Strain clamp defect detection method based on unmanned aerial vehicle digital X-ray

Abstract

The invention discloses a strain clamp defect detection method based on unmanned aerial vehicle digital X-rays, which comprises the following steps: acquiring a strain clamp image, and marking strain clamp defect characteristics off line; constructing a strain clamp image crimping defect detection model, optimizing convolution modules in the feature extraction network and the feature fusion network, optimizing a model bounding box loss function, and improving bounding box regression precision; setting training parameters of a strain clamp image crimping defect detection model, training to obtain a strain clamp image crimping defect detection model, inputting the strain clamp crimping defect image into the strain clamp image crimping defect detection model, and outputting defect confidence and a bounding box; and rating the crimping defects of the strain clamp through the related defect parameters. The method has great practical engineering application value for reducing the working intensity of strain clamp detection and improving the defect detection efficiency and accuracy.

Description

Technical Field

[0001] The present invention relates to the field of intelligent detection technology, and in particular to a method for detecting defects of tension clamps based on unmanned aerial vehicle digital X-rays. Background Art

[0002] Strain clamps are important hardware in overhead lines, and their performance is directly related to the safe and stable operation of the lines. With the continuous expansion of the scale of power grids and the increasing requirements for power supply reliability, the defect detection of strain clamps has become crucial. The strain clamp defect detection method based on drone digital X-ray can overcome the shortcomings of traditional detection methods such as complex operation, low efficiency and low accuracy, and achieve efficient, accurate and comprehensive defect detection, providing strong technical support for ensuring the safe operation of power grids.

[0003] Among the existing related technologies based on drone digital X-ray detection and tension clamp defect detection before the present invention, there are several comparative patents and documents: CN202411284457.X, CN202411173902.5, and CN202411027895.8.

[0004] The above-mentioned specific patent comparative documents are:

[0005] 1) "A method for managing detection information of tension clamps based on X-ray image recognition", patent number CN202411284457.X. The present invention relates to a method for managing detection information of tension clamps based on X-ray image recognition, which collects X-ray detection images of tension clamps in a large historical period and their corresponding defect types; pre-processes the detection images; extracts the main features of the X-ray detection images as input, and the corresponding defect types as output, and trains a convolutional neural network model; obtains the X-ray detection images of the tension clamps in real time, inputs them into the convolutional neural network model, and outputs the defect type detection results; classifies and stores the obtained X-ray detection images into the corresponding defect types according to the detection results; clusters the stored X-ray detection images of the same defect type, and divides the defect levels. The present invention solves the problem that the prior art lacks unified classification management of defect detection information of X-ray images of tension clamps, and it is difficult for people to quickly call the corresponding defect types through detection images, resulting in low detection efficiency. The defect grade classification of the present invention is different from that of the above invention. There are quantitative standards for the defects of steel anchor groove leakage and underpressure, aluminum wire not pressed, and steel anchor internal non-crimped defects, and direct assessment is performed according to the defect detection results.

[0006] 2) "An easily adjustable tension clamp and its detection method", patent number CN202411173902.5. The present invention provides an easily adjustable tension clamp and its detection method, which belongs to the field of power transmission and transformation technology. The easily adjustable tension clamp includes a detection frame, a detection slot is provided at the top of the detection frame, and two limit holes are provided on the inner walls of both sides of the detection slot. The operator controls the electric valve plate to make the tension clamp aluminum part fall into the discharge frame, and then moves the tension clamp aluminum part into the environmental sleeve through the first screw gear and rack. The temperature control device in the sleeve simulates different environments, such as high temperature, high humidity and salt spray. At the top of the high humidity connecting frame and the bearing plate, after two infrared sensors sense the tension clamp aluminum part, the environmental detection mechanism is started, and the corrosion resistance and stability of the tension clamp under different environments are evaluated through the second screw clamping device. At the same time, the conductivity test device is used to evaluate the conductivity, reflecting the purity and internal structural integrity of the aluminum part, thereby evaluating its tension resistance. This method can simultaneously collect data of the tension clamp aluminum part. The defect detection method of the present invention is different from the above-mentioned invention. The present invention adopts digital X-ray to take pictures of the tension clamp, and detects the pictures to achieve detection.

[0007] 3) "A non-destructive testing device for power transmission line tension clamp based on UAV platform", patent number CN202411027895.8. The present invention relates to the technical field of analysis equipment, specifically a non-destructive testing device for power transmission line tension clamp based on UAV platform, comprising a UAV main body, the lower ends of the wings of the UAV main body are respectively fixedly connected with four groups of first electric telescopic rods, the lower ends of the four groups of the first electric telescopic rods are all fixedly connected with airbags, the lower end of the UAV main body is fixedly connected with a mounting plate, the outer side of the mounting plate is provided with a circular slot, the front end of the mounting plate is installed with a detection mechanism, the detection mechanism comprises a C-shaped fixed plate, the inner side of the C-shaped fixed plate is fixedly connected with a ray machine, the inner side of the C-shaped fixed plate is fixedly connected with an imaging plate opposite to the ray machine, and the upper end of the second electric telescopic rod is installed with a spraying mechanism, the device can make the ray machine and the imaging plate rotate 360 ​​degrees, comprehensively detect the outer side of the tension clamp, avoid the single angle, so that the detection result has multiple groups of reference results. The detection method of the present invention is different from the above invention. The above invention is that the X-ray machine and the imaging plate rotate 360 ​​degrees to comprehensively detect the outside of the tension clamp, avoiding the single angle and making the detection result have multiple groups of reference results, but the operation is complicated and the detection time is long. Summary of the invention

[0008] In order to solve the above technical problems, the purpose of the present invention is to provide a method for detecting defects of tension clamps based on drone digital X-rays.

[0009] The purpose of the present invention is achieved through the following technical solutions:

[0010] A method for detecting defects of tension clamps based on UAV digital X-rays, comprising:

[0011] Step A, collecting images of the tension clamp and annotating the defect features of the tension clamp offline;

[0012] Step B, constructing a strain clamp image crimping defect detection model, optimizing the convolution modules in the feature extraction network and the feature fusion network, and optimizing the model bounding box loss function to improve the bounding box regression accuracy;

[0013] Step C, setting the training parameters of the tension clamp image crimping defect detection model and training to obtain the tension clamp image crimping defect detection model, inputting the tension clamp image crimping defect detection model into the tension clamp image crimping defect detection model, and outputting the defect confidence and bounding box;

[0014] Step D: Rating the tension clamp crimping defects by using relevant defect parameters.

[0015] Compared with the prior art, one or more embodiments of the present invention may have the following advantages:

[0016] The present invention can identify the defects of the tension clamp based on image detection and grade the degree of the defects, and can be effectively applied to the detection and identification of the tension clamp by the digital X-ray of the UAV. The adoption of this method has great practical engineering application value in reducing the intensity of the tension clamp detection work and improving the efficiency and accuracy of defect detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a flow chart of the method for detecting defects of tension clamps based on UAV digital X-ray;

[0018] Figure 2 This is the SIB-YOLOv8 network model diagram;

[0019] Figure 3 It is a schematic diagram of the SPD-conv method;

[0020] Figure 4 It is a schematic diagram of the BiFormer module method;

[0021] Figure 5 It is a schematic diagram of the I-IoU method;

[0022] Figure 6a , 6b and 6c are the results of the strain clamp defect detection. DETAILED DESCRIPTION

[0023] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be further described in detail below in conjunction with embodiments and drawings.

[0024] like Figure 1-6c As shown, a schematic diagram of a method for detecting defects in a tension clamp based on a UAV digital X-ray is provided, wherein the method for detecting defects in a tension clamp includes the following steps:

[0025] Step 10: Collect the image of the tension clamp and annotate the defect features of the tension clamp offline;

[0026] Step 20: construct a strain clamp image crimping defect detection model, optimize the convolution modules in the feature extraction network and the feature fusion network, and at the same time, optimize the model bounding box loss function to improve the bounding box regression accuracy;

[0027] Step 30, setting the training parameters of the tension clamp image crimping defect detection model and training to obtain the tension clamp image crimping defect detection model, inputting the tension clamp image crimping defect detection model into the tension clamp image crimping defect detection model, and outputting the defect confidence and bounding box;

[0028] Step 40: Rating the tension clamp crimping defects by using relevant defect parameters.

[0029] The above step 10 includes: collecting defect images of the transmission line tension clamp through a digital X-ray device carried by an unmanned aerial vehicle; using the open source tool labelme to mark various defects such as leakage and underpressure in the steel anchor groove, bending of the steel anchor, unpressurized aluminum wire, unpressurized internal connection of the steel anchor, broken steel core, burrs, and water accumulation in the cavity.

[0030] The above step 20 includes: the image crimping defect detection model of the tension clamp is a SIB-YOLO model, the SIB-YOLO model uses YOLOv8 as the basic network, uses SPD-Conv to optimize the CBS module of the initial network; uses the BiFomer attention mechanism to optimize the feature extraction and feature fusion network; uses I-IoU to optimize the loss function module of the initial network to complete the automatic detection of tension clamp defects. The SPD layer in the SPD-Conv module uses interval sampling to convert the input feature with the number of channels (S, S, C1) Figure 4 Divide it into (S / 2, S / 2, C1) sub-feature maps with equal number of channels. When the scale is 2, the four sub-feature mappings of the defect are as follows:

[0031] f (0,0) =X[0:S:scale,0:S:scale] (1)

[0032] f (1,0) =X[1:S:scale,0:S:scale] (2)

[0033] f (0,1)=X[0:S:scale,1:S:scale] (3)

[0034] f (1,1) =X[scale-1:S:scale,scale-1:S:scale] (4)

[0035] Where scale∈{1,2}, the sub-feature maps are concatenated along the channel dimension to obtain the intermediate feature map X′(S / 2,S / 2,4C1), and the spatial information is converted into depth information. Then the non-strided convolutional layer converts X′ into the final feature map X″(S / 2,S / 2,C2) through a convolutional layer with a stride of 1, which can not only reduce the number of channels but also retain the global spatial feature information in the channel dimension, effectively enhancing the detection performance of the defect feature information of the tension clamp.

[0036] In SIB-YOLO, the BiFormer module of the Backbone part is added to the last layer, so that the network can adaptively adjust the attention weight according to the input image features; the BiFormer module of the neck part is added before the psample structure and after the C2f module. The BiFormer processing process is as follows: for the input image X∈R H×W×C Where H is the input image height, W is the input image width, and C is the number of input image channels. It is divided into A×A different regions, each of which contains feature vector, the input image 2D feature map X r for: Combined with the linear transformation formula, the query tensor Q, key tensor K, and value tensor V after linear projection of the input image can be expressed as:

[0037]

[0038] Among them, W q , W k , W v They are respectively the query, key, and value projection weights, and we get Combined with the average values ​​of Q and K, we can get Where Q r is the region-level query tensor, K r is the region-level key tensor, the adjacency matrix A of the correlation between regions r The calculation formula is as follows:

[0039] A r =Q r (K r ) T (6)

[0040] topkIndex(·) is the index function of the topk regions, and the inter-region index routing matrix I r The calculation formula is as follows:

[0041] I r =topkIndex(A r ) (7)

[0042] Through the index matrix I r Aggregate the scattered routing areas and get the key tensor K of the aggregated area. g 、value tensor V g for:

[0043]

[0044] Combined with the local context enhancement term LCE(V), the calculation formula for the fine-grained token-to-token attention mechanism is as follows:

[0045] O=Attention(Q,K g ,V g )+LCE(V) (9)

[0046] The I-IoU definition formula in SIB-YOLO is as follows:

[0047]

[0048] Where: T r is the scale factor; b gt and b represent the true box and the predicted box respectively, as well as Respectively represent the left boundary, right boundary, upper boundary, and lower boundary of the real box, b l 、b r 、b t and b b Respectively represent the left boundary, right boundary, upper boundary, and lower boundary of the prediction box; and are the inner center points of the real frame; x c and c They are the center points inside the prediction box respectively; I is the intersection of the real box and the prediction box, U is the union of the real box and the prediction box, IoU I represents the ratio of the two; w and h are the height and width of the prediction box respectively; W gt 、h gt are the height and width of the real frame respectively.

[0049] T rThe value range is usually [0.5, 1.5]. The scale factor is used to control the scale of the auxiliary bounding box for loss calculation. The I-IoU loss function calculation formula is as follows:

[0050] L I-CIoU =L CIoU +IoU-IoU I (12)

[0051] The above step 30 specifically includes: using a self-made data set, dividing the tension clamp defects into steel anchor groove leakage and underpressure, steel anchor bending, aluminum wire not pressed, steel anchor internal not crimped, steel core fracture, flash, cavity water, and dividing the tension clamp defect image data set into 2000 defect image data sets in a ratio of 7:2:1. The model training parameter settings include the optimizer as SGD, the initial learning rate as 0.01, the batch size as 8, and the number of training rounds as 50.

[0052] The above step 40 specifically includes: according to the identified defects, the following rating parameters are calculated for the steel anchor groove leakage and under-pressure, aluminum wire not pressed, steel anchor internal not crimped, etc.:

[0053] Crimping condition between steel anchor and aluminum tube: the severity is determined by whether there is a gap in the steel anchor groove and the ratio of the number of pressure-missing grooves to the total number of grooves;

[0054] The crimping condition of aluminum tube and stranded wire: the severity is determined by the ratio of the length of the leakage between the aluminum tube and the stranded wire;

[0055] Crimping condition of anchor tube and core wire: the severity is determined by the proportion of voids in the anchor tube cavity.

[0056] Strain clamp crimp defects are graded including:

[0057] Regarding the crimping of steel anchors and aluminum tubes, if the pressure leakage ratio is <1 / 2, it is judged as a general defect and can be temporarily left untreated. If conditions permit, pressure replenishment can be carried out; if the pressure leakage ratio is ≥1 / 2, it is judged as a major defect and pressure replenishment must be carried out as soon as possible as required.

[0058] Regarding the crimping condition of the anchor pipe and the core wire, if the pressure leakage ratio is less than 15%, it is judged as a general defect and can be temporarily untreated; if the pressure leakage ratio is 15% to 30%, it is judged as a major defect and should be re-pressurized in combination with power outage; if the pressure leakage ratio is >30%, re-pressurization should be carried out immediately as required for critical defects.

[0059] Regarding the crimping situation of aluminum tubes and stranded wires, which is reflected in the crimping parts of aluminum tubes and aluminum wires of steel core aluminum stranded wires and aluminum tubes and aluminum wires of aluminum alloy core aluminum stranded wires, if the pressure leakage ratio between the aluminum tubes and the stranded wires is less than 10%, it can be temporarily unprocessed. If conditions permit, pressure replenishment can be carried out; if the pressure leakage ratio is 10% to 30%, pressure replenishment should be carried out in conjunction with power outages; if the pressure leakage ratio is 30% to 50%, pressure replenishment should be carried out as soon as possible according to the requirements for serious defects; if the pressure leakage ratio reaches 50% or above, pressure should be re-applied immediately according to the requirements for critical defects.

[0060] Although the embodiments disclosed in the present invention are as above, the above contents are only embodiments adopted for facilitating the understanding of the present invention and are not intended to limit the present invention. Any technician in the technical field to which the present invention belongs can make any modifications and changes in the form and details of the implementation without departing from the spirit and scope disclosed in the present invention, but the patent protection scope of the present invention shall still be subject to the scope defined in the attached claims.

Claims

1. A method for detecting defects of tension clamps based on UAV digital X-ray, characterized in that: The method comprises the following steps: Step A, collecting images of the tension clamp and annotating the defect features of the tension clamp offline; Step B, constructing a strain clamp image crimping defect detection model, optimizing the convolution modules in the feature extraction network and the feature fusion network, and optimizing the model bounding box loss function to improve the bounding box regression accuracy; Step C, setting the training parameters of the tension clamp image crimping defect detection model and training to obtain the tension clamp image crimping defect detection model, inputting the tension clamp image crimping defect detection model into the tension clamp image crimping defect detection model, and outputting the defect confidence and bounding box; Step D: Rating the tension clamp crimping defects by using relevant defect parameters.

2. The method for detecting defects of tension clamps based on UAV digital X-ray according to claim 1 is characterized in that: In the step A, Use drones equipped with digital X-rays to collect images of tension clamps; The defects of the tension clamp include pressure leakage and under-pressure in the steel anchor groove, bending of the steel anchor, unpressurized aluminum wire, unpressurized connection inside the steel anchor, broken steel core, burrs, and water accumulation in the cavity.

3. The method for detecting defects of tension clamps based on UAV digital X-ray according to claim 1 is characterized in that: In the step B, the tension clamp image crimping defect detection model is a SIB-YOLO model, which uses YOLOv8 as the basic network, and uses SPD-Conv to optimize the CBS module in the YOLOv8 network; uses the BiFomer attention mechanism to optimize the feature extraction network and the feature fusion network; and uses the I-IoU optimization model bounding box loss function to complete the automatic detection of tension clamp defects.

4. The method for detecting defects of tension clamps based on UAV digital X-ray according to claim 3 is characterized in that: SPD-Conv is used to replace some CBS modules in the YOLOv8 network. The SPD layer in the SPD-Conv module divides the input feature map with the number of channels (S, S, C1) into four equal parts (S / 2, S / 2, C1) sub-feature maps with the same number of channels by interval sampling. When the scale is 2, the four sub-feature mappings of the defect are obtained: f (0,0) =X[0:S:scale,0:S:scale] (1) f (1,0) =X[1:S:scale,0:S:scale] (2) f (0,1) =X[0:S:scale,1:S:scale] (3) f (1,1) =X[scale-1:S:scale,scale-1:S:scale] (4) Where scale∈{1,2}, the sub-feature maps are concatenated along the channel dimension to obtain the intermediate feature map X′(S / 2,S / 2,4C1), and the spatial information is converted into depth information. Then, a non-strided convolutional layer is used and a convolutional layer with a stride of 1 is used to convert X′ into the final feature map X″(S / 2,S / 2,C2) to reduce the number of channels and retain the global spatial feature information in the channel dimension, thereby enhancing the detection performance of the defect feature information of the tension clamp.

5. The method for detecting defects of tension clamps based on UAV digital X-ray according to claim 3 is characterized in that: In the SIB-YOLO model: in the backbone part, BiFormer is located in the last layer of the network, so that the network can adaptively adjust the attention weight according to the input image features; in the neck part, BiFormer is located before the psample structure and after the C2f module; The processing of BiFormer is as follows: Input image X∈R H×W×C , divided into A×A different areas, each area contains feature vectors, where H is the input image height, W is the input image width, and C is the number of input image channels; the input image two-dimensional feature map X r for: Combined with the linear transformation formula, the query tensor Q, key tensor K, and value tensor V after linear projection of the input image can be expressed as: Among them, W q , W k , W v They are respectively the query, key, and value projection weights, and we get Combined with the average values ​​of Q and K, we can get Where Q r is the region-level query tensor, K r is the region-level key tensor, the adjacency matrix A of the correlation between regions r The calculation formula is as follows: A r =Q r (K r ) T (6) topkIndex(·) is the index function of the topk regions, and the inter-region index routing matrix I r The calculation formula is as follows: I r =topkIndex(A r ) (7) Through the index matrix I r Aggregate the scattered routing areas and get the key tensor K of the aggregated area. g 、value tensor V g for: Combined with the local context enhancement term LCE(V), the calculation formula for the fine-grained token-to-token attention mechanism is as follows: O=Attention(Q,K g ,V g )+LCE(V) (9)。 6. The method for detecting defects of tension clamps based on UAV digital X-ray according to claim 3 is characterized in that: The I-IoU definition formula in SIB-YOLO is: Where T r is the scale factor; b gt and b represent the true box and the predicted box respectively, as well as Respectively represent the left boundary, right boundary, upper boundary, and lower boundary of the real box, b l , b r , b t and b b Respectively represent the left boundary, right boundary, upper boundary, and lower boundary of the prediction box; and are the inner center points of the real frame; x c and c They are the center points inside the prediction box respectively; I is the intersection of the real box and the prediction box, U is the union of the real box and the prediction box, IoU I represents the ratio of the two; w and h are the height and width of the prediction box respectively; W gt 、h gt are the height and width of the real frame respectively; T r The value range is usually [0.5, 1.5]. The scale factor is used to control the scale of the auxiliary bounding box for loss calculation. The I-IoU loss function calculation formula is as follows: L I-CIoU =L CIoU +IoU-IoU I (12) 7. The method for detecting defects of tension clamps based on UAV digital X-ray according to claim 2 is characterized in that: In step C: the tension clamp defect image data set is randomly divided into a training set, a validation set, and a test set according to a ratio of 7:2:1; the model training parameter settings include an optimizer of SGD, an initial learning rate of 0.01, a batch size of 8, and a number of training rounds of 50.

8. The method for detecting defects of tension clamps based on UAV digital X-ray according to claim 1 is characterized in that: In the step D, based on the identified defects, the parameters of the tension wire clamping defects such as steel anchor groove leakage and underpressure, aluminum wire not pressed, and steel anchor internal not crimped are rated, and the parameters include: The crimping condition between the steel anchor and the aluminum tube is determined by whether there is a gap in the steel anchor groove and the proportion of the number of leaked grooves to the total number of grooves. The crimping condition of the aluminum tube and the stranded wire. The severity is determined by the ratio of the length of the leak between the aluminum tube and the stranded wire. The crimping condition between the anchor tube and the core wire is determined by the proportion of voids in the anchor tube cavity.

9. The method for detecting defects of tension clamps based on UAV digital X-ray according to claim 8, characterized in that: Regarding the crimping of the steel anchor and the aluminum tube, if the pressure leakage ratio is less than 1 / 2, it is determined to be a general defect and can be temporarily left untreated and pressure replenishment is allowed; if the pressure leakage ratio is ≥ 1 / 2, it is determined to be a major defect and pressure replenishment must be carried out as soon as possible; For the crimping of anchor pipe and core wire, if the pressure leakage ratio is less than 15%, it is judged as a general defect and can be temporarily left untreated; if the pressure leakage ratio is 15% to 30%, it is judged as a major defect and should be re-pressed in combination with power outage; if the pressure leakage ratio is >30%, re-pressurize immediately according to the requirements of critical defects; Regarding the crimping of aluminum tubes and stranded wires, including the crimping of aluminum tubes and aluminum wires of steel core aluminum stranded wires and aluminum tubes and aluminum wires of aluminum alloy core aluminum stranded wires, if the pressure leakage ratio between the aluminum tubes and the stranded wires is less than 10%, no treatment is required for the time being. If conditions permit, pressure can be supplemented. If the pressure leakage ratio is 10% to 30%, pressure supplementation is carried out in conjunction with power outages. If the pressure leakage ratio is between 30% and 50%, the pressure must be replenished as soon as possible according to the requirements for serious defects; if the pressure leakage ratio reaches 50% or above, the pressure must be re-pressurized immediately according to the requirements for critical defects.

Images