Contact network dropper clamp defect state identification method and system

Abstract

The invention relates to the technical field of contact networks, in particular to a contact network dropper clamp defect state recognition method and system, and the method comprises the steps: obtaining multi-angle image data of a dropper clamp, and carrying out the preprocessing; extracting a first local image of the dropper clamp by using a target detection algorithm, and classifying the first local image; constructing a defect detection model, extracting a second local image based on a classification result of the first local image, and classifying the second local image; and fusing defect state detection results under the multi-angle image data. The objective of the invention is to solve the technical problem of insufficient detection capability of key parts of droppers.

Description

Technical Field

[0001] The invention relates to the technical field of contact networks, and in particular to a method and system for identifying defective states of contact network suspension string clamps. Background Art

[0002] The overhead contact network is one of the core components of electrified railways, and its main function is to provide a continuous and stable power supply to electric locomotives. The structural stability of the overhead contact network directly affects the operating safety of electric locomotives and the efficiency of railway transportation. In daily operation, the overhead contact network will be subject to the impact and vibration generated by the operation of the electric locomotive. These external forces may cause deformation or damage to the overhead contact network structure, thereby affecting the normal operation of the train. Among them, the suspension string is a key component in the high-speed rail overhead contact network system, responsible for connecting the contact wire and the load-bearing cable to ensure that the pantograph of the electric locomotive can smoothly draw current. The state of the suspension string is directly related to the contact quality between the pantograph and the contact wire, which in turn affects the operating stability and safety of the electric locomotive. It is particularly important to note that the suspension string bolts may become loose or fall off during long-term service due to environmental factors (such as temperature changes, humidity, corrosion) and mechanical factors (such as high-frequency vibration). These problems may affect the current efficiency of the pantograph at the least, or may cause damage to the pantograph or breakage of the contact wire at the worst, seriously affecting driving safety.

[0003] At present, my country's high-speed rail power supply safety detection system (6C system) has achieved image acquisition of key components of the overhead contact network suspension system through non-contact detection equipment. The system can obtain high-definition images of the overhead contact network without power outages, and conduct a preliminary assessment of the state of the overhead contact network through image analysis technology. However, although the existing technology has introduced image automatic recognition means to assist in the detection of the state of the suspension string, the existing image automatic recognition technology mainly focuses on the state recognition of the suspension string bolts, and lacks the ability to comprehensively detect other key components of the suspension string. Summary of the invention

[0004] In order to solve the technical problem of insufficient detection capability of key parts of the catenary, the purpose of the present invention is to provide a method and system for identifying defective states of catenary hanger wire clamps. The technical solutions adopted are as follows:

[0005] The technical solution of the first aspect of the present invention provides a method for identifying defective states of a catenary suspension wire clamp, the method comprising:

[0006] Acquire multi-angle image data of the suspension string clamp and perform preprocessing;

[0007] Extracting a first partial image of the suspension string clamp using a target detection algorithm and classifying the first partial image;

[0008] Building a defect detection model, extracting a second partial image based on the classification result of the first partial image and classifying the second partial image;

[0009] The defect status detection results under multi-angle image data are fused.

[0010] Furthermore, obtaining multi-angle image data of the suspension string clamp and performing preprocessing includes:

[0011] Acquire multi-angle image data of the dropper wire clamp by using a contact network detection device, wherein the multi-angle image data at least includes different sides of the dropper wire clamp;

[0012] Perform image scaling preprocessing on multi-angle image data.

[0013] Further, extracting the first partial image of the suspension string clamp by using the target detection algorithm and classifying the first partial image includes:

[0014] Use target detection algorithms to locate the string clamps in multi-angle image data;

[0015] According to the target detection result, a first partial image of the suspension string clamp is intercepted from the multi-angle image data;

[0016] It is determined whether the string clamp is dislodged according to the first partial image, and a first detection result is output.

[0017] Further, a defect detection model is constructed, and a second partial image is extracted based on the classification result of the first partial image and the second partial image is classified, including:

[0018] If the classification result of the first partial image is abnormal, the first detection result is directly output; if the classification result of the first partial image is normal, the second partial image is extracted and classified.

[0019] Further, the second partial image includes at least a partial image of the bolt, a partial image of the nut, a partial image of the nut, and a partial image of the stopper.

[0020] Furthermore, a defect detection model is constructed, and a second partial image is extracted based on the classification result of the first partial image and the second partial image is classified, and the method further includes:

[0021] The front and back state information of the suspension string clamp is extracted based on the second partial image, and a second detection result is output.

[0022] Furthermore, a defect detection model is constructed, and a second partial image is extracted based on the classification result of the first partial image and the second partial image is classified, and the method further includes:

[0023] The position state information of the stopper is extracted based on the second partial image, and the deviation state information of the stopper is extracted.

[0024] Furthermore, a defect detection model is constructed, and a second partial image is extracted based on the classification result of the first partial image and the second partial image is classified, and the method further includes:

[0025] The second partial image is classified based on the position state information of the stopper, and a fourth detection result is output.

[0026] Furthermore, the defect state detection results under multi-angle image data are fused, including:

[0027] The defect state detection results of the same suspension wire clamp at multiple angles are extracted, the defect state detection results at multiple angles are fused, and the final defect state detection results are extracted according to the confidence level.

[0028] The technical solution of the second aspect of the present invention provides a system for identifying defective states of catenary suspension wire clamps, the system comprising:

[0029] A data acquisition module configured to acquire multi-angle image data of the suspension string clamp and perform preprocessing;

[0030] A first detection module is configured to extract a first partial image of the suspension string clamp using a target detection algorithm and classify the first partial image;

[0031] A second detection module is configured to extract a second partial image based on the classification result of the first partial image and classify the second partial image;

[0032] The fusion module is configured to fuse the defect status detection results under multi-angle image data.

[0033] The present invention has the following beneficial effects:

[0034] The contact network string clamp defect status identification method and system provided by the present invention ensure that all parts of the string clamp can be fully observed by acquiring multi-angle images, reducing missed detections due to a single viewing angle; through secondary positioning and classification, more subtle defects can be detected, such as a stopper not being bent into place, a bolt being loose, etc., thereby improving the level of detail in detecting key parts; and finally, the multi-angle detection results are integrated to improve the reliability and robustness of defect identification, ensuring that the defect status report outputted in the end is more accurate and reliable. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only 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.

[0036] Figure 1 A method flow chart of a method for identifying defective states of a catenary suspension wire clamp provided by one embodiment of the present invention;

[0037] Figure 2 A schematic structural diagram of a contact network hanger wire clamp defect state identification system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0038] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of a method and system for identifying defective states of a contact network suspension string clamp proposed by the present invention, its specific implementation method, structure, features and effects, in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0039] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0040] The following is a detailed description of a method and system for identifying defective states of a contact network hanger wire clamp provided by the present invention in conjunction with the accompanying drawings.

[0041] See also Figure 1 , which shows a method flow chart of a method for identifying a defect state of a catenary suspension wire clamp provided by an embodiment of the present invention,

[0042] The technical solution of the first aspect of the present invention provides a method for identifying defective states of a catenary suspension wire clamp, the method comprising:

[0043] Step S100: acquiring multi-angle image data of the suspension string clamp and performing pre-processing;

[0044] Step S100 specifically includes:

[0045] Step S110: obtaining multi-angle image data of the suspension string clamp through the contact network detection equipment, the multi-angle image data at least including different sides of the suspension string clamp; specifically, the high-speed railway power supply safety detection system (6C system) can be used to perform multi-angle image acquisition on the same suspension string to obtain image data of the suspension string clamp at different angles; when the train passes through the detection point where the 6C system is installed, the system will automatically trigger the image acquisition function, and multiple cameras should be configured to aim at the same suspension string clamp at different angles to ensure that the image of the suspension string clamp is captured from multiple directions such as the front, side and inclined surfaces; through multi-angle image acquisition, the blind spot problem that may exist in single-angle shooting can be compensated, making defect detection more comprehensive, thereby improving the detection accuracy; multi-angle image data can be verified with each other, which helps to reduce false alarms caused by occlusion or light influence.

[0046] Step S120: performing image scaling preprocessing on the multi-angle image data; specifically, for each image in the multi-angle image data obtained in step S110, the target image size is set to 800x800 pixels, and the original image is scaled to 800x800 pixels using a bilinear interpolation algorithm; reducing a high-resolution image to 800x800 pixels can significantly reduce the time required for image processing; the bilinear interpolation method can better maintain the image quality during the scaling process and avoid jagged edges or blurring;

[0047] Step S200: extracting a first partial image of the suspension string clamp using a target detection algorithm and classifying the first partial image;

[0048] Step S200 specifically includes:

[0049] Step S210: Use a target detection algorithm to locate the string clamp in the multi-angle image data; specifically, a target positioning algorithm such as YOLOv5, YOLOv6, etc. can be used to obtain the position information of the string clamp in the multi-angle image data through model construction, model training target detection, post-processing and other processes; the following takes YOLOv5 as an example for explanation:

[0050] Use the LabelImg tool to annotate the multi-angle image data and generate an XML file containing the bounding box coordinates of the string clamp.

[0051] Select YOLOv5 as the target detection model, and adjust the backbone network and head network according to the actual situation. The backbone network is used as a feature extractor, and the head network is used to predict the position and category probability of the bounding box;

[0052] Use labeled datasets to train models and adjust hyperparameters to optimize model performance.

[0053] Use the test set to evaluate the performance of the model and make necessary adjustments based on the evaluation results;

[0054] Apply the trained model to the actual image data to obtain the position information of the string clamp;

[0055] In this embodiment, the position of the suspension string clamp can be accurately located through the target detection algorithm. Models such as YOLOv5 have a faster detection speed and are suitable for real-time detection scenarios.

[0056] Step S220: According to the target detection result, a first partial image of the suspension string clamp is intercepted from the multi-angle image data; specifically, the detection result obtained by YOLOv5 or other target detection algorithms in step S210 is read, and these results include the position information of the suspension string clamp in the image, which is the bounding box coordinates of the suspension string clamp in this embodiment. According to the detected bounding box coordinates, the area including the suspension string clamp is cropped from the original image, and the size of the first partial image is preferably 256*256 pixels; the YOLOv5 target positioning loss function can be expressed as:

[0057]

[0058] Where, L CIoU represents the target positioning loss function; IoU represents the intersection over union ratio of the real target and the predicted target box; ρ represents the Euclidean distance between the center point of the real target box and the center point of the predicted target box; c represents the diagonal length of the minimum circumscribed rectangle of the real target box and the predicted target box; α represents a dynamic trade-off factor used to balance the center point distance term and the aspect ratio consistency term; v is used to measure the consistency of the aspect ratio; w gt 、h gt Represent the width and height of the real target box respectively; w and h represent the width and height of the predicted target box respectively; arctan represents the forward and reverse tangent function; by introducing the target positioning loss function, YOLOv5 can better adjust the position and size of the bounding box during the training process and improve the accuracy of target detection, especially when dealing with targets with large aspect ratio differences. This is especially effective for the detection of slender objects such as catenary wire clamps; in the detection task of catenary wire clamps, since the wire clamps are usually slender objects, their performance in the image may be affected by multiple factors such as shooting angle and lighting conditions. This loss function can overcome these challenges to a certain extent by comprehensively considering IoU, center point distance and aspect ratio consistency, making the detection model more stable and reliable, and improving the accuracy of catenary wire clamp detection;

[0059] Step S230: judging whether the string clamp is out of place according to the first partial image, and outputting a first detection result; classifying the string clamp by intercepting the first partial image, judging whether the string clamp is out of place, and outputting a first detection result; specifically, the classification algorithm adopts the Inception-ResNet algorithm to classify the string clamp image into two categories: the string clamp is out of place and the string clamp is not out of place;

[0060] Step S300: constructing a defect detection model, extracting a second partial image based on the classification result of the first partial image, and classifying the second partial image;

[0061] Step S300 specifically includes:

[0062] Step S310: if the classification result of the first local image is abnormal, the first detection result is directly output; if the classification result of the first local image is normal, the second local image is extracted and the second local image is classified; specifically, when the classification result of the first local image determines that the string clamp is detached, the first detection result is directly output; when the classification result of the first local image determines that the string clamp is not detached, the second local image is further extracted from the first local image; the second local image at least includes a bolt local image, a nut local image, a nut local image and a stopper local image; in this embodiment, the first local image is detected by using the YOLOv6 model to obtain the regional position information of the bolt, nut, nut and stopper; the second local image is classified by using the classification model to determine whether there is a defect; through multi-level detection (first local image classification + second local image classification), the status of the string clamp and its components can be fully identified to improve the comprehensiveness and accuracy of the detection;

[0063] Step S320: extracting the front and back state information of the string clamp based on the second partial image, and outputting the second detection result; specifically, in step S310, extracting the second partial image of the bolt, nut, nut and stopper from the first partial image; then judging the front and back state of the string clamp based on the position information of the bolt and nut, stipulating that the side with the nut is the back side of the string clamp, and for each second partial image, analyzing the position relationship between the bolt and the nut, if the nut is located on the left side of the bolt, it is considered that the back side of the string clamp is the left side, so as to determine whether the installation direction of the string clamp is correct; judging whether the nut exists based on the partial image of the nut, if the nut is not detected in the partial image of the nut, it is determined that the nut is missing; according to the above analysis, outputting the second detection result, including information such as the front and back state of the string clamp, whether the nut is missing, and whether the stopper is installed;

[0064] Step S330: extracting the position state information of the stopper based on the second partial image, extracting the deviation state information of the stopper, and outputting a third detection result; specifically, using a target detection algorithm (such as YOLOv8) to obtain the regional position information of the stopper from the first partial image; based on the position information of the stopper, intercepting a partial image of the stopper from the first partial image;

[0065] Step S330 includes:

[0066] Step S331: By analyzing the proximity between the stopper and the bolt / nut area, it is determined whether the stopper is not bent into place.

[0067] Step S332: Calculate the distance or overlapping area ratio between the stop plate area and the bolt / nut area to evaluate the deviation state of the stop plate.

[0068] Step S333: if the length ratio of the stopper to the bolt / nut area is less than a certain value (such as 1.5), it is determined that the stopper is not bent into place;

[0069] Step S334: if the stopper is not bent into place, further calculating the offset between the stopper and the ideal position to evaluate the offset of the stopper;

[0070] Through the above steps, the position state and deviation state of the stopper can be effectively judged, and whether the stopper is correctly installed can be detected;

[0071] Step S340: classifying the second partial image based on the position status information of the stopper, and outputting a fourth detection result; specifically, if the position of the stopper is at a preset position, classifying the second partial image, determining the state of the bolt in the second partial image, and using a classification algorithm such as Resnet and Efficientnet to classify the state of the bolt, and outputting a fourth detection result, which includes the state information of the bolt;

[0072] Step S400: outputting a defect state detection result based on the classification result of the second partial image and fusing the defect state detection results at multiple angles;

[0073] Step S400 specifically includes:

[0074] Step S410: extracting the defect state detection results of the same suspension string clamp at multiple angles, fusing the defect state detection results at multiple angles, and extracting the final defect state detection result according to the confidence level; specifically, obtaining images of the suspension string clamp at different angles through images at different angles, extracting defect state information from the second local image classification results at each angle, which information includes defect type, location, confidence level and other information; for the same suspension string clamp, a comprehensive analysis is performed on the defect states detected at different angles; if the same defect is detected at multiple angles, the results are fused according to the confidence level, and only one defect is retained; if different types of defects are detected, the defect with the highest risk level is selected according to the risk level of the defect; according to the fused result, the final defect state detection result is output, including information such as whether there is a defect, defect type and location, etc.;

[0075] Step S410 specifically includes:

[0076] Step S411: merge the defect information sets under all angles into a total set; for example, there are second partial image classification results at N angles, and the result of each angle contains one or more defect detection information. The information of each defect can be represented by a triple (d i ,l i ,c i ) indicates that d i represents the defect type at the i-th angle, l i represents the defect position at the i-th angle, c i Represents the confidence at the i-th angle; for each angle i, extract the set of all detected defect information; through fusion processing (combining hard voting and soft voting), the accuracy and detection rate of defect detection are ensured.

[0077] Step S412: for each defect type, find a set of the defect type detected at all angles;

[0078] Step S413: When the defect types are the same, they are sorted according to the confidence level, and the defect record with the highest confidence level is selected as the final result;

[0079] Step S414: If there are multiple defect types, the defect record with the highest risk level or the most defect occurrence times is selected as the final result according to the risk level;

[0080] In this embodiment, by fusing the defect status detection results at multiple angles, the defect status of the suspension string clamp can be judged more accurately, reducing the misjudgment that may be caused by a single perspective; when the same defect is detected at multiple angles, only the record with the highest confidence level is retained, avoiding redundant information and simplifying the final result; when there are multiple defects, the defect with the highest risk level is selected to be retained according to the risk level, which helps to prioritize high-risk defects and improve the efficiency of maintenance work.

[0081] See also Figure 2 , which shows a result schematic diagram of a system for identifying defective states of a catenary suspension wire clamp provided by an embodiment of the present invention;

[0082] The technical solution of the second aspect of the present invention provides a system for identifying defective states of catenary suspension wire clamps, the system comprising:

[0083] A data acquisition module configured to acquire multi-angle image data of the suspension string clamp and perform preprocessing;

[0084] A first detection module is configured to extract a first partial image of the suspension string clamp using a target detection algorithm and classify the first partial image;

[0085] A second detection module is configured to extract a second partial image based on the classification result of the first partial image and classify the second partial image;

[0086] The fusion module is configured to fuse the defect status detection results under multi-angle image data.

[0087] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0088] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

Claims

1. A method for identifying defective states of overhead wire hanger wire clamps, characterized in that: The method comprises: Acquire multi-angle image data of the suspension string clamp and perform preprocessing; Extracting a first partial image of the suspension string clamp using a target detection algorithm and classifying the first partial image; Building a defect detection model, extracting a second partial image based on the classification result of the first partial image and classifying the second partial image; The defect status detection results under multi-angle image data are fused.

2. The method for identifying defective states of overhead wire hanger clamps according to claim 1, characterized in that: Acquiring multi-angle image data of the suspension string clamp and preprocessing includes: Acquire multi-angle image data of the dropper wire clamp by using a contact network detection device, wherein the multi-angle image data at least includes different sides of the dropper wire clamp; Perform image scaling preprocessing on multi-angle image data.

3. The method for identifying defective states of overhead wire hanger clamps according to claim 1, characterized in that: Extracting the first partial image of the suspension string clamp by using the target detection algorithm and classifying the first partial image includes: Use target detection algorithms to locate the string clamps in multi-angle image data; According to the target detection result, a first partial image of the suspension string clamp is intercepted from the multi-angle image data; It is determined whether the suspension string clamp is dislodged according to the first partial image, and a first detection result is output.

4. The method for identifying defective states of overhead wire hanger clamps according to claim 3, characterized in that: Constructing a defect detection model, extracting a second partial image based on the classification result of the first partial image and classifying the second partial image, including: If the classification result of the first partial image is abnormal, the first detection result is directly output; if the classification result of the first partial image is normal, the second partial image is extracted and classified.

5. The method for identifying defective states of overhead wire hanger clamps according to claim 4, characterized in that: The second partial image includes at least a bolt partial image, a nut partial image, a nut partial image, and a stopper partial image.

6. The method for identifying defective states of overhead wire hanger clamps according to any one of claims 1 to 5, characterized in that: Building a defect detection model, extracting a second partial image based on the classification result of the first partial image and classifying the second partial image, further comprising: The front and back state information of the suspension string clamp is extracted based on the second partial image, and a second detection result is output.

7. The method for identifying defective states of overhead wire hanger clamps according to claim 6, characterized in that: Building a defect detection model, extracting a second partial image based on the classification result of the first partial image and classifying the second partial image, further comprising: The position state information of the stopper is extracted based on the second partial image, and the deviation state information of the stopper is extracted.

8. The method for identifying defective states of overhead wire hanger clamps according to claim 7, characterized in that: Building a defect detection model, extracting a second partial image based on the classification result of the first partial image and classifying the second partial image, further comprising: The second partial image is classified based on the position state information of the stopper, and a fourth detection result is output.

9. The method for identifying defective states of overhead wire hanger clamps according to claim 1, characterized in that: The defect status detection results under multi-angle image data are integrated, including: The defect state detection results of the same suspension wire clamp at multiple angles are extracted, the defect state detection results at multiple angles are fused, and the final defect state detection results are extracted according to the confidence level.

10. The contact network suspension wire clamp defect state identification system is characterized by: The system comprises: A data acquisition module configured to acquire multi-angle image data of the suspension string clamp and perform preprocessing; A first detection module is configured to extract a first partial image of the suspension string clamp using a target detection algorithm and classify the first partial image; A second detection module is configured to extract a second partial image based on the classification result of the first partial image and classify the second partial image; The fusion module is configured to fuse the defect status detection results under multi-angle image data.

Images