A welding robot seam tracking control method for pipe laying

Abstract

The application discloses a welding seam tracking control method of a welding robot for pipeline laying, and belongs to the field of welding robots, which comprises the following steps: after collecting a welding seam stripe image, extracting a welding seam feature point coordinate and binding the feature point coordinate with a mileage position, and storing the feature point coordinate into a circular queue; according to the feature point coordinate in the circular queue, fitting a front welding seam curve by using an adaptive variable order spline, and obtaining a fitting curve of a continuous welding seam; correcting a skidding mileage error and a variable speed time delay, obtaining a feedback deviation, adjusting a leading step according to a local curvature of the fitting curve, and taking a difference value between a queue head feature point and a curve leading point as a leading deviation; fusing the feedback deviation and the leading deviation to obtain a total deviation, and combining with an acceleration constraint of each axis to generate a motion control instruction of the welding robot. The application can be applied to bridges, highways and culverts, can effectively inhibit tracking errors caused by track skidding, time-varying time delay and curvature mutation, and can guarantee welding seam quality.

Description

Technical Field

[0001] This application relates to the field of welding robots, and more particularly to a method for tracking and controlling weld seams in a welding robot used for pipeline laying. Background Technology

[0002] Pipeline laying is the construction process of connecting prefabricated pipe sections one by one according to the designed route to form a continuous pipeline. It is widely used in long-distance transportation fields such as oil, natural gas, water supply and chemical industry. Similarly, bridge steel structure welding (such as the assembly of large-span steel box girders and arch rib segments), highway steel beam welding (such as the butt welding of steel box girder segments), and culvert corrugated pipe welding (such as the circumferential splicing of large-diameter corrugated steel plates) also require connecting prefabricated components into a whole through circumferential butt welds. The quality of the welds also determines the load-bearing capacity, fatigue life and sealing performance of the structure.

[0003] During on-site laying and installation, due to the ellipticity of pipe sections, misalignment of bevels, manufacturing tolerances and hoisting deformation of bridge and highway steel beam sections, and unevenness of culvert corrugated plates, coupled with strong interference from arc light, smoke, and reflections in the welding environment, and the potential speed fluctuations or slippage of the welding vehicle when traveling along the pipe ring track, bridge arc track, or culvert ring track, traditional fixed-parameter automatic welding cannot guarantee that the welding torch is always aligned with the center of the weld.

[0004] Furthermore, traditional methods are inaccurate in extracting weld feature points under strong interference environments, and fixed-order spline fitting cannot simultaneously ensure smoothness in gentle areas and shape preservation in abrupt change areas. The feedback deviation caused by track slippage and speed change delay is not considered, and the fixed advance step size is difficult to adapt to curvature changes, such as the local turning points at the connection between the bridge arch rib and web plate and the corrugated plate splice joint of the culvert. The lack of ellipticity / deformation compensation and acceleration constraints affects tracking accuracy and motion stability.

[0005] No effective solutions have yet been proposed to address the problems in the relevant technologies. Summary of the Invention

[0006] To overcome the above problems, this application aims to propose a weld seam tracking control method for welding robots used in pipeline laying. The purpose is to solve the problems of inaccurate feedback deviation caused by failure to consider track slippage and speed change delay, difficulty in adapting to curvature changes with fixed advance step length, lack of ellipticity compensation and acceleration constraints, which affect tracking accuracy and motion stability.

[0007] Therefore, the specific technical solution adopted in this application is as follows: A method for weld seam tracking control of a welding robot used in pipeline laying, the method comprising: S1. Acquire weld stripe images, extract the coordinates of weld feature points and bind them to the mileage location, and store them in a circular queue with layered filtering and capacity dual threshold protection. S2. Based on the coordinates of the feature points in the circular queue, the curve of the preceding weld is fitted using an adaptive variable-order spline, and the curvature boundary constraints of the beginning and end of the circumferential weld are applied to obtain the fitted curve of the continuous weld. S3. Correct the slippage mileage error and gear shift delay, obtain the feedback deviation from the fitted curve, adjust the advance step size according to the curvature, and take the difference between the characteristic point of the front of the formation and the advance point of the curve as the advance deviation. S4. The feedback deviation and the lead deviation are fused to obtain the total deviation. The total deviation is decomposed into motion components of the circumferential, lateral and lifting axes used for pipe ellipticity compensation. The motion control commands of the welding robot are generated by combining the acceleration constraints of each axis.

[0008] Optionally, the data is stored in a circular queue with hierarchical filtering and dual-threshold capacity protection, including: The laser vision sensor of the welding robot is used to acquire laser stripe images of the weld area at a fixed frequency to obtain the initial stripe image; The initial stripe image is converted to grayscale, subjected to multi-constraint adaptive binarization, extracted from the centerline skeleton, and removed from outliers by double thresholds to obtain the coordinates of the weld feature points in the current frame. Obtain the robot's current mileage position, bind the coordinates of the weld feature points to the mileage position, and obtain bound data with spatiotemporal information; The bound data is stored in a circular queue with hierarchical filtering and dual capacity threshold protection according to the first-in-first-out rule.

[0009] Optionally, the coordinates of the weld feature points in the current frame are obtained, including: Convert the initial striped image to a grayscale image to obtain the grayscale striped image; A multi-constraint adaptive threshold binarization process is performed on the grayscale striped image to separate the laser stripes from the background, resulting in a threshold segmentation image. The centerline skeleton of the laser stripes is extracted from the threshold segmentation image and then repaired to obtain a continuous stripe centerline skeleton. Breakpoints on both sides of the bevel are identified on the stripe centerline skeleton. Interference breakpoints are eliminated by using a dual threshold of geometric distance and contour slope. The geometric midpoint of the selected valid breakpoints is used as the coordinates of the weld feature point in the current frame.

[0010] Optionally, the fitted curve of the continuous weld is obtained, including: Read the stored feature point coordinates from the circular queue, sort them by mileage position, and obtain the feature point sequence to be fitted; Analyze the local curvature changes of the feature point sequence, identify smooth sections and sections with abrupt curvature changes, assign appropriate spline orders to different sections, and construct a variable-order adaptive spline model; The feature point sequence is input into the variable-order adaptive spline model to perform fitting calculations and obtain the initial weld curve. Apply curvature continuity boundary constraints at the pipe circumferential joint overlap to the beginning and end of the initial weld curve to correct the joint distortion and obtain a weld fitting curve that meets the curvature continuity requirements.

[0011] Optionally, the expression for the variable-order adaptive spline model is: In the formula, Represents arc length parameter The weld fitting curve at the location; Indicates the total number of feature points; Indicates the first i One feature point; Indicates the first i The coordinates of the feature points; Indicates the basic order; Indicates the coefficient of magnitude of change of order; Represents the hyperbolic tangent function; Indicates the curvature sensitivity coefficient; Indicates the feature point sequence in arc length Local curvature estimation at the location; Indicates the curvature reference threshold; The parameter representing the arc length of the weld curve; Represents arc length parameter The adaptive order at the location.

[0012] Alternatively, the method for obtaining the initial weld curve is as follows: The arc length of each point is obtained by accumulating the spatial distance between adjacent points in the feature point sequence, and the local curvature of each point is analyzed to obtain a point sequence with arc length and local curvature. Substituting the arc length and local curvature at each point into the order calculation formula, and then smoothing the process, we obtain the spline order distribution that varies continuously along the weld. Based on the variable-order adaptive spline model, a variable-order spline basis function matrix is ​​constructed using the arc length of feature points, order distribution, and preset node vectors. Using the feature point coordinates as the fitting target, a set of fitting equations is established to obtain the control points. The optimal coordinates of the control points are obtained by solving the fitted equation system, and then combined with the basis function matrix to generate a continuous initial weld curve.

[0013] Alternatively, the method of taking the difference between the leading feature point and the leading point of the curve as the leading deviation is as follows: The mileage error caused by track slippage is corrected, and the transmission delay caused by changes in welding speed is compensated to obtain the corrected mileage position and delay compensation value. Based on the corrected mileage position and time delay compensation value, the corresponding preset position is obtained from the weld fitting curve, and the difference between the preset position and the actual welding gun position is analyzed to obtain the feedback deviation. Based on the local curvature of the weld at the current location, the advance step size is dynamically adjusted to obtain an adaptive advance step size value. The lead position point is determined on the fitted curve by using the adaptive lead step size value. At the same time, the feature point at the head of the circular queue is taken, and the difference between the coordinates of the two is calculated to obtain the lead deviation.

[0014] Alternatively, the method for calculating the lead deviation by measuring the difference between the two coordinates is as follows: Based on the advance step value adjusted in real time according to the local curvature of the weld, the corresponding advance arc length position is determined on the weld fitting curve, and the coordinates of this position are extracted as the coordinates of the adaptive advance reference point. From the circular queue that has been filtered by layer and protected by dual thresholds for capacity, extract the coordinates of the head feature point that represents the actual weld position of the welding torch and has eliminated false breakpoint interference. By subtracting the coordinates of the adaptive leading reference point from the coordinates of the feature point at the front of the formation, the leading deviation, which reflects the offset trend of the actual weld relative to the fitted curve, is obtained.

[0015] Optionally, motion control commands for the welding robot are generated, including: Based on the preset fuzzy control rules, the feedback deviation and the lead deviation are adaptively weighted and fused to obtain the total deviation; The total deviation is decomposed into the circumferential axis, the transverse axis, and the lifting axis used for pipe ellipticity compensation to obtain each motion component; Preset acceleration constraints are applied to the motion components in the circumferential, lateral, and vertical axes respectively to obtain smooth motion components; The motion components after acceleration constraints are converted into motion control commands for the welding robot and then issued for execution.

[0016] Compared with the prior art, this application has the following beneficial effects: 1. This application uses hierarchical filtering cyclic queue and curvature adaptive variable-order spline fitting to stably extract weld features and accurately fit bends and misalignments under strong interference environments. Combined with variable speed compensation and curvature adaptive lead deviation, it achieves feedforward prediction and feedback fusion control. A dedicated lifting shaft is designed to compensate for the ellipticity of the pipeline. It can also compensate for the local deformation of steel components of bridges and highways and corrugated plates of culverts caused by manufacturing tolerances, hoisting or assembly. This significantly improves the tracking accuracy and motion stability of the welding robot under variable speed, variable curvature and elliptical pipeline conditions. It is also applicable to welding scenarios such as bridges, highways and culverts.

[0017] 2. This application effectively suppresses strong interferences such as arc light, smoke, and reflection by using multi-constraint adaptive binarization, dual-threshold outlier removal, and hierarchical filtering cyclic queue, and stably extracts weld feature points. It uses a curvature-driven adaptive variable-order spline model to fit the weld curve and applies curvature boundary constraints at the beginning and end of the circumferential seam, thereby maintaining smoothness and noise resistance in flat areas and accurately restoring the real contour at misaligned edges or abrupt curvature changes. It can effectively cope with complex interferences and contour changes in the welding of bridge, highway steel beams and culvert corrugated pipes.

[0018] 3. This application improves feedback accuracy by correcting slippage and speed change delay. It utilizes curvature adaptive advance step size and the difference between the front feature point and the curve advance point to achieve feedforward prediction. Then, it fuses the double bias through fuzzy weighted fusion and uses a dedicated lifting axis to compensate for the ellipticity of the pipeline. At the same time, it applies acceleration constraints, thereby significantly improving the tracking accuracy and motion stability of the welding robot under variable speed, variable curvature and elliptical pipeline conditions. When applied to circumferential welding of structures such as bridges, highways and culverts, it can also effectively suppress tracking errors caused by track slippage, time-varying delay and curvature change, and ensure weld quality. Attached Figure Description

[0019] The above-mentioned features, characteristics, and advantages of this application, as well as their implementation methods, will become clearer and more understandable in conjunction with the following description of the embodiments, which are illustrated in detail with reference to the accompanying drawings. Schematic diagrams are shown here: Figure 1 This is a flowchart of the weld seam tracking and control method for a welding robot used in pipeline laying in this application. Detailed Implementation

[0020] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0021] According to embodiments of this application, a weld seam tracking control method for a welding robot used in pipeline laying is provided. Through hierarchical filtering cyclic queues and curvature adaptive variable-order spline fitting, weld seam features are stably extracted and bends and misalignments are accurately fitted under strong interference environments. Combining variable speed compensation and curvature adaptive lead deviation, feedforward prediction and feedback fusion control are achieved. A dedicated lifting axis is used to compensate for pipeline ellipticity, significantly improving the tracking accuracy and motion stability of the welding robot under variable speed, variable curvature, and elliptical pipeline conditions. Figure 1 As shown, the method includes: S1. Acquire weld stripe images, extract the coordinates of weld feature points and bind them to the mileage location, and store them in a circular queue with layered filtering and capacity dual threshold protection.

[0022] Preferably, the data is stored in a circular queue with hierarchical filtering and dual threshold capacity protection, including: The laser vision sensor of the welding robot is used to acquire laser stripe images of the weld area at a fixed frequency to obtain the initial stripe image; The initial stripe image is converted to grayscale, subjected to multi-constraint adaptive binarization, extracted from the centerline skeleton, and removed from outliers by double thresholds to obtain the coordinates of the weld feature points in the current frame. Obtain the robot's current mileage position, bind the coordinates of the weld feature points to the mileage position, and obtain bound data with spatiotemporal information; The bound data is stored in a circular queue with hierarchical filtering and dual capacity threshold protection according to the first-in-first-out rule.

[0023] Preferably, obtaining the coordinates of the weld feature points in the current frame includes: Convert the initial striped image to a grayscale image to obtain the grayscale striped image; A multi-constraint adaptive threshold binarization process is performed on the grayscale striped image to separate the laser stripes from the background, resulting in a threshold segmentation image. The centerline skeleton of the laser stripes is extracted from the threshold segmentation image and then repaired to obtain a continuous stripe centerline skeleton. Breakpoints on both sides of the bevel are identified on the stripe centerline skeleton. Interference breakpoints are eliminated by using a dual threshold of geometric distance and contour slope. The geometric midpoint of the selected valid breakpoints is used as the coordinates of the weld feature point in the current frame.

[0024] It should be explained that layered filtering refers to setting up multiple buffer layers in the circular queue, such as the original layer, smoothing layer, and feature layer. Newly stored bound data first enters the original layer, and is filtered by sliding window mean filtering to remove high-frequency jitter caused by structural vibrations in application scenarios such as bridges, highways, and culverts (such as flexible deformation and wind-induced vibration of long-span bridges, micro-vibrations caused by vehicle traffic on highway bridges, uneven settlement of culvert fill areas, or construction disturbances). The window size can be 3 to 5 frames. Then it is passed to the feature layer for subsequent curve fitting. The filtering weights can be adaptively adjusted according to the signal confidence.

[0025] The circular queue has minimum and maximum capacities: when the welding robot's travel speed decreases (e.g., when approaching a bridge diaphragm, a corner of a highway steel box girder, or a culvert corrugated plate splice joint), the target capacity may exceed the maximum capacity. In this case, the actual capacity is clamped to the maximum capacity, and an early warning is triggered. When the speed increases, the target capacity may fall below the minimum capacity. In this case, the capacity is clamped to the minimum capacity, and the oldest excess data is discarded. Dual threshold protection can prevent queue overflow or information loss, making it particularly suitable for continuous data acquisition scenarios such as long bridge welds (e.g., a whole 30m U-rib), long straight welds of highway steel beams, and large-diameter corrugated pipe circumferential joints in culverts. Multiple constraints include: 1. Gray-scale dynamic range constraint: Based on the gray-scale statistics of the laser stripes in the image, including the mean and standard deviation, an initial threshold is determined; 2. Spatial continuity constraint: The threshold change rate between adjacent frames shall not exceed 20% to avoid inter-frame flicker; 3. Arc light intensity constraint: Real-time detection of arc light interference level. When the arc light is too strong, the threshold of the previous successful frame is used and the size of the binarization open operation kernel is increased. The explanation for the dual threshold of geometric distance and contour slope to remove false breakpoints is as follows: Calculate the Euclidean distance between the breakpoints on both sides of the bevel. If the distance is less than the preset minimum width (e.g., 2mm) or greater than the maximum width (e.g., 10mm), it is identified as a false breakpoint and removed. Calculate the rate of change of the tangent slope of the weld profile at the break point. If the slope change exceeds the set threshold (e.g., the slope difference between three adjacent points is greater than 30°), it is determined to be rust or spatter interference and is rejected. Only breakpoints that simultaneously satisfy both geometric distance and slope thresholds are considered valid breakpoints, and their geometric midpoints are taken as feature points.

[0026] The mileage position is obtained through an incremental encoder mounted on the robot's wheels, with a sampling frequency synchronized with the laser vision sensor. Simultaneously, to prevent track slippage (especially in applications such as bridge inclines or curves, curved tracks on highway steel beams, and circular tracks in culverts), a fusion correction between the encoder and inertial measurement unit (IMU) is introduced: when the encoder jumps abnormally, the mileage is corrected using integrated data from the IMU. For long-span bridges (such as cable-stayed bridges and suspension bridges), long-distance steel highway bridges, and long-distance welding of large-diameter culvert circumferential joints (a single weld can reach tens of meters), this application can effectively suppress mileage accumulation errors and ensure precise binding between weld feature points and mileage positions.

[0027] S2. Based on the coordinates of the feature points in the circular queue, the curve of the preceding weld is fitted using an adaptive variable-order spline, and the curvature boundary constraints of the beginning and end of the circumferential weld are applied to obtain the fitted curve of the continuous weld.

[0028] Preferably, obtaining the fitting curve for the continuous weld seam includes: Read the stored feature point coordinates from the circular queue, sort them by mileage position, and obtain the feature point sequence to be fitted; Analyze the local curvature changes of the feature point sequence, identify smooth sections and sections with abrupt curvature changes, assign appropriate spline orders to different sections, and construct a variable-order adaptive spline model; The feature point sequence is input into the variable-order adaptive spline model to perform fitting calculations and obtain the initial weld curve. Apply curvature continuity boundary constraints at the pipe circumferential joint overlap to the beginning and end of the initial weld curve to correct the joint distortion and obtain a weld fitting curve that meets the curvature continuity requirements.

[0029] Preferably, the expression for the variable-order adaptive spline model is: In the formula, Represents arc length parameter The weld fitting curve at the location; Indicates the total number of feature points; Indicates the first i One feature point; Indicates the first i The coordinates of the feature points; Indicates the basic order; Indicates the coefficient of magnitude of change of order; Represents the hyperbolic tangent function; Indicates the curvature sensitivity coefficient; Indicates the feature point sequence in arc length Local curvature estimation at the location; Indicates the curvature reference threshold; The parameter representing the arc length of the weld curve; Represents arc length parameter The adaptive order at the location.

[0030] Preferably, the method for obtaining the initial weld curve is as follows: The arc length of each point is obtained by accumulating the spatial distance between adjacent points in the feature point sequence, and the local curvature of each point is analyzed to obtain a point sequence with arc length and local curvature. Substituting the arc length and local curvature at each point into the order calculation formula, and then smoothing the process, we obtain the spline order distribution that varies continuously along the weld. Based on the variable-order adaptive spline model, a variable-order spline basis function matrix is ​​constructed using the arc length of feature points, order distribution, and preset node vectors. Using the feature point coordinates as the fitting target, a set of fitting equations is established to obtain the control points. The optimal coordinates of the control points are obtained by solving the fitted equation system, and then combined with the basis function matrix to generate a continuous initial weld curve.

[0031] It should be explained that the scenario is set as follows: Taking the arch rib (or large bridge steel tower column) of a long-span steel arch bridge as an example, the outer diameter is 1200mm, the wall thickness is 24mm, the circumferential butt weld bevel is V-shaped, and the distance between the two edges (effective bevel width) is 3-8mm; similarly, the butt joints of highway steel box girder segments or the circumferential joints of large-diameter corrugated pipes in culverts also have similar size ranges and welding characteristics (such as the outer diameter of highway steel beams or the diameter of culvert corrugated pipes can reach 1000~2000mm, the wall thickness is 20~30mm, and the bevel form is similar); Welding robot: Walking wheels with incremental encoders, equipped with laser vision sensors, sampling frequency 10Hz (1 frame is collected every 100ms). Walking speed: 0.01m / s (10mm / s), with a step length of 1mm per frame; Objective: To track the circumferential joints of bridge arch ribs (or highway steel beam circumferential joints, culvert corrugated pipe circumferential joints), fit the weld curve, guide the welding torch to correct deviation in real time, and adapt to high-altitude wind vibration of bridges, vibration of highway bridge decks due to traffic, disturbance of culvert fill areas, slight deformation of tracks, interference from paint layer reflections, and uneven lighting in the narrow environment inside culverts.

[0032] The laser vision sensor acquires images at a frequency of 10Hz, resulting in a 640×480 pixel laser stripe image. The laser stripe image was converted to a grayscale image with a pixel value range of 0 to 255. The average grayscale value of the laser stripe area was about 180, and the average grayscale value of the background was about 40. The multi-constraint adaptive binarization settings are as follows: 1. Gray-scale dynamic range constraint: Statistical stripe region gray-scale: mean =180, standard deviation =15, initial threshold set to -2 =150; 2. Spatial continuity constraint: The threshold change rate between adjacent frames is ≤20%. The threshold of the previous frame is 150. The allowable range of the threshold in this frame is 120~180, and the final value is 155. 3. Arc light intensity constraint: The arc light interference in this frame is strong. The threshold of 150 from the previous frame is adopted, and the size of the open operation core is increased from 3×3 to 5×5 to remove arc light noise. Skeleton extraction is performed on the binarized image to obtain a discontinuous stripe skeleton; For fracture segments with a length of less than 10 elements, continuous repair is performed to obtain a complete stripe centerline skeleton; Coordinates (pixels) of the breakpoints on both sides of the bevel: The breakpoint on the left is (320, 240); The breakpoint on the right is (328, 242); Euclidean distance Pixel; Calibration ratio: 1 pixel ≈ 0.2mm → actual distance ≈ 1.65mm; Correction: If 1 pixel ≈ 0.5mm, then the actual distance ≈ 4.125mm, which falls within the preset range (2~10mm) and is valid; Profile slope change rate verification: Take three adjacent skeleton points: A(318, 238), B(320, 240), C(322, 241); Calculate the slope of segment AB: (Angle approximately 45°); Calculate the slope of segment BC: (Angle approximately 26.6°) The slope change angle difference: 45°-26.6°=18.4°<30°, which is determined to be a valid breakpoint; Feature point coordinates (geometric midpoint): The coordinates are: X = 324 × 0.5 = 162 mm, Y = 241 × 0.5 = 120.5 mm; Through the fusion correction of the walking wheel encoder and IMU, the current mileage s=1250mm (the circumference of the circumference of the circumference is about 3140mm, and the current position is 1 / 4 of the circumference). Binding data: (s=1250, X=162, Y=120.5); Store in a circular queue (layered filtering + double threshold protection): Queue settings: minimum capacity 30 frames, maximum capacity 100 frames; smoothing layer window size 5 frames; The data in this frame is first stored in the raw layer, and after being filtered by the moving average of 5 frames, it is passed to the feature layer.

[0033] Welding speed 10mm / s, 10Hz sampling, 10 frames acquired per second, current queue capacity 60 frames, within the range of 30~100; Ten consecutive feature points (sorted by mileage) were read from the circular queue, as shown in Table 1: Table 1 Feature point sequence list As shown in Table 1, the basic parameters are set as follows: Basic order =3 indicates a cubic B-spline; order variation amplitude =2 indicates that the order range is 3 to 5; Curvature sensitivity =2; Curvature reference threshold =0.005mm -1 ; Taking point 5 as an example, the adjacent points are 4, 5, and 6: Coordinates of three points: (162, 120.6) (162.3, 120.8) (162, 120.5); Calculate the second derivative to obtain the local curvature. =0.003mm -1 ; Taking point 6 as an example, the adjacent points are 5, 6, and 7: Coordinates of three points: (162.3, 120.8) (162, 120.5) (161.2, 120.1); The local curvature was calculated. =0.008mm -1 ; Adaptive order calculation to obtain point 5 ≈4.67, rounded to the 4th order; Point 6 ≈4.97, rounded to the 5th order; Order distribution: Points 1-4: Low curvature, order 3; 5 points: Order 4; 6-7 o'clock: High curvature, order 5; 8-10 o'clock: Curvature decreases, order is 3.

[0034] Based on the order distribution and the preset node vector, generate the B-spline basis function for each point; Using the coordinates of the feature points as the target, solve the system of equations to obtain the coordinates of 6 control points (example): Q 1 = (160, 120) Q 2 = (161, 120.3) Q 3 = (162.2, 120.7) Q 4 = (161.8, 120.4) Q 5 = (160.6, 119.9) Q 6 = (160, 129.5); Based on the above, an initial weld curve is generated; the obtained curve is a 3rd-order B-spline in the flat section (low curvature) and a 5th-order B-spline in the high curvature section, balancing smoothness and fitting accuracy. At the beginning and end of the pipe circumferential joint, the curvature should be continuous (both are the circumferential curvature of the pipe, k=0.02mm). -1 ); Apply curvature constraints to the beginning and end points of the curve to correct the curve at the beginning and end segments and avoid twisting and bending at the overlap.

[0035] S3. Correct the slippage mileage error and gear shift delay, obtain the feedback deviation from the fitted curve, adjust the advance step size according to the curvature, and take the difference between the feature point of the front of the formation and the advance point of the curve as the advance deviation.

[0036] Preferably, the method of taking the difference between the feature point at the head of the formation and the leading point of the curve as the leading deviation is as follows: The mileage error caused by track slippage is corrected, and the transmission delay caused by changes in welding speed is compensated to obtain the corrected mileage position and delay compensation value. Based on the corrected mileage position and time delay compensation value, the corresponding preset position is obtained from the weld fitting curve, and the difference between the preset position and the actual welding gun position is analyzed to obtain the feedback deviation. Based on the local curvature of the weld at the current location, the advance step size is dynamically adjusted to obtain an adaptive advance step size value. The lead position point is determined on the fitted curve by using the adaptive lead step size value. At the same time, the feature point at the head of the circular queue is taken, and the difference between the coordinates of the two is calculated to obtain the lead deviation.

[0037] Preferably, the method for calculating the lead deviation by measuring the difference between the two coordinates is as follows: Based on the advance step value adjusted in real time according to the local curvature of the weld, the corresponding advance arc length position is determined on the weld fitting curve, and the coordinates of this position are extracted as the coordinates of the adaptive advance reference point. From the circular queue that has been filtered by layer and protected by dual thresholds for capacity, extract the coordinates of the head feature point that represents the actual weld position of the welding torch and has eliminated false breakpoint interference. By subtracting the coordinates of the adaptive leading reference point from the coordinates of the feature point at the front of the formation, the leading deviation, which reflects the offset trend of the actual weld relative to the fitted curve, is obtained.

[0038] It should be explained that the welding robot relies on encoders on its bottom wheels to count the distance traveled. However, slight rust, weld slag protrusions, or paint peeling may occur on the steel structure surface (such as the outer wall of pipes, bridge arch ribs, highway steel beams, culvert corrugated plates, etc.), causing slight slippage when the robot moves. The encoder displays a higher value for the current distance traveled, while the onboard inertial measurement unit collects the robot's actual displacement trajectory in real time. By comparing the encoder mileage reading with the actual trajectory of the inertial unit, the positional deviation caused by slippage can be accurately identified. The mileage values ​​artificially inflated by the encoder are deducted and corrected to eliminate the positioning error caused by the robot's idle movement, so as to obtain the corrected mileage coordinates that fit the robot's actual physical position and avoid the subsequent weld seam tracking reference offset due to inaccurate positioning. The laser vision sensor captures weld seam images, performs on-site arc and spatter noise reduction, extracts feature points, fits spline curves, and transmits data to the robot controller. The entire process has an inherent time delay. At the same time, the robot welding process involves small speed increases, steady speeds, and slight speed decreases, so the time delay is not a fixed value and is a typical time-varying delay characteristic. Based on the robot's current actual walking speed, the delay time of the entire signal processing process is calculated. According to the current walking speed, the physical distance that the robot has actually traveled during the delay time period is calculated. This travel distance is superimposed on the corrected mileage position to complete the time-varying delay compensation, correct the weld seam tracking lag problem caused by signal lag, and ensure that the position information obtained by the controller is completely synchronized with the robot's real-time physical position. Based on the already fitted smooth curve of the weld, and according to the real-time mileage position after compensation and calibration, the standard reference position of the theoretical center of the weld under this coordinate is retrieved. This position is the reference point for the ideal alignment of the welding torch. At the same time, the robot's own position detection module provides real-time feedback on the actual spatial coordinates of the welding torch. The actual position of the welding torch is compared with the reference position of the theoretical center of the weld in the horizontal and vertical dimensions to accurately determine the current offset of the welding torch relative to the center of the weld, which is the real-time feedback deviation. This clearly reflects the actual state of the welding torch's leftward, rightward, or high and low offset. Currently, in some areas of bridge, highway steel beams, or culvert circumferential joints (such as the connection between the bridge arch rib and web, the arc transition section of a highway steel box girder, or the turning area of ​​the corrugated plate splice joint in a culvert), the weld direction is curved, with the local curvature being significantly greater than that of the straight weld section. The local curvature of the weld fitting curve is detected in real time, following the adjustment logic that the larger the curvature, the longer the advance step, and the smaller the curvature, the shorter the advance step: for straight and gentle weld sections, there is no need for excessively long advance predictions, and a smaller step is set to avoid unnecessary posture adjustments; for arc segments with abrupt curvature changes and rapid changes in weld direction, the advance step is automatically increased to lock the weld trend ahead in advance; Based on the adaptively adjusted lead step size, a distance is cut forward on the weld fitting curve to determine the weld lead reference point that the robot will reach in the near future; at the same time, the real weld feature point, which has been filtered and smoothed, is retrieved from the feature point circular queue of the pre-buffer, and the position of the real feature point is compared with the lead reference point to calculate the lead deviation.

[0039] S4. The feedback deviation and the lead deviation are fused to obtain the total deviation. The total deviation is decomposed into motion components of the circumferential, lateral and lifting axes used for pipe ellipticity compensation. The motion control commands of the welding robot are generated by combining the acceleration constraints of each axis.

[0040] Preferably, the motion control commands for the welding robot include: Based on the preset fuzzy control rules, the feedback deviation and the lead deviation are adaptively weighted and fused to obtain the total deviation; The total deviation is decomposed into the circumferential axis, the transverse axis, and the lifting axis used for pipe ellipticity compensation to obtain each motion component; Preset acceleration constraints are applied to the motion components in the circumferential, lateral, and vertical axes respectively to obtain smooth motion components; The motion components after acceleration constraints are converted into motion control commands for the welding robot and then issued for execution.

[0041] It should be explained that the system acquires the feedback deviation reflecting the real-time offset of the welding torch and the leading deviation reflecting the trend of the weld seam ahead. The weights of the two are automatically allocated according to the actual offset on site: when the current offset of the welding torch is large and exceeds the normal allowable error range, the weight of the real-time feedback deviation is increased to prioritize and quickly correct the current welding torch offset and ensure real-time tracking accuracy; when the current offset of the welding torch is small, but the curvature of the weld seam ahead changes drastically and the trend fluctuates significantly, the weight of the leading deviation is increased to focus on predicting future trajectory changes and fine-tuning the welding torch posture in advance. Based on the total deviation after fusion, the rate of change of weld curvature, and the robot's walking speed, the feedforward control gain parameters are autonomously tuned online: when the deviation fluctuation is gentle and the weld trend is stable (such as long-distance welding of straight sections of bridges, highway steel beams, or culverts), the feedforward gain is automatically lowered to avoid excessive control and small jitters; when the deviation changes significantly, the weld makes sharp turns, or the robot moves at different speeds (such as at the joint of bridge diaphragms, at the corner of highway steel box girders, or in the sharp bend area of ​​the corrugated plate circumferential joint of culverts), the feedforward gain is automatically increased to quickly suppress tracking errors caused by external interference (such as high-altitude wind vibration of bridges, vibration of highway bridge decks due to traffic, disturbance of culvert fill areas, or slight deformation of tracks) and mechanical lag, so that the gain does not need to be manually adjusted, making it particularly suitable for the variable construction environment of bridges, highways, and culverts. The total deviation after fusion and adjustment is broken down according to the three-axis motion structure of the welding robot (applicable to steel structure components such as bridges, highways, and culverts). It is divided into a lateral correction component perpendicular to the weld, a height adjustment component perpendicular to the surface of the steel component (adapting to changes in wall thickness, uneven coating thickness, and the concave and convex characteristics of corrugated plates), and a circumferential walking speed component along an arc track (such as the arch rib of a bridge, the arc track of a highway steel beam, or the circular track of a culvert). At the same time, combined with the physical limitations of the robot's mechanical structure, the maximum moving speed and maximum acceleration constraints of each motion axis are set to prevent excessive single correction displacement (for example, to avoid violent shaking when working at height on a bridge or in a confined space inside a culvert). If the correction offset calculated in a single cycle exceeds the maximum allowable adjustment range of the axis system, it is smoothed and evenly distributed, and the large deviation is gradually corrected by splitting it into multiple control cycles. This prevents the welding torch from swinging a large amount of time, avoiding defects such as uneven weld width and undercut during the welding process, especially ensuring the weld quality of bridges, highways, and culverts under dynamic loads or earth pressure. After deviation fusion, gain tuning, axis decomposition, and smoothing constraints, the adjustment values ​​of each axis are converted into standardized motion control commands according to the communication protocol format of the robot controller. Using the sensor's 10Hz control cycle as a unit, position fine-tuning and speed matching commands are continuously sent to each motion axis of the robot, driving the welding torch to make small corrections in the lateral and vertical directions in real time, and simultaneously fine-tuning the circumferential walking speed to ensure high-precision tracking and stable movement in the circumferential welding of bridges, highways, and culverts.

[0042] It should be noted that the calculation formulas and all parameters involved in the calculations in this application have been dimensionless beforehand. The process of dimensionless processing is well known in the industry and will not be described here.

[0043] Although the present application has disclosed the preferred embodiments above, the embodiments are merely examples for the purpose of illustration and are not intended to limit the present application. Those skilled in the art can make some modifications and refinements without departing from the spirit and scope of the present application. The scope of protection claimed by the present application should be determined by the claims.

Claims

1. A method for tracking and controlling weld seams in a welding robot used for pipeline laying, characterized in that, The method includes: S1. After acquiring the weld stripe image, extract the coordinates of the weld feature points, bind them to the mileage location, and then store them in a circular queue with layered filtering and capacity dual threshold protection. S2. Based on the coordinates of the feature points in the circular queue, the curve of the preceding weld is fitted using an adaptive variable-order spline, and the curvature boundary constraints of the beginning and end of the circumferential weld are applied to obtain the fitted curve of the continuous weld. S3. Correct the slippage mileage error and gear shift delay, obtain the feedback deviation from the fitted curve, adjust the advance step size according to the local curvature of the fitted curve, and take the difference between the feature point of the front of the formation and the advance point of the curve as the advance deviation. S4. The feedback deviation and the lead deviation are fused to obtain the total deviation. The total deviation is decomposed into motion components of the circumferential, lateral and lifting axes used for pipe ellipticity compensation. The motion control commands of the welding robot are generated by combining the acceleration constraints of each axis.

2. The welding robot weld seam tracking control method according to claim 1, characterized in that, The circular queue with hierarchical filtering and dual threshold capacity protection includes: The laser vision sensor of the welding robot is used to acquire laser stripe images of the weld area at a fixed frequency to obtain the initial stripe image; The initial stripe image is converted to grayscale, subjected to multi-constraint adaptive binarization, extracted from the centerline skeleton, and removed from outliers by double thresholds to obtain the coordinates of the weld feature points in the current frame. Obtain the robot's current mileage position, bind the coordinates of the weld feature points to the mileage position, and obtain bound data with spatiotemporal information; The bound data is stored in a circular queue with hierarchical filtering and dual capacity threshold protection according to the first-in-first-out rule.

3. The welding robot weld seam tracking control method according to claim 2, characterized in that, Obtaining the coordinates of the weld feature points in the current frame includes: Convert the initial striped image to a grayscale image to obtain the grayscale striped image; A multi-constraint adaptive threshold binarization process is performed on the grayscale striped image to separate the laser stripes from the background, resulting in a threshold segmentation image. The centerline skeleton of the laser stripes is extracted from the threshold segmentation image and then repaired to obtain a continuous stripe centerline skeleton. Breakpoints on both sides of the bevel are identified on the stripe centerline skeleton. Interference breakpoints are eliminated by using a dual threshold of geometric distance and contour slope. The geometric midpoint of the selected valid breakpoints is used as the coordinates of the weld feature point in the current frame.

4. The welding robot weld seam tracking control method according to claim 1, characterized in that, The obtained fitting curve for the continuous weld seam includes: Read the stored feature point coordinates from the circular queue, sort them by mileage position, and obtain the feature point sequence to be fitted; Analyze the local curvature changes of the feature point sequence, identify smooth sections and sections with abrupt curvature changes, assign appropriate spline orders to different sections, and construct a variable-order adaptive spline model; The feature point sequence is input into the variable-order adaptive spline model to perform fitting calculations and obtain the initial weld curve. Apply curvature continuity boundary constraints at the pipe circumferential joint overlap to the beginning and end of the initial weld curve to correct the joint distortion and obtain a weld fitting curve that meets the curvature continuity requirements.

5. The welding robot weld seam tracking control method according to claim 4, characterized in that, The expression for the variable-order adaptive spline model is: In the formula, Represents arc length parameter The weld fitting curve at the location; Indicates the total number of feature points; Indicates the first i One feature point; Indicates the first i The coordinates of the feature points; Indicates the basic order; Indicates the coefficient of magnitude of change of order; Represents the hyperbolic tangent function; Indicates the curvature sensitivity coefficient; Indicates the feature point sequence in arc length Local curvature estimation at the location; Indicates the curvature reference threshold; The parameter representing the arc length of the weld curve; Represents arc length parameter The adaptive order at the location.

6. The welding robot weld seam tracking control method according to claim 4, characterized in that, The method for obtaining the initial weld curve is as follows: The arc length of each point is obtained by accumulating the spatial distance between adjacent points in the feature point sequence, and the local curvature of each point is analyzed to obtain a point sequence with arc length and local curvature. Substituting the arc length and local curvature at each point into the order calculation formula, and then smoothing the process, we obtain the spline order distribution that varies continuously along the weld. Based on the variable-order adaptive spline model, a variable-order spline basis function matrix is ​​constructed using the arc length of feature points, order distribution, and preset node vectors. Using the feature point coordinates as the fitting target, a set of fitting equations is established to obtain the control points. The optimal coordinates of the control points are obtained by solving the fitted equation system, and then combined with the basis function matrix to generate a continuous initial weld curve.

7. The welding robot weld seam tracking control method according to claim 1, characterized in that, The method of taking the difference between the leading feature point of the queue and the leading point of the curve as the leading deviation is as follows: The mileage error caused by track slippage is corrected, and the transmission delay caused by changes in welding speed is compensated to obtain the corrected mileage position and delay compensation value. Based on the corrected mileage position and time delay compensation value, the corresponding preset position is obtained from the weld fitting curve, and the difference between the preset position and the actual welding gun position is analyzed to obtain the feedback deviation. Based on the local curvature of the weld at the current location, the advance step size is dynamically adjusted to obtain an adaptive advance step size value. The lead position point is determined on the fitted curve by using the adaptive lead step size value. At the same time, the feature point at the head of the circular queue is taken, and the difference between the coordinates of the two is calculated to obtain the lead deviation.

8. The welding robot weld seam tracking control method according to claim 7, characterized in that, The method for calculating the lead deviation by the difference between the two coordinates is as follows: Based on the advance step value adjusted in real time according to the local curvature of the weld, the corresponding advance arc length position is determined on the weld fitting curve, and the coordinates of this position are extracted as the coordinates of the adaptive advance reference point. From the circular queue that has been filtered by layer and protected by dual thresholds for capacity, extract the coordinates of the head feature point that represents the actual weld position of the welding torch and has eliminated false breakpoint interference. By subtracting the coordinates of the adaptive leading reference point from the coordinates of the feature point at the front of the formation, the leading deviation, which reflects the offset trend of the actual weld relative to the fitted curve, is obtained.

9. The welding robot weld seam tracking control method according to claim 1, characterized in that, The motion control commands for generating the welding robot include: Based on the preset fuzzy control rules, the feedback deviation and the lead deviation are adaptively weighted and fused to obtain the total deviation; The total deviation is decomposed into the circumferential axis, the transverse axis, and the lifting axis used for pipe ellipticity compensation to obtain each motion component; Preset acceleration constraints are applied to the motion components in the circumferential, lateral, and vertical axes respectively to obtain smooth motion components; The motion components after acceleration constraints are converted into motion control commands for the welding robot and then issued for execution.

Images