A catenary rail contact line abrasion detection system and method

Abstract

The rail-mounted contact wire wear detection system and method of this invention addresses the technical bottleneck of measurement distortion caused by high-frequency nonlinear vibration in existing technologies by introducing a reverse compensation mechanism. This mechanism reversely reconstructs the laser point cloud at the algorithm level, effectively eliminating vehicle vibration artifacts. Simultaneously, it solves the problem of traditional ICP easily getting trapped in local extrema when processing severely worn, straightened contact wires, reducing the repeatability error of wear detection to the sub-millimeter level and ensuring confidence. Utilizing the precisely registered 3D point cloud geometric topology, it reversely guides the generation of a 2D image mask. This significantly reduces inference time by more than 80%, perfectly meeting the high frame rate real-time detection needs of embedded terminals with limited computing power (such as Raspberry Pi and Jetson Nano), ensuring data consistency, improving overall detection accuracy, and accurately reflecting the contact wire wear state and bolt loosening state. It is suitable for large-scale industrial use and promotion.

Description

Technical Field

[0001] This invention belongs to the field of intelligent operation and maintenance and machine vision inspection technology for rail transit, specifically relating to a rail-mounted contact wire wear detection system and method. Background Technology

[0002] The busbar is located inside the tunnel, and a contact wire is installed along its length at the bottom of the busbar. This system is a key piece of equipment for the safe and stable power supply of trains. As a component that directly rubs against the pantograph, the wear condition of the contact wire directly affects the reliability of power supply and operational safety. If the wear exceeds the safety limit and is not replaced in time, it may cause serious accidents such as wire breakage and arcing, leading to the interruption of subway operations.

[0003] Existing technologies include devices capable of moving and avoiding obstacles aboard busbars, such as the rail-mounted contact wire wear detection robot clamping and walking device described in Chinese patent CN121105936B. However, in the extreme environment of subway tunnels—where space is limited, multi-source noise is coupled, and there are no external navigation signals (GPS denied)—existing detection technologies reveal the following significant drawbacks when automated contact wire inspection based on such devices is required:

[0004] First, the accuracy of dynamic measurement is low. Existing technologies use line laser profilometers to acquire three-dimensional dot matrix images. However, existing methods are highly dependent on the stability of the detection platform. The busbar is rigid and has a zigzag shape. There are also abrupt changes in stiffness at the busbar joints. This leads to distortion when the detection device crosses obstacles, misjudging vehicle vibration as changes in the contact line profile and introducing a large "pseudo-wearing" error.

[0005] Secondly, existing technologies typically use the YOLOv8-tiny target detection network to output the loosening state of bolts. While the YOLOv8-tiny target detection network is an open-source, publicly known target detection network, its operation is hampered by uneven lighting, dense dust and oil contamination within subway tunnels, and strong reflective spots on the aluminum alloy busbar surface.

[0006] Some background noise can easily be mistaken by the algorithm as bolts or cause missed detection. At the same time, the background area of ​​the contact wire detection accounts for more than 80% of the entire image. Performing convolution operations on the entire image results in extremely high computational redundancy, making it difficult to achieve high frame rate real-time processing on embedded edge devices with limited computing power (such as Raspberry Pi, Jetson, etc.).

[0007] Third, existing technologies suffer from spatiotemporal fragmentation during detection. "Contact line wear (3D geometry)" and "fastener condition (2D texture)" are treated as two completely independent and isolated tasks, with various sensors (lasers, cameras, inertial navigation modules) asynchronously acquiring data at different frequencies. This "separate-track" architecture results in the inability to accurately align multimodal data in terms of "timestamps" and "mileage coordinates".

[0008] In summary, the detection accuracy of existing technologies is insufficient, and they cannot accurately reflect the wear condition of contact wires and the loosening condition of bolts. Summary of the Invention

[0009] The purpose of this invention is to provide a rail-mounted contact wire wear detection system and method to solve the problem that the detection accuracy of existing detection methods is insufficient and cannot truly reflect the wear state of the contact wire and the loosening state of the bolts.

[0010] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0011] A rail-mounted contact wire wear detection system includes a rail-mounted contact wire wear detection device suspended from the bottom of a busbar. A contact wire extending along its length is disposed in the middle of the bottom end of the busbar, and multiple pre-tightening bolts are disposed along the length of the opposite side walls of the busbar. The rail-mounted contact wire wear detection device includes a moving mechanism and an obstacle avoidance mechanism. The moving mechanism can drive the rail-mounted contact wire wear detection device to move along the busbar, and the obstacle avoidance mechanism can drive the rail-mounted contact wire wear detection device to avoid obstacles on the busbar. The system also includes a controller, an inertial measurement unit, an absolute encoder, an area scan camera, and a line laser profilometer. The controller is electrically connected to the moving mechanism, the absolute encoder, the area scan camera, and the line laser profilometer, respectively.

[0012] The inertial measurement unit is located at the center of gravity of the rail-mounted contact wire wear detection device, and is used to detect the acceleration and angular velocity of the rail-mounted contact wire wear detection device and transmit them to the controller;

[0013] The absolute encoder is installed on the moving mechanism to detect the distance traveled by the moving mechanism and transmit it to the controller;

[0014] The area array camera is mounted on the side wall of the rail-mounted contact wire wear detection device to capture two-dimensional images of the two side walls of the busbar and transmit them to the controller; wherein, the two-dimensional image is a two-dimensional pixel image including the pre-tightening bolts on the two side walls of the busbar.

[0015] The line laser profilometer is positioned below the area array camera. The output beam of the line laser profilometer spans the contact line cross section and forms an angle with the optical axis of the area array camera. The line laser profilometer is used to: acquire a three-dimensional dot matrix image of the bottom of the busbar and transmit it to the controller.

[0016] The three-dimensional dot matrix image includes a contact line shape curve image and dot matrix images on both sides of the bottom of the busbar;

[0017] The contact line shape curve image specifically includes a curve composed of multiple points reflecting the shape of the bottom cross-section edge of the contact line; the busbar bottom side dot matrix image specifically includes a dot matrix image composed of multiple points reflecting the shape of the bottom ends of the two side walls of the busbar.

[0018] The controller is installed inside the rail-mounted contact wire wear detection device, and the controller is used for:

[0019] Control the moving mechanism to move along the busbar;

[0020] Control the obstacle avoidance mechanism to avoid obstacles;

[0021] Set the sampling step size for the line laser profilometer, area array camera, and inertial measurement unit;

[0022] The linear laser profilometer and the area array camera perform synchronous sampling at a fixed step size;

[0023] Image correction is performed on the contact wire shape curve image to obtain a standard contact wire shape curve image;

[0024] Image processing is performed on the standard contact wire shape curve image, the dot matrix image and the two-dimensional image on both sides of the bottom of the busbar to obtain the wear condition of each position of the contact wire and the loosening condition of the pre-tightening bolts.

[0025] A method for detecting contact wire wear based on the above-mentioned rail-mounted contact wire wear detection system includes the following steps:

[0026] Step 1: Set the sampling step size for the line laser profilometer and the area scan camera;

[0027] Step 2: Move the moving mechanism along the busbar at a constant speed. When the moving distance reaches the sampling step length set in Step 1, have the line laser profilometer and the area array camera sample simultaneously to acquire three-dimensional dot matrix images and two-dimensional images respectively.

[0028] Record the position at the time of this sampling as the current position;

[0029] Step 3: Perform image correction on the contact line shape curve image, and use the corrected three-dimensional dot matrix image as the standard contact line shape curve image, then proceed to Step 4.

[0030] Step 4: Set the surface curvature determination threshold for the contact line;

[0031] Based on the surface curvature of the curve in the standard contact wire shape curve image, determine whether the contact wire has experienced wear at the current position based on the relationship between the surface curvature and the surface curvature judgment threshold.

[0032] Step 5: Project the center lines of the bottom sides of the busbar in the dot matrix images on both sides of the busbar onto the corresponding two-dimensional images to serve as two-dimensional baselines;

[0033] Using the two-dimensional baseline as a reference, the area where the pre-tightening bolts are located is delineated in the two-dimensional image as the target area;

[0034] The target area is subjected to dimensionality reduction and noise reduction processing to obtain a target image containing only the pre-tightened bolts, and the loosening of the pre-tightened bolts in the target image is identified.

[0035] Step 6: Repeat steps 2-5 until the rail-mounted contact wire wear detection device reaches the end of the busbar, and obtain the wear condition of the contact wire and the loosening condition of the pre-tightening bolts at each sampling position of the busbar.

[0036] The present invention also has the following features:

[0037] Furthermore, in step 1, the sampling step size is 5-15mm.

[0038] Furthermore, step 3 specifically includes the following sub-steps:

[0039] Step 31: Using the error state extended Kalman filter algorithm, calculate the roll angle and pitch angle of the rail-mounted contact wire wear detection device based on the current position of the rail-mounted contact wire wear detection device at the time of sampling, the speed at the time of sampling, the acceleration and angular velocity at the time of sampling.

[0040] Step 32: Based on the roll angle and pitch angle calculated in step 31, construct the inverse kinematics compensation matrix for each point in the contact line shape curve image;

[0041] Step 33: For any point in the 3D point image, perform an inverse transformation using the inverse kinematics compensation matrix corresponding to that point to obtain the corrected point;

[0042] Using the above method, traverse each point in the 3D dot matrix image to obtain the corrected contact line shape curve image, and use the corrected contact line shape curve image as the standard contact line shape curve image.

[0043] Furthermore, step 4 specifically includes the following sub-steps:

[0044] Step 41, set the surface curvature determination threshold, as follows:

[0045]

[0046] in, This indicates the threshold for determining surface curvature;

[0047] The mathematical mean of the surface curvature of the standard CAD section of the contact wire;

[0048] This represents the tolerance sensitivity adjustment coefficient. ;

[0049] The standard deviation of the surface curvature of the standard CAD cross-section of the contact wire;

[0050] Step 42: For a standard contact line shape curve image, select one endpoint of the curve and construct the neighborhood point set of the selected point. Covariance matrix;

[0051] Regarding the aforementioned The covariance matrix is ​​subjected to eigenvalue decomposition to obtain three eigenvalues ​​arranged in ascending order. ;

[0052] Step 43: Calculate the surface curvature corresponding to the selected point by taking the ratio of the smallest eigenvalue to the sum of the three eigenvalues, as shown in the following formula:

[0053]

[0054] in, This indicates the surface curvature corresponding to the selected point;

[0055] Step 44, if the surface curvature of the selected point is... ≤ Surface curvature determination threshold If the selected point has been worn, then the point is designated as the wear point and proceed to step 46.

[0056] If the surface curvature of the selected point is > Surface curvature determination threshold If the selected point does not show wear, proceed to step 45.

[0057] Step 45: Select a point in the standard contact line shape curve image that is adjacent to the point selected in step 45, and repeat steps 43-44 until all points in the standard contact line shape curve image are traversed or wear points are obtained.

[0058] If all points in the standard contact wire shape curve image have been traversed and there are no wear points, then the contact wire at the current position is not worn.

[0059] If the wear points are obtained, then all surface curvatures... > Surface curvature determination threshold The point is designated as the unworn point, proceed to step 46;

[0060] Step 46: For the standard contact line shape curve image, select the point where the other endpoint of the curve is located, and repeat steps 42-45 until another wear point is obtained;

[0061] Step 47: Select the unworn points adjacent to the two worn points as the two lug points;

[0062] Step 48: Based on the standard CAD cross-section of the contact wire, use the ICP precise matching algorithm to correct all unworn points and lug points;

[0063] Connect the two corrected lug points, obtain the true coordinates of the two corrected lug points, connect the two lug points, and take the straight line formed by connecting the two corrected lug points as the wear line at the current position of the contact line. Take the area formed by the wear line and the bottom of the standard CAD section as the wear area at the current position of the contact line.

[0064] Furthermore, step 5 specifically includes the following sub-steps:

[0065] Step 51: Extract the center line of the bottom side of the busbar from the dot matrix image of that side as the spatial three-dimensional geometric reference line.

[0066] Step 52: Project the three-dimensional geometric baseline onto the two-dimensional image of the corresponding sidewall of the busbar to form a two-dimensional baseline;

[0067] Step 53: Using the two-dimensional baseline as a reference, translate in the direction of the pre-tightening bolt in the two-dimensional image until the bottom of the pre-tightening bolt is reached, and delineate the area where the pre-tightening bolt is located as the target area;

[0068] Step 54: Based on the standard design drawings of the bus, use a dynamic mask generation algorithm to identify the pre-tightening bolts in the target area;

[0069] Step 55: Use the YOLOv8-tiny target detection network to output the loosening status of the pre-tightened bolts;

[0070] Step 56: Extract the center line of the other side from the dot matrix image at the bottom of the busbar as the three-dimensional geometric reference line in space. Repeat steps 52-55 to obtain the loosening state of the pre-tightening bolts on both sides of the busbar.

[0071] Furthermore, in step 52, when projecting the spatial three-dimensional geometric baseline onto the two-dimensional image of the corresponding sidewall of the busbar, each point of the spatial three-dimensional geometric baseline is specifically projected onto the two-dimensional image of the corresponding sidewall of the busbar, forming a reference point that corresponds one-to-one with each point of the spatial three-dimensional geometric baseline, and then forming a two-dimensional baseline from multiple reference points.

[0072] The projection transformation formula for any point on the three-dimensional geometric baseline is as follows:

[0073]

[0074] in, X, Y, Z These represent the three-dimensional coordinate components of any point on the three-dimensional geometric reference line in space;

[0075] u and v These represent the coordinates of a point on a three-dimensional geometric baseline projected onto a point on a two-dimensional image;

[0076] This represents the depth value of a point projected onto a 2D image in the coordinate system of an area array camera.

[0077] K The intrinsic parameter matrix represents the area array camera;

[0078] This represents the joint extrinsic parameter matrix from the line laser profilometer coordinate system to the area array camera coordinate system.

[0079] Compared with the prior art, the present invention has the following technical effects:

[0080] The present invention relates to a rail-mounted contact wire wear detection system and method. Addressing the technical bottleneck of measurement distortion caused by high-frequency nonlinear vibration in existing technologies, this invention introduces a reverse compensation mechanism. At the algorithm level, it performs reverse reconstruction of the laser point cloud, effectively eliminating vehicle vibration artifacts. Simultaneously, it solves the problem of traditional ICP easily getting trapped in local extrema when processing severely worn, straightened contact wires, reducing the repeatability error of wear detection to the sub-millimeter level and ensuring confidence.

[0081] The rail-mounted contact wire wear detection method of this invention utilizes the precisely registered geometric topology of a 3D point cloud to guide the generation of a mask from a 2D image. This dimensionality reduction operation directly isolates the interference of tunnel background light spots at the physical level, reducing the search area of ​​the YOLOv8-tiny target detection network to less than 10% of the entire image. This not only reduces the false detection rate of bolt loosening by more than 90%, but also significantly reduces the inference time by more than 80%, perfectly meeting the high frame rate real-time detection requirements of embedded terminals with limited computing power (such as Raspberry Pi and Jetson Nano).

[0082] The rail-mounted contact wire wear detection method of the present invention utilizes the hardware timer channel of the underlying microcontroller to receive the fixed step-size pulse signal sent by the absolute mileage encoder when the moving mechanism is moving. This pulse is used as a hard trigger source to trigger the laser profilometer to emit light, the area array industrial camera to expose, and the attitude data of the latched IMU in parallel. Synchronous heterogeneous data sharing the same absolute mileage stamp and timestamp are obtained, which ensures data uniformity, improves the overall detection accuracy, and can truly reflect the wear state of the contact wire and the loosening state of the bolts. It is suitable for large-scale industrial use and promotion. Attached Figure Description

[0083] Figure 1 This is a schematic diagram of the suspension state of the rail-mounted contact wire wear detection system on the busbar in the existing technology.

[0084] Figure 2 This is an embodiment of the contact wire wear detection method of the present invention, including an absolute encoder, a line laser profilometer, and a waveform diagram of an area array camera;

[0085] Figure 3 This is a comparison image of the standard contact wire shape curve obtained by step 4 of the method of the present invention and the contact wire shape curve image obtained directly by the existing method.

[0086] Figure 4 This is a comparison diagram of the wear line obtained by step 5 of the method of the present invention and the wear line obtained by direct detection by existing methods.

[0087] The meanings of the labels in the diagram are as follows: 1. Busbar; 2. Contact wire; 3. Preload bolt. Detailed Implementation

[0088] It should be noted that, unless otherwise specified, all components in this invention are known in the prior art. For example, the controller uses a commonly known controller.

[0089] The following are specific embodiments of the present invention. It should be noted that the present invention is not limited to the following specific embodiments. All equivalent modifications made based on the technical solutions of this application fall within the protection scope of the present invention.

[0090] A rail-mounted contact wire wear detection system includes a rail-mounted contact wire wear detection device suspended at the bottom of a busbar 1. A contact wire 2 extending along its length is provided in the middle of the bottom end of the busbar 1, and multiple pre-tightening bolts 3 are provided along the length on opposite side walls of the busbar 1. The rail-mounted contact wire wear detection device includes a moving mechanism and an obstacle avoidance mechanism. The moving mechanism can drive the rail-mounted contact wire wear detection device to move along the busbar 1, and the obstacle avoidance mechanism can drive the rail-mounted contact wire wear detection device to avoid obstacles on the busbar 1. The system is characterized by further including a controller, an inertial measurement unit, an absolute encoder, an area array camera, and a line laser profilometer. The controller is electrically connected to the moving mechanism, the absolute encoder, the area array camera, and the line laser profilometer, respectively.

[0091] The absolute encoder is used to detect the distance traveled by the moving mechanism and transmit it to the controller.

[0092] The line laser profilometer is used to: acquire a three-dimensional dot matrix image of the bottom of busbar 1 and transmit it to the controller;

[0093] The three-dimensional dot matrix image includes the contact line shape curve image and the dot matrix images of the bottom two sides of busbar 1;

[0094] The contact line shape curve image specifically includes a curve composed of multiple points that reflects the shape of the bottom cross-section edge of the contact line 2;

[0095] The dot matrix images on both sides of the bottom of busbar 1 specifically include dot matrix images composed of multiple dots that reflect the shape of the bottom ends of the two side walls of the bottom of busbar 1;

[0096] The area array camera is used to: capture two-dimensional images of the two side walls of the busbar 1 and transmit them to the controller; wherein, the two-dimensional image is a two-dimensional pixel image containing the pre-tightening bolts 3 on the two side walls of the busbar 1;

[0097] The inertial measurement unit is used to detect the acceleration and angular velocity of the moving mechanism during sampling and transmit them to the controller;

[0098] The controller is used for:

[0099] Control the moving mechanism to move along busbar 1 and avoid obstacles;

[0100] Set the sampling step size of the line laser profilometer, area scan camera, and inertial measurement unit, and make the line laser profilometer, area scan camera, and inertial measurement unit sample at fixed intervals;

[0101] Image correction is performed on the contact wire shape curve image to obtain a standard contact wire shape curve image;

[0102] Image processing is performed on the standard contact wire shape curve image, the dot matrix image and the two-dimensional image of the bottom two sides of the busbar 1 to obtain the wear condition of each position of the contact wire 2 and the loosening condition of the pre-tightening bolt 3.

[0103] The following combination Figure 1 The rail-mounted contact wire wear detection system of this embodiment will be further described below. The devices used in this embodiment are all existing devices known in the art. Figure 1 The specific positions of each component on the rail-mounted contact wire wear detection device are not shown in the text. Those skilled in the art can directly adopt the installation positions described in this embodiment, or make adaptive adjustments based on conventional choices in the field and actual conditions.

[0104] Specifically:

[0105] The rail-mounted contact wire wear detection device is an existing device, such as the rail-mounted contact wire wear detection robot clamping and walking device given in Chinese patent CN121105936B. This device includes a moving mechanism and is equipped with an obstacle avoidance mechanism, which can move along the busbar 1 while avoiding obstacles.

[0106] In this embodiment, the rail-mounted contact wire wear detection device uses a 24V lithium battery as the main power source.

[0107] In this embodiment, the controller uses a Raspberry Pi and an STM32 microcontroller, and the Raspberry Pi and the STM32 microcontroller communicate via a serial port.

[0108] Specifically, the Raspberry Pi, acting as the host computer, is mainly responsible for running complex image processing algorithms, including image correction of the contact wire shape curve image to obtain a standard contact wire shape curve image.

[0109] Image processing is performed on the standard contact wire shape curve image, the dot matrix image and the two-dimensional image of the bottom two sides of the busbar 1 to obtain the wear condition of each position of the contact wire 2 and the loosening condition of the pre-tightening bolt 3.

[0110] The STM32 microcontroller acts as the lower-level machine, responsible for tasks with high real-time requirements, controlling the moving mechanism to move along busbar 1 and the obstacle avoidance mechanism to avoid obstacles.

[0111] Set the sampling step size of the line laser profilometer, area scan camera, and inertial measurement unit, so that the line laser profilometer, area scan camera, and inertial measurement unit sample at fixed intervals; and exchange instructions and data with the Raspberry Pi through the serial port.

[0112] For STM32 microcontrollers, the control methods used are all known and commonly used methods in this field, and do not involve any improvement to the algorithm.

[0113] An inertial measurement unit (IMU) is a known device in the prior art, used to measure the three-dimensional acceleration and angular velocity of an object and calculate its attitude angle. In this embodiment, the inertial measurement unit is typically installed at the center of gravity of the object being measured to improve measurement accuracy. In this embodiment, the inertial measurement unit is conventionally installed at the center of gravity of the rail-mounted contact wire wear detection device.

[0114] An absolute encoder is a known device in the prior art. It is a position sensor mounted on a moving mechanism. Its core function is to measure the distance traveled by the platform in real time and output a unique, non-lost position value.

[0115] A line laser profilometer is a known device in the prior art, and it is an optical measurement sensor based on laser triangulation. Its core function is to acquire three-dimensional contour data of a line on the surface of an object non-contactly.

[0116] In this embodiment, the line laser profilometer is installed below one side of the area array camera. The output beam of the line laser profilometer crosses the contact line 2 section and forms a 20° angle with the optical axis of the area array camera.

[0117] Area scan cameras are known devices in the prior art, specifically imaging tools that acquire images on a "plane" basis, and are widely used in industrial fields. Devices using area scan CMOS or CCD sensors for image acquisition can acquire a complete target image in a short time. Unlike line scan cameras, area scan cameras can directly generate two-dimensional images. In this embodiment, following conventional selection, the area scan camera is mounted on the side wall of the rail-mounted contact wire wear detection device. It is sufficient to capture images of the pre-tightening bolts 3 on both sides of the busbar 1; no further restrictions are placed on the more specific installation location.

[0118] A method for detecting wear of contact wire 2 based on the above-mentioned rail-mounted contact wire wear detection system includes the following steps:

[0119] Step 1, follow the fixed sampling step size The line laser profilometer and area array camera are set to operate according to this sampling step size. Perform sampling;

[0120] In this embodiment, the sampling step size is 5-15mm.

[0121] Step 2: Move the moving mechanism along busbar 1 at a constant speed. When the moving distance reaches the sampling step length set in Step 1, make the line laser profilometer and the area array camera sample simultaneously, and record the position at this sampling time as the current position.

[0122] Acquire 3D raster images, 2D images, and the acceleration and angular velocity of the moving mechanism during this sampling;

[0123] In this embodiment, whenever the moving mechanism travels... When the distance is reached, the hardware timer channel of the STM32 microcontroller sends out nanosecond-level hard trigger pulse signals in parallel. These signals synchronously drive the line laser profilometer to emit lasers, drive the shutter of the area array camera to expose, and latch the current acceleration and angular velocity of the inertial measurement unit (IMU).

[0124] This resulted in the attachment of a unified spatial mileage stamp. With timestamp Synchronized data The representation is as follows:

[0125]

[0126] in, It is a three-dimensional raster image data;

[0127] It is two-dimensional image data;

[0128] Acceleration and angular velocity measured by an inertial measurement unit (IMU).

[0129] Therefore, as Figure 2 As shown, absolute alignment of geometry and texture in the physical world coordinate system is achieved.

[0130] Step 3: Perform image correction on the standard contact line shape curve image to obtain the corrected standard contact line shape curve image, and use the corrected three-dimensional dot matrix image as the standard contact line shape curve image.

[0131] Because the moving mechanism generates vibrations during movement, especially when crossing obstacles such as busbar 1 connectors or pre-tightening bolts 3, high-frequency nonlinear vibrations can affect the 3D dot matrix image data acquired by the line laser profilometer. This produces severe serrated distortion, resulting in "pseudo-wearing" errors.

[0132] Generally speaking, before calibrating a 3D raster image, it is necessary to determine whether obstacle crossing has occurred. If no obstacle crossing has occurred, the acquired 3D raster image is considered to be standard and can be used directly; otherwise, calibration is required.

[0133] However, even if no obstacle is crossed during the movement of the mobile mechanism, errors may occur due to its own vibration. Therefore, in this embodiment, correction is performed regardless of whether an obstacle is crossed, thus maximizing the accuracy of the three-dimensional dot matrix image.

[0134] Step 4: Calculate the surface curvature of contact line 2 based on the standard contact line shape curve image;

[0135] Determine whether the contact wire 2 at the current position has experienced wear based on the surface curvature of the contact wire 2;

[0136] If wear occurs, further determine the wear condition of contact line 2 at the current position;

[0137] Step 5: Project the center line of the bottom of busbar 1 from the dot matrix images on both sides of the bottom of busbar 1 into the two-dimensional image as a two-dimensional baseline;

[0138] Using the two-dimensional baseline as a reference, the area where the pre-tightening bolt 3 is located is delineated in the two-dimensional image as the target area;

[0139] The target area is subjected to dimensionality reduction and noise reduction processing to obtain a target image containing only the pre-tightened bolt 3, and the loosening of the pre-tightened bolt 3 in the target image is identified.

[0140] In the prior art, 3D raster image processing and 2D image recognition are separate; however, this embodiment uses multi-sensor joint calibration to actively guide and limit the search area (ROI) of the 2D image by utilizing the high-precision geometric topology information of the 3D raster image, thereby physically isolating complex background noise, which will be further explained in more detail later.

[0141] Step 6: Repeat steps 2-5 until the wear condition of the contact wire 2 at each position of the busbar 1 and the loosening condition of the pre-tightening bolt 3 are obtained.

[0142] As a more specific implementation, step 3 specifically includes the following sub-steps:

[0143] Step 31: Using the error state extended Kalman filter algorithm, calculate the roll angle and pitch angle of the moving mechanism at the time of sampling based on the current position, velocity, acceleration and angular velocity of the moving mechanism at the time of sampling.

[0144] It should be noted that the Error-State Extended Kalman Filter (ES-EKF) algorithm is an important tool in the field of state estimation and is widely used in scenarios such as navigation, positioning, and sensor fusion.

[0145] Error State Extended Kalman Filter (EKF) is a widely used state estimation algorithm in inertial navigation, robot localization, and other fields. It is a variant of the classic EKF, and its core idea is to decompose the system's true state into a nominal state and an error state. The error state is then linearized and filtered for estimation, and finally, the estimated error state is used to correct the nominal state. By linearizing the error, it avoids the error amplification problem associated with directly linearizing nonlinear states. It is widely used in IMU and GPS fusion, especially for high-frequency IMU data, where error filtering corrects deviations in navigation information.

[0146] In this embodiment, step 31 is implemented in the following specific steps:

[0147] Step 3.1.1, State Definition and Mechanical Arrangement: Define the nominal state variables of the system as... , representing position, velocity, quaternion attitude, acceleration bias, and angular velocity bias, respectively.

[0148] Using high-frequency data (such as 500Hz) from the IMU (Inertial Measurement Unit), the nominal state is subjected to strapdown recursive integration based on the rigid body kinematic differential equations.

[0149] Step 3.1.2, Error State Filtering and Correction: Due to the drift of the integral, the deterministic velocity information of the moving mechanism is introduced as the observation value.

[0150] Next, a 15-dimensional error state vector is constructed. Calculate the Kalman gain and update the error state:

[0151] The error state is then injected into the nominal state, and the output data is a highly accurate smooth attitude angle (roll and pitch) at the moment of data acquisition.

[0152] Step 32: Based on the roll angle and pitch angle calculated in step 31, construct the inverse kinematics compensation matrix for each point in the contact line shape curve image;

[0153] Constructing an inverse kinematics compensation matrix based on roll angle and pitch angle is also a well-known method in the field. In this embodiment, an inverse kinematics compensation matrix is ​​constructed based on the instantaneous roll angle and pitch angle calculated in step 41, as well as the translation of the center of gravity of the rail-mounted contact wire wear detection device.

[0154] Step 33: For any point in the 3D point image, perform an inverse transformation using the inverse kinematics compensation matrix corresponding to that point;

[0155] Specifically, for those containing For each point in the contact line shape curve image of the data points, apply an inverse transformation to it;

[0156] After traversing all points, a high-precision contact wire geometric point cloud with vibration artifacts eliminated is reconstructed. At this point, the zigzag interference and high-frequency jitter in the point cloud have been completely filtered out, and the outline of contact line 2 is restored to a smooth actual geometric shape.

[0157] After traversing each point in the 3D dot matrix image using the above method, the corrected contact line shape curve image is obtained, and the corrected contact line shape curve image is used as the standard contact line shape curve image.

[0158] like Figure 3 As shown, the contact line 2 curve without wear is a gray curve.

[0159] When the rail-mounted contact wire wear detection device encounters an obstacle, the dot matrix cloud in the contact wire shape curve image obtained by directly detecting it using traditional methods is as follows: Figure 3 As shown by the green line in the upper part, there is a significant offset due to error, with a root mean square error (RMSE) of 1.32 / mm.

[0160] After adopting the correction method of this embodiment, the dot matrix cloud, as shown in the embodiment, Figure 3 As shown by the blue lines in the lower half, the dot matrix cloud is no longer discrete, with a root mean square error (RMSE) of 0.1577 mm, achieving sub-millimeter level accuracy.

[0161] As a more specific implementation, high-precision measurement of contact wire wear (3D geometric calculation) is performed. Due to pantograph friction, the bottom of contact wire 2 gradually wears down from a standard arc. Traditional ICP registration algorithms, because they lose the arc feature, are prone to getting trapped in local extrema. This invention proposes a curvature adaptive weighted ICP fine registration algorithm, with specific step 5 including the following sub-steps:

[0162] Step 41, set the surface curvature determination threshold. ;

[0163] It should be noted that the surface curvature determination threshold in this embodiment... The threshold of surface curvature varies depending on the specific specifications of the contact line 2. This threshold can be adaptively adjusted by those skilled in the art based on the actual situation. This embodiment does not impose detailed limitations.

[0164] As a more specific implementation method, this embodiment further provides a method for setting the surface curvature determination threshold:

[0165] The sliding friction between the pantograph and the contact wire causes the arc feature at the bottom of the contact wire to degenerate into a flat surface. In differential geometry topology, the flatter the surface, the closer its local curvature is to 0.

[0166] Accordingly, this embodiment introduces the confidence interval rule of normal distribution and sets a threshold for determining surface curvature, as follows:

[0167]

[0168] in, This indicates the threshold for determining surface curvature;

[0169] The mathematical mean of the surface curvature of the standard CAD section of the contact wire;

[0170] This represents the tolerance sensitivity adjustment coefficient. ;

[0171] The standard deviation of the surface curvature of the standard CAD cross-section of the contact wire;

[0172] It should be noted that the surface curvature determination threshold in step 41 and the surface curvature calculated in subsequent step 43 use the same curvature description method. This surface curvature is a normalized curvature description quantity constructed based on the eigenvalues ​​of the point neighborhood covariance matrix, used to characterize the flatness of the local shape, and is not the continuous curvature with the inverse dimension of length in the traditional analytical geometry sense.

[0173] Therefore, the surface curvature determination threshold, the mathematical expectation mean of the surface curvature of the standard CAD section, and the standard deviation in step 41 have the same dimension, all being dimensionless quantities; the tolerance sensitivity adjustment coefficient is a dimensionless adjustment coefficient.

[0174] Step 42: For the standard contact wire shape curve image, since the standard contact wire shape curve image is a curve composed of multiple points, select one endpoint of the curve, construct and calculate the covariance matrix of the selected point, and then obtain the three eigenvalues ​​of the selected point. ;

[0175] Constructing and calculating the covariance matrix and eigenvalues ​​are conventional methods in this field. In this embodiment, the covariance matrix is ​​constructed from the neighborhood point set of the selected point. Covariance matrix.

[0176] Step 43: Calculate the surface curvature corresponding to the selected point by taking the ratio of the smallest eigenvalue to the sum of the three eigenvalues, as shown in the following formula:

[0177]

[0178] in, This indicates the surface curvature corresponding to the selected point;

[0179] This represents the minimum eigenvalue of the selected point.

[0180] These represent the other two feature values ​​of the selected point.

[0181] Step 44, if the surface curvature of the selected point is... ≤ Surface curvature determination threshold Proceed to step 45;

[0182] If the surface curvature of the selected point is > Surface curvature determination threshold This point is designated as the wear point, and the process proceeds to step 46.

[0183] for The area (i.e., the non-wearable lug area that retains the original arc features) is assigned a high registration weight;

[0184] for The regions (i.e., the wear-out flat lines where features are degraded) are assigned low registration weights or are directly removed.

[0185] In the actual testing and weighted ICP registration guidance process: when the local curvature of the measured point... When the geometric curvature of the region is determined to be below the permissible lower limit of the standard lug, it is identified as a wear-resistant flat point of feature degradation; if If it is, then it is confirmed as a non-wear point that retains the original circular arc geometry.

[0186] Step 45: Select a point in the standard contact line shape curve image that is adjacent to the point selected in step 54, and repeat steps 43-44 until all points in the standard contact line shape curve image are traversed or wear points are obtained.

[0187] If no wear points are found after traversing all points in the standard contact line shape curve image, then there is no wear at position 2 of the contact line during this sampling.

[0188] If the wear point is obtained, proceed to step 46;

[0189] Step 46: For the standard contact line shape curve image, select the point where the other endpoint of the curve is located, and repeat steps 42-45 until another wear point is obtained;

[0190] Step 47: Select the unworn points adjacent to the two worn points as the two lug points;

[0191] Step 48: Based on the standard CAD cross-section of the contact wire, use the ICP precise matching algorithm to correct all unworn points and lug points;

[0192] Connect the two corrected lug points, obtain the true coordinates of the two corrected lug points, connect the two lug points, and take the straight line formed by connecting the two corrected lug points as the wear line at the current position of the contact line. Take the area formed by the wear line and the bottom of the standard CAD section as the wear area at the current position of the contact line.

[0193] It should be noted that the standard CAD cross-section of contact wire 2 is provided by the manufacturer of contact wire 2 and is a property specification that comes with contact wire 2 when it leaves the factory.

[0194] By calculating the area of ​​the difference integral between the registered point cloud and the outer envelope of the standard section, the wear depth of contact line 2 can be accurately calculated. remaining area due to wear and the grinding angle .

[0195] like Figure 4 As shown, without wear line correction, such as Figure 4 As shown by the red line in the upper part, although the wear line can be obtained through the two wear points, the wear line is severely offset, with a root mean square error (RMSE) of 0.973 / mm; it is difficult to reflect the true wear condition of contact line 2.

[0196] After using the method of this embodiment and through multiple iterations, a corrected lug wire conforming to the standard CAD cross-section of the contact wire and a corrected wear line were obtained. Among them, Figure 4 The blue line shown in the lower half is the correction loop line obtained after iterative convergence; Figure 4 The gray area in the lower half is a superposition of the wear lines calculated in each iteration, used to characterize the process of the wear line gradually converging from the initial estimate to the final corrected result. The final corrected wear line is located within this gray area and is affected by error, with a root mean square error (RMSE) of 0.166 / mm.

[0197] Furthermore, while measuring wear, the spatial three-dimensional geometric baseline of the bottom surface of the lugs on both sides of the busbar 1 was extracted. This serves as a benchmark for cross-modal guidance in step 5.

[0198] In a more specific implementation, step 5 includes the following sub-steps:

[0199] Step 51: Extract the center line of the bottom side of the busbar 1 from the dot matrix image of that side as the spatial three-dimensional geometric reference line.

[0200] Step 52: Project the three-dimensional geometric baseline onto the two-dimensional image of the corresponding sidewall of busbar 1 to form a two-dimensional baseline;

[0201] In this step, when the spatial three-dimensional geometric baseline is projected onto the two-dimensional image of the corresponding sidewall of the busbar 1, each point of the spatial three-dimensional geometric baseline is projected onto the two-dimensional image of the corresponding sidewall of the busbar 1 to form a reference point that corresponds one-to-one with each point of the spatial three-dimensional geometric baseline, and then a two-dimensional baseline is formed by multiple reference points.

[0202] Using the pre-calibrated joint extrinsic parameter matrix from the line laser coordinate system to the camera coordinate system and the camera's intrinsic parameter matrix , to establish a three-dimensional geometric baseline Two-dimensional images projected onto a two-dimensional area array camera In the process, a baseline is generated in a two-dimensional pixel coordinate system. ,in Zc This is the depth value of the point in the coordinate system of the area array camera, used for perspective division. The projection transformation formula is:

[0203]

[0204] in, X, Y, Z These represent the three-dimensional coordinate components of any point on the three-dimensional geometric reference line in space;

[0205] u and v These represent the coordinates of a point on a three-dimensional geometric baseline projected onto a point on a two-dimensional image;

[0206] This represents the depth value of a point projected onto a 2D image in the coordinate system of an area array camera.

[0207] K The intrinsic parameter matrix represents the area array camera;

[0208] This represents the joint extrinsic parameter matrix from the line laser profilometer coordinate system to the area array camera coordinate system.

[0209] For rotation matrix, It is a translation vector.

[0210] To ensure dimensional consistency, X, Y, Z, And the length quantities in t use the same unit of length.

[0211] Step 53: Using the two-dimensional baseline as a reference, translate in the direction of the pre-tightening bolt in the two-dimensional image until the pre-tightening bolt 3 is translated to the bottom, and delineate the area where the pre-tightening bolt 3 is located as the target area;

[0212] Using two-dimensional baselines Using the central axis as the pivot, shift upwards by a specific pixel distance to generate a width of... Height is The tubular dynamic wrapping region is binarized into a mask matrix, where the pixel value of the area where the pre-tightening bolt 3 may appear is 1, and the pixel value of the remaining area is 0.

[0213] Step 54: Based on the standard design drawings of busbar 1, use the dynamic mask generation algorithm to identify the pre-tightening bolts 3 in the target area;

[0214] The dynamic mask generation algorithm specifically involves transforming a two-dimensional image... Perform a pixel-by-pixel bitwise AND operation with the mask. After the operation, except for the narrow area where the pre-tightening bolt 3 is located, the pixel values ​​of all "high-brightness noise sources" such as background light tubes, water stain reflections, and cables in the tunnel are set to 0 (pure black).

[0215] Step 55: Use the YOLOv8-tiny target detection network to output the loosening status of the pre-tightened bolt 3;

[0216] The YOLOv8-tiny object detection network is a lightweight version (usually referring to YOLOv8n, the nano version) of the YOLOv8 series of object detection models released by Ultralytics in 2023. It is designed for platforms with limited computing power, such as mobile and embedded devices, and significantly reduces the number of model parameters and computational load while maintaining high detection accuracy, making it suitable for real-time edge deployment.

[0217] In this embodiment, YOLOv8-tiny is used as a classifier / detector to extract features and infer the small target (pre-tightening bolt 3) within the ROI, and output the state category and confidence level of each pre-tightening bolt 3.

[0218] Because the search area is reduced to less than 10% of the entire image and the background is clean, the network can output the connection status of bus 1 fasteners with high accuracy and extremely low computational power consumption. (Loose / Normal) and its confidence level .

[0219] It should be further explained that, regarding the loose bolt status output by the YOLOv8-tiny object detection network in step 55, in order to clarify the legitimate acquisition channel and training loop of this model, and to avoid scenario incompatibility issues caused by the generality of open-source models, this embodiment constructs the training and deployment process based on common methods in this field as follows:

[0220] (1) Legal acquisition and structural optimization of basic network architecture:

[0221] The YOLOv8-tiny object detection network used in this system was obtained through legitimate open-source channels, along with its basic architecture code and initial pre-trained weights trained on public datasets.

[0222] However, the initial open-source model only possesses general object recognition capabilities and lacks the ability to identify the state of minute fasteners in overhead contact lines. Therefore, this embodiment performs local optimizations to the open-source network structure: a channel attention mechanism module (such as CBAM or SE module) is embedded between the feature extraction backbone and the neck. This mechanism enables the network to adaptively focus on high-frequency features such as bolt metal edges and physical gaps, suppressing useless background feature responses.

[0223] (2) Construction of a pure, specialized dataset based on cross-modal physical dimensionality reduction:

[0224] To completely remove background noise, this embodiment utilizes the cross-modal physical dimensionality reduction mechanism described in the above steps to construct a training set.

[0225] Specifically: a line laser profilometer is used to calculate the three-dimensional geometric baseline in space, and then a joint extrinsic parameter matrix is ​​used. The system projects the image onto a 2D image across modalities to generate a tubular dynamic mask that precisely encloses the bolt area. The system then performs a bitwise AND operation between the original image and this dynamic mask to directly crop out clean region of interest (ROI) image blocks that have removed all tunnel background.

[0226] Engineers manually labeled these clean ROI image blocks, assigning them the true labels of "normally tight" (defined as the anti-loosening gasket and the busbar sidewall are pixel-level fit and the anti-loosening marking line is continuous) or "loosened" (defined as the presence of physical gaps in the dark field caused by relative displacement or misalignment of the anti-loosening marking line), thereby establishing a cross-modal specialized training dataset.

[0227] (3) Data augmentation, domain-specific transfer learning, and network weight iteration:

[0228] Before inputting the data into the network for training, in order to cope with the harsh working conditions of subway tunnel excavation, specialized data augmentation was first performed on the labeled clean ROI image blocks: motion blur was introduced to simulate the trailing shadow of a high-speed train inspection, Gaussian noise was injected to simulate the background noise of the dark field, and local strong light masking was applied to simulate the reflection of the tunnel wall spotlights.

[0229] Subsequently, the enhanced training dataset was input into the optimized YOLOv8-tiny network for transfer learning. Open-source pre-trained weights were loaded as initialization parameters; the forward propagation loss was calculated jointly using the bounding box regression loss function and the classification cross-entropy loss function; and the weights of the deep convolutional kernels of the network were iteratively fine-tuned using the backpropagation algorithm and optimizer, ultimately generating a customized weight file specializing in the detection of catenary pre-tightening bolt states.

[0230] (4) Lightweight compilation and real-time inference deployment on the edge:

[0231] To match the embedded hardware environment of the rail-mounted contact wire wear detection device, the trained customized model will have its backpropagation computation graph stripped and exported as a simplified edge inference computation engine format (such as TensorRT or ONNX format). During actual inspection inference, the system only needs to feed the network with tiny image patches after dynamic mask filtering generated in real time on site. With extremely low computational power consumption and memory usage, it can output the classification status of "loose bolts" or "normal" with high confidence.

[0232] This embodiment introduces a dynamic mask based on 3D physical space topological constraints, reducing the construction of training data from "blindly grabbing the whole image" to "precise targeted extraction". This method not only enables open-source basic networks to achieve industrial-grade high noise resistance recognition capabilities at low cost through slight structural improvements and training with specialized data, but also achieves deep fusion of three-dimensional geometric quantities and two-dimensional texture features at the physical level, effectively improving the model's generalization ability and detection reliability under complex subway conditions.

[0233] Step 56: Extract the center line of the other side from the dot matrix image of the bottom of the busbar 1 as the three-dimensional geometric reference line in space. Repeat steps 52-55 to obtain the loosening state of the pre-tightening bolts 3 on both sides of the busbar 1.

Claims

1. A rail-mounted contact wire wear detection system, comprising a rail-mounted contact wire wear detection device suspended from the bottom of a busbar, wherein a contact wire extending along its length is disposed at the center of the bottom end of the busbar, and multiple pre-tightening bolts are disposed along the length of the opposite side walls of the busbar; the rail-mounted contact wire wear detection device comprises a moving mechanism and an obstacle avoidance mechanism, wherein the moving mechanism is capable of moving the rail-mounted contact wire wear detection device along the busbar, and the obstacle avoidance mechanism is capable of moving the rail-mounted contact wire wear detection device to avoid obstacles on the busbar, characterized in that... It also includes a controller, an inertial measurement unit, an absolute encoder, an area array camera, and a line laser profilometer; the controller is electrically connected to the moving mechanism, the absolute encoder, the area array camera, and the line laser profilometer, respectively. The inertial measurement unit is located at the center of gravity of the rail-mounted contact wire wear detection device, and is used to detect the acceleration and angular velocity of the rail-mounted contact wire wear detection device and transmit them to the controller; The absolute encoder is installed on the moving mechanism to detect the distance traveled by the moving mechanism and transmit it to the controller; The area array camera is installed on the side wall of the rail-mounted contact wire wear detection device to capture two-dimensional images of the two side walls of the busbar and transmit them to the controller; wherein, the two-dimensional image is specifically a two-dimensional pixel image containing the pre-tightening bolts on the two side walls of the busbar. The line laser profilometer is positioned below the area array camera. The output beam of the line laser profilometer spans the contact line cross section and forms an angle with the optical axis of the area array camera. The line laser profilometer is used to: acquire a three-dimensional dot matrix image of the bottom of the busbar and transmit it to the controller. The three-dimensional dot matrix image includes a contact line shape curve image and dot matrix images on both sides of the bottom of the busbar; The contact line shape curve image specifically includes a curve composed of multiple points reflecting the shape of the bottom cross-section edge of the contact line; the busbar bottom side dot matrix image specifically includes a dot matrix image composed of multiple points reflecting the shape of the bottom ends of the two side walls of the busbar. The controller is installed inside the rail-mounted contact wire wear detection device, and the controller is used for: Control the moving mechanism to move along the busbar; Control the obstacle avoidance mechanism to avoid obstacles; Set the sampling step size for the line laser profilometer, area array camera, and inertial measurement unit; The linear laser profilometer and the area array camera perform synchronous sampling at a fixed step size; Image correction is performed on the contact wire shape curve image to obtain a standard contact wire shape curve image; Image processing is performed on the standard contact wire shape curve image, the dot matrix image and the two-dimensional image on both sides of the bottom of the busbar to obtain the wear condition of each position of the contact wire and the loosening condition of the pre-tightening bolts.

2. A method for detecting contact wire wear based on the rail-mounted contact wire wear detection system according to claim 1, characterized in that, Includes the following steps: Step 1: Set the sampling step size for the line laser profilometer and the area scan camera; Step 2: Move the moving mechanism at a constant speed along the busbar. When the moving distance reaches the sampling step length set in Step 1, have the line laser profilometer and the area scan camera sample simultaneously. Acquire 3D raster images and 2D images respectively; Record the position at the time of this sampling as the current position; Step 3: Perform image correction on the contact line shape curve image, and use the corrected three-dimensional dot matrix image as the standard contact line shape curve image, then proceed to Step 4. Step 4: Set the surface curvature determination threshold for the contact line; Based on the surface curvature of the curve in the standard contact wire shape curve image, determine whether the contact wire has experienced wear at the current position based on the relationship between the surface curvature and the surface curvature judgment threshold. Step 5: Project the center lines of the bottom sides of the busbar in the dot matrix images on both sides of the busbar onto the corresponding two-dimensional images to serve as two-dimensional baselines; Using the two-dimensional baseline as a reference, the area where the pre-tightening bolts are located is delineated in the two-dimensional image as the target area; The target area is subjected to dimensionality reduction and noise reduction processing to obtain a target image containing only the pre-tightened bolts, and the loosening of the pre-tightened bolts in the target image is identified. Step 6: Repeat steps 2-5 until the rail-mounted contact wire wear detection device reaches the end of the busbar, and obtain the wear condition of the contact wire and the loosening condition of the pre-tightening bolts at each sampling position of the busbar.

3. The contact wire wear detection method of the rail-mounted contact wire wear detection system as described in claim 2, characterized in that, In step 1, the sampling step size is 5-15mm.

4. The contact wire wear detection method of the rail-mounted contact wire wear detection system as described in claim 2, characterized in that, Step 3 specifically includes the following sub-steps: Step 31: Using the error state extended Kalman filter algorithm, calculate the roll angle and pitch angle of the rail-mounted contact wire wear detection device based on the current position of the rail-mounted contact wire wear detection device at the time of sampling, the speed at the time of sampling, the acceleration and angular velocity at the time of sampling. Step 32: Based on the roll angle and pitch angle calculated in step 31, construct the inverse kinematics compensation matrix for each point in the contact line shape curve image; Step 33: For any point in the 3D point image, perform an inverse transformation using the inverse kinematics compensation matrix corresponding to that point to obtain the corrected point; Using the above method, traverse each point in the 3D dot matrix image to obtain the corrected contact line shape curve image, and use the corrected contact line shape curve image as the standard contact line shape curve image.

5. The contact wire wear detection method of the rail-mounted contact wire wear detection system as described in claim 4, characterized in that, Step 4 specifically includes the following sub-steps: Step 41, set the surface curvature determination threshold, as follows: in, This indicates the threshold for determining surface curvature. The mathematical expectation mean of the surface curvature of the standard CAD section of the contact wire; This represents the tolerance sensitivity adjustment coefficient. ; The standard deviation of the surface curvature of the standard CAD cross-section of the contact wire; Step 42: For a standard contact line shape curve image, select one endpoint of the curve and construct the neighborhood point set of the selected point. Covariance matrix; Regarding the aforementioned The covariance matrix is ​​subjected to eigenvalue decomposition to obtain three eigenvalues ​​arranged in ascending order. ; Step 43: Calculate the surface curvature corresponding to the selected point by taking the ratio of the smallest eigenvalue to the sum of the three eigenvalues, as shown in the following formula: in, This indicates the surface curvature corresponding to the selected point; Step 44, if the surface curvature of the selected point is... ≤ Surface curvature determination threshold If the selected point has been worn, then the point is designated as the wear point and proceed to step 46. If the surface curvature of the selected point is > Surface curvature determination threshold If the selected point does not show wear, proceed to step 45. Step 45: Select a point in the standard contact line shape curve image that is adjacent to the point selected in step 45, and repeat steps 43-44 until all points in the standard contact line shape curve image are traversed or wear points are obtained. If all points in the standard contact wire shape curve image have been traversed and there are no wear points, then the contact wire at the current position is not worn. If the wear points are obtained, then all surface curvatures... > Surface curvature determination threshold The point is designated as the unworn point, proceed to step 46; Step 46: For the standard contact line shape curve image, select the point where the other endpoint of the curve is located, and repeat steps 42-45 until another wear point is obtained; Step 47: Select the unworn points adjacent to the two worn points as the two lug points; Step 48: Based on the standard CAD cross-section of the contact wire, use the ICP precise matching algorithm to correct all unworn points and lug points; Connect the two corrected lug points, obtain the true coordinates of the two corrected lug points, connect the two lug points, and take the straight line formed by connecting the two corrected lug points as the wear line at the current position of the contact line. Take the area formed by the wear line and the bottom of the standard CAD section as the wear area at the current position of the contact line.

6. The contact wire wear detection method of the rail-mounted contact wire wear detection system as described in claim 5, characterized in that, Step 5 specifically includes the following sub-steps: Step 51: Extract the center line of the bottom side of the busbar from the dot matrix image of that side as the spatial three-dimensional geometric reference line. Step 52: Project the three-dimensional geometric baseline onto the two-dimensional image of the corresponding sidewall of the busbar to form a two-dimensional baseline; Step 53: Using the two-dimensional baseline as a reference, translate in the direction of the pre-tightening bolt in the two-dimensional image until the bottom of the pre-tightening bolt is reached, and delineate the area where the pre-tightening bolt is located as the target area; Step 54: Based on the standard design drawings of the bus, use a dynamic mask generation algorithm to identify the pre-tightening bolts in the target area; Step 55: Use the YOLOv8-tiny target detection network to output the loosening status of the pre-tightened bolts; Step 56: Extract the center line of the other side from the dot matrix image at the bottom of the busbar as the three-dimensional geometric reference line in space. Repeat steps 52-55 to obtain the loosening state of the pre-tightening bolts on both sides of the busbar.

7. The contact wire wear detection method of the rail-mounted contact wire wear detection system as described in claim 6, characterized in that, Step 52 involves projecting the spatial three-dimensional geometric baseline onto the two-dimensional image of the corresponding sidewall of the busbar. Specifically, each point of the spatial three-dimensional geometric baseline is projected onto the two-dimensional image of the corresponding sidewall of the busbar, forming a reference point that corresponds one-to-one with each point of the spatial three-dimensional geometric baseline, and then forming a two-dimensional baseline from multiple reference points. The projection transformation formula for any point on the three-dimensional geometric baseline is as follows: in, X, Y, Z These represent the three-dimensional coordinate components of any point on the three-dimensional geometric reference line in space; u and v These represent the coordinates of a point on a three-dimensional geometric baseline projected onto a point on a two-dimensional image; This represents the depth value of a point projected onto a 2D image in the coordinate system of an area array camera. K The intrinsic parameter matrix represents the area array camera; This represents the joint extrinsic parameter matrix from the line laser profilometer coordinate system to the area array camera coordinate system.

Images