Cladding welding robot system based on laser vision sensing

Abstract

The invention relates to the field of industrial robots, and provides a cladding welding robot system based on laser vision sensing, comprising: a robot body for supporting and driving a welding device to move in space; the cladding welding device is arranged at the tail end of the robot body and is used for cladding a powder material to the surface of a target workpiece; the laser vision sensing system comprises at least three laser stripe projectors, at least one industrial camera and a band-pass filtering assembly; a sequential control module; the time sequence control module is used for projecting laser to the plurality of laser stripe projectors in turn according to a set time sequence; the image processing module is used for carrying out fusion processing on the multi-angle image frames at different time points; and the path generation and control module is used for planning a cladding path according to the welding seam boundary contour coordinates and outputting a control instruction to adjust the movement track and the process parameters of the welding device.

Description

Technical Field

[0001] The present invention belongs to the field of industrial robots, and in particular relates to a cladding welding robot system based on laser vision sensing. Background Art

[0002] As the demand for welding quality and efficiency continues to increase in industries such as high-end equipment manufacturing, ship repair, and metal remanufacturing, laser cladding technology, with its advantages of low dilution, high bond strength of the cladding layer, and excellent controllability, has become widely used for surface strengthening and defect repair in critical areas. By integrating a laser cladding device with a multi-axis industrial robot, the cladding welding robot system achieves high-precision material deposition on the surfaces of complex spatial structures, effectively improving the level of production automation and component serviceability.

[0003] Existing laser cladding robot systems typically rely on laser vision sensors to obtain weld contour information. They employ methods such as structured light, laser dot matrix, or line scanning to identify the workpiece boundary. Path-fitting algorithms then construct the cladding trajectory and drive the robot to perform the welding action. Common methods include single-view laser projection, fixed filter recognition, and rule-based template fitting. Some systems also extract weld contours through algorithms such as image enhancement, threshold segmentation, and edge detection.

[0004] However, existing technologies are susceptible to the high-intensity arc light, metal spatter, and smoke obstruction during welding in highly interfering environments, resulting in poor visual recognition stability, discontinuous fringe profiles, or missing information. Furthermore, most methods rely on single-frame images or fixed viewing angles to capture weld information, lacking the ability to integrate redundant information. This can lead to recognition failures in areas with large weld curvature or uneven lighting, impacting the accuracy of cladding path planning and the final weld quality. Summary of the Invention

[0005] In order to solve the problems in the prior art, the present invention provides a cladding welding robot system based on laser vision sensing, comprising:

[0006] A robot body, used to support and drive the welding device to move in space;

[0007] a cladding welding device, disposed at the end of the robot body, for cladding powder material onto the surface of a target workpiece;

[0008] A laser vision sensing system, comprising:

[0009] At least three laser stripe projectors, installed at different positions, for projecting structured light stripes onto the weld area at different angles;

[0010] at least one industrial camera, configured to capture image data containing the laser stripes;

[0011] A bandpass filter component is provided at the front end of the industrial camera, and is used to allow only light with a wavelength corresponding to the laser stripe to enter the industrial camera, thereby shielding the non-structured light interference caused by arc light and highlight splash;

[0012] a timing control module for sequentially projecting laser light from the plurality of laser stripe projectors according to a set timing sequence, and synchronously controlling the industrial camera to capture corresponding image frames in each laser projection cycle;

[0013] An image processing module is used to fuse image frames from multiple angles and at different time points and perform the following operations:

[0014] Constructing a weld response density tensor based on multiple laser stripe projection directions and multiple time-series image frames, and identifying a stable weld stripe region according to the response density tensor, thereby extracting a complete weld boundary contour;

[0015] A path generation and control module is used to plan the cladding path according to the weld boundary contour coordinates and output control instructions to adjust the motion trajectory and process parameters of the welding device.

[0016] Furthermore, the angles between the three laser stripe projectors are the same.

[0017] Furthermore, the image acquisition rate of the industrial camera is not less than 100 frames per second.

[0018] Furthermore, the optical filter assembly adopts an interference bandpass filter with a central wavelength of 660 nanometers and a bandwidth range of ±10 nanometers.

[0019] Furthermore, the timing control module triggers the working state of each laser stripe projector in sequence according to a set order, so that it emits laser signals in turn at fixed time intervals.

[0020] Furthermore, a weld response density tensor is constructed based on multiple laser stripe projection directions and multiple time-series image frames, and a stable weld stripe region is identified according to the response density tensor, thereby extracting a complete weld boundary contour. Specifically, the process includes:

[0021] Multiple laser stripe projectors are activated in sequence through the timing control module. The laser projector in each direction triggers N projections to emit light, which in turn triggers the industrial camera to collect images.

[0022] Perform normalization, Gaussian filtering, and contrast enhancement on each frame of image;

[0023] At each pixel coordinate, count the response density under each laser direction;

[0024] Perform weighted superposition of the response densities in different directions to obtain a fused response density map;

[0025] Constructing a stripe response mask based on the fused response density map, wherein stripes whose response density map is greater than a density threshold of a preset value are retained;

[0026] According to the retained stripe contour pixel set;

[0027] Each contour pixel point is back-projected into the three-dimensional weld space reference coordinate system through the structured light triangulation model according to the laser direction during acquisition to obtain the weld contour point set.

[0028] Furthermore, the response density is calculated as follows:

[0029]

[0030] in:

[0031] R(x,y,d) represents the response density of pixel (x,y) in direction d;

[0032] N represents the number of frames collected in this direction;

[0033] (x,y) represents the normalized grayscale value of the pixel (x,y) in the k-th frame in the d-th direction;

[0034] τ represents the grayscale threshold of the fringe response;

[0035] δ(condition) is an indicator function, which takes the value 1 when the condition is true and 0 otherwise;

[0036] R(x,y,d) represents the structured light response frequency of a certain pixel in direction d.

[0037] Furthermore, the response density map is calculated as follows:

[0038]

[0039] in:

[0040] Rtotal(x,y) represents the multi-view fusion response value of pixel (x,y);

[0041] ω1, ω2, and ω3 are weighting coefficients for each direction.

[0042] Furthermore, the stripe contour is extracted using the Canny operator.

[0043] Furthermore, back-projecting the structured light triangulation model into a three-dimensional weld space reference coordinate system includes:

[0044]

[0045] in:

[0046] [Xi, Yi, Zi] represents the three-dimensional coordinates of the i-th weld boundary point;

[0047] Rd is the rotation matrix of the corresponding direction (camera posture);

[0048] Td is the translation vector in the corresponding direction (camera position);

[0049] K is the camera intrinsic parameter matrix;

[0050] s is the scale factor (obtained from laser triangulation calibration);

[0051] xi, yi are the image pixel coordinates.

[0052] This invention achieves stable perception of weld contours by introducing multi-angle laser fringe projection and timing control mechanisms, combined with a bandpass filter and industrial camera to capture high-purity structured light images. During image processing, a response density tensor is constructed, integrating fringe response information from different perspectives and timings to extract weld boundary features that remain stable even in interference environments, significantly improving the system's anti-interference capabilities and recognition robustness.

[0053] The redundant completion mechanism of multi-angle images effectively avoids contour breaks at weld boundaries caused by occlusion, reflection, or localized stripe loss, enhancing the continuity and integrity of the cladding path. The output weld spatial contour coordinates can be directly used in the path generation and motion control modules, enabling high-precision cladding welding trajectory planning and improving the automation level and form quality stability of complex curved surface welding operations. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0055] Figure 1 Is a system diagram of the method of the present invention;

[0056] Figure 2 It is an image processing flow chart. DETAILED DESCRIPTION

[0057] Below, the invention is preferably described with reference to the accompanying drawings and specific embodiments.

[0058] This embodiment solves the above problem through the following steps:

[0059] In one embodiment, reference Figure 1 The present invention belongs to the field of industrial robots and provides a cladding welding robot system based on laser vision sensing, which aims to solve the problem of unstable weld identification caused by strong arc light, metal splashing and smoke obstruction in the traditional cladding welding process. The system consists of a robot body, a cladding welding device and a laser vision sensing system. The laser vision sensing system integrates laser stripe projectors, bandpass filter components and industrial cameras arranged in multiple directions. The timing control module realizes the sequential projection of lasers and the synchronous acquisition of images. The image processing module is used to perform differential superposition and redundancy completion on multiple frames of images to accurately extract the weld boundary contour. Finally, the path generation and control module realizes the linkage control of welding trajectory planning and cladding parameters, thereby effectively improving the system's weld detection accuracy and cladding stability in high-interference environments. It is suitable for automated precision repair and strengthening operations in complex welding scenarios such as hull structures and heavy-loaded workpieces.

[0060] The system includes:

[0061] A robot body, used to support and drive the welding device to move in space;

[0062] a cladding welding device, disposed at the end of the robot body, for cladding powder material onto the surface of a target workpiece;

[0063] A laser vision sensing system, comprising:

[0064] At least three laser stripe projectors, installed at different positions, for projecting structured light stripes onto the weld area at different angles;

[0065] at least one industrial camera, configured to capture image data containing the laser stripes;

[0066] A bandpass filter component is provided at the front end of the industrial camera, and is used to allow only light with a wavelength corresponding to the laser stripe to enter the industrial camera, thereby shielding the non-structured light interference caused by arc light and highlight splash;

[0067] a timing control module for sequentially projecting laser light from the plurality of laser stripe projectors according to a set timing sequence, and synchronously controlling the industrial camera to capture corresponding image frames in each laser projection cycle;

[0068] An image processing module is used to fuse image frames from multiple angles and at different time points and perform the following operations:

[0069] Constructing a weld response density tensor based on multiple laser stripe projection directions and multiple time-series image frames, and identifying a stable weld stripe region according to the response density tensor, thereby extracting a complete weld boundary contour;

[0070] A path generation and control module is used to plan the cladding path according to the weld contour coordinates and output control instructions to adjust the motion trajectory and process parameters of the welding device.

[0071] The following is a detailed description of each part of the above system.

[0072] A robot body is used to support and drive the welding device to move in space.

[0073] The robot body module supports and drives the cladding welding device along a predetermined path in three-dimensional space. By spatially positioning and adjusting the cladding welding device, it achieves coverage, tracking, and precise welding control of the weld area on the workpiece surface, ensuring uniform deposition of the cladding material at the set location, thereby forming a structurally complete and stable metallurgical bond layer. The robot body module is the motion execution core of the entire system, and its operating accuracy directly affects the accuracy of the welding path and the consistency of the cladding quality.

[0074] The robot body module may include an industrial robot arm with multiple degrees of freedom. This industrial robot arm, through a controller, achieves coordinated control of multiple rotational joints or linear axes, driving the cladding welding device mounted at its end to adjust its position and move its trajectory in space. Preferably, the industrial robot arm has a six-degree-of-freedom structure, capable of completing trajectory tracking welding tasks for complex curved workpieces and meeting the motion requirements of non-planar weld paths. A communication interface is provided between the robot body and the welding path generation module for receiving trajectory planning results in real time and converting them into joint motion commands.

[0075] In a specific example, the robot body adopts a six-axis industrial robot arm model ABBIRB4600, which completes the spatial path movement based on the weld contour under the instruction of the controller, driving the laser cladding welding head installed at the end to perform high-precision positioning movement along the longitudinal weld area of the hull steel plate, realizing a continuous and constant speed powder deposition process.

[0076] A cladding welding device is provided at the end of the robot body and is used for cladding powder material onto the surface of a target workpiece.

[0077] The cladding welding module, located at the end of the robot body, deposits powdered material onto the target workpiece surface. Energy beam heating creates a metallurgical bond between the cladding material and the substrate surface, enhancing the workpiece's functionality, repairing its structure, or improving its wear and corrosion resistance. The cladding welding module coordinates with the robot body to achieve cladding coverage of complex weld paths, ensuring continuous material deposition and consistent weld bead quality.

[0078] The cladding welding device module may include a laser cladding head, a powder feeding mechanism, a gas shielding assembly, and a cooling mechanism. The laser cladding head is used to output a high-energy laser beam. The powder feeding mechanism is used to feed metal powder into the molten pool via a coaxial or off-axis path. The gas shielding assembly is used to protect the cladding area from oxidation. The cooling mechanism is used to maintain a stable operating temperature of the equipment. Preferably, the laser cladding head adopts a coaxial powder feeding structure to improve cladding forming accuracy and powder utilization. The laser power is adjustable to adapt to different materials and working conditions.

[0079] In a specific example, continuing the situation where the aforementioned robot body adopts ABBIRB4600, the cladding welding device includes a coaxial laser cladding head connected to a fiber laser with a power of 2000W. Metal powder is sprayed into the focal area of the laser beam at a rate of 8g / min through a coaxial nozzle controlled by a pneumatic powder feeder, forming a continuous cladding layer with a width of about 2.5mm and a height of about 0.8mm on the longitudinal weld path overlapping the surface of the hull steel plate.

[0080] A laser vision sensing system, comprising:

[0081] At least three laser stripe projectors, installed at different positions, for projecting structured light stripes onto the weld area at different angles;

[0082] at least one industrial camera, configured to capture image data containing the laser stripes;

[0083] A bandpass filter component is provided at the front end of the industrial camera, and is used to allow only light with a wavelength corresponding to the laser stripes to enter the industrial camera, thereby shielding non-structured light interference caused by arc light and highlight splashes.

[0084] The laser vision sensing system is used to detect the geometric features of the weld area in real time during the cladding welding process, improving the system's weld recognition accuracy and anti-interference capabilities under complex working conditions. The system uses structured light projection and image acquisition to achieve high-precision acquisition of information such as weld boundaries, residual height, and gap width. To achieve this, the laser vision sensing system includes at least three laser stripe projectors, at least one industrial camera, and a bandpass filter. These three components work together to create a weld vision recognition module with multiple viewing angles, strong anti-interference capabilities, and high accuracy.

[0085] The laser stripe projector module is used to project structured light stripes onto the weld surface, forming a light band pattern on the workpiece. The laser stripes, after being distorted by the workpiece geometry, are captured by an industrial camera for subsequent weld profile analysis. Because welds often have irregular morphologies, complex curvatures, and are subject to obstruction, it is difficult to obtain complete information using only a single projection direction. Therefore, multiple projectors are deployed to provide spatial redundancy. This module, based on the principle of laser triangulation, uses fringe distortion information to infer surface topography.

[0086] At least three laser stripe projectors are respectively arranged in front of, above, and to the side of the welding device. The projection angles form different angles with the normal of the workpiece surface to cover multiple directions of the weld area. Each laser projector emits laser light of the same wavelength, but can be started sequentially through a timing control module to avoid interference overlap. The laser is preferably a red light semiconductor laser, which has the characteristics of compact structure, fast response, and low power consumption. The width of the projected light band is between 1 and 2 mm, which is suitable for high-precision measurement scenarios. Preferably, the angles between the three laser stripe projectors are the same.

[0087] By setting up a multi-angle laser stripe projector, the laser stripes can be effectively projected at at least one viewing angle even in the presence of spatter, obstruction or surface reflection interference, thereby improving the system's ability to fully identify welds in harsh environments and enhancing the system's spatial adaptability to complex curved welds.

[0088] The industrial camera module is used to capture image information of the weld area after structured light illumination, converting the fringe deformation image into a digital image for subsequent processing. This module aims to provide high-resolution, high-frame-rate image acquisition capabilities, ensuring stable acquisition of target information during the dynamic welding process. It uses visual measurement to extract 3D contour information by observing the contours of the image formed by the deformation of the structured light on the workpiece surface.

[0089] The industrial camera is positioned at a triangulated angle with the laser fringe projection direction to achieve optimal fringe distortion. Preferably, the industrial camera utilizes a black and white CCD sensor with low exposure time and high signal-to-noise ratio, an image acquisition rate of no less than 100 frames per second, and communicates with the image processing module via a gigabit Ethernet port. The industrial camera is fixed near the end of the robot to ensure synchronous movement with the welding trajectory.

[0090] By using a high-speed industrial camera, stable imaging of rapidly changing welding scenes is achieved, avoiding image blur caused by workpiece vibration, changing working conditions, or robot path errors. Furthermore, the high-sensitivity imaging chip, combined with subsequent image enhancement algorithms, ensures the contrast and edge clarity of fringe images, helping to improve the accuracy of weld boundary extraction.

[0091] The bandpass filter component is installed at the front end of the industrial camera. Its purpose is to use the principle of spectrally selective filtering to allow only light with the same wavelength as the laser stripe to enter the camera imaging system, thereby effectively shielding the interference of arc light, splashing fire and background scattered light on image acquisition.

[0092] The bandpass filter assembly utilizes an interference bandpass filter with a central wavelength of 660 nanometers and a bandwidth of ±10 nanometers, compatible with the operating wavelength range of laser projectors. The filter is secured to the front end of the industrial camera lens via a threaded ring or bayonet mount, providing wavelength filtering within the optical path. The filter is heat- and UV-resistant, making it suitable for long-term use in welding environments.

[0093] By limiting the wavelength of the bandpass filter, the imaging system effectively eliminates non-structured light signals, retaining a clear stripe pattern outline in the imaging image, significantly improving the image signal-to-noise ratio and the recognizability of the structured light pattern, and ensuring that weld features can still be stably extracted in a strong interference environment.

[0094] The multi-angle redundant projection of the laser stripe projector module, the high-speed imaging capabilities of the industrial camera module, and the wavelength filtering function of the bandpass filter assembly form a laser vision sensing system with strong anti-interference capabilities, high stripe recognition accuracy, and strong weld profile extraction stability. This system is suitable for working environments with high reflectivity, complex curved surfaces, and severe welding process disturbances. It can significantly improve the visual recognition performance and control accuracy of the cladding welding robot, providing a reliable perception foundation for subsequent path planning and quality control.

[0095] In a specific example, the robot body adopts an ABBIRB4600 six-axis industrial robot and the cladding welding device adopts a coaxial powder feeding laser cladding head. The laser vision sensing system is arranged on a fixed bracket 300 mm in front of the cladding welding device, and three groups of laser stripe projectors are respectively installed on the upper left, upper right and front directions of the bracket. The angles between them and the cladding welding path are 45 degrees, 60 degrees and 75 degrees respectively. The emitted laser wavelength is 660 nanometers, the stripe spacing is 1.5 mm, and the power is 20 milliwatts.

[0096] The industrial camera is a Basler ac A1920-155um monochrome industrial camera with an imaging resolution of 1920 × 1200 pixels and a frame rate of 120 frames per second. It uses a C-mount interface to connect to a 50mm fixed-focus industrial lens and communicates with the host controller via the GigE protocol. The industrial camera captures structured light images in short exposure mode (less than 1ms) and works in conjunction with a laser projector in a polling sequence to generate a time-series image sequence.

[0097] The bandpass filter assembly is an interference filter installed at the front end of an industrial camera lens. It features a central wavelength of 660 nanometers, a bandwidth of ±10 nanometers, a transmittance greater than 90%, and a stopband depth better than OD4. The filter assembly is secured to the front end of the lens via a threaded ring and is dust- and heat-resistant.

[0098] A timing control module is used to project laser light to multiple laser stripe projectors in turn according to a set timing, and synchronously control the industrial camera to collect corresponding image frames in each laser projection cycle.

[0099] The timing control module coordinates the operating sequence of multiple laser stripe projectors and synchronizes the timing with the image acquisition process of the industrial camera, ensuring that each frame corresponds to a uniquely oriented structured light stripe. This effectively avoids overlapping interference from multiple laser signals, improving the separability of image data and weld identification accuracy. This module enables precise timing control of laser projection and image acquisition, ensuring the independence of stripe features in the image sequence, which facilitates subsequent image fusion and redundant information completion.

[0100] The timing control module comprises a control motherboard and multiple output channels. After receiving a start signal from the main controller, the control motherboard triggers the operating states of each laser stripe projector in a set sequence, causing them to alternately emit laser signals at fixed intervals. Simultaneously, it sends a synchronous capture signal to the industrial camera, ensuring that each frame contains only the stripe pattern formed by a single laser projection direction. Optional implementations include using a programmable logic device (such as an FPGA), a timer interrupt controller, or a multi-channel counter to implement timing scheduling. Preferably, the timing control module has microsecond-level timing accuracy, enabling it to accommodate high-frame-rate image acquisition requirements.

[0101] In a specific example, continuing the previously described system configuration using a Baslerac A1920-155um industrial camera and three laser stripe projectors, the timing control module uses an STM32F4 series microcontroller chip as its core control. Three IO output ports connect to the three laser projector driver circuits, with each laser stripe projector turning on sequentially at 5-millisecond intervals for a duration of 2 milliseconds. The control module also sends an image trigger signal to the industrial camera via a dedicated synchronization signal interface, ensuring that each image frame strictly corresponds to the laser stripes in a specific direction. With this setup, the system stably captures three stripe image frames with a complete cycle of 15 milliseconds.

[0102] An image processing module is used to fuse image frames from multiple angles and at different time points.

[0103] The image processing module is used to fuse image frames generated by multiple laser stripe projectors at different time points to extract the complete contour information of the weld area and suppress interference stripes. Through redundant angle compensation, image difference analysis and contour geometry fitting, continuous, complete and high-precision weld boundary data is reconstructed, providing a reliable input basis for path planning and cladding control.

[0104] Specifically, refer to Figure 2 , the image processing module performs the following operations:

[0105] A weld response density tensor is constructed based on multiple laser stripe projection directions and multiple time-series image frames, and a stable weld stripe area is identified according to the response density tensor, thereby extracting a complete weld boundary contour.

[0106] Step 1: Image acquisition and data organization

[0107] The system activates multiple laser stripe projectors in sequence through a timing control module, for example, three directions numbered d=1, d=2, and d=3.

[0108] The laser projector in each direction triggers N projections to emit light, which in turn triggers the industrial camera to capture images.

[0109] A total of 3×N image frames are obtained, which are represented as Id(k)(x,y), where:

[0110] I represents an image frame;

[0111] d represents the laser projection direction (1, 2, 3);

[0112] k represents the kth frame in this direction (k ranges from 1 to N);

[0113] (x,y) represents the image pixel coordinates;

[0114] Id(k)(x,y) represents the grayscale value of the pixel (x,y) in the kth frame in the dth direction.

[0115] All images are filtered by a bandpass filter component, allowing only light near the wavelength λ (for example, λ = 660nm) to enter, shielding unstructured light interference from the source.

[0116] Step 2: Image preprocessing

[0117] Each frame image Id(k)(x,y) is processed as follows:

[0118] Normalization processing: Map the grayscale value range of each frame image to the [0,1] interval to eliminate the influence of intensity differences.

[0119] Gaussian filtering: Use a two-dimensional Gaussian filter to smooth the image, reduce random noise, and improve the stability of stripe edges.

[0120] Contrast enhancement: Perform local histogram equalization on each frame to enhance the contrast between the structured light stripes and the background.

[0121] The processed image is recorded as (x,y), providing standardized input for subsequent analysis.

[0122] Step 3: Response density tensor construction

[0123] At each pixel coordinate (x, y), the response density R(x, y, d) under each laser direction d is calculated as follows:

[0124]

[0125] in:

[0126] R(x,y,d) represents the response density of pixel (x,y) in direction d;

[0127] N represents the number of frames collected in this direction;

[0128] (x,y) represents the normalized grayscale value of the pixel (x,y) in the k-th frame in the d-th direction;

[0129] τ represents the grayscale threshold of the fringe response (e.g., τ = 0.4);

[0130] δ(condition) is an indicator function, which is 1 when the condition is true and 0 otherwise.

[0131] R(x,y,d) represents the structured light response frequency of a pixel in direction d, and its value range is [0,1]. A higher value indicates that the pixel stably presents a stripe response.

[0132] Step 4: Fusion density map calculation

[0133] Perform weighted superposition of the response densities in different directions to obtain the fused response density map Rtotal(x,y):

[0134]

[0135] in:

[0136] Rtotal(x,y) represents the multi-view fusion response value of pixel (x,y);

[0137] ω1, ω2, and ω3 are weighting coefficients for each direction (e.g., they can be equal to 1 / 3, or set according to the confidence level of the viewing angle);

[0138] R(x,y,d) represents the response density of the pixel in the dth direction.

[0139] Step 5: Stripe area mask generation

[0140] Construct the stripe response mask M(x,y) according to Rtotal(x,y):

[0141] M(x,y)=

[0142] 1. If Rtotal(x,y)≥θ

[0143] 0, otherwise

[0144] in:

[0145] θ is the response density threshold (e.g., θ = 0.6), which is used to control the fringe retention accuracy;

[0146] M(x,y)=1 indicates that the pixel (x,y) is determined to be a stripe valid area.

[0147] Step 6: Extracting the weld boundary contour

[0148] Perform the following edge processing on the mask map M(x,y):

[0149] Use the Canny operator to extract edge contours;

[0150] Use connected domain analysis to retain the largest closed area and remove isolated noise blocks;

[0151] Refine the boundaries and fit spline curves to improve continuity and smoothness;

[0152] Extract the contour pixel point set C2D={(xi,yi)}.

[0153] Step 7: 3D coordinate back-projection and fusion reconstruction

[0154] Each two-dimensional boundary point (xi, yi) is back-projected into the three-dimensional weld space reference coordinate system through the structured light triangulation model according to the laser direction during acquisition.

[0155] The three-dimensional coordinates are calculated as follows:

[0156]

[0157] in:

[0158] [Xi, Yi, Zi] represents the three-dimensional coordinates of the i-th weld boundary point;

[0159] Rd is the rotation matrix of the corresponding direction (camera posture);

[0160] Td is the translation vector in the corresponding direction (camera position);

[0161] K is the camera intrinsic parameter matrix;

[0162] s is the scale factor (obtained from laser triangulation calibration);

[0163] xi, yi are the image pixel coordinates.

[0164] Fuse the 3D point cloud sets obtained from different directions:

[0165] Identify point pairs in overlapping areas and perform density-weighted averaging;

[0166] Perform Bezier interpolation to complete the fracture boundary;

[0167] Outputs a continuous and smooth set of 3D weld boundary lines.

[0168] The final output weld contour point set C3D={(Xi,Yi,Zi)}.

[0169] In this module, by constructing a weld response density tensor and fusing multi-angle and multi-time series image information, the image fusion processing method can effectively suppress random image interference caused by arc flicker, metal splash, and smoke disturbance, significantly improving the stability and robustness of weld stripe recognition; at the same time, combining multi-view structured light information for spatial redundancy completion can accurately extract the continuous boundary of the weld contour, avoid stripe breakage and contour loss, and ensure the integrity of subsequent path planning and cladding accuracy. It is particularly suitable for weld recognition tasks under complex curved surface structures and high-interference environments.

[0170] A path generation and control module is used to plan the cladding path according to the weld contour coordinates and output control instructions to adjust the motion trajectory and process parameters of the welding device.

[0171] The path generation and control module plans the spatial motion path of the cladding welding device based on the three-dimensional weld contour coordinates output by the image processing module. It also generates corresponding control instructions in real time, guiding the robot body and cladding device to move along the set trajectory with high precision. It also dynamically adjusts process parameters such as laser power, powder feed rate, and welding speed to achieve a stable, continuous cladding operation that matches the actual workpiece topography. By converting sensory data into motion and process control instructions, the path generation and control module establishes a closed-loop control chain of "vision-planning-execution," making it a key module for automating precision welding operations.

[0172] The path generation and control module may include a path fitting unit, a trajectory optimization unit, and a motion instruction output unit, wherein the path fitting unit reconstructs the curve based on the weld boundary point cloud, preferably using a spline interpolation algorithm or a Bezier fitting algorithm to generate a smooth and continuous center path, the trajectory optimization unit combines the workpiece surface normal information with the robot kinematic constraints to generate a six-dimensional pose trajectory and decompose it into executable instructions, and the motion instruction output unit is responsible for converting the trajectory information into a control signal and sending it to the robot control system and the cladding system. Optional implementation schemes include: a path interpreter based on the industrial robot's own control platform, a trajectory converter based on G code, or a multi-channel coordination module based on PLC. Preferably, the path generation process supports the partitioning strategy and speed adaptive adjustment of the weld segment to improve process consistency and local forming quality.

[0173] It should be noted that the path tracking and process parameter control processes performed by the control module are not the technical focus of the present invention. They can be implemented using any existing mature industrial robot control algorithm, motion control card, or laser welding platform integrated control solution, and do not constitute a limitation of the inventive content of the present invention. The module's innovation lies in its ability to perform structured path construction and real-time process coordination based on weld boundary data output from a multi-view visual fusion system, ensuring a closed-loop, high-precision cladding execution process driven by perception.

[0174] The prior art mentioned in the above background technology section and specific embodiments section of the present invention can be regarded as part of the present invention and used to understand the meaning of some technical features or parameters.

Claims

1. A cladding welding robot system based on laser vision sensing, characterized in that: The system comprises: A robot body, used to support and drive the welding device to move in space; a cladding welding device, disposed at the end of the robot body, for cladding powder material onto the surface of a target workpiece; A laser vision sensing system, comprising: At least three laser stripe projectors, installed at different positions, for projecting structured light stripes onto the weld area at different angles; at least one industrial camera, configured to capture image data containing the laser stripes; A bandpass filter component is provided at the front end of the industrial camera, and is used to allow only light with a wavelength corresponding to the laser stripe to enter the industrial camera, thereby shielding the non-structured light interference caused by arc light and highlight splash; a timing control module for sequentially projecting laser light from the plurality of laser stripe projectors according to a set timing sequence, and synchronously controlling the industrial camera to capture corresponding image frames in each laser projection cycle; An image processing module is used to fuse image frames from multiple angles and at different time points and perform the following operations: Constructing a weld response density tensor based on multiple laser stripe projection directions and multiple time-series image frames, and identifying a stable weld stripe region according to the response density tensor, thereby extracting a complete weld boundary contour; A path generation and control module is used to plan the cladding path according to the weld boundary contour coordinates and output control instructions to adjust the motion trajectory and process parameters of the welding device.

2. The cladding welding robot system based on laser vision sensing according to claim 1 is characterized in that: The angles between the three laser stripe projectors are the same.

3. The cladding welding robot system based on laser vision sensing according to claim 1 is characterized in that: The image acquisition rate of the industrial camera is not less than 100 frames per second.

4. The cladding welding robot system based on laser vision sensing according to claim 1 is characterized in that: The optical filter assembly adopts an interference bandpass filter with a central wavelength of 660 nanometers and a bandwidth range of ±10 nanometers.

5. The cladding welding robot system based on laser vision sensing according to claim 1 is characterized in that: The timing control module triggers the working state of each laser stripe projector in turn according to a set sequence, so that it emits laser signals in turn at fixed time intervals.

6. The cladding welding robot system based on laser vision sensing according to claim 1 is characterized in that: A weld response density tensor is constructed based on multiple laser stripe projection directions and multiple time-series image frames, and a stable weld stripe region is identified according to the response density tensor, thereby extracting a complete weld boundary contour. Specifically, the following steps are performed: Multiple laser stripe projectors are activated in sequence through the timing control module. The laser projector in each direction triggers N projections to emit light, which in turn triggers the industrial camera to collect images. Perform normalization, Gaussian filtering, and contrast enhancement on each frame of image; At each pixel coordinate, count its response density under each laser direction; Perform weighted superposition of the response densities in different directions to obtain a fused response density map; Constructing a stripe response mask based on the fused response density map, wherein stripes whose response density map is greater than a density threshold of a preset value are retained; According to the retained stripe contour pixel set; Each contour pixel point is back-projected into the three-dimensional weld space reference coordinate system through the structured light triangulation model according to the laser direction during acquisition to obtain the weld contour point set.

7. The cladding welding robot system based on laser vision sensing according to claim 6 is characterized in that: The response density is calculated as follows: ； in: R(x,y,d) represents the response density of pixel (x,y) in direction d; N represents the number of frames collected in this direction; (x,y) represents the normalized grayscale value of the pixel (x,y) in the k-th frame in the d-th direction; τ represents the grayscale threshold of the fringe response; δ(condition) is an indicator function, which takes the value 1 when the condition is true and 0 otherwise; R(x,y,d) represents the structured light response frequency of a certain pixel in direction d.

8. The cladding welding robot system based on laser vision sensing according to claim 7 is characterized in that: The response density map is calculated as follows: Rtotal(x,y)=ω1×R(x,y,1)+ω2×R(x,y,2)+ω3×R(x,y,3) in: Rtotal(x,y) represents the multi-view fusion response value of pixel (x,y); ω1, ω2, and ω3 are weighting coefficients for each direction.

9. The cladding welding robot system based on laser vision sensing according to claim 6 is characterized in that: The stripe contour is extracted using the Canny operator.

10. The cladding welding robot system based on laser vision sensing according to claim 6, characterized in that: The back-projection of the structured light triangulation model into the 3D weld space reference coordinate system includes: [Xi,Yi,Zi] T =Rd(-1)×K(-1)×s×[xi,yi,1] T -Td; in: [Xi, Yi, Zi] represents the three-dimensional coordinates of the i-th weld boundary point; Rd is the rotation matrix of the corresponding direction (camera posture); Td is the translation vector in the corresponding direction (camera position); K is the camera intrinsic parameter matrix; s is the scale factor (obtained from laser triangulation calibration); xi, yi are the image pixel coordinates.

Images