Self-adaptive fusion steel pipe welding seam tracking method and system

Abstract

The invention relates to the technical field of welding, in particular to a self-adaptive fusion steel pipe welding seam tracking method and system.The method comprises the following steps that a visual sensing subsystem collects a welding seam original laser image of a steel pipe welding plane in real time, and the transverse deviation of a welding gun of the visual sensing subsystem is calculated; the arc sensing subsystem collects original welding current signals in real time for filtering processing, and the transverse deviation of a welding gun of the arc sensing subsystem is calculated; the decision fusion subsystem performs dynamic multi-working-condition optimal fusion on the welding gun transverse deviation of the visual sensing subsystem and the welding gun transverse deviation of the electric arc sensing subsystem to obtain a final welding gun deviation value; and the executing mechanism drives the welding gun to act according to the final welding gun deviation value, and accurate tracking of the steel pipe welding seam is achieved. According to the method and the system provided by the invention, the information of the visual sensing subsystem and the electric arc sensing subsystem is fused, the steel pipe welding seam is tracked based on visual and electric arc composite sensing, and the precision and the reliability of welding seam tracking are greatly improved.

Description

Technical Field

[0001] This invention relates to the field of welding technology, and in particular to an adaptive fusion method and system for tracking steel pipe welds. Background Technology

[0002] In recent years, my country has consistently ranked first in the world in both annual steel pipe production and consumption. Welding is the primary method used in steel pipe construction across industries such as oil and gas pipelines, offshore platforms, bridge construction, automobiles, and shipbuilding. With societal development, the demand for welders has created a significant contradiction with China's aging population and shortage of skilled workers. Against this backdrop, the need for automated and intelligent welding has become even more urgent. Achieving automated and intelligent welding requires, first and foremost, automatic identification of the workpiece bevel, followed by real-time adjustment of the welding path during the welding process.

[0003] Currently, the most commonly used weld seam tracking sensors in the field of welding automation include arc sensors and laser vision sensors. Arc sensors determine the change in arc length by collecting voltage or current changes during welding, thereby obtaining the relative position of the welding torch and the bevel. They have the advantages of good real-time performance and no need for additional equipment. However, arc sensors depend on the characteristics of the welding power source, and their sensing accuracy and reliability are directly affected by the stability of the welding process. Laser vision sensors determine the welding torch position by acquiring an image of the laser beam and analyzing its shape within the bevel. Laser vision sensors have high accuracy and are not affected by the stability of the welding process, making them convenient to use. However, laser vision sensors are not suitable for narrow and deep weld seams, cannot resist the effects of magnetic blow and wire bending, and will completely fail under strong arc light and fume interference, as well as specular reflection. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide an adaptive fusion method and system for tracking steel pipe welds. It integrates information from a vision sensing subsystem and an arc sensing subsystem, and utilizes the complementarity of the two sensing subsystems to track steel pipe welds based on vision and arc composite sensing, which greatly improves the accuracy and reliability of weld tracking.

[0005] This invention is achieved through the following technical solution: An adaptive fusion method for tracking steel pipe weld seams includes the following steps: S1: During the welding process, the visual sensing subsystem acquires the original laser image of the weld seam on the welding plane of the steel pipe in real time, and captures the weld seam features after processing the original laser image of the weld seam to obtain the processed laser image of the weld seam, obtains the pixel coordinates of key corner points, and calculates the lateral deviation of the welding torch of the visual sensing subsystem and transmits it to the decision fusion subsystem. S2: The arc sensing subsystem acquires the original welding current signal in real time, filters the original welding current signal to obtain the filtered welding current signal, and then calculates the lateral deviation of the welding torch of the arc sensing subsystem based on the filtered welding current signal and transmits it to the decision fusion subsystem. S3: The decision fusion subsystem performs dynamic multi-condition optimal fusion of the welding torch lateral deviation of the vision sensing subsystem and the welding torch lateral deviation of the arc sensing subsystem based on the CART decision tree to obtain the final welding torch deviation value. S4: The actuator drives the welding torch to move according to the final welding torch deviation value, and performs deviation correction to achieve accurate tracking of the steel pipe weld.

[0006] In the optimized version, the visual sensing subsystem in step S1 includes a line laser, an embedded industrial camera, and a visual sensing image processing module. The line laser is tilted and projected onto the welding plane. The axis of the embedded industrial camera is parallel to the axis of the welding torch. The embedded industrial camera is equipped with a narrow-band filter. The line laser and the embedded industrial camera work together to acquire the original laser image of the weld in real time. The visual sensing image processing module is used to process the original laser image of the weld to obtain the processed laser image of the weld.

[0007] The optimized visual sensing image processing module processes the original laser image of the weld seam using methods including preprocessing of the original laser image of the weld seam, selection and cropping of the region of interest, image binarization, and morphological processing.

[0008] Furthermore, the preprocessing method for the original laser image of the weld is as follows: First, based on the camera intrinsic parameter matrix obtained by calibration, the corresponding distortion model is applied for geometric correction to eliminate the geometric distortion caused by lens distortion; spatial domain filtering is used to smooth the original laser image of the weld; nonlinear contrast enhancement is performed on the smoothed laser image of the weld.

[0009] Furthermore, the image binarization method is as follows: a threshold segmentation algorithm based on the image grayscale histogram is adopted. The optimal threshold is determined by maximizing the inter-class variance of the foreground and background pixels. Pixels with grayscale values ​​higher than the optimal threshold are set as foreground, and pixels with grayscale values ​​lower than the optimal threshold are set as background.

[0010] Furthermore, the morphological processing method is as follows: the Zhang-Suen thinning algorithm is applied to extract the skeleton of the binarized laser region, and the target region is gradually converged to the central skeleton through iterative boundary erosion operation.

[0011] Furthermore, in step S1, the lateral deviation of the welding torch in the visual sensing subsystem is obtained based on the processed weld laser image using the Shi-Tomasi method, specifically including the following steps: S11: Based on the processed weld laser image, calculate the average horizontal coordinate of all effective feature points in the calibration image according to equation (1), and establish the average horizontal coordinate of the calibration pixels as the system reference position: (1); in: This represents the average value of the x-coordinate of the calibrated pixels. This represents the total number of feature points detected in the calibration image. Indicates the first in the calibration image The pixel x-coordinates of each feature point; S12: During the welding tracking process, the average value of the current pixel abscissa of all effective feature points is calculated according to equation (2) for each frame of real-time acquired image, and the average value of the current pixel abscissa is used as the center pixel position of the weld represented by the laser stripe in the current frame: (2); in: This represents the average value of the current pixel's x-coordinate. This indicates the number of valid feature points detected in the current frame. Indicates the first in the current frame i The pixel x-coordinates of each feature point; S13: Calculate the pixel offset based on the average horizontal coordinate of the current pixel and the average horizontal coordinate of the calibrated pixel according to equation (3). : (3); S14: Convert the pixel offset into the welding torch lateral deviation of the visual sensing subsystem according to equation (4). : (4); in: This represents the visual sensing scaling factor.

[0012] Furthermore, the method for calculating the lateral deviation of the welding torch in the arc sensing subsystem in step S2 is as follows: S21: Perform sliding mean filtering on the original current signal to obtain the filtered current signal, identify each current pulse in the filtered current signal and extract its peak current; S22: Based on the welding torch oscillation position, calculate the continuous position near the left and right limit positions according to formula (5). Average peak current and peak current difference of each pulse: (5); in: Indicates the continuity near the left extreme position Average peak current of each pulse Indicates the first [number]th [unit] within the left limit window Peak current of each pulse Indicates the continuity near the right extreme position Average peak current of each pulse Indicates the right limit window. Peak current of each pulse This represents the peak current difference; S23: Calculate the lateral deviation of the welding torch in the arc sensing subsystem based on the peak current difference according to equation (6). : (6); in: This represents the proportional coefficient of the arc sensing sensor. This indicates the offset.

[0013] Furthermore, the method for obtaining the final welding torch deviation value in step S3 by performing dynamic multi-condition optimal fusion based on CART decision tree is as follows: S31: Calculate the confidence levels of the visual sensing subsystem and the arc sensing subsystem; S32: Compare the confidence levels of the visual sensing subsystem and the arc sensing subsystem with the set confidence thresholds respectively. If the confidence levels of both the visual sensing subsystem and the arc sensing subsystem are greater than or equal to the set confidence thresholds, then the information from both the visual sensing subsystem and the arc sensing subsystem is valid. The weights of the welding torch lateral deviation of the visual sensing subsystem and the welding torch lateral deviation of the arc sensing subsystem are assigned using a weighted average method. If only one of the confidence levels of the visual sensing subsystem and the arc sensing subsystem is greater than or equal to the set confidence threshold, then the weight of the one greater than or equal to the set confidence threshold is 1, and the weight of the one less than the set confidence threshold is 0. If the confidence levels of both the visual sensing subsystem and the arc sensing subsystem are less than the set confidence thresholds, then both subsystems are determined to be invalid, and historical trajectory prediction is enabled. S33: Based on the determined weights, calculate the final welding torch deviation value according to equation (7): (7); in: This indicates the final welding torch deviation value. Represents the weights of the visual sensing subsystem. This indicates the lateral deviation of the welding torch in the vision sensing subsystem. This indicates the weights of the arc sensing subsystem. This indicates the lateral deviation of the welding torch in the arc sensing subsystem. An adaptive fusion steel pipe weld seam tracking system is used to execute the adaptive fusion steel pipe weld seam tracking method described in any one of the above, which includes a vision sensing subsystem, an arc sensing subsystem, a decision fusion subsystem, and an execution mechanism. The visual sensing subsystem includes a line laser, an embedded industrial camera, and a visual sensing image processing module. The line laser and the embedded industrial camera work together to acquire the original laser image of the weld in real time. The visual sensing image processing module is used to process the original laser image of the weld to obtain the processed laser image of the weld, and to obtain the pixel coordinates of key corner points and calculate the lateral deviation of the welding torch in the visual sensing subsystem. The arc sensing subsystem includes an arc sensor and an arc signal filtering and processing module. The arc sensor is used to acquire welding current signals in real time, and the arc signal filtering and processing module is used to filter the acquired welding current signals in real time to obtain filtered welding current signals, and to calculate the lateral deviation of the welding torch in the arc sensing subsystem based on the filtered welding current signals. The decision fusion subsystem is used to dynamically fuse the welding torch lateral deviation of the vision sensing subsystem and the welding torch lateral deviation of the arc sensing subsystem based on the CART decision tree to obtain the final welding torch deviation value. The actuator is used to drive the welding torch to move according to the final welding torch deviation value, perform deviation correction, and achieve accurate tracking of the steel pipe weld.

[0014] Beneficial effects of the invention: The adaptive fusion method and system for tracking steel pipe welds provided by this invention have the following advantages: When the visual sensing subsystem fails due to specular reflection or the arc sensing subsystem becomes unstable due to arc breakage, the decision fusion subsystem can automatically reduce the weight of the failed sensing subsystem and have it compensated by the arc sensing subsystem. This ensures the continuity of weld tracking and effectively suppresses runaway deviations caused by extreme conditions such as strong arc interference and arc breakage. This provides a theoretical basis for solving the pain points of weak visual anti-reflection and poor arc breakage resistance in welding, and provides a reliable basis for high-precision welding automation. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the three-sensor tracking error distribution of the present invention.

[0016] Figure 2 This is a schematic diagram of the arc sensing tracking results in a single arc sensing experiment.

[0017] Figure 3 This is a schematic diagram of the visual sensing tracking results in a single visual sensing experiment.

[0018] Figure 4 This is a schematic diagram of the composite sensing tracking results of the present invention. Detailed Implementation

[0019] An adaptive fusion method for tracking steel pipe welds includes the following steps: S1: During the welding process, the visual sensing subsystem acquires the original laser image of the weld seam on the welding plane of the steel pipe in real time, and captures the weld seam features after processing the original laser image of the weld seam to obtain the processed laser image of the weld seam, obtains the pixel coordinates of key corner points, and calculates the lateral deviation of the welding torch of the visual sensing subsystem and transmits it to the decision fusion subsystem. Specifically, the vision sensing subsystem includes a line laser, an embedded industrial camera, and a vision sensing image processing module. The line laser is tilted and projected onto the welding plane. The axis of the embedded industrial camera is parallel to the axis of the welding torch. The embedded industrial camera is equipped with a narrow-band filter. The line laser and the embedded industrial camera work together to acquire the original laser image of the weld in real time. The vision sensing image processing module is used to process the original laser image of the weld to obtain the processed laser image of the weld.

[0020] The optimized line laser can be tilted at an angle of 30° to 60° onto the welding plane, while the axis of the embedded industrial camera is parallel to the axis of the welding torch. The embedded industrial camera is equipped with a narrow-band filter, which can maximize the capture of weld features and avoid optical interference.

[0021] The optimized visual sensing image processing module processes the original laser image of the weld seam using methods including preprocessing of the original laser image of the weld seam, selection and cropping of the region of interest, image binarization, and morphological processing.

[0022] Preprocessing of the original laser image of the weld aims to improve its quality, including three key steps: distortion correction, image smoothing, and contrast enhancement.

[0023] Specifically, the preprocessing method for the original laser image of the weld is as follows: First, distortion correction is performed: Based on the camera intrinsic parameter matrix obtained from calibration, the corresponding distortion model is applied to perform geometric correction to eliminate geometric distortion caused by lens distortion. Then, spatial domain filtering is used to smooth the original laser image of the weld. This process is mainly aimed at the problems of Gaussian noise and salt-and-pepper noise introduced by environmental factors (such as electromagnetic interference, uneven lighting, dust, etc.) that the original laser image is susceptible to. Spatial domain filtering is used to smooth the image. Specifically, a Gaussian low-pass filter can be selected to achieve a good balance between effectively suppressing noise and maintaining edge details.

[0024] Finally, nonlinear contrast enhancement is performed on the smoothed weld laser image: this can enhance the distinction between the laser stripe core area (usually corresponding to the medium and low grayscale range) and the background.

[0025] Given the specific mounting positions of embedded industrial cameras, lasers, and workpieces, laser stripes are typically located in the central region of the embedded industrial camera's imaging surface. By extracting the region of interest (ROI) containing the stripes through region selection and image cropping, noise interference from non-target areas can be effectively suppressed, and the speed of subsequent image processing can be significantly improved.

[0026] Furthermore, the image binarization method is as follows: a threshold segmentation algorithm based on the image grayscale histogram is adopted. The optimal threshold is determined by maximizing the inter-class variance of the foreground and background pixels. Pixels with grayscale values ​​higher than the optimal threshold are set as foreground, and pixels with grayscale values ​​lower than the optimal threshold are set as background.

[0027] Specifically, pixels with grayscale values ​​higher than a threshold can be set as foreground (255, white), and pixels with grayscale values ​​lower than the threshold can be set as background (0, black). Using a threshold segmentation algorithm based on the image's grayscale histogram can effectively avoid the subjective errors and instability introduced by manual threshold setting.

[0028] Furthermore, the morphological processing method is as follows: the Zhang-Suen thinning algorithm is applied to extract the skeleton of the binarized laser region, and the target region is gradually converged to the central skeleton through iterative boundary erosion operation.

[0029] Furthermore, in step S1, the lateral deviation of the welding torch in the visual sensing subsystem is obtained based on the processed weld laser image using the Shi-Tomasi method, specifically including the following steps: S11: Based on the processed weld laser image, calculate the average horizontal coordinate of all effective feature points in the calibration image according to equation (1), and establish the average horizontal coordinate of the calibration pixels as the system reference position: (1); in: This represents the average value of the x-coordinate of the calibrated pixels. This represents the total number of feature points detected in the calibration image. Indicates the first in the calibration image The pixel x-coordinates of each feature point; S12: During the welding tracking process, the average value of the current pixel abscissa of all effective feature points is calculated according to equation (2) for each frame of real-time acquired image, and the average value of the current pixel abscissa is used as the center pixel position of the weld represented by the laser stripe in the current frame: (2); in: This represents the average value of the current pixel's x-coordinate. This indicates the number of valid feature points detected in the current frame. Indicates the first in the current frame iThe pixel x-coordinates of each feature point; S13: Calculate the pixel offset based on the average horizontal coordinate of the current pixel and the average horizontal coordinate of the calibrated pixel according to equation (3). : (3); S14: Convert the pixel offset into the welding torch lateral deviation of the visual sensing subsystem according to equation (4). : (4); in: This represents the visual sensing scaling factor.

[0030] By using the above method to obtain the lateral deviation of the welding torch in the vision sensing subsystem, the data is relatively accurate, which is beneficial for accurate tracking of the weld seam in the later stage.

[0031] S2: The arc sensing subsystem acquires the original welding current signal in real time, filters the original welding current signal to obtain the filtered welding current signal, and then calculates the lateral deviation of the welding torch of the arc sensing subsystem based on the filtered welding current signal and transmits it to the decision fusion subsystem. The core of the arc sensing subsystem consists of an arc sensor integrated with the main control board. The arc sensor can be a Hall effect sensor.

[0032] The raw current signal acquired by the arc sensor during welding contains not only key characteristics of the welding state but also various noise components. Therefore, signal filtering is crucial for effectively suppressing noise and improving the signal-to-noise ratio. Mean filtering, as a low-pass filtering method, is effective in suppressing high-frequency noise. Therefore, mean filtering can be used for filtering. This method calculates the continuous signal... Mean filtering is the arithmetic mean of several sampling points, which smooths out random fluctuations in the signal, thereby improving its stability and signal-to-noise ratio. Mean filtering is particularly suitable for situations containing random interference signals. The choice of value directly affects the filtering performance. Larger values ​​result in higher smoothness but lower sensitivity. When the value is small, the smoothness is low but the sensitivity is high.

[0033] Specifically, the method for calculating the lateral deviation of the welding torch in the arc sensing subsystem is as follows: S21: Perform sliding mean filtering on the original current signal to obtain the filtered current signal, identify each current pulse in the filtered current signal and extract its peak current; S22: Based on the welding torch oscillation position, calculate the continuous position near the left and right limit positions according to formula (5). Average peak current and peak current difference of each pulse: (5); in: Indicates the continuity near the left extreme position Average peak current of each pulse Indicates the first [number]th [unit] within the left limit window Peak current of each pulse Indicates the continuity near the right extreme position Average peak current of each pulse Indicates the right limit window. Peak current of each pulse This represents the peak current difference; S23: Calculate the lateral deviation of the welding torch in the arc sensing subsystem based on the peak current difference according to equation (6). : (6); in: This represents the proportional coefficient of the arc sensing, reflecting the change in lateral deviation corresponding to every 1 Hz change in pulse frequency difference, and is preferably 0.107. This represents the offset, reflecting the system's zero-point deviation or initial calibration value, and is preferably -1.588.

[0034] By using the above method to obtain the lateral deviation of the welding torch in the arc sensing subsystem, the data is relatively accurate, which is further beneficial for accurate tracking of the weld seam in the later stage.

[0035] S3: The decision fusion subsystem performs dynamic multi-condition optimal fusion of the welding torch lateral deviation of the vision sensing subsystem and the welding torch lateral deviation of the arc sensing subsystem based on the CART decision tree to obtain the final welding torch deviation value. CART decision trees are inherently compatible with vision and arc composite sensing. The lateral deviation of laser stripes in vision sensing is a high-dimensional spatial feature, while welding current is a low-dimensional temporal feature. CART decision trees use minimum Gini impurity to find split points and give both types of features a unified score. High Gini features are first screened out. If the arc signal is lost, visual weight is added, and vice versa. The generated rules are the optimal fusion strategy under multiple working conditions, which directly explains the relationship between sensor confidence and weld deviation.

[0036] Specifically, the mapping relationship between sensor failure modes and fusion weights can be learned through training data. The core components of the decision fusion subsystem include the root node (data extracted from the input visual and arc sensing features), internal nodes (sensor effectiveness detection results and dynamic weight ratio allocation results), leaf nodes (calculation and output of welding torch deviation values), and decision rules (screening conditions and allocation conditions).

[0037] The decision fusion subsystem first filters the data and then calculates two confidence levels: visual confidence level CW: 10 to 15 feature points are best, fewer will cause arc interference, more will cause splash noise; electric arc confidence level CA: the closer the left and right limit frequency difference is to 0, the more stable it is, exceeding the limit means arc breakage or short circuit.

[0038] Specifically, the method for obtaining the final welding torch deviation value by performing dynamic multi-condition optimal fusion based on CART decision trees is as follows: S31: Calculate the confidence levels of the visual sensing subsystem and the arc sensing subsystem; S32: Compare the confidence levels of the visual sensing subsystem and the arc sensing subsystem with the set confidence thresholds respectively. If the confidence levels of both the visual sensing subsystem and the arc sensing subsystem are greater than or equal to the set confidence thresholds, then the information from both the visual sensing subsystem and the arc sensing subsystem is valid. The weights of the welding torch lateral deviation of the visual sensing subsystem and the welding torch lateral deviation of the arc sensing subsystem are assigned using a weighted average method. If only one of the confidence levels of the visual sensing subsystem and the arc sensing subsystem is greater than or equal to the set confidence threshold, then the weight of the one greater than or equal to the set confidence threshold is 1, and the weight of the one less than the set confidence threshold is 0. If the confidence levels of both the visual sensing subsystem and the arc sensing subsystem are less than the set confidence thresholds, then both subsystems are determined to be invalid, and historical trajectory prediction is enabled. Specifically, the confidence threshold can preferably be set to 0.7.

[0039] S33: Based on the determined weights, calculate the final welding torch deviation value according to equation (7): (7); in: This indicates the final welding torch deviation value. Represents the weights of the visual sensing subsystem. This indicates the lateral deviation of the welding torch in the vision sensing subsystem. This indicates the weights of the arc sensing subsystem. This indicates the lateral deviation of the welding torch in the arc sensing subsystem. Using the above method, the signals of the visual sensing subsystem and the arc sensing subsystem can be effectively fused. This method is significantly better than the single sensing mode in terms of tracking accuracy and stability. When the vision fails due to specular reflection or the arc becomes unstable due to arc breakage, the composite sensing can automatically reduce the weight of the failed sensor and be compensated by another sensor. This can not only ensure the continuity of weld tracking, but also effectively suppress runaway deviations caused by extreme conditions such as strong arc light interference and arc breakage. This provides a theoretical basis for solving the pain points of weak visual anti-reflection and poor arc anti-breakage in welding, and provides a reliable basis for high-precision welding automation.

[0040] S4: The actuator drives the welding torch to move according to the final welding torch deviation value, and performs deviation correction to achieve accurate tracking of the steel pipe weld.

[0041] Specifically, the actuator can drive the motor through the oscillator to move the welding torch and achieve closed-loop control of the welding torch's position and posture.

[0042] To verify the effectiveness of the dynamic confidence assessment and CART decision tree fusion method of this invention under the standard test conditions of P-MAG welding with a narrow gap (groove angle of 5°), the following comparative test was designed: The test base material was Q235B steel, and the test plate was spliced ​​together from two 400mm×40mm×15mm steel plates with a 5° narrow gap V-shaped groove. A 1.2mm diameter welding wire and 80%Ar+20%CO2 shielding gas (flow rate 15L / min) were used. The test system was based on a Frenzynes TPS3200 welding machine (pulse MAG mode), and was compared using a single laser vision tracking subsystem, a single arc tracking subsystem, and the composite sensor tracking system described in this invention. The welding path adopted a preset "oblique line" butt track. Time-varying tracking deviations were actively introduced through parameters (lateral offset distance) and (path deviation angle) to simulate actual working conditions. Under ideal alignment, the theoretical lateral displacement of the welding torch in each oscillation cycle was determined by the following formula and used as the evaluation benchmark for tracking accuracy: ; in: This represents the theoretical lateral displacement of the welding torch. Indicates the oscillation period, Indicates the oscillation frequency. Indicates the side stop time. Indicates welding speed. Indicates the angle of deviation from the path.

[0043] The welding parameters are shown in the table below: Table 1

[0044] To verify the system's robustness, extreme interference conditions such as strong arc light interference and momentary arc interruption were artificially introduced during the welding process. The experiment simultaneously collected data on the actual position of the welding torch, the raw outputs of each sensor, the fused decision commands, and confidence level data. Quantitative analysis was performed using mean absolute error (MAE), the percentage of errors ≤0.2 mm, the maximum absolute error, and the fault recovery time as core evaluation indicators. Statistically reliable data obtained from at least three repetitions of each experiment under identical conditions were analyzed. The results are shown in Table 2. A schematic diagram of the three-sensor tracking error distribution is shown below. Figure 1 As shown, the arc sensing tracking results in the single arc sensing experiment are as follows: Figure 2As shown, the visual sensing tracking results in the single visual sensing experiment are as follows: Figure 3 As shown, the composite sensing tracking results of this invention are as follows: Figure 4 As shown.

[0045] Table 2

[0046] pass Figure 2 It can be seen that in the single arc sensing test, the arc stability drops sharply after the arc is broken and then reignited, and the sensing trajectory fluctuates greatly. This unstable state causes the calculated difference in pulse frequency between the left and right extreme positions to exceed the reasonable range, which in turn causes the amplified tracking deviation, with a maximum error of 0.6-0.7 mm.

[0047] pass Figure 3 It can be seen that in the single visual sensing experiment, strong arc light interference caused significant degradation of the extracted weld features, reduced the reliability of the visual data, and caused the calculation deviation to increase abnormally to 0.5-0.6 mm; as the arc light interference weakened, the error gradually recovered to the stable range.

[0048] pass Figure 4 As can be seen, in the composite sensing experiment of this invention, although trajectory fluctuations also occurred due to arc interruption and strong arc light interference, the fluctuation amplitude was significantly smaller than that of the single sensing mode. This keyly demonstrates the core advantage of composite sensing: when the arc sensor fails due to arc interruption or the visual sensor fails due to strong arc light interference, the fusion mechanism based on the CART decision tree can dynamically reduce the weight of the failed sensor and switch to a mode dominated by the effective sensor based on real-time confidence assessment, thereby ensuring the continuity of control. Specifically, the maximum deviation of composite sensing is effectively suppressed to about 0.3 mm, and the deviation value is mainly concentrated in the high-precision range of 0 to 0.2 mm.

[0049] Table 2 summarizes the statistical results of five sets of repeated experiments, quantitatively comparing the performance of the three sensing modes: In terms of mean absolute error (MAE), the composite sensing reduced the error by 15.2% compared to single vision sensing and by 23.1% compared to single arc sensing; In terms of error distribution concentration, the composite sensing had a higher percentage of data points with errors ≤0.2 mm than vision sensing and arc sensing by 8.34 and 13.89 percentage points, respectively; In terms of adaptability to extreme conditions, the maximum tracking deviation of the composite sensing was only 49.2% of that of arc sensing and 59.6% of that of vision sensing.

[0050] The above data strongly validates that the present invention, through a dynamic confidence assessment and decision tree fusion mechanism, can effectively suppress tracking deviations caused by single-point sensor failures, achieving a comprehensive improvement in tracking accuracy and a fundamental enhancement of system robustness. Furthermore, this fusion framework possesses the ability to optimize decision rules through data learning, demonstrating its high level of environmental adaptability and intelligent potential in handling complex and variable welding conditions.

[0051] An adaptive fusion steel pipe weld seam tracking system is used to execute the adaptive fusion steel pipe weld seam tracking method described in any one of the above, which includes a vision sensing subsystem, an arc sensing subsystem, a decision fusion subsystem, and an execution mechanism. The visual sensing subsystem includes a line laser, an embedded industrial camera, and a visual sensing image processing module. The line laser and the embedded industrial camera work together to acquire the original laser image of the weld in real time. The visual sensing image processing module is used to process the original laser image of the weld to obtain the processed laser image of the weld, and to obtain the pixel coordinates of key corner points and calculate the lateral deviation of the welding torch in the visual sensing subsystem. The arc sensing subsystem includes an arc sensor and an arc signal filtering and processing module. The arc sensor is used to acquire welding current signals in real time, and the arc signal filtering and processing module is used to filter the acquired welding current signals in real time to obtain filtered welding current signals, and to calculate the lateral deviation of the welding torch in the arc sensing subsystem based on the filtered welding current signals. The decision fusion subsystem is used to dynamically fuse the welding torch lateral deviation of the vision sensing subsystem and the welding torch lateral deviation of the arc sensing subsystem based on the CART decision tree to obtain the final welding torch deviation value. The actuator is used to drive the welding torch to move according to the final welding torch deviation value, perform deviation correction, and achieve accurate tracking of the steel pipe weld.

[0052] In summary, the adaptive fusion method and system for tracking steel pipe welds provided by this invention effectively integrates the signals of the visual sensing subsystem and the arc sensing subsystem. This ensures the continuity of weld tracking and effectively suppresses runaway deviations caused by extreme conditions such as strong arc interference and arc interruption. It solves the problems of weak visual anti-reflection and poor arc interruption resistance in welding, providing a reliable basis for high-precision welding automation.

[0053] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An adaptive fusion method for tracking steel pipe welds, characterized in that: Includes the following steps: S1: During the welding process, the visual sensing subsystem acquires the original laser image of the weld seam on the welding plane of the steel pipe in real time, and captures the weld seam features after processing the original laser image of the weld seam to obtain the processed laser image of the weld seam. Then, based on the processed laser image of the weld seam, the visual sensing subsystem obtains the lateral deviation of the welding torch and transmits it to the decision fusion subsystem. S2: The arc sensing subsystem acquires the original welding current signal in real time, filters the original welding current signal to obtain the filtered welding current signal, and then calculates the lateral deviation of the welding torch of the arc sensing subsystem based on the filtered welding current signal and transmits it to the decision fusion subsystem. S3: The decision fusion subsystem performs dynamic multi-condition optimal fusion of the welding torch lateral deviation of the vision sensing subsystem and the welding torch lateral deviation of the arc sensing subsystem based on the CART decision tree to obtain the final welding torch deviation value. S4: The actuator drives the welding torch to move according to the final welding torch deviation value, and performs deviation correction to achieve accurate tracking of the steel pipe weld.

2. The adaptive fusion method for tracking steel pipe welds according to claim 1, characterized in that: The vision sensing subsystem described in step S1 includes a line laser, an embedded industrial camera, and a vision sensing image processing module. The line laser is tilted and projected onto the welding plane. The axis of the embedded industrial camera is parallel to the axis of the welding torch. The embedded industrial camera is equipped with a narrow-band filter. The line laser and the embedded industrial camera work together to acquire the original laser image of the weld in real time. The vision sensing image processing module is used to process the original laser image of the weld to obtain the processed laser image of the weld.

3. The adaptive fusion method for tracking steel pipe welds according to claim 2, characterized in that: The visual sensing image processing module processes the original laser image of the weld seam using methods including preprocessing of the original laser image of the weld seam, selection and cropping of the region of interest, image binarization, and morphological processing.

4. The adaptive fusion method for tracking steel pipe welds according to claim 3, characterized in that: The preprocessing method for the original laser image of the weld is as follows: First, based on the camera intrinsic parameter matrix obtained by calibration, geometric correction is performed using the corresponding distortion model to eliminate the geometric distortion caused by lens distortion; spatial domain filtering is used to smooth the original laser image of the weld; nonlinear contrast enhancement is performed on the smoothed laser image of the weld.

5. The adaptive fusion method for tracking steel pipe welds according to claim 3, characterized in that: The image binarization method is as follows: a threshold segmentation algorithm based on the image grayscale histogram is adopted. The optimal threshold is determined by maximizing the inter-class variance of the foreground and background pixels. Pixels with grayscale values ​​higher than the optimal threshold are set as foreground, and pixels with grayscale values ​​lower than the optimal threshold are set as background.

6. The adaptive fusion method for tracking steel pipe welds according to claim 3, characterized in that: The morphological processing method is as follows: the Zhang-Suen thinning algorithm is applied to extract the skeleton of the binarized laser region, and the target region is gradually converged to the central skeleton through iterative boundary erosion operation.

7. The adaptive fusion method for tracking steel pipe welds according to claim 3, characterized in that: In step S1, the lateral deviation of the welding torch in the vision sensing subsystem is obtained using the Shi-Tomasi method based on the processed weld laser image. This specifically includes the following steps: S11: Based on the processed weld laser image, calculate the average horizontal coordinate of all effective feature points in the calibration image according to equation (1), and establish the average horizontal coordinate of the calibration pixels as the system reference position: （1）； in: This represents the average value of the x-coordinate of the calibrated pixels. This represents the total number of feature points detected in the calibration image. Indicates the first in the calibration image The pixel x-coordinates of each feature point; S12: During the welding tracking process, the average value of the current pixel abscissa of all effective feature points is calculated according to equation (2) for each frame of real-time acquired image, and the average value of the current pixel abscissa is used as the center pixel position of the weld represented by the laser stripe in the current frame: （2）； in: This represents the average value of the current pixel's x-coordinate. This indicates the number of valid feature points detected in the current frame. Indicates the first in the current frame i The pixel x-coordinates of each feature point; S13: Calculate the pixel offset based on the average horizontal coordinate of the current pixel and the average horizontal coordinate of the calibrated pixel according to equation (3). : （3）； S14: Convert the pixel offset into the welding torch lateral deviation of the visual sensing subsystem according to equation (4). : （4）； in: This represents the visual sensing scaling factor.

8. The adaptive fusion method for tracking steel pipe welds according to claim 1, characterized in that: The method for calculating the lateral deviation of the welding torch in the arc sensing subsystem in step S2 is as follows: S21: Perform sliding mean filtering on the original current signal to obtain the filtered current signal, identify each current pulse in the filtered current signal and extract its peak current; S22: Based on the welding torch oscillation position, calculate the continuous position near the left and right limit positions according to formula (5). Average peak current and peak current difference of each pulse: （5）； in: Indicates the continuity near the left extreme position Average peak current of each pulse Indicates the first [number]th [unit] within the left limit window Peak current of each pulse Indicates the continuity near the right extreme position Average peak current of each pulse Indicates the right limit window. Peak current of each pulse This represents the peak current difference; S23: Calculate the lateral deviation of the welding torch in the arc sensing subsystem based on the peak current difference according to equation (6). : （6）； in: This represents the proportional coefficient of the arc sensing sensor. This indicates the offset.

9. The adaptive fusion method for tracking steel pipe welds according to claim 1, characterized in that: The method for obtaining the final welding torch deviation value in step S3 by performing dynamic multi-condition optimal fusion based on CART decision tree is as follows: S31: Calculate the confidence levels of the visual sensing subsystem and the arc sensing subsystem; S32: Compare the confidence levels of the visual sensing subsystem and the arc sensing subsystem with the set confidence thresholds respectively. If the confidence levels of both the visual sensing subsystem and the arc sensing subsystem are greater than or equal to the set confidence thresholds, then the information of both the visual sensing subsystem and the arc sensing subsystem is valid. The weights of the welding torch lateral deviation of the visual sensing subsystem and the welding torch lateral deviation of the arc sensing subsystem are assigned using a weighted average method. If only one of the confidence levels of the visual sensing subsystem and the arc sensing subsystem is greater than or equal to the set confidence threshold, then the weight of the confidence level greater than or equal to the set confidence threshold is 1, and the weight of the confidence level less than the set confidence threshold is 0. If the confidence levels of both the visual sensing subsystem and the arc sensing subsystem are less than the set confidence threshold, then both subsystems are determined to be in failure, and historical trajectory prediction is enabled. S33: Based on the determined weights, calculate the final welding torch deviation value according to equation (7): （7）； in: This indicates the final welding torch deviation value. Represents the weights of the visual sensing subsystem. This indicates the lateral deviation of the welding torch in the vision sensing subsystem. This indicates the weights of the arc sensing subsystem. This indicates the lateral deviation of the welding torch in the arc sensing subsystem.

10. An adaptive fusion steel pipe weld seam tracking system, characterized in that: The method for performing an adaptive fusion steel pipe weld tracking method as described in any one of claims 1 to 9 includes a visual sensing subsystem, an arc sensing subsystem, a decision fusion subsystem, and an execution mechanism. The visual sensing subsystem includes a line laser, an embedded industrial camera, and a visual sensing image processing module. The line laser and the embedded industrial camera work together to acquire the original laser image of the weld in real time. The visual sensing image processing module is used to process the original laser image of the weld to obtain the processed laser image of the weld, and to obtain the pixel coordinates of key corner points and calculate the lateral deviation of the welding torch in the visual sensing subsystem. The arc sensing subsystem includes an arc sensor and an arc signal filtering and processing module. The arc sensor is used to acquire welding current signals in real time, and the arc signal filtering and processing module is used to filter the acquired welding current signals in real time to obtain filtered welding current signals, and to calculate the lateral deviation of the welding torch in the arc sensing subsystem based on the filtered welding current signals. The decision fusion subsystem is used to dynamically fuse the welding torch lateral deviation of the vision sensing subsystem and the welding torch lateral deviation of the arc sensing subsystem based on the CART decision tree to obtain the final welding torch deviation value. The actuator is used to drive the welding torch to move according to the final welding torch deviation value, perform deviation correction, and achieve accurate tracking of the steel pipe weld.

Images