Bullet train undercarriage anti-loose wire breakage detection method

Abstract

The application discloses a kind of motor train bottom anti-loose wire fracture detection methods, comprising: using the mode of comparing the characteristic information of the anti-loose wire to be detected with the standard characteristic information set in advance for detection, first collect the characteristic information of the anti-loose wire at the bottom of a perfect motor train as a standard reference template, and then collect the characteristic information of the anti-loose wire at the same position at the bottom of the motor train during the motor train maintenance, and compare it with the characteristic information in the standard reference template to determine whether the anti-loose wire has been broken.According to the application, without negative samples or a small amount of negative samples, the anti-loose wire can be detected to determine whether it is broken. The detection of the anti-loose wire is more convenient, and the application range is wider. The anti-loose wire detection of different equipment has promotional value.

Description

Technical Field

[0001] This invention relates to the technical field of machine learning and fault detection, and in particular to a method for detecting the breakage of anti-loosening wires under a high-speed train. Background Technology

[0002] The undercarriage of a high-speed train contains numerous running mechanisms and components, primarily connected by bolts. The safety performance of these bolts affects the overall safety of the train; loose or missing bolts can lead to the detachment or malfunction of critical components, thus compromising the train's overall safety. To ensure the safety and reliability of the train during operation, bolt components are reinforced with anti-loosening wire. The purpose of the anti-loosening wire is to prevent bolts from loosening. The breakage or loss of the anti-loosening wire can lead to bolt loosening or loss, which in turn threatens the safety of the train. Therefore, detecting whether the anti-loosening wire is broken or missing is of great importance to ensuring the overall safety of the high-speed train.

[0003] Most existing methods for detecting loosened wires rely on manual inspection or neural network classification. However, these methods have limited detection features, the wires are thin and numerous, making detection difficult. Furthermore, manual inspection under high-speed trains is susceptible to lighting conditions, leading to misjudgments. Prolonged manual inspection can also cause fatigue, resulting in missed or false positives.

[0004] When using traditional convolutional neural network classification, misclassification is prone to occur due to the limited feature information of the anti-loosening wires, the large area occupied by the wires, the abundance of surrounding interference features, and the scarcity of feature information regarding wire breakage. Furthermore, for safety reasons, providing negative samples of anti-loosening wires on the underside of trains is difficult, resulting in a limited number of negative samples. Artificially simulated fault samples also fail to accurately represent the real situation. Therefore, training sample collection is challenging. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the present invention aims to provide a method for detecting the breakage of anti-loosening wires under high-speed trains. This method can detect the breakage of anti-loosening wires without requiring or with only a few negative samples, making the detection of anti-loosening wires simpler and more widely applicable. It has promotional value for detecting anti-loosening wires in various devices. To achieve the above-mentioned objectives and other advantages of the present invention, a method for detecting the breakage of anti-loosening wires under high-speed trains is provided, comprising:

[0006] S1. Collect depth maps and 2D images of the anti-loosening wire under normal conditions;

[0007] S2. Generate 3D point cloud data of the anti-loosening wire based on the depth map and camera intrinsic parameters, and extract the straight slope of the edge of the anti-loosening wire based on the 3D point cloud data.

[0008] S3. Store the three-dimensional point cloud data of the anti-loosening wire and the slope of the anti-loosening wire under normal conditions as a standard reference template, and compare the subsequent test results with this standard reference template for judgment.

[0009] S4. Collect depth map and 2D image of the image to be tested. The image to be tested is the anti-loosening wire at the same position on the bottom of the train and at different times in the standard template.

[0010] S5. Generate corresponding 3D point cloud data based on the depth map, 2D image and camera intrinsic parameters of the anti-loosening wire to be tested. Perform point cloud matching between the 3D point cloud data of the anti-loosening wire to be tested and the 3D point cloud data in the standard reference template. Use the ICP point cloud matching algorithm to obtain the matched 3D point cloud data of the anti-loosening wire to be tested.

[0011] S6. Project the matched 3D point cloud data into a grayscale image and perform edge detection to extract edge information;

[0012] S7. Perform straight line detection based on edge information to detect the slope of the straight line of the anti-loosening wire to be tested;

[0013] S8. Compare the slope of the straight line of the anti-loosening wire to be tested with the slope of the straight line in the standard reference template. If the error is within the allowable range, the anti-loosening wire is normal; otherwise, the anti-loosening wire has broken.

[0014] Preferably, in step S2, the specific location of the anti-loosening wire in the image is identified and detected by the target detection algorithm. First, a large number of labeled samples are used for training, and then the trained target detection algorithm YOLOv5 is used for detection and identification to detect the location of the anti-loosening wire.

[0015] Preferably, in step S2, the camera intrinsic parameters are obtained using the Zhang Zhengyou calibration method, and then the 3D point cloud coordinates are calculated using the transformation formula from image points to world coordinate points. The calculation formula is as follows:

[0016]

[0017]

[0018]

[0019] In the formula, [x w y w z w ] T For world coordinates, i.e., the coordinates of the obtained 3D point cloud data, [uv] T Let Z be the image coordinates, A be the camera intrinsic parameter matrix, and Z be the coordinates of the points. c This represents the depth value.

[0020] Preferably, in step S2, the three-dimensional point cloud data is projected into a grayscale image, and edge detection is performed. The extracted edge information is then used to extract straight lines, and the slope of the straight line of the anti-loosening wire edge is extracted.

[0021] Preferably, when the 3D point cloud projection is a grayscale image in steps S2 and S6, the 3D point cloud data is first converted into a depth image according to the camera intrinsic parameters. Then, the depth values ​​are filtered to remove the maximum and minimum values ​​that occur during the conversion process. The filtered depth values ​​are then normalized to convert them into grayscale values ​​of 0-255, thus generating a grayscale image.

[0022] Preferably, in steps S5 and S6, ICP point cloud matching is first used, and the matched point cloud data is used for grayscale image projection and line detection to eliminate line detection errors caused by the shooting angle. Since the shooting angles of two cameras cannot be exactly the same, there will be a certain angle error. This angle error will cause the two 3D point cloud data to be in different world coordinate systems. This will cause errors when performing 3D projection grayscale image and line detection. Therefore, point cloud matching technology is used to ensure that the two point cloud data are in the same coordinate system, then projected into a grayscale image in the same direction, and then line detection is performed, eliminating errors caused by different shooting angles.

[0023] Compared with the prior art, the beneficial effects of this invention are:

[0024] (1) This invention eliminates the error caused by the difference in shooting angle when collecting photos of anti-loosening wire.

[0025] (2) The present invention provides a method for detecting whether the anti-loosening wire is broken without the need for negative samples, making the detection of the anti-loosening wire more convenient. Attached Figure Description

[0026] Figure 1 Two-dimensional images of the anti-loosening wire under normal conditions, based on the method for detecting breakage of anti-loosening wire under the train underside according to the present invention, are used to establish a standard reference template.

[0027] Figure 2 This document provides a standard reference template for establishing a diagram showing the depth of the anti-loosening wire under normal conditions, based on the method for detecting broken anti-loosening wires on the undercarriage of a high-speed train according to the present invention.

[0028] Figure 3 This is a schematic diagram of the target detection of the anti-loosening wire under normal conditions in the method for detecting the breakage of the anti-loosening wire under the train body according to the present invention, used to determine the position of the anti-loosening wire;

[0029] Figure 4 This is a three-dimensional point cloud diagram of the anti-loosening wire under normal conditions, based on the method for detecting breakage of anti-loosening wire under the train body according to the present invention.

[0030] Figure 5 This is a diagram showing the edge detection results of the anti-loosening wire under normal conditions using the method for detecting broken anti-loosening wires on the undercarriage of a high-speed train according to the present invention.

[0031] Figure 6 The diagram shows the results of straight-line detection of anti-loosening wires under normal conditions using the method for detecting breakage of anti-loosening wires on the undercarriage of a high-speed train according to the present invention.

[0032] Figure 7 A two-dimensional diagram of the anti-loosening wire under test in the test state according to the method for detecting the breakage of anti-loosening wire under the train body according to the present invention;

[0033] Figure 8 This is a depth diagram of the anti-loosening wire under the test state in the method for detecting the breakage of anti-loosening wires under the train under the present invention;

[0034] Figure 9 This is a schematic diagram of the detection of the anti-loosening wire target under the test state in the method for detecting the breakage of the anti-loosening wire under the train body according to the present invention;

[0035] Figure 10 This is a three-dimensional point cloud diagram of the anti-loosening wire under the test state in the method for detecting the breakage of the anti-loosening wire under the train body according to the present invention.

[0036] Figure 11 This is a three-dimensional point cloud data diagram of the method for detecting broken anti-loosening wires under the train body according to the present invention.

[0037] Figure 12 The image shows the edge detection result of the three-dimensional point cloud data after matching the anti-loosening wire under the train body, according to the method for detecting the breakage of anti-loosening wire under the train body of the present invention.

[0038] Figure 13 This is a diagram showing the straightness test results of the anti-loosening wire under the test state in accordance with the method for detecting the breakage of the anti-loosening wire under the train body according to the present invention. Detailed Implementation

[0039] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0040] Reference Figure 1-13A method for detecting broken anti-loosening wires under a high-speed train includes: First, acquiring depth maps and 2D images of the anti-loosening wires under normal conditions (meaning the wires are not broken). Then, using a target recognition algorithm, the location of the anti-loosening wire is identified, and 3D point cloud data of the anti-loosening wire is generated based on the depth map and camera intrinsic parameters. The 3D point cloud data is projected into a grayscale image, and edge detection is performed. Straight lines are extracted from the extracted edge information to obtain the slope of the straight line at the edge of the anti-loosening wire. The 3D point cloud data of the anti-loosening wire under normal conditions and the slope of the straight line are stored as a standard reference template. Furthermore, the Canny edge detection method is used to perform edge detection on the grayscale image, and the Hough LinesP probabilistic transform function is used for straight line detection to detect the slope of the straight line at the edge of the anti-loosening wire. During testing, depth maps and 2D images of the anti-loosening wire at the same location are acquired at different times. The target recognition algorithm is used to identify the location of the anti-loosening wire under test, and 3D point cloud data of the anti-loosening wire under test is generated based on the depth map and camera intrinsic parameters. Because the two shooting angles cannot be exactly the same, the 3D point cloud data are not in the same world coordinate system, and the change in angle will affect the subsequent edge detection and line detection results. To eliminate the error caused by the shooting angle, the two 3D point cloud data need to be projected into a grayscale image under the same coordinate system. Therefore, the 3D point cloud data of the anti-loosening wire to be tested is matched with the 3D point cloud data in the standard template to obtain the matched point cloud data of the anti-loosening wire to be tested. The matched point cloud data is projected into a grayscale image and edge detection is performed to extract edge information. Line detection is then performed on the edge information to extract the slope of the straight line of the anti-loosening wire to be tested. The slope of the straight line of the anti-loosening wire to be tested is compared with the slope of the straight line in the standard reference template. If the corresponding slope is within the allowable error range, the anti-loosening wire is judged to be normal. If the error exceeds the allowable range, the anti-loosening wire is considered to be broken. Currently, the common method used is a classifier trained by a neural network, which requires a large number of positive and negative samples. Positive samples are relatively easy to collect, while negative samples are destructive. Due to the special nature and safety considerations of high-speed trains, it is difficult to collect a sufficient number of negative samples, and the simulated negative samples in the experiment are not representative. However, judging by comparing the state of the wire at different times can be performed without negative samples, making this method more widely applicable.

[0041] First, point cloud matching is used. The matched point cloud data is then used for grayscale image projection and line detection. The key feature is that, since the angles of two camera shots cannot be exactly the same, there will be a certain angle error. This angle error will cause the two 3D point cloud data to be in different world coordinate systems. This will cause errors when performing 3D projection grayscale image and line detection. Therefore, point cloud matching technology is used to ensure that the two point cloud data are in the same coordinate system, then projected into a grayscale image in the same direction before line detection, eliminating the error caused by the different shooting angles.

[0042] Example 1

[0043] Step 1: Take a photo of the anti-loosening wire in its normal state to obtain a two-dimensional photo of the anti-loosening wire (see attached image). Figure 1 See attached depth map Figure 2 ;

[0044] Step 2: Use a pre-trained YOLOv5 model to identify the specific location of the anti-loosening wire in the 2D image (see attached image). Figure 3 Then, based on the depth map, the location of the loosened wire, and the camera's intrinsic parameters, 3D point cloud data of the loosened wire is generated, as shown in the appendix. Figure 4 ;

[0045] Step 3: Project the 3D point cloud data into a grayscale image, and perform edge detection on the grayscale image (see [link]). Figure 5 ;

[0046] Step 4: Extract straight lines from the detected edge information. The slope of the straight line extracted from the edge of the anti-loosening wire is shown in the appendix. Figure 6 ;

[0047] Step 5: Submit the 3D point cloud data (see attached file). Figure 4 The slope of the straight line along the edge of the anti-loosening wire is shown in the appendix. Figure 6 As a standard reference template;

[0048] Step Six: The anti-loosening wires to be tested should be anti-loosening wires located at the same position as the standard reference template after the same train has been running for a period of time. Take two-dimensional images of the anti-loosening wires at the corresponding positions. Figure 7 And see depth map Figure 8 ;

[0049] Step 7: Use a pre-trained YOLOv5 model to identify the specific location of the anti-loosening wire to be tested in the 2D image (see attached). Figure 9 Then, based on the depth map of the anti-loosening wire to be tested, the location of the anti-loosening wire, and the camera's intrinsic parameters, the 3D point cloud data of the anti-loosening wire to be tested is generated (see attached). Figure 10 ;

[0050] Step 8: The 3D point cloud data of the anti-loosening wire to be tested is attached. Figure 10The 3D point cloud data for the standard reference template is attached. Figure 4 The matching was performed using the ICP point cloud matching algorithm, and the resulting matched point cloud data is shown below. Figure 11 Point cloud matching is performed to eliminate the influence of changes in the slope of the straight line at the edge of the anti-loosening wire caused by changes in the camera's shooting angle.

[0051] Step 9: View the matched point cloud data Figure 11 The projection is a grayscale image, and edge detection is performed to extract edge information (see image below). Figure 12 ;

[0052] Step 10: Perform straight line detection on the extracted edge information, and extract the slope of the straight line of the edge of the anti-loosening wire to be tested. Figure 13 ;

[0053] Step 11: Compare the slope of the straight line at the edge of the anti-loosening wire with the slope of the straight line in the standard reference template. If it is within the allowable error range, the wire is considered normal. If it exceeds the error range, the wire is considered broken.

[0054] The number of devices and processing scale described herein are for the purpose of simplifying the description of the invention, and applications, modifications and variations thereof will be apparent to those skilled in the art.

[0055] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be applied to various fields suitable for the present invention. For those skilled in the art, other modifications can be easily made. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and illustrations shown and described herein.

Claims

1. A method for detecting the breakage of anti-loosening wires under a high-speed train, characterized in that, Includes the following steps: S1. Collect depth maps and 2D images of the anti-loosening wire under normal conditions; S2. Generate 3D point cloud data of the anti-loosening wire based on the depth map and camera intrinsic parameters, and extract the straight line slope of the anti-loosening wire edge based on the 3D point cloud data. S3. Store the three-dimensional point cloud data of the anti-loosening wire and the slope of the anti-loosening wire under normal conditions as a standard reference template, and compare the subsequent test results with this standard reference template for judgment. S4. Collect depth map and 2D image of the image to be tested. The image to be tested is the anti-loosening wire at the same position on the bottom of the train and at different times in the standard template. S5. Generate corresponding 3D point cloud data based on the depth map, 2D image and camera intrinsic parameters of the anti-loosening wire to be tested. Perform point cloud matching between the 3D point cloud data of the anti-loosening wire to be tested and the 3D point cloud data in the standard reference template. Use the ICP point cloud matching algorithm to obtain the matched 3D point cloud data of the anti-loosening wire to be tested. S6. Project the matched 3D point cloud data into a grayscale image and perform edge detection to extract edge information; S7. Perform straight line detection based on edge information to detect the slope of the straight line of the anti-loosening wire to be tested; S8. Compare the slope of the straight line of the anti-loosening wire to be tested with the slope of the straight line in the standard reference template. If the error is within the allowable range, the anti-loosening wire is normal; otherwise, the anti-loosening wire has broken.

2. The method for detecting the breakage of anti-loosening wires under a high-speed train as described in claim 1, characterized in that, In step S2, the specific location of the anti-loosening wire in the image is identified and detected by the target detection algorithm. First, a large number of labeled samples are used for training, and then the trained target detection algorithm YOLOv5 is used for detection and identification to detect the location of the anti-loosening wire.

3. The method for detecting the breakage of anti-loosening wires under a high-speed train as described in claim 1, characterized in that, In step S2, the three-dimensional point cloud data is projected into a grayscale image and edge detection is performed. The extracted edge information is then used to extract straight lines, and the slope of the straight line at the edge of the anti-loosening wire is extracted.

4. The method for detecting the breakage of anti-loosening wires under a high-speed train as described in claim 1, characterized in that, In steps S2 and S6, when the 3D point cloud is projected into a grayscale image, the 3D point cloud data is first converted into a depth map according to the camera intrinsic parameters. Then, the depth values ​​are filtered to remove the maximum and minimum values ​​that occur during the conversion process. The filtered depth values ​​are then normalized to convert them into grayscale values ​​of 0-255, thus generating a grayscale image.

5. The method for detecting the breakage of anti-loosening wires under a high-speed train as described in claim 1, characterized in that, In steps S5 and S6, ICP point cloud matching is first used, and grayscale projection and line detection are performed using the matched point cloud data to eliminate line detection errors caused by the shooting angle. The two 3D point cloud data are not in the same world coordinate system. Errors will be caused when performing 3D projection grayscale image and line detection. Point cloud matching technology is used to make the two point cloud data in the same coordinate system, and then project them into grayscale images in the same direction before performing line detection, thus eliminating errors caused by different shooting angles.

Images