Cutting equipment and cutting method for replacing and repairing side walls and end walls of gondolas

Abstract

The present invention relates to the technical field of railway open car body inspection and maintenance, and specifically to a cutting device for the replacement and inspection of open car side walls and end walls. The device comprises a robot and a control system, wherein the robot is provided with a cutting actuator and a laser scanning device; the laser scanning device is used to perform laser scanning on the open car side wall, obtain point cloud data of the open car side wall, generate a point cloud image, and send it to the control system, wherein the point cloud data includes coordinate information and depth information of each pixel point and the position data of the robot during scanning; the control system comprises a recognition module; the recognition module is used to recognize the point cloud image, reconstruct a three-dimensional model of the point cloud after processing the point cloud image according to the point cloud information, recognize the weld position, and generate a cutting path; the instruction sending module is used to generate a cutting instruction according to the cutting route and the point cloud data on the cutting path, and send the cutting instruction to the cutting actuator; the cutting actuator is used to cut the open car side wall according to the cutting instruction.

Description

Technical Field

[0001] The invention relates to the technical field of railway gondola car body maintenance, and in particular to a cutting device and a cutting method for replacing and repairing the side walls and end walls of a gondola car. Background Art

[0002] Rail transport accounts for a significant portion of my country's transportation industry due to its advantages, such as large cargo capacity and low freight rates. Gondolas are a type of railway freight car. Their all-steel welded structure consists of a chassis, side walls, end walls, and doors. They are primarily used to transport weather-resistant bulk cargo such as coal and ballast. Corrosion caused by transporting cargo in gondolas, as well as wear, oxidation, and corrosion from long-term use, pose serious risks to these vehicles, necessitating the removal and replacement of steel plates from the side walls.

[0003] Currently, flame cutting is widely used for railway freight car maintenance. However, when inspecting open wagon bodies, traditional manual flame cutting is used to cut areas that need to be patched or replaced. This results in poorly guaranteed manual cutting results and low weld positioning accuracy. Traditional flame cutting also suffers from low efficiency, high labor intensity, high energy consumption, and poor quality. While industrial robots currently use this method to locate and cut welds, the point cloud image captured by the industrial camera, which is mounted on the robot arm, is directly related to the robot's position. During the scanning process, errors caused by changes in the robot's position can affect the accuracy of the point cloud registration, resulting in errors in the final cutting result. Summary of the Invention

[0004] The present invention aims to provide a cutting device and a cutting method for the replacement and maintenance of the side walls and end walls of a gondola, so as to replace the traditional manual cutting method and ensure the cutting accuracy.

[0005] The basic solution provided by the present invention is a cutting device for the replacement and maintenance of the side walls and end walls of a gondola, comprising a robot and a control system, wherein the robot is provided with a cutting actuator and a laser scanning device;

[0006] The laser scanning device is used to perform laser scanning on the side wall of the open car, obtain point cloud data of the side wall of the open car, generate a point cloud image and send it to the control system. The point cloud data includes the coordinate information and depth information of each pixel point and the position data of the robot during scanning;

[0007] The control system includes an identification module;

[0008] The recognition module is used to recognize the point cloud image, process the point cloud image according to the point cloud information, reconstruct the three-dimensional model of the point cloud, identify the weld position, and generate the cutting path;

[0009] An instruction sending module is used to generate cutting instructions based on the cutting route and point cloud data on the cutting path, and send the cutting instructions to the cutting execution mechanism;

[0010] The cutting actuator is used to cut the side wall of the open car according to the cutting instruction.

[0011] The principle of this invention is that when cutting a gondola sidewall, a laser scanner acquires point cloud data of the gondola sidewall. This point cloud data contains the coordinates and depth information of each pixel, as well as the robot's position data during scanning. The scanned point cloud data is sent to a control system, which processes the point cloud image and reconstructs a 3D model to identify the weld locations. Based on this, a cutting path is generated, and the cutting actuator is controlled to cut the gondola sidewall according to the generated cutting path.

[0012] Compared with existing technologies, the use of robotic cutting improves cutting efficiency, achieves higher cutting precision, and achieves better cutting results. At the same time, by collecting and processing the scanned point cloud data, the accuracy of point cloud reconstruction is improved, the accumulated error in the positioning process is reduced, and the accuracy of machine recognition is improved.

[0013] Furthermore, the recognition module includes an edge extraction module and a data processing module;

[0014] The edge extraction module is used to mark the point cloud image collected by the laser scanning device as image A, perform edge extraction based on the depth information of image A, obtain the suspected weld position, and extract the suspected weld position and nearby pixels to create a mask to generate a mask image;

[0015] The data processing module is used to perform an AND operation on image A and the inverse code of the mask image and perform downsampling to obtain image B, perform an AND operation on image A and the mask image to obtain image C, perform an OR operation on image C and image B to obtain image D, and use image D as the modeling image.

[0016] Image A is edge-extracted, and using its depth information as a feature, the suspected weld locations are extracted and their approximate positions are determined. Pixels at the suspected weld locations are comprehensively utilized, and pixels near the suspected weld locations are extracted to create a mask image. An AND operation is performed between Image A and the inverse of the mask image to obtain a non-edge region that does not contain weld information. This region is then downsampled to reduce the point cloud data, resulting in Image B. An AND operation is performed between Image A and the mask image to obtain Image C, which contains all pixels at the suspected weld locations. An OR operation is performed between Image C and Image B. Compared to Image A, Image D retains the point cloud information at the weld locations, reducing the amount of point cloud data and the computational effort required for subsequent point cloud reconstruction, thereby increasing the speed of point cloud reconstruction.

[0017] Furthermore, the recognition module also includes an image grouping module, a group modeling module, a loop detection module, a fusion modeling module and a correction module;

[0018] An image grouping module is configured to identify image features in the modeling images and classify the modeling images into three categories based on the image features, wherein the image features include images containing diagonal braces, images containing vertical columns, and images without diagonal braces or vertical columns. Each type of modeling image is grouped based on the pose data so that the image features and pose data of each group of modeling images are the same.

[0019] The group modeling module builds a dense point cloud model of each group based on the modeling image of each frame and the corresponding pose data in each group;

[0020] The loop detection module is used to compare the depth information of each modeling image in a group with the reconstructed dense point cloud model, and determine whether the depth information difference between each frame modeling image and the dense point cloud model is within the set threshold. If not, the depth information and pose data of each frame modeling image in the group are corrected according to the dense geometric consistency and photometric consistency;

[0021] The fusion modeling module performs fusion reconstruction based on the image features of each group and the reconstructed dense point cloud model to obtain the overall dense point cloud model;

[0022] The loop detection module is also used to compare the depth information of the dense point cloud model of each group with the overall dense point cloud model to determine whether the depth information is within a set threshold. If not, the reconstructed point cloud and position data of each group are corrected according to the dense geometric consistency and photometric consistency.

[0023] The correction module builds the final dense point cloud model of the vehicle side wall and corrects the robot's reference position and pose data based on the loop detection results.

[0024] By extracting image features from the modeling images, the modeling images are classified into three categories: images containing diagonal braces, images containing columns, and images without columns and diagonal braces. The three categories of images are then grouped according to their pose data, so that the image features in each group of images are similar to the pose data, reducing the amount of calculation. The cumulative error is reduced by loop detection. During loop detection, the camera pose is corrected by detecting the similarity between the current position and historical data. In the present invention, the pose is corrected by using the similarity between the current position of each frame image and the position of the established dense point cloud, reducing the cumulative error of the camera pose caused by the movement of the robot hand-eye system during the scanning process. Based on the similarity between the image features and the pose of each frame after grouping, the image coordinates of a group are transformed and fused for reconstruction. Within each group, the positional information of each image frame is compared with the reconstructed dense point cloud based on dense geometric consistency and photometric consistency. The difference between the position of each data point in each frame and the reconstructed dense point cloud is used as feedback to optimize and correct the details of the depth map of each frame. This process is repeated until the position difference between the reconstructed dense point cloud and each frame is within a certain threshold. The dense point clouds created in each group are then fused using the similarity of the bag-of-words model between the groups. The previous steps are repeated to finally create the final dense point cloud of the vehicle side wall and the corrected robot baseline position and pose data.

[0025] Furthermore, the identification module also includes a depth grouping module, a plane fitting module and a weld determination module;

[0026] A depth grouping module is used to group the point cloud data of the dense point cloud model of the vehicle side wall according to the depth information, grouping the point cloud data with the same depth information into one group, and grouping the adjacent point cloud data with a depth information change greater than a preset threshold into another group;

[0027] A plane fitting module is used to perform plane fitting on the grouped point cloud data. The fitted planes include the steel plate plane, the diagonal brace vertical plane, the diagonal brace wing plane, the diagonal brace wing weld plane, the side wall column plane, the side wall column wing weld plane, and the side wall column wing plane.

[0028] The weld determination module is used to determine the coordinate information of the weld according to the intersection lines of each plane.

[0029] In the reconstructed dense point cloud model of the vehicle side wall, the pixel coordinates of the weld locations must be determined first. Based on this dense point cloud model, point cloud data with the same depth are grouped together, and point cloud data with significant depth variations are grouped together. Plane fitting is performed on the grouped point cloud data, and the fitted planes are divided into six types: steel plate plane, diagonal brace vertical plane, diagonal brace wing plane, diagonal brace wing plane weld plane, side wall column plane, side wall column wing plane weld plane, and side wall column wing plane. The final weld locations are determined based on the intersection of these planes. Welds on the side wall primarily occur near the diagonal braces and columns, and the pixels at the weld edge have a certain degree of height difference. In the point cloud image, the 3D position information of each pixel can be obtained. Plane fitting is performed on the 3D data near the weld, and the intersection of the two planes is determined as the weld location. The cutting height is determined based on the depth information in the image, and the final cutting path is planned.

[0030] Furthermore, the recognition module also includes a path planning module and a posture determination module;

[0031] The path planning module is used to determine the depth information of each pixel at the weld position based on the weld coordinate information and the dense point cloud model of the vehicle side wall;

[0032] The posture determination module is used to determine the posture data of the robot at each welding position based on the correction of the robot's reference position posture data.

[0033] After determining the cutting path, the cutting depth is determined based on the depth information of each pixel point on the cutting path. The robot's position at each weld position is determined based on the corrected robot reference position data, thereby achieving precise cutting of the open car side wall.

[0034] The present invention also discloses a cutting method for replacing and repairing the side walls and end walls of a gondola car. The method uses the above-mentioned cutting device for replacing and repairing the side walls and end walls of a gondola car. The method comprises the following steps:

[0035] Data acquisition step: performing laser scanning on the side wall of the open car using a laser scanning device to obtain point cloud data and generate a point cloud image. The point cloud data includes coordinate information and depth information of each pixel point and the position data of the robot during scanning;

[0036] Recognition step: identify the point cloud image, process the point cloud image according to the point cloud information, reconstruct the three-dimensional model, identify the weld position, and generate the cutting path;

[0037] Cutting execution steps: sending instructions to the cutting actuator, so that the cutting actuator cuts the side wall of the open car according to the cutting path.

[0038] Furthermore, in the edge extraction step, the point cloud image collected by the laser scanning device is marked as image A, and edge extraction is performed based on the depth information of image A to obtain the suspected weld position, and the suspected weld position and nearby pixels are extracted to create a mask to generate a mask image;

[0039] Data processing steps: perform an AND operation on image A and the inverse code of the mask image and perform downsampling to obtain image B, perform an AND operation on image A and the mask image to obtain image C, perform an OR operation on image C and image B to obtain image D, and use image D as the modeling image.

[0040] Furthermore, the identification step further includes the following steps:

[0041] Image grouping step: identifying image features in the modeling images and classifying the modeling images into three categories based on the image features, wherein the image features include images containing diagonal braces, images containing vertical columns, and images without diagonal braces or vertical columns. Each category of modeling images is grouped based on the pose data so that the image features and pose data of each group of modeling images are the same;

[0042] Group modeling step: Based on each frame modeling image and corresponding pose data in each group, a dense point cloud model of each group of fusion reconstruction is established;

[0043] Group loop detection step: Compare the depth information of each frame modeling image in a group with the reconstructed dense point cloud model to determine whether the depth information difference between each frame modeling image and the dense point cloud model is within the set threshold. If not, the depth information and pose data of each frame modeling image in the group are corrected according to the dense geometric consistency and photometric consistency;

[0044] In the fusion modeling step, the image features of each group and the reconstructed dense point cloud model are fused and reconstructed to obtain the overall dense point cloud model;

[0045] Overall loop detection step: Compare the depth information of each group's dense point cloud model with the overall dense point cloud model to determine whether the depth information is within the set threshold. If not, correct the reconstructed point cloud and position data of each group based on dense geometric consistency and photometric consistency;

[0046] In the correction step, based on the loop detection results, the final dense point cloud model of the vehicle side wall and the correction robot's reference position and pose data are established.

[0047] Furthermore, the identification step further includes the following steps:

[0048] Depth grouping step: group the point cloud data of the overall dense point cloud model according to the depth information, group the point cloud data with the same depth information into one group, and group the adjacent point cloud data with a depth information change greater than a preset threshold into another group;

[0049] Plane fitting step: performing plane fitting on the grouped point cloud data. The fitted planes include the steel plate plane, the diagonal brace vertical plane, the diagonal brace wing plane, the diagonal brace wing weld plane, the side wall column plane, the side wall column wing weld plane, and the side wall column wing plane.

[0050] Weld seam determination steps: Determine the coordinate information of the weld seam based on the intersection lines of each plane.

[0051] Furthermore, the identification step further includes the following steps:

[0052] In the path planning step, the depth information of each pixel at the weld position is determined based on the weld coordinate information and the dense point cloud model of the vehicle side wall.

[0053] The posture determination step determines the posture data of the robot at each weld position based on the corrected robot reference position posture data. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 This is a structural schematic diagram of an embodiment of a cutting device for replacing and repairing the side walls and end walls of a gondola car according to the present invention;

[0055] Figure 2 The present invention is a logic block diagram of an embodiment of a cutting device for replacing and repairing the side walls and end walls of a gondola. DETAILED DESCRIPTION

[0056] The following is further described in detail through specific implementation methods:

[0057] The embodiment is basically as shown in the attached Figure 1 and attached Figure 2 As shown:

[0058] A cutting device for the replacement and maintenance of open-top car side walls includes a robot and a control system. The robot is equipped with a cutting actuator and a laser scanning device. Specifically, in this embodiment, the robot is a six-section robot 1 mounted on an AGV vehicle-mounted platform 5. The laser scanning device 2 is an industrial laser camera with an accuracy of 0.1mm. The cutting actuator includes a plasma cutting saw 3, both of which are mounted on the robot arm. The AGV vehicle-mounted platform 5 also houses a control cabinet 4, specifically a programmable logic controller (PLC). The control system is housed in the control cabinet 4.

[0059] The laser scanning device is used to perform laser scanning on the side walls of a gondola, acquiring point cloud data from the side walls, generating a point cloud image, and transmitting it to the control system. The point cloud data specifically includes coordinate information and depth information for each pixel, as well as the robot's position during scanning. Coordinate information refers to the coordinate position of each pixel in the global coordinate system, depth information refers to the depth of each pixel relative to the same plane, and position data refers to the robot's posture, angle, and other postures during scanning. During laser scanning, a QR code is first placed at the center of the gondola side wall, providing the center point for scanning and determining the global coordinate reference position of the 3D model. Before scanning, the vehicle body is treated with anti-reflective treatments such as rust prevention and painting. Scanning is performed under different light intensities. In this embodiment, 30%, 50%, and 80% light intensities are selected for laser scanning to minimize the impact of external factors such as light, rust, and reflections on the laser scanning imaging.

[0060] The control system includes a recognition module for identifying point cloud images, processing the point cloud images according to the point cloud information, reconstructing the three-dimensional model of the point cloud, identifying the weld position, and generating a cutting path.

[0061] Specifically, the recognition module includes an edge extraction module, a data processing module, an image grouping module, a group modeling module, a loop detection module, a fusion modeling module, a correction module, a depth grouping module, a correction module, a plane fitting module and a weld determination module.

[0062] The edge extraction module is used to mark the point cloud image collected by the laser scanning device as image A, perform edge extraction based on the depth information of image A, obtain the suspected weld position, and extract the suspected weld position and nearby pixel points to create a mask to generate a mask image.

[0063] Specifically, first, by using a suitable edge extraction operator, the gradient of the depth information is calculated, and the depth information of image A is regarded as a two-dimensional surface. The depth mutation part is extracted, that is, when the depth information difference between adjacent pixels exceeds a certain threshold, it is considered to be a depth mutation part, and the depth mutation part is used as the suspected weld position. Then, the suspected weld position and the nearby pixels are extracted to produce a mask image. The mask operation of the image refers to recalculating the value of each pixel in the image through the mask kernel operator. The mask kernel characterizes the degree of influence of the domain pixel point on the new pixel value, and at the same time, the original pixel point is weighted averaged according to the weight factor in the mask operator. In this embodiment, a mask image is produced by a pixel domain traversal method. Based on the image domain traversal, the source data matrix is ​​operated, the center target point is calculated with the current pixel point position, the mask kernel operator template is moved pixel by pixel, the original image data is traversed, and then each pixel point value corresponding to the new image is updated.

[0064] The data processing module is used to perform an AND operation on image A and the inverse of the mask image and downsample to obtain image B. Image A and the mask image are then ANDed to obtain image C. Image C is then ORed with image B to obtain image D, which is used as the modeling image. The AND operation logic is that all 0s are 0 and all 1s are 1, while the OR operation logic is that all 1s are 1 and all 0s are 0. Therefore, image B is a non-edge region, and image B is downsampled to reduce the amount of point cloud data in this non-edge region. Image C contains all pixels at suspected weld locations, and all pixels at suspected weld locations are retained. Image B is ORed with image C to obtain image D. Compared to image A, image D retains the point cloud data at the weld locations while reducing the amount of point cloud data and the computational complexity of subsequent point cloud reconstruction, thereby increasing the speed of point cloud reconstruction.

[0065] The image grouping module is used to identify image features in the modeling images and classify them into three categories based on these features: images containing diagonal braces, images containing columns, and images without diagonal braces or columns. Each category of modeling images is then grouped based on their pose data, ensuring that the image features and pose data of each group of modeling images are consistent. Specifically, the modeling images are classified into three categories using clustering. Based on these three categories, k frames are grouped into groups based on their corresponding interdependent poses. A bag-of-visual-words model is then constructed based on the image features of each group. In this case, each group of data contains similar image features and pose data.

[0066] The group modeling module establishes a dense point cloud model of each group of fusion reconstruction based on the modeling image of each frame in each group and the corresponding pose data.

[0067] The loop detection module is used to compare the depth information of each frame modeling image in a group with the reconstructed dense point cloud model, and determine whether the depth information difference between each frame modeling image and the dense point cloud model is within the set threshold. If not, the depth information and pose data of each frame modeling image in the group are corrected according to the dense geometric consistency and photometric consistency.

[0068] Since the coordinate transformation of the point cloud data of the scanned vehicle body is performed through the reference position and the posture data of the robot during the scanning process, the multiple frames of images obtained at different times, angles, and illuminations are superimposed and matched into a unified coordinate system. The industrial camera is installed on the robot arm, and the point cloud image scanned by the industrial camera is directly related to the posture data of the robot. During the scanning process, the robot posture changes and errors will gradually accumulate during image stitching, which will have a certain impact on the final accuracy of the point cloud alignment. Therefore, loop detection is used to reduce the cumulative error. The loop detection corrects the camera posture by detecting the similarity between the current position and the historical position. In this embodiment, the similarity between the current position of each frame of the modeled image and the position of the established dense point cloud is used to correct the posture, thereby reducing the accumulated error of the camera posture caused by the movement of the robot's hand-eye system during the scanning process. Based on the similarity of image features and posture of each frame image after grouping, a group of image coordinates are transformed and fused and reconstructed. In each group, the position information of each frame image in a group is compared with the reconstructed dense point cloud based on dense geometric consistency and photometric consistency. The gap between the position of each data point in each frame image and the reconstructed dense point cloud is used as feedback to optimize the details of the depth map of each frame image. It is continuously iterated until the position gap between the reconstructed dense point cloud and each frame image is within a certain threshold.

[0069] The fusion modeling module performs fusion reconstruction based on the image features of each group and the reconstructed dense point cloud model to obtain the overall dense point cloud model.

[0070] The loop detection module is also used to compare the depth information of each group's dense point cloud model with the overall dense point cloud model to determine whether the depth information is within a set threshold. If not, the reconstructed point cloud and position data of each group are corrected according to the dense geometric consistency and photometric consistency.

[0071] The correction module builds the final dense point cloud model of the vehicle side wall and corrects the robot's reference position and pose data based on the loop detection results.

[0072] Using the similarity of the bag-of-words model between each group, the dense point cloud established in each group is fused, and the previous steps are repeated to finally establish the final dense point cloud of the vehicle side wall and the correction robot reference position and posture data.

[0073] A depth grouping module is used to group the point cloud data of the dense point cloud model of the vehicle side wall according to the depth information, grouping the point cloud data with the same depth information into one group, and grouping the adjacent point cloud data with a depth information change greater than a preset threshold into another group;

[0074] A plane fitting module is used to perform plane fitting on the grouped point cloud data. The fitted planes include the steel plate plane, the diagonal brace vertical plane, the diagonal brace wing plane, the diagonal brace wing weld plane, the side wall column plane, the side wall column wing weld plane, and the side wall column wing plane.

[0075] In the reconstructed dense point cloud model of the vehicle side wall, the pixel coordinates of the weld locations must first be determined. Because the data processing module filters the point cloud data, the point cloud data of the weld locations in the dense point cloud model of the side wall is retained intact. Based on the established dense point cloud model of the side wall, point cloud data with the same depth are grouped into one category, and point cloud data with significant depth changes are grouped into another category. Due to point cloud registration and data fusion, the error caused by this step is minimized. Plane fitting is performed on the grouped data points, and the fitted planes are divided into six types: steel plate plane, diagonal brace facade plane, diagonal brace wing plane, diagonal brace wing weld plane, side wall column plane, side wall column wing weld plane, and side wall column wing plane. The pixel coordinates of the weld on the vertical side of the diagonal brace and the weld on the wing side of the diagonal brace are determined according to the intersection line of the steel plate plane and the vertical plane of the diagonal brace and the intersection line of the diagonal brace wing plane and the weld plane of the diagonal brace wing. The pixel coordinates of the weld on the vertical side of the diagonal brace and the weld on the wing side of the diagonal brace are determined according to the intersection line of the steel plate plane and the vertical plane of the column and the intersection line of the column wing plane and the weld plane of the column wing.

[0076] The weld determination module is used to determine the coordinate information of the weld according to the intersection lines of each plane.

[0077] The instruction sending module is used to generate cutting instructions according to the cutting route and the point cloud data on the cutting path, and send the cutting instructions to the cutting execution mechanism.

[0078] The cutting actuator is used to cut the side wall of the open car according to the cutting instruction.

[0079] Based on the determined pixel coordinates of the weld edge and the dense point cloud model of the side wall, the depth information of each pixel point at the weld position is determined. The transfer matrix between the dense point cloud and the robot's reference position and posture determines the robot's posture data at each obtained weld coordinate. The instruction sending module sends the cutting path, depth and posture during cutting to the cutting actuator, allowing the robot to drive the cutting knife to cut, allowing the robot to handle tearing, bulging and deformation of the vehicle body, while ensuring that the steel plate after cutting will not be excessively deformed, and the trajectory after cutting will be smooth, which is convenient for subsequent welding steel plate maintenance.

[0080] This embodiment also discloses a method for cutting a gondola car side wall replacement area, which uses the above-mentioned gondola car side wall replacement and repair cutting device, and includes the following steps:

[0081] Data acquisition step: performing laser scanning on the side wall of the open car using a laser scanning device to obtain point cloud data and generate a point cloud image. The point cloud data includes coordinate information and depth information of each pixel point and the position data of the robot during scanning;

[0082] Recognition step: identify the point cloud image, process the point cloud image according to the point cloud information, reconstruct the three-dimensional model, identify the weld position, and generate the cutting path;

[0083] Cutting execution steps: sending instructions to the cutting actuator, so that the cutting actuator cuts the side wall of the open car according to the cutting path.

[0084] Furthermore, in the edge extraction step, the point cloud image collected by the laser scanning device is marked as image A, and edge extraction is performed based on the depth information of image A to obtain the suspected weld position, and the suspected weld position and nearby pixels are extracted to create a mask to generate a mask image;

[0085] Data processing steps: perform an AND operation on image A and the inverse code of the mask image and perform downsampling to obtain image B, perform an AND operation on image A and the mask image to obtain image C, perform an OR operation on image C and image B to obtain image D, and use image D as the modeling image.

[0086] The identifying step further comprises the following steps:

[0087] Image grouping step: identifying image features in the modeling images and classifying the modeling images into three categories based on the image features, wherein the image features include images containing diagonal braces, images containing vertical columns, and images without diagonal braces or vertical columns. Each category of modeling images is grouped based on the pose data so that the image features and pose data of each group of modeling images are the same;

[0088] Group modeling step: Based on each frame modeling image and corresponding pose data in each group, a dense point cloud model of each group of fusion reconstruction is established;

[0089] Group loop detection step: Compare the depth information of each frame modeling image in a group with the reconstructed dense point cloud model to determine whether the depth information difference between each frame modeling image and the dense point cloud model is within the set threshold. If not, the depth information and pose data of each frame modeling image in the group are corrected according to the dense geometric consistency and photometric consistency;

[0090] In the fusion modeling step, the image features of each group and the reconstructed dense point cloud model are fused and reconstructed to obtain the overall dense point cloud model;

[0091] Overall loop detection step: Compare the depth information of each group's dense point cloud model with the overall dense point cloud model to determine whether the depth information is within the set threshold. If not, correct the reconstructed point cloud and position data of each group based on dense geometric consistency and photometric consistency;

[0092] In the correction step, based on the loop detection results, the final dense point cloud model of the vehicle side wall and the correction robot's reference position and pose data are established.

[0093] The identifying step further comprises the following steps:

[0094] Depth grouping step: group the point cloud data of the overall dense point cloud model according to the depth information, group the point cloud data with the same depth information into one group, and group the adjacent point cloud data with a depth information change greater than a preset threshold into another group;

[0095] Plane fitting step: performing plane fitting on the grouped point cloud data. The fitted planes include the steel plate plane, the diagonal brace vertical plane, the diagonal brace wing plane, the diagonal brace wing weld plane, the side wall column plane, the side wall column wing weld plane, and the side wall column wing plane.

[0096] Weld seam determination steps: Determine the coordinate information of the weld seam based on the intersection lines of each plane.

[0097] The identifying step further comprises the following steps:

[0098] In the path planning step, the depth information of each pixel at the weld position is determined based on the weld coordinate information and the dense point cloud model of the vehicle side wall.

[0099] The posture determination step determines the posture data of the robot at each weld position based on the corrected robot reference position posture data.

[0100] The above are only embodiments of the present invention. It should be noted that the technical solutions of the above embodiments can be applied to the replacement and maintenance of the side walls of gondola cars, as well as the replacement and maintenance of the end walls of gondola cars. Common knowledge such as the known specific structures and characteristics in the solution are not described in detail here. Ordinary technicians in the relevant field are aware of all common technical knowledge in the technical field of the invention before the application date or priority date, can obtain all existing technologies in the field, and have the ability to apply conventional experimental means before that date. Ordinary technicians in the relevant field can improve and implement this solution based on their own abilities under the guidance of this application. Some typical known structures or known methods should not become obstacles for ordinary technicians in the relevant field to implement this application. It should be pointed out that for those skilled in the art, without departing from the structure of the present invention, several variations and improvements can be made, which should also be regarded as the scope of protection of the present invention. These will not affect the effect of the implementation of the present invention and the practicality of the patent. The scope of protection required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.

Claims

1. A cutting device for the replacement and maintenance of the side walls and end walls of a gondola, characterized by: It includes a robot and a control system, wherein the robot is provided with a cutting actuator and a laser scanning device; The laser scanning device is used to perform laser scanning on the side wall of the gondola, obtain point cloud data of the side wall of the gondola, generate a point cloud image, and send it to the control system. The point cloud data includes the coordinate information and depth information of each pixel point and the position data of the robot during scanning; The control system includes an identification module; The recognition module is used to recognize the point cloud image, process the point cloud image according to the point cloud information, reconstruct the three-dimensional model of the point cloud, identify the weld position, and generate the cutting path; An instruction sending module is used to generate cutting instructions based on the cutting route and point cloud data on the cutting path, and send the cutting instructions to the cutting execution mechanism; A cutting actuator, used to cut the side wall of the gondola according to the cutting instruction; The recognition module includes an edge extraction module and a data processing module; The edge extraction module is used to mark the point cloud image collected by the laser scanning device as image A, perform edge extraction based on the depth information of image A, obtain the suspected weld position, and extract the suspected weld position and nearby pixels to create a mask to generate a mask image; A data processing module is configured to perform an AND operation on image A and the inverse of the mask image and perform downsampling to obtain image B, perform an AND operation on image A and the mask image to obtain image C, perform an OR operation on image C and image B to obtain image D, and use image D as the modeling image; The recognition module also includes an image grouping module, a group modeling module, a loop detection module, a fusion modeling module and a correction module; An image grouping module is configured to identify image features in the modeling images and classify the modeling images into three categories based on the image features, wherein the image features include images containing diagonal braces, images containing vertical columns, and images without diagonal braces or vertical columns. Each type of modeling image is grouped based on the pose data so that the image features and pose data of each group of modeling images are the same. The group modeling module builds a dense point cloud model of each group based on the modeling image of each frame and the corresponding pose data in each group; The loop detection module is used to compare the depth information of each modeling image in a group with the reconstructed dense point cloud model, and determine whether the depth information difference between each frame modeling image and the dense point cloud model is within the set threshold. If not, the depth information and pose data of each frame modeling image in the group are corrected according to the dense geometric consistency and photometric consistency; The fusion modeling module performs fusion reconstruction based on the image features of each group and the reconstructed dense point cloud model to obtain the overall dense point cloud model; The loop detection module is also used to compare the depth information of the dense point cloud model of each group with the overall dense point cloud model to determine whether the depth information is within a set threshold. If not, the reconstructed point cloud and position data of each group are corrected according to the dense geometric consistency and photometric consistency. The correction module builds the final dense point cloud model of the vehicle side wall and corrects the robot's reference position and pose data based on the loop detection results.

2. The cutting equipment for replacing and repairing the side walls and end walls of a gondola car according to claim 1, characterized in that: The recognition module also includes a depth grouping module, a plane fitting module and a weld determination module; A depth grouping module is used to group the point cloud data of the dense point cloud model of the vehicle side wall according to the depth information, grouping the point cloud data with the same depth information into one group, and grouping the adjacent point cloud data with a depth information change greater than a preset threshold into another group; A plane fitting module is used to perform plane fitting on the grouped point cloud data. The fitted planes include the steel plate plane, the diagonal brace vertical plane, the diagonal brace wing plane, the diagonal brace wing weld plane, the side wall column plane, the side wall column wing weld plane, and the side wall column wing plane. The weld determination module is used to determine the coordinate information of the weld according to the intersection lines of each plane.

3. The cutting equipment for replacing and repairing the side walls and end walls of a gondola car according to claim 2, characterized in that: The recognition module also includes a path planning module and a posture determination module; The path planning module is used to determine the depth information of each pixel at the weld location based on the weld coordinate information and the dense point cloud model of the vehicle side wall; The posture determination module is used to determine the posture data of the robot at each welding position based on the correction of the robot's reference position posture data.

4. A method for cutting and replacing side walls and end walls of a gondola car, using the cutting device for cutting and replacing side walls and end walls of a gondola car according to any one of claims 1 to 3, characterized in that: The following steps are involved: Data acquisition step: performing laser scanning on the side wall of the open car using a laser scanning device to obtain point cloud data and generate a point cloud image. The point cloud data includes coordinate information and depth information of each pixel point and the position data of the robot during scanning; Recognition step: identify the point cloud image, process the point cloud image according to the point cloud information, reconstruct the three-dimensional model, identify the weld position, and generate the cutting path; Cutting execution steps: sending instructions to the cutting actuator, so that the cutting actuator cuts the side wall of the open car according to the cutting path.

5. A cutting method for replacing and repairing the side walls and end walls of a gondola car according to claim 4, characterized in that: The identification step comprises the following steps: Edge extraction step: The point cloud image collected by the laser scanning device is marked as image A. Edge extraction is performed based on the depth information of image A to obtain the suspected weld position. The suspected weld position and nearby pixels are extracted to create a mask and generate a mask image. Data processing steps: perform an AND operation on image A and the inverse code of the mask image and perform downsampling to obtain image B, perform an AND operation on image A and the mask image to obtain image C, perform an OR operation on image C and image B to obtain image D, and use image D as the modeling image.

6. A cutting method for replacing and repairing the side walls and end walls of a gondola car according to claim 5, characterized in that: The identifying step further comprises the following steps: Image grouping step: identifying image features in the modeling images and classifying the modeling images into three categories based on the image features, wherein the image features include images containing diagonal braces, images containing vertical columns, and images without diagonal braces or vertical columns. Each category of modeling images is grouped based on the pose data so that the image features and pose data of each group of modeling images are the same; Group modeling step: Based on each frame modeling image and corresponding pose data in each group, a dense point cloud model of each group of fusion reconstruction is established; Group loop detection step: Compare the depth information of each frame modeling image in a group with the reconstructed dense point cloud model to determine whether the depth information difference between each frame modeling image and the dense point cloud model is within the set threshold. If not, the depth information and pose data of each frame modeling image in the group are corrected according to the dense geometric consistency and photometric consistency; In the fusion modeling step, the image features of each group and the reconstructed dense point cloud model are fused and reconstructed to obtain the overall dense point cloud model; Overall loop detection step: Compare the depth information of each group's dense point cloud model with the overall dense point cloud model to determine whether the depth information is within the set threshold. If not, correct the reconstructed point cloud and position data of each group based on dense geometric consistency and photometric consistency; In the correction step, based on the loop detection results, the final dense point cloud model of the vehicle side wall and the correction robot's reference position and pose data are established.

7. A cutting method for replacing and repairing the side walls and end walls of a gondola car according to claim 6, characterized in that: The identifying step further comprises the following steps: Depth grouping step: group the point cloud data of the overall dense point cloud model according to the depth information, group the point cloud data with the same depth information into one group, and group the adjacent point cloud data with a depth information change greater than a preset threshold into another group; Plane fitting step: performing plane fitting on the grouped point cloud data. The fitted planes include the steel plate plane, the diagonal brace vertical plane, the diagonal brace wing plane, the diagonal brace wing weld plane, the side wall column plane, the side wall column wing weld plane, and the side wall column wing plane. Weld seam determination steps: Determine the coordinate information of the weld seam based on the intersection lines of each plane.

8. A cutting method for replacing and repairing the side walls and end walls of a gondola car according to claim 7, characterized in that: The identifying step further comprises the following steps: In the path planning step, the depth information of each pixel at the weld position is determined based on the weld coordinate information and the dense point cloud model of the vehicle side wall. The posture determination step determines the posture data of the robot at each weld position based on the corrected robot reference position posture data.

Images