mandrel, and determine the tilt state of the equalizing ring based on the set threshold. S5: Identify the region image of the equalizing ring in equilibrium to determine if there are any potential risks; the main steps include: S51: Identify the region image of the equalizing ring in equilibrium; S52: Record the equipotential rings with weld cracks, assess their potential risks, and generate an analysis report; specifically, this includes recording the initial crack length of the equipotential rings with cracks. L 1. The number of the equalizing ring, and based on the corrosion rate of the equalizing ring material. η According to the formula Calculate the safe service life of the equalizing ring. y ,in, L The system sets a preset safety threshold and generates an analysis report containing the equalizing ring number, area image, and safe usage time to assess potential risks.
2. The method for detecting tilt defects in equalizing rings according to claim 1, characterized in that: The steps prior to obtaining the image of the region where the equalizing ring is located also include: S11. Obtain the 3D model of the power pole / tower; S12. Generate UAV flight path and image acquisition points based on the coordinates of the equipotential rings on the power poles / towers; S13. After the UAV flies to the image acquisition point, it acquires images of the area where the equalization ring is located, obtaining images of the area where the equalization ring is located.
3. The method for detecting tilt defects in equalizing rings according to claim 2, characterized in that: The image acquisition points are located at least two of the following directions: east, west, south, north, southeast, southwest, northeast, and northwest of the equalization ring.
4. The method for detecting tilt defects in equalizing rings according to claim 1, characterized in that: The steps prior to image recognition of the equalizing ring and insulator in the regional image also include: constructing a neural network model based on a convolutional neural network, the main steps of which include: S21: Acquire images of the equalizing ring sample to obtain a first original image set, and use the first original image set to train a convolutional neural network to obtain a first training model; S22: Acquire images of insulator samples to obtain a second original image set, use the second original image set to train a convolutional neural network to obtain a second training model, and merge the data of the first training model and the second training model to obtain an initial model; S23: Collect third-party data to enhance the initial model and obtain a neural network model.
5. The method for detecting tilt defects in equalizing rings according to claim 1, characterized in that: The main steps for determining the tilt state of the equalizing ring based on the set threshold include: S41: Obtain the tilt threshold; S42: Calculate the tilt angle of the equalizing ring; S43: Obtain the tilt angle determination value through the tilt angle. If the tilt angle determination value is less than the tilt threshold, it is determined that the equalizing ring is in a balanced state. If the tilt angle determination value is greater than the tilt threshold, it is determined that the equalizing ring is in a tilted state.